Attachment 137571 Details for Bug 105200 – clover debug imagemagick.ll

clover debug imagemagick.ll

imagemagick.ll (text/plain), 564.49 KB, created by nixscripter on 2018-02-24 05:23:29 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: nixscripter

Created: 2018-02-24 05:23:29 UTC

Size: 564.49 KB

patch

obsolete

>; ModuleID = 'input.cl'
>source_filename = "input.cl"
>target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-A5"
>target triple = "r600--"
>
>%struct.mwc64x_state_t = type { i32, i32 }
>
>@WaveletDenoise.noise = private unnamed_addr addrspace(4) constant [8 x float] [float 0x3FE99B3D00000000, float 0x3FD1810620000000, float 0x3FBEC56D60000000, float 0x3FADF3B640000000, float 0x3F9DCC6400000000, float 0x3F8F212D80000000, float 0x3F80624DE0000000, float 0x3F7205BC00000000], align 4
>@WaveletDenoise.buffer = internal unnamed_addr addrspace(3) global [4096 x float] undef, align 4
>@LOG_INV_TBL = linkonce_odr dso_local local_unnamed_addr addrspace(4) constant [129 x float] [float 2.000000e+00, float 0x3FFFC07F00000000, float 0x3FFF81F820000000, float 0x3FFF4465A0000000, float 0x3FFF07C200000000, float 0x3FFECC07C0000000, float 0x3FFE9131A0000000, float 0x3FFE573AC0000000, float 0x3FFE1E1E20000000, float 0x3FFDE5D6E0000000, float 0x3FFDAE6080000000, float 0x3FFD77B660000000, float 0x3FFD41D420000000, float 0x3FFD0CB580000000, float 0x3FFCD85680000000, float 0x3FFCA4B300000000, float 0x3FFC71C720000000, float 0x3FFC3F8F00000000, float 0x3FFC0E0700000000, float 0x3FFBDD2B80000000, float 0x3FFBACF920000000, float 0x3FFB7D6C40000000, float 0x3FFB4E81C0000000, float 0x3FFB203640000000, float 0x3FFAF286C0000000, float 0x3FFAC57020000000, float 0x3FFA98EF60000000, float 0x3FFA6D01A0000000, float 0x3FFA41A420000000, float 0x3FFA16D400000000, float 0x3FF9EC8EA0000000, float 0x3FF9C2D140000000, float 0x3FF99999A0000000, float 0x3FF970E500000000, float 0x3FF948B100000000, float 0x3FF920FB40000000, float 0x3FF8F9C180000000, float 0x3FF8D30180000000, float 0x3FF8ACB900000000, float 0x3FF886E600000000, float 0x3FF8618620000000, float 0x3FF83C9780000000, float 0x3FF8181820000000, float 0x3FF7F40600000000, float 0x3FF7D05F40000000, float 0x3FF7AD2200000000, float 0x3FF78A4C80000000, float 0x3FF767DCE0000000, float 0x3FF745D180000000, float 0x3FF7242880000000, float 0x3FF702E060000000, float 0x3FF6E1F760000000, float 0x3FF6C16C20000000, float 0x3FF6A13CE0000000, float 0x3FF6816820000000, float 0x3FF661EC60000000, float 0x3FF642C860000000, float 0x3FF623FA80000000, float 0x3FF6058160000000, float 0x3FF5E75BC0000000, float 0x3FF5C98820000000, float 0x3FF5AC0560000000, float 0x3FF58ED240000000, float 0x3FF571ED40000000, float 0x3FF5555560000000, float 0x3FF5390940000000, float 0x3FF51D07E0000000, float 0x3FF5015020000000, float 0x3FF4E5E0A0000000, float 0x3FF4CAB880000000, float 0x3FF4AFD6A0000000, float 0x3FF49539E0000000, float 0x3FF47AE140000000, float 0x3FF460CBC0000000, float 0x3FF446F860000000, float 0x3FF42D6620000000, float 0x3FF4141420000000, float 0x3FF3FB0140000000, float 0x3FF3E22CC0000000, float 0x3FF3C995A0000000, float 0x3FF3B13B20000000, float 0x3FF3991C20000000, float 0x3FF3813820000000, float 0x3FF3698E00000000, float 0x3FF3521D00000000, float 0x3FF33AE460000000, float 0x3FF323E340000000, float 0x3FF30D1900000000, float 0x3FF2F684C0000000, float 0x3FF2E025C0000000, float 0x3FF2C9FB40000000, float 0x3FF2B404A0000000, float 0x3FF29E4120000000, float 0x3FF288B020000000, float 0x3FF27350C0000000, float 0x3FF25E2280000000, float 0x3FF24924A0000000, float 0x3FF2345680000000, float 0x3FF21FB780000000, float 0x3FF20B4700000000, float 0x3FF1F70480000000, float 0x3FF1E2EF40000000, float 0x3FF1CF06A0000000, float 0x3FF1BB4A40000000, float 0x3FF1A7B960000000, float 0x3FF1945380000000, float 0x3FF1811820000000, float 0x3FF16E0680000000, float 0x3FF15B1E60000000, float 0x3FF1485F00000000, float 0x3FF135C820000000, float 0x3FF12358E0000000, float 0x3FF1111120000000, float 0x3FF0FEF020000000, float 0x3FF0ECF560000000, float 0x3FF0DB20A0000000, float 0x3FF0C97140000000, float 0x3FF0B7E6E0000000, float 0x3FF0A68100000000, float 0x3FF0953F40000000, float 0x3FF0842100000000, float 0x3FF0732600000000, float 0x3FF0624DE0000000, float 0x3FF0519800000000, float 0x3FF0410420000000, float 0x3FF03091C0000000, float 0x3FF0204080000000, float 0x3FF0101020000000, float 1.000000e+00], align 4
>@LOG2_TBL = linkonce_odr dso_local local_unnamed_addr addrspace(4) constant [129 x <2 x float>] [<2 x float> zeroinitializer, <2 x float> <float 0x3F86F80000000000, float 0x3EE942DBA0000000>, <2 x float> <float 0x3F96E00000000000, float 0x3EFE5A1700000000>, <2 x float> <float 0x3FA1180000000000, float 0x3F03475440000000>, <2 x float> <float 0x3FA6B80000000000, float 0x3EF69BAC60000000>, <2 x float> <float 0x3FAC480000000000, float 0x3F07EAE420000000>, <2 x float> <float 0x3FB0E80000000000, float 0x3F09C4FD00000000>, <2 x float> <float 0x3FB3A80000000000, float 0x3F017EE920000000>, <2 x float> <float 0x3FB6600000000000, float 0x3F0FB7D640000000>, <2 x float> <float 0x3FB9180000000000, float 0x3EE42DC8C0000000>, <2 x float> <float 0x3FBBC80000000000, float 0x3ED0902B60000000>, <2 x float> <float 0x3FBE700000000000, float 0x3F07608BE0000000>, <2 x float> <float 0x3FC0880000000000, float 0x3F21623360000000>, <2 x float> <float 0x3FC1D80000000000, float 0x3F23465D40000000>, <2 x float> <float 0x3FC3280000000000, float 0x3F174F13C0000000>, <2 x float> <float 0x3FC4700000000000, float 0x3F2AA7E600000000>, <2 x float> <float 0x3FC5C00000000000, float 0x3ECA39FBC0000000>, <2 x float> <float 0x3FC7000000000000, float 0x3F2D0B53A0000000>, <2 x float> <float 0x3FC8480000000000, float 0x3F20AF40A0000000>, <2 x float> <float 0x3FC9880000000000, float 0x3F2B741DE0000000>, <2 x float> <float 0x3FCAC80000000000, float 0x3F2D78B6C0000000>, <2 x float> <float 0x3FCC080000000000, float 0x3F26DB3760000000>, <2 x float> <float 0x3FCD480000000000, float 0x3F0EE4C320000000>, <2 x float> <float 0x3FCE800000000000, float 0x3F202F9D20000000>, <2 x float> <float 0x3FCFB80000000000, float 0x3F205AE400000000>, <2 x float> <float 0x3FD0780000000000, float 0x3F10ADBB00000000>, <2 x float> <float 0x3FD1100000000000, float 0x3F283ED680000000>, <2 x float> <float 0x3FD1A80000000000, float 0x3F3016CA40000000>, <2 x float> <float 0x3FD2400000000000, float 0x3F301EAC20000000>, <2 x float> <float 0x3FD2D80000000000, float 0x3F2887E260000000>, <2 x float> <float 0x3FD3700000000000, float 0x3F124CEA40000000>, <2 x float> <float 3.125000e-01, float 0x3F3918EC60000000>, <2 x float> <float 0x3FD4980000000000, float 0x3F23C25E60000000>, <2 x float> <float 0x3FD5280000000000, float 0x3F36F7F120000000>, <2 x float> <float 0x3FD5C00000000000, float 0x3EDA39FBC0000000>, <2 x float> <float 0x3FD6500000000000, float 0x3F18FE4660000000>, <2 x float> <float 0x3FD6E00000000000, float 0x3F210E6CE0000000>, <2 x float> <float 0x3FD7700000000000, float 0x3F1D2BA7E0000000>, <2 x float> <float 3.750000e-01, float 0x3F04AC62C0000000>, <2 x float> <float 0x3FD8880000000000, float 0x3F3A71CB80000000>, <2 x float> <float 0x3FD9180000000000, float 0x3F2DD448E0000000>, <2 x float> <float 0x3FD9A80000000000, float 0x3EA1C8F100000000>, <2 x float> <float 0x3FDA300000000000, float 0x3F2BB053E0000000>, <2 x float> <float 0x3FDAB80000000000, float 0x3F3861E5E0000000>, <2 x float> <float 0x3FDB400000000000, float 0x3F3FAFDCE0000000>, <2 x float> <float 0x3FDBD00000000000, float 0x3F0E5D3CE0000000>, <2 x float> <float 0x3FDC580000000000, float 0x3F12FAD280000000>, <2 x float> <float 0x3FDCE00000000000, float 0x3F04924740000000>, <2 x float> <float 0x3FDD600000000000, float 0x3F3D4F80C0000000>, <2 x float> <float 0x3FDDE80000000000, float 0x3F34FF5100000000>, <2 x float> <float 0x3FDE700000000000, float 0x3F23550F20000000>, <2 x float> <float 0x3FDEF00000000000, float 0x3F3B59CCA0000000>, <2 x float> <float 0x3FDF780000000000, float 0x3F242B4640000000>, <2 x float> <float 0x3FDFF80000000000, float 0x3F35E66A00000000>, <2 x float> <float 0x3FE0380000000000, float 0x3F4F6A2E40000000>, <2 x float> <float 5.156250e-01, float 0x3F139E4FE0000000>, <2 x float> <float 0x3FE0C00000000000, float 0x3F20500D60000000>, <2 x float> <float 5.312500e-01, float 0x3F213B1520000000>, <2 x float> <float 0x3FE1400000000000, float 0x3F193F5420000000>, <2 x float> <float 5.468750e-01, float 0x3EF467B940000000>, <2 x float> <float 0x3FE1B80000000000, float 0x3F4CC47A40000000>, <2 x float> <float 0x3FE1F80000000000, float 0x3F478F4C20000000>, <2 x float> <float 0x3FE2380000000000, float 0x3F41075080000000>, <2 x float> <float 0x3FE2780000000000, float 0x3F32602C20000000>, <2 x float> <float 0x3FE2B80000000000, float 0x3EBA39FBC0000000>, <2 x float> <float 0x3FE2F00000000000, float 0x3F45A1D7A0000000>, <2 x float> <float 0x3FE3300000000000, float 0x3F33E355A0000000>, <2 x float> <float 0x3FE3680000000000, float 0x3F4CFFEDA0000000>, <2 x float> <float 0x3FE3A80000000000, float 0x3F3D9FD500000000>, <2 x float> <float 0x3FE3E00000000000, float 0x3F4F64DE60000000>, <2 x float> <float 0x3FE4200000000000, float 0x3F3D83F4C0000000>, <2 x float> <float 0x3FE4580000000000, float 0x3F4CEA6280000000>, <2 x float> <float 0x3FE4980000000000, float 0x3F33C25E60000000>, <2 x float> <float 0x3FE4D00000000000, float 0x3F45A96CC0000000>, <2 x float> <float 0x3FE5100000000000, float 0x3EE18708A0000000>, <2 x float> <float 0x3FE5480000000000, float 0x3F33746520000000>, <2 x float> <float 6.718750e-01, float 0x3F42089A60000000>, <2 x float> <float 0x3FE5B80000000000, float 0x3F493432C0000000>, <2 x float> <float 0x3FE5F00000000000, float 0x3F4F3FD060000000>, <2 x float> <float 0x3FE6300000000000, float 0x3F20B8F540000000>, <2 x float> <float 0x3FE6680000000000, float 0x3F30047220000000>, <2 x float> <float 0x3FE6A00000000000, float 0x3F357CF2C0000000>, <2 x float> <float 0x3FE6D80000000000, float 0x3F38CB53A0000000>, <2 x float> <float 0x3FE7100000000000, float 0x3F39F4D8A0000000>, <2 x float> <float 0x3FE7480000000000, float 0x3F38FEB260000000>, <2 x float> <float 7.343750e-01, float 0x3F35EDFEE0000000>, <2 x float> <float 0x3FE7B80000000000, float 0x3F30C7C9A0000000>, <2 x float> <float 0x3FE7F00000000000, float 0x3F23221820000000>, <2 x float> <float 0x3FE8280000000000, float 0x3ED3AB7CE0000000>, <2 x float> <float 0x3FE8580000000000, float 0x3F4A82C2C0000000>, <2 x float> <float 0x3FE8900000000000, float 0x3F43DD2C00000000>, <2 x float> <float 0x3FE8C80000000000, float 0x3F3871DA40000000>, <2 x float> <float 7.812500e-01, float 0x3F1CC2C000000000>, <2 x float> <float 0x3FE9300000000000, float 0x3F49FDB680000000>, <2 x float> <float 0x3FE9680000000000, float 0x3F3ED69560000000>, <2 x float> <float 0x3FE9A00000000000, float 0x3F1F1A7600000000>, <2 x float> <float 0x3FE9D00000000000, float 0x3F4767F540000000>, <2 x float> <float 0x3FEA080000000000, float 0x3F33F6D260000000>, <2 x float> <float 0x3FEA380000000000, float 0x3F4B9FCE20000000>, <2 x float> <float 0x3FEA700000000000, float 0x3F38AE8160000000>, <2 x float> <float 0x3FEAA00000000000, float 0x3F4C23D600000000>, <2 x float> <float 0x3FEAD80000000000, float 0x3F360F3880000000>, <2 x float> <float 0x3FEB080000000000, float 0x3F49049AE0000000>, <2 x float> <float 0x3FEB400000000000, float 0x3F28734A80000000>, <2 x float> <float 0x3FEB700000000000, float 0x3F42523D40000000>, <2 x float> <float 0x3FEBA00000000000, float 0x3F4DA6CE60000000>, <2 x float> <float 0x3FEBD80000000000, float 0x3F3038E620000000>, <2 x float> <float 0x3FEC080000000000, float 0x3F41B511E0000000>, <2 x float> <float 0x3FEC380000000000, float 0x3F4A728B80000000>, <2 x float> <float 0x3FEC700000000000, float 0x3F12B5D220000000>, <2 x float> <float 0x3FECA00000000000, float 0x3F32C6E540000000>, <2 x float> <float 0x3FECD00000000000, float 0x3F3F350640000000>, <2 x float> <float 9.062500e-01, float 0x3F44FDB480000000>, <2 x float> <float 0x3FED300000000000, float 0x3F498EC9E0000000>, <2 x float> <float 0x3FED600000000000, float 0x3F4D4F80C0000000>, <2 x float> <float 0x3FED980000000000, float 0x3EE0643D60000000>, <2 x float> <float 0x3FEDC80000000000, float 0x3F133567E0000000>, <2 x float> <float 0x3FEDF80000000000, float 0x3F1E0410C0000000>, <2 x float> <float 0x3FEE280000000000, float 0x3F2142E0E0000000>, <2 x float> <float 0x3FEE580000000000, float 0x3F2063C880000000>, <2 x float> <float 0x3FEE880000000000, float 0x3F18D66C40000000>, <2 x float> <float 0x3FEEB80000000000, float 0x3F057E32A0000000>, <2 x float> <float 0x3FEEE00000000000, float 0x3F4ED1C6C0000000>, <2 x float> <float 0x3FEF100000000000, float 0x3F4B8A0760000000>, <2 x float> <float 0x3FEF400000000000, float 0x3F47822F20000000>, <2 x float> <float 0x3FEF700000000000, float 0x3F42BBC3A0000000>, <2 x float> <float 0x3FEFA00000000000, float 0x3F3A708BA0000000>, <2 x float> <float 0x3FEFD00000000000, float 0x3F2BE4C7E0000000>, <2 x float> <float 1.000000e+00, float 0.000000e+00>], align 8
>
>; Function Attrs: norecurse nounwind readnone
>define i64 @MWC_AddMod64(i64 %a, i64 %b, i64 %M) local_unnamed_addr #0 {
>entry:
>  %add = add i64 %b, %a
>  %cmp = icmp ult i64 %add, %M
>  br i1 %cmp, label %lor.lhs.false, label %if.then
>
>lor.lhs.false:                                    ; preds = %entry
>  %0 = uitofp i64 %add to float
>  %1 = uitofp i64 %a to float
>  %cmp2 = fcmp olt float %0, %1
>  br i1 %cmp2, label %if.then, label %if.end
>
>if.then:                                          ; preds = %entry, %lor.lhs.false
>  %sub = sub i64 %add, %M
>  br label %if.end
>
>if.end:                                           ; preds = %if.then, %lor.lhs.false
>  %v.0 = phi i64 [ %sub, %if.then ], [ %add, %lor.lhs.false ]
>  ret i64 %v.0
>}
>
>; Function Attrs: argmemonly nounwind
>declare void @llvm.lifetime.start.p5i8(i64, i8 addrspace(5)* nocapture) #1
>
>; Function Attrs: argmemonly nounwind
>declare void @llvm.lifetime.end.p5i8(i64, i8 addrspace(5)* nocapture) #1
>
>; Function Attrs: norecurse nounwind readnone
>define i64 @MWC_MulMod64(i64 %a, i64 %b, i64 %M) local_unnamed_addr #0 {
>entry:
>  %cmp17 = icmp eq i64 %a, 0
>  br i1 %cmp17, label %while.end, label %while.body
>
>while.body:                                       ; preds = %entry, %MWC_AddMod64.exit16
>  %r.020 = phi i64 [ %r.1, %MWC_AddMod64.exit16 ], [ 0, %entry ]
>  %a.addr.019 = phi i64 [ %shr, %MWC_AddMod64.exit16 ], [ %a, %entry ]
>  %b.addr.018 = phi i64 [ %v.0.i15, %MWC_AddMod64.exit16 ], [ %b, %entry ]
>  %and = and i64 %a.addr.019, 1
>  %tobool = icmp eq i64 %and, 0
>  br i1 %tobool, label %if.end, label %if.then
>
>if.then:                                          ; preds = %while.body
>  %add.i = add i64 %r.020, %b.addr.018
>  %cmp.i = icmp ult i64 %add.i, %M
>  br i1 %cmp.i, label %lor.lhs.false.i, label %if.then.i
>
>lor.lhs.false.i:                                  ; preds = %if.then
>  %0 = uitofp i64 %add.i to float
>  %1 = uitofp i64 %r.020 to float
>  %cmp2.i = fcmp olt float %0, %1
>  br i1 %cmp2.i, label %if.then.i, label %if.end
>
>if.then.i:                                        ; preds = %lor.lhs.false.i, %if.then
>  %sub.i = sub i64 %add.i, %M
>  br label %if.end
>
>if.end:                                           ; preds = %if.then.i, %lor.lhs.false.i, %while.body
>  %r.1 = phi i64 [ %r.020, %while.body ], [ %sub.i, %if.then.i ], [ %add.i, %lor.lhs.false.i ]
>  %add.i9 = shl i64 %b.addr.018, 1
>  %cmp.i10 = icmp ult i64 %add.i9, %M
>  br i1 %cmp.i10, label %lor.lhs.false.i12, label %if.then.i14
>
>lor.lhs.false.i12:                                ; preds = %if.end
>  %2 = uitofp i64 %add.i9 to float
>  %3 = uitofp i64 %b.addr.018 to float
>  %cmp2.i11 = fcmp olt float %2, %3
>  br i1 %cmp2.i11, label %if.then.i14, label %MWC_AddMod64.exit16
>
>if.then.i14:                                      ; preds = %lor.lhs.false.i12, %if.end
>  %sub.i13 = sub i64 %add.i9, %M
>  br label %MWC_AddMod64.exit16
>
>MWC_AddMod64.exit16:                              ; preds = %lor.lhs.false.i12, %if.then.i14
>  %v.0.i15 = phi i64 [ %sub.i13, %if.then.i14 ], [ %add.i9, %lor.lhs.false.i12 ]
>  %shr = lshr i64 %a.addr.019, 1
>  %cmp = icmp eq i64 %shr, 0
>  br i1 %cmp, label %while.end, label %while.body
>
>while.end:                                        ; preds = %MWC_AddMod64.exit16, %entry
>  %r.0.lcssa = phi i64 [ 0, %entry ], [ %r.1, %MWC_AddMod64.exit16 ]
>  ret i64 %r.0.lcssa
>}
>
>; Function Attrs: norecurse nounwind readnone
>define i64 @MWC_PowMod64(i64 %a, i64 %e, i64 %M) local_unnamed_addr #0 {
>entry:
>  %cmp38 = icmp eq i64 %e, 0
>  br i1 %cmp38, label %while.end, label %while.body
>
>while.body:                                       ; preds = %entry, %MWC_MulMod64.exit37
>  %acc.041 = phi i64 [ %acc.1, %MWC_MulMod64.exit37 ], [ 1, %entry ]
>  %sqr.040 = phi i64 [ %r.0.lcssa.i36, %MWC_MulMod64.exit37 ], [ %a, %entry ]
>  %e.addr.039 = phi i64 [ %shr, %MWC_MulMod64.exit37 ], [ %e, %entry ]
>  %and = and i64 %e.addr.039, 1
>  %tobool = icmp eq i64 %and, 0
>  br i1 %tobool, label %if.end, label %if.then
>
>if.then:                                          ; preds = %while.body
>  %cmp17.i = icmp eq i64 %acc.041, 0
>  br i1 %cmp17.i, label %if.end, label %while.body.i
>
>while.body.i:                                     ; preds = %if.then, %MWC_AddMod64.exit16.i
>  %r.020.i = phi i64 [ %r.1.i, %MWC_AddMod64.exit16.i ], [ 0, %if.then ]
>  %a.addr.019.i = phi i64 [ %shr.i, %MWC_AddMod64.exit16.i ], [ %acc.041, %if.then ]
>  %b.addr.018.i = phi i64 [ %v.0.i15.i, %MWC_AddMod64.exit16.i ], [ %sqr.040, %if.then ]
>  %and.i = and i64 %a.addr.019.i, 1
>  %tobool.i = icmp eq i64 %and.i, 0
>  br i1 %tobool.i, label %if.end.i, label %if.then.i
>
>if.then.i:                                        ; preds = %while.body.i
>  %add.i.i = add i64 %b.addr.018.i, %r.020.i
>  %cmp.i.i = icmp ult i64 %add.i.i, %M
>  br i1 %cmp.i.i, label %lor.lhs.false.i.i, label %if.then.i.i
>
>lor.lhs.false.i.i:                                ; preds = %if.then.i
>  %0 = uitofp i64 %add.i.i to float
>  %1 = uitofp i64 %r.020.i to float
>  %cmp2.i.i = fcmp olt float %0, %1
>  br i1 %cmp2.i.i, label %if.then.i.i, label %if.end.i
>
>if.then.i.i:                                      ; preds = %lor.lhs.false.i.i, %if.then.i
>  %sub.i.i = sub i64 %add.i.i, %M
>  br label %if.end.i
>
>if.end.i:                                         ; preds = %if.then.i.i, %lor.lhs.false.i.i, %while.body.i
>  %r.1.i = phi i64 [ %r.020.i, %while.body.i ], [ %sub.i.i, %if.then.i.i ], [ %add.i.i, %lor.lhs.false.i.i ]
>  %add.i9.i = shl i64 %b.addr.018.i, 1
>  %cmp.i10.i = icmp ult i64 %add.i9.i, %M
>  br i1 %cmp.i10.i, label %lor.lhs.false.i12.i, label %if.then.i14.i
>
>lor.lhs.false.i12.i:                              ; preds = %if.end.i
>  %2 = uitofp i64 %add.i9.i to float
>  %3 = uitofp i64 %b.addr.018.i to float
>  %cmp2.i11.i = fcmp olt float %2, %3
>  br i1 %cmp2.i11.i, label %if.then.i14.i, label %MWC_AddMod64.exit16.i
>
>if.then.i14.i:                                    ; preds = %lor.lhs.false.i12.i, %if.end.i
>  %sub.i13.i = sub i64 %add.i9.i, %M
>  br label %MWC_AddMod64.exit16.i
>
>MWC_AddMod64.exit16.i:                            ; preds = %if.then.i14.i, %lor.lhs.false.i12.i
>  %v.0.i15.i = phi i64 [ %sub.i13.i, %if.then.i14.i ], [ %add.i9.i, %lor.lhs.false.i12.i ]
>  %shr.i = lshr i64 %a.addr.019.i, 1
>  %cmp.i = icmp eq i64 %shr.i, 0
>  br i1 %cmp.i, label %if.end, label %while.body.i
>
>if.end:                                           ; preds = %MWC_AddMod64.exit16.i, %if.then, %while.body
>  %acc.1 = phi i64 [ %acc.041, %while.body ], [ 0, %if.then ], [ %r.1.i, %MWC_AddMod64.exit16.i ]
>  %cmp17.i10 = icmp eq i64 %sqr.040, 0
>  br i1 %cmp17.i10, label %MWC_MulMod64.exit37, label %while.body.i16
>
>while.body.i16:                                   ; preds = %if.end, %MWC_AddMod64.exit16.i35
>  %r.020.i11 = phi i64 [ %r.1.i24, %MWC_AddMod64.exit16.i35 ], [ 0, %if.end ]
>  %a.addr.019.i12 = phi i64 [ %shr.i33, %MWC_AddMod64.exit16.i35 ], [ %sqr.040, %if.end ]
>  %b.addr.018.i13 = phi i64 [ %v.0.i15.i32, %MWC_AddMod64.exit16.i35 ], [ %sqr.040, %if.end ]
>  %and.i14 = and i64 %a.addr.019.i12, 1
>  %tobool.i15 = icmp eq i64 %and.i14, 0
>  br i1 %tobool.i15, label %if.end.i27, label %if.then.i19
>
>if.then.i19:                                      ; preds = %while.body.i16
>  %add.i.i17 = add i64 %b.addr.018.i13, %r.020.i11
>  %cmp.i.i18 = icmp ult i64 %add.i.i17, %M
>  br i1 %cmp.i.i18, label %lor.lhs.false.i.i21, label %if.then.i.i23
>
>lor.lhs.false.i.i21:                              ; preds = %if.then.i19
>  %4 = uitofp i64 %add.i.i17 to float
>  %5 = uitofp i64 %r.020.i11 to float
>  %cmp2.i.i20 = fcmp olt float %4, %5
>  br i1 %cmp2.i.i20, label %if.then.i.i23, label %if.end.i27
>
>if.then.i.i23:                                    ; preds = %lor.lhs.false.i.i21, %if.then.i19
>  %sub.i.i22 = sub i64 %add.i.i17, %M
>  br label %if.end.i27
>
>if.end.i27:                                       ; preds = %if.then.i.i23, %lor.lhs.false.i.i21, %while.body.i16
>  %r.1.i24 = phi i64 [ %r.020.i11, %while.body.i16 ], [ %sub.i.i22, %if.then.i.i23 ], [ %add.i.i17, %lor.lhs.false.i.i21 ]
>  %add.i9.i25 = shl i64 %b.addr.018.i13, 1
>  %cmp.i10.i26 = icmp ult i64 %add.i9.i25, %M
>  br i1 %cmp.i10.i26, label %lor.lhs.false.i12.i29, label %if.then.i14.i31
>
>lor.lhs.false.i12.i29:                            ; preds = %if.end.i27
>  %6 = uitofp i64 %add.i9.i25 to float
>  %7 = uitofp i64 %b.addr.018.i13 to float
>  %cmp2.i11.i28 = fcmp olt float %6, %7
>  br i1 %cmp2.i11.i28, label %if.then.i14.i31, label %MWC_AddMod64.exit16.i35
>
>if.then.i14.i31:                                  ; preds = %lor.lhs.false.i12.i29, %if.end.i27
>  %sub.i13.i30 = sub i64 %add.i9.i25, %M
>  br label %MWC_AddMod64.exit16.i35
>
>MWC_AddMod64.exit16.i35:                          ; preds = %if.then.i14.i31, %lor.lhs.false.i12.i29
>  %v.0.i15.i32 = phi i64 [ %sub.i13.i30, %if.then.i14.i31 ], [ %add.i9.i25, %lor.lhs.false.i12.i29 ]
>  %shr.i33 = lshr i64 %a.addr.019.i12, 1
>  %cmp.i34 = icmp eq i64 %shr.i33, 0
>  br i1 %cmp.i34, label %MWC_MulMod64.exit37, label %while.body.i16
>
>MWC_MulMod64.exit37:                              ; preds = %MWC_AddMod64.exit16.i35, %if.end
>  %r.0.lcssa.i36 = phi i64 [ 0, %if.end ], [ %r.1.i24, %MWC_AddMod64.exit16.i35 ]
>  %shr = lshr i64 %e.addr.039, 1
>  %cmp = icmp eq i64 %shr, 0
>  br i1 %cmp, label %while.end, label %while.body
>
>while.end:                                        ; preds = %MWC_MulMod64.exit37, %entry
>  %acc.0.lcssa = phi i64 [ 1, %entry ], [ %acc.1, %MWC_MulMod64.exit37 ]
>  ret i64 %acc.0.lcssa
>}
>
>; Function Attrs: norecurse nounwind readnone
>define <2 x i32> @MWC_SkipImpl_Mod64(<2 x i32> %curr, i64 %A, i64 %M, i64 %distance) local_unnamed_addr #0 {
>entry:
>  %cmp38.i = icmp eq i64 %distance, 0
>  br i1 %cmp38.i, label %MWC_PowMod64.exit, label %while.body.i
>
>while.body.i:                                     ; preds = %entry, %MWC_MulMod64.exit37.i
>  %acc.041.i = phi i64 [ %acc.1.i, %MWC_MulMod64.exit37.i ], [ 1, %entry ]
>  %sqr.040.i = phi i64 [ %r.0.lcssa.i36.i, %MWC_MulMod64.exit37.i ], [ %A, %entry ]
>  %e.addr.039.i = phi i64 [ %shr.i, %MWC_MulMod64.exit37.i ], [ %distance, %entry ]
>  %and.i = and i64 %e.addr.039.i, 1
>  %tobool.i = icmp eq i64 %and.i, 0
>  br i1 %tobool.i, label %if.end.i, label %if.then.i
>
>if.then.i:                                        ; preds = %while.body.i
>  %cmp17.i.i = icmp eq i64 %acc.041.i, 0
>  br i1 %cmp17.i.i, label %if.end.i, label %while.body.i.i
>
>while.body.i.i:                                   ; preds = %if.then.i, %MWC_AddMod64.exit16.i.i
>  %r.020.i.i = phi i64 [ %r.1.i.i, %MWC_AddMod64.exit16.i.i ], [ 0, %if.then.i ]
>  %a.addr.019.i.i = phi i64 [ %shr.i.i, %MWC_AddMod64.exit16.i.i ], [ %acc.041.i, %if.then.i ]
>  %b.addr.018.i.i = phi i64 [ %v.0.i15.i.i, %MWC_AddMod64.exit16.i.i ], [ %sqr.040.i, %if.then.i ]
>  %and.i.i = and i64 %a.addr.019.i.i, 1
>  %tobool.i.i = icmp eq i64 %and.i.i, 0
>  br i1 %tobool.i.i, label %if.end.i.i, label %if.then.i.i
>
>if.then.i.i:                                      ; preds = %while.body.i.i
>  %add.i.i.i = add i64 %b.addr.018.i.i, %r.020.i.i
>  %cmp.i.i.i = icmp ult i64 %add.i.i.i, %M
>  br i1 %cmp.i.i.i, label %lor.lhs.false.i.i.i, label %if.then.i.i.i
>
>lor.lhs.false.i.i.i:                              ; preds = %if.then.i.i
>  %0 = uitofp i64 %add.i.i.i to float
>  %1 = uitofp i64 %r.020.i.i to float
>  %cmp2.i.i.i = fcmp olt float %0, %1
>  br i1 %cmp2.i.i.i, label %if.then.i.i.i, label %if.end.i.i
>
>if.then.i.i.i:                                    ; preds = %lor.lhs.false.i.i.i, %if.then.i.i
>  %sub.i.i.i = sub i64 %add.i.i.i, %M
>  br label %if.end.i.i
>
>if.end.i.i:                                       ; preds = %if.then.i.i.i, %lor.lhs.false.i.i.i, %while.body.i.i
>  %r.1.i.i = phi i64 [ %r.020.i.i, %while.body.i.i ], [ %sub.i.i.i, %if.then.i.i.i ], [ %add.i.i.i, %lor.lhs.false.i.i.i ]
>  %add.i9.i.i = shl i64 %b.addr.018.i.i, 1
>  %cmp.i10.i.i = icmp ult i64 %add.i9.i.i, %M
>  br i1 %cmp.i10.i.i, label %lor.lhs.false.i12.i.i, label %if.then.i14.i.i
>
>lor.lhs.false.i12.i.i:                            ; preds = %if.end.i.i
>  %2 = uitofp i64 %add.i9.i.i to float
>  %3 = uitofp i64 %b.addr.018.i.i to float
>  %cmp2.i11.i.i = fcmp olt float %2, %3
>  br i1 %cmp2.i11.i.i, label %if.then.i14.i.i, label %MWC_AddMod64.exit16.i.i
>
>if.then.i14.i.i:                                  ; preds = %lor.lhs.false.i12.i.i, %if.end.i.i
>  %sub.i13.i.i = sub i64 %add.i9.i.i, %M
>  br label %MWC_AddMod64.exit16.i.i
>
>MWC_AddMod64.exit16.i.i:                          ; preds = %if.then.i14.i.i, %lor.lhs.false.i12.i.i
>  %v.0.i15.i.i = phi i64 [ %sub.i13.i.i, %if.then.i14.i.i ], [ %add.i9.i.i, %lor.lhs.false.i12.i.i ]
>  %shr.i.i = lshr i64 %a.addr.019.i.i, 1
>  %cmp.i.i = icmp eq i64 %shr.i.i, 0
>  br i1 %cmp.i.i, label %if.end.i, label %while.body.i.i
>
>if.end.i:                                         ; preds = %MWC_AddMod64.exit16.i.i, %if.then.i, %while.body.i
>  %acc.1.i = phi i64 [ %acc.041.i, %while.body.i ], [ 0, %if.then.i ], [ %r.1.i.i, %MWC_AddMod64.exit16.i.i ]
>  %cmp17.i10.i = icmp eq i64 %sqr.040.i, 0
>  br i1 %cmp17.i10.i, label %MWC_MulMod64.exit37.i, label %while.body.i16.i
>
>while.body.i16.i:                                 ; preds = %if.end.i, %MWC_AddMod64.exit16.i35.i
>  %r.020.i11.i = phi i64 [ %r.1.i24.i, %MWC_AddMod64.exit16.i35.i ], [ 0, %if.end.i ]
>  %a.addr.019.i12.i = phi i64 [ %shr.i33.i, %MWC_AddMod64.exit16.i35.i ], [ %sqr.040.i, %if.end.i ]
>  %b.addr.018.i13.i = phi i64 [ %v.0.i15.i32.i, %MWC_AddMod64.exit16.i35.i ], [ %sqr.040.i, %if.end.i ]
>  %and.i14.i = and i64 %a.addr.019.i12.i, 1
>  %tobool.i15.i = icmp eq i64 %and.i14.i, 0
>  br i1 %tobool.i15.i, label %if.end.i27.i, label %if.then.i19.i
>
>if.then.i19.i:                                    ; preds = %while.body.i16.i
>  %add.i.i17.i = add i64 %b.addr.018.i13.i, %r.020.i11.i
>  %cmp.i.i18.i = icmp ult i64 %add.i.i17.i, %M
>  br i1 %cmp.i.i18.i, label %lor.lhs.false.i.i21.i, label %if.then.i.i23.i
>
>lor.lhs.false.i.i21.i:                            ; preds = %if.then.i19.i
>  %4 = uitofp i64 %add.i.i17.i to float
>  %5 = uitofp i64 %r.020.i11.i to float
>  %cmp2.i.i20.i = fcmp olt float %4, %5
>  br i1 %cmp2.i.i20.i, label %if.then.i.i23.i, label %if.end.i27.i
>
>if.then.i.i23.i:                                  ; preds = %lor.lhs.false.i.i21.i, %if.then.i19.i
>  %sub.i.i22.i = sub i64 %add.i.i17.i, %M
>  br label %if.end.i27.i
>
>if.end.i27.i:                                     ; preds = %if.then.i.i23.i, %lor.lhs.false.i.i21.i, %while.body.i16.i
>  %r.1.i24.i = phi i64 [ %r.020.i11.i, %while.body.i16.i ], [ %sub.i.i22.i, %if.then.i.i23.i ], [ %add.i.i17.i, %lor.lhs.false.i.i21.i ]
>  %add.i9.i25.i = shl i64 %b.addr.018.i13.i, 1
>  %cmp.i10.i26.i = icmp ult i64 %add.i9.i25.i, %M
>  br i1 %cmp.i10.i26.i, label %lor.lhs.false.i12.i29.i, label %if.then.i14.i31.i
>
>lor.lhs.false.i12.i29.i:                          ; preds = %if.end.i27.i
>  %6 = uitofp i64 %add.i9.i25.i to float
>  %7 = uitofp i64 %b.addr.018.i13.i to float
>  %cmp2.i11.i28.i = fcmp olt float %6, %7
>  br i1 %cmp2.i11.i28.i, label %if.then.i14.i31.i, label %MWC_AddMod64.exit16.i35.i
>
>if.then.i14.i31.i:                                ; preds = %lor.lhs.false.i12.i29.i, %if.end.i27.i
>  %sub.i13.i30.i = sub i64 %add.i9.i25.i, %M
>  br label %MWC_AddMod64.exit16.i35.i
>
>MWC_AddMod64.exit16.i35.i:                        ; preds = %if.then.i14.i31.i, %lor.lhs.false.i12.i29.i
>  %v.0.i15.i32.i = phi i64 [ %sub.i13.i30.i, %if.then.i14.i31.i ], [ %add.i9.i25.i, %lor.lhs.false.i12.i29.i ]
>  %shr.i33.i = lshr i64 %a.addr.019.i12.i, 1
>  %cmp.i34.i = icmp eq i64 %shr.i33.i, 0
>  br i1 %cmp.i34.i, label %MWC_MulMod64.exit37.i, label %while.body.i16.i
>
>MWC_MulMod64.exit37.i:                            ; preds = %MWC_AddMod64.exit16.i35.i, %if.end.i
>  %r.0.lcssa.i36.i = phi i64 [ 0, %if.end.i ], [ %r.1.i24.i, %MWC_AddMod64.exit16.i35.i ]
>  %shr.i = lshr i64 %e.addr.039.i, 1
>  %cmp.i = icmp eq i64 %shr.i, 0
>  br i1 %cmp.i, label %MWC_PowMod64.exit, label %while.body.i
>
>MWC_PowMod64.exit:                                ; preds = %MWC_MulMod64.exit37.i, %entry
>  %acc.0.lcssa.i = phi i64 [ 1, %entry ], [ %acc.1.i, %MWC_MulMod64.exit37.i ]
>  %8 = extractelement <2 x i32> %curr, i32 0
>  %conv = zext i32 %8 to i64
>  %mul = mul i64 %conv, %A
>  %9 = extractelement <2 x i32> %curr, i32 1
>  %conv1 = zext i32 %9 to i64
>  %add = add i64 %mul, %conv1
>  %cmp17.i = icmp eq i64 %add, 0
>  br i1 %cmp17.i, label %MWC_MulMod64.exit, label %while.body.i17
>
>while.body.i17:                                   ; preds = %MWC_PowMod64.exit, %MWC_AddMod64.exit16.i
>  %r.020.i = phi i64 [ %r.1.i, %MWC_AddMod64.exit16.i ], [ 0, %MWC_PowMod64.exit ]
>  %a.addr.019.i = phi i64 [ %shr.i22, %MWC_AddMod64.exit16.i ], [ %add, %MWC_PowMod64.exit ]
>  %b.addr.018.i = phi i64 [ %v.0.i15.i, %MWC_AddMod64.exit16.i ], [ %acc.0.lcssa.i, %MWC_PowMod64.exit ]
>  %and.i15 = and i64 %a.addr.019.i, 1
>  %tobool.i16 = icmp eq i64 %and.i15, 0
>  br i1 %tobool.i16, label %if.end.i21, label %if.then.i19
>
>if.then.i19:                                      ; preds = %while.body.i17
>  %add.i.i = add i64 %b.addr.018.i, %r.020.i
>  %cmp.i.i18 = icmp ult i64 %add.i.i, %M
>  br i1 %cmp.i.i18, label %lor.lhs.false.i.i, label %if.then.i.i20
>
>lor.lhs.false.i.i:                                ; preds = %if.then.i19
>  %10 = uitofp i64 %add.i.i to float
>  %11 = uitofp i64 %r.020.i to float
>  %cmp2.i.i = fcmp olt float %10, %11
>  br i1 %cmp2.i.i, label %if.then.i.i20, label %if.end.i21
>
>if.then.i.i20:                                    ; preds = %lor.lhs.false.i.i, %if.then.i19
>  %sub.i.i = sub i64 %add.i.i, %M
>  br label %if.end.i21
>
>if.end.i21:                                       ; preds = %if.then.i.i20, %lor.lhs.false.i.i, %while.body.i17
>  %r.1.i = phi i64 [ %r.020.i, %while.body.i17 ], [ %sub.i.i, %if.then.i.i20 ], [ %add.i.i, %lor.lhs.false.i.i ]
>  %add.i9.i = shl i64 %b.addr.018.i, 1
>  %cmp.i10.i = icmp ult i64 %add.i9.i, %M
>  br i1 %cmp.i10.i, label %lor.lhs.false.i12.i, label %if.then.i14.i
>
>lor.lhs.false.i12.i:                              ; preds = %if.end.i21
>  %12 = uitofp i64 %add.i9.i to float
>  %13 = uitofp i64 %b.addr.018.i to float
>  %cmp2.i11.i = fcmp olt float %12, %13
>  br i1 %cmp2.i11.i, label %if.then.i14.i, label %MWC_AddMod64.exit16.i
>
>if.then.i14.i:                                    ; preds = %lor.lhs.false.i12.i, %if.end.i21
>  %sub.i13.i = sub i64 %add.i9.i, %M
>  br label %MWC_AddMod64.exit16.i
>
>MWC_AddMod64.exit16.i:                            ; preds = %if.then.i14.i, %lor.lhs.false.i12.i
>  %v.0.i15.i = phi i64 [ %sub.i13.i, %if.then.i14.i ], [ %add.i9.i, %lor.lhs.false.i12.i ]
>  %shr.i22 = lshr i64 %a.addr.019.i, 1
>  %cmp.i23 = icmp eq i64 %shr.i22, 0
>  br i1 %cmp.i23, label %MWC_MulMod64.exit, label %while.body.i17
>
>MWC_MulMod64.exit:                                ; preds = %MWC_AddMod64.exit16.i, %MWC_PowMod64.exit
>  %r.0.lcssa.i = phi i64 [ 0, %MWC_PowMod64.exit ], [ %r.1.i, %MWC_AddMod64.exit16.i ]
>  %div = udiv i64 %r.0.lcssa.i, %A
>  %conv3 = trunc i64 %div to i32
>  %vecinit = insertelement <2 x i32> undef, i32 %conv3, i32 0
>  %14 = mul i64 %div, %A
>  %15 = sub i64 %r.0.lcssa.i, %14
>  %conv4 = trunc i64 %15 to i32
>  %vecinit5 = insertelement <2 x i32> %vecinit, i32 %conv4, i32 1
>  ret <2 x i32> %vecinit5
>}
>
>; Function Attrs: convergent nounwind readonly
>define <2 x i32> @MWC_SeedImpl_Mod64(i64 %A, i64 %M, i32 %vecSize, i32 %vecOffset, i64 %streamBase, i64 %streamGap) local_unnamed_addr #2 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %mul = mul i32 %8, %vecSize
>  %add = add i32 %mul, %vecOffset
>  %conv = zext i32 %add to i64
>  %mul1 = mul i64 %conv, %streamGap
>  %add2 = add i64 %mul1, %streamBase
>  %cmp38.i = icmp eq i64 %add2, 0
>  br i1 %cmp38.i, label %while.body.i.preheader, label %while.body.i17
>
>while.body.i17:                                   ; preds = %entry, %MWC_MulMod64.exit37.i
>  %acc.041.i = phi i64 [ %acc.1.i, %MWC_MulMod64.exit37.i ], [ 1, %entry ]
>  %sqr.040.i = phi i64 [ %r.0.lcssa.i36.i, %MWC_MulMod64.exit37.i ], [ %A, %entry ]
>  %e.addr.039.i = phi i64 [ %shr.i22, %MWC_MulMod64.exit37.i ], [ %add2, %entry ]
>  %and.i15 = and i64 %e.addr.039.i, 1
>  %tobool.i16 = icmp eq i64 %and.i15, 0
>  br i1 %tobool.i16, label %if.end.i21, label %if.then.i18
>
>if.then.i18:                                      ; preds = %while.body.i17
>  %cmp17.i.i = icmp eq i64 %acc.041.i, 0
>  br i1 %cmp17.i.i, label %if.end.i21, label %while.body.i.i
>
>while.body.i.i:                                   ; preds = %if.then.i18, %MWC_AddMod64.exit16.i.i
>  %r.020.i.i = phi i64 [ %r.1.i.i, %MWC_AddMod64.exit16.i.i ], [ 0, %if.then.i18 ]
>  %a.addr.019.i.i = phi i64 [ %shr.i.i, %MWC_AddMod64.exit16.i.i ], [ %acc.041.i, %if.then.i18 ]
>  %b.addr.018.i.i = phi i64 [ %v.0.i15.i.i, %MWC_AddMod64.exit16.i.i ], [ %sqr.040.i, %if.then.i18 ]
>  %and.i.i = and i64 %a.addr.019.i.i, 1
>  %tobool.i.i = icmp eq i64 %and.i.i, 0
>  br i1 %tobool.i.i, label %if.end.i.i, label %if.then.i.i19
>
>if.then.i.i19:                                    ; preds = %while.body.i.i
>  %add.i.i.i = add i64 %b.addr.018.i.i, %r.020.i.i
>  %cmp.i.i.i = icmp ult i64 %add.i.i.i, %M
>  br i1 %cmp.i.i.i, label %lor.lhs.false.i.i.i, label %if.then.i.i.i
>
>lor.lhs.false.i.i.i:                              ; preds = %if.then.i.i19
>  %9 = uitofp i64 %add.i.i.i to float
>  %10 = uitofp i64 %r.020.i.i to float
>  %cmp2.i.i.i = fcmp olt float %9, %10
>  br i1 %cmp2.i.i.i, label %if.then.i.i.i, label %if.end.i.i
>
>if.then.i.i.i:                                    ; preds = %lor.lhs.false.i.i.i, %if.then.i.i19
>  %sub.i.i.i = sub i64 %add.i.i.i, %M
>  br label %if.end.i.i
>
>if.end.i.i:                                       ; preds = %if.then.i.i.i, %lor.lhs.false.i.i.i, %while.body.i.i
>  %r.1.i.i = phi i64 [ %r.020.i.i, %while.body.i.i ], [ %sub.i.i.i, %if.then.i.i.i ], [ %add.i.i.i, %lor.lhs.false.i.i.i ]
>  %add.i9.i.i = shl i64 %b.addr.018.i.i, 1
>  %cmp.i10.i.i = icmp ult i64 %add.i9.i.i, %M
>  br i1 %cmp.i10.i.i, label %lor.lhs.false.i12.i.i, label %if.then.i14.i.i
>
>lor.lhs.false.i12.i.i:                            ; preds = %if.end.i.i
>  %11 = uitofp i64 %add.i9.i.i to float
>  %12 = uitofp i64 %b.addr.018.i.i to float
>  %cmp2.i11.i.i = fcmp olt float %11, %12
>  br i1 %cmp2.i11.i.i, label %if.then.i14.i.i, label %MWC_AddMod64.exit16.i.i
>
>if.then.i14.i.i:                                  ; preds = %lor.lhs.false.i12.i.i, %if.end.i.i
>  %sub.i13.i.i = sub i64 %add.i9.i.i, %M
>  br label %MWC_AddMod64.exit16.i.i
>
>MWC_AddMod64.exit16.i.i:                          ; preds = %if.then.i14.i.i, %lor.lhs.false.i12.i.i
>  %v.0.i15.i.i = phi i64 [ %sub.i13.i.i, %if.then.i14.i.i ], [ %add.i9.i.i, %lor.lhs.false.i12.i.i ]
>  %shr.i.i = lshr i64 %a.addr.019.i.i, 1
>  %cmp.i.i20 = icmp eq i64 %shr.i.i, 0
>  br i1 %cmp.i.i20, label %if.end.i21, label %while.body.i.i
>
>if.end.i21:                                       ; preds = %MWC_AddMod64.exit16.i.i, %if.then.i18, %while.body.i17
>  %acc.1.i = phi i64 [ %acc.041.i, %while.body.i17 ], [ 0, %if.then.i18 ], [ %r.1.i.i, %MWC_AddMod64.exit16.i.i ]
>  %cmp17.i10.i = icmp eq i64 %sqr.040.i, 0
>  br i1 %cmp17.i10.i, label %MWC_MulMod64.exit37.i, label %while.body.i16.i
>
>while.body.i16.i:                                 ; preds = %if.end.i21, %MWC_AddMod64.exit16.i35.i
>  %r.020.i11.i = phi i64 [ %r.1.i24.i, %MWC_AddMod64.exit16.i35.i ], [ 0, %if.end.i21 ]
>  %a.addr.019.i12.i = phi i64 [ %shr.i33.i, %MWC_AddMod64.exit16.i35.i ], [ %sqr.040.i, %if.end.i21 ]
>  %b.addr.018.i13.i = phi i64 [ %v.0.i15.i32.i, %MWC_AddMod64.exit16.i35.i ], [ %sqr.040.i, %if.end.i21 ]
>  %and.i14.i = and i64 %a.addr.019.i12.i, 1
>  %tobool.i15.i = icmp eq i64 %and.i14.i, 0
>  br i1 %tobool.i15.i, label %if.end.i27.i, label %if.then.i19.i
>
>if.then.i19.i:                                    ; preds = %while.body.i16.i
>  %add.i.i17.i = add i64 %b.addr.018.i13.i, %r.020.i11.i
>  %cmp.i.i18.i = icmp ult i64 %add.i.i17.i, %M
>  br i1 %cmp.i.i18.i, label %lor.lhs.false.i.i21.i, label %if.then.i.i23.i
>
>lor.lhs.false.i.i21.i:                            ; preds = %if.then.i19.i
>  %13 = uitofp i64 %add.i.i17.i to float
>  %14 = uitofp i64 %r.020.i11.i to float
>  %cmp2.i.i20.i = fcmp olt float %13, %14
>  br i1 %cmp2.i.i20.i, label %if.then.i.i23.i, label %if.end.i27.i
>
>if.then.i.i23.i:                                  ; preds = %lor.lhs.false.i.i21.i, %if.then.i19.i
>  %sub.i.i22.i = sub i64 %add.i.i17.i, %M
>  br label %if.end.i27.i
>
>if.end.i27.i:                                     ; preds = %if.then.i.i23.i, %lor.lhs.false.i.i21.i, %while.body.i16.i
>  %r.1.i24.i = phi i64 [ %r.020.i11.i, %while.body.i16.i ], [ %sub.i.i22.i, %if.then.i.i23.i ], [ %add.i.i17.i, %lor.lhs.false.i.i21.i ]
>  %add.i9.i25.i = shl i64 %b.addr.018.i13.i, 1
>  %cmp.i10.i26.i = icmp ult i64 %add.i9.i25.i, %M
>  br i1 %cmp.i10.i26.i, label %lor.lhs.false.i12.i29.i, label %if.then.i14.i31.i
>
>lor.lhs.false.i12.i29.i:                          ; preds = %if.end.i27.i
>  %15 = uitofp i64 %add.i9.i25.i to float
>  %16 = uitofp i64 %b.addr.018.i13.i to float
>  %cmp2.i11.i28.i = fcmp olt float %15, %16
>  br i1 %cmp2.i11.i28.i, label %if.then.i14.i31.i, label %MWC_AddMod64.exit16.i35.i
>
>if.then.i14.i31.i:                                ; preds = %lor.lhs.false.i12.i29.i, %if.end.i27.i
>  %sub.i13.i30.i = sub i64 %add.i9.i25.i, %M
>  br label %MWC_AddMod64.exit16.i35.i
>
>MWC_AddMod64.exit16.i35.i:                        ; preds = %if.then.i14.i31.i, %lor.lhs.false.i12.i29.i
>  %v.0.i15.i32.i = phi i64 [ %sub.i13.i30.i, %if.then.i14.i31.i ], [ %add.i9.i25.i, %lor.lhs.false.i12.i29.i ]
>  %shr.i33.i = lshr i64 %a.addr.019.i12.i, 1
>  %cmp.i34.i = icmp eq i64 %shr.i33.i, 0
>  br i1 %cmp.i34.i, label %MWC_MulMod64.exit37.i, label %while.body.i16.i
>
>MWC_MulMod64.exit37.i:                            ; preds = %MWC_AddMod64.exit16.i35.i, %if.end.i21
>  %r.0.lcssa.i36.i = phi i64 [ 0, %if.end.i21 ], [ %r.1.i24.i, %MWC_AddMod64.exit16.i35.i ]
>  %shr.i22 = lshr i64 %e.addr.039.i, 1
>  %cmp.i23 = icmp eq i64 %shr.i22, 0
>  br i1 %cmp.i23, label %while.body.i.preheader, label %while.body.i17
>
>while.body.i.preheader:                           ; preds = %MWC_MulMod64.exit37.i, %entry
>  %b.addr.018.i.ph = phi i64 [ 1, %entry ], [ %acc.1.i, %MWC_MulMod64.exit37.i ]
>  br label %while.body.i
>
>while.body.i:                                     ; preds = %MWC_AddMod64.exit16.i.1, %while.body.i.preheader
>  %r.020.i = phi i64 [ 0, %while.body.i.preheader ], [ %r.1.i.1, %MWC_AddMod64.exit16.i.1 ]
>  %a.addr.019.i = phi i64 [ 4077358422479273989, %while.body.i.preheader ], [ %shr.i, %MWC_AddMod64.exit16.i.1 ]
>  %b.addr.018.i = phi i64 [ %b.addr.018.i.ph, %while.body.i.preheader ], [ %v.0.i15.i.1, %MWC_AddMod64.exit16.i.1 ]
>  %and.i = and i64 %a.addr.019.i, 1
>  %tobool.i = icmp eq i64 %and.i, 0
>  br i1 %tobool.i, label %if.end.i, label %if.then.i
>
>if.then.i:                                        ; preds = %while.body.i
>  %add.i.i = add i64 %b.addr.018.i, %r.020.i
>  %cmp.i.i = icmp ult i64 %add.i.i, %M
>  br i1 %cmp.i.i, label %lor.lhs.false.i.i, label %if.then.i.i
>
>lor.lhs.false.i.i:                                ; preds = %if.then.i
>  %17 = uitofp i64 %add.i.i to float
>  %18 = uitofp i64 %r.020.i to float
>  %cmp2.i.i = fcmp olt float %17, %18
>  br i1 %cmp2.i.i, label %if.then.i.i, label %if.end.i
>
>if.then.i.i:                                      ; preds = %lor.lhs.false.i.i, %if.then.i
>  %sub.i.i = sub i64 %add.i.i, %M
>  br label %if.end.i
>
>if.end.i:                                         ; preds = %if.then.i.i, %lor.lhs.false.i.i, %while.body.i
>  %r.1.i = phi i64 [ %r.020.i, %while.body.i ], [ %sub.i.i, %if.then.i.i ], [ %add.i.i, %lor.lhs.false.i.i ]
>  %add.i9.i = shl i64 %b.addr.018.i, 1
>  %cmp.i10.i = icmp ult i64 %add.i9.i, %M
>  br i1 %cmp.i10.i, label %lor.lhs.false.i12.i, label %if.then.i14.i
>
>lor.lhs.false.i12.i:                              ; preds = %if.end.i
>  %19 = uitofp i64 %add.i9.i to float
>  %20 = uitofp i64 %b.addr.018.i to float
>  %cmp2.i11.i = fcmp olt float %19, %20
>  br i1 %cmp2.i11.i, label %if.then.i14.i, label %MWC_AddMod64.exit16.i
>
>if.then.i14.i:                                    ; preds = %lor.lhs.false.i12.i, %if.end.i
>  %sub.i13.i = sub i64 %add.i9.i, %M
>  br label %MWC_AddMod64.exit16.i
>
>MWC_AddMod64.exit16.i:                            ; preds = %if.then.i14.i, %lor.lhs.false.i12.i
>  %v.0.i15.i = phi i64 [ %sub.i13.i, %if.then.i14.i ], [ %add.i9.i, %lor.lhs.false.i12.i ]
>  %shr.i = lshr i64 %a.addr.019.i, 2
>  %and.i.1 = and i64 %a.addr.019.i, 2
>  %tobool.i.1 = icmp eq i64 %and.i.1, 0
>  br i1 %tobool.i.1, label %if.end.i.1, label %if.then.i.1
>
>MWC_MulMod64.exit:                                ; preds = %MWC_AddMod64.exit16.i.1
>  %div = udiv i64 %r.1.i.1, %A
>  %conv5 = trunc i64 %div to i32
>  %vecinit = insertelement <2 x i32> undef, i32 %conv5, i32 0
>  %21 = mul i64 %div, %A
>  %22 = sub i64 %r.1.i.1, %21
>  %conv6 = trunc i64 %22 to i32
>  %vecinit7 = insertelement <2 x i32> %vecinit, i32 %conv6, i32 1
>  ret <2 x i32> %vecinit7
>
>if.then.i.1:                                      ; preds = %MWC_AddMod64.exit16.i
>  %add.i.i.1 = add i64 %v.0.i15.i, %r.1.i
>  %cmp.i.i.1 = icmp ult i64 %add.i.i.1, %M
>  br i1 %cmp.i.i.1, label %lor.lhs.false.i.i.1, label %if.then.i.i.1
>
>lor.lhs.false.i.i.1:                              ; preds = %if.then.i.1
>  %23 = uitofp i64 %add.i.i.1 to float
>  %24 = uitofp i64 %r.1.i to float
>  %cmp2.i.i.1 = fcmp olt float %23, %24
>  br i1 %cmp2.i.i.1, label %if.then.i.i.1, label %if.end.i.1
>
>if.then.i.i.1:                                    ; preds = %lor.lhs.false.i.i.1, %if.then.i.1
>  %sub.i.i.1 = sub i64 %add.i.i.1, %M
>  br label %if.end.i.1
>
>if.end.i.1:                                       ; preds = %if.then.i.i.1, %lor.lhs.false.i.i.1, %MWC_AddMod64.exit16.i
>  %r.1.i.1 = phi i64 [ %r.1.i, %MWC_AddMod64.exit16.i ], [ %sub.i.i.1, %if.then.i.i.1 ], [ %add.i.i.1, %lor.lhs.false.i.i.1 ]
>  %add.i9.i.1 = shl i64 %v.0.i15.i, 1
>  %cmp.i10.i.1 = icmp ult i64 %add.i9.i.1, %M
>  br i1 %cmp.i10.i.1, label %lor.lhs.false.i12.i.1, label %if.then.i14.i.1
>
>lor.lhs.false.i12.i.1:                            ; preds = %if.end.i.1
>  %25 = uitofp i64 %add.i9.i.1 to float
>  %26 = uitofp i64 %v.0.i15.i to float
>  %cmp2.i11.i.1 = fcmp olt float %25, %26
>  br i1 %cmp2.i11.i.1, label %if.then.i14.i.1, label %MWC_AddMod64.exit16.i.1
>
>if.then.i14.i.1:                                  ; preds = %lor.lhs.false.i12.i.1, %if.end.i.1
>  %sub.i13.i.1 = sub i64 %add.i9.i.1, %M
>  br label %MWC_AddMod64.exit16.i.1
>
>MWC_AddMod64.exit16.i.1:                          ; preds = %if.then.i14.i.1, %lor.lhs.false.i12.i.1
>  %v.0.i15.i.1 = phi i64 [ %sub.i13.i.1, %if.then.i14.i.1 ], [ %add.i9.i.1, %lor.lhs.false.i12.i.1 ]
>  %cmp.i.1 = icmp eq i64 %shr.i, 0
>  br i1 %cmp.i.1, label %MWC_MulMod64.exit, label %while.body.i
>}
>
>; Function Attrs: norecurse nounwind
>define void @MWC64X_Step(%struct.mwc64x_state_t addrspace(5)* nocapture %s) local_unnamed_addr #3 {
>entry:
>  %x = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %s, i32 0, i32 0
>  %0 = load i32, i32 addrspace(5)* %x, align 4, !tbaa !8
>  %c = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %s, i32 0, i32 1
>  %1 = load i32, i32 addrspace(5)* %c, align 4, !tbaa !10
>  %mul = mul i32 %0, -83941
>  %add = add i32 %mul, %1
>  %cmp = icmp ult i32 %add, %1
>  %conv = zext i1 %cmp to i32
>  %2 = zext i32 %0 to i64
>  %3 = mul nuw i64 %2, 4294883355
>  %4 = lshr i64 %3, 32
>  %5 = trunc i64 %4 to i32
>  %add1 = add i32 %5, %conv
>  store i32 %add, i32 addrspace(5)* %x, align 4, !tbaa !8
>  store i32 %add1, i32 addrspace(5)* %c, align 4, !tbaa !10
>  ret void
>}
>
>; Function Attrs: norecurse nounwind
>define void @MWC64X_Skip(%struct.mwc64x_state_t addrspace(5)* nocapture %s, i64 %distance) local_unnamed_addr #3 {
>entry:
>  %x = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %s, i32 0, i32 0
>  %0 = load i32, i32 addrspace(5)* %x, align 4, !tbaa !8
>  %c = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %s, i32 0, i32 1
>  %1 = load i32, i32 addrspace(5)* %c, align 4, !tbaa !10
>  %cmp38.i.i = icmp eq i64 %distance, 0
>  br i1 %cmp38.i.i, label %MWC_PowMod64.exit.i, label %while.body.i.i
>
>while.body.i.i:                                   ; preds = %entry, %MWC_MulMod64.exit37.i.i
>  %acc.041.i.i = phi i64 [ %acc.1.i.i, %MWC_MulMod64.exit37.i.i ], [ 1, %entry ]
>  %sqr.040.i.i = phi i64 [ %r.0.lcssa.i36.i.i, %MWC_MulMod64.exit37.i.i ], [ 4294883355, %entry ]
>  %e.addr.039.i.i = phi i64 [ %shr.i.i, %MWC_MulMod64.exit37.i.i ], [ %distance, %entry ]
>  %and.i.i = and i64 %e.addr.039.i.i, 1
>  %tobool.i.i = icmp eq i64 %and.i.i, 0
>  br i1 %tobool.i.i, label %if.end.i.i, label %if.then.i.i
>
>if.then.i.i:                                      ; preds = %while.body.i.i
>  %cmp17.i.i.i = icmp eq i64 %acc.041.i.i, 0
>  br i1 %cmp17.i.i.i, label %if.end.i.i, label %while.body.i.i.i
>
>while.body.i.i.i:                                 ; preds = %if.then.i.i, %MWC_AddMod64.exit16.i.i.i
>  %r.020.i.i.i = phi i64 [ %r.1.i.i.i, %MWC_AddMod64.exit16.i.i.i ], [ 0, %if.then.i.i ]
>  %a.addr.019.i.i.i = phi i64 [ %shr.i.i.i, %MWC_AddMod64.exit16.i.i.i ], [ %acc.041.i.i, %if.then.i.i ]
>  %b.addr.018.i.i.i = phi i64 [ %v.0.i15.i.i.i, %MWC_AddMod64.exit16.i.i.i ], [ %sqr.040.i.i, %if.then.i.i ]
>  %and.i.i.i = and i64 %a.addr.019.i.i.i, 1
>  %tobool.i.i.i = icmp eq i64 %and.i.i.i, 0
>  br i1 %tobool.i.i.i, label %if.end.i.i.i, label %if.then.i.i.i
>
>if.then.i.i.i:                                    ; preds = %while.body.i.i.i
>  %add.i.i.i.i = add i64 %b.addr.018.i.i.i, %r.020.i.i.i
>  %cmp.i.i.i.i = icmp ult i64 %add.i.i.i.i, -360523849793537
>  br i1 %cmp.i.i.i.i, label %lor.lhs.false.i.i.i.i, label %if.then.i.i.i.i
>
>lor.lhs.false.i.i.i.i:                            ; preds = %if.then.i.i.i
>  %2 = uitofp i64 %add.i.i.i.i to float
>  %3 = uitofp i64 %r.020.i.i.i to float
>  %cmp2.i.i.i.i = fcmp olt float %2, %3
>  br i1 %cmp2.i.i.i.i, label %if.then.i.i.i.i, label %if.end.i.i.i
>
>if.then.i.i.i.i:                                  ; preds = %lor.lhs.false.i.i.i.i, %if.then.i.i.i
>  %sub.i.i.i.i = add i64 %add.i.i.i.i, 360523849793537
>  br label %if.end.i.i.i
>
>if.end.i.i.i:                                     ; preds = %if.then.i.i.i.i, %lor.lhs.false.i.i.i.i, %while.body.i.i.i
>  %r.1.i.i.i = phi i64 [ %r.020.i.i.i, %while.body.i.i.i ], [ %sub.i.i.i.i, %if.then.i.i.i.i ], [ %add.i.i.i.i, %lor.lhs.false.i.i.i.i ]
>  %add.i9.i.i.i = shl i64 %b.addr.018.i.i.i, 1
>  %cmp.i10.i.i.i = icmp ult i64 %add.i9.i.i.i, -360523849793537
>  br i1 %cmp.i10.i.i.i, label %lor.lhs.false.i12.i.i.i, label %if.then.i14.i.i.i
>
>lor.lhs.false.i12.i.i.i:                          ; preds = %if.end.i.i.i
>  %4 = uitofp i64 %add.i9.i.i.i to float
>  %5 = uitofp i64 %b.addr.018.i.i.i to float
>  %cmp2.i11.i.i.i = fcmp olt float %4, %5
>  br i1 %cmp2.i11.i.i.i, label %if.then.i14.i.i.i, label %MWC_AddMod64.exit16.i.i.i
>
>if.then.i14.i.i.i:                                ; preds = %lor.lhs.false.i12.i.i.i, %if.end.i.i.i
>  %sub.i13.i.i.i = add i64 %add.i9.i.i.i, 360523849793537
>  br label %MWC_AddMod64.exit16.i.i.i
>
>MWC_AddMod64.exit16.i.i.i:                        ; preds = %if.then.i14.i.i.i, %lor.lhs.false.i12.i.i.i
>  %v.0.i15.i.i.i = phi i64 [ %sub.i13.i.i.i, %if.then.i14.i.i.i ], [ %add.i9.i.i.i, %lor.lhs.false.i12.i.i.i ]
>  %shr.i.i.i = lshr i64 %a.addr.019.i.i.i, 1
>  %cmp.i.i.i = icmp eq i64 %shr.i.i.i, 0
>  br i1 %cmp.i.i.i, label %if.end.i.i, label %while.body.i.i.i
>
>if.end.i.i:                                       ; preds = %MWC_AddMod64.exit16.i.i.i, %if.then.i.i, %while.body.i.i
>  %acc.1.i.i = phi i64 [ %acc.041.i.i, %while.body.i.i ], [ 0, %if.then.i.i ], [ %r.1.i.i.i, %MWC_AddMod64.exit16.i.i.i ]
>  %cmp17.i10.i.i = icmp eq i64 %sqr.040.i.i, 0
>  br i1 %cmp17.i10.i.i, label %MWC_MulMod64.exit37.i.i, label %while.body.i16.i.i
>
>while.body.i16.i.i:                               ; preds = %if.end.i.i, %MWC_AddMod64.exit16.i35.i.i
>  %r.020.i11.i.i = phi i64 [ %r.1.i24.i.i, %MWC_AddMod64.exit16.i35.i.i ], [ 0, %if.end.i.i ]
>  %a.addr.019.i12.i.i = phi i64 [ %shr.i33.i.i, %MWC_AddMod64.exit16.i35.i.i ], [ %sqr.040.i.i, %if.end.i.i ]
>  %b.addr.018.i13.i.i = phi i64 [ %v.0.i15.i32.i.i, %MWC_AddMod64.exit16.i35.i.i ], [ %sqr.040.i.i, %if.end.i.i ]
>  %and.i14.i.i = and i64 %a.addr.019.i12.i.i, 1
>  %tobool.i15.i.i = icmp eq i64 %and.i14.i.i, 0
>  br i1 %tobool.i15.i.i, label %if.end.i27.i.i, label %if.then.i19.i.i
>
>if.then.i19.i.i:                                  ; preds = %while.body.i16.i.i
>  %add.i.i17.i.i = add i64 %b.addr.018.i13.i.i, %r.020.i11.i.i
>  %cmp.i.i18.i.i = icmp ult i64 %add.i.i17.i.i, -360523849793537
>  br i1 %cmp.i.i18.i.i, label %lor.lhs.false.i.i21.i.i, label %if.then.i.i23.i.i
>
>lor.lhs.false.i.i21.i.i:                          ; preds = %if.then.i19.i.i
>  %6 = uitofp i64 %add.i.i17.i.i to float
>  %7 = uitofp i64 %r.020.i11.i.i to float
>  %cmp2.i.i20.i.i = fcmp olt float %6, %7
>  br i1 %cmp2.i.i20.i.i, label %if.then.i.i23.i.i, label %if.end.i27.i.i
>
>if.then.i.i23.i.i:                                ; preds = %lor.lhs.false.i.i21.i.i, %if.then.i19.i.i
>  %sub.i.i22.i.i = add i64 %add.i.i17.i.i, 360523849793537
>  br label %if.end.i27.i.i
>
>if.end.i27.i.i:                                   ; preds = %if.then.i.i23.i.i, %lor.lhs.false.i.i21.i.i, %while.body.i16.i.i
>  %r.1.i24.i.i = phi i64 [ %r.020.i11.i.i, %while.body.i16.i.i ], [ %sub.i.i22.i.i, %if.then.i.i23.i.i ], [ %add.i.i17.i.i, %lor.lhs.false.i.i21.i.i ]
>  %add.i9.i25.i.i = shl i64 %b.addr.018.i13.i.i, 1
>  %cmp.i10.i26.i.i = icmp ult i64 %add.i9.i25.i.i, -360523849793537
>  br i1 %cmp.i10.i26.i.i, label %lor.lhs.false.i12.i29.i.i, label %if.then.i14.i31.i.i
>
>lor.lhs.false.i12.i29.i.i:                        ; preds = %if.end.i27.i.i
>  %8 = uitofp i64 %add.i9.i25.i.i to float
>  %9 = uitofp i64 %b.addr.018.i13.i.i to float
>  %cmp2.i11.i28.i.i = fcmp olt float %8, %9
>  br i1 %cmp2.i11.i28.i.i, label %if.then.i14.i31.i.i, label %MWC_AddMod64.exit16.i35.i.i
>
>if.then.i14.i31.i.i:                              ; preds = %lor.lhs.false.i12.i29.i.i, %if.end.i27.i.i
>  %sub.i13.i30.i.i = add i64 %add.i9.i25.i.i, 360523849793537
>  br label %MWC_AddMod64.exit16.i35.i.i
>
>MWC_AddMod64.exit16.i35.i.i:                      ; preds = %if.then.i14.i31.i.i, %lor.lhs.false.i12.i29.i.i
>  %v.0.i15.i32.i.i = phi i64 [ %sub.i13.i30.i.i, %if.then.i14.i31.i.i ], [ %add.i9.i25.i.i, %lor.lhs.false.i12.i29.i.i ]
>  %shr.i33.i.i = lshr i64 %a.addr.019.i12.i.i, 1
>  %cmp.i34.i.i = icmp eq i64 %shr.i33.i.i, 0
>  br i1 %cmp.i34.i.i, label %MWC_MulMod64.exit37.i.i, label %while.body.i16.i.i
>
>MWC_MulMod64.exit37.i.i:                          ; preds = %MWC_AddMod64.exit16.i35.i.i, %if.end.i.i
>  %r.0.lcssa.i36.i.i = phi i64 [ 0, %if.end.i.i ], [ %r.1.i24.i.i, %MWC_AddMod64.exit16.i35.i.i ]
>  %shr.i.i = lshr i64 %e.addr.039.i.i, 1
>  %cmp.i.i = icmp eq i64 %shr.i.i, 0
>  br i1 %cmp.i.i, label %MWC_PowMod64.exit.i, label %while.body.i.i
>
>MWC_PowMod64.exit.i:                              ; preds = %MWC_MulMod64.exit37.i.i, %entry
>  %acc.0.lcssa.i.i = phi i64 [ 1, %entry ], [ %acc.1.i.i, %MWC_MulMod64.exit37.i.i ]
>  %conv.i = zext i32 %0 to i64
>  %mul.i = mul nuw i64 %conv.i, 4294883355
>  %conv1.i = zext i32 %1 to i64
>  %add.i = add i64 %mul.i, %conv1.i
>  %cmp17.i.i = icmp eq i64 %add.i, 0
>  br i1 %cmp17.i.i, label %MWC_SkipImpl_Mod64.exit, label %while.body.i17.i
>
>while.body.i17.i:                                 ; preds = %MWC_PowMod64.exit.i, %MWC_AddMod64.exit16.i.i
>  %r.020.i.i = phi i64 [ %r.1.i.i, %MWC_AddMod64.exit16.i.i ], [ 0, %MWC_PowMod64.exit.i ]
>  %a.addr.019.i.i = phi i64 [ %shr.i22.i, %MWC_AddMod64.exit16.i.i ], [ %add.i, %MWC_PowMod64.exit.i ]
>  %b.addr.018.i.i = phi i64 [ %v.0.i15.i.i, %MWC_AddMod64.exit16.i.i ], [ %acc.0.lcssa.i.i, %MWC_PowMod64.exit.i ]
>  %and.i15.i = and i64 %a.addr.019.i.i, 1
>  %tobool.i16.i = icmp eq i64 %and.i15.i, 0
>  br i1 %tobool.i16.i, label %if.end.i21.i, label %if.then.i19.i
>
>if.then.i19.i:                                    ; preds = %while.body.i17.i
>  %add.i.i.i = add i64 %b.addr.018.i.i, %r.020.i.i
>  %cmp.i.i18.i = icmp ult i64 %add.i.i.i, -360523849793537
>  br i1 %cmp.i.i18.i, label %lor.lhs.false.i.i.i, label %if.then.i.i20.i
>
>lor.lhs.false.i.i.i:                              ; preds = %if.then.i19.i
>  %10 = uitofp i64 %add.i.i.i to float
>  %11 = uitofp i64 %r.020.i.i to float
>  %cmp2.i.i.i = fcmp olt float %10, %11
>  br i1 %cmp2.i.i.i, label %if.then.i.i20.i, label %if.end.i21.i
>
>if.then.i.i20.i:                                  ; preds = %lor.lhs.false.i.i.i, %if.then.i19.i
>  %sub.i.i.i = add i64 %add.i.i.i, 360523849793537
>  br label %if.end.i21.i
>
>if.end.i21.i:                                     ; preds = %if.then.i.i20.i, %lor.lhs.false.i.i.i, %while.body.i17.i
>  %r.1.i.i = phi i64 [ %r.020.i.i, %while.body.i17.i ], [ %sub.i.i.i, %if.then.i.i20.i ], [ %add.i.i.i, %lor.lhs.false.i.i.i ]
>  %add.i9.i.i = shl i64 %b.addr.018.i.i, 1
>  %cmp.i10.i.i = icmp ult i64 %add.i9.i.i, -360523849793537
>  br i1 %cmp.i10.i.i, label %lor.lhs.false.i12.i.i, label %if.then.i14.i.i
>
>lor.lhs.false.i12.i.i:                            ; preds = %if.end.i21.i
>  %12 = uitofp i64 %add.i9.i.i to float
>  %13 = uitofp i64 %b.addr.018.i.i to float
>  %cmp2.i11.i.i = fcmp olt float %12, %13
>  br i1 %cmp2.i11.i.i, label %if.then.i14.i.i, label %MWC_AddMod64.exit16.i.i
>
>if.then.i14.i.i:                                  ; preds = %lor.lhs.false.i12.i.i, %if.end.i21.i
>  %sub.i13.i.i = add i64 %add.i9.i.i, 360523849793537
>  br label %MWC_AddMod64.exit16.i.i
>
>MWC_AddMod64.exit16.i.i:                          ; preds = %if.then.i14.i.i, %lor.lhs.false.i12.i.i
>  %v.0.i15.i.i = phi i64 [ %sub.i13.i.i, %if.then.i14.i.i ], [ %add.i9.i.i, %lor.lhs.false.i12.i.i ]
>  %shr.i22.i = lshr i64 %a.addr.019.i.i, 1
>  %cmp.i23.i = icmp eq i64 %shr.i22.i, 0
>  br i1 %cmp.i23.i, label %MWC_SkipImpl_Mod64.exit, label %while.body.i17.i
>
>MWC_SkipImpl_Mod64.exit:                          ; preds = %MWC_AddMod64.exit16.i.i, %MWC_PowMod64.exit.i
>  %r.0.lcssa.i.i = phi i64 [ 0, %MWC_PowMod64.exit.i ], [ %r.1.i.i, %MWC_AddMod64.exit16.i.i ]
>  %div.i = udiv i64 %r.0.lcssa.i.i, 4294883355
>  %conv3.i = trunc i64 %div.i to i32
>  %14 = mul i64 %div.i, 4294883355
>  %15 = sub i64 %r.0.lcssa.i.i, %14
>  %conv4.i = trunc i64 %15 to i32
>  store i32 %conv3.i, i32 addrspace(5)* %x, align 4, !tbaa !8
>  store i32 %conv4.i, i32 addrspace(5)* %c, align 4, !tbaa !10
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define void @MWC64X_SeedStreams(%struct.mwc64x_state_t addrspace(5)* nocapture %s, i64 %baseOffset, i64 %perStreamOffset) local_unnamed_addr #4 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %conv.i = zext i32 %8 to i64
>  %mul1.i = mul i64 %conv.i, %perStreamOffset
>  %add2.i = add i64 %mul1.i, %baseOffset
>  %cmp38.i.i = icmp eq i64 %add2.i, 0
>  br i1 %cmp38.i.i, label %while.body.i.i.preheader, label %while.body.i17.i
>
>while.body.i17.i:                                 ; preds = %entry, %MWC_MulMod64.exit37.i.i
>  %acc.041.i.i = phi i64 [ %acc.1.i.i, %MWC_MulMod64.exit37.i.i ], [ 1, %entry ]
>  %sqr.040.i.i = phi i64 [ %r.0.lcssa.i36.i.i, %MWC_MulMod64.exit37.i.i ], [ 4294883355, %entry ]
>  %e.addr.039.i.i = phi i64 [ %shr.i22.i, %MWC_MulMod64.exit37.i.i ], [ %add2.i, %entry ]
>  %and.i15.i = and i64 %e.addr.039.i.i, 1
>  %tobool.i16.i = icmp eq i64 %and.i15.i, 0
>  br i1 %tobool.i16.i, label %if.end.i21.i, label %if.then.i18.i
>
>if.then.i18.i:                                    ; preds = %while.body.i17.i
>  %cmp17.i.i.i = icmp eq i64 %acc.041.i.i, 0
>  br i1 %cmp17.i.i.i, label %if.end.i21.i, label %while.body.i.i.i
>
>while.body.i.i.i:                                 ; preds = %if.then.i18.i, %MWC_AddMod64.exit16.i.i.i
>  %r.020.i.i.i = phi i64 [ %r.1.i.i.i, %MWC_AddMod64.exit16.i.i.i ], [ 0, %if.then.i18.i ]
>  %a.addr.019.i.i.i = phi i64 [ %shr.i.i.i, %MWC_AddMod64.exit16.i.i.i ], [ %acc.041.i.i, %if.then.i18.i ]
>  %b.addr.018.i.i.i = phi i64 [ %v.0.i15.i.i.i, %MWC_AddMod64.exit16.i.i.i ], [ %sqr.040.i.i, %if.then.i18.i ]
>  %and.i.i.i = and i64 %a.addr.019.i.i.i, 1
>  %tobool.i.i.i = icmp eq i64 %and.i.i.i, 0
>  br i1 %tobool.i.i.i, label %if.end.i.i.i, label %if.then.i.i19.i
>
>if.then.i.i19.i:                                  ; preds = %while.body.i.i.i
>  %add.i.i.i.i = add i64 %b.addr.018.i.i.i, %r.020.i.i.i
>  %cmp.i.i.i.i = icmp ult i64 %add.i.i.i.i, -360523849793537
>  br i1 %cmp.i.i.i.i, label %lor.lhs.false.i.i.i.i, label %if.then.i.i.i.i
>
>lor.lhs.false.i.i.i.i:                            ; preds = %if.then.i.i19.i
>  %9 = uitofp i64 %add.i.i.i.i to float
>  %10 = uitofp i64 %r.020.i.i.i to float
>  %cmp2.i.i.i.i = fcmp olt float %9, %10
>  br i1 %cmp2.i.i.i.i, label %if.then.i.i.i.i, label %if.end.i.i.i
>
>if.then.i.i.i.i:                                  ; preds = %lor.lhs.false.i.i.i.i, %if.then.i.i19.i
>  %sub.i.i.i.i = add i64 %add.i.i.i.i, 360523849793537
>  br label %if.end.i.i.i
>
>if.end.i.i.i:                                     ; preds = %if.then.i.i.i.i, %lor.lhs.false.i.i.i.i, %while.body.i.i.i
>  %r.1.i.i.i = phi i64 [ %r.020.i.i.i, %while.body.i.i.i ], [ %sub.i.i.i.i, %if.then.i.i.i.i ], [ %add.i.i.i.i, %lor.lhs.false.i.i.i.i ]
>  %add.i9.i.i.i = shl i64 %b.addr.018.i.i.i, 1
>  %cmp.i10.i.i.i = icmp ult i64 %add.i9.i.i.i, -360523849793537
>  br i1 %cmp.i10.i.i.i, label %lor.lhs.false.i12.i.i.i, label %if.then.i14.i.i.i
>
>lor.lhs.false.i12.i.i.i:                          ; preds = %if.end.i.i.i
>  %11 = uitofp i64 %add.i9.i.i.i to float
>  %12 = uitofp i64 %b.addr.018.i.i.i to float
>  %cmp2.i11.i.i.i = fcmp olt float %11, %12
>  br i1 %cmp2.i11.i.i.i, label %if.then.i14.i.i.i, label %MWC_AddMod64.exit16.i.i.i
>
>if.then.i14.i.i.i:                                ; preds = %lor.lhs.false.i12.i.i.i, %if.end.i.i.i
>  %sub.i13.i.i.i = add i64 %add.i9.i.i.i, 360523849793537
>  br label %MWC_AddMod64.exit16.i.i.i
>
>MWC_AddMod64.exit16.i.i.i:                        ; preds = %if.then.i14.i.i.i, %lor.lhs.false.i12.i.i.i
>  %v.0.i15.i.i.i = phi i64 [ %sub.i13.i.i.i, %if.then.i14.i.i.i ], [ %add.i9.i.i.i, %lor.lhs.false.i12.i.i.i ]
>  %shr.i.i.i = lshr i64 %a.addr.019.i.i.i, 1
>  %cmp.i.i20.i = icmp eq i64 %shr.i.i.i, 0
>  br i1 %cmp.i.i20.i, label %if.end.i21.i, label %while.body.i.i.i
>
>if.end.i21.i:                                     ; preds = %MWC_AddMod64.exit16.i.i.i, %if.then.i18.i, %while.body.i17.i
>  %acc.1.i.i = phi i64 [ %acc.041.i.i, %while.body.i17.i ], [ 0, %if.then.i18.i ], [ %r.1.i.i.i, %MWC_AddMod64.exit16.i.i.i ]
>  %cmp17.i10.i.i = icmp eq i64 %sqr.040.i.i, 0
>  br i1 %cmp17.i10.i.i, label %MWC_MulMod64.exit37.i.i, label %while.body.i16.i.i
>
>while.body.i16.i.i:                               ; preds = %if.end.i21.i, %MWC_AddMod64.exit16.i35.i.i
>  %r.020.i11.i.i = phi i64 [ %r.1.i24.i.i, %MWC_AddMod64.exit16.i35.i.i ], [ 0, %if.end.i21.i ]
>  %a.addr.019.i12.i.i = phi i64 [ %shr.i33.i.i, %MWC_AddMod64.exit16.i35.i.i ], [ %sqr.040.i.i, %if.end.i21.i ]
>  %b.addr.018.i13.i.i = phi i64 [ %v.0.i15.i32.i.i, %MWC_AddMod64.exit16.i35.i.i ], [ %sqr.040.i.i, %if.end.i21.i ]
>  %and.i14.i.i = and i64 %a.addr.019.i12.i.i, 1
>  %tobool.i15.i.i = icmp eq i64 %and.i14.i.i, 0
>  br i1 %tobool.i15.i.i, label %if.end.i27.i.i, label %if.then.i19.i.i
>
>if.then.i19.i.i:                                  ; preds = %while.body.i16.i.i
>  %add.i.i17.i.i = add i64 %b.addr.018.i13.i.i, %r.020.i11.i.i
>  %cmp.i.i18.i.i = icmp ult i64 %add.i.i17.i.i, -360523849793537
>  br i1 %cmp.i.i18.i.i, label %lor.lhs.false.i.i21.i.i, label %if.then.i.i23.i.i
>
>lor.lhs.false.i.i21.i.i:                          ; preds = %if.then.i19.i.i
>  %13 = uitofp i64 %add.i.i17.i.i to float
>  %14 = uitofp i64 %r.020.i11.i.i to float
>  %cmp2.i.i20.i.i = fcmp olt float %13, %14
>  br i1 %cmp2.i.i20.i.i, label %if.then.i.i23.i.i, label %if.end.i27.i.i
>
>if.then.i.i23.i.i:                                ; preds = %lor.lhs.false.i.i21.i.i, %if.then.i19.i.i
>  %sub.i.i22.i.i = add i64 %add.i.i17.i.i, 360523849793537
>  br label %if.end.i27.i.i
>
>if.end.i27.i.i:                                   ; preds = %if.then.i.i23.i.i, %lor.lhs.false.i.i21.i.i, %while.body.i16.i.i
>  %r.1.i24.i.i = phi i64 [ %r.020.i11.i.i, %while.body.i16.i.i ], [ %sub.i.i22.i.i, %if.then.i.i23.i.i ], [ %add.i.i17.i.i, %lor.lhs.false.i.i21.i.i ]
>  %add.i9.i25.i.i = shl i64 %b.addr.018.i13.i.i, 1
>  %cmp.i10.i26.i.i = icmp ult i64 %add.i9.i25.i.i, -360523849793537
>  br i1 %cmp.i10.i26.i.i, label %lor.lhs.false.i12.i29.i.i, label %if.then.i14.i31.i.i
>
>lor.lhs.false.i12.i29.i.i:                        ; preds = %if.end.i27.i.i
>  %15 = uitofp i64 %add.i9.i25.i.i to float
>  %16 = uitofp i64 %b.addr.018.i13.i.i to float
>  %cmp2.i11.i28.i.i = fcmp olt float %15, %16
>  br i1 %cmp2.i11.i28.i.i, label %if.then.i14.i31.i.i, label %MWC_AddMod64.exit16.i35.i.i
>
>if.then.i14.i31.i.i:                              ; preds = %lor.lhs.false.i12.i29.i.i, %if.end.i27.i.i
>  %sub.i13.i30.i.i = add i64 %add.i9.i25.i.i, 360523849793537
>  br label %MWC_AddMod64.exit16.i35.i.i
>
>MWC_AddMod64.exit16.i35.i.i:                      ; preds = %if.then.i14.i31.i.i, %lor.lhs.false.i12.i29.i.i
>  %v.0.i15.i32.i.i = phi i64 [ %sub.i13.i30.i.i, %if.then.i14.i31.i.i ], [ %add.i9.i25.i.i, %lor.lhs.false.i12.i29.i.i ]
>  %shr.i33.i.i = lshr i64 %a.addr.019.i12.i.i, 1
>  %cmp.i34.i.i = icmp eq i64 %shr.i33.i.i, 0
>  br i1 %cmp.i34.i.i, label %MWC_MulMod64.exit37.i.i, label %while.body.i16.i.i
>
>MWC_MulMod64.exit37.i.i:                          ; preds = %MWC_AddMod64.exit16.i35.i.i, %if.end.i21.i
>  %r.0.lcssa.i36.i.i = phi i64 [ 0, %if.end.i21.i ], [ %r.1.i24.i.i, %MWC_AddMod64.exit16.i35.i.i ]
>  %shr.i22.i = lshr i64 %e.addr.039.i.i, 1
>  %cmp.i23.i = icmp eq i64 %shr.i22.i, 0
>  br i1 %cmp.i23.i, label %while.body.i.i.preheader, label %while.body.i17.i
>
>while.body.i.i.preheader:                         ; preds = %MWC_MulMod64.exit37.i.i, %entry
>  %b.addr.018.i.i.ph = phi i64 [ 1, %entry ], [ %acc.1.i.i, %MWC_MulMod64.exit37.i.i ]
>  br label %while.body.i.i
>
>while.body.i.i:                                   ; preds = %MWC_AddMod64.exit16.i.i.1, %while.body.i.i.preheader
>  %r.020.i.i = phi i64 [ 0, %while.body.i.i.preheader ], [ %r.1.i.i.1, %MWC_AddMod64.exit16.i.i.1 ]
>  %a.addr.019.i.i = phi i64 [ 4077358422479273989, %while.body.i.i.preheader ], [ %shr.i.i, %MWC_AddMod64.exit16.i.i.1 ]
>  %b.addr.018.i.i = phi i64 [ %b.addr.018.i.i.ph, %while.body.i.i.preheader ], [ %v.0.i15.i.i.1, %MWC_AddMod64.exit16.i.i.1 ]
>  %and.i.i = and i64 %a.addr.019.i.i, 1
>  %tobool.i.i = icmp eq i64 %and.i.i, 0
>  br i1 %tobool.i.i, label %if.end.i.i, label %if.then.i.i
>
>if.then.i.i:                                      ; preds = %while.body.i.i
>  %add.i.i.i = add i64 %b.addr.018.i.i, %r.020.i.i
>  %cmp.i.i.i = icmp ult i64 %add.i.i.i, -360523849793537
>  br i1 %cmp.i.i.i, label %lor.lhs.false.i.i.i, label %if.then.i.i.i
>
>lor.lhs.false.i.i.i:                              ; preds = %if.then.i.i
>  %17 = uitofp i64 %add.i.i.i to float
>  %18 = uitofp i64 %r.020.i.i to float
>  %cmp2.i.i.i = fcmp olt float %17, %18
>  br i1 %cmp2.i.i.i, label %if.then.i.i.i, label %if.end.i.i
>
>if.then.i.i.i:                                    ; preds = %lor.lhs.false.i.i.i, %if.then.i.i
>  %sub.i.i.i = add i64 %add.i.i.i, 360523849793537
>  br label %if.end.i.i
>
>if.end.i.i:                                       ; preds = %if.then.i.i.i, %lor.lhs.false.i.i.i, %while.body.i.i
>  %r.1.i.i = phi i64 [ %r.020.i.i, %while.body.i.i ], [ %sub.i.i.i, %if.then.i.i.i ], [ %add.i.i.i, %lor.lhs.false.i.i.i ]
>  %add.i9.i.i = shl i64 %b.addr.018.i.i, 1
>  %cmp.i10.i.i = icmp ult i64 %add.i9.i.i, -360523849793537
>  br i1 %cmp.i10.i.i, label %lor.lhs.false.i12.i.i, label %if.then.i14.i.i
>
>lor.lhs.false.i12.i.i:                            ; preds = %if.end.i.i
>  %19 = uitofp i64 %add.i9.i.i to float
>  %20 = uitofp i64 %b.addr.018.i.i to float
>  %cmp2.i11.i.i = fcmp olt float %19, %20
>  br i1 %cmp2.i11.i.i, label %if.then.i14.i.i, label %MWC_AddMod64.exit16.i.i
>
>if.then.i14.i.i:                                  ; preds = %lor.lhs.false.i12.i.i, %if.end.i.i
>  %sub.i13.i.i = add i64 %add.i9.i.i, 360523849793537
>  br label %MWC_AddMod64.exit16.i.i
>
>MWC_AddMod64.exit16.i.i:                          ; preds = %if.then.i14.i.i, %lor.lhs.false.i12.i.i
>  %v.0.i15.i.i = phi i64 [ %sub.i13.i.i, %if.then.i14.i.i ], [ %add.i9.i.i, %lor.lhs.false.i12.i.i ]
>  %shr.i.i = lshr i64 %a.addr.019.i.i, 2
>  %and.i.i.1 = and i64 %a.addr.019.i.i, 2
>  %tobool.i.i.1 = icmp eq i64 %and.i.i.1, 0
>  br i1 %tobool.i.i.1, label %if.end.i.i.1, label %if.then.i.i.1
>
>MWC_SeedImpl_Mod64.exit:                          ; preds = %MWC_AddMod64.exit16.i.i.1
>  %div.i = udiv i64 %r.1.i.i.1, 4294883355
>  %conv5.i = trunc i64 %div.i to i32
>  %21 = mul i64 %div.i, 4294883355
>  %22 = sub i64 %r.1.i.i.1, %21
>  %conv6.i = trunc i64 %22 to i32
>  %x = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %s, i32 0, i32 0
>  store i32 %conv5.i, i32 addrspace(5)* %x, align 4, !tbaa !8
>  %c = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %s, i32 0, i32 1
>  store i32 %conv6.i, i32 addrspace(5)* %c, align 4, !tbaa !10
>  ret void
>
>if.then.i.i.1:                                    ; preds = %MWC_AddMod64.exit16.i.i
>  %add.i.i.i.1 = add i64 %v.0.i15.i.i, %r.1.i.i
>  %cmp.i.i.i.1 = icmp ult i64 %add.i.i.i.1, -360523849793537
>  br i1 %cmp.i.i.i.1, label %lor.lhs.false.i.i.i.1, label %if.then.i.i.i.1
>
>lor.lhs.false.i.i.i.1:                            ; preds = %if.then.i.i.1
>  %23 = uitofp i64 %add.i.i.i.1 to float
>  %24 = uitofp i64 %r.1.i.i to float
>  %cmp2.i.i.i.1 = fcmp olt float %23, %24
>  br i1 %cmp2.i.i.i.1, label %if.then.i.i.i.1, label %if.end.i.i.1
>
>if.then.i.i.i.1:                                  ; preds = %lor.lhs.false.i.i.i.1, %if.then.i.i.1
>  %sub.i.i.i.1 = add i64 %add.i.i.i.1, 360523849793537
>  br label %if.end.i.i.1
>
>if.end.i.i.1:                                     ; preds = %if.then.i.i.i.1, %lor.lhs.false.i.i.i.1, %MWC_AddMod64.exit16.i.i
>  %r.1.i.i.1 = phi i64 [ %r.1.i.i, %MWC_AddMod64.exit16.i.i ], [ %sub.i.i.i.1, %if.then.i.i.i.1 ], [ %add.i.i.i.1, %lor.lhs.false.i.i.i.1 ]
>  %add.i9.i.i.1 = shl i64 %v.0.i15.i.i, 1
>  %cmp.i10.i.i.1 = icmp ult i64 %add.i9.i.i.1, -360523849793537
>  br i1 %cmp.i10.i.i.1, label %lor.lhs.false.i12.i.i.1, label %if.then.i14.i.i.1
>
>lor.lhs.false.i12.i.i.1:                          ; preds = %if.end.i.i.1
>  %25 = uitofp i64 %add.i9.i.i.1 to float
>  %26 = uitofp i64 %v.0.i15.i.i to float
>  %cmp2.i11.i.i.1 = fcmp olt float %25, %26
>  br i1 %cmp2.i11.i.i.1, label %if.then.i14.i.i.1, label %MWC_AddMod64.exit16.i.i.1
>
>if.then.i14.i.i.1:                                ; preds = %lor.lhs.false.i12.i.i.1, %if.end.i.i.1
>  %sub.i13.i.i.1 = add i64 %add.i9.i.i.1, 360523849793537
>  br label %MWC_AddMod64.exit16.i.i.1
>
>MWC_AddMod64.exit16.i.i.1:                        ; preds = %if.then.i14.i.i.1, %lor.lhs.false.i12.i.i.1
>  %v.0.i15.i.i.1 = phi i64 [ %sub.i13.i.i.1, %if.then.i14.i.i.1 ], [ %add.i9.i.i.1, %lor.lhs.false.i12.i.i.1 ]
>  %cmp.i.i.1 = icmp eq i64 %shr.i.i, 0
>  br i1 %cmp.i.i.1, label %MWC_SeedImpl_Mod64.exit, label %while.body.i.i
>}
>
>; Function Attrs: norecurse nounwind
>define i32 @MWC64X_NextUint(%struct.mwc64x_state_t addrspace(5)* nocapture %s) local_unnamed_addr #3 {
>entry:
>  %x = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %s, i32 0, i32 0
>  %0 = load i32, i32 addrspace(5)* %x, align 4, !tbaa !8
>  %c = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %s, i32 0, i32 1
>  %1 = load i32, i32 addrspace(5)* %c, align 4, !tbaa !10
>  %xor = xor i32 %1, %0
>  %mul.i = mul i32 %0, -83941
>  %add.i = add i32 %mul.i, %1
>  %cmp.i = icmp ult i32 %add.i, %1
>  %conv.i = zext i1 %cmp.i to i32
>  %2 = zext i32 %0 to i64
>  %3 = mul nuw i64 %2, 4294883355
>  %4 = lshr i64 %3, 32
>  %5 = trunc i64 %4 to i32
>  %add1.i = add i32 %5, %conv.i
>  store i32 %add.i, i32 addrspace(5)* %x, align 4, !tbaa !8
>  store i32 %add1.i, i32 addrspace(5)* %c, align 4, !tbaa !10
>  ret i32 %xor
>}
>
>; Function Attrs: norecurse nounwind
>define float @mwcReadPseudoRandomValue(%struct.mwc64x_state_t addrspace(5)* nocapture %rng) local_unnamed_addr #3 {
>entry:
>  %x.i = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %rng, i32 0, i32 0
>  %0 = load i32, i32 addrspace(5)* %x.i, align 4, !tbaa !8
>  %c.i = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %rng, i32 0, i32 1
>  %1 = load i32, i32 addrspace(5)* %c.i, align 4, !tbaa !10
>  %xor.i = xor i32 %1, %0
>  %mul.i.i = mul i32 %0, -83941
>  %add.i.i = add i32 %mul.i.i, %1
>  %cmp.i.i = icmp ult i32 %add.i.i, %1
>  %conv.i.i = zext i1 %cmp.i.i to i32
>  %2 = zext i32 %0 to i64
>  %3 = mul nuw i64 %2, 4294883355
>  %4 = lshr i64 %3, 32
>  %5 = trunc i64 %4 to i32
>  %add1.i.i = add i32 %5, %conv.i.i
>  store i32 %add.i.i, i32 addrspace(5)* %x.i, align 4, !tbaa !8
>  store i32 %add1.i.i, i32 addrspace(5)* %c.i, align 4, !tbaa !10
>  %conv = uitofp i32 %xor.i to float
>  %div = fmul float %conv, 0x3DF0000000000000
>  ret float %div
>}
>
>; Function Attrs: convergent nounwind
>define float @mwcGenerateDifferentialNoise(%struct.mwc64x_state_t addrspace(5)* nocapture %r, float %pixel, i32 %noise_type, float %attenuate) local_unnamed_addr #4 {
>entry:
>  %x.i.i = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %r, i32 0, i32 0
>  %0 = load i32, i32 addrspace(5)* %x.i.i, align 4, !tbaa !8
>  %c.i.i = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %r, i32 0, i32 1
>  %1 = load i32, i32 addrspace(5)* %c.i.i, align 4, !tbaa !10
>  %xor.i.i = xor i32 %1, %0
>  %mul.i.i.i = mul i32 %0, -83941
>  %add.i.i.i = add i32 %mul.i.i.i, %1
>  %cmp.i.i.i = icmp ult i32 %add.i.i.i, %1
>  %conv.i.i.i = zext i1 %cmp.i.i.i to i32
>  %2 = zext i32 %0 to i64
>  %3 = mul nuw i64 %2, 4294883355
>  %4 = lshr i64 %3, 32
>  %5 = trunc i64 %4 to i32
>  %add1.i.i.i = add i32 %5, %conv.i.i.i
>  store i32 %add.i.i.i, i32 addrspace(5)* %x.i.i, align 4, !tbaa !8
>  store i32 %add1.i.i.i, i32 addrspace(5)* %c.i.i, align 4, !tbaa !10
>  %conv.i = uitofp i32 %xor.i.i to float
>  %div.i = fmul float %conv.i, 0x3DF0000000000000
>  switch i32 %noise_type, label %sw.default [
>    i32 7, label %sw.bb87
>    i32 2, label %sw.bb3
>    i32 4, label %sw.bb21
>    i32 5, label %sw.bb33
>    i32 3, label %sw.bb59
>    i32 6, label %sw.bb75
>  ]
>
>sw.default:                                       ; preds = %entry
>  %mul = fmul float %attenuate, 1.562500e-02
>  %mul1 = fmul float %mul, 6.553500e+04
>  %sub = fadd float %div.i, -5.000000e-01
>  %6 = tail call float @llvm.fmuladd.f32(float %mul1, float %sub, float %pixel)
>  br label %sw.epilog
>
>sw.bb3:                                           ; preds = %entry
>  %cmp = fcmp oeq float %div.i, 0.000000e+00
>  %alpha.0 = select i1 %cmp, float 1.000000e+00, float %div.i
>  %xor.i.i167 = xor i32 %add1.i.i.i, %add.i.i.i
>  %mul.i.i.i168 = mul i32 %add.i.i.i, -83941
>  %add.i.i.i169 = add i32 %add1.i.i.i, %mul.i.i.i168
>  %cmp.i.i.i170 = icmp ult i32 %add.i.i.i169, %add1.i.i.i
>  %conv.i.i.i171 = zext i1 %cmp.i.i.i170 to i32
>  %7 = zext i32 %add.i.i.i to i64
>  %8 = mul nuw i64 %7, 4294883355
>  %9 = lshr i64 %8, 32
>  %10 = trunc i64 %9 to i32
>  %add1.i.i.i172 = add i32 %conv.i.i.i171, %10
>  store i32 %add.i.i.i169, i32 addrspace(5)* %x.i.i, align 4, !tbaa !8
>  store i32 %add1.i.i.i172, i32 addrspace(5)* %c.i.i, align 4, !tbaa !10
>  %conv.i173 = uitofp i32 %xor.i.i167 to float
>  %div.i174 = fmul float %conv.i173, 0x3DF0000000000000
>  %11 = bitcast float %alpha.0 to i32
>  %12 = and i32 %11, 2147483647
>  %13 = fadd float %alpha.0, -1.000000e+00
>  %14 = tail call float @llvm.fabs.f32(float %13) #14
>  %15 = fcmp olt float %14, 6.250000e-02
>  %16 = fadd float %13, 2.000000e+00
>  %17 = fdiv float %13, %16, !fpmath !11
>  %18 = fmul float %13, %17
>  %19 = fadd float %17, %17
>  %20 = fmul float %19, %19
>  %21 = tail call float @llvm.fmuladd.f32(float %20, float 0x3F899999A0000000, float 0x3FB5555560000000) #13
>  %22 = fmul float %20, %21
>  %23 = fsub float -0.000000e+00, %18
>  %24 = tail call float @llvm.fmuladd.f32(float %19, float %22, float %23) #13
>  %25 = bitcast float %13 to i32
>  %26 = and i32 %25, -65536
>  %27 = bitcast i32 %26 to float
>  %28 = fsub float %13, %27
>  %29 = fadd float %28, %24
>  %30 = fmul float %29, 0x3F75476520000000
>  %31 = tail call float @llvm.fmuladd.f32(float %27, float 0x3F75476520000000, float %30) #13
>  %32 = tail call float @llvm.fmuladd.f32(float %29, float 1.437500e+00, float %31) #13
>  %33 = tail call float @llvm.fmuladd.f32(float %27, float 1.437500e+00, float %32) #13
>  %34 = lshr i32 %11, 23
>  %35 = add nsw i32 %34, -127
>  %36 = or i32 %11, 1065353216
>  %37 = bitcast i32 %36 to float
>  %38 = fadd float %37, -1.000000e+00
>  %39 = bitcast float %38 to i32
>  %40 = lshr i32 %39, 23
>  %41 = add nsw i32 %40, -253
>  %42 = icmp eq i32 %34, 0
>  %43 = select i1 %42, i32 %41, i32 %35
>  %44 = select i1 %42, i32 %39, i32 %11
>  %45 = sitofp i32 %43 to float
>  %46 = and i32 %44, 8323072
>  %47 = shl i32 %44, 1
>  %48 = and i32 %47, 65536
>  %49 = add nuw nsw i32 %48, %46
>  %50 = or i32 %49, 1056964608
>  %51 = bitcast i32 %50 to float
>  %52 = and i32 %44, 8388607
>  %53 = or i32 %52, 1056964608
>  %54 = bitcast i32 %53 to float
>  %55 = fsub float %51, %54
>  %56 = lshr exact i32 %49, 16
>  %57 = getelementptr inbounds [129 x float], [129 x float] addrspace(4)* @LOG_INV_TBL, i32 0, i32 %56
>  %58 = load float, float addrspace(4)* %57, align 4, !tbaa !12
>  %59 = fmul float %55, %58
>  %60 = tail call float @llvm.fmuladd.f32(float %59, float 0x3FD5555560000000, float 5.000000e-01) #13
>  %61 = fmul float %59, %59
>  %62 = tail call float @llvm.fmuladd.f32(float %60, float %61, float %59) #13
>  %63 = getelementptr inbounds [129 x <2 x float>], [129 x <2 x float>] addrspace(4)* @LOG2_TBL, i32 0, i32 %56
>  %64 = load <2 x float>, <2 x float> addrspace(4)* %63, align 8, !tbaa !14
>  %65 = extractelement <2 x float> %64, i32 0
>  %66 = fadd float %65, %45
>  %67 = extractelement <2 x float> %64, i32 1
>  %68 = tail call float @llvm.fmuladd.f32(float %62, float 0xBFF7154760000000, float %67) #13
>  %69 = fadd float %66, %68
>  %70 = select i1 %15, float %33, float %69
>  %71 = icmp ugt i32 %12, 2139095039
>  %72 = select i1 %71, float %alpha.0, float %70
>  %73 = icmp eq i32 %12, %11
>  %74 = icmp eq i32 %12, 0
>  %.op1.i163 = fmul float %72, 0x3FE62E4300000000
>  %.op1.i163.op = fmul float %.op1.i163, -2.000000e+00
>  %.op.i164.op = select i1 %73, float %.op1.i163.op, float 0x7FF8000000000000
>  %mul6 = select i1 %74, float 0x7FF0000000000000, float %.op.i164.op
>  %75 = fcmp olt float %mul6, 0.000000e+00
>  %76 = tail call float @llvm.sqrt.f32(float %mul6) #14
>  %77 = select i1 %75, float 0x7FF8000000000000, float %76
>  %mul8 = fmul float %div.i174, 2.000000e+00
>  %78 = bitcast float %mul8 to i32
>  %79 = and i32 %78, 2147483647
>  %80 = bitcast i32 %79 to float
>  %81 = fptosi float %80 to i32
>  %82 = sitofp i32 %81 to float
>  %83 = fsub float %80, %82
>  %84 = shl i32 %81, 31
>  %85 = icmp ult i32 %79, 2139095040
>  %86 = select i1 %85, i32 1065353216, i32 2143289344
>  %87 = icmp ult i32 %79, 1266679808
>  %88 = or i32 %84, 1065353216
>  %89 = select i1 %87, i32 %88, i32 %86
>  %90 = fsub float 1.000000e+00, %83
>  %91 = xor i32 %84, -2147483648
>  %92 = fcmp ole float %83, 7.500000e-01
>  %93 = fadd float %83, -5.000000e-01
>  %94 = select i1 %92, float %93, float %90
>  %95 = xor i1 %92, true
>  %96 = fcmp olt float %83, 5.000000e-01
>  %97 = fsub float 5.000000e-01, %83
>  %98 = select i1 %96, float %97, float %94
>  %99 = select i1 %96, i32 %84, i32 %91
>  %100 = fcmp ole float %83, 2.500000e-01
>  %101 = select i1 %100, float %83, float %98
>  %102 = or i1 %100, %95
>  %103 = fmul float %101, 0x400921FB60000000
>  %104 = fmul float %103, %103
>  %105 = fmul float %103, %104
>  %106 = tail call float @llvm.fmuladd.f32(float %104, float 0x3EC6DBE4A0000000, float 0xBF2A013A80000000) #13
>  %107 = tail call float @llvm.fmuladd.f32(float %104, float %106, float 0x3F811110E0000000) #13
>  %108 = tail call float @llvm.fmuladd.f32(float %104, float %107, float 0xBFC5555560000000) #13
>  %109 = tail call float @llvm.fmuladd.f32(float %105, float %108, float %103) #13
>  %110 = fmul float %104, %104
>  %111 = tail call float @llvm.fmuladd.f32(float %104, float 0xBE92524740000000, float 0x3EFA015C60000000) #13
>  %112 = tail call float @llvm.fmuladd.f32(float %104, float %111, float 0xBF56C16C00000000) #13
>  %113 = tail call float @llvm.fmuladd.f32(float %104, float %112, float 0x3FA5555560000000) #13
>  %114 = tail call float @llvm.fmuladd.f32(float %104, float -5.000000e-01, float 1.000000e+00) #13
>  %115 = tail call float @llvm.fmuladd.f32(float %110, float %113, float %114) #13
>  %116 = select i1 %102, float %115, float %109
>  %117 = bitcast float %116 to i32
>  %118 = xor i32 %99, %117
>  %119 = icmp ult i32 %79, 1258291200
>  %120 = select i1 %119, i32 %118, i32 %89
>  %121 = bitcast i32 %120 to float
>  %mul10 = fmul float %77, %121
>  %122 = and i32 %78, -2147483648
>  %123 = xor i32 %84, %122
>  %124 = select i1 %85, i32 %122, i32 2143289344
>  %125 = xor i1 %100, true
>  %126 = and i1 %92, %125
>  %127 = select i1 %126, float %115, float %109
>  %128 = bitcast float %127 to i32
>  %129 = xor i32 %123, %128
>  %130 = select i1 %119, i32 %129, i32 %124
>  %131 = bitcast i32 %130 to float
>  %mul13 = fmul float %77, %131
>  %132 = fcmp olt float %pixel, 0.000000e+00
>  %133 = tail call float @llvm.sqrt.f32(float %pixel) #14
>  %134 = select i1 %132, float 0x7FF8000000000000, float %133
>  %mul15 = fmul float %attenuate, 1.562500e-02
>  %mul16 = fmul float %134, %mul15
>  %135 = tail call float @llvm.fmuladd.f32(float %mul16, float %mul10, float %pixel)
>  %mul18 = fmul float %attenuate, 7.812500e-02
>  %mul19 = fmul float %mul18, 6.553500e+04
>  %136 = tail call float @llvm.fmuladd.f32(float %mul19, float %mul13, float %135)
>  br label %sw.epilog
>
>sw.bb21:                                          ; preds = %entry
>  %mul22 = fmul float %attenuate, 0x3FB99999A0000000
>  %div = fmul float %mul22, 5.000000e-01
>  %cmp23 = fcmp olt float %div.i, %div
>  br i1 %cmp23, label %sw.epilog, label %if.else
>
>if.else:                                          ; preds = %sw.bb21
>  %sub27 = fsub float 1.000000e+00, %div
>  %cmp28 = fcmp ult float %div.i, %sub27
>  %pixel. = select i1 %cmp28, float %pixel, float 6.553500e+04
>  br label %sw.epilog
>
>sw.bb33:                                          ; preds = %entry
>  %cmp34 = fcmp ugt float %div.i, 5.000000e-01
>  br i1 %cmp34, label %if.end46, label %if.then35
>
>if.then35:                                        ; preds = %sw.bb33
>  %cmp36 = fcmp ugt float %div.i, 0.000000e+00
>  br i1 %cmp36, label %if.else39, label %if.then37
>
>if.then37:                                        ; preds = %if.then35
>  %sub38 = fadd float %pixel, -6.553500e+04
>  br label %sw.epilog
>
>if.else39:                                        ; preds = %if.then35
>  %mul40 = fmul float %attenuate, 3.906250e-02
>  %mul41 = fmul float %mul40, 6.553500e+04
>  %mul42 = fmul float %div.i, 2.000000e+00
>  %137 = bitcast float %mul42 to i32
>  %138 = and i32 %137, 2147483647
>  %139 = fadd float %mul42, -1.000000e+00
>  %140 = tail call float @llvm.fabs.f32(float %139) #14
>  %141 = fcmp olt float %140, 6.250000e-02
>  %142 = fadd float %139, 2.000000e+00
>  %143 = fdiv float %139, %142, !fpmath !11
>  %144 = fmul float %139, %143
>  %145 = fadd float %143, %143
>  %146 = fmul float %145, %145
>  %147 = tail call float @llvm.fmuladd.f32(float %146, float 0x3F899999A0000000, float 0x3FB5555560000000) #13
>  %148 = fmul float %146, %147
>  %149 = fsub float -0.000000e+00, %144
>  %150 = tail call float @llvm.fmuladd.f32(float %145, float %148, float %149) #13
>  %151 = bitcast float %139 to i32
>  %152 = and i32 %151, -65536
>  %153 = bitcast i32 %152 to float
>  %154 = fsub float %139, %153
>  %155 = fadd float %154, %150
>  %156 = fmul float %155, 0x3F75476520000000
>  %157 = tail call float @llvm.fmuladd.f32(float %153, float 0x3F75476520000000, float %156) #13
>  %158 = tail call float @llvm.fmuladd.f32(float %155, float 1.437500e+00, float %157) #13
>  %159 = tail call float @llvm.fmuladd.f32(float %153, float 1.437500e+00, float %158) #13
>  %160 = lshr i32 %137, 23
>  %161 = add nsw i32 %160, -127
>  %162 = or i32 %137, 1065353216
>  %163 = bitcast i32 %162 to float
>  %164 = fadd float %163, -1.000000e+00
>  %165 = bitcast float %164 to i32
>  %166 = lshr i32 %165, 23
>  %167 = add nsw i32 %166, -253
>  %168 = icmp eq i32 %160, 0
>  %169 = select i1 %168, i32 %167, i32 %161
>  %170 = select i1 %168, i32 %165, i32 %137
>  %171 = sitofp i32 %169 to float
>  %172 = and i32 %170, 8323072
>  %173 = shl i32 %170, 1
>  %174 = and i32 %173, 65536
>  %175 = add nuw nsw i32 %174, %172
>  %176 = or i32 %175, 1056964608
>  %177 = bitcast i32 %176 to float
>  %178 = and i32 %170, 8388607
>  %179 = or i32 %178, 1056964608
>  %180 = bitcast i32 %179 to float
>  %181 = fsub float %177, %180
>  %182 = lshr exact i32 %175, 16
>  %183 = getelementptr inbounds [129 x float], [129 x float] addrspace(4)* @LOG_INV_TBL, i32 0, i32 %182
>  %184 = load float, float addrspace(4)* %183, align 4, !tbaa !12
>  %185 = fmul float %181, %184
>  %186 = tail call float @llvm.fmuladd.f32(float %185, float 0x3FD5555560000000, float 5.000000e-01) #13
>  %187 = fmul float %185, %185
>  %188 = tail call float @llvm.fmuladd.f32(float %186, float %187, float %185) #13
>  %189 = getelementptr inbounds [129 x <2 x float>], [129 x <2 x float>] addrspace(4)* @LOG2_TBL, i32 0, i32 %182
>  %190 = load <2 x float>, <2 x float> addrspace(4)* %189, align 8, !tbaa !14
>  %191 = extractelement <2 x float> %190, i32 0
>  %192 = fadd float %191, %171
>  %193 = extractelement <2 x float> %190, i32 1
>  %194 = tail call float @llvm.fmuladd.f32(float %188, float 0xBFF7154760000000, float %193) #13
>  %195 = fadd float %192, %194
>  %196 = select i1 %141, float %159, float %195
>  %197 = icmp ugt i32 %138, 2139095039
>  %198 = select i1 %197, float %mul42, float %196
>  %199 = icmp eq i32 %138, %137
>  %200 = icmp eq i32 %138, 0
>  %.op1.i161 = fmul float %198, 0x3FE62E4300000000
>  %.op.i162 = select i1 %199, float %.op1.i161, float 0x7FF8000000000000
>  %201 = select i1 %200, float 0xFFF0000000000000, float %.op.i162
>  %202 = tail call float @llvm.fmuladd.f32(float %mul41, float %201, float %pixel)
>  %add = fadd float %202, 5.000000e-01
>  br label %sw.epilog
>
>if.end46:                                         ; preds = %sw.bb33
>  %sub47 = fsub float 1.000000e+00, %div.i
>  %cmp48 = fcmp ugt float %sub47, 0.000000e+00
>  br i1 %cmp48, label %if.else51, label %if.then49
>
>if.then49:                                        ; preds = %if.end46
>  %add50 = fadd float %pixel, 6.553500e+04
>  br label %sw.epilog
>
>if.else51:                                        ; preds = %if.end46
>  %mul52 = fmul float %attenuate, 3.906250e-02
>  %mul53 = fmul float %mul52, 6.553500e+04
>  %mul54 = fmul float %sub47, 2.000000e+00
>  %203 = bitcast float %mul54 to i32
>  %204 = and i32 %203, 2147483647
>  %205 = fadd float %mul54, -1.000000e+00
>  %206 = tail call float @llvm.fabs.f32(float %205) #14
>  %207 = fcmp olt float %206, 6.250000e-02
>  %208 = fadd float %205, 2.000000e+00
>  %209 = fdiv float %205, %208, !fpmath !11
>  %210 = fmul float %205, %209
>  %211 = fadd float %209, %209
>  %212 = fmul float %211, %211
>  %213 = tail call float @llvm.fmuladd.f32(float %212, float 0x3F899999A0000000, float 0x3FB5555560000000) #13
>  %214 = fmul float %212, %213
>  %215 = fsub float -0.000000e+00, %210
>  %216 = tail call float @llvm.fmuladd.f32(float %211, float %214, float %215) #13
>  %217 = bitcast float %205 to i32
>  %218 = and i32 %217, -65536
>  %219 = bitcast i32 %218 to float
>  %220 = fsub float %205, %219
>  %221 = fadd float %220, %216
>  %222 = fmul float %221, 0x3F75476520000000
>  %223 = tail call float @llvm.fmuladd.f32(float %219, float 0x3F75476520000000, float %222) #13
>  %224 = tail call float @llvm.fmuladd.f32(float %221, float 1.437500e+00, float %223) #13
>  %225 = tail call float @llvm.fmuladd.f32(float %219, float 1.437500e+00, float %224) #13
>  %226 = lshr i32 %203, 23
>  %227 = add nsw i32 %226, -127
>  %228 = or i32 %203, 1065353216
>  %229 = bitcast i32 %228 to float
>  %230 = fadd float %229, -1.000000e+00
>  %231 = bitcast float %230 to i32
>  %232 = lshr i32 %231, 23
>  %233 = add nsw i32 %232, -253
>  %234 = icmp eq i32 %226, 0
>  %235 = select i1 %234, i32 %233, i32 %227
>  %236 = select i1 %234, i32 %231, i32 %203
>  %237 = sitofp i32 %235 to float
>  %238 = and i32 %236, 8323072
>  %239 = shl i32 %236, 1
>  %240 = and i32 %239, 65536
>  %241 = add nuw nsw i32 %240, %238
>  %242 = or i32 %241, 1056964608
>  %243 = bitcast i32 %242 to float
>  %244 = and i32 %236, 8388607
>  %245 = or i32 %244, 1056964608
>  %246 = bitcast i32 %245 to float
>  %247 = fsub float %243, %246
>  %248 = lshr exact i32 %241, 16
>  %249 = getelementptr inbounds [129 x float], [129 x float] addrspace(4)* @LOG_INV_TBL, i32 0, i32 %248
>  %250 = load float, float addrspace(4)* %249, align 4, !tbaa !12
>  %251 = fmul float %247, %250
>  %252 = tail call float @llvm.fmuladd.f32(float %251, float 0x3FD5555560000000, float 5.000000e-01) #13
>  %253 = fmul float %251, %251
>  %254 = tail call float @llvm.fmuladd.f32(float %252, float %253, float %251) #13
>  %255 = getelementptr inbounds [129 x <2 x float>], [129 x <2 x float>] addrspace(4)* @LOG2_TBL, i32 0, i32 %248
>  %256 = load <2 x float>, <2 x float> addrspace(4)* %255, align 8, !tbaa !14
>  %257 = extractelement <2 x float> %256, i32 0
>  %258 = fadd float %257, %237
>  %259 = extractelement <2 x float> %256, i32 1
>  %260 = tail call float @llvm.fmuladd.f32(float %254, float 0xBFF7154760000000, float %259) #13
>  %261 = fadd float %258, %260
>  %262 = select i1 %207, float %225, float %261
>  %263 = icmp ugt i32 %204, 2139095039
>  %264 = select i1 %263, float %mul54, float %262
>  %265 = icmp eq i32 %204, %203
>  %266 = icmp eq i32 %204, 0
>  %.op1.i159 = fmul float %264, 0x3FE62E4300000000
>  %.op.i160 = select i1 %265, float %.op1.i159, float 0x7FF8000000000000
>  %267 = select i1 %266, float 0xFFF0000000000000, float %.op.i160
>  %neg = fsub float -0.000000e+00, %mul53
>  %268 = tail call float @llvm.fmuladd.f32(float %neg, float %267, float %pixel)
>  %add57 = fadd float %268, 5.000000e-01
>  br label %sw.epilog
>
>sw.bb59:                                          ; preds = %entry
>  %cmp60 = fcmp ogt float %div.i, 0.000000e+00
>  br i1 %cmp60, label %if.then61, label %if.end65
>
>if.then61:                                        ; preds = %sw.bb59
>  %269 = bitcast float %div.i to i32
>  %270 = and i32 %269, 2147483647
>  %271 = fadd float %div.i, -1.000000e+00
>  %272 = tail call float @llvm.fabs.f32(float %271) #14
>  %273 = fcmp olt float %272, 6.250000e-02
>  %274 = fadd float %271, 2.000000e+00
>  %275 = fdiv float %271, %274, !fpmath !11
>  %276 = fmul float %271, %275
>  %277 = fadd float %275, %275
>  %278 = fmul float %277, %277
>  %279 = tail call float @llvm.fmuladd.f32(float %278, float 0x3F899999A0000000, float 0x3FB5555560000000) #13
>  %280 = fmul float %278, %279
>  %281 = fsub float -0.000000e+00, %276
>  %282 = tail call float @llvm.fmuladd.f32(float %277, float %280, float %281) #13
>  %283 = bitcast float %271 to i32
>  %284 = and i32 %283, -65536
>  %285 = bitcast i32 %284 to float
>  %286 = fsub float %271, %285
>  %287 = fadd float %286, %282
>  %288 = fmul float %287, 0x3F75476520000000
>  %289 = tail call float @llvm.fmuladd.f32(float %285, float 0x3F75476520000000, float %288) #13
>  %290 = tail call float @llvm.fmuladd.f32(float %287, float 1.437500e+00, float %289) #13
>  %291 = tail call float @llvm.fmuladd.f32(float %285, float 1.437500e+00, float %290) #13
>  %292 = lshr i32 %269, 23
>  %293 = add nsw i32 %292, -127
>  %294 = or i32 %269, 1065353216
>  %295 = bitcast i32 %294 to float
>  %296 = fadd float %295, -1.000000e+00
>  %297 = bitcast float %296 to i32
>  %298 = lshr i32 %297, 23
>  %299 = add nsw i32 %298, -253
>  %300 = icmp eq i32 %292, 0
>  %301 = select i1 %300, i32 %299, i32 %293
>  %302 = select i1 %300, i32 %297, i32 %269
>  %303 = sitofp i32 %301 to float
>  %304 = and i32 %302, 8323072
>  %305 = shl i32 %302, 1
>  %306 = and i32 %305, 65536
>  %307 = add nuw nsw i32 %306, %304
>  %308 = or i32 %307, 1056964608
>  %309 = bitcast i32 %308 to float
>  %310 = and i32 %302, 8388607
>  %311 = or i32 %310, 1056964608
>  %312 = bitcast i32 %311 to float
>  %313 = fsub float %309, %312
>  %314 = lshr exact i32 %307, 16
>  %315 = getelementptr inbounds [129 x float], [129 x float] addrspace(4)* @LOG_INV_TBL, i32 0, i32 %314
>  %316 = load float, float addrspace(4)* %315, align 4, !tbaa !12
>  %317 = fmul float %313, %316
>  %318 = tail call float @llvm.fmuladd.f32(float %317, float 0x3FD5555560000000, float 5.000000e-01) #13
>  %319 = fmul float %317, %317
>  %320 = tail call float @llvm.fmuladd.f32(float %318, float %319, float %317) #13
>  %321 = getelementptr inbounds [129 x <2 x float>], [129 x <2 x float>] addrspace(4)* @LOG2_TBL, i32 0, i32 %314
>  %322 = load <2 x float>, <2 x float> addrspace(4)* %321, align 8, !tbaa !14
>  %323 = extractelement <2 x float> %322, i32 0
>  %324 = fadd float %323, %303
>  %325 = extractelement <2 x float> %322, i32 1
>  %326 = tail call float @llvm.fmuladd.f32(float %320, float 0xBFF7154760000000, float %325) #13
>  %327 = fadd float %324, %326
>  %328 = select i1 %273, float %291, float %327
>  %329 = icmp ugt i32 %270, 2139095039
>  %330 = select i1 %329, float %div.i, float %328
>  %331 = icmp eq i32 %270, %269
>  %332 = icmp eq i32 %270, 0
>  %.op1.i = fmul float %330, 0x3FE62E4300000000
>  %.op1.i.op = fmul float %.op1.i, -2.000000e+00
>  %.op.i.op = select i1 %331, float %.op1.i.op, float 0x7FF8000000000000
>  %mul63 = select i1 %332, float 0x7FF0000000000000, float %.op.i.op
>  %333 = fcmp olt float %mul63, 0.000000e+00
>  %334 = tail call float @llvm.sqrt.f32(float %mul63) #14
>  %335 = select i1 %333, float 0x7FF8000000000000, float %334
>  br label %if.end65
>
>if.end65:                                         ; preds = %if.then61, %sw.bb59
>  %sigma.0 = phi float [ %335, %if.then61 ], [ 1.000000e+00, %sw.bb59 ]
>  %xor.i.i151 = xor i32 %add1.i.i.i, %add.i.i.i
>  %mul.i.i.i152 = mul i32 %add.i.i.i, -83941
>  %add.i.i.i153 = add i32 %add1.i.i.i, %mul.i.i.i152
>  %cmp.i.i.i154 = icmp ult i32 %add.i.i.i153, %add1.i.i.i
>  %conv.i.i.i155 = zext i1 %cmp.i.i.i154 to i32
>  %336 = zext i32 %add.i.i.i to i64
>  %337 = mul nuw i64 %336, 4294883355
>  %338 = lshr i64 %337, 32
>  %339 = trunc i64 %338 to i32
>  %add1.i.i.i156 = add i32 %conv.i.i.i155, %339
>  store i32 %add.i.i.i153, i32 addrspace(5)* %x.i.i, align 4, !tbaa !8
>  store i32 %add1.i.i.i156, i32 addrspace(5)* %c.i.i, align 4, !tbaa !10
>  %conv.i157 = uitofp i32 %xor.i.i151 to float
>  %div.i158 = fmul float %conv.i157, 0x3DF0000000000000
>  %mul67 = fmul float %attenuate, 5.000000e-01
>  %mul68 = fmul float %mul67, %pixel
>  %mul69 = fmul float %mul68, %sigma.0
>  %mul70 = fmul float %div.i158, 2.000000e+00
>  %340 = bitcast float %mul70 to i32
>  %341 = and i32 %340, 2147483647
>  %342 = bitcast i32 %341 to float
>  %343 = fptosi float %342 to i32
>  %344 = sitofp i32 %343 to float
>  %345 = fsub float %342, %344
>  %346 = shl i32 %343, 31
>  %347 = icmp ult i32 %341, 2139095040
>  %348 = select i1 %347, i32 1065353216, i32 2143289344
>  %349 = icmp ult i32 %341, 1266679808
>  %350 = or i32 %346, 1065353216
>  %351 = select i1 %349, i32 %350, i32 %348
>  %352 = fsub float 1.000000e+00, %345
>  %353 = xor i32 %346, -2147483648
>  %354 = fcmp ole float %345, 7.500000e-01
>  %355 = fadd float %345, -5.000000e-01
>  %356 = select i1 %354, float %355, float %352
>  %357 = xor i1 %354, true
>  %358 = fcmp olt float %345, 5.000000e-01
>  %359 = fsub float 5.000000e-01, %345
>  %360 = select i1 %358, float %359, float %356
>  %361 = select i1 %358, i32 %346, i32 %353
>  %362 = fcmp ole float %345, 2.500000e-01
>  %363 = select i1 %362, float %345, float %360
>  %364 = or i1 %362, %357
>  %365 = fmul float %363, 0x400921FB60000000
>  %366 = fmul float %365, %365
>  %367 = fmul float %365, %366
>  %368 = tail call float @llvm.fmuladd.f32(float %366, float 0x3EC6DBE4A0000000, float 0xBF2A013A80000000) #13
>  %369 = tail call float @llvm.fmuladd.f32(float %366, float %368, float 0x3F811110E0000000) #13
>  %370 = tail call float @llvm.fmuladd.f32(float %366, float %369, float 0xBFC5555560000000) #13
>  %371 = tail call float @llvm.fmuladd.f32(float %367, float %370, float %365) #13
>  %372 = fmul float %366, %366
>  %373 = tail call float @llvm.fmuladd.f32(float %366, float 0xBE92524740000000, float 0x3EFA015C60000000) #13
>  %374 = tail call float @llvm.fmuladd.f32(float %366, float %373, float 0xBF56C16C00000000) #13
>  %375 = tail call float @llvm.fmuladd.f32(float %366, float %374, float 0x3FA5555560000000) #13
>  %376 = tail call float @llvm.fmuladd.f32(float %366, float -5.000000e-01, float 1.000000e+00) #13
>  %377 = tail call float @llvm.fmuladd.f32(float %372, float %375, float %376) #13
>  %378 = select i1 %364, float %377, float %371
>  %379 = bitcast float %378 to i32
>  %380 = xor i32 %361, %379
>  %381 = icmp ult i32 %341, 1258291200
>  %382 = select i1 %381, i32 %380, i32 %351
>  %383 = bitcast i32 %382 to float
>  %mul72 = fmul float %mul69, %383
>  %div73 = fmul float %mul72, 5.000000e-01
>  %add74 = fadd float %div73, %pixel
>  br label %sw.epilog
>
>sw.bb75:                                          ; preds = %entry
>  %mul76 = fmul float %attenuate, 1.250000e+01
>  %mul78 = fmul float %mul76, 0xBEEF751040000000
>  %mul79 = fmul float %mul78, %pixel
>  %384 = fcmp olt float %mul79, 0.000000e+00
>  %385 = select i1 %384, float -5.000000e-01, float 5.000000e-01
>  %386 = tail call float @llvm.fmuladd.f32(float %mul79, float 0x3FF7154760000000, float %385) #13
>  %387 = fptosi float %386 to i32
>  %388 = sitofp i32 %387 to float
>  %389 = tail call float @llvm.fmuladd.f32(float %388, float 0xBFE62E3000000000, float %mul79) #13
>  %390 = fmul float %388, 0xBEE2FEFA20000000
>  %391 = fadd float %389, %390
>  %392 = fmul float %391, %391
>  %393 = tail call float @llvm.fmuladd.f32(float %392, float 0x3E66376980000000, float 0xBEBBBD41C0000000) #13
>  %394 = tail call float @llvm.fmuladd.f32(float %392, float %393, float 0x3F11566AA0000000) #13
>  %395 = tail call float @llvm.fmuladd.f32(float %392, float %394, float 0xBF66C16C20000000) #13
>  %396 = tail call float @llvm.fmuladd.f32(float %392, float %395, float 0x3FC5555560000000) #13
>  %397 = fsub float -0.000000e+00, %396
>  %398 = tail call float @llvm.fmuladd.f32(float %392, float %397, float %391) #13
>  %399 = fsub float -0.000000e+00, %390
>  %400 = fmul float %391, %398
>  %401 = fsub float 2.000000e+00, %398
>  %402 = fdiv float %400, %401, !fpmath !11
>  %403 = fsub float %399, %402
>  %404 = fsub float %403, %389
>  %405 = fsub float 1.000000e+00, %404
>  %406 = bitcast float %405 to i32
>  %407 = shl i32 %387, 23
>  %408 = add nsw i32 %407, %406
>  %409 = bitcast i32 %408 to float
>  %410 = fcmp olt float %mul79, 0xC055D589E0000000
>  %411 = select i1 %410, float 0.000000e+00, float %409
>  %412 = fcmp olt float %mul79, 0x40562E4300000000
>  %413 = select i1 %412, float %411, float 0x7FF0000000000000
>  %414 = fcmp ord float %mul79, 0.000000e+00
>  %415 = select i1 %414, float %413, float %mul79
>  %cmp81175 = fcmp ogt float %div.i, %415
>  br i1 %cmp81175, label %for.body, label %for.end
>
>for.body:                                         ; preds = %sw.bb75, %for.body
>  %416 = phi i32 [ %add1.i.i.i146, %for.body ], [ %add1.i.i.i, %sw.bb75 ]
>  %417 = phi i32 [ %add.i.i.i143, %for.body ], [ %add.i.i.i, %sw.bb75 ]
>  %i.0177 = phi i32 [ %inc, %for.body ], [ 0, %sw.bb75 ]
>  %alpha.1176 = phi float [ %mul83, %for.body ], [ %div.i, %sw.bb75 ]
>  %xor.i.i141 = xor i32 %416, %417
>  %mul.i.i.i142 = mul i32 %417, -83941
>  %add.i.i.i143 = add i32 %mul.i.i.i142, %416
>  %cmp.i.i.i144 = icmp ult i32 %add.i.i.i143, %416
>  %conv.i.i.i145 = zext i1 %cmp.i.i.i144 to i32
>  %418 = zext i32 %417 to i64
>  %419 = mul nuw i64 %418, 4294883355
>  %420 = lshr i64 %419, 32
>  %421 = trunc i64 %420 to i32
>  %add1.i.i.i146 = add i32 %421, %conv.i.i.i145
>  %conv.i147 = uitofp i32 %xor.i.i141 to float
>  %div.i148 = fmul float %conv.i147, 0x3DF0000000000000
>  %mul83 = fmul float %alpha.1176, %div.i148
>  %inc = add i32 %i.0177, 1
>  %cmp81 = fcmp ogt float %mul83, %415
>  br i1 %cmp81, label %for.body, label %for.cond.for.end_crit_edge
>
>for.cond.for.end_crit_edge:                       ; preds = %for.body
>  store i32 %add.i.i.i143, i32 addrspace(5)* %x.i.i, align 4, !tbaa !8
>  store i32 %add1.i.i.i146, i32 addrspace(5)* %c.i.i, align 4, !tbaa !10
>  %phitmp = uitofp i32 %inc to float
>  %phitmp178 = fmul float %phitmp, 6.553500e+04
>  br label %for.end
>
>for.end:                                          ; preds = %for.cond.for.end_crit_edge, %sw.bb75
>  %i.0.lcssa = phi float [ %phitmp178, %for.cond.for.end_crit_edge ], [ 0.000000e+00, %sw.bb75 ]
>  %div86 = fdiv float %i.0.lcssa, %mul76, !fpmath !11
>  br label %sw.epilog
>
>sw.bb87:                                          ; preds = %entry
>  %mul88 = fmul float %attenuate, 6.553500e+04
>  %mul89 = fmul float %mul88, %div.i
>  br label %sw.epilog
>
>sw.epilog:                                        ; preds = %if.else, %sw.bb21, %if.then49, %if.else51, %if.then37, %if.else39, %sw.bb87, %for.end, %if.end65, %sw.bb3, %sw.default
>  %noise.0 = phi float [ %6, %sw.default ], [ %div86, %for.end ], [ %add74, %if.end65 ], [ %sub38, %if.then37 ], [ %add, %if.else39 ], [ %add50, %if.then49 ], [ %add57, %if.else51 ], [ %136, %sw.bb3 ], [ %mul89, %sw.bb87 ], [ 0.000000e+00, %sw.bb21 ], [ %pixel., %if.else ]
>  ret float %noise.0
>}
>
>; Function Attrs: nounwind readnone speculatable
>declare float @llvm.fmuladd.f32(float, float, float) #5
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @AddNoise(float addrspace(1)* nocapture readonly %image, i32 %number_channels, i32 %channel, i32 %length, i32 %pixelsPerWorkItem, i32 %noise_type, float %attenuate, i32 %seed0, i32 %seed1, i32 %numRandomNumbersPerPixel, float addrspace(1)* nocapture %filteredImage) local_unnamed_addr #4 !kernel_arg_addr_space !15 !kernel_arg_access_qual !16 !kernel_arg_type !17 !kernel_arg_base_type !17 !kernel_arg_type_qual !18 {
>entry:
>  %rng = alloca %struct.mwc64x_state_t, align 4, addrspace(5)
>  %0 = bitcast %struct.mwc64x_state_t addrspace(5)* %rng to i8 addrspace(5)*
>  call void @llvm.lifetime.start.p5i8(i64 8, i8 addrspace(5)* %0) #13
>  %x = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %rng, i32 0, i32 0
>  store i32 %seed0, i32 addrspace(5)* %x, align 4, !tbaa !8
>  %c = getelementptr inbounds %struct.mwc64x_state_t, %struct.mwc64x_state_t addrspace(5)* %rng, i32 0, i32 1
>  store i32 %seed1, i32 addrspace(5)* %c, align 4, !tbaa !10
>  %mul = mul i32 %numRandomNumbersPerPixel, %pixelsPerWorkItem
>  %x.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %mul1 = mul i32 %1, %x.i
>  %mul3 = mul i32 %mul1, %mul
>  %conv = zext i32 %mul3 to i64
>  %conv4 = zext i32 %mul to i64
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %mul1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %conv.i.i = zext i32 %8 to i64
>  %mul1.i.i = mul nuw i64 %conv.i.i, %conv4
>  %add2.i.i = add i64 %mul1.i.i, %conv
>  %cmp38.i.i.i = icmp eq i64 %add2.i.i, 0
>  br i1 %cmp38.i.i.i, label %while.body.i.i.i.preheader, label %while.body.i17.i.i
>
>while.body.i17.i.i:                               ; preds = %entry, %MWC_MulMod64.exit37.i.i.i
>  %acc.041.i.i.i = phi i64 [ %acc.1.i.i.i, %MWC_MulMod64.exit37.i.i.i ], [ 1, %entry ]
>  %sqr.040.i.i.i = phi i64 [ %r.0.lcssa.i36.i.i.i, %MWC_MulMod64.exit37.i.i.i ], [ 4294883355, %entry ]
>  %e.addr.039.i.i.i = phi i64 [ %shr.i22.i.i, %MWC_MulMod64.exit37.i.i.i ], [ %add2.i.i, %entry ]
>  %and.i15.i.i = and i64 %e.addr.039.i.i.i, 1
>  %tobool.i16.i.i = icmp eq i64 %and.i15.i.i, 0
>  br i1 %tobool.i16.i.i, label %if.end.i21.i.i, label %if.then.i18.i.i
>
>if.then.i18.i.i:                                  ; preds = %while.body.i17.i.i
>  %cmp17.i.i.i.i = icmp eq i64 %acc.041.i.i.i, 0
>  br i1 %cmp17.i.i.i.i, label %if.end.i21.i.i, label %while.body.i.i.i.i
>
>while.body.i.i.i.i:                               ; preds = %if.then.i18.i.i, %MWC_AddMod64.exit16.i.i.i.i
>  %r.020.i.i.i.i = phi i64 [ %r.1.i.i.i.i, %MWC_AddMod64.exit16.i.i.i.i ], [ 0, %if.then.i18.i.i ]
>  %a.addr.019.i.i.i.i = phi i64 [ %shr.i.i.i.i, %MWC_AddMod64.exit16.i.i.i.i ], [ %acc.041.i.i.i, %if.then.i18.i.i ]
>  %b.addr.018.i.i.i.i = phi i64 [ %v.0.i15.i.i.i.i, %MWC_AddMod64.exit16.i.i.i.i ], [ %sqr.040.i.i.i, %if.then.i18.i.i ]
>  %and.i.i.i.i = and i64 %a.addr.019.i.i.i.i, 1
>  %tobool.i.i.i.i = icmp eq i64 %and.i.i.i.i, 0
>  br i1 %tobool.i.i.i.i, label %if.end.i.i.i.i, label %if.then.i.i19.i.i
>
>if.then.i.i19.i.i:                                ; preds = %while.body.i.i.i.i
>  %add.i.i.i.i.i = add i64 %b.addr.018.i.i.i.i, %r.020.i.i.i.i
>  %cmp.i.i.i.i.i = icmp ult i64 %add.i.i.i.i.i, -360523849793537
>  br i1 %cmp.i.i.i.i.i, label %lor.lhs.false.i.i.i.i.i, label %if.then.i.i.i.i.i
>
>lor.lhs.false.i.i.i.i.i:                          ; preds = %if.then.i.i19.i.i
>  %9 = uitofp i64 %add.i.i.i.i.i to float
>  %10 = uitofp i64 %r.020.i.i.i.i to float
>  %cmp2.i.i.i.i.i = fcmp olt float %9, %10
>  br i1 %cmp2.i.i.i.i.i, label %if.then.i.i.i.i.i, label %if.end.i.i.i.i
>
>if.then.i.i.i.i.i:                                ; preds = %lor.lhs.false.i.i.i.i.i, %if.then.i.i19.i.i
>  %sub.i.i.i.i.i = add i64 %add.i.i.i.i.i, 360523849793537
>  br label %if.end.i.i.i.i
>
>if.end.i.i.i.i:                                   ; preds = %if.then.i.i.i.i.i, %lor.lhs.false.i.i.i.i.i, %while.body.i.i.i.i
>  %r.1.i.i.i.i = phi i64 [ %r.020.i.i.i.i, %while.body.i.i.i.i ], [ %sub.i.i.i.i.i, %if.then.i.i.i.i.i ], [ %add.i.i.i.i.i, %lor.lhs.false.i.i.i.i.i ]
>  %add.i9.i.i.i.i = shl i64 %b.addr.018.i.i.i.i, 1
>  %cmp.i10.i.i.i.i = icmp ult i64 %add.i9.i.i.i.i, -360523849793537
>  br i1 %cmp.i10.i.i.i.i, label %lor.lhs.false.i12.i.i.i.i, label %if.then.i14.i.i.i.i
>
>lor.lhs.false.i12.i.i.i.i:                        ; preds = %if.end.i.i.i.i
>  %11 = uitofp i64 %add.i9.i.i.i.i to float
>  %12 = uitofp i64 %b.addr.018.i.i.i.i to float
>  %cmp2.i11.i.i.i.i = fcmp olt float %11, %12
>  br i1 %cmp2.i11.i.i.i.i, label %if.then.i14.i.i.i.i, label %MWC_AddMod64.exit16.i.i.i.i
>
>if.then.i14.i.i.i.i:                              ; preds = %lor.lhs.false.i12.i.i.i.i, %if.end.i.i.i.i
>  %sub.i13.i.i.i.i = add i64 %add.i9.i.i.i.i, 360523849793537
>  br label %MWC_AddMod64.exit16.i.i.i.i
>
>MWC_AddMod64.exit16.i.i.i.i:                      ; preds = %if.then.i14.i.i.i.i, %lor.lhs.false.i12.i.i.i.i
>  %v.0.i15.i.i.i.i = phi i64 [ %sub.i13.i.i.i.i, %if.then.i14.i.i.i.i ], [ %add.i9.i.i.i.i, %lor.lhs.false.i12.i.i.i.i ]
>  %shr.i.i.i.i = lshr i64 %a.addr.019.i.i.i.i, 1
>  %cmp.i.i20.i.i = icmp eq i64 %shr.i.i.i.i, 0
>  br i1 %cmp.i.i20.i.i, label %if.end.i21.i.i, label %while.body.i.i.i.i
>
>if.end.i21.i.i:                                   ; preds = %MWC_AddMod64.exit16.i.i.i.i, %if.then.i18.i.i, %while.body.i17.i.i
>  %acc.1.i.i.i = phi i64 [ %acc.041.i.i.i, %while.body.i17.i.i ], [ 0, %if.then.i18.i.i ], [ %r.1.i.i.i.i, %MWC_AddMod64.exit16.i.i.i.i ]
>  %cmp17.i10.i.i.i = icmp eq i64 %sqr.040.i.i.i, 0
>  br i1 %cmp17.i10.i.i.i, label %MWC_MulMod64.exit37.i.i.i, label %while.body.i16.i.i.i
>
>while.body.i16.i.i.i:                             ; preds = %if.end.i21.i.i, %MWC_AddMod64.exit16.i35.i.i.i
>  %r.020.i11.i.i.i = phi i64 [ %r.1.i24.i.i.i, %MWC_AddMod64.exit16.i35.i.i.i ], [ 0, %if.end.i21.i.i ]
>  %a.addr.019.i12.i.i.i = phi i64 [ %shr.i33.i.i.i, %MWC_AddMod64.exit16.i35.i.i.i ], [ %sqr.040.i.i.i, %if.end.i21.i.i ]
>  %b.addr.018.i13.i.i.i = phi i64 [ %v.0.i15.i32.i.i.i, %MWC_AddMod64.exit16.i35.i.i.i ], [ %sqr.040.i.i.i, %if.end.i21.i.i ]
>  %and.i14.i.i.i = and i64 %a.addr.019.i12.i.i.i, 1
>  %tobool.i15.i.i.i = icmp eq i64 %and.i14.i.i.i, 0
>  br i1 %tobool.i15.i.i.i, label %if.end.i27.i.i.i, label %if.then.i19.i.i.i
>
>if.then.i19.i.i.i:                                ; preds = %while.body.i16.i.i.i
>  %add.i.i17.i.i.i = add i64 %b.addr.018.i13.i.i.i, %r.020.i11.i.i.i
>  %cmp.i.i18.i.i.i = icmp ult i64 %add.i.i17.i.i.i, -360523849793537
>  br i1 %cmp.i.i18.i.i.i, label %lor.lhs.false.i.i21.i.i.i, label %if.then.i.i23.i.i.i
>
>lor.lhs.false.i.i21.i.i.i:                        ; preds = %if.then.i19.i.i.i
>  %13 = uitofp i64 %add.i.i17.i.i.i to float
>  %14 = uitofp i64 %r.020.i11.i.i.i to float
>  %cmp2.i.i20.i.i.i = fcmp olt float %13, %14
>  br i1 %cmp2.i.i20.i.i.i, label %if.then.i.i23.i.i.i, label %if.end.i27.i.i.i
>
>if.then.i.i23.i.i.i:                              ; preds = %lor.lhs.false.i.i21.i.i.i, %if.then.i19.i.i.i
>  %sub.i.i22.i.i.i = add i64 %add.i.i17.i.i.i, 360523849793537
>  br label %if.end.i27.i.i.i
>
>if.end.i27.i.i.i:                                 ; preds = %if.then.i.i23.i.i.i, %lor.lhs.false.i.i21.i.i.i, %while.body.i16.i.i.i
>  %r.1.i24.i.i.i = phi i64 [ %r.020.i11.i.i.i, %while.body.i16.i.i.i ], [ %sub.i.i22.i.i.i, %if.then.i.i23.i.i.i ], [ %add.i.i17.i.i.i, %lor.lhs.false.i.i21.i.i.i ]
>  %add.i9.i25.i.i.i = shl i64 %b.addr.018.i13.i.i.i, 1
>  %cmp.i10.i26.i.i.i = icmp ult i64 %add.i9.i25.i.i.i, -360523849793537
>  br i1 %cmp.i10.i26.i.i.i, label %lor.lhs.false.i12.i29.i.i.i, label %if.then.i14.i31.i.i.i
>
>lor.lhs.false.i12.i29.i.i.i:                      ; preds = %if.end.i27.i.i.i
>  %15 = uitofp i64 %add.i9.i25.i.i.i to float
>  %16 = uitofp i64 %b.addr.018.i13.i.i.i to float
>  %cmp2.i11.i28.i.i.i = fcmp olt float %15, %16
>  br i1 %cmp2.i11.i28.i.i.i, label %if.then.i14.i31.i.i.i, label %MWC_AddMod64.exit16.i35.i.i.i
>
>if.then.i14.i31.i.i.i:                            ; preds = %lor.lhs.false.i12.i29.i.i.i, %if.end.i27.i.i.i
>  %sub.i13.i30.i.i.i = add i64 %add.i9.i25.i.i.i, 360523849793537
>  br label %MWC_AddMod64.exit16.i35.i.i.i
>
>MWC_AddMod64.exit16.i35.i.i.i:                    ; preds = %if.then.i14.i31.i.i.i, %lor.lhs.false.i12.i29.i.i.i
>  %v.0.i15.i32.i.i.i = phi i64 [ %sub.i13.i30.i.i.i, %if.then.i14.i31.i.i.i ], [ %add.i9.i25.i.i.i, %lor.lhs.false.i12.i29.i.i.i ]
>  %shr.i33.i.i.i = lshr i64 %a.addr.019.i12.i.i.i, 1
>  %cmp.i34.i.i.i = icmp eq i64 %shr.i33.i.i.i, 0
>  br i1 %cmp.i34.i.i.i, label %MWC_MulMod64.exit37.i.i.i, label %while.body.i16.i.i.i
>
>MWC_MulMod64.exit37.i.i.i:                        ; preds = %MWC_AddMod64.exit16.i35.i.i.i, %if.end.i21.i.i
>  %r.0.lcssa.i36.i.i.i = phi i64 [ 0, %if.end.i21.i.i ], [ %r.1.i24.i.i.i, %MWC_AddMod64.exit16.i35.i.i.i ]
>  %shr.i22.i.i = lshr i64 %e.addr.039.i.i.i, 1
>  %cmp.i23.i.i = icmp eq i64 %shr.i22.i.i, 0
>  br i1 %cmp.i23.i.i, label %while.body.i.i.i.preheader, label %while.body.i17.i.i
>
>while.body.i.i.i.preheader:                       ; preds = %MWC_MulMod64.exit37.i.i.i, %entry
>  %b.addr.018.i.i.i.ph = phi i64 [ 1, %entry ], [ %acc.1.i.i.i, %MWC_MulMod64.exit37.i.i.i ]
>  br label %while.body.i.i.i
>
>while.body.i.i.i:                                 ; preds = %MWC_AddMod64.exit16.i.i.i.1, %while.body.i.i.i.preheader
>  %r.020.i.i.i = phi i64 [ 0, %while.body.i.i.i.preheader ], [ %r.1.i.i.i.1, %MWC_AddMod64.exit16.i.i.i.1 ]
>  %a.addr.019.i.i.i = phi i64 [ 4077358422479273989, %while.body.i.i.i.preheader ], [ %shr.i.i.i, %MWC_AddMod64.exit16.i.i.i.1 ]
>  %b.addr.018.i.i.i = phi i64 [ %b.addr.018.i.i.i.ph, %while.body.i.i.i.preheader ], [ %v.0.i15.i.i.i.1, %MWC_AddMod64.exit16.i.i.i.1 ]
>  %and.i.i.i = and i64 %a.addr.019.i.i.i, 1
>  %tobool.i.i.i = icmp eq i64 %and.i.i.i, 0
>  br i1 %tobool.i.i.i, label %if.end.i.i.i, label %if.then.i.i.i
>
>if.then.i.i.i:                                    ; preds = %while.body.i.i.i
>  %add.i.i.i.i = add i64 %b.addr.018.i.i.i, %r.020.i.i.i
>  %cmp.i.i.i.i = icmp ult i64 %add.i.i.i.i, -360523849793537
>  br i1 %cmp.i.i.i.i, label %lor.lhs.false.i.i.i.i, label %if.then.i.i.i.i
>
>lor.lhs.false.i.i.i.i:                            ; preds = %if.then.i.i.i
>  %17 = uitofp i64 %add.i.i.i.i to float
>  %18 = uitofp i64 %r.020.i.i.i to float
>  %cmp2.i.i.i.i = fcmp olt float %17, %18
>  br i1 %cmp2.i.i.i.i, label %if.then.i.i.i.i, label %if.end.i.i.i
>
>if.then.i.i.i.i:                                  ; preds = %lor.lhs.false.i.i.i.i, %if.then.i.i.i
>  %sub.i.i.i.i = add i64 %add.i.i.i.i, 360523849793537
>  br label %if.end.i.i.i
>
>if.end.i.i.i:                                     ; preds = %if.then.i.i.i.i, %lor.lhs.false.i.i.i.i, %while.body.i.i.i
>  %r.1.i.i.i = phi i64 [ %r.020.i.i.i, %while.body.i.i.i ], [ %sub.i.i.i.i, %if.then.i.i.i.i ], [ %add.i.i.i.i, %lor.lhs.false.i.i.i.i ]
>  %add.i9.i.i.i = shl i64 %b.addr.018.i.i.i, 1
>  %cmp.i10.i.i.i = icmp ult i64 %add.i9.i.i.i, -360523849793537
>  br i1 %cmp.i10.i.i.i, label %lor.lhs.false.i12.i.i.i, label %if.then.i14.i.i.i
>
>lor.lhs.false.i12.i.i.i:                          ; preds = %if.end.i.i.i
>  %19 = uitofp i64 %add.i9.i.i.i to float
>  %20 = uitofp i64 %b.addr.018.i.i.i to float
>  %cmp2.i11.i.i.i = fcmp olt float %19, %20
>  br i1 %cmp2.i11.i.i.i, label %if.then.i14.i.i.i, label %MWC_AddMod64.exit16.i.i.i
>
>if.then.i14.i.i.i:                                ; preds = %lor.lhs.false.i12.i.i.i, %if.end.i.i.i
>  %sub.i13.i.i.i = add i64 %add.i9.i.i.i, 360523849793537
>  br label %MWC_AddMod64.exit16.i.i.i
>
>MWC_AddMod64.exit16.i.i.i:                        ; preds = %if.then.i14.i.i.i, %lor.lhs.false.i12.i.i.i
>  %v.0.i15.i.i.i = phi i64 [ %sub.i13.i.i.i, %if.then.i14.i.i.i ], [ %add.i9.i.i.i, %lor.lhs.false.i12.i.i.i ]
>  %shr.i.i.i = lshr i64 %a.addr.019.i.i.i, 2
>  %and.i.i.i.1 = and i64 %a.addr.019.i.i.i, 2
>  %tobool.i.i.i.1 = icmp eq i64 %and.i.i.i.1, 0
>  br i1 %tobool.i.i.i.1, label %if.end.i.i.i.1, label %if.then.i.i.i.1
>
>MWC64X_SeedStreams.exit:                          ; preds = %MWC_AddMod64.exit16.i.i.i.1
>  %div.i.i = udiv i64 %r.1.i.i.i.1, 4294883355
>  %conv5.i.i = trunc i64 %div.i.i to i32
>  %21 = mul i64 %div.i.i, 4294883355
>  %22 = sub i64 %r.1.i.i.i.1, %21
>  %conv6.i.i = trunc i64 %22 to i32
>  store i32 %conv5.i.i, i32 addrspace(5)* %x, align 4, !tbaa !8
>  store i32 %conv6.i.i, i32 addrspace(5)* %c, align 4, !tbaa !10
>  %mul8 = mul i32 %mul1, %pixelsPerWorkItem
>  %mul979 = add i32 %mul8, %2
>  %add = mul i32 %mul979, %number_channels
>  %cmp110 = icmp ne i32 %pixelsPerWorkItem, 0
>  %cmp13111 = icmp ult i32 %add, %length
>  %or.cond80112 = and i1 %cmp110, %cmp13111
>  br i1 %or.cond80112, label %while.body.lr.ph, label %while.end
>
>while.body.lr.ph:                                 ; preds = %MWC64X_SeedStreams.exit
>  %and.i82 = and i32 %channel, 1
>  %cmp.i83 = icmp eq i32 %and.i82, 0
>  %cmp1.i85 = icmp ugt i32 %number_channels, 2
>  %and3.i87 = and i32 %channel, 2
>  %cmp4.i88 = icmp eq i32 %and3.i87, 0
>  %and8.i91 = and i32 %channel, 4
>  %cmp9.i92 = icmp eq i32 %and8.i91, 0
>  %and16.i96 = and i32 %channel, 16
>  %cmp17.i97 = icmp eq i32 %and16.i96, 0
>  %mul46 = mul i32 %x.i, %number_channels
>  br label %while.body
>
>while.body:                                       ; preds = %while.body.lr.ph, %WriteChannels.exit
>  %pos.0118 = phi i32 [ %add, %while.body.lr.ph ], [ %add47, %WriteChannels.exit ]
>  %count.0117 = phi i32 [ %pixelsPerWorkItem, %while.body.lr.ph ], [ %dec, %WriteChannels.exit ]
>  %alpha.sroa.0.0116 = phi i32 [ undef, %while.body.lr.ph ], [ %alpha.sroa.0.2, %WriteChannels.exit ]
>  %23 = phi float [ undef, %while.body.lr.ph ], [ %54, %WriteChannels.exit ]
>  %blue.sroa.0.0115 = phi i32 [ undef, %while.body.lr.ph ], [ %blue.sroa.0.2, %WriteChannels.exit ]
>  %24 = phi float [ undef, %while.body.lr.ph ], [ %52, %WriteChannels.exit ]
>  %green.sroa.0.0114 = phi i32 [ undef, %while.body.lr.ph ], [ %green.sroa.0.4, %WriteChannels.exit ]
>  %25 = phi float [ undef, %while.body.lr.ph ], [ %51, %WriteChannels.exit ]
>  %red.sroa.0.0113 = phi i32 [ undef, %while.body.lr.ph ], [ %red.sroa.0.2, %WriteChannels.exit ]
>  %26 = phi float [ undef, %while.body.lr.ph ], [ %45, %WriteChannels.exit ]
>  %add.ptr = getelementptr inbounds float, float addrspace(1)* %image, i32 %pos.0118
>  %add.ptr15 = getelementptr inbounds float, float addrspace(1)* %filteredImage, i32 %pos.0118
>  br i1 %cmp.i83, label %if.end.i86, label %if.then.i84
>
>if.then.i84:                                      ; preds = %while.body
>  %27 = bitcast float addrspace(1)* %add.ptr to i32 addrspace(1)*
>  %28 = load i32, i32 addrspace(1)* %27, align 4, !tbaa !12
>  %29 = bitcast i32 %28 to float
>  br label %if.end.i86
>
>if.end.i86:                                       ; preds = %if.then.i84, %while.body
>  %30 = phi float [ %26, %while.body ], [ %29, %if.then.i84 ]
>  %red.sroa.0.1 = phi i32 [ %red.sroa.0.0113, %while.body ], [ %28, %if.then.i84 ]
>  br i1 %cmp1.i85, label %if.then2.i89, label %if.end13.i95
>
>if.then2.i89:                                     ; preds = %if.end.i86
>  br i1 %cmp4.i88, label %if.end7.i93, label %if.then5.i90
>
>if.then5.i90:                                     ; preds = %if.then2.i89
>  %add.ptr.i31.i = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 1
>  %31 = bitcast float addrspace(1)* %add.ptr.i31.i to i32 addrspace(1)*
>  %32 = load i32, i32 addrspace(1)* %31, align 4, !tbaa !12
>  %33 = bitcast i32 %32 to float
>  br label %if.end7.i93
>
>if.end7.i93:                                      ; preds = %if.then5.i90, %if.then2.i89
>  %34 = phi float [ %25, %if.then2.i89 ], [ %33, %if.then5.i90 ]
>  %green.sroa.0.1 = phi i32 [ %green.sroa.0.0114, %if.then2.i89 ], [ %32, %if.then5.i90 ]
>  br i1 %cmp9.i92, label %if.end13.i95, label %if.then10.i94
>
>if.then10.i94:                                    ; preds = %if.end7.i93
>  %add.ptr.i30.i = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 2
>  %35 = bitcast float addrspace(1)* %add.ptr.i30.i to i32 addrspace(1)*
>  %36 = load i32, i32 addrspace(1)* %35, align 4, !tbaa !12
>  %37 = bitcast i32 %36 to float
>  br label %if.end13.i95
>
>if.end13.i95:                                     ; preds = %if.then10.i94, %if.end7.i93, %if.end.i86
>  %38 = phi float [ %34, %if.end7.i93 ], [ %34, %if.then10.i94 ], [ %25, %if.end.i86 ]
>  %green.sroa.0.2 = phi i32 [ %green.sroa.0.1, %if.end7.i93 ], [ %green.sroa.0.1, %if.then10.i94 ], [ %green.sroa.0.0114, %if.end.i86 ]
>  %39 = phi float [ %24, %if.end7.i93 ], [ %37, %if.then10.i94 ], [ %24, %if.end.i86 ]
>  %blue.sroa.0.1 = phi i32 [ %blue.sroa.0.0115, %if.end7.i93 ], [ %36, %if.then10.i94 ], [ %blue.sroa.0.0115, %if.end.i86 ]
>  switch i32 %number_channels, label %ReadChannels.exit [
>    i32 4, label %land.lhs.true.i98
>    i32 2, label %land.lhs.true.i98
>  ]
>
>land.lhs.true.i98:                                ; preds = %if.end13.i95, %if.end13.i95
>  br i1 %cmp17.i97, label %ReadChannels.exit, label %if.then18.i101
>
>if.then18.i101:                                   ; preds = %land.lhs.true.i98
>  %add.ptr.i.i99 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %number_channels
>  %add.ptr1.i.i100 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i99, i32 -1
>  %40 = bitcast float addrspace(1)* %add.ptr1.i.i100 to i32 addrspace(1)*
>  %41 = load i32, i32 addrspace(1)* %40, align 4, !tbaa !12
>  %42 = bitcast i32 %41 to float
>  br label %ReadChannels.exit
>
>ReadChannels.exit:                                ; preds = %if.end13.i95, %land.lhs.true.i98, %if.then18.i101
>  %43 = phi float [ %23, %if.end13.i95 ], [ %23, %land.lhs.true.i98 ], [ %42, %if.then18.i101 ]
>  %alpha.sroa.0.1 = phi i32 [ %alpha.sroa.0.0116, %if.end13.i95 ], [ %alpha.sroa.0.0116, %land.lhs.true.i98 ], [ %41, %if.then18.i101 ]
>  br i1 %cmp.i83, label %if.end, label %if.then
>
>if.then:                                          ; preds = %ReadChannels.exit
>  %call18 = call float @mwcGenerateDifferentialNoise(%struct.mwc64x_state_t addrspace(5)* %rng, float %30, i32 %noise_type, float %attenuate) #15
>  %44 = bitcast float %call18 to i32
>  br label %if.end
>
>if.end:                                           ; preds = %ReadChannels.exit, %if.then
>  %45 = phi float [ %30, %ReadChannels.exit ], [ %call18, %if.then ]
>  %red.sroa.0.2 = phi i32 [ %red.sroa.0.1, %ReadChannels.exit ], [ %44, %if.then ]
>  br i1 %cmp1.i85, label %if.then21, label %if.end34
>
>if.then21:                                        ; preds = %if.end
>  br i1 %cmp4.i88, label %if.end27, label %if.then25
>
>if.then25:                                        ; preds = %if.then21
>  %46 = bitcast i32 %green.sroa.0.2 to float
>  %call26 = call float @mwcGenerateDifferentialNoise(%struct.mwc64x_state_t addrspace(5)* %rng, float %46, i32 %noise_type, float %attenuate) #15
>  %47 = bitcast float %call26 to i32
>  br label %if.end27
>
>if.end27:                                         ; preds = %if.then21, %if.then25
>  %48 = phi float [ %38, %if.then21 ], [ %call26, %if.then25 ]
>  %green.sroa.0.3 = phi i32 [ %green.sroa.0.2, %if.then21 ], [ %47, %if.then25 ]
>  br i1 %cmp9.i92, label %if.end34, label %if.then31
>
>if.then31:                                        ; preds = %if.end27
>  %49 = bitcast i32 %blue.sroa.0.1 to float
>  %call32 = call float @mwcGenerateDifferentialNoise(%struct.mwc64x_state_t addrspace(5)* %rng, float %49, i32 %noise_type, float %attenuate) #15
>  %50 = bitcast float %call32 to i32
>  br label %if.end34
>
>if.end34:                                         ; preds = %if.end27, %if.then31, %if.end
>  %51 = phi float [ %48, %if.end27 ], [ %48, %if.then31 ], [ %38, %if.end ]
>  %green.sroa.0.4 = phi i32 [ %green.sroa.0.3, %if.end27 ], [ %green.sroa.0.3, %if.then31 ], [ %green.sroa.0.2, %if.end ]
>  %52 = phi float [ %39, %if.end27 ], [ %call32, %if.then31 ], [ %39, %if.end ]
>  %blue.sroa.0.2 = phi i32 [ %blue.sroa.0.1, %if.end27 ], [ %50, %if.then31 ], [ %blue.sroa.0.1, %if.end ]
>  switch i32 %number_channels, label %if.end44 [
>    i32 4, label %land.lhs.true
>    i32 2, label %land.lhs.true
>  ]
>
>land.lhs.true:                                    ; preds = %if.end34, %if.end34
>  br i1 %cmp17.i97, label %if.end44, label %if.then42
>
>if.then42:                                        ; preds = %land.lhs.true
>  %call43 = call float @mwcGenerateDifferentialNoise(%struct.mwc64x_state_t addrspace(5)* %rng, float %43, i32 %noise_type, float %attenuate) #15
>  %53 = bitcast float %call43 to i32
>  br label %if.end44
>
>if.end44:                                         ; preds = %if.end34, %land.lhs.true, %if.then42
>  %54 = phi float [ %43, %if.end34 ], [ %43, %land.lhs.true ], [ %call43, %if.then42 ]
>  %alpha.sroa.0.2 = phi i32 [ %alpha.sroa.0.1, %if.end34 ], [ %alpha.sroa.0.1, %land.lhs.true ], [ %53, %if.then42 ]
>  %55 = bitcast i32 %red.sroa.0.2 to float
>  %56 = bitcast i32 %alpha.sroa.0.2 to float
>  br i1 %cmp.i83, label %if.end.i, label %if.then.i
>
>if.then.i:                                        ; preds = %if.end44
>  %57 = fcmp ogt float %55, 6.553500e+04
>  %58 = fcmp olt float %55, 0.000000e+00
>  %59 = select i1 %58, float 0.000000e+00, float %55
>  %.op.i.i = fadd float %59, 5.000000e-01
>  %add.i.i = select i1 %57, float 6.553550e+04, float %.op.i.i
>  store float %add.i.i, float addrspace(1)* %add.ptr15, align 4, !tbaa !12
>  br label %if.end.i
>
>if.end.i:                                         ; preds = %if.then.i, %if.end44
>  br i1 %cmp1.i85, label %if.then2.i, label %if.end13.i
>
>if.then2.i:                                       ; preds = %if.end.i
>  br i1 %cmp4.i88, label %if.end7.i, label %if.then5.i
>
>if.then5.i:                                       ; preds = %if.then2.i
>  %60 = fcmp ogt float %51, 6.553500e+04
>  %61 = fcmp olt float %51, 0.000000e+00
>  %62 = select i1 %61, float 0.000000e+00, float %51
>  %.op.i36.i = fadd float %62, 5.000000e-01
>  %add.i37.i = select i1 %60, float 6.553550e+04, float %.op.i36.i
>  %add.ptr.i35.i = getelementptr inbounds float, float addrspace(1)* %add.ptr15, i32 1
>  store float %add.i37.i, float addrspace(1)* %add.ptr.i35.i, align 4, !tbaa !12
>  br label %if.end7.i
>
>if.end7.i:                                        ; preds = %if.then5.i, %if.then2.i
>  br i1 %cmp9.i92, label %if.end13.i, label %if.then10.i
>
>if.then10.i:                                      ; preds = %if.end7.i
>  %63 = fcmp ogt float %52, 6.553500e+04
>  %64 = fcmp olt float %52, 0.000000e+00
>  %65 = select i1 %64, float 0.000000e+00, float %52
>  %.op.i33.i = fadd float %65, 5.000000e-01
>  %add.i34.i = select i1 %63, float 6.553550e+04, float %.op.i33.i
>  %add.ptr.i32.i = getelementptr inbounds float, float addrspace(1)* %add.ptr15, i32 2
>  store float %add.i34.i, float addrspace(1)* %add.ptr.i32.i, align 4, !tbaa !12
>  br label %if.end13.i
>
>if.end13.i:                                       ; preds = %if.then10.i, %if.end7.i, %if.end.i
>  switch i32 %number_channels, label %WriteChannels.exit [
>    i32 4, label %land.lhs.true.i
>    i32 2, label %land.lhs.true.i
>  ]
>
>land.lhs.true.i:                                  ; preds = %if.end13.i, %if.end13.i
>  br i1 %cmp17.i97, label %WriteChannels.exit, label %if.then18.i
>
>if.then18.i:                                      ; preds = %land.lhs.true.i
>  %66 = fcmp ogt float %56, 6.553500e+04
>  %67 = fcmp olt float %56, 0.000000e+00
>  %68 = select i1 %67, float 0.000000e+00, float %56
>  %.op.i30.i = fadd float %68, 5.000000e-01
>  %add.i31.i = select i1 %66, float 6.553550e+04, float %.op.i30.i
>  %add.ptr.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr15, i32 %number_channels
>  %add.ptr1.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i, i32 -1
>  store float %add.i31.i, float addrspace(1)* %add.ptr1.i.i, align 4, !tbaa !12
>  br label %WriteChannels.exit
>
>WriteChannels.exit:                               ; preds = %if.end13.i, %land.lhs.true.i, %if.then18.i
>  %add47 = add i32 %pos.0118, %mul46
>  %dec = add i32 %count.0117, -1
>  %cmp = icmp ne i32 %dec, 0
>  %cmp13 = icmp ult i32 %add47, %length
>  %or.cond80 = and i1 %cmp, %cmp13
>  br i1 %or.cond80, label %while.body, label %while.end
>
>while.end:                                        ; preds = %WriteChannels.exit, %MWC64X_SeedStreams.exit
>  call void @llvm.lifetime.end.p5i8(i64 8, i8 addrspace(5)* %0) #13
>  ret void
>
>if.then.i.i.i.1:                                  ; preds = %MWC_AddMod64.exit16.i.i.i
>  %add.i.i.i.i.1 = add i64 %v.0.i15.i.i.i, %r.1.i.i.i
>  %cmp.i.i.i.i.1 = icmp ult i64 %add.i.i.i.i.1, -360523849793537
>  br i1 %cmp.i.i.i.i.1, label %lor.lhs.false.i.i.i.i.1, label %if.then.i.i.i.i.1
>
>lor.lhs.false.i.i.i.i.1:                          ; preds = %if.then.i.i.i.1
>  %69 = uitofp i64 %add.i.i.i.i.1 to float
>  %70 = uitofp i64 %r.1.i.i.i to float
>  %cmp2.i.i.i.i.1 = fcmp olt float %69, %70
>  br i1 %cmp2.i.i.i.i.1, label %if.then.i.i.i.i.1, label %if.end.i.i.i.1
>
>if.then.i.i.i.i.1:                                ; preds = %lor.lhs.false.i.i.i.i.1, %if.then.i.i.i.1
>  %sub.i.i.i.i.1 = add i64 %add.i.i.i.i.1, 360523849793537
>  br label %if.end.i.i.i.1
>
>if.end.i.i.i.1:                                   ; preds = %if.then.i.i.i.i.1, %lor.lhs.false.i.i.i.i.1, %MWC_AddMod64.exit16.i.i.i
>  %r.1.i.i.i.1 = phi i64 [ %r.1.i.i.i, %MWC_AddMod64.exit16.i.i.i ], [ %sub.i.i.i.i.1, %if.then.i.i.i.i.1 ], [ %add.i.i.i.i.1, %lor.lhs.false.i.i.i.i.1 ]
>  %add.i9.i.i.i.1 = shl i64 %v.0.i15.i.i.i, 1
>  %cmp.i10.i.i.i.1 = icmp ult i64 %add.i9.i.i.i.1, -360523849793537
>  br i1 %cmp.i10.i.i.i.1, label %lor.lhs.false.i12.i.i.i.1, label %if.then.i14.i.i.i.1
>
>lor.lhs.false.i12.i.i.i.1:                        ; preds = %if.end.i.i.i.1
>  %71 = uitofp i64 %add.i9.i.i.i.1 to float
>  %72 = uitofp i64 %v.0.i15.i.i.i to float
>  %cmp2.i11.i.i.i.1 = fcmp olt float %71, %72
>  br i1 %cmp2.i11.i.i.i.1, label %if.then.i14.i.i.i.1, label %MWC_AddMod64.exit16.i.i.i.1
>
>if.then.i14.i.i.i.1:                              ; preds = %lor.lhs.false.i12.i.i.i.1, %if.end.i.i.i.1
>  %sub.i13.i.i.i.1 = add i64 %add.i9.i.i.i.1, 360523849793537
>  br label %MWC_AddMod64.exit16.i.i.i.1
>
>MWC_AddMod64.exit16.i.i.i.1:                      ; preds = %if.then.i14.i.i.i.1, %lor.lhs.false.i12.i.i.i.1
>  %v.0.i15.i.i.i.1 = phi i64 [ %sub.i13.i.i.i.1, %if.then.i14.i.i.i.1 ], [ %add.i9.i.i.i.1, %lor.lhs.false.i12.i.i.i.1 ]
>  %cmp.i.i.i.1 = icmp eq i64 %shr.i.i.i, 0
>  br i1 %cmp.i.i.i.1, label %MWC64X_SeedStreams.exit, label %while.body.i.i.i
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @BlurRow(float addrspace(1)* nocapture readonly %image, i32 %number_channels, i32 %channel, float addrspace(4)* nocapture readonly %filter, i32 %width, i32 %imageColumns, i32 %imageRows, <4 x float> addrspace(3)* nocapture %temp, <4 x float> addrspace(1)* nocapture %tempImage) local_unnamed_addr #4 !kernel_arg_addr_space !19 !kernel_arg_access_qual !20 !kernel_arg_type !21 !kernel_arg_base_type !22 !kernel_arg_type_qual !23 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %7, %3
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %add = add i32 %x.i.i, %width
>  %cmp98 = icmp ult i32 %2, %add
>  br i1 %cmp98, label %for.body.lr.ph, label %for.cond.cleanup
>
>for.body.lr.ph:                                   ; preds = %entry
>  %sub = add i32 %width, -1
>  %div = lshr i32 %sub, 1
>  %add6 = sub i32 %1, %div
>  %sub.i = add nsw i32 %imageColumns, -1
>  %mul1.i.i = mul i32 %16, %imageColumns
>  %and.i.i = and i32 %channel, 1
>  %cmp.i.i = icmp eq i32 %and.i.i, 0
>  %cmp1.i.i = icmp ugt i32 %number_channels, 2
>  %and3.i.i = and i32 %channel, 2
>  %cmp4.i.i = icmp eq i32 %and3.i.i, 0
>  %and8.i.i = and i32 %channel, 4
>  %cmp9.i.i = icmp eq i32 %and8.i.i, 0
>  %and16.i.i = and i32 %channel, 16
>  %cmp17.i.i = icmp eq i32 %and16.i.i, 0
>  br label %for.body
>
>for.cond.cleanup:                                 ; preds = %ReadFloat4.exit, %entry
>  tail call void @llvm.r600.group.barrier() #13
>  %17 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %18 = add i32 %17, %3
>  %cmp13 = icmp ult i32 %18, %imageColumns
>  br i1 %cmp13, label %for.cond15.preheader, label %if.end
>
>for.cond15.preheader:                             ; preds = %for.cond.cleanup
>  %cmp1793 = icmp ugt i32 %width, 7
>  br i1 %cmp1793, label %for.cond19.preheader, label %for.cond34.preheader
>
>for.body:                                         ; preds = %for.body.lr.ph, %ReadFloat4.exit
>  %i.099 = phi i32 [ %2, %for.body.lr.ph ], [ %add11, %ReadFloat4.exit ]
>  %sub7 = add i32 %add6, %i.099
>  %19 = icmp slt i32 %sub7, %imageColumns
>  %20 = icmp sgt i32 %sub7, 0
>  %21 = select i1 %20, i32 %sub7, i32 0
>  %22 = select i1 %19, i32 %21, i32 %sub.i
>  %mul4.i.i = add i32 %22, %mul1.i.i
>  %add.i.i = mul i32 %mul4.i.i, %number_channels
>  %add.ptr.i = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i
>  br i1 %cmp.i.i, label %if.end.i.i, label %if.then.i.i
>
>if.then.i.i:                                      ; preds = %for.body
>  %23 = load float, float addrspace(1)* %add.ptr.i, align 4, !tbaa !12
>  br label %if.end.i.i
>
>if.end.i.i:                                       ; preds = %if.then.i.i, %for.body
>  %24 = phi float [ 0.000000e+00, %for.body ], [ %23, %if.then.i.i ]
>  br i1 %cmp1.i.i, label %if.then2.i.i, label %if.end13.i.i
>
>if.then2.i.i:                                     ; preds = %if.end.i.i
>  br i1 %cmp4.i.i, label %if.end7.i.i, label %if.then5.i.i
>
>if.then5.i.i:                                     ; preds = %if.then2.i.i
>  %add.ptr.i31.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 1
>  %25 = load float, float addrspace(1)* %add.ptr.i31.i.i, align 4, !tbaa !12
>  br label %if.end7.i.i
>
>if.end7.i.i:                                      ; preds = %if.then5.i.i, %if.then2.i.i
>  %26 = phi float [ 0.000000e+00, %if.then2.i.i ], [ %25, %if.then5.i.i ]
>  br i1 %cmp9.i.i, label %if.end13.i.i, label %if.then10.i.i
>
>if.then10.i.i:                                    ; preds = %if.end7.i.i
>  %add.ptr.i30.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 2
>  %27 = load float, float addrspace(1)* %add.ptr.i30.i.i, align 4, !tbaa !12
>  br label %if.end13.i.i
>
>if.end13.i.i:                                     ; preds = %if.then10.i.i, %if.end7.i.i, %if.end.i.i
>  %28 = phi float [ %26, %if.end7.i.i ], [ %26, %if.then10.i.i ], [ 0.000000e+00, %if.end.i.i ]
>  %29 = phi float [ 0.000000e+00, %if.end7.i.i ], [ %27, %if.then10.i.i ], [ 0.000000e+00, %if.end.i.i ]
>  switch i32 %number_channels, label %ReadFloat4.exit [
>    i32 4, label %land.lhs.true.i.i
>    i32 2, label %land.lhs.true.i.i
>  ]
>
>land.lhs.true.i.i:                                ; preds = %if.end13.i.i, %if.end13.i.i
>  br i1 %cmp17.i.i, label %ReadFloat4.exit, label %if.then18.i.i
>
>if.then18.i.i:                                    ; preds = %land.lhs.true.i.i
>  %add.ptr.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 %number_channels
>  %add.ptr1.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i, i32 -1
>  %30 = load float, float addrspace(1)* %add.ptr1.i.i.i, align 4, !tbaa !12
>  br label %ReadFloat4.exit
>
>ReadFloat4.exit:                                  ; preds = %if.end13.i.i, %land.lhs.true.i.i, %if.then18.i.i
>  %31 = phi float [ 0.000000e+00, %if.end13.i.i ], [ 0.000000e+00, %land.lhs.true.i.i ], [ %30, %if.then18.i.i ]
>  %vecinit.i = insertelement <4 x float> undef, float %24, i32 0
>  %vecinit1.i = insertelement <4 x float> %vecinit.i, float %28, i32 1
>  %vecinit2.i = insertelement <4 x float> %vecinit1.i, float %29, i32 2
>  %vecinit3.i = insertelement <4 x float> %vecinit2.i, float %31, i32 3
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %i.099
>  store <4 x float> %vecinit3.i, <4 x float> addrspace(3)* %arrayidx, align 16, !tbaa !14
>  %add11 = add i32 %i.099, %x.i.i
>  %cmp = icmp ult i32 %add11, %add
>  br i1 %cmp, label %for.body, label %for.cond.cleanup
>
>for.cond19.preheader:                             ; preds = %for.cond15.preheader, %for.cond19.preheader
>  %i14.095 = phi i32 [ %add32, %for.cond19.preheader ], [ 0, %for.cond15.preheader ]
>  %result.094 = phi <4 x float> [ %55, %for.cond19.preheader ], [ zeroinitializer, %for.cond15.preheader ]
>  %arrayidx24 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %i14.095
>  %32 = load float, float addrspace(4)* %arrayidx24, align 4, !tbaa !12
>  %splat.splatinsert = insertelement <4 x float> undef, float %32, i32 0
>  %splat.splat = shufflevector <4 x float> %splat.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
>  %add27 = add nuw i32 %i14.095, %2
>  %arrayidx28 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add27
>  %33 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx28, align 16, !tbaa !14
>  %34 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat, <4 x float> %33, <4 x float> %result.094)
>  %add23.1 = or i32 %i14.095, 1
>  %arrayidx24.1 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add23.1
>  %35 = load float, float addrspace(4)* %arrayidx24.1, align 4, !tbaa !12
>  %splat.splatinsert.1 = insertelement <4 x float> undef, float %35, i32 0
>  %splat.splat.1 = shufflevector <4 x float> %splat.splatinsert.1, <4 x float> undef, <4 x i32> zeroinitializer
>  %add27.1 = add nuw i32 %add23.1, %2
>  %arrayidx28.1 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add27.1
>  %36 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx28.1, align 16, !tbaa !14
>  %37 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.1, <4 x float> %36, <4 x float> %34)
>  %add23.2 = or i32 %i14.095, 2
>  %arrayidx24.2 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add23.2
>  %38 = load float, float addrspace(4)* %arrayidx24.2, align 4, !tbaa !12
>  %splat.splatinsert.2 = insertelement <4 x float> undef, float %38, i32 0
>  %splat.splat.2 = shufflevector <4 x float> %splat.splatinsert.2, <4 x float> undef, <4 x i32> zeroinitializer
>  %add27.2 = add nuw i32 %add23.2, %2
>  %arrayidx28.2 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add27.2
>  %39 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx28.2, align 16, !tbaa !14
>  %40 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.2, <4 x float> %39, <4 x float> %37)
>  %add23.3 = or i32 %i14.095, 3
>  %arrayidx24.3 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add23.3
>  %41 = load float, float addrspace(4)* %arrayidx24.3, align 4, !tbaa !12
>  %splat.splatinsert.3 = insertelement <4 x float> undef, float %41, i32 0
>  %splat.splat.3 = shufflevector <4 x float> %splat.splatinsert.3, <4 x float> undef, <4 x i32> zeroinitializer
>  %add27.3 = add nuw i32 %add23.3, %2
>  %arrayidx28.3 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add27.3
>  %42 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx28.3, align 16, !tbaa !14
>  %43 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.3, <4 x float> %42, <4 x float> %40)
>  %add23.4 = or i32 %i14.095, 4
>  %arrayidx24.4 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add23.4
>  %44 = load float, float addrspace(4)* %arrayidx24.4, align 4, !tbaa !12
>  %splat.splatinsert.4 = insertelement <4 x float> undef, float %44, i32 0
>  %splat.splat.4 = shufflevector <4 x float> %splat.splatinsert.4, <4 x float> undef, <4 x i32> zeroinitializer
>  %add27.4 = add nuw i32 %add23.4, %2
>  %arrayidx28.4 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add27.4
>  %45 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx28.4, align 16, !tbaa !14
>  %46 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.4, <4 x float> %45, <4 x float> %43)
>  %add23.5 = or i32 %i14.095, 5
>  %arrayidx24.5 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add23.5
>  %47 = load float, float addrspace(4)* %arrayidx24.5, align 4, !tbaa !12
>  %splat.splatinsert.5 = insertelement <4 x float> undef, float %47, i32 0
>  %splat.splat.5 = shufflevector <4 x float> %splat.splatinsert.5, <4 x float> undef, <4 x i32> zeroinitializer
>  %add27.5 = add nuw i32 %add23.5, %2
>  %arrayidx28.5 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add27.5
>  %48 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx28.5, align 16, !tbaa !14
>  %49 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.5, <4 x float> %48, <4 x float> %46)
>  %add23.6 = or i32 %i14.095, 6
>  %arrayidx24.6 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add23.6
>  %50 = load float, float addrspace(4)* %arrayidx24.6, align 4, !tbaa !12
>  %splat.splatinsert.6 = insertelement <4 x float> undef, float %50, i32 0
>  %splat.splat.6 = shufflevector <4 x float> %splat.splatinsert.6, <4 x float> undef, <4 x i32> zeroinitializer
>  %add27.6 = add nuw i32 %add23.6, %2
>  %arrayidx28.6 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add27.6
>  %51 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx28.6, align 16, !tbaa !14
>  %52 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.6, <4 x float> %51, <4 x float> %49)
>  %add23.7 = or i32 %i14.095, 7
>  %arrayidx24.7 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add23.7
>  %53 = load float, float addrspace(4)* %arrayidx24.7, align 4, !tbaa !12
>  %splat.splatinsert.7 = insertelement <4 x float> undef, float %53, i32 0
>  %splat.splat.7 = shufflevector <4 x float> %splat.splatinsert.7, <4 x float> undef, <4 x i32> zeroinitializer
>  %add27.7 = add nuw i32 %add23.7, %2
>  %arrayidx28.7 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add27.7
>  %54 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx28.7, align 16, !tbaa !14
>  %55 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.7, <4 x float> %54, <4 x float> %52)
>  %add32 = add nuw nsw i32 %i14.095, 8
>  %add16 = or i32 %add32, 7
>  %cmp17 = icmp ult i32 %add16, %width
>  br i1 %cmp17, label %for.cond19.preheader, label %for.cond34.preheader.loopexit
>
>for.cond34.preheader.loopexit:                    ; preds = %for.cond19.preheader
>  %56 = and i32 %width, -8
>  br label %for.cond34.preheader
>
>for.cond34.preheader:                             ; preds = %for.cond34.preheader.loopexit, %for.cond15.preheader
>  %result.0.lcssa = phi <4 x float> [ zeroinitializer, %for.cond15.preheader ], [ %55, %for.cond34.preheader.loopexit ]
>  %i14.0.lcssa = phi i32 [ 0, %for.cond15.preheader ], [ %56, %for.cond34.preheader.loopexit ]
>  %cmp3588 = icmp ult i32 %i14.0.lcssa, %width
>  br i1 %cmp3588, label %for.body36, label %for.end46
>
>for.body36:                                       ; preds = %for.cond34.preheader, %for.body36
>  %i14.190 = phi i32 [ %inc45, %for.body36 ], [ %i14.0.lcssa, %for.cond34.preheader ]
>  %result.289 = phi <4 x float> [ %59, %for.body36 ], [ %result.0.lcssa, %for.cond34.preheader ]
>  %arrayidx37 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %i14.190
>  %57 = load float, float addrspace(4)* %arrayidx37, align 4, !tbaa !12
>  %splat.splatinsert38 = insertelement <4 x float> undef, float %57, i32 0
>  %splat.splat39 = shufflevector <4 x float> %splat.splatinsert38, <4 x float> undef, <4 x i32> zeroinitializer
>  %add41 = add nuw i32 %i14.190, %2
>  %arrayidx42 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add41
>  %58 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx42, align 16, !tbaa !14
>  %59 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat39, <4 x float> %58, <4 x float> %result.289)
>  %inc45 = add nuw nsw i32 %i14.190, 1
>  %exitcond = icmp eq i32 %inc45, %width
>  br i1 %exitcond, label %for.end46, label %for.body36
>
>for.end46:                                        ; preds = %for.body36, %for.cond34.preheader
>  %result.2.lcssa = phi <4 x float> [ %result.0.lcssa, %for.cond34.preheader ], [ %59, %for.body36 ]
>  %mul47 = mul nsw i32 %16, %imageColumns
>  %add48 = add nsw i32 %8, %mul47
>  %arrayidx49 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %tempImage, i32 %add48
>  store <4 x float> %result.2.lcssa, <4 x float> addrspace(1)* %arrayidx49, align 16, !tbaa !14
>  br label %if.end
>
>if.end:                                           ; preds = %for.end46, %for.cond.cleanup
>  ret void
>}
>
>; Function Attrs: nounwind readnone speculatable
>declare <4 x float> @llvm.fmuladd.v4f32(<4 x float>, <4 x float>, <4 x float>) #5
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @BlurColumn(<4 x float> addrspace(1)* nocapture readonly %blurRowData, i32 %number_channels, i32 %channel, float addrspace(4)* nocapture readonly %filter, i32 %width, i32 %imageColumns, i32 %imageRows, <4 x float> addrspace(3)* nocapture %temp, float addrspace(1)* nocapture %filteredImage) local_unnamed_addr #4 !kernel_arg_addr_space !19 !kernel_arg_access_qual !20 !kernel_arg_type !24 !kernel_arg_base_type !25 !kernel_arg_type_qual !23 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i88 = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i88, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %15, %12
>  %add = add i32 %y.i.i88, %width
>  %cmp99 = icmp ult i32 %11, %add
>  br i1 %cmp99, label %for.body.lr.ph, label %for.cond.cleanup
>
>for.body.lr.ph:                                   ; preds = %entry
>  %sub = add i32 %width, -1
>  %div = lshr i32 %sub, 1
>  %add9 = sub i32 %10, %div
>  %sub.i = add nsw i32 %imageRows, -1
>  br label %for.body
>
>for.cond.cleanup:                                 ; preds = %for.body, %entry
>  tail call void @llvm.r600.group.barrier() #13
>  %17 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %18 = add i32 %17, %12
>  %cmp18 = icmp ult i32 %18, %imageRows
>  br i1 %cmp18, label %for.cond20.preheader, label %if.end
>
>for.cond20.preheader:                             ; preds = %for.cond.cleanup
>  %cmp2294 = icmp ugt i32 %width, 7
>  br i1 %cmp2294, label %for.cond24.preheader, label %for.cond39.preheader
>
>for.body:                                         ; preds = %for.body.lr.ph, %for.body
>  %i.0100 = phi i32 [ %11, %for.body.lr.ph ], [ %add16, %for.body ]
>  %sub10 = add i32 %add9, %i.0100
>  %19 = icmp slt i32 %sub10, %imageRows
>  %20 = icmp sgt i32 %sub10, 0
>  %21 = select i1 %20, i32 %sub10, i32 0
>  %22 = select i1 %19, i32 %21, i32 %sub.i
>  %mul12 = mul nsw i32 %22, %imageColumns
>  %add13 = add nsw i32 %mul12, %1
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %blurRowData, i32 %add13
>  %23 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %arrayidx14 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %i.0100
>  store <4 x float> %23, <4 x float> addrspace(3)* %arrayidx14, align 16, !tbaa !14
>  %add16 = add i32 %i.0100, %y.i.i88
>  %cmp = icmp ult i32 %add16, %add
>  br i1 %cmp, label %for.body, label %for.cond.cleanup
>
>for.cond24.preheader:                             ; preds = %for.cond20.preheader, %for.cond24.preheader
>  %i19.096 = phi i32 [ %add37, %for.cond24.preheader ], [ 0, %for.cond20.preheader ]
>  %result.095 = phi <4 x float> [ %47, %for.cond24.preheader ], [ zeroinitializer, %for.cond20.preheader ]
>  %arrayidx29 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %i19.096
>  %24 = load float, float addrspace(4)* %arrayidx29, align 4, !tbaa !12
>  %splat.splatinsert = insertelement <4 x float> undef, float %24, i32 0
>  %splat.splat = shufflevector <4 x float> %splat.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
>  %add32 = add nuw i32 %i19.096, %11
>  %arrayidx33 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add32
>  %25 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx33, align 16, !tbaa !14
>  %26 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat, <4 x float> %25, <4 x float> %result.095)
>  %add28.1 = or i32 %i19.096, 1
>  %arrayidx29.1 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add28.1
>  %27 = load float, float addrspace(4)* %arrayidx29.1, align 4, !tbaa !12
>  %splat.splatinsert.1 = insertelement <4 x float> undef, float %27, i32 0
>  %splat.splat.1 = shufflevector <4 x float> %splat.splatinsert.1, <4 x float> undef, <4 x i32> zeroinitializer
>  %add32.1 = add nuw i32 %add28.1, %11
>  %arrayidx33.1 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add32.1
>  %28 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx33.1, align 16, !tbaa !14
>  %29 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.1, <4 x float> %28, <4 x float> %26)
>  %add28.2 = or i32 %i19.096, 2
>  %arrayidx29.2 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add28.2
>  %30 = load float, float addrspace(4)* %arrayidx29.2, align 4, !tbaa !12
>  %splat.splatinsert.2 = insertelement <4 x float> undef, float %30, i32 0
>  %splat.splat.2 = shufflevector <4 x float> %splat.splatinsert.2, <4 x float> undef, <4 x i32> zeroinitializer
>  %add32.2 = add nuw i32 %add28.2, %11
>  %arrayidx33.2 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add32.2
>  %31 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx33.2, align 16, !tbaa !14
>  %32 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.2, <4 x float> %31, <4 x float> %29)
>  %add28.3 = or i32 %i19.096, 3
>  %arrayidx29.3 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add28.3
>  %33 = load float, float addrspace(4)* %arrayidx29.3, align 4, !tbaa !12
>  %splat.splatinsert.3 = insertelement <4 x float> undef, float %33, i32 0
>  %splat.splat.3 = shufflevector <4 x float> %splat.splatinsert.3, <4 x float> undef, <4 x i32> zeroinitializer
>  %add32.3 = add nuw i32 %add28.3, %11
>  %arrayidx33.3 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add32.3
>  %34 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx33.3, align 16, !tbaa !14
>  %35 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.3, <4 x float> %34, <4 x float> %32)
>  %add28.4 = or i32 %i19.096, 4
>  %arrayidx29.4 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add28.4
>  %36 = load float, float addrspace(4)* %arrayidx29.4, align 4, !tbaa !12
>  %splat.splatinsert.4 = insertelement <4 x float> undef, float %36, i32 0
>  %splat.splat.4 = shufflevector <4 x float> %splat.splatinsert.4, <4 x float> undef, <4 x i32> zeroinitializer
>  %add32.4 = add nuw i32 %add28.4, %11
>  %arrayidx33.4 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add32.4
>  %37 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx33.4, align 16, !tbaa !14
>  %38 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.4, <4 x float> %37, <4 x float> %35)
>  %add28.5 = or i32 %i19.096, 5
>  %arrayidx29.5 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add28.5
>  %39 = load float, float addrspace(4)* %arrayidx29.5, align 4, !tbaa !12
>  %splat.splatinsert.5 = insertelement <4 x float> undef, float %39, i32 0
>  %splat.splat.5 = shufflevector <4 x float> %splat.splatinsert.5, <4 x float> undef, <4 x i32> zeroinitializer
>  %add32.5 = add nuw i32 %add28.5, %11
>  %arrayidx33.5 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add32.5
>  %40 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx33.5, align 16, !tbaa !14
>  %41 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.5, <4 x float> %40, <4 x float> %38)
>  %add28.6 = or i32 %i19.096, 6
>  %arrayidx29.6 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add28.6
>  %42 = load float, float addrspace(4)* %arrayidx29.6, align 4, !tbaa !12
>  %splat.splatinsert.6 = insertelement <4 x float> undef, float %42, i32 0
>  %splat.splat.6 = shufflevector <4 x float> %splat.splatinsert.6, <4 x float> undef, <4 x i32> zeroinitializer
>  %add32.6 = add nuw i32 %add28.6, %11
>  %arrayidx33.6 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add32.6
>  %43 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx33.6, align 16, !tbaa !14
>  %44 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.6, <4 x float> %43, <4 x float> %41)
>  %add28.7 = or i32 %i19.096, 7
>  %arrayidx29.7 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add28.7
>  %45 = load float, float addrspace(4)* %arrayidx29.7, align 4, !tbaa !12
>  %splat.splatinsert.7 = insertelement <4 x float> undef, float %45, i32 0
>  %splat.splat.7 = shufflevector <4 x float> %splat.splatinsert.7, <4 x float> undef, <4 x i32> zeroinitializer
>  %add32.7 = add nuw i32 %add28.7, %11
>  %arrayidx33.7 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add32.7
>  %46 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx33.7, align 16, !tbaa !14
>  %47 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.7, <4 x float> %46, <4 x float> %44)
>  %add37 = add nuw nsw i32 %i19.096, 8
>  %add21 = or i32 %add37, 7
>  %cmp22 = icmp ult i32 %add21, %width
>  br i1 %cmp22, label %for.cond24.preheader, label %for.cond39.preheader.loopexit
>
>for.cond39.preheader.loopexit:                    ; preds = %for.cond24.preheader
>  %48 = and i32 %width, -8
>  br label %for.cond39.preheader
>
>for.cond39.preheader:                             ; preds = %for.cond39.preheader.loopexit, %for.cond20.preheader
>  %result.0.lcssa = phi <4 x float> [ zeroinitializer, %for.cond20.preheader ], [ %47, %for.cond39.preheader.loopexit ]
>  %i19.0.lcssa = phi i32 [ 0, %for.cond20.preheader ], [ %48, %for.cond39.preheader.loopexit ]
>  %cmp4089 = icmp ult i32 %i19.0.lcssa, %width
>  br i1 %cmp4089, label %for.body41, label %for.end51
>
>for.body41:                                       ; preds = %for.cond39.preheader, %for.body41
>  %i19.191 = phi i32 [ %inc50, %for.body41 ], [ %i19.0.lcssa, %for.cond39.preheader ]
>  %result.290 = phi <4 x float> [ %51, %for.body41 ], [ %result.0.lcssa, %for.cond39.preheader ]
>  %arrayidx42 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %i19.191
>  %49 = load float, float addrspace(4)* %arrayidx42, align 4, !tbaa !12
>  %splat.splatinsert43 = insertelement <4 x float> undef, float %49, i32 0
>  %splat.splat44 = shufflevector <4 x float> %splat.splatinsert43, <4 x float> undef, <4 x i32> zeroinitializer
>  %add46 = add nuw i32 %i19.191, %11
>  %arrayidx47 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %temp, i32 %add46
>  %50 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx47, align 16, !tbaa !14
>  %51 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat44, <4 x float> %50, <4 x float> %result.290)
>  %inc50 = add nuw nsw i32 %i19.191, 1
>  %exitcond = icmp eq i32 %inc50, %width
>  br i1 %exitcond, label %for.end51, label %for.body41
>
>for.end51:                                        ; preds = %for.body41, %for.cond39.preheader
>  %result.2.lcssa = phi <4 x float> [ %result.0.lcssa, %for.cond39.preheader ], [ %51, %for.body41 ]
>  %mul1.i.i = mul i32 %16, %imageColumns
>  %mul4.i.i = add i32 %8, %mul1.i.i
>  %add.i.i = mul i32 %mul4.i.i, %number_channels
>  %add.ptr.i = getelementptr inbounds float, float addrspace(1)* %filteredImage, i32 %add.i.i
>  %52 = extractelement <4 x float> %result.2.lcssa, i32 0
>  %53 = extractelement <4 x float> %result.2.lcssa, i32 1
>  %54 = extractelement <4 x float> %result.2.lcssa, i32 2
>  %55 = extractelement <4 x float> %result.2.lcssa, i32 3
>  %and.i.i = and i32 %channel, 1
>  %cmp.i.i = icmp eq i32 %and.i.i, 0
>  br i1 %cmp.i.i, label %if.end.i.i, label %if.then.i.i
>
>if.then.i.i:                                      ; preds = %for.end51
>  %56 = fcmp ogt float %52, 6.553500e+04
>  %57 = fcmp olt float %52, 0.000000e+00
>  %58 = select i1 %57, float 0.000000e+00, float %52
>  %.op.i.i.i = fadd float %58, 5.000000e-01
>  %add.i.i.i = select i1 %56, float 6.553550e+04, float %.op.i.i.i
>  store float %add.i.i.i, float addrspace(1)* %add.ptr.i, align 4, !tbaa !12
>  br label %if.end.i.i
>
>if.end.i.i:                                       ; preds = %if.then.i.i, %for.end51
>  %cmp1.i.i = icmp ugt i32 %number_channels, 2
>  br i1 %cmp1.i.i, label %if.then2.i.i, label %if.end13.i.i
>
>if.then2.i.i:                                     ; preds = %if.end.i.i
>  %and3.i.i = and i32 %channel, 2
>  %cmp4.i.i = icmp eq i32 %and3.i.i, 0
>  br i1 %cmp4.i.i, label %if.end7.i.i, label %if.then5.i.i
>
>if.then5.i.i:                                     ; preds = %if.then2.i.i
>  %add.ptr.i35.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 1
>  %59 = fcmp ogt float %53, 6.553500e+04
>  %60 = fcmp olt float %53, 0.000000e+00
>  %61 = select i1 %60, float 0.000000e+00, float %53
>  %.op.i36.i.i = fadd float %61, 5.000000e-01
>  %add.i37.i.i = select i1 %59, float 6.553550e+04, float %.op.i36.i.i
>  store float %add.i37.i.i, float addrspace(1)* %add.ptr.i35.i.i, align 4, !tbaa !12
>  br label %if.end7.i.i
>
>if.end7.i.i:                                      ; preds = %if.then5.i.i, %if.then2.i.i
>  %and8.i.i = and i32 %channel, 4
>  %cmp9.i.i = icmp eq i32 %and8.i.i, 0
>  br i1 %cmp9.i.i, label %if.end13.i.i, label %if.then10.i.i
>
>if.then10.i.i:                                    ; preds = %if.end7.i.i
>  %add.ptr.i32.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 2
>  %62 = fcmp ogt float %54, 6.553500e+04
>  %63 = fcmp olt float %54, 0.000000e+00
>  %64 = select i1 %63, float 0.000000e+00, float %54
>  %.op.i33.i.i = fadd float %64, 5.000000e-01
>  %add.i34.i.i = select i1 %62, float 6.553550e+04, float %.op.i33.i.i
>  store float %add.i34.i.i, float addrspace(1)* %add.ptr.i32.i.i, align 4, !tbaa !12
>  br label %if.end13.i.i
>
>if.end13.i.i:                                     ; preds = %if.then10.i.i, %if.end7.i.i, %if.end.i.i
>  switch i32 %number_channels, label %if.end [
>    i32 4, label %land.lhs.true.i.i
>    i32 2, label %land.lhs.true.i.i
>  ]
>
>land.lhs.true.i.i:                                ; preds = %if.end13.i.i, %if.end13.i.i
>  %and16.i.i = and i32 %channel, 16
>  %cmp17.i.i = icmp eq i32 %and16.i.i, 0
>  br i1 %cmp17.i.i, label %if.end, label %if.then18.i.i
>
>if.then18.i.i:                                    ; preds = %land.lhs.true.i.i
>  %add.ptr.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 %number_channels
>  %add.ptr1.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i, i32 -1
>  %65 = fcmp ogt float %55, 6.553500e+04
>  %66 = fcmp olt float %55, 0.000000e+00
>  %67 = select i1 %66, float 0.000000e+00, float %55
>  %.op.i30.i.i = fadd float %67, 5.000000e-01
>  %add.i31.i.i = select i1 %65, float 6.553550e+04, float %.op.i30.i.i
>  store float %add.i31.i.i, float addrspace(1)* %add.ptr1.i.i.i, align 4, !tbaa !12
>  br label %if.end
>
>if.end:                                           ; preds = %if.then18.i.i, %land.lhs.true.i.i, %if.end13.i.i, %for.cond.cleanup
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @Contrast(float addrspace(1)* nocapture %image, i32 %number_channels, i32 %sign) local_unnamed_addr #4 !kernel_arg_addr_space !26 !kernel_arg_access_qual !27 !kernel_arg_type !28 !kernel_arg_base_type !28 !kernel_arg_type_qual !29 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %x.i = tail call i32 @llvm.r600.read.global.size.x() #16
>  %mul1.i.i41 = mul i32 %16, %x.i
>  %mul4.i.i42 = add i32 %8, %mul1.i.i41
>  %add.i.i43 = mul i32 %mul4.i.i42, %number_channels
>  %add.ptr.i44 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i43
>  %17 = load float, float addrspace(1)* %add.ptr.i44, align 4, !tbaa !12
>  %18 = insertelement <4 x float> undef, float %17, i32 0
>  %cmp.i45 = icmp ugt i32 %number_channels, 2
>  br i1 %cmp.i45, label %if.then.i46, label %if.end.i47
>
>if.then.i46:                                      ; preds = %entry
>  %add.ptr.i23.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i44, i32 1
>  %19 = load float, float addrspace(1)* %add.ptr.i23.i, align 4, !tbaa !12
>  %20 = insertelement <4 x float> %18, float %19, i32 1
>  %add.ptr.i22.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i44, i32 2
>  %21 = load float, float addrspace(1)* %add.ptr.i22.i, align 4, !tbaa !12
>  %22 = insertelement <4 x float> %20, float %21, i32 2
>  br label %if.end.i47
>
>if.end.i47:                                       ; preds = %if.then.i46, %entry
>  %pixel.0.i = phi <4 x float> [ %22, %if.then.i46 ], [ %18, %entry ]
>  switch i32 %number_channels, label %ReadAllChannels.exit [
>    i32 4, label %if.then6.i50
>    i32 2, label %if.then6.i50
>  ]
>
>if.then6.i50:                                     ; preds = %if.end.i47, %if.end.i47
>  %add.ptr.i.i48 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i44, i32 %number_channels
>  %add.ptr1.i.i49 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i48, i32 -1
>  %23 = load float, float addrspace(1)* %add.ptr1.i.i49, align 4, !tbaa !12
>  %24 = insertelement <4 x float> %pixel.0.i, float %23, i32 3
>  br label %ReadAllChannels.exit
>
>ReadAllChannels.exit:                             ; preds = %if.end.i47, %if.then6.i50
>  %pixel.1.i = phi <4 x float> [ %24, %if.then6.i50 ], [ %pixel.0.i, %if.end.i47 ]
>  %cmp = icmp ult i32 %number_channels, 3
>  %25 = shufflevector <4 x float> %pixel.1.i, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 3>
>  %spec.select = select i1 %cmp, <4 x float> %25, <4 x float> %pixel.1.i
>  %26 = shufflevector <4 x float> <float 0.000000e+00, float undef, float undef, float undef>, <4 x float> %spec.select, <4 x i32> <i32 0, i32 0, i32 0, i32 7>
>  %27 = extractelement <4 x float> %spec.select, i32 0
>  %28 = extractelement <4 x float> %spec.select, i32 1
>  %cmp.i36 = fcmp ogt float %27, %28
>  %cond.i = select i1 %cmp.i36, float %27, float %28
>  %29 = extractelement <4 x float> %spec.select, i32 2
>  %cmp1.i = fcmp ogt float %cond.i, %29
>  %cond..i = select i1 %cmp1.i, float %cond.i, float %29
>  %cmp11.i = fcmp une float %cond..i, 0.000000e+00
>  br i1 %cmp11.i, label %if.then.i39, label %ConvertRGBToHSB.exit
>
>if.then.i39:                                      ; preds = %ReadAllChannels.exit
>  %cmp12.i = fcmp olt float %27, %28
>  %cond16.i = select i1 %cmp12.i, float %27, float %28
>  %cmp17.i = fcmp olt float %cond16.i, %29
>  %cond16..i = select i1 %cmp17.i, float %cond16.i, float %29
>  %sub.i37 = fsub float %cond..i, %cond16..i
>  %div.i = fdiv float %sub.i37, %cond..i, !fpmath !11
>  %30 = insertelement <4 x float> %26, float %div.i, i32 1
>  %mul.i38 = fmul float %cond..i, 0x3EEF751040000000
>  %31 = insertelement <4 x float> %30, float %mul.i38, i32 2
>  %cmp27.i = fcmp une float %sub.i37, 0.000000e+00
>  br i1 %cmp27.i, label %if.then28.i, label %ConvertRGBToHSB.exit
>
>if.then28.i:                                      ; preds = %if.then.i39
>  %cmp29.i = fcmp oeq float %27, %cond..i
>  %cmp32.i = fcmp oeq float %28, %cond..i
>  %cond33.i = select i1 %cmp32.i, float 2.000000e+00, float 4.000000e+00
>  %cond35.i = select i1 %cmp29.i, float 0.000000e+00, float %cond33.i
>  br i1 %cmp29.i, label %cond.true37.i, label %cond.false39.i
>
>cond.true37.i:                                    ; preds = %if.then28.i
>  %sub38.i = fsub float %28, %29
>  br label %cond.end47.i
>
>cond.false39.i:                                   ; preds = %if.then28.i
>  br i1 %cmp32.i, label %cond.true41.i, label %cond.false43.i
>
>cond.true41.i:                                    ; preds = %cond.false39.i
>  %sub42.i = fsub float %29, %27
>  br label %cond.end47.i
>
>cond.false43.i:                                   ; preds = %cond.false39.i
>  %sub44.i = fsub float %27, %28
>  br label %cond.end47.i
>
>cond.end47.i:                                     ; preds = %cond.false43.i, %cond.true41.i, %cond.true37.i
>  %cond48.i = phi float [ %sub38.i, %cond.true37.i ], [ %sub42.i, %cond.true41.i ], [ %sub44.i, %cond.false43.i ]
>  %div49.i = fdiv float %cond48.i, %sub.i37, !fpmath !11
>  %add.i = fadd float %cond35.i, %div49.i
>  %div50.i = fdiv float %add.i, 6.000000e+00, !fpmath !11
>  %cmp51.i = fcmp olt float %div50.i, 0.000000e+00
>  %cond52.i = select i1 %cmp51.i, float 0.000000e+00, float 1.000000e+00
>  %add53.i = fadd float %div50.i, %cond52.i
>  %32 = insertelement <4 x float> %31, float %add53.i, i32 0
>  br label %ConvertRGBToHSB.exit
>
>ConvertRGBToHSB.exit:                             ; preds = %ReadAllChannels.exit, %if.then.i39, %cond.end47.i
>  %result.1.i40 = phi <4 x float> [ %26, %ReadAllChannels.exit ], [ %32, %cond.end47.i ], [ %31, %if.then.i39 ]
>  %33 = extractelement <4 x float> %result.1.i40, i32 2
>  %conv = sitofp i32 %sign to float
>  %mul = fmul float %conv, 5.000000e-01
>  %sub = fadd float %33, -5.000000e-01
>  %34 = bitcast float %sub to i32
>  %35 = and i32 %34, -2147483648
>  %36 = and i32 %34, 2147483647
>  %37 = bitcast i32 %36 to float
>  %38 = fptosi float %37 to i32
>  %39 = sitofp i32 %38 to float
>  %40 = fsub float %37, %39
>  %41 = shl i32 %38, 31
>  %42 = xor i32 %41, %35
>  %43 = icmp ult i32 %36, 2139095040
>  %44 = select i1 %43, i32 %35, i32 2143289344
>  %45 = fsub float 1.000000e+00, %40
>  %46 = fcmp ole float %40, 7.500000e-01
>  %47 = fadd float %40, -5.000000e-01
>  %48 = select i1 %46, float %47, float %45
>  %49 = fcmp olt float %40, 5.000000e-01
>  %50 = fsub float 5.000000e-01, %40
>  %51 = select i1 %49, float %50, float %48
>  %52 = fcmp ole float %40, 2.500000e-01
>  %53 = select i1 %52, float %40, float %51
>  %54 = xor i1 %52, true
>  %55 = and i1 %46, %54
>  %56 = fmul float %53, 0x400921FB60000000
>  %57 = fmul float %56, %56
>  %58 = fmul float %56, %57
>  %59 = tail call float @llvm.fmuladd.f32(float %57, float 0x3EC6DBE4A0000000, float 0xBF2A013A80000000) #13
>  %60 = tail call float @llvm.fmuladd.f32(float %57, float %59, float 0x3F811110E0000000) #13
>  %61 = tail call float @llvm.fmuladd.f32(float %57, float %60, float 0xBFC5555560000000) #13
>  %62 = tail call float @llvm.fmuladd.f32(float %58, float %61, float %56) #13
>  %63 = fmul float %57, %57
>  %64 = tail call float @llvm.fmuladd.f32(float %57, float 0xBE92524740000000, float 0x3EFA015C60000000) #13
>  %65 = tail call float @llvm.fmuladd.f32(float %57, float %64, float 0xBF56C16C00000000) #13
>  %66 = tail call float @llvm.fmuladd.f32(float %57, float %65, float 0x3FA5555560000000) #13
>  %67 = tail call float @llvm.fmuladd.f32(float %57, float -5.000000e-01, float 1.000000e+00) #13
>  %68 = tail call float @llvm.fmuladd.f32(float %63, float %66, float %67) #13
>  %69 = select i1 %55, float %68, float %62
>  %70 = bitcast float %69 to i32
>  %71 = xor i32 %42, %70
>  %72 = icmp ult i32 %36, 1258291200
>  %73 = select i1 %72, i32 %71, i32 %44
>  %74 = bitcast i32 %73 to float
>  %add = fadd float %74, 1.000000e+00
>  %neg = fsub float -0.000000e+00, %33
>  %75 = tail call float @llvm.fmuladd.f32(float %add, float 5.000000e-01, float %neg)
>  %76 = tail call float @llvm.fmuladd.f32(float %mul, float %75, float %33)
>  %77 = fcmp ogt float %76, 1.000000e+00
>  %78 = fcmp olt float %76, 0.000000e+00
>  %79 = select i1 %78, float 0.000000e+00, float %76
>  %80 = select i1 %77, float 1.000000e+00, float %79
>  %81 = insertelement <4 x float> %result.1.i40, float %80, i32 2
>  %82 = extractelement <4 x float> %result.1.i40, i32 0
>  %83 = extractelement <4 x float> %result.1.i40, i32 1
>  %cmp.i32 = fcmp oeq float %83, 0.000000e+00
>  br i1 %cmp.i32, label %if.then.i35, label %if.else.i
>
>if.then.i35:                                      ; preds = %ConvertRGBToHSB.exit
>  %mul.i = fmul float %80, 6.553500e+04
>  %84 = fcmp ogt float %mul.i, 6.553500e+04
>  %85 = fcmp olt float %mul.i, 0.000000e+00
>  %86 = select i1 %85, float 0.000000e+00, float %mul.i
>  %.op.i.i33 = fadd float %86, 5.000000e-01
>  %add.i.i34 = select i1 %84, float 6.553550e+04, float %.op.i.i33
>  %87 = insertelement <4 x float> %81, float %add.i.i34, i32 2
>  %88 = insertelement <4 x float> %87, float %add.i.i34, i32 1
>  %89 = insertelement <4 x float> %88, float %add.i.i34, i32 0
>  br label %ConvertHSBToRGB.exit
>
>if.else.i:                                        ; preds = %ConvertRGBToHSB.exit
>  %call1.i = tail call float @llvm.floor.f32(float %82) #17
>  %sub.i = fsub float %82, %call1.i
>  %mul2.i = fmul float %sub.i, 6.000000e+00
>  %call3.i = tail call float @llvm.floor.f32(float %mul2.i) #17
>  %sub4.i = fsub float %mul2.i, %call3.i
>  %sub5.i = fsub float 1.000000e+00, %83
>  %mul6.i = fmul float %sub5.i, %80
>  %neg.i = fsub float -0.000000e+00, %83
>  %90 = tail call float @llvm.fmuladd.f32(float %neg.i, float %sub4.i, float 1.000000e+00) #13
>  %mul8.i = fmul float %90, %80
>  %sub9.i = fsub float 1.000000e+00, %sub4.i
>  %91 = tail call float @llvm.fmuladd.f32(float %neg.i, float %sub9.i, float 1.000000e+00) #13
>  %mul12.i = fmul float %91, %80
>  %conv.i = fptosi float %mul2.i to i32
>  switch i32 %conv.i, label %if.else52.i [
>    i32 1, label %if.then15.i
>    i32 2, label %if.then25.i
>    i32 3, label %if.then35.i
>    i32 4, label %if.then45.i
>  ]
>
>if.then15.i:                                      ; preds = %if.else.i
>  %mul16.i = fmul float %mul8.i, 6.553500e+04
>  %92 = fcmp ogt float %mul16.i, 6.553500e+04
>  %93 = fcmp olt float %mul16.i, 0.000000e+00
>  %94 = select i1 %93, float 0.000000e+00, float %mul16.i
>  %.op.i167.i = fadd float %94, 5.000000e-01
>  %add.i168.i = select i1 %92, float 6.553550e+04, float %.op.i167.i
>  %95 = insertelement <4 x float> %81, float %add.i168.i, i32 0
>  %mul18.i = fmul float %80, 6.553500e+04
>  %96 = fcmp ogt float %mul18.i, 6.553500e+04
>  %97 = fcmp olt float %mul18.i, 0.000000e+00
>  %98 = select i1 %97, float 0.000000e+00, float %mul18.i
>  %.op.i165.i = fadd float %98, 5.000000e-01
>  %add.i166.i = select i1 %96, float 6.553550e+04, float %.op.i165.i
>  %99 = insertelement <4 x float> %95, float %add.i166.i, i32 1
>  %mul20.i = fmul float %mul6.i, 6.553500e+04
>  %100 = fcmp ogt float %mul20.i, 6.553500e+04
>  %101 = fcmp olt float %mul20.i, 0.000000e+00
>  %102 = select i1 %101, float 0.000000e+00, float %mul20.i
>  %.op.i163.i = fadd float %102, 5.000000e-01
>  %add.i164.i = select i1 %100, float 6.553550e+04, float %.op.i163.i
>  %103 = insertelement <4 x float> %99, float %add.i164.i, i32 2
>  br label %ConvertHSBToRGB.exit
>
>if.then25.i:                                      ; preds = %if.else.i
>  %mul26.i = fmul float %mul6.i, 6.553500e+04
>  %104 = fcmp ogt float %mul26.i, 6.553500e+04
>  %105 = fcmp olt float %mul26.i, 0.000000e+00
>  %106 = select i1 %105, float 0.000000e+00, float %mul26.i
>  %.op.i161.i = fadd float %106, 5.000000e-01
>  %add.i162.i = select i1 %104, float 6.553550e+04, float %.op.i161.i
>  %107 = insertelement <4 x float> %81, float %add.i162.i, i32 0
>  %mul28.i = fmul float %80, 6.553500e+04
>  %108 = fcmp ogt float %mul28.i, 6.553500e+04
>  %109 = fcmp olt float %mul28.i, 0.000000e+00
>  %110 = select i1 %109, float 0.000000e+00, float %mul28.i
>  %.op.i159.i = fadd float %110, 5.000000e-01
>  %add.i160.i = select i1 %108, float 6.553550e+04, float %.op.i159.i
>  %111 = insertelement <4 x float> %107, float %add.i160.i, i32 1
>  %mul30.i = fmul float %mul12.i, 6.553500e+04
>  %112 = fcmp ogt float %mul30.i, 6.553500e+04
>  %113 = fcmp olt float %mul30.i, 0.000000e+00
>  %114 = select i1 %113, float 0.000000e+00, float %mul30.i
>  %.op.i157.i = fadd float %114, 5.000000e-01
>  %add.i158.i = select i1 %112, float 6.553550e+04, float %.op.i157.i
>  %115 = insertelement <4 x float> %111, float %add.i158.i, i32 2
>  br label %ConvertHSBToRGB.exit
>
>if.then35.i:                                      ; preds = %if.else.i
>  %mul36.i = fmul float %mul6.i, 6.553500e+04
>  %116 = fcmp ogt float %mul36.i, 6.553500e+04
>  %117 = fcmp olt float %mul36.i, 0.000000e+00
>  %118 = select i1 %117, float 0.000000e+00, float %mul36.i
>  %.op.i155.i = fadd float %118, 5.000000e-01
>  %add.i156.i = select i1 %116, float 6.553550e+04, float %.op.i155.i
>  %119 = insertelement <4 x float> %81, float %add.i156.i, i32 0
>  %mul38.i = fmul float %mul8.i, 6.553500e+04
>  %120 = fcmp ogt float %mul38.i, 6.553500e+04
>  %121 = fcmp olt float %mul38.i, 0.000000e+00
>  %122 = select i1 %121, float 0.000000e+00, float %mul38.i
>  %.op.i153.i = fadd float %122, 5.000000e-01
>  %add.i154.i = select i1 %120, float 6.553550e+04, float %.op.i153.i
>  %123 = insertelement <4 x float> %119, float %add.i154.i, i32 1
>  %mul40.i = fmul float %80, 6.553500e+04
>  %124 = fcmp ogt float %mul40.i, 6.553500e+04
>  %125 = fcmp olt float %mul40.i, 0.000000e+00
>  %126 = select i1 %125, float 0.000000e+00, float %mul40.i
>  %.op.i151.i = fadd float %126, 5.000000e-01
>  %add.i152.i = select i1 %124, float 6.553550e+04, float %.op.i151.i
>  %127 = insertelement <4 x float> %123, float %add.i152.i, i32 2
>  br label %ConvertHSBToRGB.exit
>
>if.then45.i:                                      ; preds = %if.else.i
>  %mul46.i = fmul float %mul12.i, 6.553500e+04
>  %128 = fcmp ogt float %mul46.i, 6.553500e+04
>  %129 = fcmp olt float %mul46.i, 0.000000e+00
>  %130 = select i1 %129, float 0.000000e+00, float %mul46.i
>  %.op.i149.i = fadd float %130, 5.000000e-01
>  %add.i150.i = select i1 %128, float 6.553550e+04, float %.op.i149.i
>  %131 = insertelement <4 x float> %81, float %add.i150.i, i32 0
>  %mul48.i = fmul float %mul6.i, 6.553500e+04
>  %132 = fcmp ogt float %mul48.i, 6.553500e+04
>  %133 = fcmp olt float %mul48.i, 0.000000e+00
>  %134 = select i1 %133, float 0.000000e+00, float %mul48.i
>  %.op.i147.i = fadd float %134, 5.000000e-01
>  %add.i148.i = select i1 %132, float 6.553550e+04, float %.op.i147.i
>  %135 = insertelement <4 x float> %131, float %add.i148.i, i32 1
>  %mul50.i = fmul float %80, 6.553500e+04
>  %136 = fcmp ogt float %mul50.i, 6.553500e+04
>  %137 = fcmp olt float %mul50.i, 0.000000e+00
>  %138 = select i1 %137, float 0.000000e+00, float %mul50.i
>  %.op.i145.i = fadd float %138, 5.000000e-01
>  %add.i146.i = select i1 %136, float 6.553550e+04, float %.op.i145.i
>  %139 = insertelement <4 x float> %135, float %add.i146.i, i32 2
>  br label %ConvertHSBToRGB.exit
>
>if.else52.i:                                      ; preds = %if.else.i
>  %cmp53.i = icmp eq i32 %conv.i, 5
>  %mul56.i = fmul float %80, 6.553500e+04
>  %140 = fcmp ogt float %mul56.i, 6.553500e+04
>  %141 = fcmp olt float %mul56.i, 0.000000e+00
>  %142 = select i1 %141, float 0.000000e+00, float %mul56.i
>  %.op.i143.i = fadd float %142, 5.000000e-01
>  %add.i144.i = select i1 %140, float 6.553550e+04, float %.op.i143.i
>  %143 = insertelement <4 x float> %81, float %add.i144.i, i32 0
>  br i1 %cmp53.i, label %if.then55.i, label %if.else62.i
>
>if.then55.i:                                      ; preds = %if.else52.i
>  %mul58.i = fmul float %mul6.i, 6.553500e+04
>  %144 = fcmp ogt float %mul58.i, 6.553500e+04
>  %145 = fcmp olt float %mul58.i, 0.000000e+00
>  %146 = select i1 %145, float 0.000000e+00, float %mul58.i
>  %.op.i141.i = fadd float %146, 5.000000e-01
>  %add.i142.i = select i1 %144, float 6.553550e+04, float %.op.i141.i
>  %147 = insertelement <4 x float> %143, float %add.i142.i, i32 1
>  %mul60.i = fmul float %mul8.i, 6.553500e+04
>  %148 = fcmp ogt float %mul60.i, 6.553500e+04
>  %149 = fcmp olt float %mul60.i, 0.000000e+00
>  %150 = select i1 %149, float 0.000000e+00, float %mul60.i
>  %.op.i139.i = fadd float %150, 5.000000e-01
>  %add.i140.i = select i1 %148, float 6.553550e+04, float %.op.i139.i
>  %151 = insertelement <4 x float> %147, float %add.i140.i, i32 2
>  br label %ConvertHSBToRGB.exit
>
>if.else62.i:                                      ; preds = %if.else52.i
>  %mul65.i = fmul float %mul12.i, 6.553500e+04
>  %152 = fcmp ogt float %mul65.i, 6.553500e+04
>  %153 = fcmp olt float %mul65.i, 0.000000e+00
>  %154 = select i1 %153, float 0.000000e+00, float %mul65.i
>  %.op.i137.i = fadd float %154, 5.000000e-01
>  %add.i138.i = select i1 %152, float 6.553550e+04, float %.op.i137.i
>  %155 = insertelement <4 x float> %143, float %add.i138.i, i32 1
>  %mul67.i = fmul float %mul6.i, 6.553500e+04
>  %156 = fcmp ogt float %mul67.i, 6.553500e+04
>  %157 = fcmp olt float %mul67.i, 0.000000e+00
>  %158 = select i1 %157, float 0.000000e+00, float %mul67.i
>  %.op.i135.i = fadd float %158, 5.000000e-01
>  %add.i136.i = select i1 %156, float 6.553550e+04, float %.op.i135.i
>  %159 = insertelement <4 x float> %155, float %add.i136.i, i32 2
>  br label %ConvertHSBToRGB.exit
>
>ConvertHSBToRGB.exit:                             ; preds = %if.then.i35, %if.then15.i, %if.then25.i, %if.then35.i, %if.then45.i, %if.then55.i, %if.else62.i
>  %result.1.i = phi <4 x float> [ %89, %if.then.i35 ], [ %103, %if.then15.i ], [ %115, %if.then25.i ], [ %127, %if.then35.i ], [ %139, %if.then45.i ], [ %151, %if.then55.i ], [ %159, %if.else62.i ]
>  %160 = extractelement <4 x float> %result.1.i, i32 0
>  %161 = fcmp ogt float %160, 6.553500e+04
>  %162 = fcmp olt float %160, 0.000000e+00
>  %163 = select i1 %162, float 0.000000e+00, float %160
>  %.op.i27.i = fadd float %163, 5.000000e-01
>  %add.i28.i = select i1 %161, float 6.553550e+04, float %.op.i27.i
>  store float %add.i28.i, float addrspace(1)* %add.ptr.i44, align 4, !tbaa !12
>  br i1 %cmp.i45, label %if.then.i, label %if.end.i
>
>if.then.i:                                        ; preds = %ConvertHSBToRGB.exit
>  %164 = extractelement <4 x float> %result.1.i, i32 1
>  %165 = fcmp ogt float %164, 6.553500e+04
>  %166 = fcmp olt float %164, 0.000000e+00
>  %167 = select i1 %166, float 0.000000e+00, float %164
>  %.op.i25.i = fadd float %167, 5.000000e-01
>  %add.i26.i = select i1 %165, float 6.553550e+04, float %.op.i25.i
>  %add.ptr.i24.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i44, i32 1
>  store float %add.i26.i, float addrspace(1)* %add.ptr.i24.i, align 4, !tbaa !12
>  %168 = extractelement <4 x float> %result.1.i, i32 2
>  %169 = fcmp ogt float %168, 6.553500e+04
>  %170 = fcmp olt float %168, 0.000000e+00
>  %171 = select i1 %170, float 0.000000e+00, float %168
>  %.op.i22.i = fadd float %171, 5.000000e-01
>  %add.i23.i = select i1 %169, float 6.553550e+04, float %.op.i22.i
>  %add.ptr.i21.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i44, i32 2
>  store float %add.i23.i, float addrspace(1)* %add.ptr.i21.i, align 4, !tbaa !12
>  br label %if.end.i
>
>if.end.i:                                         ; preds = %if.then.i, %ConvertHSBToRGB.exit
>  switch i32 %number_channels, label %WriteAllChannels.exit [
>    i32 4, label %if.then6.i
>    i32 2, label %if.then6.i
>  ]
>
>if.then6.i:                                       ; preds = %if.end.i, %if.end.i
>  %172 = extractelement <4 x float> %result.1.i, i32 3
>  %173 = fcmp ogt float %172, 6.553500e+04
>  %174 = fcmp olt float %172, 0.000000e+00
>  %175 = select i1 %174, float 0.000000e+00, float %172
>  %.op.i.i = fadd float %175, 5.000000e-01
>  %add.i20.i = select i1 %173, float 6.553550e+04, float %.op.i.i
>  %add.ptr.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i44, i32 %number_channels
>  %add.ptr1.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i, i32 -1
>  store float %add.i20.i, float addrspace(1)* %add.ptr1.i.i, align 4, !tbaa !12
>  br label %WriteAllChannels.exit
>
>WriteAllChannels.exit:                            ; preds = %if.end.i, %if.then6.i
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @Histogram(<4 x float> addrspace(1)* noalias nocapture readonly %im, i32 %channel, i32 %colorspace, i32 %method, <4 x i32> addrspace(1)* noalias nocapture %histogram) local_unnamed_addr #4 !kernel_arg_addr_space !30 !kernel_arg_access_qual !31 !kernel_arg_type !32 !kernel_arg_base_type !33 !kernel_arg_type_qual !34 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %2 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %3 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %4 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %x.i = tail call i32 @llvm.r600.read.global.size.x() #16
>  %and = and i32 %channel, 131072
>  %cmp = icmp eq i32 %and, 0
>  br i1 %cmp, label %if.end, label %if.then
>
>if.then:                                          ; preds = %entry
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %2, i32 8
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = mul i32 %y.i.i, %3
>  %9 = add i32 %8, %4
>  %10 = add i32 %9, %7
>  %11 = getelementptr inbounds i8, i8 addrspace(7)* %2, i32 4
>  %12 = bitcast i8 addrspace(7)* %11 to i32 addrspace(7)*
>  %13 = load i32, i32 addrspace(7)* %12, align 4, !tbaa !4
>  %14 = mul i32 %x.i.i, %0
>  %15 = add i32 %14, %1
>  %16 = add i32 %15, %13
>  %mul = mul nsw i32 %10, %x.i
>  %add = add nsw i32 %16, %mul
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %im, i32 %add
>  %17 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %18 = extractelement <4 x float> %17, i32 2
>  %19 = extractelement <4 x float> %17, i32 1
>  %20 = extractelement <4 x float> %17, i32 0
>  switch i32 %colorspace, label %if.end.i [
>    i32 33, label %GetPixelIntensity.exit
>    i32 3, label %GetPixelIntensity.exit
>  ]
>
>if.end.i:                                         ; preds = %if.then
>  switch i32 %method, label %sw.default.i [
>    i32 1, label %sw.bb.i
>    i32 2, label %sw.bb3.i
>    i32 3, label %sw.bb15.i
>    i32 4, label %sw.bb48.i
>    i32 5, label %sw.bb51.i
>    i32 6, label %sw.bb53.i
>    i32 9, label %sw.bb59.i
>    i32 8, label %sw.bb57.i
>  ]
>
>sw.bb.i:                                          ; preds = %if.end.i
>  %add.i26 = fadd float %18, %19
>  %add2.i = fadd float %20, %add.i26
>  %div.i = fdiv float %add2.i, 3.000000e+00, !fpmath !11
>  br label %GetPixelIntensity.exit
>
>sw.bb3.i:                                         ; preds = %if.end.i
>  %cmp4.i = fcmp ogt float %18, %19
>  %cond.i = select i1 %cmp4.i, float %18, float %19
>  %cmp5.i = fcmp ogt float %cond.i, %20
>  %cond.blue.i = select i1 %cmp5.i, float %cond.i, float %20
>  br label %GetPixelIntensity.exit
>
>sw.bb15.i:                                        ; preds = %if.end.i
>  %cmp16.i = fcmp olt float %18, %19
>  %cond20.i = select i1 %cmp16.i, float %18, float %19
>  %cmp21.i = fcmp olt float %cond20.i, %20
>  %cond20.blue.i = select i1 %cmp21.i, float %cond20.i, float %20
>  %cmp31.i = fcmp ogt float %18, %19
>  %cond35.i = select i1 %cmp31.i, float %18, float %19
>  %cmp36.i = fcmp ogt float %cond35.i, %20
>  %cond45.i = select i1 %cmp36.i, float %cond35.i, float %20
>  %add46.i = fadd float %cond20.blue.i, %cond45.i
>  %div47.i = fmul float %add46.i, 5.000000e-01
>  br label %GetPixelIntensity.exit
>
>sw.bb48.i:                                        ; preds = %if.end.i
>  %mul49.i = fmul float %19, %19
>  %21 = tail call float @llvm.fmuladd.f32(float %18, float %18, float %mul49.i) #13
>  %22 = tail call float @llvm.fmuladd.f32(float %20, float %20, float %21) #13
>  %div50.i = fdiv float %22, 1.966050e+05, !fpmath !11
>  br label %GetPixelIntensity.exit
>
>sw.bb51.i:                                        ; preds = %if.end.i
>  %mul52.i = fmul float %19, 0x3FE2C727E0000000
>  %23 = tail call float @llvm.fmuladd.f32(float %18, float 0x3FD3202DA0000000, float %mul52.i) #13
>  %24 = tail call float @llvm.fmuladd.f32(float %20, float 0x3FBD460AA0000000, float %23) #13
>  br label %GetPixelIntensity.exit
>
>sw.bb53.i:                                        ; preds = %if.end.i
>  %mul54.i = fmul float %19, 0x3FE2C727E0000000
>  %25 = tail call float @llvm.fmuladd.f32(float %18, float 0x3FD3202DA0000000, float %mul54.i) #13
>  %26 = tail call float @llvm.fmuladd.f32(float %20, float 0x3FBD460AA0000000, float %25) #13
>  br label %GetPixelIntensity.exit
>
>sw.default.i:                                     ; preds = %if.end.i
>  %mul56.i = fmul float %19, 0x3FE6E29300000000
>  %27 = tail call float @llvm.fmuladd.f32(float %18, float 0x3FCB384FE0000000, float %mul56.i) #13
>  %28 = tail call float @llvm.fmuladd.f32(float %20, float 0x3FB27AC820000000, float %27) #13
>  br label %GetPixelIntensity.exit
>
>sw.bb57.i:                                        ; preds = %if.end.i
>  %mul58.i = fmul float %19, 0x3FE6E29300000000
>  %29 = tail call float @llvm.fmuladd.f32(float %18, float 0x3FCB384FE0000000, float %mul58.i) #13
>  %30 = tail call float @llvm.fmuladd.f32(float %20, float 0x3FB27AC820000000, float %29) #13
>  br label %GetPixelIntensity.exit
>
>sw.bb59.i:                                        ; preds = %if.end.i
>  %mul60.i = fmul float %19, %19
>  %31 = tail call float @llvm.fmuladd.f32(float %18, float %18, float %mul60.i) #13
>  %32 = tail call float @llvm.fmuladd.f32(float %20, float %20, float %31) #13
>  %33 = tail call float @llvm.sqrt.f32(float %32) #14
>  %34 = tail call float @llvm.sqrt.f32(float 3.000000e+00) #14
>  %div62.i = fdiv float %33, %34, !fpmath !11
>  br label %GetPixelIntensity.exit
>
>GetPixelIntensity.exit:                           ; preds = %if.then, %if.then, %sw.bb.i, %sw.bb3.i, %sw.bb15.i, %sw.bb48.i, %sw.bb51.i, %sw.bb53.i, %sw.default.i, %sw.bb57.i, %sw.bb59.i
>  %retval.0.i = phi float [ %18, %if.then ], [ %18, %if.then ], [ %28, %sw.default.i ], [ %30, %sw.bb57.i ], [ %div62.i, %sw.bb59.i ], [ %26, %sw.bb53.i ], [ %24, %sw.bb51.i ], [ %div50.i, %sw.bb48.i ], [ %div47.i, %sw.bb15.i ], [ %cond.blue.i, %sw.bb3.i ], [ %div.i, %sw.bb.i ]
>  %35 = fcmp ogt float %retval.0.i, 6.553500e+04
>  %36 = fcmp olt float %retval.0.i, 0.000000e+00
>  %37 = select i1 %36, float 0.000000e+00, float %retval.0.i
>  %.op.i = fadd float %37, 5.000000e-01
>  %add.i = select i1 %35, float 6.553550e+04, float %.op.i
>  %.inv.i = fcmp oge float %add.i, 6.553500e+04
>  %retval.02.i = select i1 %.inv.i, float 6.553500e+04, float %add.i
>  %38 = fptoui float %retval.02.i to i32
>  %add.ptr = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %histogram, i32 %38, i32 2
>  %39 = atomicrmw add i32 addrspace(1)* %add.ptr, i32 1 seq_cst
>  br label %if.end
>
>if.end:                                           ; preds = %entry, %GetPixelIntensity.exit
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @ContrastStretch(<4 x float> addrspace(1)* noalias nocapture %im, i32 %channel, <4 x float> addrspace(1)* noalias nocapture readonly %stretch_map, <4 x float> %white, <4 x float> %black) local_unnamed_addr #4 !kernel_arg_addr_space !35 !kernel_arg_access_qual !31 !kernel_arg_type !36 !kernel_arg_base_type !37 !kernel_arg_type_qual !38 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %x.i = tail call i32 @llvm.r600.read.global.size.x() #16
>  %mul = mul nsw i32 %16, %x.i
>  %add = add nsw i32 %8, %mul
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %im, i32 %add
>  %17 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %and = and i32 %channel, 1
>  %cmp = icmp eq i32 %and, 0
>  br i1 %cmp, label %if.end11, label %if.then
>
>if.then:                                          ; preds = %entry
>  %18 = extractelement <4 x float> %white, i32 2
>  %19 = extractelement <4 x float> %black, i32 2
>  %cmp5 = fcmp une float %18, %19
>  br i1 %cmp5, label %if.then6, label %if.end11
>
>if.then6:                                         ; preds = %if.then
>  %20 = extractelement <4 x float> %17, i32 2
>  %.inv.i93 = fcmp oge float %20, 6.553500e+04
>  %retval.02.i94 = select i1 %.inv.i93, float 6.553500e+04, float %20
>  %21 = fptoui float %retval.02.i94 to i32
>  %arrayidx9 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %stretch_map, i32 %21
>  %22 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx9, align 16, !tbaa !14
>  %23 = extractelement <4 x float> %22, i32 2
>  br label %if.end11
>
>if.end11:                                         ; preds = %entry, %if.then, %if.then6
>  %red.0 = phi float [ %23, %if.then6 ], [ undef, %if.then ], [ undef, %entry ]
>  %and12 = and i32 %channel, 2
>  %cmp13 = icmp eq i32 %and12, 0
>  br i1 %cmp13, label %if.end24, label %if.then14
>
>if.then14:                                        ; preds = %if.end11
>  %24 = extractelement <4 x float> %white, i32 1
>  %25 = extractelement <4 x float> %black, i32 1
>  %cmp17 = fcmp une float %24, %25
>  br i1 %cmp17, label %if.then18, label %if.end24
>
>if.then18:                                        ; preds = %if.then14
>  %26 = extractelement <4 x float> %17, i32 1
>  %.inv.i91 = fcmp oge float %26, 6.553500e+04
>  %retval.02.i92 = select i1 %.inv.i91, float 6.553500e+04, float %26
>  %27 = fptoui float %retval.02.i92 to i32
>  %arrayidx21 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %stretch_map, i32 %27
>  %28 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx21, align 16, !tbaa !14
>  %29 = extractelement <4 x float> %28, i32 1
>  br label %if.end24
>
>if.end24:                                         ; preds = %if.end11, %if.then14, %if.then18
>  %green.0 = phi float [ %29, %if.then18 ], [ undef, %if.then14 ], [ undef, %if.end11 ]
>  %and25 = and i32 %channel, 4
>  %cmp26 = icmp eq i32 %and25, 0
>  br i1 %cmp26, label %if.end37, label %if.then27
>
>if.then27:                                        ; preds = %if.end24
>  %30 = extractelement <4 x float> %white, i32 0
>  %31 = extractelement <4 x float> %black, i32 0
>  %cmp30 = fcmp une float %30, %31
>  br i1 %cmp30, label %if.then31, label %if.end37
>
>if.then31:                                        ; preds = %if.then27
>  %32 = extractelement <4 x float> %17, i32 0
>  %.inv.i89 = fcmp oge float %32, 6.553500e+04
>  %retval.02.i90 = select i1 %.inv.i89, float 6.553500e+04, float %32
>  %33 = fptoui float %retval.02.i90 to i32
>  %arrayidx34 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %stretch_map, i32 %33
>  %34 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx34, align 16, !tbaa !14
>  %35 = extractelement <4 x float> %34, i32 0
>  br label %if.end37
>
>if.end37:                                         ; preds = %if.end24, %if.then27, %if.then31
>  %blue.0 = phi float [ %35, %if.then31 ], [ undef, %if.then27 ], [ undef, %if.end24 ]
>  %and38 = and i32 %channel, 16
>  %cmp39 = icmp eq i32 %and38, 0
>  br i1 %cmp39, label %if.end50, label %if.then40
>
>if.then40:                                        ; preds = %if.end37
>  %36 = extractelement <4 x float> %white, i32 3
>  %37 = extractelement <4 x float> %black, i32 3
>  %cmp43 = fcmp une float %36, %37
>  br i1 %cmp43, label %if.then44, label %if.end50
>
>if.then44:                                        ; preds = %if.then40
>  %38 = extractelement <4 x float> %17, i32 3
>  %.inv.i = fcmp oge float %38, 6.553500e+04
>  %retval.02.i = select i1 %.inv.i, float 6.553500e+04, float %38
>  %39 = fptoui float %retval.02.i to i32
>  %arrayidx47 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %stretch_map, i32 %39
>  %40 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx47, align 16, !tbaa !14
>  %41 = extractelement <4 x float> %40, i32 3
>  br label %if.end50
>
>if.end50:                                         ; preds = %if.end37, %if.then40, %if.then44
>  %alpha.0 = phi float [ %41, %if.then44 ], [ undef, %if.then40 ], [ undef, %if.end37 ]
>  %vecinit = insertelement <4 x float> undef, float %blue.0, i32 0
>  %vecinit51 = insertelement <4 x float> %vecinit, float %green.0, i32 1
>  %vecinit52 = insertelement <4 x float> %vecinit51, float %red.0, i32 2
>  %vecinit53 = insertelement <4 x float> %vecinit52, float %alpha.0, i32 3
>  store <4 x float> %vecinit53, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @ConvolveOptimized(<4 x float> addrspace(1)* nocapture readonly %input, <4 x float> addrspace(1)* nocapture %output, i32 %imageWidth, i32 %imageHeight, float addrspace(4)* nocapture readonly %filter, i32 %filterWidth, i32 %filterHeight, i32 %matte, i32 %channel, <4 x float> addrspace(3)* nocapture %pixelLocalCache, float addrspace(3)* nocapture %filterCache) local_unnamed_addr #4 !kernel_arg_addr_space !39 !kernel_arg_access_qual !16 !kernel_arg_type !40 !kernel_arg_base_type !41 !kernel_arg_type_qual !42 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %17 = urem i32 %8, %x.i.i
>  %mul = sub i32 %8, %17
>  %18 = insertelement <2 x i32> undef, i32 %mul, i32 0
>  %19 = urem i32 %16, %y.i.i
>  %mul7 = sub i32 %16, %19
>  %20 = insertelement <2 x i32> %18, i32 %mul7, i32 1
>  %sub = add i32 %filterWidth, -1
>  %div8 = lshr i32 %sub, 1
>  %21 = insertelement <2 x i32> undef, i32 %div8, i32 0
>  %sub9 = add i32 %filterHeight, -1
>  %div10 = lshr i32 %sub9, 1
>  %22 = insertelement <2 x i32> %21, i32 %div10, i32 1
>  %sub11 = sub <2 x i32> %20, %22
>  %sub13 = add i32 %sub, %x.i.i
>  %sub16 = add i32 %sub9, %y.i.i
>  %mul19 = mul i32 %11, %x.i.i
>  %add21 = add i32 %mul19, %2
>  %mul22 = mul nsw i32 %sub16, %sub13
>  %mul25 = mul i32 %x.i.i, %y.i.i
>  %cmp327 = icmp slt i32 %add21, %mul22
>  br i1 %cmp327, label %for.body.lr.ph, label %for.cond35.preheader
>
>for.body.lr.ph:                                   ; preds = %entry
>  %sub.i292 = add nsw i32 %imageWidth, -1
>  %sub.i = add nsw i32 %imageHeight, -1
>  br label %for.body
>
>for.cond35.preheader:                             ; preds = %for.body, %entry
>  %mul36 = mul i32 %filterHeight, %filterWidth
>  %cmp37325 = icmp ult i32 %add21, %mul36
>  br i1 %cmp37325, label %for.body39, label %for.cond.cleanup38
>
>for.body:                                         ; preds = %for.body.lr.ph, %for.body
>  %i.0328 = phi i32 [ %add21, %for.body.lr.ph ], [ %add33, %for.body ]
>  %div26 = sdiv i32 %i.0328, %sub13
>  %23 = mul i32 %div26, %sub13
>  %24 = sub i32 %i.0328, %23
>  %25 = insertelement <2 x i32> undef, i32 %24, i32 0
>  %26 = insertelement <2 x i32> %25, i32 %div26, i32 1
>  %add27 = add <2 x i32> %26, %sub11
>  %27 = extractelement <2 x i32> %add27, i32 0
>  %28 = icmp slt i32 %27, %imageWidth
>  %29 = icmp sgt i32 %27, 0
>  %30 = select i1 %29, i32 %27, i32 0
>  %31 = select i1 %28, i32 %30, i32 %sub.i292
>  %32 = extractelement <2 x i32> %add27, i32 1
>  %33 = icmp slt i32 %32, %imageHeight
>  %34 = icmp sgt i32 %32, 0
>  %35 = select i1 %34, i32 %32, i32 0
>  %36 = select i1 %33, i32 %35, i32 %sub.i
>  %mul30 = mul i32 %36, %imageWidth
>  %add31 = add i32 %mul30, %31
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %input, i32 %add31
>  %37 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %arrayidx32 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixelLocalCache, i32 %i.0328
>  store <4 x float> %37, <4 x float> addrspace(3)* %arrayidx32, align 16, !tbaa !14
>  %add33 = add nsw i32 %i.0328, %mul25
>  %cmp = icmp slt i32 %add33, %mul22
>  br i1 %cmp, label %for.body, label %for.cond35.preheader
>
>for.cond.cleanup38:                               ; preds = %for.body39, %for.cond35.preheader
>  tail call void @llvm.r600.group.barrier() #13
>  %add46 = add i32 %mul, %2
>  %add48 = add i32 %mul7, %11
>  %cmp49 = icmp ult i32 %add46, %imageWidth
>  %cmp50 = icmp ult i32 %add48, %imageHeight
>  %or.cond285 = and i1 %cmp49, %cmp50
>  br i1 %or.cond285, label %if.end, label %cleanup
>
>for.body39:                                       ; preds = %for.cond35.preheader, %for.body39
>  %i34.0326 = phi i32 [ %add43, %for.body39 ], [ %add21, %for.cond35.preheader ]
>  %arrayidx40 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %i34.0326
>  %38 = bitcast float addrspace(4)* %arrayidx40 to i32 addrspace(4)*
>  %39 = load i32, i32 addrspace(4)* %38, align 4, !tbaa !12
>  %arrayidx41 = getelementptr inbounds float, float addrspace(3)* %filterCache, i32 %i34.0326
>  %40 = bitcast float addrspace(3)* %arrayidx41 to i32 addrspace(3)*
>  store i32 %39, i32 addrspace(3)* %40, align 4, !tbaa !12
>  %add43 = add nsw i32 %i34.0326, %mul25
>  %cmp37 = icmp ult i32 %add43, %mul36
>  br i1 %cmp37, label %for.body39, label %for.cond.cleanup38
>
>if.end:                                           ; preds = %for.cond.cleanup38
>  %and = and i32 %channel, 16
>  %cmp51 = icmp eq i32 %and, 0
>  %cmp53 = icmp eq i32 %matte, 0
>  %or.cond = or i1 %cmp53, %cmp51
>  %cmp57301 = icmp ne i32 %filterHeight, 0
>  br i1 %or.cond, label %for.cond56.preheader, label %for.cond86.preheader
>
>for.cond86.preheader:                             ; preds = %if.end
>  br i1 %cmp57301, label %for.cond93.preheader.lr.ph, label %for.cond.cleanup88
>
>for.cond93.preheader.lr.ph:                       ; preds = %for.cond86.preheader
>  %cmp94308 = icmp eq i32 %filterWidth, 0
>  br label %for.cond93.preheader
>
>for.cond56.preheader:                             ; preds = %if.end
>  br i1 %cmp57301, label %for.cond62.preheader.lr.ph, label %if.end121
>
>for.cond62.preheader.lr.ph:                       ; preds = %for.cond56.preheader
>  %cmp63293 = icmp eq i32 %filterWidth, 0
>  br label %for.cond62.preheader
>
>for.cond62.preheader:                             ; preds = %for.cond.cleanup64, %for.cond62.preheader.lr.ph
>  %j.0306 = phi i32 [ 0, %for.cond62.preheader.lr.ph ], [ %inc81, %for.cond.cleanup64 ]
>  %cacheIndexY.0305 = phi i32 [ %11, %for.cond62.preheader.lr.ph ], [ %inc79, %for.cond.cleanup64 ]
>  %sum.0303 = phi <4 x float> [ zeroinitializer, %for.cond62.preheader.lr.ph ], [ %sum.1.lcssa, %for.cond.cleanup64 ]
>  %filterIndex.0302 = phi i32 [ 0, %for.cond62.preheader.lr.ph ], [ %filterIndex.1.lcssa, %for.cond.cleanup64 ]
>  br i1 %cmp63293, label %for.cond.cleanup64, label %for.body65.lr.ph
>
>for.body65.lr.ph:                                 ; preds = %for.cond62.preheader
>  %mul66 = mul nsw i32 %cacheIndexY.0305, %sub13
>  br label %for.body65
>
>for.cond.cleanup64.loopexit:                      ; preds = %for.body65
>  %41 = add i32 %filterIndex.0302, %filterWidth
>  br label %for.cond.cleanup64
>
>for.cond.cleanup64:                               ; preds = %for.cond.cleanup64.loopexit, %for.cond62.preheader
>  %filterIndex.1.lcssa = phi i32 [ %filterIndex.0302, %for.cond62.preheader ], [ %41, %for.cond.cleanup64.loopexit ]
>  %sum.1.lcssa = phi <4 x float> [ %sum.0303, %for.cond62.preheader ], [ %59, %for.cond.cleanup64.loopexit ]
>  %inc79 = add nuw nsw i32 %cacheIndexY.0305, 1
>  %inc81 = add nuw nsw i32 %j.0306, 1
>  %exitcond330 = icmp eq i32 %inc81, %filterHeight
>  br i1 %exitcond330, label %if.end121, label %for.cond62.preheader
>
>for.body65:                                       ; preds = %for.body65, %for.body65.lr.ph
>  %i61.0298 = phi i32 [ 0, %for.body65.lr.ph ], [ %inc77, %for.body65 ]
>  %cacheIndexX.0297 = phi i32 [ %2, %for.body65.lr.ph ], [ %inc75, %for.body65 ]
>  %sum.1295 = phi <4 x float> [ %sum.0303, %for.body65.lr.ph ], [ %59, %for.body65 ]
>  %filterIndex.1294 = phi i32 [ %filterIndex.0302, %for.body65.lr.ph ], [ %inc, %for.body65 ]
>  %add67 = add nsw i32 %cacheIndexX.0297, %mul66
>  %arrayidx68 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixelLocalCache, i32 %add67
>  %42 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx68, align 16, !tbaa !14
>  %arrayidx69 = getelementptr inbounds float, float addrspace(3)* %filterCache, i32 %filterIndex.1294
>  %43 = load float, float addrspace(3)* %arrayidx69, align 4, !tbaa !12
>  %44 = extractelement <4 x float> %42, i32 0
>  %45 = extractelement <4 x float> %sum.1295, i32 0
>  %46 = tail call float @llvm.fmuladd.f32(float %43, float %44, float %45)
>  %47 = insertelement <4 x float> undef, float %46, i32 0
>  %48 = extractelement <4 x float> %42, i32 1
>  %49 = extractelement <4 x float> %sum.1295, i32 1
>  %50 = tail call float @llvm.fmuladd.f32(float %43, float %48, float %49)
>  %51 = insertelement <4 x float> %47, float %50, i32 1
>  %52 = extractelement <4 x float> %42, i32 2
>  %53 = extractelement <4 x float> %sum.1295, i32 2
>  %54 = tail call float @llvm.fmuladd.f32(float %43, float %52, float %53)
>  %55 = insertelement <4 x float> %51, float %54, i32 2
>  %56 = extractelement <4 x float> %42, i32 3
>  %57 = extractelement <4 x float> %sum.1295, i32 3
>  %58 = tail call float @llvm.fmuladd.f32(float %43, float %56, float %57)
>  %59 = insertelement <4 x float> %55, float %58, i32 3
>  %inc = add nsw i32 %filterIndex.1294, 1
>  %inc75 = add nuw nsw i32 %cacheIndexX.0297, 1
>  %inc77 = add nuw nsw i32 %i61.0298, 1
>  %exitcond = icmp eq i32 %inc77, %filterWidth
>  br i1 %exitcond, label %for.cond.cleanup64.loopexit, label %for.body65
>
>for.cond93.preheader:                             ; preds = %for.cond.cleanup95, %for.cond93.preheader.lr.ph
>  %j85.0322 = phi i32 [ 0, %for.cond93.preheader.lr.ph ], [ %inc117, %for.cond.cleanup95 ]
>  %cacheIndexY83.0321 = phi i32 [ %11, %for.cond93.preheader.lr.ph ], [ %inc115, %for.cond.cleanup95 ]
>  %gamma.2320 = phi float [ 0.000000e+00, %for.cond93.preheader.lr.ph ], [ %gamma.3.lcssa, %for.cond.cleanup95 ]
>  %sum.2319 = phi <4 x float> [ zeroinitializer, %for.cond93.preheader.lr.ph ], [ %sum.3.lcssa, %for.cond.cleanup95 ]
>  %filterIndex.2318 = phi i32 [ 0, %for.cond93.preheader.lr.ph ], [ %filterIndex.3.lcssa, %for.cond.cleanup95 ]
>  br i1 %cmp94308, label %for.cond.cleanup95, label %for.body96.lr.ph
>
>for.body96.lr.ph:                                 ; preds = %for.cond93.preheader
>  %mul98 = mul nsw i32 %cacheIndexY83.0321, %sub13
>  br label %for.body96
>
>for.cond.cleanup88:                               ; preds = %for.cond.cleanup95, %for.cond86.preheader
>  %sum.2.lcssa = phi <4 x float> [ zeroinitializer, %for.cond86.preheader ], [ %sum.3.lcssa, %for.cond.cleanup95 ]
>  %gamma.2.lcssa = phi float [ 0.000000e+00, %for.cond86.preheader ], [ %gamma.3.lcssa, %for.cond.cleanup95 ]
>  %cmp.i = fcmp olt float %gamma.2.lcssa, 0.000000e+00
>  %cond.i = select i1 %cmp.i, float -1.000000e+00, float 1.000000e+00
>  %mul.i = fmul float %gamma.2.lcssa, %cond.i
>  %cmp1.i = fcmp ult float %mul.i, 0.000000e+00
>  %div.i = fdiv float 1.000000e+00, %gamma.2.lcssa
>  %mul2.i = fmul float %cond.i, 0x7FF0000000000000
>  %cond3.i = select i1 %cmp1.i, float %mul2.i, float %div.i
>  %splat.splatinsert = insertelement <3 x float> undef, float %cond3.i, i32 0
>  %splat.splat = shufflevector <3 x float> %splat.splatinsert, <3 x float> undef, <3 x i32> zeroinitializer
>  %60 = shufflevector <4 x float> %sum.2.lcssa, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
>  %mul120 = fmul <3 x float> %60, %splat.splat
>  %61 = shufflevector <3 x float> %mul120, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
>  %62 = shufflevector <4 x float> %sum.2.lcssa, <4 x float> %61, <4 x i32> <i32 4, i32 5, i32 6, i32 3>
>  br label %if.end121
>
>for.cond.cleanup95.loopexit:                      ; preds = %for.body96
>  %63 = add i32 %filterIndex.2318, %filterWidth
>  br label %for.cond.cleanup95
>
>for.cond.cleanup95:                               ; preds = %for.cond.cleanup95.loopexit, %for.cond93.preheader
>  %filterIndex.3.lcssa = phi i32 [ %filterIndex.2318, %for.cond93.preheader ], [ %63, %for.cond.cleanup95.loopexit ]
>  %sum.3.lcssa = phi <4 x float> [ %sum.2319, %for.cond93.preheader ], [ %81, %for.cond.cleanup95.loopexit ]
>  %gamma.3.lcssa = phi float [ %gamma.2320, %for.cond93.preheader ], [ %add109, %for.cond.cleanup95.loopexit ]
>  %inc115 = add nuw nsw i32 %cacheIndexY83.0321, 1
>  %inc117 = add nuw nsw i32 %j85.0322, 1
>  %exitcond332 = icmp eq i32 %inc117, %filterHeight
>  br i1 %exitcond332, label %for.cond.cleanup88, label %for.cond93.preheader
>
>for.body96:                                       ; preds = %for.body96, %for.body96.lr.ph
>  %i92.0313 = phi i32 [ 0, %for.body96.lr.ph ], [ %inc113, %for.body96 ]
>  %cacheIndexX90.0312 = phi i32 [ %2, %for.body96.lr.ph ], [ %inc111, %for.body96 ]
>  %gamma.3311 = phi float [ %gamma.2320, %for.body96.lr.ph ], [ %add109, %for.body96 ]
>  %sum.3310 = phi <4 x float> [ %sum.2319, %for.body96.lr.ph ], [ %81, %for.body96 ]
>  %filterIndex.3309 = phi i32 [ %filterIndex.2318, %for.body96.lr.ph ], [ %inc110, %for.body96 ]
>  %add99 = add nsw i32 %cacheIndexX90.0312, %mul98
>  %arrayidx100 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixelLocalCache, i32 %add99
>  %64 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx100, align 16, !tbaa !14
>  %65 = extractelement <4 x float> %64, i32 3
>  %mul101 = fmul float %65, 0x3EEF751040000000
>  %arrayidx103 = getelementptr inbounds float, float addrspace(3)* %filterCache, i32 %filterIndex.3309
>  %66 = load float, float addrspace(3)* %arrayidx103, align 4, !tbaa !12
>  %mul104 = fmul float %66, %mul101
>  %67 = extractelement <4 x float> %64, i32 0
>  %68 = extractelement <4 x float> %sum.3310, i32 0
>  %69 = tail call float @llvm.fmuladd.f32(float %mul104, float %67, float %68)
>  %70 = insertelement <4 x float> undef, float %69, i32 0
>  %71 = extractelement <4 x float> %64, i32 1
>  %72 = extractelement <4 x float> %sum.3310, i32 1
>  %73 = tail call float @llvm.fmuladd.f32(float %mul104, float %71, float %72)
>  %74 = insertelement <4 x float> %70, float %73, i32 1
>  %75 = extractelement <4 x float> %64, i32 2
>  %76 = extractelement <4 x float> %sum.3310, i32 2
>  %77 = tail call float @llvm.fmuladd.f32(float %mul104, float %75, float %76)
>  %78 = insertelement <4 x float> %74, float %77, i32 2
>  %79 = extractelement <4 x float> %sum.3310, i32 3
>  %80 = tail call float @llvm.fmuladd.f32(float %66, float %65, float %79)
>  %81 = insertelement <4 x float> %78, float %80, i32 3
>  %add109 = fadd float %gamma.3311, %mul104
>  %inc110 = add nsw i32 %filterIndex.3309, 1
>  %inc111 = add nuw nsw i32 %cacheIndexX90.0312, 1
>  %inc113 = add nuw nsw i32 %i92.0313, 1
>  %exitcond331 = icmp eq i32 %inc113, %filterWidth
>  br i1 %exitcond331, label %for.cond.cleanup95.loopexit, label %for.body96
>
>if.end121:                                        ; preds = %for.cond.cleanup64, %for.cond56.preheader, %for.cond.cleanup88
>  %sum.4 = phi <4 x float> [ %62, %for.cond.cleanup88 ], [ zeroinitializer, %for.cond56.preheader ], [ %sum.1.lcssa, %for.cond.cleanup64 ]
>  %82 = extractelement <4 x float> %sum.4, i32 0
>  %83 = fcmp ogt float %82, 6.553500e+04
>  %84 = fcmp olt float %82, 0.000000e+00
>  %85 = select i1 %84, float 0.000000e+00, float %82
>  %.op.i290 = fadd float %85, 5.000000e-01
>  %add.i291 = select i1 %83, float 6.553550e+04, float %.op.i290
>  %86 = insertelement <4 x float> undef, float %add.i291, i32 0
>  %87 = extractelement <4 x float> %sum.4, i32 1
>  %88 = fcmp ogt float %87, 6.553500e+04
>  %89 = fcmp olt float %87, 0.000000e+00
>  %90 = select i1 %89, float 0.000000e+00, float %87
>  %.op.i288 = fadd float %90, 5.000000e-01
>  %add.i289 = select i1 %88, float 6.553550e+04, float %.op.i288
>  %91 = insertelement <4 x float> %86, float %add.i289, i32 1
>  %92 = extractelement <4 x float> %sum.4, i32 2
>  %93 = fcmp ogt float %92, 6.553500e+04
>  %94 = fcmp olt float %92, 0.000000e+00
>  %95 = select i1 %94, float 0.000000e+00, float %92
>  %.op.i286 = fadd float %95, 5.000000e-01
>  %add.i287 = select i1 %93, float 6.553550e+04, float %.op.i286
>  %96 = insertelement <4 x float> %91, float %add.i287, i32 2
>  br i1 %cmp51, label %cond.false, label %cond.true
>
>cond.true:                                        ; preds = %if.end121
>  %97 = extractelement <4 x float> %sum.4, i32 3
>  %98 = fcmp ogt float %97, 6.553500e+04
>  %99 = fcmp olt float %97, 0.000000e+00
>  %100 = select i1 %99, float 0.000000e+00, float %97
>  %.op.i = fadd float %100, 5.000000e-01
>  %add.i = select i1 %98, float 6.553550e+04, float %.op.i
>  %.pre = mul i32 %add48, %imageWidth
>  %.pre333 = add i32 %.pre, %add46
>  br label %cond.end
>
>cond.false:                                       ; preds = %if.end121
>  %mul128 = mul i32 %add48, %imageWidth
>  %add129 = add i32 %mul128, %add46
>  %arrayidx130 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %input, i32 %add129
>  %101 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx130, align 16
>  %102 = extractelement <4 x float> %101, i32 3
>  br label %cond.end
>
>cond.end:                                         ; preds = %cond.false, %cond.true
>  %add132.pre-phi = phi i32 [ %add129, %cond.false ], [ %.pre333, %cond.true ]
>  %cond = phi float [ %102, %cond.false ], [ %add.i, %cond.true ]
>  %103 = insertelement <4 x float> %96, float %cond, i32 3
>  %arrayidx133 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %output, i32 %add132.pre-phi
>  store <4 x float> %103, <4 x float> addrspace(1)* %arrayidx133, align 16, !tbaa !14
>  br label %cleanup
>
>cleanup:                                          ; preds = %for.cond.cleanup38, %cond.end
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @Convolve(<4 x float> addrspace(1)* nocapture readonly %input, <4 x float> addrspace(1)* nocapture %output, i32 %imageWidth, i32 %imageHeight, float addrspace(4)* nocapture readonly %filter, i32 %filterWidth, i32 %filterHeight, i32 %matte, i32 %channel) local_unnamed_addr #4 !kernel_arg_addr_space !43 !kernel_arg_access_qual !20 !kernel_arg_type !44 !kernel_arg_base_type !45 !kernel_arg_type_qual !46 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %cmp = icmp ult i32 %8, %imageWidth
>  %cmp2 = icmp ult i32 %16, %imageHeight
>  %or.cond200 = and i1 %cmp, %cmp2
>  br i1 %or.cond200, label %if.end, label %cleanup
>
>if.end:                                           ; preds = %entry
>  %sub = add i32 %filterWidth, -1
>  %div = lshr i32 %sub, 1
>  %sub3 = add i32 %filterHeight, -1
>  %div4 = lshr i32 %sub3, 1
>  %and = and i32 %channel, 16
>  %cmp5 = icmp eq i32 %and, 0
>  %cmp7 = icmp eq i32 %matte, 0
>  %or.cond = or i1 %cmp7, %cmp5
>  %cmp9219 = icmp ne i32 %filterHeight, 0
>  br i1 %or.cond, label %for.cond.preheader, label %for.cond31.preheader
>
>for.cond31.preheader:                             ; preds = %if.end
>  br i1 %cmp9219, label %for.body34.lr.ph, label %for.cond.cleanup33
>
>for.body34.lr.ph:                                 ; preds = %for.cond31.preheader
>  %sub36 = sub nsw i32 %16, %div4
>  %sub.i207 = add nsw i32 %imageHeight, -1
>  %cmp41226 = icmp eq i32 %filterWidth, 0
>  %sub44 = sub nsw i32 %8, %div
>  %sub.i = add nsw i32 %imageWidth, -1
>  br label %for.body34
>
>for.cond.preheader:                               ; preds = %if.end
>  br i1 %cmp9219, label %for.body.lr.ph, label %if.end69
>
>for.body.lr.ph:                                   ; preds = %for.cond.preheader
>  %sub10 = sub nsw i32 %16, %div4
>  %sub.i209 = add nsw i32 %imageHeight, -1
>  %cmp13210 = icmp eq i32 %filterWidth, 0
>  %sub16 = sub nsw i32 %8, %div
>  %sub.i208 = add nsw i32 %imageWidth, -1
>  br label %for.body
>
>for.body:                                         ; preds = %for.cond.cleanup14, %for.body.lr.ph
>  %filterIndex.0224 = phi i32 [ 0, %for.body.lr.ph ], [ %filterIndex.1.lcssa, %for.cond.cleanup14 ]
>  %sum.0223 = phi <4 x float> [ zeroinitializer, %for.body.lr.ph ], [ %sum.1.lcssa, %for.cond.cleanup14 ]
>  %inputPixelIndex.0221 = phi <2 x i32> [ undef, %for.body.lr.ph ], [ %inputPixelIndex.1.lcssa, %for.cond.cleanup14 ]
>  %j.0220 = phi i32 [ 0, %for.body.lr.ph ], [ %inc28, %for.cond.cleanup14 ]
>  %add = add nsw i32 %j.0220, %sub10
>  %17 = icmp slt i32 %add, %imageHeight
>  %18 = icmp sgt i32 %add, 0
>  %19 = select i1 %18, i32 %add, i32 0
>  %20 = select i1 %17, i32 %19, i32 %sub.i209
>  %21 = insertelement <2 x i32> %inputPixelIndex.0221, i32 %20, i32 1
>  br i1 %cmp13210, label %for.cond.cleanup14, label %for.body15
>
>for.cond.cleanup14.loopexit:                      ; preds = %for.body15
>  %22 = add i32 %filterIndex.0224, %filterWidth
>  br label %for.cond.cleanup14
>
>for.cond.cleanup14:                               ; preds = %for.cond.cleanup14.loopexit, %for.body
>  %inputPixelIndex.1.lcssa = phi <2 x i32> [ %21, %for.body ], [ %27, %for.cond.cleanup14.loopexit ]
>  %sum.1.lcssa = phi <4 x float> [ %sum.0223, %for.body ], [ %46, %for.cond.cleanup14.loopexit ]
>  %filterIndex.1.lcssa = phi i32 [ %filterIndex.0224, %for.body ], [ %22, %for.cond.cleanup14.loopexit ]
>  %inc28 = add nuw nsw i32 %j.0220, 1
>  %exitcond247 = icmp eq i32 %inc28, %filterHeight
>  br i1 %exitcond247, label %if.end69, label %for.body
>
>for.body15:                                       ; preds = %for.body, %for.body15
>  %filterIndex.1215 = phi i32 [ %inc, %for.body15 ], [ %filterIndex.0224, %for.body ]
>  %sum.1214 = phi <4 x float> [ %46, %for.body15 ], [ %sum.0223, %for.body ]
>  %i.0212 = phi i32 [ %inc26, %for.body15 ], [ 0, %for.body ]
>  %inputPixelIndex.1211 = phi <2 x i32> [ %27, %for.body15 ], [ %21, %for.body ]
>  %add17 = add nsw i32 %i.0212, %sub16
>  %23 = icmp slt i32 %add17, %imageWidth
>  %24 = icmp sgt i32 %add17, 0
>  %25 = select i1 %24, i32 %add17, i32 0
>  %26 = select i1 %23, i32 %25, i32 %sub.i208
>  %27 = insertelement <2 x i32> %inputPixelIndex.1211, i32 %26, i32 0
>  %28 = extractelement <2 x i32> %inputPixelIndex.1211, i32 1
>  %mul = mul i32 %28, %imageWidth
>  %add19 = add i32 %26, %mul
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %input, i32 %add19
>  %29 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %arrayidx20 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %filterIndex.1215
>  %30 = load float, float addrspace(4)* %arrayidx20, align 4, !tbaa !12
>  %31 = extractelement <4 x float> %29, i32 0
>  %32 = extractelement <4 x float> %sum.1214, i32 0
>  %33 = tail call float @llvm.fmuladd.f32(float %30, float %31, float %32)
>  %34 = insertelement <4 x float> undef, float %33, i32 0
>  %35 = extractelement <4 x float> %29, i32 1
>  %36 = extractelement <4 x float> %sum.1214, i32 1
>  %37 = tail call float @llvm.fmuladd.f32(float %30, float %35, float %36)
>  %38 = insertelement <4 x float> %34, float %37, i32 1
>  %39 = extractelement <4 x float> %29, i32 2
>  %40 = extractelement <4 x float> %sum.1214, i32 2
>  %41 = tail call float @llvm.fmuladd.f32(float %30, float %39, float %40)
>  %42 = insertelement <4 x float> %38, float %41, i32 2
>  %43 = extractelement <4 x float> %29, i32 3
>  %44 = extractelement <4 x float> %sum.1214, i32 3
>  %45 = tail call float @llvm.fmuladd.f32(float %30, float %43, float %44)
>  %46 = insertelement <4 x float> %42, float %45, i32 3
>  %inc = add nsw i32 %filterIndex.1215, 1
>  %inc26 = add nuw nsw i32 %i.0212, 1
>  %exitcond = icmp eq i32 %inc26, %filterWidth
>  br i1 %exitcond, label %for.cond.cleanup14.loopexit, label %for.body15
>
>for.cond.cleanup33:                               ; preds = %for.cond.cleanup42, %for.cond31.preheader
>  %gamma.2.lcssa = phi float [ 0.000000e+00, %for.cond31.preheader ], [ %gamma.3.lcssa, %for.cond.cleanup42 ]
>  %sum.2.lcssa = phi <4 x float> [ zeroinitializer, %for.cond31.preheader ], [ %sum.3.lcssa, %for.cond.cleanup42 ]
>  %cmp.i = fcmp olt float %gamma.2.lcssa, 0.000000e+00
>  %cond.i = select i1 %cmp.i, float -1.000000e+00, float 1.000000e+00
>  %mul.i = fmul float %gamma.2.lcssa, %cond.i
>  %cmp1.i = fcmp ult float %mul.i, 0.000000e+00
>  %div.i = fdiv float 1.000000e+00, %gamma.2.lcssa
>  %mul2.i = fmul float %cond.i, 0x7FF0000000000000
>  %cond3.i = select i1 %cmp1.i, float %mul2.i, float %div.i
>  %splat.splatinsert = insertelement <3 x float> undef, float %cond3.i, i32 0
>  %splat.splat = shufflevector <3 x float> %splat.splatinsert, <3 x float> undef, <3 x i32> zeroinitializer
>  %47 = shufflevector <4 x float> %sum.2.lcssa, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
>  %mul68 = fmul <3 x float> %47, %splat.splat
>  %48 = shufflevector <3 x float> %mul68, <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef>
>  %49 = shufflevector <4 x float> %sum.2.lcssa, <4 x float> %48, <4 x i32> <i32 4, i32 5, i32 6, i32 3>
>  br label %if.end69
>
>for.body34:                                       ; preds = %for.cond.cleanup42, %for.body34.lr.ph
>  %filterIndex.2241 = phi i32 [ 0, %for.body34.lr.ph ], [ %filterIndex.3.lcssa, %for.cond.cleanup42 ]
>  %inputPixelIndex35.0240 = phi <2 x i32> [ undef, %for.body34.lr.ph ], [ %inputPixelIndex35.1.lcssa, %for.cond.cleanup42 ]
>  %j30.0239 = phi i32 [ 0, %for.body34.lr.ph ], [ %inc65, %for.cond.cleanup42 ]
>  %sum.2238 = phi <4 x float> [ zeroinitializer, %for.body34.lr.ph ], [ %sum.3.lcssa, %for.cond.cleanup42 ]
>  %gamma.2237 = phi float [ 0.000000e+00, %for.body34.lr.ph ], [ %gamma.3.lcssa, %for.cond.cleanup42 ]
>  %add37 = add nsw i32 %j30.0239, %sub36
>  %50 = icmp slt i32 %add37, %imageHeight
>  %51 = icmp sgt i32 %add37, 0
>  %52 = select i1 %51, i32 %add37, i32 0
>  %53 = select i1 %50, i32 %52, i32 %sub.i207
>  %54 = insertelement <2 x i32> %inputPixelIndex35.0240, i32 %53, i32 1
>  br i1 %cmp41226, label %for.cond.cleanup42, label %for.body43
>
>for.cond.cleanup42.loopexit:                      ; preds = %for.body43
>  %55 = add i32 %filterIndex.2241, %filterWidth
>  br label %for.cond.cleanup42
>
>for.cond.cleanup42:                               ; preds = %for.cond.cleanup42.loopexit, %for.body34
>  %gamma.3.lcssa = phi float [ %gamma.2237, %for.body34 ], [ %add59, %for.cond.cleanup42.loopexit ]
>  %sum.3.lcssa = phi <4 x float> [ %sum.2238, %for.body34 ], [ %79, %for.cond.cleanup42.loopexit ]
>  %inputPixelIndex35.1.lcssa = phi <2 x i32> [ %54, %for.body34 ], [ %60, %for.cond.cleanup42.loopexit ]
>  %filterIndex.3.lcssa = phi i32 [ %filterIndex.2241, %for.body34 ], [ %55, %for.cond.cleanup42.loopexit ]
>  %inc65 = add nuw nsw i32 %j30.0239, 1
>  %exitcond249 = icmp eq i32 %inc65, %filterHeight
>  br i1 %exitcond249, label %for.cond.cleanup33, label %for.body34
>
>for.body43:                                       ; preds = %for.body34, %for.body43
>  %filterIndex.3231 = phi i32 [ %inc60, %for.body43 ], [ %filterIndex.2241, %for.body34 ]
>  %i39.0230 = phi i32 [ %inc62, %for.body43 ], [ 0, %for.body34 ]
>  %inputPixelIndex35.1229 = phi <2 x i32> [ %60, %for.body43 ], [ %54, %for.body34 ]
>  %sum.3228 = phi <4 x float> [ %79, %for.body43 ], [ %sum.2238, %for.body34 ]
>  %gamma.3227 = phi float [ %add59, %for.body43 ], [ %gamma.2237, %for.body34 ]
>  %add45 = add nsw i32 %i39.0230, %sub44
>  %56 = icmp slt i32 %add45, %imageWidth
>  %57 = icmp sgt i32 %add45, 0
>  %58 = select i1 %57, i32 %add45, i32 0
>  %59 = select i1 %56, i32 %58, i32 %sub.i
>  %60 = insertelement <2 x i32> %inputPixelIndex35.1229, i32 %59, i32 0
>  %61 = extractelement <2 x i32> %inputPixelIndex35.1229, i32 1
>  %mul48 = mul i32 %61, %imageWidth
>  %add49 = add i32 %59, %mul48
>  %arrayidx50 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %input, i32 %add49
>  %62 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx50, align 16, !tbaa !14
>  %63 = extractelement <4 x float> %62, i32 3
>  %mul51 = fmul float %63, 0x3EEF751040000000
>  %arrayidx53 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %filterIndex.3231
>  %64 = load float, float addrspace(4)* %arrayidx53, align 4, !tbaa !12
>  %mul54 = fmul float %64, %mul51
>  %65 = extractelement <4 x float> %62, i32 0
>  %66 = extractelement <4 x float> %sum.3228, i32 0
>  %67 = tail call float @llvm.fmuladd.f32(float %mul54, float %65, float %66)
>  %68 = insertelement <4 x float> undef, float %67, i32 0
>  %69 = extractelement <4 x float> %62, i32 1
>  %70 = extractelement <4 x float> %sum.3228, i32 1
>  %71 = tail call float @llvm.fmuladd.f32(float %mul54, float %69, float %70)
>  %72 = insertelement <4 x float> %68, float %71, i32 1
>  %73 = extractelement <4 x float> %62, i32 2
>  %74 = extractelement <4 x float> %sum.3228, i32 2
>  %75 = tail call float @llvm.fmuladd.f32(float %mul54, float %73, float %74)
>  %76 = insertelement <4 x float> %72, float %75, i32 2
>  %77 = extractelement <4 x float> %sum.3228, i32 3
>  %78 = tail call float @llvm.fmuladd.f32(float %64, float %63, float %77)
>  %79 = insertelement <4 x float> %76, float %78, i32 3
>  %add59 = fadd float %gamma.3227, %mul54
>  %inc60 = add nsw i32 %filterIndex.3231, 1
>  %inc62 = add nuw nsw i32 %i39.0230, 1
>  %exitcond248 = icmp eq i32 %inc62, %filterWidth
>  br i1 %exitcond248, label %for.cond.cleanup42.loopexit, label %for.body43
>
>if.end69:                                         ; preds = %for.cond.cleanup14, %for.cond.preheader, %for.cond.cleanup33
>  %sum.4 = phi <4 x float> [ %49, %for.cond.cleanup33 ], [ zeroinitializer, %for.cond.preheader ], [ %sum.1.lcssa, %for.cond.cleanup14 ]
>  %80 = extractelement <4 x float> %sum.4, i32 0
>  %81 = fcmp ogt float %80, 6.553500e+04
>  %82 = fcmp olt float %80, 0.000000e+00
>  %83 = select i1 %82, float 0.000000e+00, float %80
>  %.op.i205 = fadd float %83, 5.000000e-01
>  %add.i206 = select i1 %81, float 6.553550e+04, float %.op.i205
>  %84 = insertelement <4 x float> undef, float %add.i206, i32 0
>  %85 = extractelement <4 x float> %sum.4, i32 1
>  %86 = fcmp ogt float %85, 6.553500e+04
>  %87 = fcmp olt float %85, 0.000000e+00
>  %88 = select i1 %87, float 0.000000e+00, float %85
>  %.op.i203 = fadd float %88, 5.000000e-01
>  %add.i204 = select i1 %86, float 6.553550e+04, float %.op.i203
>  %89 = insertelement <4 x float> %84, float %add.i204, i32 1
>  %90 = extractelement <4 x float> %sum.4, i32 2
>  %91 = fcmp ogt float %90, 6.553500e+04
>  %92 = fcmp olt float %90, 0.000000e+00
>  %93 = select i1 %92, float 0.000000e+00, float %90
>  %.op.i201 = fadd float %93, 5.000000e-01
>  %add.i202 = select i1 %91, float 6.553550e+04, float %.op.i201
>  %94 = insertelement <4 x float> %89, float %add.i202, i32 2
>  br i1 %cmp5, label %cond.false, label %cond.true
>
>cond.true:                                        ; preds = %if.end69
>  %95 = extractelement <4 x float> %sum.4, i32 3
>  %96 = fcmp ogt float %95, 6.553500e+04
>  %97 = fcmp olt float %95, 0.000000e+00
>  %98 = select i1 %97, float 0.000000e+00, float %95
>  %.op.i = fadd float %98, 5.000000e-01
>  %add.i = select i1 %96, float 6.553550e+04, float %.op.i
>  %.pre = mul i32 %16, %imageWidth
>  %.pre250 = add i32 %.pre, %8
>  br label %cond.end
>
>cond.false:                                       ; preds = %if.end69
>  %mul76 = mul i32 %16, %imageWidth
>  %add77 = add i32 %mul76, %8
>  %arrayidx78 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %input, i32 %add77
>  %99 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx78, align 16
>  %100 = extractelement <4 x float> %99, i32 3
>  br label %cond.end
>
>cond.end:                                         ; preds = %cond.false, %cond.true
>  %add80.pre-phi = phi i32 [ %add77, %cond.false ], [ %.pre250, %cond.true ]
>  %cond = phi float [ %100, %cond.false ], [ %add.i, %cond.true ]
>  %101 = insertelement <4 x float> %94, float %cond, i32 3
>  %arrayidx81 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %output, i32 %add80.pre-phi
>  store <4 x float> %101, <4 x float> addrspace(1)* %arrayidx81, align 16, !tbaa !14
>  br label %cleanup
>
>cleanup:                                          ; preds = %entry, %cond.end
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @HullPass1(<4 x float> addrspace(1)* nocapture readonly %inputImage, <4 x float> addrspace(1)* nocapture %outputImage, i32 %imageWidth, i32 %imageHeight, <2 x i32> %offset, i32 %polarity, i32 %matte) local_unnamed_addr #4 !kernel_arg_addr_space !47 !kernel_arg_access_qual !48 !kernel_arg_type !49 !kernel_arg_base_type !50 !kernel_arg_type_qual !51 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %mul = mul i32 %16, %imageWidth
>  %add = add i32 %mul, %8
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %inputImage, i32 %add
>  %17 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %18 = extractelement <2 x i32> %offset, i32 1
>  %add2 = add nsw i32 %16, %18
>  %19 = extractelement <2 x i32> %offset, i32 0
>  %add3 = add nsw i32 %8, %19
>  %sub.i140 = add nsw i32 %imageWidth, -1
>  %20 = icmp slt i32 %add3, %imageWidth
>  %21 = icmp sgt i32 %add3, 0
>  %22 = select i1 %21, i32 %add3, i32 0
>  %23 = select i1 %20, i32 %22, i32 %sub.i140
>  %sub.i = add nsw i32 %imageHeight, -1
>  %24 = icmp slt i32 %add2, %imageHeight
>  %25 = icmp sgt i32 %add2, 0
>  %26 = select i1 %25, i32 %add2, i32 0
>  %27 = select i1 %24, i32 %26, i32 %sub.i
>  %cmp = icmp eq i32 %23, %add3
>  %cmp6 = icmp eq i32 %27, %add2
>  %or.cond = and i1 %cmp, %cmp6
>  br i1 %or.cond, label %cond.true, label %cond.end
>
>cond.true:                                        ; preds = %entry
>  %mul7 = mul i32 %add2, %imageWidth
>  %add8 = add i32 %mul7, %add3
>  %arrayidx9 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %inputImage, i32 %add8
>  %28 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx9, align 16, !tbaa !14
>  br label %cond.end
>
>cond.end:                                         ; preds = %entry, %cond.true
>  %cond = phi <4 x float> [ %28, %cond.true ], [ zeroinitializer, %entry ]
>  %29 = extractelement <4 x float> %17, i32 0
>  %conv = fptosi float %29 to i32
>  %30 = extractelement <4 x float> %17, i32 1
>  %conv11 = fptosi float %30 to i32
>  %31 = extractelement <4 x float> %17, i32 2
>  %conv13 = fptosi float %31 to i32
>  %32 = extractelement <4 x float> %17, i32 3
>  %conv15 = fptosi float %32 to i32
>  %33 = extractelement <4 x float> %cond, i32 0
>  %conv17 = fptosi float %33 to i32
>  %34 = extractelement <4 x float> %cond, i32 1
>  %conv19 = fptosi float %34 to i32
>  %35 = extractelement <4 x float> %cond, i32 2
>  %conv21 = fptosi float %35 to i32
>  %36 = extractelement <4 x float> %cond, i32 3
>  %conv23 = fptosi float %36 to i32
>  %cmp25 = icmp sgt i32 %polarity, 0
>  %conv30 = sitofp i32 %conv17 to float
>  %conv32 = sitofp i32 %conv to float
>  br i1 %cmp25, label %for.body.preheader, label %for.body54.preheader
>
>for.body54.preheader:                             ; preds = %cond.end
>  %sub = fadd float %conv32, -5.140000e+02
>  %cmp60 = fcmp ult float %sub, %conv30
>  %sub66 = fadd float %conv32, -2.570000e+02
>  %cond71 = select i1 %cmp60, float %conv32, float %sub66
>  %conv56.1 = sitofp i32 %conv19 to float
>  %conv58.1 = sitofp i32 %conv11 to float
>  %sub.1 = fadd float %conv58.1, -5.140000e+02
>  %cmp60.1 = fcmp ult float %sub.1, %conv56.1
>  %sub66.1 = fadd float %conv58.1, -2.570000e+02
>  %cond71.1 = select i1 %cmp60.1, float %conv58.1, float %sub66.1
>  %conv56.2 = sitofp i32 %conv21 to float
>  %conv58.2 = sitofp i32 %conv13 to float
>  %sub.2 = fadd float %conv58.2, -5.140000e+02
>  %cmp60.2 = fcmp ult float %sub.2, %conv56.2
>  %sub66.2 = fadd float %conv58.2, -2.570000e+02
>  %cond71.2 = select i1 %cmp60.2, float %conv58.2, float %sub66.2
>  %conv56.3 = sitofp i32 %conv23 to float
>  %conv58.3 = sitofp i32 %conv15 to float
>  %sub.3 = fadd float %conv58.3, -5.140000e+02
>  %cmp60.3 = fcmp ult float %sub.3, %conv56.3
>  %sub66.3 = fadd float %conv58.3, -2.570000e+02
>  %cond71.3 = select i1 %cmp60.3, float %conv58.3, float %sub66.3
>  br label %if.end
>
>for.body.preheader:                               ; preds = %cond.end
>  %add34 = fadd float %conv32, 5.140000e+02
>  %cmp35 = fcmp ugt float %add34, %conv30
>  %add41 = fadd float %conv32, 2.570000e+02
>  %cond46 = select i1 %cmp35, float %conv32, float %add41
>  %conv30.1 = sitofp i32 %conv19 to float
>  %conv32.1 = sitofp i32 %conv11 to float
>  %add34.1 = fadd float %conv32.1, 5.140000e+02
>  %cmp35.1 = fcmp ugt float %add34.1, %conv30.1
>  %add41.1 = fadd float %conv32.1, 2.570000e+02
>  %cond46.1 = select i1 %cmp35.1, float %conv32.1, float %add41.1
>  %conv30.2 = sitofp i32 %conv21 to float
>  %conv32.2 = sitofp i32 %conv13 to float
>  %add34.2 = fadd float %conv32.2, 5.140000e+02
>  %cmp35.2 = fcmp ugt float %add34.2, %conv30.2
>  %add41.2 = fadd float %conv32.2, 2.570000e+02
>  %cond46.2 = select i1 %cmp35.2, float %conv32.2, float %add41.2
>  %conv30.3 = sitofp i32 %conv23 to float
>  %conv32.3 = sitofp i32 %conv15 to float
>  %add34.3 = fadd float %conv32.3, 5.140000e+02
>  %cmp35.3 = fcmp ugt float %add34.3, %conv30.3
>  %add41.3 = fadd float %conv32.3, 2.570000e+02
>  %cond46.3 = select i1 %cmp35.3, float %conv32.3, float %add41.3
>  br label %if.end
>
>if.end:                                           ; preds = %for.body54.preheader, %for.body.preheader
>  %.in = phi float [ %cond71.3, %for.body54.preheader ], [ %cond46.3, %for.body.preheader ]
>  %.in145 = phi float [ %cond71.2, %for.body54.preheader ], [ %cond46.2, %for.body.preheader ]
>  %.in146 = phi float [ %cond71.1, %for.body54.preheader ], [ %cond46.1, %for.body.preheader ]
>  %.in147 = phi float [ %cond71, %for.body54.preheader ], [ %cond46, %for.body.preheader ]
>  %37 = fptosi float %.in147 to i32
>  %38 = fptosi float %.in146 to i32
>  %39 = fptosi float %.in145 to i32
>  %conv78 = sitofp i32 %37 to float
>  %40 = insertelement <4 x float> %17, float %conv78, i32 0
>  %conv80 = sitofp i32 %38 to float
>  %41 = insertelement <4 x float> %40, float %conv80, i32 1
>  %conv82 = sitofp i32 %39 to float
>  %42 = insertelement <4 x float> %41, float %conv82, i32 2
>  %cmp83 = icmp eq i32 %matte, 0
>  br i1 %cmp83, label %if.end88, label %if.then85
>
>if.then85:                                        ; preds = %if.end
>  %43 = fptosi float %.in to i32
>  %conv87 = sitofp i32 %43 to float
>  %44 = insertelement <4 x float> %42, float %conv87, i32 3
>  br label %if.end88
>
>if.end88:                                         ; preds = %if.end, %if.then85
>  %v.0 = phi <4 x float> [ %44, %if.then85 ], [ %42, %if.end ]
>  %arrayidx91 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %outputImage, i32 %add
>  store <4 x float> %v.0, <4 x float> addrspace(1)* %arrayidx91, align 16, !tbaa !14
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @HullPass2(<4 x float> addrspace(1)* nocapture readonly %inputImage, <4 x float> addrspace(1)* nocapture %outputImage, i32 %imageWidth, i32 %imageHeight, <2 x i32> %offset, i32 %polarity, i32 %matte) local_unnamed_addr #4 !kernel_arg_addr_space !47 !kernel_arg_access_qual !48 !kernel_arg_type !49 !kernel_arg_base_type !50 !kernel_arg_type_qual !51 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %mul = mul i32 %16, %imageWidth
>  %add = add i32 %mul, %8
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %inputImage, i32 %add
>  %17 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %18 = extractelement <2 x i32> %offset, i32 1
>  %add2 = add nsw i32 %16, %18
>  %19 = extractelement <2 x i32> %offset, i32 0
>  %add3 = add nsw i32 %8, %19
>  %sub.i207 = add nsw i32 %imageWidth, -1
>  %20 = icmp slt i32 %add3, %imageWidth
>  %21 = icmp sgt i32 %add3, 0
>  %22 = select i1 %21, i32 %add3, i32 0
>  %23 = select i1 %20, i32 %22, i32 %sub.i207
>  %sub.i206 = add nsw i32 %imageHeight, -1
>  %24 = icmp slt i32 %add2, %imageHeight
>  %25 = icmp sgt i32 %add2, 0
>  %26 = select i1 %25, i32 %add2, i32 0
>  %27 = select i1 %24, i32 %26, i32 %sub.i206
>  %cmp = icmp eq i32 %23, %add3
>  %cmp6 = icmp eq i32 %27, %add2
>  %or.cond = and i1 %cmp, %cmp6
>  br i1 %or.cond, label %cond.true, label %cond.end
>
>cond.true:                                        ; preds = %entry
>  %mul7 = mul i32 %add2, %imageWidth
>  %add8 = add i32 %mul7, %add3
>  %arrayidx9 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %inputImage, i32 %add8
>  %28 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx9, align 16, !tbaa !14
>  br label %cond.end
>
>cond.end:                                         ; preds = %entry, %cond.true
>  %cond = phi <4 x float> [ %28, %cond.true ], [ zeroinitializer, %entry ]
>  %sub = sub nsw i32 %16, %18
>  %sub10 = sub nsw i32 %8, %19
>  %29 = icmp slt i32 %sub10, %imageWidth
>  %30 = icmp sgt i32 %sub10, 0
>  %31 = select i1 %30, i32 %sub10, i32 0
>  %32 = select i1 %29, i32 %31, i32 %sub.i207
>  %33 = icmp slt i32 %sub, %imageHeight
>  %34 = icmp sgt i32 %sub, 0
>  %35 = select i1 %34, i32 %sub, i32 0
>  %36 = select i1 %33, i32 %35, i32 %sub.i206
>  %cmp13 = icmp eq i32 %32, %sub10
>  %cmp15 = icmp eq i32 %36, %sub
>  %or.cond204 = and i1 %cmp13, %cmp15
>  br i1 %or.cond204, label %cond.true16, label %cond.end21
>
>cond.true16:                                      ; preds = %cond.end
>  %mul17 = mul i32 %sub, %imageWidth
>  %add18 = add i32 %mul17, %sub10
>  %arrayidx19 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %inputImage, i32 %add18
>  %37 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx19, align 16, !tbaa !14
>  br label %cond.end21
>
>cond.end21:                                       ; preds = %cond.end, %cond.true16
>  %cond22 = phi <4 x float> [ %37, %cond.true16 ], [ zeroinitializer, %cond.end ]
>  %38 = extractelement <4 x float> %17, i32 0
>  %conv = fptosi float %38 to i32
>  %39 = extractelement <4 x float> %17, i32 1
>  %conv24 = fptosi float %39 to i32
>  %40 = extractelement <4 x float> %17, i32 2
>  %conv26 = fptosi float %40 to i32
>  %41 = extractelement <4 x float> %17, i32 3
>  %conv28 = fptosi float %41 to i32
>  %42 = extractelement <4 x float> %cond, i32 0
>  %conv30 = fptosi float %42 to i32
>  %43 = extractelement <4 x float> %cond, i32 1
>  %conv32 = fptosi float %43 to i32
>  %44 = extractelement <4 x float> %cond, i32 2
>  %conv34 = fptosi float %44 to i32
>  %45 = extractelement <4 x float> %cond, i32 3
>  %conv36 = fptosi float %45 to i32
>  %46 = extractelement <4 x float> %cond22, i32 0
>  %conv38 = fptosi float %46 to i32
>  %47 = extractelement <4 x float> %cond22, i32 1
>  %conv40 = fptosi float %47 to i32
>  %48 = extractelement <4 x float> %cond22, i32 2
>  %conv42 = fptosi float %48 to i32
>  %49 = extractelement <4 x float> %cond22, i32 3
>  %conv44 = fptosi float %49 to i32
>  %cmp46 = icmp sgt i32 %polarity, 0
>  %conv51 = sitofp i32 %conv38 to float
>  %conv53 = sitofp i32 %conv to float
>  br i1 %cmp46, label %for.body.preheader, label %for.body82.preheader
>
>for.body82.preheader:                             ; preds = %cond.end21
>  %sub88 = fadd float %conv53, -5.140000e+02
>  %cmp89 = fcmp olt float %sub88, %conv51
>  %conv90 = zext i1 %cmp89 to i32
>  %cmp93 = icmp sge i32 %conv30, %conv
>  %add95 = sext i1 %cmp93 to i32
>  %cmp96 = icmp eq i32 %conv90, %add95
>  %sub105 = fadd float %conv53, -2.570000e+02
>  %spec.select208 = select i1 %cmp96, float %sub105, float %conv53
>  %conv84.1 = sitofp i32 %conv40 to float
>  %conv86.1 = sitofp i32 %conv24 to float
>  %sub88.1 = fadd float %conv86.1, -5.140000e+02
>  %cmp89.1 = fcmp olt float %sub88.1, %conv84.1
>  %conv90.1 = zext i1 %cmp89.1 to i32
>  %cmp93.1 = icmp sge i32 %conv32, %conv24
>  %add95.1 = sext i1 %cmp93.1 to i32
>  %cmp96.1 = icmp eq i32 %conv90.1, %add95.1
>  %sub105.1 = fadd float %conv86.1, -2.570000e+02
>  %spec.select208.1 = select i1 %cmp96.1, float %sub105.1, float %conv86.1
>  %conv84.2 = sitofp i32 %conv42 to float
>  %conv86.2 = sitofp i32 %conv26 to float
>  %sub88.2 = fadd float %conv86.2, -5.140000e+02
>  %cmp89.2 = fcmp olt float %sub88.2, %conv84.2
>  %conv90.2 = zext i1 %cmp89.2 to i32
>  %cmp93.2 = icmp sge i32 %conv34, %conv26
>  %add95.2 = sext i1 %cmp93.2 to i32
>  %cmp96.2 = icmp eq i32 %conv90.2, %add95.2
>  %sub105.2 = fadd float %conv86.2, -2.570000e+02
>  %spec.select208.2 = select i1 %cmp96.2, float %sub105.2, float %conv86.2
>  %conv84.3 = sitofp i32 %conv44 to float
>  %conv86.3 = sitofp i32 %conv28 to float
>  %sub88.3 = fadd float %conv86.3, -5.140000e+02
>  %cmp89.3 = fcmp olt float %sub88.3, %conv84.3
>  %conv90.3 = zext i1 %cmp89.3 to i32
>  %cmp93.3 = icmp sge i32 %conv36, %conv28
>  %add95.3 = sext i1 %cmp93.3 to i32
>  %cmp96.3 = icmp eq i32 %conv90.3, %add95.3
>  %sub105.3 = fadd float %conv86.3, -2.570000e+02
>  br label %if.end
>
>for.body.preheader:                               ; preds = %cond.end21
>  %add55 = fadd float %conv53, 5.140000e+02
>  %cmp56 = fcmp ogt float %add55, %conv51
>  %conv57 = zext i1 %cmp56 to i32
>  %cmp60 = icmp sle i32 %conv30, %conv
>  %add62 = sext i1 %cmp60 to i32
>  %cmp63 = icmp eq i32 %conv57, %add62
>  %add72 = fadd float %conv53, 2.570000e+02
>  %spec.select = select i1 %cmp63, float %add72, float %conv53
>  %conv51.1 = sitofp i32 %conv40 to float
>  %conv53.1 = sitofp i32 %conv24 to float
>  %add55.1 = fadd float %conv53.1, 5.140000e+02
>  %cmp56.1 = fcmp ogt float %add55.1, %conv51.1
>  %conv57.1 = zext i1 %cmp56.1 to i32
>  %cmp60.1 = icmp sle i32 %conv32, %conv24
>  %add62.1 = sext i1 %cmp60.1 to i32
>  %cmp63.1 = icmp eq i32 %conv57.1, %add62.1
>  %add72.1 = fadd float %conv53.1, 2.570000e+02
>  %spec.select.1 = select i1 %cmp63.1, float %add72.1, float %conv53.1
>  %conv51.2 = sitofp i32 %conv42 to float
>  %conv53.2 = sitofp i32 %conv26 to float
>  %add55.2 = fadd float %conv53.2, 5.140000e+02
>  %cmp56.2 = fcmp ogt float %add55.2, %conv51.2
>  %conv57.2 = zext i1 %cmp56.2 to i32
>  %cmp60.2 = icmp sle i32 %conv34, %conv26
>  %add62.2 = sext i1 %cmp60.2 to i32
>  %cmp63.2 = icmp eq i32 %conv57.2, %add62.2
>  %add72.2 = fadd float %conv53.2, 2.570000e+02
>  %spec.select.2 = select i1 %cmp63.2, float %add72.2, float %conv53.2
>  %conv51.3 = sitofp i32 %conv44 to float
>  %conv53.3 = sitofp i32 %conv28 to float
>  %add55.3 = fadd float %conv53.3, 5.140000e+02
>  %cmp56.3 = fcmp ogt float %add55.3, %conv51.3
>  %conv57.3 = zext i1 %cmp56.3 to i32
>  %cmp60.3 = icmp sle i32 %conv36, %conv28
>  %add62.3 = sext i1 %cmp60.3 to i32
>  %cmp63.3 = icmp eq i32 %conv57.3, %add62.3
>  %add72.3 = fadd float %conv53.3, 2.570000e+02
>  br label %if.end
>
>if.end:                                           ; preds = %for.body82.preheader, %for.body.preheader
>  %conv86.3.sink = phi float [ %conv86.3, %for.body82.preheader ], [ %conv53.3, %for.body.preheader ]
>  %sub105.3.sink = phi float [ %sub105.3, %for.body82.preheader ], [ %add72.3, %for.body.preheader ]
>  %cmp96.3.sink = phi i1 [ %cmp96.3, %for.body82.preheader ], [ %cmp63.3, %for.body.preheader ]
>  %.in213 = phi float [ %spec.select208.2, %for.body82.preheader ], [ %spec.select.2, %for.body.preheader ]
>  %.in214 = phi float [ %spec.select208.1, %for.body82.preheader ], [ %spec.select.1, %for.body.preheader ]
>  %.in215 = phi float [ %spec.select208, %for.body82.preheader ], [ %spec.select, %for.body.preheader ]
>  %50 = fptosi float %.in215 to i32
>  %51 = fptosi float %.in214 to i32
>  %52 = fptosi float %.in213 to i32
>  %conv114 = sitofp i32 %50 to float
>  %53 = insertelement <4 x float> %17, float %conv114, i32 0
>  %conv116 = sitofp i32 %51 to float
>  %54 = insertelement <4 x float> %53, float %conv116, i32 1
>  %conv118 = sitofp i32 %52 to float
>  %55 = insertelement <4 x float> %54, float %conv118, i32 2
>  %cmp119 = icmp eq i32 %matte, 0
>  br i1 %cmp119, label %if.end124, label %if.then121
>
>if.then121:                                       ; preds = %if.end
>  %spec.select208.3 = select i1 %cmp96.3.sink, float %sub105.3.sink, float %conv86.3.sink
>  %56 = fptosi float %spec.select208.3 to i32
>  %conv123 = sitofp i32 %56 to float
>  %57 = insertelement <4 x float> %55, float %conv123, i32 3
>  br label %if.end124
>
>if.end124:                                        ; preds = %if.end, %if.then121
>  %v.0 = phi <4 x float> [ %57, %if.then121 ], [ %55, %if.end ]
>  %arrayidx127 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %outputImage, i32 %add
>  store <4 x float> %v.0, <4 x float> addrspace(1)* %arrayidx127, align 16, !tbaa !14
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @Equalize(<4 x float> addrspace(1)* noalias nocapture %im, i32 %channel, <4 x float> addrspace(1)* noalias nocapture readonly %equalize_map, <4 x float> %white, <4 x float> %black) local_unnamed_addr #4 !kernel_arg_addr_space !35 !kernel_arg_access_qual !31 !kernel_arg_type !36 !kernel_arg_base_type !37 !kernel_arg_type_qual !38 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %x.i = tail call i32 @llvm.r600.read.global.size.x() #16
>  %mul = mul nsw i32 %16, %x.i
>  %add = add nsw i32 %8, %mul
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %im, i32 %add
>  %17 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %and = and i32 %channel, 131072
>  %cmp = icmp eq i32 %and, 0
>  br i1 %cmp, label %if.end27, label %if.then
>
>if.then:                                          ; preds = %entry
>  %18 = extractelement <4 x float> %white, i32 2
>  %19 = extractelement <4 x float> %black, i32 2
>  %cmp5 = fcmp une float %18, %19
>  br i1 %cmp5, label %if.then6, label %if.end27
>
>if.then6:                                         ; preds = %if.then
>  %20 = extractelement <4 x float> %17, i32 2
>  %.inv.i57 = fcmp oge float %20, 6.553500e+04
>  %retval.02.i58 = select i1 %.inv.i57, float 6.553500e+04, float %20
>  %21 = fptoui float %retval.02.i58 to i32
>  %arrayidx9 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %equalize_map, i32 %21
>  %22 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx9, align 16, !tbaa !14
>  %23 = extractelement <4 x float> %17, i32 1
>  %.inv.i55 = fcmp oge float %23, 6.553500e+04
>  %retval.02.i56 = select i1 %.inv.i55, float 6.553500e+04, float %23
>  %24 = fptoui float %retval.02.i56 to i32
>  %arrayidx13 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %equalize_map, i32 %24
>  %25 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx13, align 16, !tbaa !14
>  %26 = extractelement <4 x float> %17, i32 0
>  %.inv.i53 = fcmp oge float %26, 6.553500e+04
>  %retval.02.i54 = select i1 %.inv.i53, float 6.553500e+04, float %26
>  %27 = fptoui float %retval.02.i54 to i32
>  %arrayidx17 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %equalize_map, i32 %27
>  %28 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx17, align 16, !tbaa !14
>  %29 = extractelement <4 x float> %17, i32 3
>  %.inv.i = fcmp oge float %29, 6.553500e+04
>  %retval.02.i = select i1 %.inv.i, float 6.553500e+04, float %29
>  %30 = fptoui float %retval.02.i to i32
>  %arrayidx21 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %equalize_map, i32 %30
>  %31 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx21, align 16, !tbaa !14
>  %vecinit23 = shufflevector <4 x float> %28, <4 x float> %25, <4 x i32> <i32 2, i32 6, i32 undef, i32 undef>
>  %vecinit24 = shufflevector <4 x float> %vecinit23, <4 x float> %22, <4 x i32> <i32 0, i32 1, i32 6, i32 undef>
>  %vecinit25 = shufflevector <4 x float> %vecinit24, <4 x float> %31, <4 x i32> <i32 0, i32 1, i32 2, i32 6>
>  store <4 x float> %vecinit25, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  br label %if.end27
>
>if.end27:                                         ; preds = %entry, %if.then, %if.then6
>  ret void
>}
>
>; Function Attrs: convergent nounwind readonly
>define float @ApplyFunction(float %pixel, i32 %function, i32 %number_parameters, float addrspace(4)* nocapture readonly %parameters) local_unnamed_addr #2 {
>entry:
>  switch i32 %function, label %sw.epilog [
>    i32 3, label %for.cond.preheader
>    i32 4, label %sw.bb3
>    i32 1, label %sw.bb29
>    i32 2, label %sw.bb74
>  ]
>
>for.cond.preheader:                               ; preds = %entry
>  %cmp195 = icmp eq i32 %number_parameters, 0
>  br i1 %cmp195, label %sw.epilog, label %for.body
>
>for.cond.cleanup.loopexit:                        ; preds = %for.body
>  %phitmp198 = fmul float %1, 6.553500e+04
>  br label %sw.epilog
>
>for.body:                                         ; preds = %for.cond.preheader, %for.body
>  %result.0197 = phi float [ %1, %for.body ], [ 0.000000e+00, %for.cond.preheader ]
>  %i.0196 = phi i32 [ %inc, %for.body ], [ 0, %for.cond.preheader ]
>  %mul = fmul float %result.0197, 0x3EEF751040000000
>  %arrayidx = getelementptr inbounds float, float addrspace(4)* %parameters, i32 %i.0196
>  %0 = load float, float addrspace(4)* %arrayidx, align 4, !tbaa !12
>  %1 = tail call float @llvm.fmuladd.f32(float %mul, float %pixel, float %0)
>  %inc = add nuw i32 %i.0196, 1
>  %exitcond = icmp eq i32 %inc, %number_parameters
>  br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body
>
>sw.bb3:                                           ; preds = %entry
>  %cmp4 = icmp eq i32 %number_parameters, 0
>  br i1 %cmp4, label %cond.end22, label %cond.end
>
>cond.end:                                         ; preds = %sw.bb3
>  %2 = load float, float addrspace(4)* %parameters, align 4, !tbaa !12
>  %phitmp167 = fmul float %2, 0x3EEF751040000000
>  %cmp6 = icmp eq i32 %number_parameters, 1
>  br i1 %cmp6, label %cond.end22, label %cond.end10
>
>cond.end10:                                       ; preds = %cond.end
>  %arrayidx8 = getelementptr inbounds float, float addrspace(4)* %parameters, i32 1
>  %3 = load float, float addrspace(4)* %arrayidx8, align 4, !tbaa !12
>  %cmp12 = icmp ugt i32 %number_parameters, 2
>  br i1 %cmp12, label %cond.end16, label %cond.end22
>
>cond.end16:                                       ; preds = %cond.end10
>  %arrayidx14 = getelementptr inbounds float, float addrspace(4)* %parameters, i32 2
>  %4 = load float, float addrspace(4)* %arrayidx14, align 4, !tbaa !12
>  %cmp18 = icmp eq i32 %number_parameters, 3
>  br i1 %cmp18, label %cond.end22, label %cond.true19
>
>cond.true19:                                      ; preds = %cond.end16
>  %arrayidx20 = getelementptr inbounds float, float addrspace(4)* %parameters, i32 3
>  %5 = load float, float addrspace(4)* %arrayidx20, align 4, !tbaa !12
>  br label %cond.end22
>
>cond.end22:                                       ; preds = %cond.end16, %cond.end, %sw.bb3, %cond.end10, %cond.true19
>  %cond17176 = phi float [ %4, %cond.true19 ], [ %4, %cond.end16 ], [ 5.000000e-01, %cond.end10 ], [ 5.000000e-01, %cond.end ], [ 5.000000e-01, %sw.bb3 ]
>  %cond169171175 = phi float [ %phitmp167, %cond.true19 ], [ %phitmp167, %cond.end16 ], [ %phitmp167, %cond.end10 ], [ %phitmp167, %cond.end ], [ 0x3EEF751040000000, %sw.bb3 ]
>  %cond11172174 = phi float [ %3, %cond.true19 ], [ %3, %cond.end16 ], [ %3, %cond.end10 ], [ 0.000000e+00, %cond.end ], [ 0.000000e+00, %sw.bb3 ]
>  %cond23 = phi float [ %5, %cond.true19 ], [ 5.000000e-01, %cond.end16 ], [ 5.000000e-01, %cond.end10 ], [ 5.000000e-01, %cond.end ], [ 5.000000e-01, %sw.bb3 ]
>  %div = fdiv float %cond11172174, 3.600000e+02, !fpmath !11
>  %6 = tail call float @llvm.fmuladd.f32(float %cond169171175, float %pixel, float %div)
>  %mul26 = fmul float %6, 0x401921FB80000000
>  %7 = bitcast float %mul26 to i32
>  %8 = and i32 %7, 2147483647
>  %9 = bitcast i32 %8 to float
>  %10 = fcmp olt float %9, 0x4160000000000000
>  br i1 %10, label %11, label %52
>
>; <label>:11:                                     ; preds = %cond.end22
>  %12 = tail call float @llvm.fmuladd.f32(float %9, float 0x3FE45F3060000000, float 5.000000e-01) #13
>  %13 = tail call float @llvm.trunc.f32(float %12) #14
>  %14 = bitcast float %13 to i32
>  %15 = and i32 %14, -4096
>  %16 = bitcast i32 %15 to float
>  %17 = fsub float %13, %16
>  %18 = fmul float %13, 0x3FF921FB40000000
>  %19 = fsub float -0.000000e+00, %18
>  %20 = tail call float @llvm.fmuladd.f32(float %16, float 0x3FF9200000000000, float %19) #13
>  %21 = tail call float @llvm.fmuladd.f32(float %16, float 0x3F3FB40000000000, float %20) #13
>  %22 = tail call float @llvm.fmuladd.f32(float %17, float 0x3FF9200000000000, float %21) #13
>  %23 = tail call float @llvm.fmuladd.f32(float %17, float 0x3F3FB40000000000, float %22) #13
>  %24 = fsub float %9, %18
>  %25 = fsub float %9, %24
>  %26 = fsub float %25, %18
>  %27 = fsub float %26, %23
>  %28 = fadd float %24, %27
>  %29 = fmul float %13, 0x3E74442D00000000
>  %30 = fsub float -0.000000e+00, %29
>  %31 = tail call float @llvm.fmuladd.f32(float %16, float 0x3E74440000000000, float %30) #13
>  %32 = tail call float @llvm.fmuladd.f32(float %16, float 0x3D86800000000000, float %31) #13
>  %33 = tail call float @llvm.fmuladd.f32(float %17, float 0x3E74440000000000, float %32) #13
>  %34 = tail call float @llvm.fmuladd.f32(float %17, float 0x3D86800000000000, float %33) #13
>  %35 = fsub float %28, %29
>  %36 = fsub float %28, %35
>  %37 = fsub float %36, %29
>  %38 = fsub float %37, %34
>  %39 = fadd float %35, %38
>  %40 = fmul float %13, 0x3CF8469880000000
>  %41 = fsub float -0.000000e+00, %40
>  %42 = tail call float @llvm.fmuladd.f32(float %16, float 0x3CF8460000000000, float %41) #13
>  %43 = tail call float @llvm.fmuladd.f32(float %16, float 0x3C23100000000000, float %42) #13
>  %44 = tail call float @llvm.fmuladd.f32(float %17, float 0x3CF8460000000000, float %43) #13
>  %45 = tail call float @llvm.fmuladd.f32(float %17, float 0x3C23100000000000, float %44) #13
>  %46 = fsub float %39, %40
>  %47 = fsub float %39, %46
>  %48 = fsub float %47, %40
>  %49 = fadd float %46, %48
>  %50 = fsub float -0.000000e+00, %45
>  %51 = fptosi float %13 to i32
>  br label %_Z3sinf.exit
>
>; <label>:52:                                     ; preds = %cond.end22
>  %53 = lshr i32 %8, 23
>  %54 = and i32 %7, 8388607
>  %55 = or i32 %54, 8388608
>  %56 = mul i32 %55, -28220501
>  %57 = zext i32 %55 to i64
>  %58 = mul nuw nsw i64 %57, 4266746795
>  %59 = lshr i64 %58, 32
>  %60 = trunc i64 %59 to i32
>  %61 = mul i32 %55, 1011060801
>  %62 = add i32 %61, %60
>  %63 = mul nuw nsw i64 %57, 1011060801
>  %64 = lshr i64 %63, 32
>  %65 = trunc i64 %64 to i32
>  %66 = icmp ult i32 %62, %60
>  %67 = zext i1 %66 to i32
>  %68 = add nuw nsw i32 %67, %65
>  %69 = mul i32 %55, -614296167
>  %70 = add i32 %68, %69
>  %71 = mul nuw nsw i64 %57, 3680671129
>  %72 = lshr i64 %71, 32
>  %73 = trunc i64 %72 to i32
>  %74 = icmp ult i32 %70, %68
>  %75 = zext i1 %74 to i32
>  %76 = add nuw nsw i32 %75, %73
>  %77 = mul i32 %55, -181084736
>  %78 = add i32 %76, %77
>  %79 = mul nuw nsw i64 %57, 4113882560
>  %80 = lshr i64 %79, 32
>  %81 = trunc i64 %80 to i32
>  %82 = icmp ult i32 %78, %76
>  %83 = zext i1 %82 to i32
>  %84 = add nuw nsw i32 %83, %81
>  %85 = mul i32 %55, -64530479
>  %86 = add i32 %84, %85
>  %87 = mul nuw nsw i64 %57, 4230436817
>  %88 = lshr i64 %87, 32
>  %89 = trunc i64 %88 to i32
>  %90 = icmp ult i32 %86, %84
>  %91 = zext i1 %90 to i32
>  %92 = add nuw nsw i32 %91, %89
>  %93 = mul i32 %55, 1313084713
>  %94 = add i32 %92, %93
>  %95 = mul nuw nsw i64 %57, 1313084713
>  %96 = lshr i64 %95, 32
>  %97 = trunc i64 %96 to i32
>  %98 = icmp ult i32 %94, %92
>  %99 = zext i1 %98 to i32
>  %100 = add nuw nsw i32 %99, %97
>  %101 = mul i32 %55, -1560706194
>  %102 = add i32 %100, %101
>  %103 = mul nuw nsw i64 %57, 2734261102
>  %104 = lshr i64 %103, 32
>  %105 = trunc i64 %104 to i32
>  %106 = icmp ult i32 %102, %100
>  %107 = zext i1 %106 to i32
>  %108 = add nuw nsw i32 %107, %105
>  %109 = add nsw i32 %53, -120
>  %110 = icmp ugt i32 %109, 31
>  %111 = select i1 %110, i32 %102, i32 %108
>  %112 = select i1 %110, i32 %94, i32 %102
>  %113 = select i1 %110, i32 %86, i32 %94
>  %114 = select i1 %110, i32 %78, i32 %86
>  %115 = select i1 %110, i32 %70, i32 %78
>  %116 = select i1 %110, i32 %62, i32 %70
>  %117 = select i1 %110, i32 %56, i32 %62
>  %118 = select i1 %110, i32 32, i32 0
>  %119 = sub nsw i32 %109, %118
>  %120 = icmp ugt i32 %119, 31
>  %121 = select i1 %120, i32 %112, i32 %111
>  %122 = select i1 %120, i32 %113, i32 %112
>  %123 = select i1 %120, i32 %114, i32 %113
>  %124 = select i1 %120, i32 %115, i32 %114
>  %125 = select i1 %120, i32 %116, i32 %115
>  %126 = select i1 %120, i32 %117, i32 %116
>  %127 = select i1 %120, i32 32, i32 0
>  %128 = sub nsw i32 %119, %127
>  %129 = icmp ugt i32 %128, 31
>  %130 = select i1 %129, i32 %122, i32 %121
>  %131 = select i1 %129, i32 %123, i32 %122
>  %132 = select i1 %129, i32 %124, i32 %123
>  %133 = select i1 %129, i32 %125, i32 %124
>  %134 = select i1 %129, i32 %126, i32 %125
>  %135 = select i1 %129, i32 32, i32 0
>  %136 = sub nsw i32 %128, %135
>  %137 = icmp ugt i32 %136, 31
>  %138 = select i1 %137, i32 %131, i32 %130
>  %139 = select i1 %137, i32 %132, i32 %131
>  %140 = select i1 %137, i32 %133, i32 %132
>  %141 = select i1 %137, i32 %134, i32 %133
>  %142 = select i1 %137, i32 32, i32 0
>  %143 = sub nsw i32 %136, %142
>  %144 = icmp ne i32 %143, 0
>  %145 = sub nsw i32 0, %143
>  %146 = and i32 %143, 31
>  %147 = shl i32 %138, %146
>  %148 = and i32 %145, 31
>  %149 = lshr i32 %139, %148
>  %150 = or i32 %147, %149
>  %151 = shl i32 %139, %146
>  %152 = lshr i32 %140, %148
>  %153 = or i32 %151, %152
>  %154 = shl i32 %140, %146
>  %155 = lshr i32 %141, %148
>  %156 = or i32 %154, %155
>  %157 = select i1 %144, i32 %150, i32 %138
>  %158 = select i1 %144, i32 %153, i32 %139
>  %159 = select i1 %144, i32 %156, i32 %140
>  %160 = lshr i32 %157, 29
>  %161 = shl i32 %157, 2
>  %162 = lshr i32 %158, 30
>  %163 = or i32 %161, %162
>  %164 = shl i32 %158, 2
>  %165 = lshr i32 %159, 30
>  %166 = or i32 %164, %165
>  %167 = shl i32 %159, 2
>  %168 = lshr i32 %141, 30
>  %169 = or i32 %167, %168
>  %170 = and i32 %160, 1
>  %171 = shl i32 %160, 31
>  %172 = ashr exact i32 %171, 31
>  %173 = xor i32 %172, %163
>  %174 = xor i32 %172, %166
>  %175 = xor i32 %172, %169
>  %176 = tail call i32 @llvm.ctlz.i32(i32 %173, i1 false) #13, !range !52
>  %177 = add nuw nsw i32 %176, 1
>  %178 = sub nsw i32 31, %176
>  %179 = and i32 %177, 31
>  %180 = shl i32 %173, %179
>  %181 = and i32 %178, 31
>  %182 = lshr i32 %174, %181
>  %183 = or i32 %180, %182
>  %184 = shl i32 %174, %179
>  %185 = lshr i32 %175, %181
>  %186 = or i32 %184, %185
>  %187 = shl nuw nsw i32 %176, 23
>  %188 = sub nsw i32 1056964608, %187
>  %189 = or i32 %188, %171
>  %190 = lshr i32 %183, 9
>  %191 = or i32 %189, %190
>  %192 = bitcast i32 %191 to float
>  %193 = shl i32 %183, 23
>  %194 = lshr i32 %186, 9
>  %195 = or i32 %193, %194
>  %196 = tail call i32 @llvm.ctlz.i32(i32 %195, i1 false) #13, !range !52
>  %197 = add nuw nsw i32 %196, 1
>  %198 = sub nsw i32 31, %196
>  %199 = and i32 %197, 31
>  %200 = shl i32 %195, %199
>  %201 = and i32 %198, 31
>  %202 = lshr i32 %186, %201
>  %203 = or i32 %200, %202
>  %204 = add nuw nsw i32 %176, 24
>  %205 = add nuw nsw i32 %204, %197
>  %206 = shl nuw nsw i32 %205, 23
>  %207 = sub nsw i32 1065353216, %206
>  %208 = or i32 %207, %171
>  %209 = lshr i32 %203, 9
>  %210 = or i32 %208, %209
>  %211 = bitcast i32 %210 to float
>  %212 = and i32 %191, -1073745920
>  %213 = bitcast i32 %212 to float
>  %214 = fsub float %192, %213
>  %215 = fmul float %192, 0x3FF921FB40000000
>  %216 = fsub float -0.000000e+00, %215
>  %217 = tail call float @llvm.fmuladd.f32(float %213, float 0x3FF9200000000000, float %216) #13
>  %218 = tail call float @llvm.fmuladd.f32(float %213, float 0x3F3FB40000000000, float %217) #13
>  %219 = tail call float @llvm.fmuladd.f32(float %214, float 0x3FF9200000000000, float %218) #13
>  %220 = tail call float @llvm.fmuladd.f32(float %214, float 0x3F3FB40000000000, float %219) #13
>  %221 = tail call float @llvm.fmuladd.f32(float %192, float 0x3E74442D00000000, float %220) #13
>  %222 = tail call float @llvm.fmuladd.f32(float %211, float 0x3FF921FB40000000, float %221) #13
>  %223 = fadd float %215, %222
>  %224 = fsub float %223, %215
>  %225 = fsub float %222, %224
>  %226 = lshr i32 %157, 30
>  %227 = add nuw nsw i32 %170, %226
>  br label %_Z3sinf.exit
>
>_Z3sinf.exit:                                     ; preds = %11, %52
>  %.05.i = phi float [ %49, %11 ], [ %223, %52 ]
>  %.0.i = phi float [ %50, %11 ], [ %225, %52 ]
>  %228 = phi i32 [ %51, %11 ], [ %227, %52 ]
>  %229 = fmul float %.05.i, %.05.i
>  %230 = fmul float %.05.i, %229
>  %231 = tail call float @llvm.fmuladd.f32(float %229, float 0x3DE5D93A60000000, float 0xBE5AE5E680000000) #13
>  %232 = tail call float @llvm.fmuladd.f32(float %229, float %231, float 0x3EC6DBE4A0000000) #13
>  %233 = tail call float @llvm.fmuladd.f32(float %229, float %232, float 0xBF2A013A80000000) #13
>  %234 = tail call float @llvm.fmuladd.f32(float %229, float %233, float 0x3F811110E0000000) #13
>  %235 = fmul float %230, %234
>  %236 = fsub float -0.000000e+00, %235
>  %237 = tail call float @llvm.fmuladd.f32(float %.0.i, float 5.000000e-01, float %236) #13
>  %238 = fsub float -0.000000e+00, %.0.i
>  %239 = tail call float @llvm.fmuladd.f32(float %229, float %237, float %238) #13
>  %240 = tail call float @llvm.fmuladd.f32(float %230, float 0x3FC5555560000000, float %239) #13
>  %241 = fsub float %.05.i, %240
>  %242 = tail call float @llvm.fmuladd.f32(float %229, float 0xBDA8FAE9C0000000, float 0x3E21EE9EC0000000) #13
>  %243 = tail call float @llvm.fmuladd.f32(float %229, float %242, float 0xBE92524740000000) #13
>  %244 = tail call float @llvm.fmuladd.f32(float %229, float %243, float 0x3EFA015C40000000) #13
>  %245 = tail call float @llvm.fmuladd.f32(float %229, float %244, float 0xBF56C16C00000000) #13
>  %246 = tail call float @llvm.fmuladd.f32(float %229, float %245, float 0x3FA5555560000000) #13
>  %247 = fmul float %229, %246
>  %248 = bitcast float %.05.i to i32
>  %249 = and i32 %248, 2147483647
>  %250 = add nsw i32 %249, -16777216
>  %251 = bitcast i32 %250 to float
>  %252 = add nsw i32 %249, -1050253722
>  %253 = icmp ult i32 %252, 11429479
>  %254 = select i1 %253, float %251, float 0.000000e+00
>  %255 = icmp ugt i32 %249, 1061683200
>  %256 = select i1 %255, float 2.812500e-01, float %254
>  %257 = fsub float -0.000000e+00, %256
>  %258 = tail call float @llvm.fmuladd.f32(float %229, float 5.000000e-01, float %257) #13
>  %259 = fsub float 1.000000e+00, %256
>  %260 = fmul float %.05.i, %.0.i
>  %261 = fsub float -0.000000e+00, %260
>  %262 = tail call float @llvm.fmuladd.f32(float %229, float %247, float %261) #13
>  %263 = fsub float %258, %262
>  %264 = fsub float %259, %263
>  %265 = and i32 %228, 1
>  %266 = icmp eq i32 %265, 0
>  %267 = select i1 %266, float %241, float %264
>  %268 = bitcast float %267 to i32
>  %269 = lshr i32 %228, 1
>  %.lobit.i = shl i32 %269, 31
>  %270 = and i32 %7, -2147483648
>  %271 = xor i32 %.lobit.i, %270
>  %272 = xor i32 %271, %268
>  %273 = bitcast i32 %272 to float
>  %274 = icmp ugt i32 %8, 2139095039
>  %275 = select i1 %274, float 0x7FF8000000000000, float %273
>  %276 = fcmp oeq float %mul26, 0.000000e+00
>  %277 = select i1 %276, float %mul26, float %275
>  %278 = tail call float @llvm.fmuladd.f32(float %cond17176, float %277, float %cond23)
>  %mul28 = fmul float %278, 6.553500e+04
>  br label %sw.epilog
>
>sw.bb29:                                          ; preds = %entry
>  %cmp31 = icmp eq i32 %number_parameters, 0
>  br i1 %cmp31, label %cond.end53, label %cond.end35
>
>cond.end35:                                       ; preds = %sw.bb29
>  %279 = load float, float addrspace(4)* %parameters, align 4, !tbaa !12
>  %cmp37 = icmp eq i32 %number_parameters, 1
>  br i1 %cmp37, label %cond.end53, label %cond.end41
>
>cond.end41:                                       ; preds = %cond.end35
>  %arrayidx39 = getelementptr inbounds float, float addrspace(4)* %parameters, i32 1
>  %280 = load float, float addrspace(4)* %arrayidx39, align 4, !tbaa !12
>  %cmp43 = icmp ugt i32 %number_parameters, 2
>  br i1 %cmp43, label %cond.end47, label %cond.end53
>
>cond.end47:                                       ; preds = %cond.end41
>  %arrayidx45 = getelementptr inbounds float, float addrspace(4)* %parameters, i32 2
>  %281 = load float, float addrspace(4)* %arrayidx45, align 4, !tbaa !12
>  %cmp49 = icmp eq i32 %number_parameters, 3
>  br i1 %cmp49, label %cond.end53, label %cond.true50
>
>cond.true50:                                      ; preds = %cond.end47
>  %arrayidx51 = getelementptr inbounds float, float addrspace(4)* %parameters, i32 3
>  %282 = load float, float addrspace(4)* %arrayidx51, align 4, !tbaa !12
>  br label %cond.end53
>
>cond.end53:                                       ; preds = %cond.end47, %cond.end35, %sw.bb29, %cond.end41, %cond.true50
>  %cond48185 = phi float [ %281, %cond.true50 ], [ %281, %cond.end47 ], [ 1.000000e+00, %cond.end41 ], [ 1.000000e+00, %cond.end35 ], [ 1.000000e+00, %sw.bb29 ]
>  %cond36178180184 = phi float [ %279, %cond.true50 ], [ %279, %cond.end47 ], [ %279, %cond.end41 ], [ %279, %cond.end35 ], [ 1.000000e+00, %sw.bb29 ]
>  %cond42181183 = phi float [ %280, %cond.true50 ], [ %280, %cond.end47 ], [ %280, %cond.end41 ], [ 5.000000e-01, %cond.end35 ], [ 5.000000e-01, %sw.bb29 ]
>  %cond54 = phi float [ %282, %cond.true50 ], [ 5.000000e-01, %cond.end47 ], [ 5.000000e-01, %cond.end41 ], [ 5.000000e-01, %cond.end35 ], [ 5.000000e-01, %sw.bb29 ]
>  %div55 = fdiv float 2.000000e+00, %cond36178180184, !fpmath !11
>  %neg = fsub float -0.000000e+00, %cond42181183
>  %283 = tail call float @llvm.fmuladd.f32(float %pixel, float 0x3EEF751040000000, float %neg)
>  %mul57 = fmul float %div55, %283
>  %div58 = fdiv float %cond48185, 0x400921FB80000000, !fpmath !11
>  %284 = fsub float -0.000000e+00, %mul57
>  %285 = tail call float @llvm.fmuladd.f32(float %284, float %mul57, float 1.000000e+00) #13
>  %286 = fcmp olt float %285, 0.000000e+00
>  %287 = tail call float @llvm.sqrt.f32(float %285) #14
>  %288 = select i1 %286, float 0x7FF8000000000000, float %287
>  %289 = tail call float @llvm.fabs.f32(float %288) #14
>  %290 = tail call float @llvm.fabs.f32(float %mul57) #14
>  %291 = fcmp olt float %290, %289
>  %292 = select i1 %291, float %290, float %289
>  %293 = select i1 %291, float %289, float %290
>  %294 = fcmp ogt float %293, 0x45F0000000000000
>  %295 = select i1 %294, float 0x3DF0000000000000, float 1.000000e+00
>  %296 = fmul float %293, %295
>  %297 = fdiv float %292, %296, !fpmath !11
>  %298 = fmul float %295, %297
>  %299 = fmul float %298, %298
>  %300 = tail call float @llvm.fmuladd.f32(float %299, float 0xBF67E1F780000000, float 0xBFC7D1B980000000) #13
>  %301 = tail call float @llvm.fmuladd.f32(float %299, float %300, float 0xBFD5554D00000000) #13
>  %302 = fmul float %299, %301
>  %303 = fmul float %298, %302
>  %304 = tail call float @llvm.fmuladd.f32(float %299, float 0x3FD1A714C0000000, float 0x3FF287C560000000) #13
>  %305 = tail call float @llvm.fmuladd.f32(float %299, float %304, float 1.000000e+00) #13
>  %306 = fdiv float 1.000000e+00, %305, !fpmath !11
>  %307 = tail call float @llvm.fmuladd.f32(float %303, float %306, float %298) #13
>  %308 = fsub float 0x3FF921FB60000000, %307
>  %309 = fcmp ogt float %290, %289
>  %310 = select i1 %309, float %308, float %307
>  %311 = bitcast float %288 to i32
>  %312 = icmp slt i32 %311, 0
>  %313 = select i1 %312, float 0x400921FB60000000, float 0.000000e+00
>  %314 = fcmp oeq float %mul57, 0.000000e+00
>  %315 = select i1 %314, float %313, float %310
>  %316 = fcmp ogt float %288, 0.000000e+00
>  %317 = select i1 %316, float 0x3FE921FB60000000, float 0x4002D97C80000000
>  %318 = fcmp oeq float %289, 0x7FF0000000000000
>  %319 = fcmp oeq float %290, 0x7FF0000000000000
>  %320 = and i1 %319, %318
>  %321 = select i1 %320, float %317, float %315
>  %322 = fcmp ord float %288, %mul57
>  %323 = select i1 %322, float %321, float 0x7FF8000000000000
>  %324 = tail call float @llvm.copysign.f32(float %323, float %mul57) #13
>  %325 = tail call float @llvm.fmuladd.f32(float %div58, float %324, float %cond54)
>  %cmp61 = fcmp ugt float %325, -1.000000e+00
>  %div63 = fmul float %cond48185, 5.000000e-01
>  %sub = fsub float %cond54, %div63
>  %cond66 = select i1 %cmp61, float %325, float %sub
>  %cmp67 = fcmp ult float %cond66, 1.000000e+00
>  %add = fadd float %div63, %cond54
>  %cond72 = select i1 %cmp67, float %cond66, float %add
>  %mul73 = fmul float %cond72, 6.553500e+04
>  br label %sw.epilog
>
>sw.bb74:                                          ; preds = %entry
>  %cmp78 = icmp eq i32 %number_parameters, 0
>  br i1 %cmp78, label %cond.end100, label %cond.end82
>
>cond.end82:                                       ; preds = %sw.bb74
>  %326 = load float, float addrspace(4)* %parameters, align 4, !tbaa !12
>  %phitmp = fmul float %326, 0x400921FB80000000
>  %cmp84 = icmp eq i32 %number_parameters, 1
>  br i1 %cmp84, label %cond.end100, label %cond.end88
>
>cond.end88:                                       ; preds = %cond.end82
>  %arrayidx86 = getelementptr inbounds float, float addrspace(4)* %parameters, i32 1
>  %327 = load float, float addrspace(4)* %arrayidx86, align 4, !tbaa !12
>  %cmp90 = icmp ugt i32 %number_parameters, 2
>  br i1 %cmp90, label %cond.end94, label %cond.end100
>
>cond.end94:                                       ; preds = %cond.end88
>  %arrayidx92 = getelementptr inbounds float, float addrspace(4)* %parameters, i32 2
>  %328 = load float, float addrspace(4)* %arrayidx92, align 4, !tbaa !12
>  %cmp96 = icmp eq i32 %number_parameters, 3
>  br i1 %cmp96, label %cond.end100, label %cond.true97
>
>cond.true97:                                      ; preds = %cond.end94
>  %arrayidx98 = getelementptr inbounds float, float addrspace(4)* %parameters, i32 3
>  %329 = load float, float addrspace(4)* %arrayidx98, align 4, !tbaa !12
>  br label %cond.end100
>
>cond.end100:                                      ; preds = %cond.end94, %cond.end82, %sw.bb74, %cond.end88, %cond.true97
>  %cond95194 = phi float [ %328, %cond.true97 ], [ %328, %cond.end94 ], [ 1.000000e+00, %cond.end88 ], [ 1.000000e+00, %cond.end82 ], [ 1.000000e+00, %sw.bb74 ]
>  %cond83187189193 = phi float [ %phitmp, %cond.true97 ], [ %phitmp, %cond.end94 ], [ %phitmp, %cond.end88 ], [ %phitmp, %cond.end82 ], [ 0x400921FB80000000, %sw.bb74 ]
>  %cond89190192 = phi float [ %327, %cond.true97 ], [ %327, %cond.end94 ], [ %327, %cond.end88 ], [ 5.000000e-01, %cond.end82 ], [ 5.000000e-01, %sw.bb74 ]
>  %cond101 = phi float [ %329, %cond.true97 ], [ 5.000000e-01, %cond.end94 ], [ 5.000000e-01, %cond.end88 ], [ 5.000000e-01, %cond.end82 ], [ 5.000000e-01, %sw.bb74 ]
>  %neg104 = fsub float -0.000000e+00, %cond89190192
>  %330 = tail call float @llvm.fmuladd.f32(float %pixel, float 0x3EEF751040000000, float %neg104)
>  %mul105 = fmul float %cond83187189193, %330
>  %div106 = fdiv float %cond95194, 0x400921FB80000000, !fpmath !11
>  %331 = bitcast float %mul105 to i32
>  %332 = and i32 %331, 2147483647
>  %333 = and i32 %331, -2147483648
>  %334 = or i32 %333, 1070141403
>  %335 = bitcast i32 %334 to float
>  %336 = bitcast i32 %332 to float
>  %337 = icmp ult i32 %332, 2139095041
>  %338 = select i1 %337, float %335, float %mul105
>  %339 = fdiv float 1.000000e+00, %336, !fpmath !11
>  %340 = fsub float -0.000000e+00, %339
>  %341 = icmp ult i32 %332, 1075576832
>  %342 = fadd float %336, -1.500000e+00
>  %343 = tail call float @llvm.fmuladd.f32(float %336, float 1.500000e+00, float 1.000000e+00) #13
>  %344 = fdiv float %342, %343, !fpmath !11
>  %345 = select i1 %341, float %344, float %340
>  %346 = select i1 %341, float 0x3FEF730BE0000000, float 0x3FF921FB60000000
>  %347 = icmp ult i32 %332, 1066926080
>  %348 = fadd float %336, -1.000000e+00
>  %349 = fadd float %336, 1.000000e+00
>  %350 = fdiv float %348, %349, !fpmath !11
>  %351 = select i1 %347, float %350, float %345
>  %352 = select i1 %347, float 0x3FE921FB60000000, float %346
>  %353 = icmp ult i32 %332, 1060110336
>  %354 = tail call float @llvm.fmuladd.f32(float %336, float 2.000000e+00, float -1.000000e+00) #13
>  %355 = fadd float %336, 2.000000e+00
>  %356 = fdiv float %354, %355, !fpmath !11
>  %357 = select i1 %353, float %356, float %351
>  %358 = select i1 %353, float 0x3FDDAC6700000000, float %352
>  %359 = icmp ult i32 %332, 1054867456
>  %360 = select i1 %359, float %336, float %357
>  %361 = select i1 %359, float 0.000000e+00, float %358
>  %362 = fmul float %360, %360
>  %363 = tail call float @llvm.fmuladd.f32(float %362, float 0x3F73476A80000000, float 0x3FC89E1740000000) #13
>  %364 = tail call float @llvm.fmuladd.f32(float %362, float %363, float 0x3FD2FA5320000000) #13
>  %365 = tail call float @llvm.fmuladd.f32(float %362, float 0x3FD327E3E0000000, float 0x3FF1C587A0000000) #13
>  %366 = tail call float @llvm.fmuladd.f32(float %362, float %365, float 0x3FEC777CA0000000) #13
>  %367 = fmul float %360, %362
>  %368 = fdiv float %364, %366, !fpmath !11
>  %369 = fmul float %367, %368
>  %370 = fsub float %369, %360
>  %371 = fsub float %361, %370
>  %372 = bitcast float %371 to i32
>  %373 = or i32 %333, %372
>  %374 = bitcast i32 %373 to float
>  %375 = icmp ult i32 %332, 1283457024
>  %376 = select i1 %375, float %374, float %338
>  %377 = icmp ult i32 %332, 905969664
>  %378 = select i1 %377, float %mul105, float %376
>  %379 = tail call float @llvm.fmuladd.f32(float %div106, float %378, float %cond101)
>  %mul109 = fmul float %379, 6.553500e+04
>  br label %sw.epilog
>
>sw.epilog:                                        ; preds = %for.cond.preheader, %for.cond.cleanup.loopexit, %entry, %cond.end100, %cond.end53, %_Z3sinf.exit
>  %result.1 = phi float [ 0.000000e+00, %entry ], [ %mul109, %cond.end100 ], [ %mul73, %cond.end53 ], [ %mul28, %_Z3sinf.exit ], [ 0.000000e+00, %for.cond.preheader ], [ %phitmp198, %for.cond.cleanup.loopexit ]
>  %380 = fcmp ogt float %result.1, 6.553500e+04
>  %381 = fcmp olt float %result.1, 0.000000e+00
>  %382 = select i1 %381, float 0.000000e+00, float %result.1
>  %.op.i = fadd float %382, 5.000000e-01
>  %add.i = select i1 %380, float 6.553550e+04, float %.op.i
>  ret float %add.i
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @ComputeFunction(float addrspace(1)* nocapture %image, i32 %number_channels, i32 %channel, i32 %function, i32 %number_parameters, float addrspace(4)* nocapture readonly %parameters) local_unnamed_addr #4 !kernel_arg_addr_space !53 !kernel_arg_access_qual !54 !kernel_arg_type !55 !kernel_arg_base_type !55 !kernel_arg_type_qual !56 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %x.i = tail call i32 @llvm.r600.read.global.size.x() #16
>  %mul1.i = mul i32 %16, %x.i
>  %mul4.i = add i32 %8, %mul1.i
>  %add.i = mul i32 %mul4.i, %number_channels
>  %add.ptr = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i
>  %and.i49 = and i32 %channel, 1
>  %cmp.i50 = icmp eq i32 %and.i49, 0
>  br i1 %cmp.i50, label %if.end.i53, label %if.then.i51
>
>if.then.i51:                                      ; preds = %entry
>  %17 = bitcast float addrspace(1)* %add.ptr to i32 addrspace(1)*
>  %18 = load i32, i32 addrspace(1)* %17, align 4, !tbaa !12
>  %19 = bitcast i32 %18 to float
>  br label %if.end.i53
>
>if.end.i53:                                       ; preds = %if.then.i51, %entry
>  %20 = phi float [ undef, %entry ], [ %19, %if.then.i51 ]
>  %red.sroa.0.0 = phi i32 [ undef, %entry ], [ %18, %if.then.i51 ]
>  %cmp1.i52 = icmp ugt i32 %number_channels, 2
>  br i1 %cmp1.i52, label %if.then2.i56, label %if.end13.i62
>
>if.then2.i56:                                     ; preds = %if.end.i53
>  %and3.i54 = and i32 %channel, 2
>  %cmp4.i55 = icmp eq i32 %and3.i54, 0
>  br i1 %cmp4.i55, label %if.end7.i60, label %if.then5.i57
>
>if.then5.i57:                                     ; preds = %if.then2.i56
>  %add.ptr.i31.i = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 1
>  %21 = bitcast float addrspace(1)* %add.ptr.i31.i to i32 addrspace(1)*
>  %22 = load i32, i32 addrspace(1)* %21, align 4, !tbaa !12
>  %23 = bitcast i32 %22 to float
>  br label %if.end7.i60
>
>if.end7.i60:                                      ; preds = %if.then5.i57, %if.then2.i56
>  %24 = phi float [ undef, %if.then2.i56 ], [ %23, %if.then5.i57 ]
>  %green.sroa.0.0 = phi i32 [ undef, %if.then2.i56 ], [ %22, %if.then5.i57 ]
>  %and8.i58 = and i32 %channel, 4
>  %cmp9.i59 = icmp eq i32 %and8.i58, 0
>  br i1 %cmp9.i59, label %if.end13.i62, label %if.then10.i61
>
>if.then10.i61:                                    ; preds = %if.end7.i60
>  %add.ptr.i30.i = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 2
>  %25 = load float, float addrspace(1)* %add.ptr.i30.i, align 4, !tbaa !12
>  br label %if.end13.i62
>
>if.end13.i62:                                     ; preds = %if.then10.i61, %if.end7.i60, %if.end.i53
>  %26 = phi float [ %24, %if.end7.i60 ], [ %24, %if.then10.i61 ], [ undef, %if.end.i53 ]
>  %green.sroa.0.1 = phi i32 [ %green.sroa.0.0, %if.end7.i60 ], [ %green.sroa.0.0, %if.then10.i61 ], [ undef, %if.end.i53 ]
>  %27 = phi float [ undef, %if.end7.i60 ], [ %25, %if.then10.i61 ], [ undef, %if.end.i53 ]
>  switch i32 %number_channels, label %ReadChannels.exit [
>    i32 4, label %land.lhs.true.i65
>    i32 2, label %land.lhs.true.i65
>  ]
>
>land.lhs.true.i65:                                ; preds = %if.end13.i62, %if.end13.i62
>  %and16.i63 = and i32 %channel, 16
>  %cmp17.i64 = icmp eq i32 %and16.i63, 0
>  br i1 %cmp17.i64, label %ReadChannels.exit, label %if.then18.i68
>
>if.then18.i68:                                    ; preds = %land.lhs.true.i65
>  %add.ptr.i.i66 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %number_channels
>  %add.ptr1.i.i67 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i66, i32 -1
>  %28 = bitcast float addrspace(1)* %add.ptr1.i.i67 to i32 addrspace(1)*
>  %29 = load i32, i32 addrspace(1)* %28, align 4, !tbaa !12
>  %30 = bitcast i32 %29 to float
>  br label %ReadChannels.exit
>
>ReadChannels.exit:                                ; preds = %if.end13.i62, %land.lhs.true.i65, %if.then18.i68
>  %31 = phi float [ undef, %if.end13.i62 ], [ undef, %land.lhs.true.i65 ], [ %30, %if.then18.i68 ]
>  %alpha.sroa.0.0 = phi i32 [ undef, %if.end13.i62 ], [ undef, %land.lhs.true.i65 ], [ %29, %if.then18.i68 ]
>  br i1 %cmp.i50, label %if.end, label %if.then
>
>if.then:                                          ; preds = %ReadChannels.exit
>  %call4 = tail call float @ApplyFunction(float %20, i32 %function, i32 %number_parameters, float addrspace(4)* %parameters) #15
>  %32 = bitcast float %call4 to i32
>  br label %if.end
>
>if.end:                                           ; preds = %ReadChannels.exit, %if.then
>  %red.sroa.0.1 = phi i32 [ %red.sroa.0.0, %ReadChannels.exit ], [ %32, %if.then ]
>  br i1 %cmp1.i52, label %if.then6, label %if.end17
>
>if.then6:                                         ; preds = %if.end
>  %and7 = and i32 %channel, 2
>  %cmp8 = icmp eq i32 %and7, 0
>  br i1 %cmp8, label %if.end11, label %if.then9
>
>if.then9:                                         ; preds = %if.then6
>  %33 = bitcast i32 %green.sroa.0.1 to float
>  %call10 = tail call float @ApplyFunction(float %33, i32 %function, i32 %number_parameters, float addrspace(4)* %parameters) #15
>  br label %if.end11
>
>if.end11:                                         ; preds = %if.then6, %if.then9
>  %34 = phi float [ %26, %if.then6 ], [ %call10, %if.then9 ]
>  %and12 = and i32 %channel, 4
>  %cmp13 = icmp eq i32 %and12, 0
>  br i1 %cmp13, label %if.end17, label %if.then14
>
>if.then14:                                        ; preds = %if.end11
>  %call15 = tail call float @ApplyFunction(float %27, i32 %function, i32 %number_parameters, float addrspace(4)* %parameters) #15
>  br label %if.end17
>
>if.end17:                                         ; preds = %if.end11, %if.then14, %if.end
>  %35 = phi float [ %34, %if.end11 ], [ %34, %if.then14 ], [ %26, %if.end ]
>  %36 = phi float [ %27, %if.end11 ], [ %call15, %if.then14 ], [ %27, %if.end ]
>  switch i32 %number_channels, label %if.end24 [
>    i32 4, label %land.lhs.true
>    i32 2, label %land.lhs.true
>  ]
>
>land.lhs.true:                                    ; preds = %if.end17, %if.end17
>  %and20 = and i32 %channel, 16
>  %cmp21 = icmp eq i32 %and20, 0
>  br i1 %cmp21, label %if.end24, label %if.then22
>
>if.then22:                                        ; preds = %land.lhs.true
>  %call23 = tail call float @ApplyFunction(float %31, i32 %function, i32 %number_parameters, float addrspace(4)* %parameters) #15
>  %37 = bitcast float %call23 to i32
>  br label %if.end24
>
>if.end24:                                         ; preds = %if.end17, %land.lhs.true, %if.then22
>  %alpha.sroa.0.1 = phi i32 [ %alpha.sroa.0.0, %if.end17 ], [ %alpha.sroa.0.0, %land.lhs.true ], [ %37, %if.then22 ]
>  %38 = bitcast i32 %red.sroa.0.1 to float
>  %39 = bitcast i32 %alpha.sroa.0.1 to float
>  br i1 %cmp.i50, label %if.end.i, label %if.then.i
>
>if.then.i:                                        ; preds = %if.end24
>  %40 = fcmp ogt float %38, 6.553500e+04
>  %41 = fcmp olt float %38, 0.000000e+00
>  %42 = select i1 %41, float 0.000000e+00, float %38
>  %.op.i.i = fadd float %42, 5.000000e-01
>  %add.i.i = select i1 %40, float 6.553550e+04, float %.op.i.i
>  store float %add.i.i, float addrspace(1)* %add.ptr, align 4, !tbaa !12
>  br label %if.end.i
>
>if.end.i:                                         ; preds = %if.then.i, %if.end24
>  br i1 %cmp1.i52, label %if.then2.i, label %if.end13.i
>
>if.then2.i:                                       ; preds = %if.end.i
>  %and3.i = and i32 %channel, 2
>  %cmp4.i = icmp eq i32 %and3.i, 0
>  br i1 %cmp4.i, label %if.end7.i, label %if.then5.i
>
>if.then5.i:                                       ; preds = %if.then2.i
>  %43 = fcmp ogt float %35, 6.553500e+04
>  %44 = fcmp olt float %35, 0.000000e+00
>  %45 = select i1 %44, float 0.000000e+00, float %35
>  %.op.i36.i = fadd float %45, 5.000000e-01
>  %add.i37.i = select i1 %43, float 6.553550e+04, float %.op.i36.i
>  %add.ptr.i35.i = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 1
>  store float %add.i37.i, float addrspace(1)* %add.ptr.i35.i, align 4, !tbaa !12
>  br label %if.end7.i
>
>if.end7.i:                                        ; preds = %if.then5.i, %if.then2.i
>  %and8.i = and i32 %channel, 4
>  %cmp9.i = icmp eq i32 %and8.i, 0
>  br i1 %cmp9.i, label %if.end13.i, label %if.then10.i
>
>if.then10.i:                                      ; preds = %if.end7.i
>  %46 = fcmp ogt float %36, 6.553500e+04
>  %47 = fcmp olt float %36, 0.000000e+00
>  %48 = select i1 %47, float 0.000000e+00, float %36
>  %.op.i33.i = fadd float %48, 5.000000e-01
>  %add.i34.i = select i1 %46, float 6.553550e+04, float %.op.i33.i
>  %add.ptr.i32.i = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 2
>  store float %add.i34.i, float addrspace(1)* %add.ptr.i32.i, align 4, !tbaa !12
>  br label %if.end13.i
>
>if.end13.i:                                       ; preds = %if.then10.i, %if.end7.i, %if.end.i
>  switch i32 %number_channels, label %WriteChannels.exit [
>    i32 4, label %land.lhs.true.i
>    i32 2, label %land.lhs.true.i
>  ]
>
>land.lhs.true.i:                                  ; preds = %if.end13.i, %if.end13.i
>  %and16.i = and i32 %channel, 16
>  %cmp17.i = icmp eq i32 %and16.i, 0
>  br i1 %cmp17.i, label %WriteChannels.exit, label %if.then18.i
>
>if.then18.i:                                      ; preds = %land.lhs.true.i
>  %49 = fcmp ogt float %39, 6.553500e+04
>  %50 = fcmp olt float %39, 0.000000e+00
>  %51 = select i1 %50, float 0.000000e+00, float %39
>  %.op.i30.i = fadd float %51, 5.000000e-01
>  %add.i31.i = select i1 %49, float 6.553550e+04, float %.op.i30.i
>  %add.ptr.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %number_channels
>  %add.ptr1.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i, i32 -1
>  store float %add.i31.i, float addrspace(1)* %add.ptr1.i.i, align 4, !tbaa !12
>  br label %WriteChannels.exit
>
>WriteChannels.exit:                               ; preds = %if.end13.i, %land.lhs.true.i, %if.then18.i
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @Grayscale(float addrspace(1)* nocapture %image, i32 %number_channels, i32 %colorspace, i32 %method) local_unnamed_addr #4 !kernel_arg_addr_space !57 !kernel_arg_access_qual !58 !kernel_arg_type !59 !kernel_arg_base_type !59 !kernel_arg_type_qual !60 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %x.i = tail call i32 @llvm.r600.read.global.size.x() #16
>  %mul1.i = mul i32 %16, %x.i
>  %mul4.i = add i32 %8, %mul1.i
>  %add.i28 = mul i32 %mul4.i, %number_channels
>  %add.ptr = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i28
>  %17 = load float, float addrspace(1)* %add.ptr, align 4, !tbaa !12
>  %add.ptr.i27 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 1
>  %18 = load float, float addrspace(1)* %add.ptr.i27, align 4, !tbaa !12
>  %add.ptr.i26 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 2
>  %19 = load float, float addrspace(1)* %add.ptr.i26, align 4, !tbaa !12
>  switch i32 %colorspace, label %if.end.i [
>    i32 33, label %GetPixelIntensity.exit
>    i32 3, label %GetPixelIntensity.exit
>  ]
>
>if.end.i:                                         ; preds = %entry
>  switch i32 %method, label %sw.default.i [
>    i32 1, label %sw.bb.i
>    i32 2, label %sw.bb3.i
>    i32 3, label %sw.bb15.i
>    i32 4, label %sw.bb48.i
>    i32 5, label %sw.bb51.i
>    i32 6, label %sw.bb53.i
>    i32 9, label %sw.bb59.i
>    i32 8, label %sw.bb57.i
>  ]
>
>sw.bb.i:                                          ; preds = %if.end.i
>  %add.i25 = fadd float %17, %18
>  %add2.i = fadd float %add.i25, %19
>  %div.i = fdiv float %add2.i, 3.000000e+00, !fpmath !11
>  br label %GetPixelIntensity.exit
>
>sw.bb3.i:                                         ; preds = %if.end.i
>  %cmp4.i = fcmp ogt float %17, %18
>  %cond.i = select i1 %cmp4.i, float %17, float %18
>  %cmp5.i = fcmp ogt float %cond.i, %19
>  %cond.blue.i = select i1 %cmp5.i, float %cond.i, float %19
>  br label %GetPixelIntensity.exit
>
>sw.bb15.i:                                        ; preds = %if.end.i
>  %cmp16.i = fcmp olt float %17, %18
>  %cond20.i = select i1 %cmp16.i, float %17, float %18
>  %cmp21.i = fcmp olt float %cond20.i, %19
>  %cond20.blue.i = select i1 %cmp21.i, float %cond20.i, float %19
>  %cmp31.i = fcmp ogt float %17, %18
>  %cond35.i = select i1 %cmp31.i, float %17, float %18
>  %cmp36.i = fcmp ogt float %cond35.i, %19
>  %cond45.i = select i1 %cmp36.i, float %cond35.i, float %19
>  %add46.i = fadd float %cond20.blue.i, %cond45.i
>  %div47.i = fmul float %add46.i, 5.000000e-01
>  br label %GetPixelIntensity.exit
>
>sw.bb48.i:                                        ; preds = %if.end.i
>  %mul49.i = fmul float %18, %18
>  %20 = tail call float @llvm.fmuladd.f32(float %17, float %17, float %mul49.i) #13
>  %21 = tail call float @llvm.fmuladd.f32(float %19, float %19, float %20) #13
>  %div50.i = fdiv float %21, 1.966050e+05, !fpmath !11
>  br label %GetPixelIntensity.exit
>
>sw.bb51.i:                                        ; preds = %if.end.i
>  %mul52.i = fmul float %18, 0x3FE2C727E0000000
>  %22 = tail call float @llvm.fmuladd.f32(float %17, float 0x3FD3202DA0000000, float %mul52.i) #13
>  %23 = tail call float @llvm.fmuladd.f32(float %19, float 0x3FBD460AA0000000, float %22) #13
>  br label %GetPixelIntensity.exit
>
>sw.bb53.i:                                        ; preds = %if.end.i
>  %mul54.i = fmul float %18, 0x3FE2C727E0000000
>  %24 = tail call float @llvm.fmuladd.f32(float %17, float 0x3FD3202DA0000000, float %mul54.i) #13
>  %25 = tail call float @llvm.fmuladd.f32(float %19, float 0x3FBD460AA0000000, float %24) #13
>  br label %GetPixelIntensity.exit
>
>sw.default.i:                                     ; preds = %if.end.i
>  %mul56.i = fmul float %18, 0x3FE6E29300000000
>  %26 = tail call float @llvm.fmuladd.f32(float %17, float 0x3FCB384FE0000000, float %mul56.i) #13
>  %27 = tail call float @llvm.fmuladd.f32(float %19, float 0x3FB27AC820000000, float %26) #13
>  br label %GetPixelIntensity.exit
>
>sw.bb57.i:                                        ; preds = %if.end.i
>  %mul58.i = fmul float %18, 0x3FE6E29300000000
>  %28 = tail call float @llvm.fmuladd.f32(float %17, float 0x3FCB384FE0000000, float %mul58.i) #13
>  %29 = tail call float @llvm.fmuladd.f32(float %19, float 0x3FB27AC820000000, float %28) #13
>  br label %GetPixelIntensity.exit
>
>sw.bb59.i:                                        ; preds = %if.end.i
>  %mul60.i = fmul float %18, %18
>  %30 = tail call float @llvm.fmuladd.f32(float %17, float %17, float %mul60.i) #13
>  %31 = tail call float @llvm.fmuladd.f32(float %19, float %19, float %30) #13
>  %32 = tail call float @llvm.sqrt.f32(float %31) #14
>  %33 = tail call float @llvm.sqrt.f32(float 3.000000e+00) #14
>  %div62.i = fdiv float %32, %33, !fpmath !11
>  br label %GetPixelIntensity.exit
>
>GetPixelIntensity.exit:                           ; preds = %entry, %entry, %sw.bb.i, %sw.bb3.i, %sw.bb15.i, %sw.bb48.i, %sw.bb51.i, %sw.bb53.i, %sw.default.i, %sw.bb57.i, %sw.bb59.i
>  %retval.0.i = phi float [ %17, %entry ], [ %17, %entry ], [ %27, %sw.default.i ], [ %29, %sw.bb57.i ], [ %div62.i, %sw.bb59.i ], [ %25, %sw.bb53.i ], [ %23, %sw.bb51.i ], [ %div50.i, %sw.bb48.i ], [ %div47.i, %sw.bb15.i ], [ %cond.blue.i, %sw.bb3.i ], [ %div.i, %sw.bb.i ]
>  %34 = fcmp ogt float %retval.0.i, 6.553500e+04
>  %35 = fcmp olt float %retval.0.i, 0.000000e+00
>  %36 = select i1 %35, float 0.000000e+00, float %retval.0.i
>  %.op.i = fadd float %36, 5.000000e-01
>  %add.i = select i1 %34, float 6.553550e+04, float %.op.i
>  store float %add.i, float addrspace(1)* %add.ptr, align 4, !tbaa !12
>  store float %add.i, float addrspace(1)* %add.ptr.i27, align 4, !tbaa !12
>  store float %add.i, float addrspace(1)* %add.ptr.i26, align 4, !tbaa !12
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @LocalContrastBlurRow(<4 x float> addrspace(1)* nocapture readonly %srcImage, <4 x float> addrspace(1)* nocapture readnone %dstImage, float addrspace(1)* nocapture %tmpImage, i32 %radius, i32 %imageWidth, i32 %imageHeight) local_unnamed_addr #4 !kernel_arg_addr_space !61 !kernel_arg_access_qual !54 !kernel_arg_type !62 !kernel_arg_base_type !63 !kernel_arg_type_qual !64 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %1 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %2 = mul i32 %y.i.i, %1
>  %3 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %4 = add i32 %2, %3
>  %5 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %6 = getelementptr inbounds i8, i8 addrspace(7)* %5, i32 8
>  %7 = bitcast i8 addrspace(7)* %6 to i32 addrspace(7)*
>  %8 = load i32, i32 addrspace(7)* %7, align 4, !tbaa !4
>  %9 = add i32 %4, %8
>  %cmp = icmp slt i32 %0, %imageWidth
>  %cmp2 = icmp slt i32 %9, %imageHeight
>  %or.cond = and i1 %cmp, %cmp2
>  br i1 %or.cond, label %if.end, label %cleanup
>
>if.end:                                           ; preds = %entry
>  %mul = mul nsw i32 %9, %imageWidth
>  %add.ptr = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %srcImage, i32 %mul
>  %mul.i142 = shl i32 %imageWidth, 1
>  %add68 = add nsw i32 %radius, 1
>  %mul70 = mul nsw i32 %add68, %add68
>  %conv71 = sitofp i32 %mul70 to float
>  %x.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %10 = add i32 %0, %radius
>  br label %for.body
>
>for.body:                                         ; preds = %if.end, %while.end67
>  %indvars.iv = phi i32 [ %10, %if.end ], [ %indvars.iv.next, %while.end67 ]
>  %i.0184 = phi i32 [ %0, %if.end ], [ %add77, %while.end67 ]
>  %sub = sub nsw i32 %i.0184, %radius
>  %add153 = add nsw i32 %sub, 7
>  %cmp4154 = icmp slt i32 %add153, %i.0184
>  br i1 %cmp4154, label %for.cond5.preheader, label %while.cond17.preheader
>
>while.cond17.preheader:                           ; preds = %for.cond5.preheader, %for.body
>  %sum.0.lcssa = phi float [ 0.000000e+00, %for.body ], [ %90, %for.cond5.preheader ]
>  %weight.0.lcssa = phi float [ 1.000000e+00, %for.body ], [ %add15, %for.cond5.preheader ]
>  %j.0.lcssa = phi i32 [ %sub, %for.body ], [ %add16, %for.cond5.preheader ]
>  %cmp18160 = icmp slt i32 %j.0.lcssa, %i.0184
>  br i1 %cmp18160, label %while.body20, label %while.cond29.preheader
>
>for.cond5.preheader:                              ; preds = %for.body, %for.cond5.preheader
>  %j.0157 = phi i32 [ %add16, %for.cond5.preheader ], [ %sub, %for.body ]
>  %weight.0156 = phi float [ %add15, %for.cond5.preheader ], [ 1.000000e+00, %for.body ]
>  %sum.0155 = phi float [ %90, %for.cond5.preheader ], [ 0.000000e+00, %for.body ]
>  %add9 = fadd float %weight.0156, 0.000000e+00
>  %cmp.i148 = icmp slt i32 %j.0157, 0
>  %sub.i149 = sub nsw i32 0, %j.0157
>  %cond.i150 = select i1 %cmp.i148, i32 %sub.i149, i32 %j.0157
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %cond.i150
>  %11 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %12 = extractelement <4 x float> %11, i32 0
>  %13 = extractelement <4 x float> %11, i32 1
>  %14 = fmul float %13, 0x3FE6E2EB20000000
>  %15 = tail call float @llvm.fmuladd.f32(float %12, float 0x3FCB367A00000000, float %14) #13
>  %16 = extractelement <4 x float> %11, i32 2
>  %17 = tail call float @llvm.fmuladd.f32(float %16, float 0x3FB27BB300000000, float %15) #13
>  %18 = extractelement <4 x float> %11, i32 3
>  %19 = tail call float @llvm.fmuladd.f32(float %18, float 0.000000e+00, float %17) #13
>  %20 = tail call float @llvm.fmuladd.f32(float %add9, float %19, float %sum.0155)
>  %add9.1 = fadd float %weight.0156, 1.000000e+00
>  %add10.1 = add nsw i32 %j.0157, 1
>  %cmp.i148.1 = icmp slt i32 %j.0157, -1
>  %sub.i149.1 = xor i32 %j.0157, -1
>  %cond.i150.1 = select i1 %cmp.i148.1, i32 %sub.i149.1, i32 %add10.1
>  %arrayidx.1 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %cond.i150.1
>  %21 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx.1, align 16, !tbaa !14
>  %22 = extractelement <4 x float> %21, i32 0
>  %23 = extractelement <4 x float> %21, i32 1
>  %24 = fmul float %23, 0x3FE6E2EB20000000
>  %25 = tail call float @llvm.fmuladd.f32(float %22, float 0x3FCB367A00000000, float %24) #13
>  %26 = extractelement <4 x float> %21, i32 2
>  %27 = tail call float @llvm.fmuladd.f32(float %26, float 0x3FB27BB300000000, float %25) #13
>  %28 = extractelement <4 x float> %21, i32 3
>  %29 = tail call float @llvm.fmuladd.f32(float %28, float 0.000000e+00, float %27) #13
>  %30 = tail call float @llvm.fmuladd.f32(float %add9.1, float %29, float %20)
>  %add9.2 = fadd float %weight.0156, 2.000000e+00
>  %add10.2 = add nsw i32 %j.0157, 2
>  %cmp.i148.2 = icmp slt i32 %j.0157, -2
>  %sub.i149.2 = sub i32 -2, %j.0157
>  %cond.i150.2 = select i1 %cmp.i148.2, i32 %sub.i149.2, i32 %add10.2
>  %arrayidx.2 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %cond.i150.2
>  %31 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx.2, align 16, !tbaa !14
>  %32 = extractelement <4 x float> %31, i32 0
>  %33 = extractelement <4 x float> %31, i32 1
>  %34 = fmul float %33, 0x3FE6E2EB20000000
>  %35 = tail call float @llvm.fmuladd.f32(float %32, float 0x3FCB367A00000000, float %34) #13
>  %36 = extractelement <4 x float> %31, i32 2
>  %37 = tail call float @llvm.fmuladd.f32(float %36, float 0x3FB27BB300000000, float %35) #13
>  %38 = extractelement <4 x float> %31, i32 3
>  %39 = tail call float @llvm.fmuladd.f32(float %38, float 0.000000e+00, float %37) #13
>  %40 = tail call float @llvm.fmuladd.f32(float %add9.2, float %39, float %30)
>  %add9.3 = fadd float %weight.0156, 3.000000e+00
>  %add10.3 = add nsw i32 %j.0157, 3
>  %cmp.i148.3 = icmp slt i32 %j.0157, -3
>  %sub.i149.3 = sub i32 -3, %j.0157
>  %cond.i150.3 = select i1 %cmp.i148.3, i32 %sub.i149.3, i32 %add10.3
>  %arrayidx.3 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %cond.i150.3
>  %41 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx.3, align 16, !tbaa !14
>  %42 = extractelement <4 x float> %41, i32 0
>  %43 = extractelement <4 x float> %41, i32 1
>  %44 = fmul float %43, 0x3FE6E2EB20000000
>  %45 = tail call float @llvm.fmuladd.f32(float %42, float 0x3FCB367A00000000, float %44) #13
>  %46 = extractelement <4 x float> %41, i32 2
>  %47 = tail call float @llvm.fmuladd.f32(float %46, float 0x3FB27BB300000000, float %45) #13
>  %48 = extractelement <4 x float> %41, i32 3
>  %49 = tail call float @llvm.fmuladd.f32(float %48, float 0.000000e+00, float %47) #13
>  %50 = tail call float @llvm.fmuladd.f32(float %add9.3, float %49, float %40)
>  %add9.4 = fadd float %weight.0156, 4.000000e+00
>  %add10.4 = add nsw i32 %j.0157, 4
>  %cmp.i148.4 = icmp slt i32 %j.0157, -4
>  %sub.i149.4 = sub i32 -4, %j.0157
>  %cond.i150.4 = select i1 %cmp.i148.4, i32 %sub.i149.4, i32 %add10.4
>  %arrayidx.4 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %cond.i150.4
>  %51 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx.4, align 16, !tbaa !14
>  %52 = extractelement <4 x float> %51, i32 0
>  %53 = extractelement <4 x float> %51, i32 1
>  %54 = fmul float %53, 0x3FE6E2EB20000000
>  %55 = tail call float @llvm.fmuladd.f32(float %52, float 0x3FCB367A00000000, float %54) #13
>  %56 = extractelement <4 x float> %51, i32 2
>  %57 = tail call float @llvm.fmuladd.f32(float %56, float 0x3FB27BB300000000, float %55) #13
>  %58 = extractelement <4 x float> %51, i32 3
>  %59 = tail call float @llvm.fmuladd.f32(float %58, float 0.000000e+00, float %57) #13
>  %60 = tail call float @llvm.fmuladd.f32(float %add9.4, float %59, float %50)
>  %add9.5 = fadd float %weight.0156, 5.000000e+00
>  %add10.5 = add nsw i32 %j.0157, 5
>  %cmp.i148.5 = icmp slt i32 %j.0157, -5
>  %sub.i149.5 = sub i32 -5, %j.0157
>  %cond.i150.5 = select i1 %cmp.i148.5, i32 %sub.i149.5, i32 %add10.5
>  %arrayidx.5 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %cond.i150.5
>  %61 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx.5, align 16, !tbaa !14
>  %62 = extractelement <4 x float> %61, i32 0
>  %63 = extractelement <4 x float> %61, i32 1
>  %64 = fmul float %63, 0x3FE6E2EB20000000
>  %65 = tail call float @llvm.fmuladd.f32(float %62, float 0x3FCB367A00000000, float %64) #13
>  %66 = extractelement <4 x float> %61, i32 2
>  %67 = tail call float @llvm.fmuladd.f32(float %66, float 0x3FB27BB300000000, float %65) #13
>  %68 = extractelement <4 x float> %61, i32 3
>  %69 = tail call float @llvm.fmuladd.f32(float %68, float 0.000000e+00, float %67) #13
>  %70 = tail call float @llvm.fmuladd.f32(float %add9.5, float %69, float %60)
>  %add9.6 = fadd float %weight.0156, 6.000000e+00
>  %add10.6 = add nsw i32 %j.0157, 6
>  %cmp.i148.6 = icmp slt i32 %j.0157, -6
>  %sub.i149.6 = sub i32 -6, %j.0157
>  %cond.i150.6 = select i1 %cmp.i148.6, i32 %sub.i149.6, i32 %add10.6
>  %arrayidx.6 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %cond.i150.6
>  %71 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx.6, align 16, !tbaa !14
>  %72 = extractelement <4 x float> %71, i32 0
>  %73 = extractelement <4 x float> %71, i32 1
>  %74 = fmul float %73, 0x3FE6E2EB20000000
>  %75 = tail call float @llvm.fmuladd.f32(float %72, float 0x3FCB367A00000000, float %74) #13
>  %76 = extractelement <4 x float> %71, i32 2
>  %77 = tail call float @llvm.fmuladd.f32(float %76, float 0x3FB27BB300000000, float %75) #13
>  %78 = extractelement <4 x float> %71, i32 3
>  %79 = tail call float @llvm.fmuladd.f32(float %78, float 0.000000e+00, float %77) #13
>  %80 = tail call float @llvm.fmuladd.f32(float %add9.6, float %79, float %70)
>  %add9.7 = fadd float %weight.0156, 7.000000e+00
>  %add10.7 = add nsw i32 %j.0157, 7
>  %cmp.i148.7 = icmp slt i32 %j.0157, -7
>  %sub.i149.7 = sub i32 -7, %j.0157
>  %cond.i150.7 = select i1 %cmp.i148.7, i32 %sub.i149.7, i32 %add10.7
>  %arrayidx.7 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %cond.i150.7
>  %81 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx.7, align 16, !tbaa !14
>  %82 = extractelement <4 x float> %81, i32 0
>  %83 = extractelement <4 x float> %81, i32 1
>  %84 = fmul float %83, 0x3FE6E2EB20000000
>  %85 = tail call float @llvm.fmuladd.f32(float %82, float 0x3FCB367A00000000, float %84) #13
>  %86 = extractelement <4 x float> %81, i32 2
>  %87 = tail call float @llvm.fmuladd.f32(float %86, float 0x3FB27BB300000000, float %85) #13
>  %88 = extractelement <4 x float> %81, i32 3
>  %89 = tail call float @llvm.fmuladd.f32(float %88, float 0.000000e+00, float %87) #13
>  %90 = tail call float @llvm.fmuladd.f32(float %add9.7, float %89, float %80)
>  %add15 = fadd float %weight.0156, 8.000000e+00
>  %add16 = add nsw i32 %j.0157, 8
>  %add = add nsw i32 %j.0157, 15
>  %cmp4 = icmp slt i32 %add, %i.0184
>  br i1 %cmp4, label %for.cond5.preheader, label %while.cond17.preheader
>
>while.cond29.preheader:                           ; preds = %while.body20, %while.cond17.preheader
>  %sum.2.lcssa = phi float [ %sum.0.lcssa, %while.cond17.preheader ], [ %100, %while.body20 ]
>  %weight.1.lcssa = phi float [ %weight.0.lcssa, %while.cond17.preheader ], [ %add26, %while.body20 ]
>  %j.1.lcssa = phi i32 [ %j.0.lcssa, %while.cond17.preheader ], [ %i.0184, %while.body20 ]
>  %add30169 = add nsw i32 %j.1.lcssa, 7
>  %add31 = add nsw i32 %i.0184, %radius
>  %cmp32170 = icmp slt i32 %add30169, %add31
>  br i1 %cmp32170, label %for.cond36.preheader, label %while.cond55.preheader
>
>while.body20:                                     ; preds = %while.cond17.preheader, %while.body20
>  %j.1163 = phi i32 [ %inc27, %while.body20 ], [ %j.0.lcssa, %while.cond17.preheader ]
>  %weight.1162 = phi float [ %add26, %while.body20 ], [ %weight.0.lcssa, %while.cond17.preheader ]
>  %sum.2161 = phi float [ %100, %while.body20 ], [ %sum.0.lcssa, %while.cond17.preheader ]
>  %cmp.i146 = icmp slt i32 %j.1163, 0
>  %sub.i147 = sub nsw i32 0, %j.1163
>  %cond.i = select i1 %cmp.i146, i32 %sub.i147, i32 %j.1163
>  %arrayidx22 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %cond.i
>  %91 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx22, align 16, !tbaa !14
>  %92 = extractelement <4 x float> %91, i32 0
>  %93 = extractelement <4 x float> %91, i32 1
>  %94 = fmul float %93, 0x3FE6E2EB20000000
>  %95 = tail call float @llvm.fmuladd.f32(float %92, float 0x3FCB367A00000000, float %94) #13
>  %96 = extractelement <4 x float> %91, i32 2
>  %97 = tail call float @llvm.fmuladd.f32(float %96, float 0x3FB27BB300000000, float %95) #13
>  %98 = extractelement <4 x float> %91, i32 3
>  %99 = tail call float @llvm.fmuladd.f32(float %98, float 0.000000e+00, float %97) #13
>  %100 = tail call float @llvm.fmuladd.f32(float %weight.1162, float %99, float %sum.2161)
>  %add26 = fadd float %weight.1162, 1.000000e+00
>  %inc27 = add nsw i32 %j.1163, 1
>  %exitcond = icmp eq i32 %inc27, %i.0184
>  br i1 %exitcond, label %while.cond29.preheader, label %while.body20
>
>while.cond55.preheader:                           ; preds = %for.cond36.preheader, %while.cond29.preheader
>  %sum.3.lcssa = phi float [ %sum.2.lcssa, %while.cond29.preheader ], [ %180, %for.cond36.preheader ]
>  %weight.2.lcssa = phi float [ %weight.1.lcssa, %while.cond29.preheader ], [ %sub52, %for.cond36.preheader ]
>  %j.2.lcssa = phi i32 [ %j.1.lcssa, %while.cond29.preheader ], [ %add53, %for.cond36.preheader ]
>  %cmp57178 = icmp slt i32 %j.2.lcssa, %add31
>  br i1 %cmp57178, label %while.body59, label %while.end67
>
>for.cond36.preheader:                             ; preds = %while.cond29.preheader, %for.cond36.preheader
>  %j.2173 = phi i32 [ %add53, %for.cond36.preheader ], [ %j.1.lcssa, %while.cond29.preheader ]
>  %weight.2172 = phi float [ %sub52, %for.cond36.preheader ], [ %weight.1.lcssa, %while.cond29.preheader ]
>  %sum.3171 = phi float [ %180, %for.cond36.preheader ], [ %sum.2.lcssa, %while.cond29.preheader ]
>  %cmp.i141 = icmp slt i32 %j.2173, %imageWidth
>  %sub.i143 = xor i32 %j.2173, -1
>  %sub1.i144 = add i32 %mul.i142, %sub.i143
>  %spec.select.i145 = select i1 %cmp.i141, i32 %j.2173, i32 %sub1.i144
>  %arrayidx45 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %spec.select.i145
>  %101 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx45, align 16, !tbaa !14
>  %102 = extractelement <4 x float> %101, i32 0
>  %103 = extractelement <4 x float> %101, i32 1
>  %104 = fmul float %103, 0x3FE6E2EB20000000
>  %105 = tail call float @llvm.fmuladd.f32(float %102, float 0x3FCB367A00000000, float %104) #13
>  %106 = extractelement <4 x float> %101, i32 2
>  %107 = tail call float @llvm.fmuladd.f32(float %106, float 0x3FB27BB300000000, float %105) #13
>  %108 = extractelement <4 x float> %101, i32 3
>  %109 = tail call float @llvm.fmuladd.f32(float %108, float 0.000000e+00, float %107) #13
>  %110 = tail call float @llvm.fmuladd.f32(float %weight.2172, float %109, float %sum.3171)
>  %sub42.1 = fadd float %weight.2172, -1.000000e+00
>  %add43.1 = add nsw i32 %j.2173, 1
>  %cmp.i141.1 = icmp slt i32 %add43.1, %imageWidth
>  %sub.i143.1 = sub i32 -2, %j.2173
>  %sub1.i144.1 = add i32 %mul.i142, %sub.i143.1
>  %spec.select.i145.1 = select i1 %cmp.i141.1, i32 %add43.1, i32 %sub1.i144.1
>  %arrayidx45.1 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %spec.select.i145.1
>  %111 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx45.1, align 16, !tbaa !14
>  %112 = extractelement <4 x float> %111, i32 0
>  %113 = extractelement <4 x float> %111, i32 1
>  %114 = fmul float %113, 0x3FE6E2EB20000000
>  %115 = tail call float @llvm.fmuladd.f32(float %112, float 0x3FCB367A00000000, float %114) #13
>  %116 = extractelement <4 x float> %111, i32 2
>  %117 = tail call float @llvm.fmuladd.f32(float %116, float 0x3FB27BB300000000, float %115) #13
>  %118 = extractelement <4 x float> %111, i32 3
>  %119 = tail call float @llvm.fmuladd.f32(float %118, float 0.000000e+00, float %117) #13
>  %120 = tail call float @llvm.fmuladd.f32(float %sub42.1, float %119, float %110)
>  %sub42.2 = fadd float %weight.2172, -2.000000e+00
>  %add43.2 = add nsw i32 %j.2173, 2
>  %cmp.i141.2 = icmp slt i32 %add43.2, %imageWidth
>  %sub.i143.2 = sub i32 -3, %j.2173
>  %sub1.i144.2 = add i32 %mul.i142, %sub.i143.2
>  %spec.select.i145.2 = select i1 %cmp.i141.2, i32 %add43.2, i32 %sub1.i144.2
>  %arrayidx45.2 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %spec.select.i145.2
>  %121 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx45.2, align 16, !tbaa !14
>  %122 = extractelement <4 x float> %121, i32 0
>  %123 = extractelement <4 x float> %121, i32 1
>  %124 = fmul float %123, 0x3FE6E2EB20000000
>  %125 = tail call float @llvm.fmuladd.f32(float %122, float 0x3FCB367A00000000, float %124) #13
>  %126 = extractelement <4 x float> %121, i32 2
>  %127 = tail call float @llvm.fmuladd.f32(float %126, float 0x3FB27BB300000000, float %125) #13
>  %128 = extractelement <4 x float> %121, i32 3
>  %129 = tail call float @llvm.fmuladd.f32(float %128, float 0.000000e+00, float %127) #13
>  %130 = tail call float @llvm.fmuladd.f32(float %sub42.2, float %129, float %120)
>  %sub42.3 = fadd float %weight.2172, -3.000000e+00
>  %add43.3 = add nsw i32 %j.2173, 3
>  %cmp.i141.3 = icmp slt i32 %add43.3, %imageWidth
>  %sub.i143.3 = sub i32 -4, %j.2173
>  %sub1.i144.3 = add i32 %mul.i142, %sub.i143.3
>  %spec.select.i145.3 = select i1 %cmp.i141.3, i32 %add43.3, i32 %sub1.i144.3
>  %arrayidx45.3 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %spec.select.i145.3
>  %131 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx45.3, align 16, !tbaa !14
>  %132 = extractelement <4 x float> %131, i32 0
>  %133 = extractelement <4 x float> %131, i32 1
>  %134 = fmul float %133, 0x3FE6E2EB20000000
>  %135 = tail call float @llvm.fmuladd.f32(float %132, float 0x3FCB367A00000000, float %134) #13
>  %136 = extractelement <4 x float> %131, i32 2
>  %137 = tail call float @llvm.fmuladd.f32(float %136, float 0x3FB27BB300000000, float %135) #13
>  %138 = extractelement <4 x float> %131, i32 3
>  %139 = tail call float @llvm.fmuladd.f32(float %138, float 0.000000e+00, float %137) #13
>  %140 = tail call float @llvm.fmuladd.f32(float %sub42.3, float %139, float %130)
>  %sub42.4 = fadd float %weight.2172, -4.000000e+00
>  %add43.4 = add nsw i32 %j.2173, 4
>  %cmp.i141.4 = icmp slt i32 %add43.4, %imageWidth
>  %sub.i143.4 = sub i32 -5, %j.2173
>  %sub1.i144.4 = add i32 %mul.i142, %sub.i143.4
>  %spec.select.i145.4 = select i1 %cmp.i141.4, i32 %add43.4, i32 %sub1.i144.4
>  %arrayidx45.4 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %spec.select.i145.4
>  %141 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx45.4, align 16, !tbaa !14
>  %142 = extractelement <4 x float> %141, i32 0
>  %143 = extractelement <4 x float> %141, i32 1
>  %144 = fmul float %143, 0x3FE6E2EB20000000
>  %145 = tail call float @llvm.fmuladd.f32(float %142, float 0x3FCB367A00000000, float %144) #13
>  %146 = extractelement <4 x float> %141, i32 2
>  %147 = tail call float @llvm.fmuladd.f32(float %146, float 0x3FB27BB300000000, float %145) #13
>  %148 = extractelement <4 x float> %141, i32 3
>  %149 = tail call float @llvm.fmuladd.f32(float %148, float 0.000000e+00, float %147) #13
>  %150 = tail call float @llvm.fmuladd.f32(float %sub42.4, float %149, float %140)
>  %sub42.5 = fadd float %weight.2172, -5.000000e+00
>  %add43.5 = add nsw i32 %j.2173, 5
>  %cmp.i141.5 = icmp slt i32 %add43.5, %imageWidth
>  %sub.i143.5 = sub i32 -6, %j.2173
>  %sub1.i144.5 = add i32 %mul.i142, %sub.i143.5
>  %spec.select.i145.5 = select i1 %cmp.i141.5, i32 %add43.5, i32 %sub1.i144.5
>  %arrayidx45.5 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %spec.select.i145.5
>  %151 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx45.5, align 16, !tbaa !14
>  %152 = extractelement <4 x float> %151, i32 0
>  %153 = extractelement <4 x float> %151, i32 1
>  %154 = fmul float %153, 0x3FE6E2EB20000000
>  %155 = tail call float @llvm.fmuladd.f32(float %152, float 0x3FCB367A00000000, float %154) #13
>  %156 = extractelement <4 x float> %151, i32 2
>  %157 = tail call float @llvm.fmuladd.f32(float %156, float 0x3FB27BB300000000, float %155) #13
>  %158 = extractelement <4 x float> %151, i32 3
>  %159 = tail call float @llvm.fmuladd.f32(float %158, float 0.000000e+00, float %157) #13
>  %160 = tail call float @llvm.fmuladd.f32(float %sub42.5, float %159, float %150)
>  %sub42.6 = fadd float %weight.2172, -6.000000e+00
>  %add43.6 = add nsw i32 %j.2173, 6
>  %cmp.i141.6 = icmp slt i32 %add43.6, %imageWidth
>  %sub.i143.6 = sub i32 -7, %j.2173
>  %sub1.i144.6 = add i32 %mul.i142, %sub.i143.6
>  %spec.select.i145.6 = select i1 %cmp.i141.6, i32 %add43.6, i32 %sub1.i144.6
>  %arrayidx45.6 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %spec.select.i145.6
>  %161 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx45.6, align 16, !tbaa !14
>  %162 = extractelement <4 x float> %161, i32 0
>  %163 = extractelement <4 x float> %161, i32 1
>  %164 = fmul float %163, 0x3FE6E2EB20000000
>  %165 = tail call float @llvm.fmuladd.f32(float %162, float 0x3FCB367A00000000, float %164) #13
>  %166 = extractelement <4 x float> %161, i32 2
>  %167 = tail call float @llvm.fmuladd.f32(float %166, float 0x3FB27BB300000000, float %165) #13
>  %168 = extractelement <4 x float> %161, i32 3
>  %169 = tail call float @llvm.fmuladd.f32(float %168, float 0.000000e+00, float %167) #13
>  %170 = tail call float @llvm.fmuladd.f32(float %sub42.6, float %169, float %160)
>  %sub42.7 = fadd float %weight.2172, -7.000000e+00
>  %add43.7 = add nsw i32 %j.2173, 7
>  %cmp.i141.7 = icmp slt i32 %add43.7, %imageWidth
>  %sub.i143.7 = sub i32 -8, %j.2173
>  %sub1.i144.7 = add i32 %mul.i142, %sub.i143.7
>  %spec.select.i145.7 = select i1 %cmp.i141.7, i32 %add43.7, i32 %sub1.i144.7
>  %arrayidx45.7 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %spec.select.i145.7
>  %171 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx45.7, align 16, !tbaa !14
>  %172 = extractelement <4 x float> %171, i32 0
>  %173 = extractelement <4 x float> %171, i32 1
>  %174 = fmul float %173, 0x3FE6E2EB20000000
>  %175 = tail call float @llvm.fmuladd.f32(float %172, float 0x3FCB367A00000000, float %174) #13
>  %176 = extractelement <4 x float> %171, i32 2
>  %177 = tail call float @llvm.fmuladd.f32(float %176, float 0x3FB27BB300000000, float %175) #13
>  %178 = extractelement <4 x float> %171, i32 3
>  %179 = tail call float @llvm.fmuladd.f32(float %178, float 0.000000e+00, float %177) #13
>  %180 = tail call float @llvm.fmuladd.f32(float %sub42.7, float %179, float %170)
>  %sub52 = fadd float %weight.2172, -8.000000e+00
>  %add53 = add nsw i32 %j.2173, 8
>  %add30 = add nsw i32 %j.2173, 15
>  %cmp32 = icmp slt i32 %add30, %add31
>  br i1 %cmp32, label %for.cond36.preheader, label %while.cond55.preheader
>
>while.body59:                                     ; preds = %while.cond55.preheader, %while.body59
>  %j.3181 = phi i32 [ %inc66, %while.body59 ], [ %j.2.lcssa, %while.cond55.preheader ]
>  %weight.3180 = phi float [ %sub65, %while.body59 ], [ %weight.2.lcssa, %while.cond55.preheader ]
>  %sum.5179 = phi float [ %190, %while.body59 ], [ %sum.3.lcssa, %while.cond55.preheader ]
>  %cmp.i = icmp slt i32 %j.3181, %imageWidth
>  %sub.i = xor i32 %j.3181, -1
>  %sub1.i = add i32 %mul.i142, %sub.i
>  %spec.select.i = select i1 %cmp.i, i32 %j.3181, i32 %sub1.i
>  %arrayidx61 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %spec.select.i
>  %181 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx61, align 16, !tbaa !14
>  %182 = extractelement <4 x float> %181, i32 0
>  %183 = extractelement <4 x float> %181, i32 1
>  %184 = fmul float %183, 0x3FE6E2EB20000000
>  %185 = tail call float @llvm.fmuladd.f32(float %182, float 0x3FCB367A00000000, float %184) #13
>  %186 = extractelement <4 x float> %181, i32 2
>  %187 = tail call float @llvm.fmuladd.f32(float %186, float 0x3FB27BB300000000, float %185) #13
>  %188 = extractelement <4 x float> %181, i32 3
>  %189 = tail call float @llvm.fmuladd.f32(float %188, float 0.000000e+00, float %187) #13
>  %190 = tail call float @llvm.fmuladd.f32(float %weight.3180, float %189, float %sum.5179)
>  %sub65 = fadd float %weight.3180, -1.000000e+00
>  %inc66 = add nsw i32 %j.3181, 1
>  %exitcond189 = icmp eq i32 %inc66, %indvars.iv
>  br i1 %exitcond189, label %while.end67, label %while.body59
>
>while.end67:                                      ; preds = %while.body59, %while.cond55.preheader
>  %sum.5.lcssa = phi float [ %sum.3.lcssa, %while.cond55.preheader ], [ %190, %while.body59 ]
>  %div = fdiv float %sum.5.lcssa, %conv71, !fpmath !11
>  %add73 = add nsw i32 %i.0184, %mul
>  %arrayidx74 = getelementptr inbounds float, float addrspace(1)* %tmpImage, i32 %add73
>  store float %div, float addrspace(1)* %arrayidx74, align 4, !tbaa !12
>  %add77 = add i32 %i.0184, %x.i
>  %cmp3 = icmp slt i32 %add77, %imageWidth
>  %indvars.iv.next = add i32 %indvars.iv, %x.i
>  br i1 %cmp3, label %for.body, label %cleanup
>
>cleanup:                                          ; preds = %while.end67, %entry
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @LocalContrastBlurApplyColumn(<4 x float> addrspace(1)* nocapture readonly %srcImage, <4 x float> addrspace(1)* nocapture %dstImage, float addrspace(1)* nocapture readonly %blurImage, i32 %radius, float %strength, i32 %imageWidth, i32 %imageHeight) local_unnamed_addr #4 !kernel_arg_addr_space !65 !kernel_arg_access_qual !48 !kernel_arg_type !66 !kernel_arg_base_type !67 !kernel_arg_type_qual !68 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %cmp = icmp slt i32 %8, %imageWidth
>  %cmp2 = icmp slt i32 %16, %imageHeight
>  %or.cond = and i1 %cmp, %cmp2
>  br i1 %or.cond, label %if.end, label %cleanup
>
>if.end:                                           ; preds = %entry
>  %add.ptr = getelementptr inbounds float, float addrspace(1)* %blurImage, i32 %8
>  %sub = sub nsw i32 %16, %radius
>  %add199 = add nsw i32 %sub, 7
>  %cmp3200 = icmp slt i32 %add199, %16
>  br i1 %cmp3200, label %for.cond.preheader, label %while.cond11.preheader
>
>for.cond.preheader:                               ; preds = %if.end, %for.cond.preheader
>  %sum.0203 = phi float [ %32, %for.cond.preheader ], [ 0.000000e+00, %if.end ]
>  %weight.0202 = phi float [ %add9, %for.cond.preheader ], [ 1.000000e+00, %if.end ]
>  %j.0201 = phi i32 [ %add10, %for.cond.preheader ], [ %sub, %if.end ]
>  %add5 = fadd float %weight.0202, 0.000000e+00
>  %cmp.i172 = icmp slt i32 %j.0201, 0
>  %sub.i173 = sub nsw i32 0, %j.0201
>  %cond.i174 = select i1 %cmp.i172, i32 %sub.i173, i32 %j.0201
>  %mul = mul nsw i32 %cond.i174, %imageWidth
>  %arrayidx = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul
>  %17 = load float, float addrspace(1)* %arrayidx, align 4, !tbaa !12
>  %18 = tail call float @llvm.fmuladd.f32(float %add5, float %17, float %sum.0203)
>  %add5.1 = fadd float %weight.0202, 1.000000e+00
>  %add6.1 = add nsw i32 %j.0201, 1
>  %cmp.i172.1 = icmp slt i32 %j.0201, -1
>  %sub.i173.1 = xor i32 %j.0201, -1
>  %cond.i174.1 = select i1 %cmp.i172.1, i32 %sub.i173.1, i32 %add6.1
>  %mul.1 = mul nsw i32 %cond.i174.1, %imageWidth
>  %arrayidx.1 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul.1
>  %19 = load float, float addrspace(1)* %arrayidx.1, align 4, !tbaa !12
>  %20 = tail call float @llvm.fmuladd.f32(float %add5.1, float %19, float %18)
>  %add5.2 = fadd float %weight.0202, 2.000000e+00
>  %add6.2 = add nsw i32 %j.0201, 2
>  %cmp.i172.2 = icmp slt i32 %j.0201, -2
>  %sub.i173.2 = sub i32 -2, %j.0201
>  %cond.i174.2 = select i1 %cmp.i172.2, i32 %sub.i173.2, i32 %add6.2
>  %mul.2 = mul nsw i32 %cond.i174.2, %imageWidth
>  %arrayidx.2 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul.2
>  %21 = load float, float addrspace(1)* %arrayidx.2, align 4, !tbaa !12
>  %22 = tail call float @llvm.fmuladd.f32(float %add5.2, float %21, float %20)
>  %add5.3 = fadd float %weight.0202, 3.000000e+00
>  %add6.3 = add nsw i32 %j.0201, 3
>  %cmp.i172.3 = icmp slt i32 %j.0201, -3
>  %sub.i173.3 = sub i32 -3, %j.0201
>  %cond.i174.3 = select i1 %cmp.i172.3, i32 %sub.i173.3, i32 %add6.3
>  %mul.3 = mul nsw i32 %cond.i174.3, %imageWidth
>  %arrayidx.3 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul.3
>  %23 = load float, float addrspace(1)* %arrayidx.3, align 4, !tbaa !12
>  %24 = tail call float @llvm.fmuladd.f32(float %add5.3, float %23, float %22)
>  %add5.4 = fadd float %weight.0202, 4.000000e+00
>  %add6.4 = add nsw i32 %j.0201, 4
>  %cmp.i172.4 = icmp slt i32 %j.0201, -4
>  %sub.i173.4 = sub i32 -4, %j.0201
>  %cond.i174.4 = select i1 %cmp.i172.4, i32 %sub.i173.4, i32 %add6.4
>  %mul.4 = mul nsw i32 %cond.i174.4, %imageWidth
>  %arrayidx.4 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul.4
>  %25 = load float, float addrspace(1)* %arrayidx.4, align 4, !tbaa !12
>  %26 = tail call float @llvm.fmuladd.f32(float %add5.4, float %25, float %24)
>  %add5.5 = fadd float %weight.0202, 5.000000e+00
>  %add6.5 = add nsw i32 %j.0201, 5
>  %cmp.i172.5 = icmp slt i32 %j.0201, -5
>  %sub.i173.5 = sub i32 -5, %j.0201
>  %cond.i174.5 = select i1 %cmp.i172.5, i32 %sub.i173.5, i32 %add6.5
>  %mul.5 = mul nsw i32 %cond.i174.5, %imageWidth
>  %arrayidx.5 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul.5
>  %27 = load float, float addrspace(1)* %arrayidx.5, align 4, !tbaa !12
>  %28 = tail call float @llvm.fmuladd.f32(float %add5.5, float %27, float %26)
>  %add5.6 = fadd float %weight.0202, 6.000000e+00
>  %add6.6 = add nsw i32 %j.0201, 6
>  %cmp.i172.6 = icmp slt i32 %j.0201, -6
>  %sub.i173.6 = sub i32 -6, %j.0201
>  %cond.i174.6 = select i1 %cmp.i172.6, i32 %sub.i173.6, i32 %add6.6
>  %mul.6 = mul nsw i32 %cond.i174.6, %imageWidth
>  %arrayidx.6 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul.6
>  %29 = load float, float addrspace(1)* %arrayidx.6, align 4, !tbaa !12
>  %30 = tail call float @llvm.fmuladd.f32(float %add5.6, float %29, float %28)
>  %add5.7 = fadd float %weight.0202, 7.000000e+00
>  %add6.7 = add nsw i32 %j.0201, 7
>  %cmp.i172.7 = icmp slt i32 %j.0201, -7
>  %sub.i173.7 = sub i32 -7, %j.0201
>  %cond.i174.7 = select i1 %cmp.i172.7, i32 %sub.i173.7, i32 %add6.7
>  %mul.7 = mul nsw i32 %cond.i174.7, %imageWidth
>  %arrayidx.7 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul.7
>  %31 = load float, float addrspace(1)* %arrayidx.7, align 4, !tbaa !12
>  %32 = tail call float @llvm.fmuladd.f32(float %add5.7, float %31, float %30)
>  %add9 = fadd float %weight.0202, 8.000000e+00
>  %add10 = add nsw i32 %j.0201, 8
>  %add = add nsw i32 %j.0201, 15
>  %cmp3 = icmp slt i32 %add, %16
>  br i1 %cmp3, label %for.cond.preheader, label %while.cond11.preheader
>
>while.cond11.preheader:                           ; preds = %for.cond.preheader, %if.end
>  %j.0.lcssa = phi i32 [ %sub, %if.end ], [ %add10, %for.cond.preheader ]
>  %weight.0.lcssa = phi float [ 1.000000e+00, %if.end ], [ %add9, %for.cond.preheader ]
>  %sum.0.lcssa = phi float [ 0.000000e+00, %if.end ], [ %32, %for.cond.preheader ]
>  %cmp12190 = icmp slt i32 %j.0.lcssa, %16
>  br i1 %cmp12190, label %while.body14, label %while.cond22.preheader
>
>while.cond22.preheader:                           ; preds = %while.body14, %while.cond11.preheader
>  %j.1.lcssa = phi i32 [ %j.0.lcssa, %while.cond11.preheader ], [ %16, %while.body14 ]
>  %weight.1.lcssa = phi float [ %weight.0.lcssa, %while.cond11.preheader ], [ %add19, %while.body14 ]
>  %sum.2.lcssa = phi float [ %sum.0.lcssa, %while.cond11.preheader ], [ %34, %while.body14 ]
>  %add23181 = add nsw i32 %j.1.lcssa, 7
>  %add24 = add nsw i32 %16, %radius
>  %cmp25182 = icmp slt i32 %add23181, %add24
>  br i1 %cmp25182, label %for.cond29.preheader.lr.ph, label %while.cond47.preheader
>
>for.cond29.preheader.lr.ph:                       ; preds = %while.cond22.preheader
>  %mul.i166 = shl i32 %imageHeight, 1
>  br label %for.cond29.preheader
>
>while.body14:                                     ; preds = %while.cond11.preheader, %while.body14
>  %sum.2193 = phi float [ %34, %while.body14 ], [ %sum.0.lcssa, %while.cond11.preheader ]
>  %weight.1192 = phi float [ %add19, %while.body14 ], [ %weight.0.lcssa, %while.cond11.preheader ]
>  %j.1191 = phi i32 [ %inc20, %while.body14 ], [ %j.0.lcssa, %while.cond11.preheader ]
>  %cmp.i170 = icmp slt i32 %j.1191, 0
>  %sub.i171 = sub nsw i32 0, %j.1191
>  %cond.i = select i1 %cmp.i170, i32 %sub.i171, i32 %j.1191
>  %mul16 = mul nsw i32 %cond.i, %imageWidth
>  %arrayidx17 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul16
>  %33 = load float, float addrspace(1)* %arrayidx17, align 4, !tbaa !12
>  %34 = tail call float @llvm.fmuladd.f32(float %weight.1192, float %33, float %sum.2193)
>  %add19 = fadd float %weight.1192, 1.000000e+00
>  %inc20 = add nsw i32 %j.1191, 1
>  %exitcond211 = icmp eq i32 %inc20, %16
>  br i1 %exitcond211, label %while.cond22.preheader, label %while.body14
>
>for.cond29.preheader:                             ; preds = %for.cond29.preheader.lr.ph, %for.cond29.preheader
>  %sum.3185 = phi float [ %sum.2.lcssa, %for.cond29.preheader.lr.ph ], [ %50, %for.cond29.preheader ]
>  %weight.2184 = phi float [ %weight.1.lcssa, %for.cond29.preheader.lr.ph ], [ %sub44, %for.cond29.preheader ]
>  %j.2183 = phi i32 [ %j.1.lcssa, %for.cond29.preheader.lr.ph ], [ %add45, %for.cond29.preheader ]
>  %cmp.i165 = icmp slt i32 %j.2183, %imageHeight
>  %sub.i167 = xor i32 %j.2183, -1
>  %sub1.i168 = add i32 %mul.i166, %sub.i167
>  %spec.select.i169 = select i1 %cmp.i165, i32 %j.2183, i32 %sub1.i168
>  %mul38 = mul nsw i32 %spec.select.i169, %imageWidth
>  %arrayidx39 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul38
>  %35 = load float, float addrspace(1)* %arrayidx39, align 4, !tbaa !12
>  %36 = tail call float @llvm.fmuladd.f32(float %weight.2184, float %35, float %sum.3185)
>  %sub35.1 = fadd float %weight.2184, -1.000000e+00
>  %add36.1 = add nsw i32 %j.2183, 1
>  %cmp.i165.1 = icmp slt i32 %add36.1, %imageHeight
>  %sub.i167.1 = sub i32 -2, %j.2183
>  %sub1.i168.1 = add i32 %mul.i166, %sub.i167.1
>  %spec.select.i169.1 = select i1 %cmp.i165.1, i32 %add36.1, i32 %sub1.i168.1
>  %mul38.1 = mul nsw i32 %spec.select.i169.1, %imageWidth
>  %arrayidx39.1 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul38.1
>  %37 = load float, float addrspace(1)* %arrayidx39.1, align 4, !tbaa !12
>  %38 = tail call float @llvm.fmuladd.f32(float %sub35.1, float %37, float %36)
>  %sub35.2 = fadd float %weight.2184, -2.000000e+00
>  %add36.2 = add nsw i32 %j.2183, 2
>  %cmp.i165.2 = icmp slt i32 %add36.2, %imageHeight
>  %sub.i167.2 = sub i32 -3, %j.2183
>  %sub1.i168.2 = add i32 %mul.i166, %sub.i167.2
>  %spec.select.i169.2 = select i1 %cmp.i165.2, i32 %add36.2, i32 %sub1.i168.2
>  %mul38.2 = mul nsw i32 %spec.select.i169.2, %imageWidth
>  %arrayidx39.2 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul38.2
>  %39 = load float, float addrspace(1)* %arrayidx39.2, align 4, !tbaa !12
>  %40 = tail call float @llvm.fmuladd.f32(float %sub35.2, float %39, float %38)
>  %sub35.3 = fadd float %weight.2184, -3.000000e+00
>  %add36.3 = add nsw i32 %j.2183, 3
>  %cmp.i165.3 = icmp slt i32 %add36.3, %imageHeight
>  %sub.i167.3 = sub i32 -4, %j.2183
>  %sub1.i168.3 = add i32 %mul.i166, %sub.i167.3
>  %spec.select.i169.3 = select i1 %cmp.i165.3, i32 %add36.3, i32 %sub1.i168.3
>  %mul38.3 = mul nsw i32 %spec.select.i169.3, %imageWidth
>  %arrayidx39.3 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul38.3
>  %41 = load float, float addrspace(1)* %arrayidx39.3, align 4, !tbaa !12
>  %42 = tail call float @llvm.fmuladd.f32(float %sub35.3, float %41, float %40)
>  %sub35.4 = fadd float %weight.2184, -4.000000e+00
>  %add36.4 = add nsw i32 %j.2183, 4
>  %cmp.i165.4 = icmp slt i32 %add36.4, %imageHeight
>  %sub.i167.4 = sub i32 -5, %j.2183
>  %sub1.i168.4 = add i32 %mul.i166, %sub.i167.4
>  %spec.select.i169.4 = select i1 %cmp.i165.4, i32 %add36.4, i32 %sub1.i168.4
>  %mul38.4 = mul nsw i32 %spec.select.i169.4, %imageWidth
>  %arrayidx39.4 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul38.4
>  %43 = load float, float addrspace(1)* %arrayidx39.4, align 4, !tbaa !12
>  %44 = tail call float @llvm.fmuladd.f32(float %sub35.4, float %43, float %42)
>  %sub35.5 = fadd float %weight.2184, -5.000000e+00
>  %add36.5 = add nsw i32 %j.2183, 5
>  %cmp.i165.5 = icmp slt i32 %add36.5, %imageHeight
>  %sub.i167.5 = sub i32 -6, %j.2183
>  %sub1.i168.5 = add i32 %mul.i166, %sub.i167.5
>  %spec.select.i169.5 = select i1 %cmp.i165.5, i32 %add36.5, i32 %sub1.i168.5
>  %mul38.5 = mul nsw i32 %spec.select.i169.5, %imageWidth
>  %arrayidx39.5 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul38.5
>  %45 = load float, float addrspace(1)* %arrayidx39.5, align 4, !tbaa !12
>  %46 = tail call float @llvm.fmuladd.f32(float %sub35.5, float %45, float %44)
>  %sub35.6 = fadd float %weight.2184, -6.000000e+00
>  %add36.6 = add nsw i32 %j.2183, 6
>  %cmp.i165.6 = icmp slt i32 %add36.6, %imageHeight
>  %sub.i167.6 = sub i32 -7, %j.2183
>  %sub1.i168.6 = add i32 %mul.i166, %sub.i167.6
>  %spec.select.i169.6 = select i1 %cmp.i165.6, i32 %add36.6, i32 %sub1.i168.6
>  %mul38.6 = mul nsw i32 %spec.select.i169.6, %imageWidth
>  %arrayidx39.6 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul38.6
>  %47 = load float, float addrspace(1)* %arrayidx39.6, align 4, !tbaa !12
>  %48 = tail call float @llvm.fmuladd.f32(float %sub35.6, float %47, float %46)
>  %sub35.7 = fadd float %weight.2184, -7.000000e+00
>  %add36.7 = add nsw i32 %j.2183, 7
>  %cmp.i165.7 = icmp slt i32 %add36.7, %imageHeight
>  %sub.i167.7 = sub i32 -8, %j.2183
>  %sub1.i168.7 = add i32 %mul.i166, %sub.i167.7
>  %spec.select.i169.7 = select i1 %cmp.i165.7, i32 %add36.7, i32 %sub1.i168.7
>  %mul38.7 = mul nsw i32 %spec.select.i169.7, %imageWidth
>  %arrayidx39.7 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul38.7
>  %49 = load float, float addrspace(1)* %arrayidx39.7, align 4, !tbaa !12
>  %50 = tail call float @llvm.fmuladd.f32(float %sub35.7, float %49, float %48)
>  %sub44 = fadd float %weight.2184, -8.000000e+00
>  %add45 = add nsw i32 %j.2183, 8
>  %add23 = add nsw i32 %j.2183, 15
>  %cmp25 = icmp slt i32 %add23, %add24
>  br i1 %cmp25, label %for.cond29.preheader, label %while.cond47.preheader
>
>while.cond47.preheader:                           ; preds = %for.cond29.preheader, %while.cond22.preheader
>  %j.2.lcssa = phi i32 [ %j.1.lcssa, %while.cond22.preheader ], [ %add45, %for.cond29.preheader ]
>  %weight.2.lcssa = phi float [ %weight.1.lcssa, %while.cond22.preheader ], [ %sub44, %for.cond29.preheader ]
>  %sum.3.lcssa = phi float [ %sum.2.lcssa, %while.cond22.preheader ], [ %50, %for.cond29.preheader ]
>  %cmp49175 = icmp slt i32 %j.2.lcssa, %add24
>  br i1 %cmp49175, label %while.body51.lr.ph, label %while.end58
>
>while.body51.lr.ph:                               ; preds = %while.cond47.preheader
>  %mul.i = shl i32 %imageHeight, 1
>  %51 = add i32 %11, %15
>  %52 = add i32 %51, %radius
>  %53 = add i32 %52, %10
>  br label %while.body51
>
>while.body51:                                     ; preds = %while.body51, %while.body51.lr.ph
>  %sum.5178 = phi float [ %sum.3.lcssa, %while.body51.lr.ph ], [ %55, %while.body51 ]
>  %weight.3177 = phi float [ %weight.2.lcssa, %while.body51.lr.ph ], [ %sub56, %while.body51 ]
>  %j.3176 = phi i32 [ %j.2.lcssa, %while.body51.lr.ph ], [ %inc57, %while.body51 ]
>  %cmp.i = icmp slt i32 %j.3176, %imageHeight
>  %sub.i = xor i32 %j.3176, -1
>  %sub1.i = add i32 %mul.i, %sub.i
>  %spec.select.i = select i1 %cmp.i, i32 %j.3176, i32 %sub1.i
>  %mul53 = mul nsw i32 %spec.select.i, %imageWidth
>  %arrayidx54 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %mul53
>  %54 = load float, float addrspace(1)* %arrayidx54, align 4, !tbaa !12
>  %55 = tail call float @llvm.fmuladd.f32(float %weight.3177, float %54, float %sum.5178)
>  %sub56 = fadd float %weight.3177, -1.000000e+00
>  %inc57 = add nsw i32 %j.3176, 1
>  %exitcond = icmp eq i32 %inc57, %53
>  br i1 %exitcond, label %while.end58, label %while.body51
>
>while.end58:                                      ; preds = %while.body51, %while.cond47.preheader
>  %sum.5.lcssa = phi float [ %sum.3.lcssa, %while.cond47.preheader ], [ %55, %while.body51 ]
>  %mul59 = mul nsw i32 %16, %imageWidth
>  %add60 = add nsw i32 %mul59, %8
>  %arrayidx61 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %srcImage, i32 %add60
>  %56 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx61, align 16, !tbaa !14
>  %57 = extractelement <4 x float> %56, i32 0
>  %58 = extractelement <4 x float> %56, i32 1
>  %59 = fmul float %58, 0x3FE6E2EB20000000
>  %60 = tail call float @llvm.fmuladd.f32(float %57, float 0x3FCB367A00000000, float %59) #13
>  %61 = extractelement <4 x float> %56, i32 2
>  %62 = tail call float @llvm.fmuladd.f32(float %61, float 0x3FB27BB300000000, float %60) #13
>  %63 = extractelement <4 x float> %56, i32 3
>  %64 = tail call float @llvm.fmuladd.f32(float %63, float 0.000000e+00, float %62) #13
>  %add64 = add nsw i32 %radius, 1
>  %mul66 = mul nsw i32 %add64, %add64
>  %conv67 = sitofp i32 %mul66 to float
>  %div = fdiv float %sum.5.lcssa, %conv67, !fpmath !11
>  %sub68 = fsub float %64, %div
>  %div69 = fdiv float %strength, 1.000000e+02, !fpmath !11
>  %mul70 = fmul float %div69, %sub68
>  %add71 = fadd float %64, %mul70
>  %div72 = fdiv float %add71, %64, !fpmath !11
>  %mul73 = fmul float %57, %div72
>  %65 = fcmp ogt float %mul73, 6.553500e+04
>  %66 = fcmp olt float %mul73, 0.000000e+00
>  %67 = select i1 %66, float 0.000000e+00, float %mul73
>  %.op.i163 = fadd float %67, 5.000000e-01
>  %add.i164 = select i1 %65, float 6.553550e+04, float %.op.i163
>  %68 = insertelement <4 x float> %56, float %add.i164, i32 0
>  %mul75 = fmul float %58, %div72
>  %69 = fcmp ogt float %mul75, 6.553500e+04
>  %70 = fcmp olt float %mul75, 0.000000e+00
>  %71 = select i1 %70, float 0.000000e+00, float %mul75
>  %.op.i161 = fadd float %71, 5.000000e-01
>  %add.i162 = select i1 %69, float 6.553550e+04, float %.op.i161
>  %72 = insertelement <4 x float> %68, float %add.i162, i32 1
>  %mul77 = fmul float %61, %div72
>  %73 = fcmp ogt float %mul77, 6.553500e+04
>  %74 = fcmp olt float %mul77, 0.000000e+00
>  %75 = select i1 %74, float 0.000000e+00, float %mul77
>  %.op.i = fadd float %75, 5.000000e-01
>  %add.i = select i1 %73, float 6.553550e+04, float %.op.i
>  %76 = insertelement <4 x float> %72, float %add.i, i32 2
>  %arrayidx81 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %dstImage, i32 %add60
>  store <4 x float> %76, <4 x float> addrspace(1)* %arrayidx81, align 16, !tbaa !14
>  br label %cleanup
>
>cleanup:                                          ; preds = %entry, %while.end58
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @Modulate(<4 x float> addrspace(1)* nocapture %im, float %percent_brightness, float %percent_hue, float %percent_saturation, i32 %colorspace) local_unnamed_addr #4 !kernel_arg_addr_space !69 !kernel_arg_access_qual !31 !kernel_arg_type !70 !kernel_arg_base_type !71 !kernel_arg_type_qual !72 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %x.i = tail call i32 @llvm.r600.read.global.size.x() #16
>  %mul = mul nsw i32 %16, %x.i
>  %add = add nsw i32 %8, %mul
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %im, i32 %add
>  %17 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %18 = extractelement <4 x float> %17, i32 2
>  %19 = extractelement <4 x float> %17, i32 1
>  %20 = extractelement <4 x float> %17, i32 0
>  %mul.i.i = fmul float %18, 0x3EEF751040000000
>  %mul1.i.i = fmul float %19, 0x3EEF751040000000
>  %mul2.i.i = fmul float %20, 0x3EEF751040000000
>  %cmp.i.i = fcmp ogt float %mul1.i.i, %mul2.i.i
>  %cond.i.i = select i1 %cmp.i.i, float %mul1.i.i, float %mul2.i.i
>  %cmp5.i.i = fcmp ogt float %mul.i.i, %cond.i.i
>  %mul.cond.i.i = select i1 %cmp5.i.i, float %mul.i.i, float %cond.i.i
>  %cmp23.i.i = fcmp olt float %mul1.i.i, %mul2.i.i
>  %cond29.i.i = select i1 %cmp23.i.i, float %mul1.i.i, float %mul2.i.i
>  %cmp30.i.i = fcmp olt float %mul.i.i, %cond29.i.i
>  %cond44.i.i = select i1 %cmp30.i.i, float %mul.i.i, float %cond29.i.i
>  %sub.i.i = fsub float %mul.cond.i.i, %cond44.i.i
>  %add.i.i = fadd float %mul.cond.i.i, %cond44.i.i
>  %div.i.i = fmul float %add.i.i, 5.000000e-01
>  %cmp45.i.i = fcmp ugt float %sub.i.i, 0.000000e+00
>  br i1 %cmp45.i.i, label %if.end.i.i, label %ConvertRGBToHSL.exit.i
>
>if.end.i.i:                                       ; preds = %entry
>  %cmp47.i.i = fcmp oeq float %mul.cond.i.i, %mul.i.i
>  br i1 %cmp47.i.i, label %if.then48.i.i, label %if.else.i.i
>
>if.then48.i.i:                                    ; preds = %if.end.i.i
>  %neg.i.i = fsub float -0.000000e+00, %mul2.i.i
>  %21 = tail call float @llvm.fmuladd.f32(float %19, float 0x3EEF751040000000, float %neg.i.i) #13
>  %div51.i.i = fdiv float %21, %sub.i.i, !fpmath !11
>  br i1 %cmp23.i.i, label %if.then55.i.i, label %if.end73.i.i
>
>if.then55.i.i:                                    ; preds = %if.then48.i.i
>  %add56.i.i = fadd float %div51.i.i, 6.000000e+00
>  br label %if.end73.i.i
>
>if.else.i.i:                                      ; preds = %if.end.i.i
>  %cmp59.i.i = fcmp oeq float %mul.cond.i.i, %mul1.i.i
>  br i1 %cmp59.i.i, label %if.then60.i.i, label %if.else66.i.i
>
>if.then60.i.i:                                    ; preds = %if.else.i.i
>  %neg63.i.i = fsub float -0.000000e+00, %mul.i.i
>  %22 = tail call float @llvm.fmuladd.f32(float %20, float 0x3EEF751040000000, float %neg63.i.i) #13
>  %div64.i.i = fdiv float %22, %sub.i.i, !fpmath !11
>  %add65.i.i = fadd float %div64.i.i, 2.000000e+00
>  br label %if.end73.i.i
>
>if.else66.i.i:                                    ; preds = %if.else.i.i
>  %neg69.i.i = fsub float -0.000000e+00, %mul1.i.i
>  %23 = tail call float @llvm.fmuladd.f32(float %18, float 0x3EEF751040000000, float %neg69.i.i) #13
>  %div70.i.i = fdiv float %23, %sub.i.i, !fpmath !11
>  %add71.i.i = fadd float %div70.i.i, 4.000000e+00
>  br label %if.end73.i.i
>
>if.end73.i.i:                                     ; preds = %if.else66.i.i, %if.then60.i.i, %if.then55.i.i, %if.then48.i.i
>  %24 = phi float [ %add65.i.i, %if.then60.i.i ], [ %add71.i.i, %if.else66.i.i ], [ %div51.i.i, %if.then48.i.i ], [ %add56.i.i, %if.then55.i.i ]
>  %mul74.i.i = fmul float %24, 0x3FC5555560000000
>  %cmp75.i.i = fcmp ugt float %div.i.i, 5.000000e-01
>  %25 = tail call float @llvm.fmuladd.f32(float %div.i.i, float -2.000000e+00, float 2.000000e+00) #13
>  %mul77.i.i = fmul float %div.i.i, 2.000000e+00
>  %.pn.i.i = select i1 %cmp75.i.i, float %25, float %mul77.i.i
>  %storemerge.i.i = fdiv float %sub.i.i, %.pn.i.i
>  br label %ConvertRGBToHSL.exit.i
>
>ConvertRGBToHSL.exit.i:                           ; preds = %if.end73.i.i, %entry
>  %hue.0.i = phi float [ %mul74.i.i, %if.end73.i.i ], [ 0.000000e+00, %entry ]
>  %storemerge132.i.i = phi float [ %storemerge.i.i, %if.end73.i.i ], [ 0.000000e+00, %entry ]
>  %26 = tail call float @llvm.fmuladd.f32(float %percent_hue, float 0x3F847AE140000000, float -1.000000e+00) #13
>  %27 = tail call float @llvm.fmuladd.f32(float %26, float 5.000000e-01, float %hue.0.i) #13
>  %cmp25.i = fcmp olt float %27, 0.000000e+00
>  br i1 %cmp25.i, label %while.body.i, label %while.cond1.preheader.i
>
>while.cond1.preheader.i:                          ; preds = %while.body.i, %ConvertRGBToHSL.exit.i
>  %storemerge.lcssa.i = phi float [ %27, %ConvertRGBToHSL.exit.i ], [ %add.i, %while.body.i ]
>  %cmp223.i = fcmp ult float %storemerge.lcssa.i, 1.000000e+00
>  br i1 %cmp223.i, label %while.end4.i, label %while.body3.i
>
>while.body.i:                                     ; preds = %ConvertRGBToHSL.exit.i, %while.body.i
>  %storemerge26.i = phi float [ %add.i, %while.body.i ], [ %27, %ConvertRGBToHSL.exit.i ]
>  %add.i = fadd float %storemerge26.i, 1.000000e+00
>  %cmp.i = fcmp olt float %add.i, 0.000000e+00
>  br i1 %cmp.i, label %while.body.i, label %while.cond1.preheader.i
>
>while.body3.i:                                    ; preds = %while.cond1.preheader.i, %while.body3.i
>  %hue.124.i = phi float [ %sub.i, %while.body3.i ], [ %storemerge.lcssa.i, %while.cond1.preheader.i ]
>  %sub.i = fadd float %hue.124.i, -1.000000e+00
>  %cmp2.i = fcmp ult float %sub.i, 1.000000e+00
>  br i1 %cmp2.i, label %while.end4.i, label %while.body3.i
>
>while.end4.i:                                     ; preds = %while.body3.i, %while.cond1.preheader.i
>  %hue.1.lcssa.i = phi float [ %storemerge.lcssa.i, %while.cond1.preheader.i ], [ %sub.i, %while.body3.i ]
>  %mul.i = fmul float %percent_saturation, 0x3F847AE140000000
>  %mul5.i = fmul float %mul.i, %storemerge132.i.i
>  %mul6.i = fmul float %percent_brightness, 0x3F847AE140000000
>  %mul7.i = fmul float %mul6.i, %div.i.i
>  %mul.i11.i = fmul float %hue.1.lcssa.i, 3.600000e+02
>  %cmp.i12.i = fcmp ugt float %mul7.i, 5.000000e-01
>  %mul1.i13.i = fmul float %mul7.i, 2.000000e+00
>  %28 = tail call float @llvm.fmuladd.f32(float %mul7.i, float -2.000000e+00, float 2.000000e+00) #13
>  %mul1.pn.i.i = select i1 %cmp.i12.i, float %28, float %mul1.i13.i
>  %c.0.i.i = fmul float %mul1.pn.i.i, %mul5.i
>  %29 = tail call float @llvm.fmuladd.f32(float %c.0.i.i, float -5.000000e-01, float %mul7.i) #13
>  %div.i14.i = fdiv float %mul.i11.i, 3.600000e+02, !fpmath !11
>  %call.i.i = tail call float @llvm.floor.f32(float %div.i14.i) #17
>  %30 = tail call float @llvm.fmuladd.f32(float %call.i.i, float -3.600000e+02, float %mul.i11.i) #13
>  %div7.i.i = fdiv float %30, 6.000000e+01, !fpmath !11
>  %div8.i.i = fmul float %div7.i.i, 5.000000e-01
>  %call9.i.i = tail call float @llvm.floor.f32(float %div8.i.i) #17
>  %31 = tail call float @llvm.fmuladd.f32(float %call9.i.i, float -2.000000e+00, float %div7.i.i) #13
>  %sub.i15.i = fadd float %31, -1.000000e+00
>  %call11.i.i = tail call float @llvm.fabs.f32(float %sub.i15.i) #17
>  %sub12.i.i = fsub float 1.000000e+00, %call11.i.i
>  %mul13.i.i = fmul float %c.0.i.i, %sub12.i.i
>  %call14.i.i = tail call float @llvm.floor.f32(float %div7.i.i) #17
>  %conv.i.i = fptosi float %call14.i.i to i32
>  %rem.i.i = srem i32 %conv.i.i, 6
>  switch i32 %rem.i.i, label %ModulateHSL.exit [
>    i32 0, label %sw.bb.i.i
>    i32 1, label %sw.bb16.i.i
>    i32 2, label %sw.bb19.i.i
>    i32 3, label %sw.bb22.i.i
>    i32 4, label %sw.bb25.i.i
>    i32 5, label %sw.bb28.i.i
>  ]
>
>sw.bb.i.i:                                        ; preds = %while.end4.i
>  %add.i16.i = fadd float %c.0.i.i, %29
>  %add15.i.i = fadd float %29, %mul13.i.i
>  br label %ModulateHSL.exit
>
>sw.bb16.i.i:                                      ; preds = %while.end4.i
>  %add17.i.i = fadd float %29, %mul13.i.i
>  %add18.i.i = fadd float %c.0.i.i, %29
>  br label %ModulateHSL.exit
>
>sw.bb19.i.i:                                      ; preds = %while.end4.i
>  %add20.i.i = fadd float %c.0.i.i, %29
>  %add21.i.i = fadd float %29, %mul13.i.i
>  br label %ModulateHSL.exit
>
>sw.bb22.i.i:                                      ; preds = %while.end4.i
>  %add23.i.i = fadd float %29, %mul13.i.i
>  %add24.i.i = fadd float %c.0.i.i, %29
>  br label %ModulateHSL.exit
>
>sw.bb25.i.i:                                      ; preds = %while.end4.i
>  %add26.i.i = fadd float %29, %mul13.i.i
>  %add27.i.i = fadd float %c.0.i.i, %29
>  br label %ModulateHSL.exit
>
>sw.bb28.i.i:                                      ; preds = %while.end4.i
>  %add29.i.i = fadd float %c.0.i.i, %29
>  %add30.i.i = fadd float %29, %mul13.i.i
>  br label %ModulateHSL.exit
>
>ModulateHSL.exit:                                 ; preds = %while.end4.i, %sw.bb.i.i, %sw.bb16.i.i, %sw.bb19.i.i, %sw.bb22.i.i, %sw.bb25.i.i, %sw.bb28.i.i
>  %g.0.i.i = phi float [ %29, %sw.bb28.i.i ], [ %29, %sw.bb25.i.i ], [ %add23.i.i, %sw.bb22.i.i ], [ %add20.i.i, %sw.bb19.i.i ], [ %add18.i.i, %sw.bb16.i.i ], [ %add15.i.i, %sw.bb.i.i ], [ 0.000000e+00, %while.end4.i ]
>  %r.0.i.i = phi float [ %add29.i.i, %sw.bb28.i.i ], [ %add26.i.i, %sw.bb25.i.i ], [ %29, %sw.bb22.i.i ], [ %29, %sw.bb19.i.i ], [ %add17.i.i, %sw.bb16.i.i ], [ %add.i16.i, %sw.bb.i.i ], [ 0.000000e+00, %while.end4.i ]
>  %b.0.i.i = phi float [ %add30.i.i, %sw.bb28.i.i ], [ %add27.i.i, %sw.bb25.i.i ], [ %add24.i.i, %sw.bb22.i.i ], [ %add21.i.i, %sw.bb19.i.i ], [ %29, %sw.bb16.i.i ], [ %29, %sw.bb.i.i ], [ 0.000000e+00, %while.end4.i ]
>  %mul31.i.i = fmul float %r.0.i.i, 6.553500e+04
>  %32 = fcmp ogt float %mul31.i.i, 6.553500e+04
>  %33 = fcmp olt float %mul31.i.i, 0.000000e+00
>  %34 = select i1 %33, float 0.000000e+00, float %mul31.i.i
>  %.op.i.i.i = fadd float %34, 5.000000e-01
>  %add.i.i.i = select i1 %32, float 6.553550e+04, float %.op.i.i.i
>  %mul33.i.i = fmul float %g.0.i.i, 6.553500e+04
>  %35 = fcmp ogt float %mul33.i.i, 6.553500e+04
>  %36 = fcmp olt float %mul33.i.i, 0.000000e+00
>  %37 = select i1 %36, float 0.000000e+00, float %mul33.i.i
>  %.op.i83.i.i = fadd float %37, 5.000000e-01
>  %add.i84.i.i = select i1 %35, float 6.553550e+04, float %.op.i83.i.i
>  %mul35.i.i = fmul float %b.0.i.i, 6.553500e+04
>  %38 = fcmp ogt float %mul35.i.i, 6.553500e+04
>  %39 = fcmp olt float %mul35.i.i, 0.000000e+00
>  %40 = select i1 %39, float 0.000000e+00, float %mul35.i.i
>  %.op.i81.i.i = fadd float %40, 5.000000e-01
>  %add.i82.i.i = select i1 %38, float 6.553550e+04, float %.op.i81.i.i
>  %41 = insertelement <4 x float> undef, float %add.i.i.i, i32 2
>  %42 = insertelement <4 x float> %41, float %add.i84.i.i, i32 1
>  %43 = insertelement <4 x float> %42, float %add.i82.i.i, i32 0
>  %44 = shufflevector <4 x float> %43, <4 x float> %17, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
>  store <4 x float> %44, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @MotionBlur(<4 x float> addrspace(1)* nocapture readonly %input, <4 x float> addrspace(1)* nocapture %output, i32 %imageWidth, i32 %imageHeight, float addrspace(1)* nocapture readonly %filter, i32 %width, <2 x i32> addrspace(1)* nocapture readonly %offset, <4 x float> %bias, i32 %channel, i32 %matte) local_unnamed_addr #4 !kernel_arg_addr_space !73 !kernel_arg_access_qual !74 !kernel_arg_type !75 !kernel_arg_base_type !76 !kernel_arg_type_qual !77 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = insertelement <2 x i32> undef, i32 %8, i32 0
>  %10 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %11 = mul i32 %y.i.i, %10
>  %12 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %13 = add i32 %11, %12
>  %14 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %15 = bitcast i8 addrspace(7)* %14 to i32 addrspace(7)*
>  %16 = load i32, i32 addrspace(7)* %15, align 4, !tbaa !4
>  %17 = add i32 %13, %16
>  %18 = insertelement <2 x i32> %9, i32 %17, i32 1
>  %cmp = icmp ult i32 %8, %imageWidth
>  %cmp2 = icmp ult i32 %17, %imageHeight
>  %or.cond180 = and i1 %cmp, %cmp2
>  br i1 %or.cond180, label %if.end, label %cleanup
>
>if.end:                                           ; preds = %entry
>  %and = and i32 %channel, 16
>  %cmp3 = icmp eq i32 %and, 0
>  %cmp5 = icmp eq i32 %matte, 0
>  %or.cond = or i1 %cmp3, %cmp5
>  %cmp7198 = icmp ne i32 %width, 0
>  br i1 %or.cond, label %for.cond.preheader, label %for.cond28.preheader
>
>for.cond28.preheader:                             ; preds = %if.end
>  br i1 %cmp7198, label %for.body31.lr.ph, label %for.cond.cleanup30
>
>for.body31.lr.ph:                                 ; preds = %for.cond28.preheader
>  %sub.i181 = add nsw i32 %imageWidth, -1
>  %sub.i = add nsw i32 %imageHeight, -1
>  br label %for.body31
>
>for.cond.preheader:                               ; preds = %if.end
>  br i1 %cmp7198, label %for.body.lr.ph, label %for.cond.cleanup
>
>for.body.lr.ph:                                   ; preds = %for.cond.preheader
>  %sub.i189 = add nsw i32 %imageWidth, -1
>  %sub.i188 = add nsw i32 %imageHeight, -1
>  br label %for.body
>
>for.cond.cleanup:                                 ; preds = %for.body, %for.cond.preheader
>  %pixel.0.lcssa = phi <4 x float> [ %bias, %for.cond.preheader ], [ %58, %for.body ]
>  %19 = extractelement <4 x float> %pixel.0.lcssa, i32 0
>  %20 = fcmp ogt float %19, 6.553500e+04
>  %21 = fcmp olt float %19, 0.000000e+00
>  %22 = select i1 %21, float 0.000000e+00, float %19
>  %.op.i196 = fadd float %22, 5.000000e-01
>  %add.i197 = select i1 %20, float 6.553550e+04, float %.op.i196
>  %23 = insertelement <4 x float> undef, float %add.i197, i32 0
>  %24 = extractelement <4 x float> %pixel.0.lcssa, i32 1
>  %25 = fcmp ogt float %24, 6.553500e+04
>  %26 = fcmp olt float %24, 0.000000e+00
>  %27 = select i1 %26, float 0.000000e+00, float %24
>  %.op.i194 = fadd float %27, 5.000000e-01
>  %add.i195 = select i1 %25, float 6.553550e+04, float %.op.i194
>  %28 = insertelement <4 x float> %23, float %add.i195, i32 1
>  %29 = extractelement <4 x float> %pixel.0.lcssa, i32 2
>  br label %cleanup.sink.split
>
>for.body:                                         ; preds = %for.body, %for.body.lr.ph
>  %pixel.0200 = phi <4 x float> [ %bias, %for.body.lr.ph ], [ %58, %for.body ]
>  %i.0199 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
>  %arrayidx = getelementptr inbounds <2 x i32>, <2 x i32> addrspace(1)* %offset, i32 %i.0199
>  %30 = load <2 x i32>, <2 x i32> addrspace(1)* %arrayidx, align 8, !tbaa !14
>  %add = add <2 x i32> %30, %18
>  %31 = extractelement <2 x i32> %add, i32 0
>  %32 = icmp slt i32 %31, %imageWidth
>  %33 = icmp sgt i32 %31, 0
>  %34 = select i1 %33, i32 %31, i32 0
>  %35 = select i1 %32, i32 %34, i32 %sub.i189
>  %36 = extractelement <2 x i32> %add, i32 1
>  %37 = icmp slt i32 %36, %imageHeight
>  %38 = icmp sgt i32 %36, 0
>  %39 = select i1 %38, i32 %36, i32 0
>  %40 = select i1 %37, i32 %39, i32 %sub.i188
>  %mul = mul i32 %40, %imageWidth
>  %add10 = add i32 %mul, %35
>  %arrayidx11 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %input, i32 %add10
>  %41 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx11, align 16, !tbaa !14
>  %arrayidx12 = getelementptr inbounds float, float addrspace(1)* %filter, i32 %i.0199
>  %42 = load float, float addrspace(1)* %arrayidx12, align 4, !tbaa !12
>  %43 = extractelement <4 x float> %41, i32 0
>  %44 = extractelement <4 x float> %pixel.0200, i32 0
>  %45 = tail call float @llvm.fmuladd.f32(float %42, float %43, float %44)
>  %46 = insertelement <4 x float> undef, float %45, i32 0
>  %47 = extractelement <4 x float> %41, i32 1
>  %48 = extractelement <4 x float> %pixel.0200, i32 1
>  %49 = tail call float @llvm.fmuladd.f32(float %42, float %47, float %48)
>  %50 = insertelement <4 x float> %46, float %49, i32 1
>  %51 = extractelement <4 x float> %41, i32 2
>  %52 = extractelement <4 x float> %pixel.0200, i32 2
>  %53 = tail call float @llvm.fmuladd.f32(float %42, float %51, float %52)
>  %54 = insertelement <4 x float> %50, float %53, i32 2
>  %55 = extractelement <4 x float> %41, i32 3
>  %56 = extractelement <4 x float> %pixel.0200, i32 3
>  %57 = tail call float @llvm.fmuladd.f32(float %42, float %55, float %56)
>  %58 = insertelement <4 x float> %54, float %57, i32 3
>  %inc = add nuw nsw i32 %i.0199, 1
>  %exitcond = icmp eq i32 %inc, %width
>  br i1 %exitcond, label %for.cond.cleanup, label %for.body
>
>for.cond.cleanup30:                               ; preds = %for.body31, %for.cond28.preheader
>  %gamma.0.lcssa = phi float [ 0.000000e+00, %for.cond28.preheader ], [ %100, %for.body31 ]
>  %pixel.1.lcssa = phi <4 x float> [ %bias, %for.cond28.preheader ], [ %99, %for.body31 ]
>  %cmp.i = fcmp olt float %gamma.0.lcssa, 0.000000e+00
>  %cond.i = select i1 %cmp.i, float -1.000000e+00, float 1.000000e+00
>  %mul.i = fmul float %gamma.0.lcssa, %cond.i
>  %cmp1.i = fcmp ult float %mul.i, 0.000000e+00
>  %div.i = fdiv float 1.000000e+00, %gamma.0.lcssa
>  %mul2.i = fmul float %cond.i, 0x7FF0000000000000
>  %cond3.i = select i1 %cmp1.i, float %mul2.i, float %div.i
>  %splat.splatinsert = insertelement <3 x float> undef, float %cond3.i, i32 0
>  %splat.splat = shufflevector <3 x float> %splat.splatinsert, <3 x float> undef, <3 x i32> zeroinitializer
>  %59 = shufflevector <4 x float> %pixel.1.lcssa, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
>  %mul56 = fmul <3 x float> %59, %splat.splat
>  %60 = extractelement <3 x float> %mul56, i32 0
>  %61 = fcmp ogt float %60, 6.553500e+04
>  %62 = fcmp olt float %60, 0.000000e+00
>  %63 = select i1 %62, float 0.000000e+00, float %60
>  %.op.i186 = fadd float %63, 5.000000e-01
>  %add.i187 = select i1 %61, float 6.553550e+04, float %.op.i186
>  %64 = insertelement <4 x float> undef, float %add.i187, i32 0
>  %65 = extractelement <3 x float> %mul56, i32 1
>  %66 = fcmp ogt float %65, 6.553500e+04
>  %67 = fcmp olt float %65, 0.000000e+00
>  %68 = select i1 %67, float 0.000000e+00, float %65
>  %.op.i184 = fadd float %68, 5.000000e-01
>  %add.i185 = select i1 %66, float 6.553550e+04, float %.op.i184
>  %69 = insertelement <4 x float> %64, float %add.i185, i32 1
>  %70 = extractelement <3 x float> %mul56, i32 2
>  br label %cleanup.sink.split
>
>for.body31:                                       ; preds = %for.body31, %for.body31.lr.ph
>  %pixel.1204 = phi <4 x float> [ %bias, %for.body31.lr.ph ], [ %99, %for.body31 ]
>  %i27.0203 = phi i32 [ 0, %for.body31.lr.ph ], [ %inc53, %for.body31 ]
>  %gamma.0202 = phi float [ 0.000000e+00, %for.body31.lr.ph ], [ %100, %for.body31 ]
>  %arrayidx33 = getelementptr inbounds <2 x i32>, <2 x i32> addrspace(1)* %offset, i32 %i27.0203
>  %71 = load <2 x i32>, <2 x i32> addrspace(1)* %arrayidx33, align 8, !tbaa !14
>  %add34 = add <2 x i32> %71, %18
>  %72 = extractelement <2 x i32> %add34, i32 0
>  %73 = icmp slt i32 %72, %imageWidth
>  %74 = icmp sgt i32 %72, 0
>  %75 = select i1 %74, i32 %72, i32 0
>  %76 = select i1 %73, i32 %75, i32 %sub.i181
>  %77 = extractelement <2 x i32> %add34, i32 1
>  %78 = icmp slt i32 %77, %imageHeight
>  %79 = icmp sgt i32 %77, 0
>  %80 = select i1 %79, i32 %77, i32 0
>  %81 = select i1 %78, i32 %80, i32 %sub.i
>  %mul38 = mul i32 %81, %imageWidth
>  %add39 = add i32 %mul38, %76
>  %arrayidx40 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %input, i32 %add39
>  %82 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx40, align 16, !tbaa !14
>  %83 = extractelement <4 x float> %82, i32 3
>  %mul41 = fmul float %83, 0x3EEF751040000000
>  %arrayidx42 = getelementptr inbounds float, float addrspace(1)* %filter, i32 %i27.0203
>  %84 = load float, float addrspace(1)* %arrayidx42, align 4, !tbaa !12
>  %85 = extractelement <4 x float> %pixel.1204, i32 0
>  %mul43 = fmul float %84, %mul41
>  %86 = extractelement <4 x float> %82, i32 0
>  %87 = tail call float @llvm.fmuladd.f32(float %mul43, float %86, float %85)
>  %88 = insertelement <4 x float> undef, float %87, i32 0
>  %89 = extractelement <4 x float> %pixel.1204, i32 1
>  %90 = extractelement <4 x float> %82, i32 1
>  %91 = tail call float @llvm.fmuladd.f32(float %mul43, float %90, float %89)
>  %92 = insertelement <4 x float> %88, float %91, i32 1
>  %93 = extractelement <4 x float> %pixel.1204, i32 2
>  %94 = extractelement <4 x float> %82, i32 2
>  %95 = tail call float @llvm.fmuladd.f32(float %mul43, float %94, float %93)
>  %96 = insertelement <4 x float> %92, float %95, i32 2
>  %97 = extractelement <4 x float> %pixel.1204, i32 3
>  %98 = tail call float @llvm.fmuladd.f32(float %mul43, float %83, float %97)
>  %99 = insertelement <4 x float> %96, float %98, i32 3
>  %100 = tail call float @llvm.fmuladd.f32(float %84, float %mul41, float %gamma.0202)
>  %inc53 = add nuw nsw i32 %i27.0203, 1
>  %exitcond209 = icmp eq i32 %inc53, %width
>  br i1 %exitcond209, label %for.cond.cleanup30, label %for.body31
>
>cleanup.sink.split:                               ; preds = %for.cond.cleanup30, %for.cond.cleanup
>  %.sink225 = phi float [ %70, %for.cond.cleanup30 ], [ %29, %for.cond.cleanup ]
>  %.sink = phi <4 x float> [ %69, %for.cond.cleanup30 ], [ %28, %for.cond.cleanup ]
>  %pixel.1.lcssa.sink = phi <4 x float> [ %pixel.1.lcssa, %for.cond.cleanup30 ], [ %pixel.0.lcssa, %for.cond.cleanup ]
>  %101 = fcmp ogt float %.sink225, 6.553500e+04
>  %102 = fcmp olt float %.sink225, 0.000000e+00
>  %103 = select i1 %102, float 0.000000e+00, float %.sink225
>  %.op.i182 = fadd float %103, 5.000000e-01
>  %add.i183 = select i1 %101, float 6.553550e+04, float %.op.i182
>  %104 = insertelement <4 x float> %.sink, float %add.i183, i32 2
>  %105 = extractelement <4 x float> %pixel.1.lcssa.sink, i32 3
>  %106 = fcmp ogt float %105, 6.553500e+04
>  %107 = fcmp olt float %105, 0.000000e+00
>  %108 = select i1 %107, float 0.000000e+00, float %105
>  %.op.i190 = fadd float %108, 5.000000e-01
>  %add.i191 = select i1 %106, float 6.553550e+04, float %.op.i190
>  %109 = insertelement <4 x float> %104, float %add.i191, i32 3
>  %mul24 = mul i32 %17, %imageWidth
>  %add25 = add i32 %mul24, %8
>  %arrayidx26 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %output, i32 %add25
>  store <4 x float> %109, <4 x float> addrspace(1)* %arrayidx26, align 16, !tbaa !14
>  br label %cleanup
>
>cleanup:                                          ; preds = %cleanup.sink.split, %entry
>  ret void
>}
>
>; Function Attrs: norecurse nounwind readnone
>define float @BoxResizeFilter(float %x) local_unnamed_addr #0 {
>entry:
>  ret float 1.000000e+00
>}
>
>; Function Attrs: nounwind readonly
>define float @CubicBC(float %x, float addrspace(1)* nocapture readonly %resizeFilterCoefficients) local_unnamed_addr #6 {
>entry:
>  %cmp = fcmp olt float %x, 1.000000e+00
>  br i1 %cmp, label %if.then, label %if.end
>
>if.then:                                          ; preds = %entry
>  %0 = load float, float addrspace(1)* %resizeFilterCoefficients, align 4, !tbaa !12
>  %arrayidx1 = getelementptr inbounds float, float addrspace(1)* %resizeFilterCoefficients, i32 1
>  %1 = load float, float addrspace(1)* %arrayidx1, align 4, !tbaa !12
>  %arrayidx2 = getelementptr inbounds float, float addrspace(1)* %resizeFilterCoefficients, i32 2
>  %2 = load float, float addrspace(1)* %arrayidx2, align 4, !tbaa !12
>  %3 = tail call float @llvm.fmuladd.f32(float %x, float %2, float %1)
>  %mul = fmul float %3, %x
>  %4 = tail call float @llvm.fmuladd.f32(float %x, float %mul, float %0)
>  br label %return
>
>if.end:                                           ; preds = %entry
>  %cmp4 = fcmp olt float %x, 2.000000e+00
>  br i1 %cmp4, label %if.then5, label %return
>
>if.then5:                                         ; preds = %if.end
>  %arrayidx6 = getelementptr inbounds float, float addrspace(1)* %resizeFilterCoefficients, i32 3
>  %5 = load float, float addrspace(1)* %arrayidx6, align 4, !tbaa !12
>  %arrayidx7 = getelementptr inbounds float, float addrspace(1)* %resizeFilterCoefficients, i32 4
>  %6 = load float, float addrspace(1)* %arrayidx7, align 4, !tbaa !12
>  %arrayidx8 = getelementptr inbounds float, float addrspace(1)* %resizeFilterCoefficients, i32 5
>  %7 = load float, float addrspace(1)* %arrayidx8, align 4, !tbaa !12
>  %arrayidx9 = getelementptr inbounds float, float addrspace(1)* %resizeFilterCoefficients, i32 6
>  %8 = load float, float addrspace(1)* %arrayidx9, align 4, !tbaa !12
>  %9 = tail call float @llvm.fmuladd.f32(float %x, float %8, float %7)
>  %10 = tail call float @llvm.fmuladd.f32(float %x, float %9, float %6)
>  %11 = tail call float @llvm.fmuladd.f32(float %x, float %10, float %5)
>  br label %return
>
>return:                                           ; preds = %if.end, %if.then5, %if.then
>  %retval.0 = phi float [ %4, %if.then ], [ %11, %if.then5 ], [ 0.000000e+00, %if.end ]
>  ret float %retval.0
>}
>
>; Function Attrs: nounwind readnone
>define float @Sinc(float %x) local_unnamed_addr #7 {
>entry:
>  %cmp = fcmp une float %x, 0.000000e+00
>  br i1 %cmp, label %if.then, label %return
>
>if.then:                                          ; preds = %entry
>  %mul = fmul float %x, 0x400921FB80000000
>  %0 = bitcast float %x to i32
>  %1 = and i32 %0, -2147483648
>  %2 = and i32 %0, 2147483647
>  %3 = bitcast i32 %2 to float
>  %4 = fptosi float %3 to i32
>  %5 = sitofp i32 %4 to float
>  %6 = fsub float %3, %5
>  %7 = shl i32 %4, 31
>  %8 = xor i32 %7, %1
>  %9 = icmp ult i32 %2, 2139095040
>  %10 = select i1 %9, i32 %1, i32 2143289344
>  %11 = fsub float 1.000000e+00, %6
>  %12 = fcmp ole float %6, 7.500000e-01
>  %13 = fadd float %6, -5.000000e-01
>  %14 = select i1 %12, float %13, float %11
>  %15 = fcmp olt float %6, 5.000000e-01
>  %16 = fsub float 5.000000e-01, %6
>  %17 = select i1 %15, float %16, float %14
>  %18 = fcmp ole float %6, 2.500000e-01
>  %19 = select i1 %18, float %6, float %17
>  %20 = xor i1 %18, true
>  %21 = and i1 %12, %20
>  %22 = fmul float %19, 0x400921FB60000000
>  %23 = fmul float %22, %22
>  %24 = fmul float %22, %23
>  %25 = tail call float @llvm.fmuladd.f32(float %23, float 0x3EC6DBE4A0000000, float 0xBF2A013A80000000) #13
>  %26 = tail call float @llvm.fmuladd.f32(float %23, float %25, float 0x3F811110E0000000) #13
>  %27 = tail call float @llvm.fmuladd.f32(float %23, float %26, float 0xBFC5555560000000) #13
>  %28 = tail call float @llvm.fmuladd.f32(float %24, float %27, float %22) #13
>  %29 = fmul float %23, %23
>  %30 = tail call float @llvm.fmuladd.f32(float %23, float 0xBE92524740000000, float 0x3EFA015C60000000) #13
>  %31 = tail call float @llvm.fmuladd.f32(float %23, float %30, float 0xBF56C16C00000000) #13
>  %32 = tail call float @llvm.fmuladd.f32(float %23, float %31, float 0x3FA5555560000000) #13
>  %33 = tail call float @llvm.fmuladd.f32(float %23, float -5.000000e-01, float 1.000000e+00) #13
>  %34 = tail call float @llvm.fmuladd.f32(float %29, float %32, float %33) #13
>  %35 = select i1 %21, float %34, float %28
>  %36 = bitcast float %35 to i32
>  %37 = xor i32 %8, %36
>  %38 = icmp ult i32 %2, 1258291200
>  %39 = select i1 %38, i32 %37, i32 %10
>  %40 = bitcast i32 %39 to float
>  %div = fdiv float %40, %mul, !fpmath !11
>  br label %return
>
>return:                                           ; preds = %entry, %if.then
>  %retval.0 = phi float [ %div, %if.then ], [ 1.000000e+00, %entry ]
>  ret float %retval.0
>}
>
>; Function Attrs: norecurse nounwind readnone
>define float @Triangle(float %x) local_unnamed_addr #0 {
>entry:
>  %cmp = fcmp olt float %x, 1.000000e+00
>  %sub = fsub float 1.000000e+00, %x
>  %cond = select i1 %cmp, float %sub, float 0.000000e+00
>  ret float %cond
>}
>
>; Function Attrs: convergent nounwind readnone
>define float @Hann(float %x) local_unnamed_addr #8 {
>entry:
>  %mul = fmul float %x, 0x400921FB80000000
>  %0 = bitcast float %mul to i32
>  %1 = and i32 %0, 2147483647
>  %2 = bitcast i32 %1 to float
>  %3 = fcmp olt float %2, 0x4160000000000000
>  br i1 %3, label %4, label %45
>
>; <label>:4:                                      ; preds = %entry
>  %5 = tail call float @llvm.fmuladd.f32(float %2, float 0x3FE45F3060000000, float 5.000000e-01) #13
>  %6 = tail call float @llvm.trunc.f32(float %5) #14
>  %7 = bitcast float %6 to i32
>  %8 = and i32 %7, -4096
>  %9 = bitcast i32 %8 to float
>  %10 = fsub float %6, %9
>  %11 = fmul float %6, 0x3FF921FB40000000
>  %12 = fsub float -0.000000e+00, %11
>  %13 = tail call float @llvm.fmuladd.f32(float %9, float 0x3FF9200000000000, float %12) #13
>  %14 = tail call float @llvm.fmuladd.f32(float %9, float 0x3F3FB40000000000, float %13) #13
>  %15 = tail call float @llvm.fmuladd.f32(float %10, float 0x3FF9200000000000, float %14) #13
>  %16 = tail call float @llvm.fmuladd.f32(float %10, float 0x3F3FB40000000000, float %15) #13
>  %17 = fsub float %2, %11
>  %18 = fsub float %2, %17
>  %19 = fsub float %18, %11
>  %20 = fsub float %19, %16
>  %21 = fadd float %17, %20
>  %22 = fmul float %6, 0x3E74442D00000000
>  %23 = fsub float -0.000000e+00, %22
>  %24 = tail call float @llvm.fmuladd.f32(float %9, float 0x3E74440000000000, float %23) #13
>  %25 = tail call float @llvm.fmuladd.f32(float %9, float 0x3D86800000000000, float %24) #13
>  %26 = tail call float @llvm.fmuladd.f32(float %10, float 0x3E74440000000000, float %25) #13
>  %27 = tail call float @llvm.fmuladd.f32(float %10, float 0x3D86800000000000, float %26) #13
>  %28 = fsub float %21, %22
>  %29 = fsub float %21, %28
>  %30 = fsub float %29, %22
>  %31 = fsub float %30, %27
>  %32 = fadd float %28, %31
>  %33 = fmul float %6, 0x3CF8469880000000
>  %34 = fsub float -0.000000e+00, %33
>  %35 = tail call float @llvm.fmuladd.f32(float %9, float 0x3CF8460000000000, float %34) #13
>  %36 = tail call float @llvm.fmuladd.f32(float %9, float 0x3C23100000000000, float %35) #13
>  %37 = tail call float @llvm.fmuladd.f32(float %10, float 0x3CF8460000000000, float %36) #13
>  %38 = tail call float @llvm.fmuladd.f32(float %10, float 0x3C23100000000000, float %37) #13
>  %39 = fsub float %32, %33
>  %40 = fsub float %32, %39
>  %41 = fsub float %40, %33
>  %42 = fadd float %39, %41
>  %43 = fsub float -0.000000e+00, %38
>  %44 = fptosi float %6 to i32
>  br label %_Z3cosf.exit
>
>; <label>:45:                                     ; preds = %entry
>  %46 = lshr i32 %1, 23
>  %47 = and i32 %0, 8388607
>  %48 = or i32 %47, 8388608
>  %49 = mul i32 %48, -28220501
>  %50 = zext i32 %48 to i64
>  %51 = mul nuw nsw i64 %50, 4266746795
>  %52 = lshr i64 %51, 32
>  %53 = trunc i64 %52 to i32
>  %54 = mul i32 %48, 1011060801
>  %55 = add i32 %54, %53
>  %56 = mul nuw nsw i64 %50, 1011060801
>  %57 = lshr i64 %56, 32
>  %58 = trunc i64 %57 to i32
>  %59 = icmp ult i32 %55, %53
>  %60 = zext i1 %59 to i32
>  %61 = add nuw nsw i32 %60, %58
>  %62 = mul i32 %48, -614296167
>  %63 = add i32 %61, %62
>  %64 = mul nuw nsw i64 %50, 3680671129
>  %65 = lshr i64 %64, 32
>  %66 = trunc i64 %65 to i32
>  %67 = icmp ult i32 %63, %61
>  %68 = zext i1 %67 to i32
>  %69 = add nuw nsw i32 %68, %66
>  %70 = mul i32 %48, -181084736
>  %71 = add i32 %69, %70
>  %72 = mul nuw nsw i64 %50, 4113882560
>  %73 = lshr i64 %72, 32
>  %74 = trunc i64 %73 to i32
>  %75 = icmp ult i32 %71, %69
>  %76 = zext i1 %75 to i32
>  %77 = add nuw nsw i32 %76, %74
>  %78 = mul i32 %48, -64530479
>  %79 = add i32 %77, %78
>  %80 = mul nuw nsw i64 %50, 4230436817
>  %81 = lshr i64 %80, 32
>  %82 = trunc i64 %81 to i32
>  %83 = icmp ult i32 %79, %77
>  %84 = zext i1 %83 to i32
>  %85 = add nuw nsw i32 %84, %82
>  %86 = mul i32 %48, 1313084713
>  %87 = add i32 %85, %86
>  %88 = mul nuw nsw i64 %50, 1313084713
>  %89 = lshr i64 %88, 32
>  %90 = trunc i64 %89 to i32
>  %91 = icmp ult i32 %87, %85
>  %92 = zext i1 %91 to i32
>  %93 = add nuw nsw i32 %92, %90
>  %94 = mul i32 %48, -1560706194
>  %95 = add i32 %93, %94
>  %96 = mul nuw nsw i64 %50, 2734261102
>  %97 = lshr i64 %96, 32
>  %98 = trunc i64 %97 to i32
>  %99 = icmp ult i32 %95, %93
>  %100 = zext i1 %99 to i32
>  %101 = add nuw nsw i32 %100, %98
>  %102 = add nsw i32 %46, -120
>  %103 = icmp ugt i32 %102, 31
>  %104 = select i1 %103, i32 %95, i32 %101
>  %105 = select i1 %103, i32 %87, i32 %95
>  %106 = select i1 %103, i32 %79, i32 %87
>  %107 = select i1 %103, i32 %71, i32 %79
>  %108 = select i1 %103, i32 %63, i32 %71
>  %109 = select i1 %103, i32 %55, i32 %63
>  %110 = select i1 %103, i32 %49, i32 %55
>  %111 = select i1 %103, i32 32, i32 0
>  %112 = sub nsw i32 %102, %111
>  %113 = icmp ugt i32 %112, 31
>  %114 = select i1 %113, i32 %105, i32 %104
>  %115 = select i1 %113, i32 %106, i32 %105
>  %116 = select i1 %113, i32 %107, i32 %106
>  %117 = select i1 %113, i32 %108, i32 %107
>  %118 = select i1 %113, i32 %109, i32 %108
>  %119 = select i1 %113, i32 %110, i32 %109
>  %120 = select i1 %113, i32 32, i32 0
>  %121 = sub nsw i32 %112, %120
>  %122 = icmp ugt i32 %121, 31
>  %123 = select i1 %122, i32 %115, i32 %114
>  %124 = select i1 %122, i32 %116, i32 %115
>  %125 = select i1 %122, i32 %117, i32 %116
>  %126 = select i1 %122, i32 %118, i32 %117
>  %127 = select i1 %122, i32 %119, i32 %118
>  %128 = select i1 %122, i32 32, i32 0
>  %129 = sub nsw i32 %121, %128
>  %130 = icmp ugt i32 %129, 31
>  %131 = select i1 %130, i32 %124, i32 %123
>  %132 = select i1 %130, i32 %125, i32 %124
>  %133 = select i1 %130, i32 %126, i32 %125
>  %134 = select i1 %130, i32 %127, i32 %126
>  %135 = select i1 %130, i32 32, i32 0
>  %136 = sub nsw i32 %129, %135
>  %137 = icmp ne i32 %136, 0
>  %138 = sub nsw i32 0, %136
>  %139 = and i32 %136, 31
>  %140 = shl i32 %131, %139
>  %141 = and i32 %138, 31
>  %142 = lshr i32 %132, %141
>  %143 = or i32 %140, %142
>  %144 = shl i32 %132, %139
>  %145 = lshr i32 %133, %141
>  %146 = or i32 %144, %145
>  %147 = shl i32 %133, %139
>  %148 = lshr i32 %134, %141
>  %149 = or i32 %147, %148
>  %150 = select i1 %137, i32 %143, i32 %131
>  %151 = select i1 %137, i32 %146, i32 %132
>  %152 = select i1 %137, i32 %149, i32 %133
>  %153 = lshr i32 %150, 29
>  %154 = shl i32 %150, 2
>  %155 = lshr i32 %151, 30
>  %156 = or i32 %154, %155
>  %157 = shl i32 %151, 2
>  %158 = lshr i32 %152, 30
>  %159 = or i32 %157, %158
>  %160 = shl i32 %152, 2
>  %161 = lshr i32 %134, 30
>  %162 = or i32 %160, %161
>  %163 = and i32 %153, 1
>  %164 = shl i32 %153, 31
>  %165 = ashr exact i32 %164, 31
>  %166 = xor i32 %165, %156
>  %167 = xor i32 %165, %159
>  %168 = xor i32 %165, %162
>  %169 = tail call i32 @llvm.ctlz.i32(i32 %166, i1 false) #13, !range !52
>  %170 = add nuw nsw i32 %169, 1
>  %171 = sub nsw i32 31, %169
>  %172 = and i32 %170, 31
>  %173 = shl i32 %166, %172
>  %174 = and i32 %171, 31
>  %175 = lshr i32 %167, %174
>  %176 = or i32 %173, %175
>  %177 = shl i32 %167, %172
>  %178 = lshr i32 %168, %174
>  %179 = or i32 %177, %178
>  %180 = shl nuw nsw i32 %169, 23
>  %181 = sub nsw i32 1056964608, %180
>  %182 = or i32 %181, %164
>  %183 = lshr i32 %176, 9
>  %184 = or i32 %182, %183
>  %185 = bitcast i32 %184 to float
>  %186 = shl i32 %176, 23
>  %187 = lshr i32 %179, 9
>  %188 = or i32 %186, %187
>  %189 = tail call i32 @llvm.ctlz.i32(i32 %188, i1 false) #13, !range !52
>  %190 = add nuw nsw i32 %189, 1
>  %191 = sub nsw i32 31, %189
>  %192 = and i32 %190, 31
>  %193 = shl i32 %188, %192
>  %194 = and i32 %191, 31
>  %195 = lshr i32 %179, %194
>  %196 = or i32 %193, %195
>  %197 = add nuw nsw i32 %169, 24
>  %198 = add nuw nsw i32 %197, %190
>  %199 = shl nuw nsw i32 %198, 23
>  %200 = sub nsw i32 1065353216, %199
>  %201 = or i32 %200, %164
>  %202 = lshr i32 %196, 9
>  %203 = or i32 %201, %202
>  %204 = bitcast i32 %203 to float
>  %205 = and i32 %184, -1073745920
>  %206 = bitcast i32 %205 to float
>  %207 = fsub float %185, %206
>  %208 = fmul float %185, 0x3FF921FB40000000
>  %209 = fsub float -0.000000e+00, %208
>  %210 = tail call float @llvm.fmuladd.f32(float %206, float 0x3FF9200000000000, float %209) #13
>  %211 = tail call float @llvm.fmuladd.f32(float %206, float 0x3F3FB40000000000, float %210) #13
>  %212 = tail call float @llvm.fmuladd.f32(float %207, float 0x3FF9200000000000, float %211) #13
>  %213 = tail call float @llvm.fmuladd.f32(float %207, float 0x3F3FB40000000000, float %212) #13
>  %214 = tail call float @llvm.fmuladd.f32(float %185, float 0x3E74442D00000000, float %213) #13
>  %215 = tail call float @llvm.fmuladd.f32(float %204, float 0x3FF921FB40000000, float %214) #13
>  %216 = fadd float %208, %215
>  %217 = fsub float %216, %208
>  %218 = fsub float %215, %217
>  %219 = lshr i32 %150, 30
>  %220 = add nuw nsw i32 %163, %219
>  br label %_Z3cosf.exit
>
>_Z3cosf.exit:                                     ; preds = %4, %45
>  %.05.i = phi float [ %42, %4 ], [ %216, %45 ]
>  %.0.i = phi float [ %43, %4 ], [ %218, %45 ]
>  %221 = phi i32 [ %44, %4 ], [ %220, %45 ]
>  %222 = fmul float %.05.i, %.05.i
>  %223 = fmul float %.05.i, %222
>  %224 = tail call float @llvm.fmuladd.f32(float %222, float 0x3DE5D93A60000000, float 0xBE5AE5E680000000) #13
>  %225 = tail call float @llvm.fmuladd.f32(float %222, float %224, float 0x3EC6DBE4A0000000) #13
>  %226 = tail call float @llvm.fmuladd.f32(float %222, float %225, float 0xBF2A013A80000000) #13
>  %227 = tail call float @llvm.fmuladd.f32(float %222, float %226, float 0x3F811110E0000000) #13
>  %228 = fmul float %223, %227
>  %229 = fsub float -0.000000e+00, %228
>  %230 = tail call float @llvm.fmuladd.f32(float %.0.i, float 5.000000e-01, float %229) #13
>  %231 = fsub float -0.000000e+00, %.0.i
>  %232 = tail call float @llvm.fmuladd.f32(float %222, float %230, float %231) #13
>  %233 = tail call float @llvm.fmuladd.f32(float %223, float 0x3FC5555560000000, float %232) #13
>  %234 = fsub float %.05.i, %233
>  %235 = fsub float -0.000000e+00, %234
>  %236 = tail call float @llvm.fmuladd.f32(float %222, float 0xBDA8FAE9C0000000, float 0x3E21EE9EC0000000) #13
>  %237 = tail call float @llvm.fmuladd.f32(float %222, float %236, float 0xBE92524740000000) #13
>  %238 = tail call float @llvm.fmuladd.f32(float %222, float %237, float 0x3EFA015C40000000) #13
>  %239 = tail call float @llvm.fmuladd.f32(float %222, float %238, float 0xBF56C16C00000000) #13
>  %240 = tail call float @llvm.fmuladd.f32(float %222, float %239, float 0x3FA5555560000000) #13
>  %241 = fmul float %222, %240
>  %242 = bitcast float %.05.i to i32
>  %243 = and i32 %242, 2147483647
>  %244 = add nsw i32 %243, -16777216
>  %245 = bitcast i32 %244 to float
>  %246 = add nsw i32 %243, -1050253722
>  %247 = icmp ult i32 %246, 11429479
>  %248 = select i1 %247, float %245, float 0.000000e+00
>  %249 = icmp ugt i32 %243, 1061683200
>  %250 = select i1 %249, float 2.812500e-01, float %248
>  %251 = fsub float -0.000000e+00, %250
>  %252 = tail call float @llvm.fmuladd.f32(float %222, float 5.000000e-01, float %251) #13
>  %253 = fsub float 1.000000e+00, %250
>  %254 = fmul float %.05.i, %.0.i
>  %255 = fsub float -0.000000e+00, %254
>  %256 = tail call float @llvm.fmuladd.f32(float %222, float %241, float %255) #13
>  %257 = fsub float %252, %256
>  %258 = fsub float %253, %257
>  %259 = and i32 %221, 1
>  %260 = icmp eq i32 %259, 0
>  %261 = select i1 %260, float %258, float %235
>  %262 = bitcast float %261 to i32
>  %263 = lshr i32 %221, 1
>  %.lobit.i = shl i32 %263, 31
>  %264 = xor i32 %.lobit.i, %262
>  %265 = bitcast i32 %264 to float
>  %266 = icmp ugt i32 %1, 2139095039
>  %267 = select i1 %266, float 0x7FF8000000000000, float %265
>  %268 = tail call float @llvm.fmuladd.f32(float %267, float 5.000000e-01, float 5.000000e-01)
>  ret float %268
>}
>
>; Function Attrs: convergent nounwind readnone
>define float @Hamming(float %x) local_unnamed_addr #8 {
>entry:
>  %mul = fmul float %x, 0x400921FB80000000
>  %0 = bitcast float %mul to i32
>  %1 = and i32 %0, 2147483647
>  %2 = bitcast i32 %1 to float
>  %3 = fcmp olt float %2, 0x4160000000000000
>  br i1 %3, label %4, label %45
>
>; <label>:4:                                      ; preds = %entry
>  %5 = tail call float @llvm.fmuladd.f32(float %2, float 0x3FE45F3060000000, float 5.000000e-01) #13
>  %6 = tail call float @llvm.trunc.f32(float %5) #14
>  %7 = bitcast float %6 to i32
>  %8 = and i32 %7, -4096
>  %9 = bitcast i32 %8 to float
>  %10 = fsub float %6, %9
>  %11 = fmul float %6, 0x3FF921FB40000000
>  %12 = fsub float -0.000000e+00, %11
>  %13 = tail call float @llvm.fmuladd.f32(float %9, float 0x3FF9200000000000, float %12) #13
>  %14 = tail call float @llvm.fmuladd.f32(float %9, float 0x3F3FB40000000000, float %13) #13
>  %15 = tail call float @llvm.fmuladd.f32(float %10, float 0x3FF9200000000000, float %14) #13
>  %16 = tail call float @llvm.fmuladd.f32(float %10, float 0x3F3FB40000000000, float %15) #13
>  %17 = fsub float %2, %11
>  %18 = fsub float %2, %17
>  %19 = fsub float %18, %11
>  %20 = fsub float %19, %16
>  %21 = fadd float %17, %20
>  %22 = fmul float %6, 0x3E74442D00000000
>  %23 = fsub float -0.000000e+00, %22
>  %24 = tail call float @llvm.fmuladd.f32(float %9, float 0x3E74440000000000, float %23) #13
>  %25 = tail call float @llvm.fmuladd.f32(float %9, float 0x3D86800000000000, float %24) #13
>  %26 = tail call float @llvm.fmuladd.f32(float %10, float 0x3E74440000000000, float %25) #13
>  %27 = tail call float @llvm.fmuladd.f32(float %10, float 0x3D86800000000000, float %26) #13
>  %28 = fsub float %21, %22
>  %29 = fsub float %21, %28
>  %30 = fsub float %29, %22
>  %31 = fsub float %30, %27
>  %32 = fadd float %28, %31
>  %33 = fmul float %6, 0x3CF8469880000000
>  %34 = fsub float -0.000000e+00, %33
>  %35 = tail call float @llvm.fmuladd.f32(float %9, float 0x3CF8460000000000, float %34) #13
>  %36 = tail call float @llvm.fmuladd.f32(float %9, float 0x3C23100000000000, float %35) #13
>  %37 = tail call float @llvm.fmuladd.f32(float %10, float 0x3CF8460000000000, float %36) #13
>  %38 = tail call float @llvm.fmuladd.f32(float %10, float 0x3C23100000000000, float %37) #13
>  %39 = fsub float %32, %33
>  %40 = fsub float %32, %39
>  %41 = fsub float %40, %33
>  %42 = fadd float %39, %41
>  %43 = fsub float -0.000000e+00, %38
>  %44 = fptosi float %6 to i32
>  br label %_Z3cosf.exit
>
>; <label>:45:                                     ; preds = %entry
>  %46 = lshr i32 %1, 23
>  %47 = and i32 %0, 8388607
>  %48 = or i32 %47, 8388608
>  %49 = mul i32 %48, -28220501
>  %50 = zext i32 %48 to i64
>  %51 = mul nuw nsw i64 %50, 4266746795
>  %52 = lshr i64 %51, 32
>  %53 = trunc i64 %52 to i32
>  %54 = mul i32 %48, 1011060801
>  %55 = add i32 %54, %53
>  %56 = mul nuw nsw i64 %50, 1011060801
>  %57 = lshr i64 %56, 32
>  %58 = trunc i64 %57 to i32
>  %59 = icmp ult i32 %55, %53
>  %60 = zext i1 %59 to i32
>  %61 = add nuw nsw i32 %60, %58
>  %62 = mul i32 %48, -614296167
>  %63 = add i32 %61, %62
>  %64 = mul nuw nsw i64 %50, 3680671129
>  %65 = lshr i64 %64, 32
>  %66 = trunc i64 %65 to i32
>  %67 = icmp ult i32 %63, %61
>  %68 = zext i1 %67 to i32
>  %69 = add nuw nsw i32 %68, %66
>  %70 = mul i32 %48, -181084736
>  %71 = add i32 %69, %70
>  %72 = mul nuw nsw i64 %50, 4113882560
>  %73 = lshr i64 %72, 32
>  %74 = trunc i64 %73 to i32
>  %75 = icmp ult i32 %71, %69
>  %76 = zext i1 %75 to i32
>  %77 = add nuw nsw i32 %76, %74
>  %78 = mul i32 %48, -64530479
>  %79 = add i32 %77, %78
>  %80 = mul nuw nsw i64 %50, 4230436817
>  %81 = lshr i64 %80, 32
>  %82 = trunc i64 %81 to i32
>  %83 = icmp ult i32 %79, %77
>  %84 = zext i1 %83 to i32
>  %85 = add nuw nsw i32 %84, %82
>  %86 = mul i32 %48, 1313084713
>  %87 = add i32 %85, %86
>  %88 = mul nuw nsw i64 %50, 1313084713
>  %89 = lshr i64 %88, 32
>  %90 = trunc i64 %89 to i32
>  %91 = icmp ult i32 %87, %85
>  %92 = zext i1 %91 to i32
>  %93 = add nuw nsw i32 %92, %90
>  %94 = mul i32 %48, -1560706194
>  %95 = add i32 %93, %94
>  %96 = mul nuw nsw i64 %50, 2734261102
>  %97 = lshr i64 %96, 32
>  %98 = trunc i64 %97 to i32
>  %99 = icmp ult i32 %95, %93
>  %100 = zext i1 %99 to i32
>  %101 = add nuw nsw i32 %100, %98
>  %102 = add nsw i32 %46, -120
>  %103 = icmp ugt i32 %102, 31
>  %104 = select i1 %103, i32 %95, i32 %101
>  %105 = select i1 %103, i32 %87, i32 %95
>  %106 = select i1 %103, i32 %79, i32 %87
>  %107 = select i1 %103, i32 %71, i32 %79
>  %108 = select i1 %103, i32 %63, i32 %71
>  %109 = select i1 %103, i32 %55, i32 %63
>  %110 = select i1 %103, i32 %49, i32 %55
>  %111 = select i1 %103, i32 32, i32 0
>  %112 = sub nsw i32 %102, %111
>  %113 = icmp ugt i32 %112, 31
>  %114 = select i1 %113, i32 %105, i32 %104
>  %115 = select i1 %113, i32 %106, i32 %105
>  %116 = select i1 %113, i32 %107, i32 %106
>  %117 = select i1 %113, i32 %108, i32 %107
>  %118 = select i1 %113, i32 %109, i32 %108
>  %119 = select i1 %113, i32 %110, i32 %109
>  %120 = select i1 %113, i32 32, i32 0
>  %121 = sub nsw i32 %112, %120
>  %122 = icmp ugt i32 %121, 31
>  %123 = select i1 %122, i32 %115, i32 %114
>  %124 = select i1 %122, i32 %116, i32 %115
>  %125 = select i1 %122, i32 %117, i32 %116
>  %126 = select i1 %122, i32 %118, i32 %117
>  %127 = select i1 %122, i32 %119, i32 %118
>  %128 = select i1 %122, i32 32, i32 0
>  %129 = sub nsw i32 %121, %128
>  %130 = icmp ugt i32 %129, 31
>  %131 = select i1 %130, i32 %124, i32 %123
>  %132 = select i1 %130, i32 %125, i32 %124
>  %133 = select i1 %130, i32 %126, i32 %125
>  %134 = select i1 %130, i32 %127, i32 %126
>  %135 = select i1 %130, i32 32, i32 0
>  %136 = sub nsw i32 %129, %135
>  %137 = icmp ne i32 %136, 0
>  %138 = sub nsw i32 0, %136
>  %139 = and i32 %136, 31
>  %140 = shl i32 %131, %139
>  %141 = and i32 %138, 31
>  %142 = lshr i32 %132, %141
>  %143 = or i32 %140, %142
>  %144 = shl i32 %132, %139
>  %145 = lshr i32 %133, %141
>  %146 = or i32 %144, %145
>  %147 = shl i32 %133, %139
>  %148 = lshr i32 %134, %141
>  %149 = or i32 %147, %148
>  %150 = select i1 %137, i32 %143, i32 %131
>  %151 = select i1 %137, i32 %146, i32 %132
>  %152 = select i1 %137, i32 %149, i32 %133
>  %153 = lshr i32 %150, 29
>  %154 = shl i32 %150, 2
>  %155 = lshr i32 %151, 30
>  %156 = or i32 %154, %155
>  %157 = shl i32 %151, 2
>  %158 = lshr i32 %152, 30
>  %159 = or i32 %157, %158
>  %160 = shl i32 %152, 2
>  %161 = lshr i32 %134, 30
>  %162 = or i32 %160, %161
>  %163 = and i32 %153, 1
>  %164 = shl i32 %153, 31
>  %165 = ashr exact i32 %164, 31
>  %166 = xor i32 %165, %156
>  %167 = xor i32 %165, %159
>  %168 = xor i32 %165, %162
>  %169 = tail call i32 @llvm.ctlz.i32(i32 %166, i1 false) #13, !range !52
>  %170 = add nuw nsw i32 %169, 1
>  %171 = sub nsw i32 31, %169
>  %172 = and i32 %170, 31
>  %173 = shl i32 %166, %172
>  %174 = and i32 %171, 31
>  %175 = lshr i32 %167, %174
>  %176 = or i32 %173, %175
>  %177 = shl i32 %167, %172
>  %178 = lshr i32 %168, %174
>  %179 = or i32 %177, %178
>  %180 = shl nuw nsw i32 %169, 23
>  %181 = sub nsw i32 1056964608, %180
>  %182 = or i32 %181, %164
>  %183 = lshr i32 %176, 9
>  %184 = or i32 %182, %183
>  %185 = bitcast i32 %184 to float
>  %186 = shl i32 %176, 23
>  %187 = lshr i32 %179, 9
>  %188 = or i32 %186, %187
>  %189 = tail call i32 @llvm.ctlz.i32(i32 %188, i1 false) #13, !range !52
>  %190 = add nuw nsw i32 %189, 1
>  %191 = sub nsw i32 31, %189
>  %192 = and i32 %190, 31
>  %193 = shl i32 %188, %192
>  %194 = and i32 %191, 31
>  %195 = lshr i32 %179, %194
>  %196 = or i32 %193, %195
>  %197 = add nuw nsw i32 %169, 24
>  %198 = add nuw nsw i32 %197, %190
>  %199 = shl nuw nsw i32 %198, 23
>  %200 = sub nsw i32 1065353216, %199
>  %201 = or i32 %200, %164
>  %202 = lshr i32 %196, 9
>  %203 = or i32 %201, %202
>  %204 = bitcast i32 %203 to float
>  %205 = and i32 %184, -1073745920
>  %206 = bitcast i32 %205 to float
>  %207 = fsub float %185, %206
>  %208 = fmul float %185, 0x3FF921FB40000000
>  %209 = fsub float -0.000000e+00, %208
>  %210 = tail call float @llvm.fmuladd.f32(float %206, float 0x3FF9200000000000, float %209) #13
>  %211 = tail call float @llvm.fmuladd.f32(float %206, float 0x3F3FB40000000000, float %210) #13
>  %212 = tail call float @llvm.fmuladd.f32(float %207, float 0x3FF9200000000000, float %211) #13
>  %213 = tail call float @llvm.fmuladd.f32(float %207, float 0x3F3FB40000000000, float %212) #13
>  %214 = tail call float @llvm.fmuladd.f32(float %185, float 0x3E74442D00000000, float %213) #13
>  %215 = tail call float @llvm.fmuladd.f32(float %204, float 0x3FF921FB40000000, float %214) #13
>  %216 = fadd float %208, %215
>  %217 = fsub float %216, %208
>  %218 = fsub float %215, %217
>  %219 = lshr i32 %150, 30
>  %220 = add nuw nsw i32 %163, %219
>  br label %_Z3cosf.exit
>
>_Z3cosf.exit:                                     ; preds = %4, %45
>  %.05.i = phi float [ %42, %4 ], [ %216, %45 ]
>  %.0.i = phi float [ %43, %4 ], [ %218, %45 ]
>  %221 = phi i32 [ %44, %4 ], [ %220, %45 ]
>  %222 = fmul float %.05.i, %.05.i
>  %223 = fmul float %.05.i, %222
>  %224 = tail call float @llvm.fmuladd.f32(float %222, float 0x3DE5D93A60000000, float 0xBE5AE5E680000000) #13
>  %225 = tail call float @llvm.fmuladd.f32(float %222, float %224, float 0x3EC6DBE4A0000000) #13
>  %226 = tail call float @llvm.fmuladd.f32(float %222, float %225, float 0xBF2A013A80000000) #13
>  %227 = tail call float @llvm.fmuladd.f32(float %222, float %226, float 0x3F811110E0000000) #13
>  %228 = fmul float %223, %227
>  %229 = fsub float -0.000000e+00, %228
>  %230 = tail call float @llvm.fmuladd.f32(float %.0.i, float 5.000000e-01, float %229) #13
>  %231 = fsub float -0.000000e+00, %.0.i
>  %232 = tail call float @llvm.fmuladd.f32(float %222, float %230, float %231) #13
>  %233 = tail call float @llvm.fmuladd.f32(float %223, float 0x3FC5555560000000, float %232) #13
>  %234 = fsub float %.05.i, %233
>  %235 = fsub float -0.000000e+00, %234
>  %236 = tail call float @llvm.fmuladd.f32(float %222, float 0xBDA8FAE9C0000000, float 0x3E21EE9EC0000000) #13
>  %237 = tail call float @llvm.fmuladd.f32(float %222, float %236, float 0xBE92524740000000) #13
>  %238 = tail call float @llvm.fmuladd.f32(float %222, float %237, float 0x3EFA015C40000000) #13
>  %239 = tail call float @llvm.fmuladd.f32(float %222, float %238, float 0xBF56C16C00000000) #13
>  %240 = tail call float @llvm.fmuladd.f32(float %222, float %239, float 0x3FA5555560000000) #13
>  %241 = fmul float %222, %240
>  %242 = bitcast float %.05.i to i32
>  %243 = and i32 %242, 2147483647
>  %244 = add nsw i32 %243, -16777216
>  %245 = bitcast i32 %244 to float
>  %246 = add nsw i32 %243, -1050253722
>  %247 = icmp ult i32 %246, 11429479
>  %248 = select i1 %247, float %245, float 0.000000e+00
>  %249 = icmp ugt i32 %243, 1061683200
>  %250 = select i1 %249, float 2.812500e-01, float %248
>  %251 = fsub float -0.000000e+00, %250
>  %252 = tail call float @llvm.fmuladd.f32(float %222, float 5.000000e-01, float %251) #13
>  %253 = fsub float 1.000000e+00, %250
>  %254 = fmul float %.05.i, %.0.i
>  %255 = fsub float -0.000000e+00, %254
>  %256 = tail call float @llvm.fmuladd.f32(float %222, float %241, float %255) #13
>  %257 = fsub float %252, %256
>  %258 = fsub float %253, %257
>  %259 = and i32 %221, 1
>  %260 = icmp eq i32 %259, 0
>  %261 = select i1 %260, float %258, float %235
>  %262 = bitcast float %261 to i32
>  %263 = lshr i32 %221, 1
>  %.lobit.i = shl i32 %263, 31
>  %264 = xor i32 %.lobit.i, %262
>  %265 = bitcast i32 %264 to float
>  %266 = icmp ugt i32 %1, 2139095039
>  %267 = select i1 %266, float 0x7FF8000000000000, float %265
>  %268 = tail call float @llvm.fmuladd.f32(float %267, float 0x3FDD70A3E0000000, float 0x3FE147AE20000000)
>  ret float %268
>}
>
>; Function Attrs: convergent nounwind readnone
>define float @Blackman(float %x) local_unnamed_addr #8 {
>entry:
>  %mul = fmul float %x, 0x400921FB80000000
>  %0 = bitcast float %mul to i32
>  %1 = and i32 %0, 2147483647
>  %2 = bitcast i32 %1 to float
>  %3 = fcmp olt float %2, 0x4160000000000000
>  br i1 %3, label %4, label %45
>
>; <label>:4:                                      ; preds = %entry
>  %5 = tail call float @llvm.fmuladd.f32(float %2, float 0x3FE45F3060000000, float 5.000000e-01) #13
>  %6 = tail call float @llvm.trunc.f32(float %5) #14
>  %7 = bitcast float %6 to i32
>  %8 = and i32 %7, -4096
>  %9 = bitcast i32 %8 to float
>  %10 = fsub float %6, %9
>  %11 = fmul float %6, 0x3FF921FB40000000
>  %12 = fsub float -0.000000e+00, %11
>  %13 = tail call float @llvm.fmuladd.f32(float %9, float 0x3FF9200000000000, float %12) #13
>  %14 = tail call float @llvm.fmuladd.f32(float %9, float 0x3F3FB40000000000, float %13) #13
>  %15 = tail call float @llvm.fmuladd.f32(float %10, float 0x3FF9200000000000, float %14) #13
>  %16 = tail call float @llvm.fmuladd.f32(float %10, float 0x3F3FB40000000000, float %15) #13
>  %17 = fsub float %2, %11
>  %18 = fsub float %2, %17
>  %19 = fsub float %18, %11
>  %20 = fsub float %19, %16
>  %21 = fadd float %17, %20
>  %22 = fmul float %6, 0x3E74442D00000000
>  %23 = fsub float -0.000000e+00, %22
>  %24 = tail call float @llvm.fmuladd.f32(float %9, float 0x3E74440000000000, float %23) #13
>  %25 = tail call float @llvm.fmuladd.f32(float %9, float 0x3D86800000000000, float %24) #13
>  %26 = tail call float @llvm.fmuladd.f32(float %10, float 0x3E74440000000000, float %25) #13
>  %27 = tail call float @llvm.fmuladd.f32(float %10, float 0x3D86800000000000, float %26) #13
>  %28 = fsub float %21, %22
>  %29 = fsub float %21, %28
>  %30 = fsub float %29, %22
>  %31 = fsub float %30, %27
>  %32 = fadd float %28, %31
>  %33 = fmul float %6, 0x3CF8469880000000
>  %34 = fsub float -0.000000e+00, %33
>  %35 = tail call float @llvm.fmuladd.f32(float %9, float 0x3CF8460000000000, float %34) #13
>  %36 = tail call float @llvm.fmuladd.f32(float %9, float 0x3C23100000000000, float %35) #13
>  %37 = tail call float @llvm.fmuladd.f32(float %10, float 0x3CF8460000000000, float %36) #13
>  %38 = tail call float @llvm.fmuladd.f32(float %10, float 0x3C23100000000000, float %37) #13
>  %39 = fsub float %32, %33
>  %40 = fsub float %32, %39
>  %41 = fsub float %40, %33
>  %42 = fadd float %39, %41
>  %43 = fsub float -0.000000e+00, %38
>  %44 = fptosi float %6 to i32
>  br label %_Z3cosf.exit
>
>; <label>:45:                                     ; preds = %entry
>  %46 = lshr i32 %1, 23
>  %47 = and i32 %0, 8388607
>  %48 = or i32 %47, 8388608
>  %49 = mul i32 %48, -28220501
>  %50 = zext i32 %48 to i64
>  %51 = mul nuw nsw i64 %50, 4266746795
>  %52 = lshr i64 %51, 32
>  %53 = trunc i64 %52 to i32
>  %54 = mul i32 %48, 1011060801
>  %55 = add i32 %54, %53
>  %56 = mul nuw nsw i64 %50, 1011060801
>  %57 = lshr i64 %56, 32
>  %58 = trunc i64 %57 to i32
>  %59 = icmp ult i32 %55, %53
>  %60 = zext i1 %59 to i32
>  %61 = add nuw nsw i32 %60, %58
>  %62 = mul i32 %48, -614296167
>  %63 = add i32 %61, %62
>  %64 = mul nuw nsw i64 %50, 3680671129
>  %65 = lshr i64 %64, 32
>  %66 = trunc i64 %65 to i32
>  %67 = icmp ult i32 %63, %61
>  %68 = zext i1 %67 to i32
>  %69 = add nuw nsw i32 %68, %66
>  %70 = mul i32 %48, -181084736
>  %71 = add i32 %69, %70
>  %72 = mul nuw nsw i64 %50, 4113882560
>  %73 = lshr i64 %72, 32
>  %74 = trunc i64 %73 to i32
>  %75 = icmp ult i32 %71, %69
>  %76 = zext i1 %75 to i32
>  %77 = add nuw nsw i32 %76, %74
>  %78 = mul i32 %48, -64530479
>  %79 = add i32 %77, %78
>  %80 = mul nuw nsw i64 %50, 4230436817
>  %81 = lshr i64 %80, 32
>  %82 = trunc i64 %81 to i32
>  %83 = icmp ult i32 %79, %77
>  %84 = zext i1 %83 to i32
>  %85 = add nuw nsw i32 %84, %82
>  %86 = mul i32 %48, 1313084713
>  %87 = add i32 %85, %86
>  %88 = mul nuw nsw i64 %50, 1313084713
>  %89 = lshr i64 %88, 32
>  %90 = trunc i64 %89 to i32
>  %91 = icmp ult i32 %87, %85
>  %92 = zext i1 %91 to i32
>  %93 = add nuw nsw i32 %92, %90
>  %94 = mul i32 %48, -1560706194
>  %95 = add i32 %93, %94
>  %96 = mul nuw nsw i64 %50, 2734261102
>  %97 = lshr i64 %96, 32
>  %98 = trunc i64 %97 to i32
>  %99 = icmp ult i32 %95, %93
>  %100 = zext i1 %99 to i32
>  %101 = add nuw nsw i32 %100, %98
>  %102 = add nsw i32 %46, -120
>  %103 = icmp ugt i32 %102, 31
>  %104 = select i1 %103, i32 %95, i32 %101
>  %105 = select i1 %103, i32 %87, i32 %95
>  %106 = select i1 %103, i32 %79, i32 %87
>  %107 = select i1 %103, i32 %71, i32 %79
>  %108 = select i1 %103, i32 %63, i32 %71
>  %109 = select i1 %103, i32 %55, i32 %63
>  %110 = select i1 %103, i32 %49, i32 %55
>  %111 = select i1 %103, i32 32, i32 0
>  %112 = sub nsw i32 %102, %111
>  %113 = icmp ugt i32 %112, 31
>  %114 = select i1 %113, i32 %105, i32 %104
>  %115 = select i1 %113, i32 %106, i32 %105
>  %116 = select i1 %113, i32 %107, i32 %106
>  %117 = select i1 %113, i32 %108, i32 %107
>  %118 = select i1 %113, i32 %109, i32 %108
>  %119 = select i1 %113, i32 %110, i32 %109
>  %120 = select i1 %113, i32 32, i32 0
>  %121 = sub nsw i32 %112, %120
>  %122 = icmp ugt i32 %121, 31
>  %123 = select i1 %122, i32 %115, i32 %114
>  %124 = select i1 %122, i32 %116, i32 %115
>  %125 = select i1 %122, i32 %117, i32 %116
>  %126 = select i1 %122, i32 %118, i32 %117
>  %127 = select i1 %122, i32 %119, i32 %118
>  %128 = select i1 %122, i32 32, i32 0
>  %129 = sub nsw i32 %121, %128
>  %130 = icmp ugt i32 %129, 31
>  %131 = select i1 %130, i32 %124, i32 %123
>  %132 = select i1 %130, i32 %125, i32 %124
>  %133 = select i1 %130, i32 %126, i32 %125
>  %134 = select i1 %130, i32 %127, i32 %126
>  %135 = select i1 %130, i32 32, i32 0
>  %136 = sub nsw i32 %129, %135
>  %137 = icmp ne i32 %136, 0
>  %138 = sub nsw i32 0, %136
>  %139 = and i32 %136, 31
>  %140 = shl i32 %131, %139
>  %141 = and i32 %138, 31
>  %142 = lshr i32 %132, %141
>  %143 = or i32 %140, %142
>  %144 = shl i32 %132, %139
>  %145 = lshr i32 %133, %141
>  %146 = or i32 %144, %145
>  %147 = shl i32 %133, %139
>  %148 = lshr i32 %134, %141
>  %149 = or i32 %147, %148
>  %150 = select i1 %137, i32 %143, i32 %131
>  %151 = select i1 %137, i32 %146, i32 %132
>  %152 = select i1 %137, i32 %149, i32 %133
>  %153 = lshr i32 %150, 29
>  %154 = shl i32 %150, 2
>  %155 = lshr i32 %151, 30
>  %156 = or i32 %154, %155
>  %157 = shl i32 %151, 2
>  %158 = lshr i32 %152, 30
>  %159 = or i32 %157, %158
>  %160 = shl i32 %152, 2
>  %161 = lshr i32 %134, 30
>  %162 = or i32 %160, %161
>  %163 = and i32 %153, 1
>  %164 = shl i32 %153, 31
>  %165 = ashr exact i32 %164, 31
>  %166 = xor i32 %165, %156
>  %167 = xor i32 %165, %159
>  %168 = xor i32 %165, %162
>  %169 = tail call i32 @llvm.ctlz.i32(i32 %166, i1 false) #13, !range !52
>  %170 = add nuw nsw i32 %169, 1
>  %171 = sub nsw i32 31, %169
>  %172 = and i32 %170, 31
>  %173 = shl i32 %166, %172
>  %174 = and i32 %171, 31
>  %175 = lshr i32 %167, %174
>  %176 = or i32 %173, %175
>  %177 = shl i32 %167, %172
>  %178 = lshr i32 %168, %174
>  %179 = or i32 %177, %178
>  %180 = shl nuw nsw i32 %169, 23
>  %181 = sub nsw i32 1056964608, %180
>  %182 = or i32 %181, %164
>  %183 = lshr i32 %176, 9
>  %184 = or i32 %182, %183
>  %185 = bitcast i32 %184 to float
>  %186 = shl i32 %176, 23
>  %187 = lshr i32 %179, 9
>  %188 = or i32 %186, %187
>  %189 = tail call i32 @llvm.ctlz.i32(i32 %188, i1 false) #13, !range !52
>  %190 = add nuw nsw i32 %189, 1
>  %191 = sub nsw i32 31, %189
>  %192 = and i32 %190, 31
>  %193 = shl i32 %188, %192
>  %194 = and i32 %191, 31
>  %195 = lshr i32 %179, %194
>  %196 = or i32 %193, %195
>  %197 = add nuw nsw i32 %169, 24
>  %198 = add nuw nsw i32 %197, %190
>  %199 = shl nuw nsw i32 %198, 23
>  %200 = sub nsw i32 1065353216, %199
>  %201 = or i32 %200, %164
>  %202 = lshr i32 %196, 9
>  %203 = or i32 %201, %202
>  %204 = bitcast i32 %203 to float
>  %205 = and i32 %184, -1073745920
>  %206 = bitcast i32 %205 to float
>  %207 = fsub float %185, %206
>  %208 = fmul float %185, 0x3FF921FB40000000
>  %209 = fsub float -0.000000e+00, %208
>  %210 = tail call float @llvm.fmuladd.f32(float %206, float 0x3FF9200000000000, float %209) #13
>  %211 = tail call float @llvm.fmuladd.f32(float %206, float 0x3F3FB40000000000, float %210) #13
>  %212 = tail call float @llvm.fmuladd.f32(float %207, float 0x3FF9200000000000, float %211) #13
>  %213 = tail call float @llvm.fmuladd.f32(float %207, float 0x3F3FB40000000000, float %212) #13
>  %214 = tail call float @llvm.fmuladd.f32(float %185, float 0x3E74442D00000000, float %213) #13
>  %215 = tail call float @llvm.fmuladd.f32(float %204, float 0x3FF921FB40000000, float %214) #13
>  %216 = fadd float %208, %215
>  %217 = fsub float %216, %208
>  %218 = fsub float %215, %217
>  %219 = lshr i32 %150, 30
>  %220 = add nuw nsw i32 %163, %219
>  br label %_Z3cosf.exit
>
>_Z3cosf.exit:                                     ; preds = %4, %45
>  %.05.i = phi float [ %42, %4 ], [ %216, %45 ]
>  %.0.i = phi float [ %43, %4 ], [ %218, %45 ]
>  %221 = phi i32 [ %44, %4 ], [ %220, %45 ]
>  %222 = fmul float %.05.i, %.05.i
>  %223 = fmul float %.05.i, %222
>  %224 = tail call float @llvm.fmuladd.f32(float %222, float 0x3DE5D93A60000000, float 0xBE5AE5E680000000) #13
>  %225 = tail call float @llvm.fmuladd.f32(float %222, float %224, float 0x3EC6DBE4A0000000) #13
>  %226 = tail call float @llvm.fmuladd.f32(float %222, float %225, float 0xBF2A013A80000000) #13
>  %227 = tail call float @llvm.fmuladd.f32(float %222, float %226, float 0x3F811110E0000000) #13
>  %228 = fmul float %223, %227
>  %229 = fsub float -0.000000e+00, %228
>  %230 = tail call float @llvm.fmuladd.f32(float %.0.i, float 5.000000e-01, float %229) #13
>  %231 = fsub float -0.000000e+00, %.0.i
>  %232 = tail call float @llvm.fmuladd.f32(float %222, float %230, float %231) #13
>  %233 = tail call float @llvm.fmuladd.f32(float %223, float 0x3FC5555560000000, float %232) #13
>  %234 = fsub float %.05.i, %233
>  %235 = fsub float -0.000000e+00, %234
>  %236 = tail call float @llvm.fmuladd.f32(float %222, float 0xBDA8FAE9C0000000, float 0x3E21EE9EC0000000) #13
>  %237 = tail call float @llvm.fmuladd.f32(float %222, float %236, float 0xBE92524740000000) #13
>  %238 = tail call float @llvm.fmuladd.f32(float %222, float %237, float 0x3EFA015C40000000) #13
>  %239 = tail call float @llvm.fmuladd.f32(float %222, float %238, float 0xBF56C16C00000000) #13
>  %240 = tail call float @llvm.fmuladd.f32(float %222, float %239, float 0x3FA5555560000000) #13
>  %241 = fmul float %222, %240
>  %242 = bitcast float %.05.i to i32
>  %243 = and i32 %242, 2147483647
>  %244 = add nsw i32 %243, -16777216
>  %245 = bitcast i32 %244 to float
>  %246 = add nsw i32 %243, -1050253722
>  %247 = icmp ult i32 %246, 11429479
>  %248 = select i1 %247, float %245, float 0.000000e+00
>  %249 = icmp ugt i32 %243, 1061683200
>  %250 = select i1 %249, float 2.812500e-01, float %248
>  %251 = fsub float -0.000000e+00, %250
>  %252 = tail call float @llvm.fmuladd.f32(float %222, float 5.000000e-01, float %251) #13
>  %253 = fsub float 1.000000e+00, %250
>  %254 = fmul float %.05.i, %.0.i
>  %255 = fsub float -0.000000e+00, %254
>  %256 = tail call float @llvm.fmuladd.f32(float %222, float %241, float %255) #13
>  %257 = fsub float %252, %256
>  %258 = fsub float %253, %257
>  %259 = and i32 %221, 1
>  %260 = icmp eq i32 %259, 0
>  %261 = select i1 %260, float %258, float %235
>  %262 = bitcast float %261 to i32
>  %263 = lshr i32 %221, 1
>  %.lobit.i = shl i32 %263, 31
>  %264 = xor i32 %.lobit.i, %262
>  %265 = bitcast i32 %264 to float
>  %266 = icmp ugt i32 %1, 2139095039
>  %267 = select i1 %266, float 0x7FF8000000000000, float %265
>  %268 = tail call float @llvm.fmuladd.f32(float %267, float 0x3FC47AE140000000, float 5.000000e-01)
>  %269 = tail call float @llvm.fmuladd.f32(float %267, float %268, float 0x3FD5C28F60000000)
>  ret float %269
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @ResizeHorizontalFilter(float addrspace(1)* nocapture readonly %inputImage, i32 %number_channels, i32 %inputColumns, i32 %inputRows, float addrspace(1)* nocapture %filteredImage, i32 %filteredColumns, i32 %filteredRows, float %xFactor, i32 %resizeFilterType, i32 %resizeWindowType, float addrspace(1)* %resizeFilterCubicCoefficients, float %resizeFilterScale, float %resizeFilterSupport, float %resizeFilterWindowSupport, float %resizeFilterBlur, float addrspace(3)* nocapture %inputImageCache, i32 %numCachedPixels, i32 %pixelPerWorkgroup, i32 %pixelChunkSize, <4 x float> addrspace(3)* nocapture %outputPixelCache, float addrspace(3)* nocapture %densityCache, float addrspace(3)* nocapture %gammaCache) local_unnamed_addr #9 !kernel_arg_addr_space !78 !kernel_arg_access_qual !79 !kernel_arg_type !80 !kernel_arg_base_type !81 !kernel_arg_type_qual !82 !reqd_work_group_size !83 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %mul = mul i32 %0, %pixelPerWorkgroup
>  %add = add i32 %mul, %pixelPerWorkgroup
>  %cmp = icmp ult i32 %add, %filteredColumns
>  %add.filteredColumns = select i1 %cmp, i32 %add, i32 %filteredColumns
>  %div = fdiv float 1.000000e+00, %xFactor, !fpmath !11
>  %add2 = fadd float %div, 0.000000e+00
>  %cmp3 = fcmp ogt float %add2, 1.000000e+00
>  %cond9 = select i1 %cmp3, float %add2, float 1.000000e+00
>  %mul10 = fmul float %cond9, %resizeFilterSupport
>  %cmp11 = fcmp ogt float %mul10, 5.000000e-01
>  %cond16 = select i1 %cmp11, float %mul10, float 5.000000e-01
>  %cmp.i455 = fcmp olt float %cond9, 0.000000e+00
>  %cond.i456 = select i1 %cmp.i455, float -1.000000e+00, float 1.000000e+00
>  %mul.i457 = fmul float %cond9, %cond.i456
>  %cmp1.i458 = fcmp ult float %mul.i457, 0.000000e+00
>  %div.i459 = fdiv float 1.000000e+00, %cond9
>  %mul2.i460 = fmul float %cond.i456, 0x7FF0000000000000
>  %cond3.i461 = select i1 %cmp1.i458, float %mul2.i460, float %div.i459
>  %conv = uitofp i32 %mul to float
>  %add18 = fadd float %conv, 5.000000e-01
>  %div19 = fdiv float %add18, %xFactor, !fpmath !11
>  %add20 = fadd float %div19, 0.000000e+00
>  %sub21 = fsub float %add20, %cond16
>  %add22 = fadd float %sub21, 5.000000e-01
>  %conv23 = fptosi float %add22 to i32
>  %cmp24 = icmp sgt i32 %conv23, 0
>  %spec.select = select i1 %cmp24, i32 %conv23, i32 0
>  %add37 = add nsw i32 %spec.select, %numCachedPixels
>  %cmp38 = icmp slt i32 %add37, %inputColumns
>  %cond44 = select i1 %cmp38, i32 %add37, i32 %inputColumns
>  %1 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %2 = mul i32 %y.i.i, %1
>  %3 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %4 = add i32 %2, %3
>  %5 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %6 = getelementptr inbounds i8, i8 addrspace(7)* %5, i32 8
>  %7 = bitcast i8 addrspace(7)* %6 to i32 addrspace(7)*
>  %8 = load i32, i32 addrspace(7)* %7, align 4, !tbaa !4
>  %9 = add i32 %4, %8
>  %mul1.i = mul i32 %9, %inputColumns
>  %mul4.i466 = add i32 %mul1.i, %spec.select
>  %add.i = mul i32 %mul4.i466, %number_channels
>  %sub47 = sub nsw i32 %cond44, %spec.select
>  %mul48 = mul i32 %sub47, %number_channels
>  %add.ptr = getelementptr inbounds float, float addrspace(1)* %inputImage, i32 %add.i
>  %x.i.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %z.i.i.i = tail call i32 @llvm.r600.read.local.size.z() #13
>  %10 = mul i32 %z.i.i.i, %y.i.i
>  %11 = mul i32 %10, %x.i.i.i
>  %12 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %13 = mul i32 %12, %y.i.i
>  %14 = tail call i32 @llvm.r600.read.tidig.z() #13, !range !3
>  %reass.add.i.i = add i32 %13, %3
>  %reass.mul.i.i = mul i32 %reass.add.i.i, %z.i.i.i
>  %15 = add i32 %reass.mul.i.i, %14
>  %16 = icmp ult i32 %15, %mul48
>  br i1 %16, label %.preheader.i.i, label %_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit
>
>.preheader.i.i:                                   ; preds = %entry, %.preheader.i.i
>  %17 = phi i32 [ %23, %.preheader.i.i ], [ %15, %entry ]
>  %18 = getelementptr inbounds float, float addrspace(1)* %add.ptr, i32 %17
>  %19 = bitcast float addrspace(1)* %18 to i32 addrspace(1)*
>  %20 = load i32, i32 addrspace(1)* %19, align 4, !tbaa !12
>  %21 = getelementptr inbounds float, float addrspace(3)* %inputImageCache, i32 %17
>  %22 = bitcast float addrspace(3)* %21 to i32 addrspace(3)*
>  store i32 %20, i32 addrspace(3)* %22, align 4, !tbaa !12
>  %23 = add i32 %17, %11
>  %24 = icmp ult i32 %23, %mul48
>  br i1 %24, label %.preheader.i.i, label %_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit
>
>_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit: ; preds = %.preheader.i.i, %entry
>  tail call void @llvm.r600.group.barrier() #13
>  %cmp50 = icmp eq i32 %number_channels, 4
>  %cmp52 = icmp eq i32 %number_channels, 2
>  %or.cond = or i1 %cmp50, %cmp52
>  %sub55 = add i32 %number_channels, -1
>  %cond58 = select i1 %or.cond, i32 %sub55, i32 0
>  %sub = xor i32 %mul, -1
>  %add59 = add i32 %sub, %pixelChunkSize
>  %sub60 = add i32 %add59, %add.filteredColumns
>  %div61 = udiv i32 %sub60, %pixelChunkSize
>  %cmp62478 = icmp ult i32 %sub60, %pixelChunkSize
>  br i1 %cmp62478, label %for.cond.cleanup, label %for.body.lr.ph
>
>for.body.lr.ph:                                   ; preds = %_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit
>  %conv100 = uitofp i32 %inputColumns to float
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %outputPixelCache, i32 %12
>  %arrayidx170 = getelementptr inbounds float, float addrspace(3)* %densityCache, i32 %12
>  %cmp171 = icmp eq i32 %cond58, 0
>  %cmp.i451 = fcmp olt float %resizeFilterWindowSupport, 0.000000e+00
>  %cmp1.i452 = icmp eq i32 %resizeWindowType, 0
>  %or.cond.i = or i1 %cmp1.i452, %cmp.i451
>  %cmp145 = icmp ugt i32 %number_channels, 2
>  %arrayidx174 = getelementptr inbounds float, float addrspace(3)* %gammaCache, i32 %12
>  %cmp211 = icmp ne i32 %cond58, 0
>  %mul1.i.i = mul i32 %9, %filteredColumns
>  %add241 = add i32 %mul1.i.i, %12
>  br label %for.body
>
>for.cond.cleanup:                                 ; preds = %if.end242, %_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit
>  ret void
>
>for.body:                                         ; preds = %for.body.lr.ph, %if.end242
>  %chunk.0479 = phi i32 [ 0, %for.body.lr.ph ], [ %inc244, %if.end242 ]
>  %mul64 = mul i32 %chunk.0479, %pixelChunkSize
>  %add65 = add i32 %mul64, %mul
>  %add66 = add i32 %add65, %pixelChunkSize
>  %cmp67 = icmp ult i32 %add66, %add.filteredColumns
>  %add66.add.filteredColumns = select i1 %cmp67, i32 %add66, i32 %add.filteredColumns
>  %sub74 = sub i32 %add66.add.filteredColumns, %add65
>  %div.i465 = udiv i32 %x.i.i.i, %sub74
>  %div.i462 = udiv i32 %12, %div.i465
>  %cmp.i463 = icmp ult i32 %div.i462, %sub74
>  %cond.i464 = select i1 %cmp.i463, i32 %div.i462, i32 -1
>  br i1 %cmp.i463, label %if.then, label %if.end166
>
>if.then:                                          ; preds = %for.body
>  %add82 = add i32 %cond.i464, %add65
>  %conv83 = sitofp i32 %add82 to float
>  %add84 = fadd float %conv83, 5.000000e-01
>  %div85 = fdiv float %add84, %xFactor, !fpmath !11
>  %add86 = fadd float %div85, 0.000000e+00
>  %sub87 = fsub float %add86, %cond16
>  %add88 = fadd float %sub87, 5.000000e-01
>  %cmp89 = fcmp ogt float %add88, 0.000000e+00
>  %cond96 = select i1 %cmp89, float %add88, float 0.000000e+00
>  %conv97 = fptoui float %cond96 to i32
>  %add98 = fadd float %cond16, %add86
>  %add99 = fadd float %add98, 5.000000e-01
>  %cmp101 = fcmp olt float %add99, %conv100
>  %add99.conv100 = select i1 %cmp101, float %add99, float %conv100
>  %conv110 = fptoui float %add99.conv100 to i32
>  %sub111 = sub i32 %conv110, %conv97
>  %div112 = udiv i32 %sub111, %div.i465
>  %mul113 = mul i32 %div112, %div.i465
>  %cmp114 = icmp ne i32 %mul113, %sub111
>  %cond116 = zext i1 %cmp114 to i32
>  %add117 = add i32 %div112, %cond116
>  %rem = urem i32 %12, %div.i465
>  %mul118 = mul i32 %add117, %rem
>  %cmp119 = icmp ult i32 %mul118, %sub111
>  br i1 %cmp119, label %if.then121, label %if.end166
>
>if.then121:                                       ; preds = %if.then
>  %add122 = add i32 %mul118, %add117
>  %cmp123 = icmp ult i32 %add122, %sub111
>  %add122.sub111 = select i1 %cmp123, i32 %add122, i32 %sub111
>  %cmp133468 = icmp ult i32 %mul118, %add122.sub111
>  br i1 %cmp133468, label %for.body136.preheader, label %if.end166
>
>for.body136.preheader:                            ; preds = %if.then121
>  %add130 = sub i32 %conv97, %spec.select
>  %sub131 = add i32 %add130, %mul118
>  br label %for.body136
>
>for.body136:                                      ; preds = %for.body136.preheader, %if.end162
>  %i.0473 = phi i32 [ %inc, %if.end162 ], [ %mul118, %for.body136.preheader ]
>  %cacheIndex.0472 = phi i32 [ %inc164, %if.end162 ], [ %sub131, %for.body136.preheader ]
>  %filteredPixel.0471 = phi <4 x float> [ %filteredPixel.1, %if.end162 ], [ zeroinitializer, %for.body136.preheader ]
>  %density.0470 = phi float [ %add163, %if.end162 ], [ 0.000000e+00, %for.body136.preheader ]
>  %gamma.0469 = phi float [ %gamma.1, %if.end162 ], [ 0.000000e+00, %for.body136.preheader ]
>  %add137 = add i32 %i.0473, %conv97
>  %conv138 = uitofp i32 %add137 to float
>  %sub139 = fsub float %conv138, %add86
>  %add140 = fadd float %sub139, 5.000000e-01
>  %mul141 = fmul float %cond3.i461, %add140
>  %div.i450 = fdiv float %mul141, %resizeFilterBlur, !fpmath !11
>  %call.i = tail call float @llvm.fabs.f32(float %div.i450) #17
>  br i1 %or.cond.i, label %getResizeFilterWeight.exit, label %if.else.i
>
>if.else.i:                                        ; preds = %for.body136
>  %mul.i453 = fmul float %call.i, %resizeFilterScale
>  %call2.i = tail call float @applyResizeFilter(float %mul.i453, i32 %resizeWindowType, float addrspace(1)* %resizeFilterCubicCoefficients) #17
>  br label %getResizeFilterWeight.exit
>
>getResizeFilterWeight.exit:                       ; preds = %for.body136, %if.else.i
>  %scale.0.i = phi float [ %call2.i, %if.else.i ], [ 1.000000e+00, %for.body136 ]
>  %call3.i = tail call float @applyResizeFilter(float %call.i, i32 %resizeFilterType, float addrspace(1)* %resizeFilterCubicCoefficients) #17
>  %mul4.i = fmul float %scale.0.i, %call3.i
>  %mul143 = mul i32 %cacheIndex.0472, %number_channels
>  %add.ptr144 = getelementptr inbounds float, float addrspace(3)* %inputImageCache, i32 %mul143
>  %25 = load float, float addrspace(3)* %add.ptr144, align 4, !tbaa !12
>  %26 = insertelement <4 x float> <float undef, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00>, float %25, i32 0
>  br i1 %cmp145, label %if.then147, label %if.end
>
>if.then147:                                       ; preds = %getResizeFilterWeight.exit
>  %add.ptr148 = getelementptr inbounds float, float addrspace(3)* %add.ptr144, i32 1
>  %27 = load float, float addrspace(3)* %add.ptr148, align 4, !tbaa !12
>  %28 = insertelement <4 x float> %26, float %27, i32 1
>  %add.ptr149 = getelementptr inbounds float, float addrspace(3)* %add.ptr144, i32 2
>  %29 = load float, float addrspace(3)* %add.ptr149, align 4, !tbaa !12
>  %30 = insertelement <4 x float> %28, float %29, i32 2
>  br label %if.end
>
>if.end:                                           ; preds = %if.then147, %getResizeFilterWeight.exit
>  %cp.0 = phi <4 x float> [ %30, %if.then147 ], [ %26, %getResizeFilterWeight.exit ]
>  br i1 %cmp171, label %if.else, label %if.then152
>
>if.then152:                                       ; preds = %if.end
>  %add.ptr153 = getelementptr inbounds float, float addrspace(3)* %add.ptr144, i32 %cond58
>  %31 = load float, float addrspace(3)* %add.ptr153, align 4, !tbaa !12
>  %mul154 = fmul float %mul4.i, 0x3EEF751040000000
>  %mul155 = fmul float %mul154, %31
>  %32 = extractelement <4 x float> %cp.0, i32 0
>  %33 = extractelement <4 x float> %filteredPixel.0471, i32 0
>  %34 = tail call float @llvm.fmuladd.f32(float %mul155, float %32, float %33)
>  %35 = insertelement <4 x float> undef, float %34, i32 0
>  %36 = extractelement <4 x float> %cp.0, i32 1
>  %37 = extractelement <4 x float> %filteredPixel.0471, i32 1
>  %38 = tail call float @llvm.fmuladd.f32(float %mul155, float %36, float %37)
>  %39 = insertelement <4 x float> %35, float %38, i32 1
>  %40 = extractelement <4 x float> %cp.0, i32 2
>  %41 = extractelement <4 x float> %filteredPixel.0471, i32 2
>  %42 = tail call float @llvm.fmuladd.f32(float %mul155, float %40, float %41)
>  %43 = insertelement <4 x float> %39, float %42, i32 2
>  %44 = extractelement <4 x float> %filteredPixel.0471, i32 3
>  %45 = tail call float @llvm.fmuladd.f32(float %mul4.i, float %31, float %44)
>  %46 = insertelement <4 x float> %43, float %45, i32 3
>  %add160 = fadd float %gamma.0469, %mul155
>  br label %if.end162
>
>if.else:                                          ; preds = %if.end
>  %splat.splatinsert = insertelement <4 x float> undef, float %mul4.i, i32 0
>  %splat.splat = shufflevector <4 x float> %splat.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
>  %47 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat, <4 x float> %cp.0, <4 x float> %filteredPixel.0471)
>  br label %if.end162
>
>if.end162:                                        ; preds = %if.else, %if.then152
>  %gamma.1 = phi float [ %add160, %if.then152 ], [ %gamma.0469, %if.else ]
>  %filteredPixel.1 = phi <4 x float> [ %46, %if.then152 ], [ %47, %if.else ]
>  %add163 = fadd float %density.0470, %mul4.i
>  %inc = add nuw i32 %i.0473, 1
>  %inc164 = add i32 %cacheIndex.0472, 1
>  %cmp133 = icmp ult i32 %inc, %add122.sub111
>  br i1 %cmp133, label %for.body136, label %if.end166
>
>if.end166:                                        ; preds = %if.end162, %if.then121, %if.then, %for.body
>  %gamma.3 = phi float [ 0.000000e+00, %for.body ], [ 0.000000e+00, %if.then ], [ 0.000000e+00, %if.then121 ], [ %gamma.1, %if.end162 ]
>  %density.2 = phi float [ 0.000000e+00, %for.body ], [ 0.000000e+00, %if.then ], [ 0.000000e+00, %if.then121 ], [ %add163, %if.end162 ]
>  %filteredPixel.3 = phi <4 x float> [ zeroinitializer, %for.body ], [ zeroinitializer, %if.then ], [ zeroinitializer, %if.then121 ], [ %filteredPixel.1, %if.end162 ]
>  %cmp167 = icmp ult i32 %12, %sub74
>  br i1 %cmp167, label %if.then169, label %if.end176
>
>if.then169:                                       ; preds = %if.end166
>  store <4 x float> zeroinitializer, <4 x float> addrspace(3)* %arrayidx, align 16, !tbaa !14
>  store float 0.000000e+00, float addrspace(3)* %arrayidx170, align 4, !tbaa !12
>  br i1 %cmp171, label %if.end176, label %if.then173
>
>if.then173:                                       ; preds = %if.then169
>  store float 0.000000e+00, float addrspace(3)* %arrayidx174, align 4, !tbaa !12
>  br label %if.end176
>
>if.end176:                                        ; preds = %if.then169, %if.then173, %if.end166
>  tail call void @llvm.r600.group.barrier() #13
>  %cmp179476 = icmp ugt i32 %sub74, %x.i.i.i
>  br i1 %cmp179476, label %for.cond.cleanup181, label %for.body182.lr.ph
>
>for.body182.lr.ph:                                ; preds = %if.end176
>  %arrayidx190 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %outputPixelCache, i32 %cond.i464
>  %arrayidx192 = getelementptr inbounds float, float addrspace(3)* %densityCache, i32 %cond.i464
>  %arrayidx197 = getelementptr inbounds float, float addrspace(3)* %gammaCache, i32 %cond.i464
>  br label %for.body182
>
>for.cond.cleanup181:                              ; preds = %if.end201, %if.end176
>  br i1 %cmp167, label %if.then207, label %if.end242
>
>for.body182:                                      ; preds = %for.body182.lr.ph, %if.end201
>  %i177.0477 = phi i32 [ 0, %for.body182.lr.ph ], [ %inc203, %if.end201 ]
>  br i1 %cmp.i463, label %if.then185, label %if.end201
>
>if.then185:                                       ; preds = %for.body182
>  %48 = mul i32 %div.i462, %div.i465
>  %49 = sub i32 %12, %48
>  %cmp187 = icmp eq i32 %49, %i177.0477
>  br i1 %cmp187, label %if.then189, label %if.end201
>
>if.then189:                                       ; preds = %if.then185
>  %50 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx190, align 16, !tbaa !14
>  %add191 = fadd <4 x float> %filteredPixel.3, %50
>  store <4 x float> %add191, <4 x float> addrspace(3)* %arrayidx190, align 16, !tbaa !14
>  %51 = load float, float addrspace(3)* %arrayidx192, align 4, !tbaa !12
>  %add193 = fadd float %density.2, %51
>  store float %add193, float addrspace(3)* %arrayidx192, align 4, !tbaa !12
>  br i1 %cmp171, label %if.end201, label %if.then196
>
>if.then196:                                       ; preds = %if.then189
>  %52 = load float, float addrspace(3)* %arrayidx197, align 4, !tbaa !12
>  %add198 = fadd float %gamma.3, %52
>  store float %add198, float addrspace(3)* %arrayidx197, align 4, !tbaa !12
>  br label %if.end201
>
>if.end201:                                        ; preds = %if.then189, %if.then185, %if.then196, %for.body182
>  tail call void @llvm.r600.group.barrier() #13
>  %inc203 = add nuw i32 %i177.0477, 1
>  %cmp179 = icmp ult i32 %inc203, %div.i465
>  br i1 %cmp179, label %for.body182, label %for.cond.cleanup181
>
>if.then207:                                       ; preds = %for.cond.cleanup181
>  %53 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx, align 16, !tbaa !14
>  br i1 %cmp211, label %if.then213, label %if.end215
>
>if.then213:                                       ; preds = %if.then207
>  %54 = load float, float addrspace(3)* %arrayidx174, align 4, !tbaa !12
>  br label %if.end215
>
>if.end215:                                        ; preds = %if.then213, %if.then207
>  %gamma210.0 = phi float [ %54, %if.then213 ], [ 0.000000e+00, %if.then207 ]
>  %55 = load float, float addrspace(3)* %arrayidx170, align 4, !tbaa !12
>  %cmp218 = fcmp une float %55, 0.000000e+00
>  %cmp220 = fcmp une float %55, 1.000000e+00
>  %or.cond246 = and i1 %cmp218, %cmp220
>  br i1 %or.cond246, label %if.then222, label %if.end232
>
>if.then222:                                       ; preds = %if.end215
>  %cmp.i443 = fcmp olt float %55, 0.000000e+00
>  %cond.i444 = select i1 %cmp.i443, float -1.000000e+00, float 1.000000e+00
>  %mul.i445 = fmul float %55, %cond.i444
>  %cmp1.i446 = fcmp ult float %mul.i445, 0.000000e+00
>  %div.i447 = fdiv float 1.000000e+00, %55
>  %mul2.i448 = fmul float %cond.i444, 0x7FF0000000000000
>  %cond3.i449 = select i1 %cmp1.i446, float %mul2.i448, float %div.i447
>  %splat.splatinsert224 = insertelement <4 x float> undef, float %cond3.i449, i32 0
>  %splat.splat225 = shufflevector <4 x float> %splat.splatinsert224, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul226 = fmul <4 x float> %53, %splat.splat225
>  %mul230 = fmul float %gamma210.0, %cond3.i449
>  %spec.select441 = select i1 %cmp211, float %mul230, float %gamma210.0
>  br label %if.end232
>
>if.end232:                                        ; preds = %if.then222, %if.end215
>  %filteredPixel208.0 = phi <4 x float> [ %53, %if.end215 ], [ %mul226, %if.then222 ]
>  %gamma210.1 = phi float [ %gamma210.0, %if.end215 ], [ %spec.select441, %if.then222 ]
>  br i1 %cmp211, label %if.then235, label %if.end240
>
>if.then235:                                       ; preds = %if.end232
>  %cmp.i442 = fcmp olt float %gamma210.1, 0.000000e+00
>  %cond.i = select i1 %cmp.i442, float -1.000000e+00, float 1.000000e+00
>  %mul.i = fmul float %gamma210.1, %cond.i
>  %cmp1.i = fcmp ult float %mul.i, 0.000000e+00
>  %div.i = fdiv float 1.000000e+00, %gamma210.1
>  %mul2.i = fmul float %cond.i, 0x7FF0000000000000
>  %cond3.i = select i1 %cmp1.i, float %mul2.i, float %div.i
>  %56 = extractelement <4 x float> %filteredPixel208.0, i32 0
>  %mul237 = fmul float %56, %cond3.i
>  %57 = insertelement <4 x float> %filteredPixel208.0, float %mul237, i32 0
>  %58 = extractelement <4 x float> %filteredPixel208.0, i32 1
>  %mul238 = fmul float %58, %cond3.i
>  %59 = insertelement <4 x float> %57, float %mul238, i32 1
>  %60 = extractelement <4 x float> %filteredPixel208.0, i32 2
>  %mul239 = fmul float %60, %cond3.i
>  %61 = insertelement <4 x float> %59, float %mul239, i32 2
>  br label %if.end240
>
>if.end240:                                        ; preds = %if.then235, %if.end232
>  %filteredPixel208.1 = phi <4 x float> [ %61, %if.then235 ], [ %filteredPixel208.0, %if.end232 ]
>  %mul4.i.i = add i32 %add241, %add65
>  %add.i.i = mul i32 %mul4.i.i, %number_channels
>  %add.ptr.i = getelementptr inbounds float, float addrspace(1)* %filteredImage, i32 %add.i.i
>  %62 = extractelement <4 x float> %filteredPixel208.1, i32 0
>  %63 = fcmp ogt float %62, 6.553500e+04
>  %64 = fcmp olt float %62, 0.000000e+00
>  %65 = select i1 %64, float 0.000000e+00, float %62
>  %.op.i27.i = fadd float %65, 5.000000e-01
>  %add.i28.i = select i1 %63, float 6.553550e+04, float %.op.i27.i
>  store float %add.i28.i, float addrspace(1)* %add.ptr.i, align 4, !tbaa !12
>  br i1 %cmp145, label %if.then.i, label %if.end.i
>
>if.then.i:                                        ; preds = %if.end240
>  %66 = extractelement <4 x float> %filteredPixel208.1, i32 1
>  %67 = fcmp ogt float %66, 6.553500e+04
>  %68 = fcmp olt float %66, 0.000000e+00
>  %69 = select i1 %68, float 0.000000e+00, float %66
>  %.op.i25.i = fadd float %69, 5.000000e-01
>  %add.i26.i = select i1 %67, float 6.553550e+04, float %.op.i25.i
>  %add.ptr.i24.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 1
>  store float %add.i26.i, float addrspace(1)* %add.ptr.i24.i, align 4, !tbaa !12
>  %70 = extractelement <4 x float> %filteredPixel208.1, i32 2
>  %71 = fcmp ogt float %70, 6.553500e+04
>  %72 = fcmp olt float %70, 0.000000e+00
>  %73 = select i1 %72, float 0.000000e+00, float %70
>  %.op.i22.i = fadd float %73, 5.000000e-01
>  %add.i23.i = select i1 %71, float 6.553550e+04, float %.op.i22.i
>  %add.ptr.i21.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 2
>  store float %add.i23.i, float addrspace(1)* %add.ptr.i21.i, align 4, !tbaa !12
>  br label %if.end.i
>
>if.end.i:                                         ; preds = %if.then.i, %if.end240
>  switch i32 %number_channels, label %if.end242 [
>    i32 4, label %if.then6.i
>    i32 2, label %if.then6.i
>  ]
>
>if.then6.i:                                       ; preds = %if.end.i, %if.end.i
>  %74 = extractelement <4 x float> %filteredPixel208.1, i32 3
>  %75 = fcmp ogt float %74, 6.553500e+04
>  %76 = fcmp olt float %74, 0.000000e+00
>  %77 = select i1 %76, float 0.000000e+00, float %74
>  %.op.i.i = fadd float %77, 5.000000e-01
>  %add.i20.i = select i1 %75, float 6.553550e+04, float %.op.i.i
>  %add.ptr.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 %number_channels
>  %add.ptr1.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i, i32 -1
>  store float %add.i20.i, float addrspace(1)* %add.ptr1.i.i, align 4, !tbaa !12
>  br label %if.end242
>
>if.end242:                                        ; preds = %if.then6.i, %if.end.i, %for.cond.cleanup181
>  %inc244 = add nuw i32 %chunk.0479, 1
>  %cmp62 = icmp ult i32 %inc244, %div61
>  br i1 %cmp62, label %for.body, label %for.cond.cleanup
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @ResizeVerticalFilter(float addrspace(1)* nocapture readonly %inputImage, i32 %number_channels, i32 %inputColumns, i32 %inputRows, float addrspace(1)* nocapture %filteredImage, i32 %filteredColumns, i32 %filteredRows, float %yFactor, i32 %resizeFilterType, i32 %resizeWindowType, float addrspace(1)* %resizeFilterCubicCoefficients, float %resizeFilterScale, float %resizeFilterSupport, float %resizeFilterWindowSupport, float %resizeFilterBlur, float addrspace(3)* nocapture %inputImageCache, i32 %numCachedPixels, i32 %pixelPerWorkgroup, i32 %pixelChunkSize, <4 x float> addrspace(3)* nocapture %outputPixelCache, float addrspace(3)* nocapture %densityCache, float addrspace(3)* nocapture %gammaCache) local_unnamed_addr #9 !kernel_arg_addr_space !78 !kernel_arg_access_qual !79 !kernel_arg_type !80 !kernel_arg_base_type !81 !kernel_arg_type_qual !82 !reqd_work_group_size !84 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %mul = mul i32 %0, %pixelPerWorkgroup
>  %add = add i32 %mul, %pixelPerWorkgroup
>  %cmp = icmp ult i32 %add, %filteredRows
>  %add.filteredRows = select i1 %cmp, i32 %add, i32 %filteredRows
>  %div = fdiv float 1.000000e+00, %yFactor, !fpmath !11
>  %add2 = fadd float %div, 0.000000e+00
>  %cmp3 = fcmp ogt float %add2, 1.000000e+00
>  %cond9 = select i1 %cmp3, float %add2, float 1.000000e+00
>  %mul10 = fmul float %cond9, %resizeFilterSupport
>  %cmp11 = fcmp ogt float %mul10, 5.000000e-01
>  %cond16 = select i1 %cmp11, float %mul10, float 5.000000e-01
>  %cmp.i476 = fcmp olt float %cond9, 0.000000e+00
>  %cond.i477 = select i1 %cmp.i476, float -1.000000e+00, float 1.000000e+00
>  %mul.i478 = fmul float %cond9, %cond.i477
>  %cmp1.i479 = fcmp ult float %mul.i478, 0.000000e+00
>  %div.i480 = fdiv float 1.000000e+00, %cond9
>  %mul2.i481 = fmul float %cond.i477, 0x7FF0000000000000
>  %cond3.i482 = select i1 %cmp1.i479, float %mul2.i481, float %div.i480
>  %conv = uitofp i32 %mul to float
>  %add18 = fadd float %conv, 5.000000e-01
>  %div19 = fdiv float %add18, %yFactor, !fpmath !11
>  %add20 = fadd float %div19, 0.000000e+00
>  %sub21 = fsub float %add20, %cond16
>  %add22 = fadd float %sub21, 5.000000e-01
>  %conv23 = fptosi float %add22 to i32
>  %cmp24 = icmp sgt i32 %conv23, 0
>  %spec.select = select i1 %cmp24, i32 %conv23, i32 0
>  %add37 = add nsw i32 %spec.select, %numCachedPixels
>  %cmp38 = icmp slt i32 %add37, %inputRows
>  %cond44 = select i1 %cmp38, i32 %add37, i32 %inputRows
>  %1 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %2 = mul i32 %x.i.i, %1
>  %3 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %4 = add i32 %2, %3
>  %5 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %6 = getelementptr inbounds i8, i8 addrspace(7)* %5, i32 4
>  %7 = bitcast i8 addrspace(7)* %6 to i32 addrspace(7)*
>  %8 = load i32, i32 addrspace(7)* %7, align 4, !tbaa !4
>  %9 = add i32 %4, %8
>  %sub47 = sub nsw i32 %cond44, %spec.select
>  %mul48 = mul i32 %inputColumns, %number_channels
>  %cmp49502 = icmp eq i32 %number_channels, 0
>  br i1 %cmp49502, label %for.cond.cleanup, label %for.body.lr.ph
>
>for.body.lr.ph:                                   ; preds = %entry
>  %mul1.i = mul i32 %spec.select, %inputColumns
>  %mul4.i488 = add i32 %9, %mul1.i
>  %add.i = mul i32 %mul4.i488, %number_channels
>  %add.ptr52 = getelementptr inbounds float, float addrspace(1)* %inputImage, i32 %add.i
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %z.i.i = tail call i32 @llvm.r600.read.local.size.z() #13
>  %10 = mul i32 %z.i.i, %y.i.i
>  %11 = mul i32 %10, %x.i.i
>  %12 = mul i32 %3, %y.i.i
>  %13 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %14 = tail call i32 @llvm.r600.read.tidig.z() #13, !range !3
>  %reass.add.i = add i32 %12, %13
>  %reass.mul.i = mul i32 %reass.add.i, %z.i.i
>  %15 = add i32 %reass.mul.i, %14
>  %16 = icmp ult i32 %15, %sub47
>  br label %for.body
>
>for.cond.cleanup:                                 ; preds = %_Z29async_work_group_strided_copyPU3AS3fPU3AS1Kfjj9ocl_event.exit, %entry
>  %cmp55 = icmp eq i32 %number_channels, 4
>  %cmp57 = icmp eq i32 %number_channels, 2
>  %or.cond = or i1 %cmp55, %cmp57
>  %sub60 = add i32 %number_channels, -1
>  %cond63 = select i1 %or.cond, i32 %sub60, i32 0
>  %sub = xor i32 %mul, -1
>  %add64 = add i32 %sub, %pixelChunkSize
>  %sub65 = add i32 %add64, %add.filteredRows
>  %div66 = udiv i32 %sub65, %pixelChunkSize
>  %cmp68500 = icmp ult i32 %sub65, %pixelChunkSize
>  br i1 %cmp68500, label %for.cond.cleanup70, label %for.body71.lr.ph
>
>for.body71.lr.ph:                                 ; preds = %for.cond.cleanup
>  %17 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %y.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %conv108 = uitofp i32 %inputRows to float
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %outputPixelCache, i32 %17
>  %arrayidx183 = getelementptr inbounds float, float addrspace(3)* %densityCache, i32 %17
>  %cmp184 = icmp eq i32 %cond63, 0
>  %cmp.i472 = fcmp olt float %resizeFilterWindowSupport, 0.000000e+00
>  %cmp1.i473 = icmp eq i32 %resizeWindowType, 0
>  %or.cond.i = or i1 %cmp1.i473, %cmp.i472
>  %cmp153 = icmp ugt i32 %number_channels, 2
>  %mul157 = shl i32 %sub47, 1
>  %mul162 = mul i32 %sub47, %cond63
>  %arrayidx187 = getelementptr inbounds float, float addrspace(3)* %gammaCache, i32 %17
>  %cmp224 = icmp ne i32 %cond63, 0
>  br label %for.body71
>
>for.body:                                         ; preds = %_Z29async_work_group_strided_copyPU3AS3fPU3AS1Kfjj9ocl_event.exit, %for.body.lr.ph
>  %i.0503 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %_Z29async_work_group_strided_copyPU3AS3fPU3AS1Kfjj9ocl_event.exit ]
>  %mul51 = mul i32 %i.0503, %sub47
>  %add.ptr = getelementptr inbounds float, float addrspace(3)* %inputImageCache, i32 %mul51
>  %add.ptr53 = getelementptr inbounds float, float addrspace(1)* %add.ptr52, i32 %i.0503
>  br i1 %16, label %.preheader.i, label %_Z29async_work_group_strided_copyPU3AS3fPU3AS1Kfjj9ocl_event.exit
>
>.preheader.i:                                     ; preds = %for.body, %.preheader.i
>  %18 = phi i32 [ %25, %.preheader.i ], [ %15, %for.body ]
>  %19 = mul i32 %mul48, %18
>  %20 = getelementptr inbounds float, float addrspace(1)* %add.ptr53, i32 %19
>  %21 = bitcast float addrspace(1)* %20 to i32 addrspace(1)*
>  %22 = load i32, i32 addrspace(1)* %21, align 4, !tbaa !12
>  %23 = getelementptr inbounds float, float addrspace(3)* %add.ptr, i32 %18
>  %24 = bitcast float addrspace(3)* %23 to i32 addrspace(3)*
>  store i32 %22, i32 addrspace(3)* %24, align 4, !tbaa !12
>  %25 = add i32 %18, %11
>  %26 = icmp ult i32 %25, %sub47
>  br i1 %26, label %.preheader.i, label %_Z29async_work_group_strided_copyPU3AS3fPU3AS1Kfjj9ocl_event.exit
>
>_Z29async_work_group_strided_copyPU3AS3fPU3AS1Kfjj9ocl_event.exit: ; preds = %.preheader.i, %for.body
>  tail call void @llvm.r600.group.barrier() #13
>  %inc = add nuw i32 %i.0503, 1
>  %exitcond = icmp eq i32 %inc, %number_channels
>  br i1 %exitcond, label %for.cond.cleanup, label %for.body
>
>for.cond.cleanup70:                               ; preds = %if.end255, %for.cond.cleanup
>  ret void
>
>for.body71:                                       ; preds = %for.body71.lr.ph, %if.end255
>  %chunk.0501 = phi i32 [ 0, %for.body71.lr.ph ], [ %inc257, %if.end255 ]
>  %mul72 = mul i32 %chunk.0501, %pixelChunkSize
>  %add73 = add i32 %mul72, %mul
>  %add74 = add i32 %add73, %pixelChunkSize
>  %cmp75 = icmp ult i32 %add74, %add.filteredRows
>  %add74.add.filteredRows = select i1 %cmp75, i32 %add74, i32 %add.filteredRows
>  %sub82 = sub i32 %add74.add.filteredRows, %add73
>  %div.i486 = udiv i32 %y.i, %sub82
>  %div.i483 = udiv i32 %17, %div.i486
>  %cmp.i484 = icmp ult i32 %div.i483, %sub82
>  %cond.i485 = select i1 %cmp.i484, i32 %div.i483, i32 -1
>  br i1 %cmp.i484, label %if.then, label %if.end179
>
>if.then:                                          ; preds = %for.body71
>  %add90 = add i32 %cond.i485, %add73
>  %conv91 = sitofp i32 %add90 to float
>  %add92 = fadd float %conv91, 5.000000e-01
>  %div93 = fdiv float %add92, %yFactor, !fpmath !11
>  %add94 = fadd float %div93, 0.000000e+00
>  %sub95 = fsub float %add94, %cond16
>  %add96 = fadd float %sub95, 5.000000e-01
>  %cmp97 = fcmp ogt float %add96, 0.000000e+00
>  %cond104 = select i1 %cmp97, float %add96, float 0.000000e+00
>  %conv105 = fptoui float %cond104 to i32
>  %add106 = fadd float %cond16, %add94
>  %add107 = fadd float %add106, 5.000000e-01
>  %cmp109 = fcmp olt float %add107, %conv108
>  %add107.conv108 = select i1 %cmp109, float %add107, float %conv108
>  %conv118 = fptoui float %add107.conv108 to i32
>  %sub119 = sub i32 %conv118, %conv105
>  %div120 = udiv i32 %sub119, %div.i486
>  %mul121 = mul i32 %div120, %div.i486
>  %cmp122 = icmp ne i32 %mul121, %sub119
>  %cond124 = zext i1 %cmp122 to i32
>  %add125 = add i32 %div120, %cond124
>  %rem = urem i32 %17, %div.i486
>  %mul126 = mul i32 %add125, %rem
>  %cmp127 = icmp ult i32 %mul126, %sub119
>  br i1 %cmp127, label %if.then129, label %if.end179
>
>if.then129:                                       ; preds = %if.then
>  %add130 = add i32 %mul126, %add125
>  %cmp131 = icmp ult i32 %add130, %sub119
>  %add130.sub119 = select i1 %cmp131, i32 %add130, i32 %sub119
>  %cmp142490 = icmp ult i32 %mul126, %add130.sub119
>  br i1 %cmp142490, label %for.body145.preheader, label %if.end179
>
>for.body145.preheader:                            ; preds = %if.then129
>  %add138 = sub i32 %conv105, %spec.select
>  %sub139 = add i32 %add138, %mul126
>  br label %for.body145
>
>for.body145:                                      ; preds = %for.body145.preheader, %if.end172
>  %i140.0495 = phi i32 [ %inc175, %if.end172 ], [ %mul126, %for.body145.preheader ]
>  %cacheIndex.0494 = phi i32 [ %inc176, %if.end172 ], [ %sub139, %for.body145.preheader ]
>  %filteredPixel.0493 = phi <4 x float> [ %filteredPixel.1, %if.end172 ], [ zeroinitializer, %for.body145.preheader ]
>  %density.0492 = phi float [ %add173, %if.end172 ], [ 0.000000e+00, %for.body145.preheader ]
>  %gamma.0491 = phi float [ %gamma.1, %if.end172 ], [ 0.000000e+00, %for.body145.preheader ]
>  %add146 = add i32 %i140.0495, %conv105
>  %conv147 = uitofp i32 %add146 to float
>  %sub148 = fsub float %conv147, %add94
>  %add149 = fadd float %sub148, 5.000000e-01
>  %mul150 = fmul float %cond3.i482, %add149
>  %div.i471 = fdiv float %mul150, %resizeFilterBlur, !fpmath !11
>  %call.i = tail call float @llvm.fabs.f32(float %div.i471) #17
>  br i1 %or.cond.i, label %getResizeFilterWeight.exit, label %if.else.i
>
>if.else.i:                                        ; preds = %for.body145
>  %mul.i474 = fmul float %call.i, %resizeFilterScale
>  %call2.i = tail call float @applyResizeFilter(float %mul.i474, i32 %resizeWindowType, float addrspace(1)* %resizeFilterCubicCoefficients) #17
>  br label %getResizeFilterWeight.exit
>
>getResizeFilterWeight.exit:                       ; preds = %for.body145, %if.else.i
>  %scale.0.i = phi float [ %call2.i, %if.else.i ], [ 1.000000e+00, %for.body145 ]
>  %call3.i = tail call float @applyResizeFilter(float %call.i, i32 %resizeFilterType, float addrspace(1)* %resizeFilterCubicCoefficients) #17
>  %mul4.i = fmul float %scale.0.i, %call3.i
>  %add.ptr152 = getelementptr inbounds float, float addrspace(3)* %inputImageCache, i32 %cacheIndex.0494
>  %27 = load float, float addrspace(3)* %add.ptr152, align 4, !tbaa !12
>  %28 = insertelement <4 x float> <float undef, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00>, float %27, i32 0
>  br i1 %cmp153, label %if.then155, label %if.end
>
>if.then155:                                       ; preds = %getResizeFilterWeight.exit
>  %add.ptr158 = getelementptr inbounds float, float addrspace(3)* %add.ptr152, i32 %mul157
>  %add.ptr156 = getelementptr inbounds float, float addrspace(3)* %add.ptr152, i32 %sub47
>  %29 = load float, float addrspace(3)* %add.ptr156, align 4, !tbaa !12
>  %30 = insertelement <4 x float> %28, float %29, i32 1
>  %31 = load float, float addrspace(3)* %add.ptr158, align 4, !tbaa !12
>  %32 = insertelement <4 x float> %30, float %31, i32 2
>  br label %if.end
>
>if.end:                                           ; preds = %if.then155, %getResizeFilterWeight.exit
>  %cp.0 = phi <4 x float> [ %32, %if.then155 ], [ %28, %getResizeFilterWeight.exit ]
>  br i1 %cmp184, label %if.else, label %if.then161
>
>if.then161:                                       ; preds = %if.end
>  %add.ptr163 = getelementptr inbounds float, float addrspace(3)* %add.ptr152, i32 %mul162
>  %33 = load float, float addrspace(3)* %add.ptr163, align 4, !tbaa !12
>  %mul164 = fmul float %mul4.i, 0x3EEF751040000000
>  %mul165 = fmul float %mul164, %33
>  %34 = extractelement <4 x float> %cp.0, i32 0
>  %35 = extractelement <4 x float> %filteredPixel.0493, i32 0
>  %36 = tail call float @llvm.fmuladd.f32(float %mul165, float %34, float %35)
>  %37 = insertelement <4 x float> undef, float %36, i32 0
>  %38 = extractelement <4 x float> %cp.0, i32 1
>  %39 = extractelement <4 x float> %filteredPixel.0493, i32 1
>  %40 = tail call float @llvm.fmuladd.f32(float %mul165, float %38, float %39)
>  %41 = insertelement <4 x float> %37, float %40, i32 1
>  %42 = extractelement <4 x float> %cp.0, i32 2
>  %43 = extractelement <4 x float> %filteredPixel.0493, i32 2
>  %44 = tail call float @llvm.fmuladd.f32(float %mul165, float %42, float %43)
>  %45 = insertelement <4 x float> %41, float %44, i32 2
>  %46 = extractelement <4 x float> %filteredPixel.0493, i32 3
>  %47 = tail call float @llvm.fmuladd.f32(float %mul4.i, float %33, float %46)
>  %48 = insertelement <4 x float> %45, float %47, i32 3
>  %add170 = fadd float %gamma.0491, %mul165
>  br label %if.end172
>
>if.else:                                          ; preds = %if.end
>  %splat.splatinsert = insertelement <4 x float> undef, float %mul4.i, i32 0
>  %splat.splat = shufflevector <4 x float> %splat.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
>  %49 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat, <4 x float> %cp.0, <4 x float> %filteredPixel.0493)
>  br label %if.end172
>
>if.end172:                                        ; preds = %if.else, %if.then161
>  %gamma.1 = phi float [ %add170, %if.then161 ], [ %gamma.0491, %if.else ]
>  %filteredPixel.1 = phi <4 x float> [ %48, %if.then161 ], [ %49, %if.else ]
>  %add173 = fadd float %density.0492, %mul4.i
>  %inc175 = add nuw i32 %i140.0495, 1
>  %inc176 = add i32 %cacheIndex.0494, 1
>  %cmp142 = icmp ult i32 %inc175, %add130.sub119
>  br i1 %cmp142, label %for.body145, label %if.end179
>
>if.end179:                                        ; preds = %if.end172, %if.then129, %if.then, %for.body71
>  %gamma.3 = phi float [ 0.000000e+00, %for.body71 ], [ 0.000000e+00, %if.then ], [ 0.000000e+00, %if.then129 ], [ %gamma.1, %if.end172 ]
>  %density.2 = phi float [ 0.000000e+00, %for.body71 ], [ 0.000000e+00, %if.then ], [ 0.000000e+00, %if.then129 ], [ %add173, %if.end172 ]
>  %filteredPixel.3 = phi <4 x float> [ zeroinitializer, %for.body71 ], [ zeroinitializer, %if.then ], [ zeroinitializer, %if.then129 ], [ %filteredPixel.1, %if.end172 ]
>  %cmp180 = icmp ult i32 %17, %sub82
>  br i1 %cmp180, label %if.then182, label %if.end189
>
>if.then182:                                       ; preds = %if.end179
>  store <4 x float> zeroinitializer, <4 x float> addrspace(3)* %arrayidx, align 16, !tbaa !14
>  store float 0.000000e+00, float addrspace(3)* %arrayidx183, align 4, !tbaa !12
>  br i1 %cmp184, label %if.end189, label %if.then186
>
>if.then186:                                       ; preds = %if.then182
>  store float 0.000000e+00, float addrspace(3)* %arrayidx187, align 4, !tbaa !12
>  br label %if.end189
>
>if.end189:                                        ; preds = %if.then182, %if.then186, %if.end179
>  tail call void @llvm.r600.group.barrier() #13
>  %cmp192498 = icmp ugt i32 %sub82, %y.i
>  br i1 %cmp192498, label %for.cond.cleanup194, label %for.body195.lr.ph
>
>for.body195.lr.ph:                                ; preds = %if.end189
>  %arrayidx203 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %outputPixelCache, i32 %cond.i485
>  %arrayidx205 = getelementptr inbounds float, float addrspace(3)* %densityCache, i32 %cond.i485
>  %arrayidx210 = getelementptr inbounds float, float addrspace(3)* %gammaCache, i32 %cond.i485
>  br label %for.body195
>
>for.cond.cleanup194:                              ; preds = %if.end214, %if.end189
>  br i1 %cmp180, label %if.then220, label %if.end255
>
>for.body195:                                      ; preds = %for.body195.lr.ph, %if.end214
>  %i190.0499 = phi i32 [ 0, %for.body195.lr.ph ], [ %inc216, %if.end214 ]
>  br i1 %cmp.i484, label %if.then198, label %if.end214
>
>if.then198:                                       ; preds = %for.body195
>  %50 = mul i32 %div.i483, %div.i486
>  %51 = sub i32 %17, %50
>  %cmp200 = icmp eq i32 %51, %i190.0499
>  br i1 %cmp200, label %if.then202, label %if.end214
>
>if.then202:                                       ; preds = %if.then198
>  %52 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx203, align 16, !tbaa !14
>  %add204 = fadd <4 x float> %filteredPixel.3, %52
>  store <4 x float> %add204, <4 x float> addrspace(3)* %arrayidx203, align 16, !tbaa !14
>  %53 = load float, float addrspace(3)* %arrayidx205, align 4, !tbaa !12
>  %add206 = fadd float %density.2, %53
>  store float %add206, float addrspace(3)* %arrayidx205, align 4, !tbaa !12
>  br i1 %cmp184, label %if.end214, label %if.then209
>
>if.then209:                                       ; preds = %if.then202
>  %54 = load float, float addrspace(3)* %arrayidx210, align 4, !tbaa !12
>  %add211 = fadd float %gamma.3, %54
>  store float %add211, float addrspace(3)* %arrayidx210, align 4, !tbaa !12
>  br label %if.end214
>
>if.end214:                                        ; preds = %if.then202, %if.then198, %if.then209, %for.body195
>  tail call void @llvm.r600.group.barrier() #13
>  %inc216 = add nuw i32 %i190.0499, 1
>  %cmp192 = icmp ult i32 %inc216, %div.i486
>  br i1 %cmp192, label %for.body195, label %for.cond.cleanup194
>
>if.then220:                                       ; preds = %for.cond.cleanup194
>  %55 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx, align 16, !tbaa !14
>  br i1 %cmp224, label %if.then226, label %if.end228
>
>if.then226:                                       ; preds = %if.then220
>  %56 = load float, float addrspace(3)* %arrayidx187, align 4, !tbaa !12
>  br label %if.end228
>
>if.end228:                                        ; preds = %if.then226, %if.then220
>  %gamma223.0 = phi float [ %56, %if.then226 ], [ 0.000000e+00, %if.then220 ]
>  %57 = load float, float addrspace(3)* %arrayidx183, align 4, !tbaa !12
>  %cmp231 = fcmp une float %57, 0.000000e+00
>  %cmp233 = fcmp une float %57, 1.000000e+00
>  %or.cond259 = and i1 %cmp231, %cmp233
>  br i1 %or.cond259, label %if.then235, label %if.end245
>
>if.then235:                                       ; preds = %if.end228
>  %cmp.i464 = fcmp olt float %57, 0.000000e+00
>  %cond.i465 = select i1 %cmp.i464, float -1.000000e+00, float 1.000000e+00
>  %mul.i466 = fmul float %57, %cond.i465
>  %cmp1.i467 = fcmp ult float %mul.i466, 0.000000e+00
>  %div.i468 = fdiv float 1.000000e+00, %57
>  %mul2.i469 = fmul float %cond.i465, 0x7FF0000000000000
>  %cond3.i470 = select i1 %cmp1.i467, float %mul2.i469, float %div.i468
>  %splat.splatinsert237 = insertelement <4 x float> undef, float %cond3.i470, i32 0
>  %splat.splat238 = shufflevector <4 x float> %splat.splatinsert237, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul239 = fmul <4 x float> %55, %splat.splat238
>  %mul243 = fmul float %gamma223.0, %cond3.i470
>  %spec.select462 = select i1 %cmp224, float %mul243, float %gamma223.0
>  br label %if.end245
>
>if.end245:                                        ; preds = %if.then235, %if.end228
>  %filteredPixel221.0 = phi <4 x float> [ %55, %if.end228 ], [ %mul239, %if.then235 ]
>  %gamma223.1 = phi float [ %gamma223.0, %if.end228 ], [ %spec.select462, %if.then235 ]
>  br i1 %cmp224, label %if.then248, label %if.end253
>
>if.then248:                                       ; preds = %if.end245
>  %cmp.i463 = fcmp olt float %gamma223.1, 0.000000e+00
>  %cond.i = select i1 %cmp.i463, float -1.000000e+00, float 1.000000e+00
>  %mul.i = fmul float %gamma223.1, %cond.i
>  %cmp1.i = fcmp ult float %mul.i, 0.000000e+00
>  %div.i = fdiv float 1.000000e+00, %gamma223.1
>  %mul2.i = fmul float %cond.i, 0x7FF0000000000000
>  %cond3.i = select i1 %cmp1.i, float %mul2.i, float %div.i
>  %58 = extractelement <4 x float> %filteredPixel221.0, i32 0
>  %mul250 = fmul float %58, %cond3.i
>  %59 = insertelement <4 x float> %filteredPixel221.0, float %mul250, i32 0
>  %60 = extractelement <4 x float> %filteredPixel221.0, i32 1
>  %mul251 = fmul float %60, %cond3.i
>  %61 = insertelement <4 x float> %59, float %mul251, i32 1
>  %62 = extractelement <4 x float> %filteredPixel221.0, i32 2
>  %mul252 = fmul float %62, %cond3.i
>  %63 = insertelement <4 x float> %61, float %mul252, i32 2
>  br label %if.end253
>
>if.end253:                                        ; preds = %if.then248, %if.end245
>  %filteredPixel221.1 = phi <4 x float> [ %63, %if.then248 ], [ %filteredPixel221.0, %if.end245 ]
>  %add254 = add i32 %add73, %17
>  %mul1.i.i = mul i32 %add254, %filteredColumns
>  %mul4.i.i = add i32 %mul1.i.i, %9
>  %add.i.i = mul i32 %mul4.i.i, %number_channels
>  %add.ptr.i = getelementptr inbounds float, float addrspace(1)* %filteredImage, i32 %add.i.i
>  %64 = extractelement <4 x float> %filteredPixel221.1, i32 0
>  %65 = fcmp ogt float %64, 6.553500e+04
>  %66 = fcmp olt float %64, 0.000000e+00
>  %67 = select i1 %66, float 0.000000e+00, float %64
>  %.op.i27.i = fadd float %67, 5.000000e-01
>  %add.i28.i = select i1 %65, float 6.553550e+04, float %.op.i27.i
>  store float %add.i28.i, float addrspace(1)* %add.ptr.i, align 4, !tbaa !12
>  br i1 %cmp153, label %if.then.i, label %if.end.i
>
>if.then.i:                                        ; preds = %if.end253
>  %68 = extractelement <4 x float> %filteredPixel221.1, i32 1
>  %69 = fcmp ogt float %68, 6.553500e+04
>  %70 = fcmp olt float %68, 0.000000e+00
>  %71 = select i1 %70, float 0.000000e+00, float %68
>  %.op.i25.i = fadd float %71, 5.000000e-01
>  %add.i26.i = select i1 %69, float 6.553550e+04, float %.op.i25.i
>  %add.ptr.i24.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 1
>  store float %add.i26.i, float addrspace(1)* %add.ptr.i24.i, align 4, !tbaa !12
>  %72 = extractelement <4 x float> %filteredPixel221.1, i32 2
>  %73 = fcmp ogt float %72, 6.553500e+04
>  %74 = fcmp olt float %72, 0.000000e+00
>  %75 = select i1 %74, float 0.000000e+00, float %72
>  %.op.i22.i = fadd float %75, 5.000000e-01
>  %add.i23.i = select i1 %73, float 6.553550e+04, float %.op.i22.i
>  %add.ptr.i21.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 2
>  store float %add.i23.i, float addrspace(1)* %add.ptr.i21.i, align 4, !tbaa !12
>  br label %if.end.i
>
>if.end.i:                                         ; preds = %if.then.i, %if.end253
>  switch i32 %number_channels, label %if.end255 [
>    i32 4, label %if.then6.i
>    i32 2, label %if.then6.i
>  ]
>
>if.then6.i:                                       ; preds = %if.end.i, %if.end.i
>  %76 = extractelement <4 x float> %filteredPixel221.1, i32 3
>  %77 = fcmp ogt float %76, 6.553500e+04
>  %78 = fcmp olt float %76, 0.000000e+00
>  %79 = select i1 %78, float 0.000000e+00, float %76
>  %.op.i.i = fadd float %79, 5.000000e-01
>  %add.i20.i = select i1 %77, float 6.553550e+04, float %.op.i.i
>  %add.ptr.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 %number_channels
>  %add.ptr1.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i, i32 -1
>  store float %add.i20.i, float addrspace(1)* %add.ptr1.i.i, align 4, !tbaa !12
>  br label %if.end255
>
>if.end255:                                        ; preds = %if.then6.i, %if.end.i, %for.cond.cleanup194
>  %inc257 = add nuw i32 %chunk.0501, 1
>  %cmp68 = icmp ult i32 %inc257, %div66
>  br i1 %cmp68, label %for.body71, label %for.cond.cleanup70
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @RotationalBlur(float addrspace(1)* nocapture readonly %image, i32 %number_channels, i32 %channel, <2 x float> %blurCenter, float addrspace(4)* nocapture readonly %cos_theta, float addrspace(4)* nocapture readonly %sin_theta, i32 %cossin_theta_size, float addrspace(1)* nocapture %filteredImage) local_unnamed_addr #4 !kernel_arg_addr_space !85 !kernel_arg_access_qual !86 !kernel_arg_type !87 !kernel_arg_base_type !88 !kernel_arg_type_qual !89 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %x.i = tail call i32 @llvm.r600.read.global.size.x() #16
>  %y.i = tail call i32 @llvm.r600.read.global.size.y() #16
>  %conv = sitofp i32 %8 to float
>  %17 = extractelement <2 x float> %blurCenter, i32 0
>  %sub = fsub float %conv, %17
>  %conv4 = sitofp i32 %16 to float
>  %18 = extractelement <2 x float> %blurCenter, i32 1
>  %sub5 = fsub float %conv4, %18
>  %19 = fmul float %sub5, %sub5
>  %20 = tail call float @llvm.fmuladd.f32(float %sub, float %sub, float %19) #13
>  %21 = tail call float @llvm.sqrt.f32(float %20) #14
>  %22 = fmul float %18, %18
>  %23 = tail call float @llvm.fmuladd.f32(float %17, float %17, float %22) #13
>  %24 = tail call float @llvm.sqrt.f32(float %23) #14
>  %cmp = fcmp ogt float %21, 0.000000e+00
>  br i1 %cmp, label %if.then, label %if.end18
>
>if.then:                                          ; preds = %entry
>  %div = fdiv float %24, %21, !fpmath !11
>  %conv9 = fptoui float %div to i32
>  %cmp10 = icmp eq i32 %conv9, 0
>  %spec.store.select = select i1 %cmp10, i32 1, i32 %conv9
>  %cmp13 = icmp ult i32 %spec.store.select, %cossin_theta_size
>  %sub16 = add i32 %cossin_theta_size, -1
>  %spec.select = select i1 %cmp13, i32 %spec.store.select, i32 %sub16
>  br label %if.end18
>
>if.end18:                                         ; preds = %if.then, %entry
>  %step.0 = phi i32 [ 1, %entry ], [ %spec.select, %if.then ]
>  %cmp19153 = icmp eq i32 %cossin_theta_size, 0
>  br i1 %cmp19153, label %for.cond.cleanup, label %for.body.lr.ph
>
>for.body.lr.ph:                                   ; preds = %if.end18
>  %neg = fsub float -0.000000e+00, %sub5
>  %sub.i145 = add nsw i32 %x.i, -1
>  %sub.i = add nsw i32 %y.i, -1
>  %cmp.i144 = icmp ugt i32 %number_channels, 2
>  br label %for.body
>
>for.cond.cleanup:                                 ; preds = %if.end41, %if.end18
>  %gamma.0.lcssa = phi float [ 0.000000e+00, %if.end18 ], [ %gamma.1, %if.end41 ]
>  %normalize.0.lcssa = phi float [ 0.000000e+00, %if.end18 ], [ %add42, %if.end41 ]
>  %result.0.lcssa = phi <4 x float> [ zeroinitializer, %if.end18 ], [ %result.1, %if.end41 ]
>  %cmp.i146 = fcmp olt float %normalize.0.lcssa, 0.000000e+00
>  %cond.i147 = select i1 %cmp.i146, float -1.000000e+00, float 1.000000e+00
>  %mul.i148 = fmul float %normalize.0.lcssa, %cond.i147
>  %cmp1.i149 = fcmp ult float %mul.i148, 0.000000e+00
>  %div.i150 = fdiv float 1.000000e+00, %normalize.0.lcssa
>  %mul2.i151 = fmul float %cond.i147, 0x7FF0000000000000
>  %cond3.i152 = select i1 %cmp1.i149, float %mul2.i151, float %div.i150
>  switch i32 %number_channels, label %if.else56 [
>    i32 4, label %if.then50
>    i32 2, label %if.then50
>  ]
>
>for.body:                                         ; preds = %for.body.lr.ph, %if.end41
>  %result.0157 = phi <4 x float> [ zeroinitializer, %for.body.lr.ph ], [ %result.1, %if.end41 ]
>  %normalize.0156 = phi float [ 0.000000e+00, %for.body.lr.ph ], [ %add42, %if.end41 ]
>  %gamma.0155 = phi float [ 0.000000e+00, %for.body.lr.ph ], [ %gamma.1, %if.end41 ]
>  %i.0154 = phi i32 [ 0, %for.body.lr.ph ], [ %add43, %if.end41 ]
>  %arrayidx = getelementptr inbounds float, float addrspace(4)* %cos_theta, i32 %i.0154
>  %25 = load float, float addrspace(4)* %arrayidx, align 4, !tbaa !12
>  %26 = tail call float @llvm.fmuladd.f32(float %sub, float %25, float %17)
>  %arrayidx21 = getelementptr inbounds float, float addrspace(4)* %sin_theta, i32 %i.0154
>  %27 = load float, float addrspace(4)* %arrayidx21, align 4, !tbaa !12
>  %28 = tail call float @llvm.fmuladd.f32(float %neg, float %27, float %26)
>  %add = fadd float %28, 5.000000e-01
>  %conv22 = fptosi float %add to i32
>  %29 = icmp sgt i32 %x.i, %conv22
>  %30 = icmp sgt i32 %conv22, 0
>  %31 = select i1 %30, i32 %conv22, i32 0
>  %32 = select i1 %29, i32 %31, i32 %sub.i145
>  %33 = tail call float @llvm.fmuladd.f32(float %sub, float %27, float %18)
>  %34 = tail call float @llvm.fmuladd.f32(float %sub5, float %25, float %33)
>  %add26 = fadd float %34, 5.000000e-01
>  %conv27 = fptosi float %add26 to i32
>  %35 = icmp sgt i32 %y.i, %conv27
>  %36 = icmp sgt i32 %conv27, 0
>  %37 = select i1 %36, i32 %conv27, i32 0
>  %38 = select i1 %35, i32 %37, i32 %sub.i
>  %mul1.i.i140 = mul i32 %38, %x.i
>  %mul4.i.i141 = add i32 %mul1.i.i140, %32
>  %add.i.i142 = mul i32 %mul4.i.i141, %number_channels
>  %add.ptr.i143 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i142
>  %39 = load float, float addrspace(1)* %add.ptr.i143, align 4, !tbaa !12
>  %40 = insertelement <4 x float> undef, float %39, i32 0
>  br i1 %cmp.i144, label %if.then.i, label %if.end.i
>
>if.then.i:                                        ; preds = %for.body
>  %add.ptr.i23.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i143, i32 1
>  %41 = load float, float addrspace(1)* %add.ptr.i23.i, align 4, !tbaa !12
>  %42 = insertelement <4 x float> %40, float %41, i32 1
>  %add.ptr.i22.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i143, i32 2
>  %43 = load float, float addrspace(1)* %add.ptr.i22.i, align 4, !tbaa !12
>  %44 = insertelement <4 x float> %42, float %43, i32 2
>  br label %if.end.i
>
>if.end.i:                                         ; preds = %if.then.i, %for.body
>  %pixel.0.i = phi <4 x float> [ %44, %if.then.i ], [ %40, %for.body ]
>  switch i32 %number_channels, label %ReadAllChannels.exit [
>    i32 4, label %if.then6.i
>    i32 2, label %if.then6.i
>  ]
>
>if.then6.i:                                       ; preds = %if.end.i, %if.end.i
>  %add.ptr.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i143, i32 %number_channels
>  %add.ptr1.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i, i32 -1
>  %45 = load float, float addrspace(1)* %add.ptr1.i.i, align 4, !tbaa !12
>  %46 = insertelement <4 x float> %pixel.0.i, float %45, i32 3
>  br label %ReadAllChannels.exit
>
>ReadAllChannels.exit:                             ; preds = %if.end.i, %if.then6.i
>  %pixel.1.i = phi <4 x float> [ %46, %if.then6.i ], [ %pixel.0.i, %if.end.i ]
>  switch i32 %number_channels, label %if.else [
>    i32 4, label %if.then34
>    i32 2, label %if.then34
>  ]
>
>if.then34:                                        ; preds = %ReadAllChannels.exit, %ReadAllChannels.exit
>  %47 = extractelement <4 x float> %pixel.1.i, i32 3
>  %mul = fmul float %47, 0x3EEF751040000000
>  %add35 = fadd float %gamma.0155, %mul
>  %48 = extractelement <4 x float> %pixel.1.i, i32 0
>  %49 = extractelement <4 x float> %result.0157, i32 0
>  %50 = tail call float @llvm.fmuladd.f32(float %mul, float %48, float %49)
>  %51 = insertelement <4 x float> undef, float %50, i32 0
>  %52 = extractelement <4 x float> %pixel.1.i, i32 1
>  %53 = extractelement <4 x float> %result.0157, i32 1
>  %54 = tail call float @llvm.fmuladd.f32(float %mul, float %52, float %53)
>  %55 = insertelement <4 x float> %51, float %54, i32 1
>  %56 = extractelement <4 x float> %pixel.1.i, i32 2
>  %57 = extractelement <4 x float> %result.0157, i32 2
>  %58 = tail call float @llvm.fmuladd.f32(float %mul, float %56, float %57)
>  %59 = insertelement <4 x float> %55, float %58, i32 2
>  %60 = extractelement <4 x float> %result.0157, i32 3
>  %add39 = fadd float %60, %47
>  %61 = insertelement <4 x float> %59, float %add39, i32 3
>  br label %if.end41
>
>if.else:                                          ; preds = %ReadAllChannels.exit
>  %add40 = fadd <4 x float> %result.0157, %pixel.1.i
>  br label %if.end41
>
>if.end41:                                         ; preds = %if.else, %if.then34
>  %gamma.1 = phi float [ %add35, %if.then34 ], [ %gamma.0155, %if.else ]
>  %result.1 = phi <4 x float> [ %61, %if.then34 ], [ %add40, %if.else ]
>  %add42 = fadd float %normalize.0156, 1.000000e+00
>  %add43 = add i32 %i.0154, %step.0
>  %cmp19 = icmp ult i32 %add43, %cossin_theta_size
>  br i1 %cmp19, label %for.body, label %for.cond.cleanup
>
>if.then50:                                        ; preds = %for.cond.cleanup, %for.cond.cleanup
>  %cmp.i = fcmp olt float %gamma.0.lcssa, 0.000000e+00
>  %cond.i = select i1 %cmp.i, float -1.000000e+00, float 1.000000e+00
>  %mul.i = fmul float %gamma.0.lcssa, %cond.i
>  %cmp1.i = fcmp ult float %mul.i, 0.000000e+00
>  %div.i = fdiv float 1.000000e+00, %gamma.0.lcssa
>  %mul2.i = fmul float %cond.i, 0x7FF0000000000000
>  %cond3.i = select i1 %cmp1.i, float %mul2.i, float %div.i
>  %62 = extractelement <4 x float> %result.0.lcssa, i32 0
>  %mul52 = fmul float %62, %cond3.i
>  %63 = insertelement <4 x float> undef, float %mul52, i32 0
>  %64 = extractelement <4 x float> %result.0.lcssa, i32 1
>  %mul53 = fmul float %64, %cond3.i
>  %65 = insertelement <4 x float> %63, float %mul53, i32 1
>  %66 = extractelement <4 x float> %result.0.lcssa, i32 2
>  %mul54 = fmul float %66, %cond3.i
>  %67 = insertelement <4 x float> %65, float %mul54, i32 2
>  %68 = extractelement <4 x float> %result.0.lcssa, i32 3
>  %mul55 = fmul float %68, %cond3.i152
>  %69 = insertelement <4 x float> %67, float %mul55, i32 3
>  br label %if.end58
>
>if.else56:                                        ; preds = %for.cond.cleanup
>  %splat.splatinsert = insertelement <4 x float> undef, float %cond3.i152, i32 0
>  %splat.splat = shufflevector <4 x float> %splat.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul57 = fmul <4 x float> %result.0.lcssa, %splat.splat
>  br label %if.end58
>
>if.end58:                                         ; preds = %if.else56, %if.then50
>  %result.2 = phi <4 x float> [ %69, %if.then50 ], [ %mul57, %if.else56 ]
>  %mul1.i.i = mul i32 %16, %x.i
>  %mul4.i.i = add i32 %mul1.i.i, %8
>  %add.i.i = mul i32 %mul4.i.i, %number_channels
>  %add.ptr.i = getelementptr inbounds float, float addrspace(1)* %filteredImage, i32 %add.i.i
>  %70 = extractelement <4 x float> %result.2, i32 0
>  %71 = extractelement <4 x float> %result.2, i32 1
>  %72 = extractelement <4 x float> %result.2, i32 2
>  %73 = extractelement <4 x float> %result.2, i32 3
>  %and.i.i = and i32 %channel, 1
>  %cmp.i.i = icmp eq i32 %and.i.i, 0
>  br i1 %cmp.i.i, label %if.end.i.i, label %if.then.i.i
>
>if.then.i.i:                                      ; preds = %if.end58
>  %74 = fcmp ogt float %70, 6.553500e+04
>  %75 = fcmp olt float %70, 0.000000e+00
>  %76 = select i1 %75, float 0.000000e+00, float %70
>  %.op.i.i.i = fadd float %76, 5.000000e-01
>  %add.i.i.i = select i1 %74, float 6.553550e+04, float %.op.i.i.i
>  store float %add.i.i.i, float addrspace(1)* %add.ptr.i, align 4, !tbaa !12
>  br label %if.end.i.i
>
>if.end.i.i:                                       ; preds = %if.then.i.i, %if.end58
>  %cmp1.i.i = icmp ugt i32 %number_channels, 2
>  br i1 %cmp1.i.i, label %if.then2.i.i, label %if.end13.i.i
>
>if.then2.i.i:                                     ; preds = %if.end.i.i
>  %and3.i.i = and i32 %channel, 2
>  %cmp4.i.i = icmp eq i32 %and3.i.i, 0
>  br i1 %cmp4.i.i, label %if.end7.i.i, label %if.then5.i.i
>
>if.then5.i.i:                                     ; preds = %if.then2.i.i
>  %add.ptr.i35.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 1
>  %77 = fcmp ogt float %71, 6.553500e+04
>  %78 = fcmp olt float %71, 0.000000e+00
>  %79 = select i1 %78, float 0.000000e+00, float %71
>  %.op.i36.i.i = fadd float %79, 5.000000e-01
>  %add.i37.i.i = select i1 %77, float 6.553550e+04, float %.op.i36.i.i
>  store float %add.i37.i.i, float addrspace(1)* %add.ptr.i35.i.i, align 4, !tbaa !12
>  br label %if.end7.i.i
>
>if.end7.i.i:                                      ; preds = %if.then5.i.i, %if.then2.i.i
>  %and8.i.i = and i32 %channel, 4
>  %cmp9.i.i = icmp eq i32 %and8.i.i, 0
>  br i1 %cmp9.i.i, label %if.end13.i.i, label %if.then10.i.i
>
>if.then10.i.i:                                    ; preds = %if.end7.i.i
>  %add.ptr.i32.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 2
>  %80 = fcmp ogt float %72, 6.553500e+04
>  %81 = fcmp olt float %72, 0.000000e+00
>  %82 = select i1 %81, float 0.000000e+00, float %72
>  %.op.i33.i.i = fadd float %82, 5.000000e-01
>  %add.i34.i.i = select i1 %80, float 6.553550e+04, float %.op.i33.i.i
>  store float %add.i34.i.i, float addrspace(1)* %add.ptr.i32.i.i, align 4, !tbaa !12
>  br label %if.end13.i.i
>
>if.end13.i.i:                                     ; preds = %if.then10.i.i, %if.end7.i.i, %if.end.i.i
>  switch i32 %number_channels, label %WriteFloat4.exit [
>    i32 4, label %land.lhs.true.i.i
>    i32 2, label %land.lhs.true.i.i
>  ]
>
>land.lhs.true.i.i:                                ; preds = %if.end13.i.i, %if.end13.i.i
>  %and16.i.i = and i32 %channel, 16
>  %cmp17.i.i = icmp eq i32 %and16.i.i, 0
>  br i1 %cmp17.i.i, label %WriteFloat4.exit, label %if.then18.i.i
>
>if.then18.i.i:                                    ; preds = %land.lhs.true.i.i
>  %add.ptr.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 %number_channels
>  %add.ptr1.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i, i32 -1
>  %83 = fcmp ogt float %73, 6.553500e+04
>  %84 = fcmp olt float %73, 0.000000e+00
>  %85 = select i1 %84, float 0.000000e+00, float %73
>  %.op.i30.i.i = fadd float %85, 5.000000e-01
>  %add.i31.i.i = select i1 %83, float 6.553550e+04, float %.op.i30.i.i
>  store float %add.i31.i.i, float addrspace(1)* %add.ptr1.i.i.i, align 4, !tbaa !12
>  br label %WriteFloat4.exit
>
>WriteFloat4.exit:                                 ; preds = %if.end13.i.i, %land.lhs.true.i.i, %if.then18.i.i
>  ret void
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @UnsharpMaskBlurColumn(float addrspace(1)* nocapture readonly %image, <4 x float> addrspace(1)* nocapture readonly %blurRowData, i32 %number_channels, i32 %channel, i32 %columns, i32 %rows, <4 x float> addrspace(3)* nocapture %cachedData, float addrspace(3)* nocapture %cachedFilter, float addrspace(1)* nocapture readonly %filter, i32 %width, float %gain, float %threshold, float addrspace(1)* nocapture %filteredImage) local_unnamed_addr #4 !kernel_arg_addr_space !90 !kernel_arg_access_qual !91 !kernel_arg_type !92 !kernel_arg_base_type !93 !kernel_arg_type_qual !94 {
>entry:
>  %sub = add i32 %width, -1
>  %div = lshr i32 %sub, 1
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %1 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %mul = mul i32 %1, %y.i
>  %sub3 = sub i32 %mul, %div
>  %add = add i32 %1, 1
>  %mul6 = mul i32 %add, %y.i
>  %add7 = add i32 %div, %mul6
>  %cmp = icmp sgt i32 %sub3, -1
>  %cmp8 = icmp ult i32 %add7, %rows
>  %or.cond = and i1 %cmp, %cmp8
>  br i1 %or.cond, label %if.then, label %if.else
>
>if.then:                                          ; preds = %entry
>  %mul9 = mul i32 %sub3, %columns
>  %add.ptr = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %blurRowData, i32 %mul9
>  %add.ptr10 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr, i32 %0
>  %sub11 = sub nsw i32 %add7, %sub3
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %z.i.i = tail call i32 @llvm.r600.read.local.size.z() #13
>  %2 = mul i32 %z.i.i, %y.i
>  %3 = mul i32 %2, %x.i.i
>  %4 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %5 = mul i32 %4, %y.i
>  %6 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %7 = tail call i32 @llvm.r600.read.tidig.z() #13, !range !3
>  %reass.add.i = add i32 %5, %6
>  %reass.mul.i = mul i32 %reass.add.i, %z.i.i
>  %8 = add i32 %reass.mul.i, %7
>  %9 = icmp ult i32 %8, %sub11
>  br i1 %9, label %.preheader.i, label %_Z29async_work_group_strided_copyPU3AS3Dv4_fPU3AS1KS_jj9ocl_event.exit
>
>.preheader.i:                                     ; preds = %if.then, %.preheader.i
>  %10 = phi i32 [ %15, %.preheader.i ], [ %8, %if.then ]
>  %11 = mul i32 %10, %columns
>  %12 = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %add.ptr10, i32 %11
>  %13 = load <4 x float>, <4 x float> addrspace(1)* %12, align 16, !tbaa !14
>  %14 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %10
>  store <4 x float> %13, <4 x float> addrspace(3)* %14, align 16, !tbaa !14
>  %15 = add i32 %10, %3
>  %16 = icmp ult i32 %15, %sub11
>  br i1 %16, label %.preheader.i, label %_Z29async_work_group_strided_copyPU3AS3Dv4_fPU3AS1KS_jj9ocl_event.exit
>
>_Z29async_work_group_strided_copyPU3AS3Dv4_fPU3AS1KS_jj9ocl_event.exit: ; preds = %.preheader.i, %if.then
>  tail call void @llvm.r600.group.barrier() #13
>  br label %if.end
>
>if.else:                                          ; preds = %entry
>  %17 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %sub14 = sub nsw i32 %add7, %sub3
>  %cmp15172 = icmp slt i32 %17, %sub14
>  br i1 %cmp15172, label %for.body.lr.ph, label %for.cond.cleanup
>
>for.body.lr.ph:                                   ; preds = %if.else
>  %sub.i = add nsw i32 %rows, -1
>  br label %for.body
>
>for.cond.cleanup:                                 ; preds = %for.body, %if.else
>  tail call void @llvm.r600.group.barrier() #13
>  %.pre = tail call i32 @llvm.r600.read.local.size.x() #13
>  %.pre176 = tail call i32 @llvm.r600.read.local.size.z() #13
>  %.pre177 = mul i32 %.pre176, %y.i
>  %.pre178 = mul i32 %.pre177, %.pre
>  %.pre180 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %.pre182 = mul i32 %.pre180, %y.i
>  %.pre185 = tail call i32 @llvm.r600.read.tidig.z() #13, !range !3
>  %.pre187 = add i32 %.pre182, %17
>  %.pre188 = mul i32 %.pre187, %.pre176
>  %.pre189 = add i32 %.pre188, %.pre185
>  br label %if.end
>
>for.body:                                         ; preds = %for.body.lr.ph, %for.body
>  %i.0173 = phi i32 [ %17, %for.body.lr.ph ], [ %add22, %for.body ]
>  %add16 = add nsw i32 %i.0173, %sub3
>  %18 = icmp slt i32 %add16, %rows
>  %19 = icmp sgt i32 %add16, 0
>  %20 = select i1 %19, i32 %add16, i32 0
>  %21 = select i1 %18, i32 %20, i32 %sub.i
>  %mul18 = mul i32 %21, %columns
>  %add19 = add i32 %mul18, %0
>  %arrayidx = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %blurRowData, i32 %add19
>  %22 = load <4 x float>, <4 x float> addrspace(1)* %arrayidx, align 16, !tbaa !14
>  %arrayidx20 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %i.0173
>  store <4 x float> %22, <4 x float> addrspace(3)* %arrayidx20, align 16, !tbaa !14
>  %add22 = add i32 %i.0173, %y.i
>  %cmp15 = icmp slt i32 %add22, %sub14
>  br i1 %cmp15, label %for.body, label %for.cond.cleanup
>
>if.end:                                           ; preds = %for.cond.cleanup, %_Z29async_work_group_strided_copyPU3AS3Dv4_fPU3AS1KS_jj9ocl_event.exit
>  %.pre-phi190 = phi i32 [ %.pre189, %for.cond.cleanup ], [ %8, %_Z29async_work_group_strided_copyPU3AS3Dv4_fPU3AS1KS_jj9ocl_event.exit ]
>  %.pre-phi184 = phi i32 [ %17, %for.cond.cleanup ], [ %6, %_Z29async_work_group_strided_copyPU3AS3Dv4_fPU3AS1KS_jj9ocl_event.exit ]
>  %.pre-phi179 = phi i32 [ %.pre178, %for.cond.cleanup ], [ %3, %_Z29async_work_group_strided_copyPU3AS3Dv4_fPU3AS1KS_jj9ocl_event.exit ]
>  %23 = icmp ult i32 %.pre-phi190, %width
>  br i1 %23, label %.preheader.i.i, label %_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit
>
>.preheader.i.i:                                   ; preds = %if.end, %.preheader.i.i
>  %24 = phi i32 [ %30, %.preheader.i.i ], [ %.pre-phi190, %if.end ]
>  %25 = getelementptr inbounds float, float addrspace(1)* %filter, i32 %24
>  %26 = bitcast float addrspace(1)* %25 to i32 addrspace(1)*
>  %27 = load i32, i32 addrspace(1)* %26, align 4, !tbaa !12
>  %28 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %24
>  %29 = bitcast float addrspace(3)* %28 to i32 addrspace(3)*
>  store i32 %27, i32 addrspace(3)* %29, align 4, !tbaa !12
>  %30 = add i32 %24, %.pre-phi179
>  %31 = icmp ult i32 %30, %width
>  br i1 %31, label %.preheader.i.i, label %_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit
>
>_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit: ; preds = %.preheader.i.i, %if.end
>  tail call void @llvm.r600.group.barrier() #13
>  %32 = add i32 %mul, %.pre-phi184
>  %33 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %34 = getelementptr inbounds i8, i8 addrspace(7)* %33, i32 8
>  %35 = bitcast i8 addrspace(7)* %34 to i32 addrspace(7)*
>  %36 = load i32, i32 addrspace(7)* %35, align 4, !tbaa !4
>  %37 = add i32 %32, %36
>  %cmp26 = icmp ult i32 %37, %rows
>  br i1 %cmp26, label %for.cond29.preheader, label %if.end72
>
>for.cond29.preheader:                             ; preds = %_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit
>  %cmp31167 = icmp ugt i32 %width, 7
>  br i1 %cmp31167, label %for.cond33.preheader, label %for.cond48.preheader
>
>for.cond33.preheader:                             ; preds = %for.cond29.preheader, %for.cond33.preheader
>  %blurredPixel.0169 = phi <4 x float> [ %62, %for.cond33.preheader ], [ zeroinitializer, %for.cond29.preheader ]
>  %i28.0168 = phi i32 [ %63, %for.cond33.preheader ], [ 0, %for.cond29.preheader ]
>  %38 = or i32 %i28.0168, 7
>  %arrayidx38 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %i28.0168
>  %39 = load float, float addrspace(3)* %arrayidx38, align 4, !tbaa !12
>  %splat.splatinsert = insertelement <4 x float> undef, float %39, i32 0
>  %splat.splat = shufflevector <4 x float> %splat.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
>  %add41 = add i32 %i28.0168, %.pre-phi184
>  %arrayidx42 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %add41
>  %40 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx42, align 16, !tbaa !14
>  %41 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat, <4 x float> %40, <4 x float> %blurredPixel.0169)
>  %inc45 = or i32 %i28.0168, 1
>  %add37.1 = add nsw i32 %inc45, 1
>  %arrayidx38.1 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %add37.1
>  %42 = load float, float addrspace(3)* %arrayidx38.1, align 4, !tbaa !12
>  %splat.splatinsert.1 = insertelement <4 x float> undef, float %42, i32 0
>  %splat.splat.1 = shufflevector <4 x float> %splat.splatinsert.1, <4 x float> undef, <4 x i32> zeroinitializer
>  %add41.1 = add i32 %add37.1, %.pre-phi184
>  %arrayidx42.1 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %add41.1
>  %43 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx42.1, align 16, !tbaa !14
>  %44 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.1, <4 x float> %43, <4 x float> %41)
>  %add37.2 = add nsw i32 %inc45, 3
>  %arrayidx38.2 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %add37.2
>  %45 = load float, float addrspace(3)* %arrayidx38.2, align 4, !tbaa !12
>  %splat.splatinsert.2 = insertelement <4 x float> undef, float %45, i32 0
>  %splat.splat.2 = shufflevector <4 x float> %splat.splatinsert.2, <4 x float> undef, <4 x i32> zeroinitializer
>  %add41.2 = add i32 %add37.2, %.pre-phi184
>  %arrayidx42.2 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %add41.2
>  %46 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx42.2, align 16, !tbaa !14
>  %47 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.2, <4 x float> %46, <4 x float> %44)
>  %inc45.2 = or i32 %i28.0168, 3
>  %add37.3 = add nsw i32 %inc45.2, 3
>  %arrayidx38.3 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %add37.3
>  %48 = load float, float addrspace(3)* %arrayidx38.3, align 4, !tbaa !12
>  %splat.splatinsert.3 = insertelement <4 x float> undef, float %48, i32 0
>  %splat.splat.3 = shufflevector <4 x float> %splat.splatinsert.3, <4 x float> undef, <4 x i32> zeroinitializer
>  %add41.3 = add i32 %add37.3, %.pre-phi184
>  %arrayidx42.3 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %add41.3
>  %49 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx42.3, align 16, !tbaa !14
>  %50 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.3, <4 x float> %49, <4 x float> %47)
>  %add37.4 = add nsw i32 %inc45.2, 5
>  %arrayidx38.4 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %add37.4
>  %51 = load float, float addrspace(3)* %arrayidx38.4, align 4, !tbaa !12
>  %splat.splatinsert.4 = insertelement <4 x float> undef, float %51, i32 0
>  %splat.splat.4 = shufflevector <4 x float> %splat.splatinsert.4, <4 x float> undef, <4 x i32> zeroinitializer
>  %add41.4 = add i32 %add37.4, %.pre-phi184
>  %arrayidx42.4 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %add41.4
>  %52 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx42.4, align 16, !tbaa !14
>  %53 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.4, <4 x float> %52, <4 x float> %50)
>  %add37.5 = add nsw i32 %inc45.2, 7
>  %arrayidx38.5 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %add37.5
>  %54 = load float, float addrspace(3)* %arrayidx38.5, align 4, !tbaa !12
>  %splat.splatinsert.5 = insertelement <4 x float> undef, float %54, i32 0
>  %splat.splat.5 = shufflevector <4 x float> %splat.splatinsert.5, <4 x float> undef, <4 x i32> zeroinitializer
>  %add41.5 = add i32 %add37.5, %.pre-phi184
>  %arrayidx42.5 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %add41.5
>  %55 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx42.5, align 16, !tbaa !14
>  %56 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.5, <4 x float> %55, <4 x float> %53)
>  %add37.6 = add nsw i32 %inc45.2, 9
>  %arrayidx38.6 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %add37.6
>  %57 = load float, float addrspace(3)* %arrayidx38.6, align 4, !tbaa !12
>  %splat.splatinsert.6 = insertelement <4 x float> undef, float %57, i32 0
>  %splat.splat.6 = shufflevector <4 x float> %splat.splatinsert.6, <4 x float> undef, <4 x i32> zeroinitializer
>  %add41.6 = add i32 %add37.6, %.pre-phi184
>  %arrayidx42.6 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %add41.6
>  %58 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx42.6, align 16, !tbaa !14
>  %59 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.6, <4 x float> %58, <4 x float> %56)
>  %inc45.6 = or i32 %i28.0168, 7
>  %add37.7 = add nsw i32 %inc45.6, 7
>  %arrayidx38.7 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %add37.7
>  %60 = load float, float addrspace(3)* %arrayidx38.7, align 4, !tbaa !12
>  %splat.splatinsert.7 = insertelement <4 x float> undef, float %60, i32 0
>  %splat.splat.7 = shufflevector <4 x float> %splat.splatinsert.7, <4 x float> undef, <4 x i32> zeroinitializer
>  %add41.7 = add i32 %add37.7, %.pre-phi184
>  %arrayidx42.7 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %add41.7
>  %61 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx42.7, align 16, !tbaa !14
>  %62 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.7, <4 x float> %61, <4 x float> %59)
>  %63 = add i32 %i28.0168, 8
>  %add30 = add nsw i32 %38, 8
>  %cmp31 = icmp ult i32 %add30, %width
>  br i1 %cmp31, label %for.cond33.preheader, label %for.cond48.preheader
>
>for.cond48.preheader:                             ; preds = %for.cond33.preheader, %for.cond29.preheader
>  %i28.0.lcssa = phi i32 [ 0, %for.cond29.preheader ], [ %63, %for.cond33.preheader ]
>  %blurredPixel.0.lcssa = phi <4 x float> [ zeroinitializer, %for.cond29.preheader ], [ %62, %for.cond33.preheader ]
>  %cmp49161 = icmp ult i32 %i28.0.lcssa, %width
>  br i1 %cmp49161, label %for.body50, label %for.end60
>
>for.body50:                                       ; preds = %for.cond48.preheader, %for.body50
>  %blurredPixel.2163 = phi <4 x float> [ %66, %for.body50 ], [ %blurredPixel.0.lcssa, %for.cond48.preheader ]
>  %i28.2162 = phi i32 [ %inc59, %for.body50 ], [ %i28.0.lcssa, %for.cond48.preheader ]
>  %arrayidx51 = getelementptr inbounds float, float addrspace(3)* %cachedFilter, i32 %i28.2162
>  %64 = load float, float addrspace(3)* %arrayidx51, align 4, !tbaa !12
>  %splat.splatinsert52 = insertelement <4 x float> undef, float %64, i32 0
>  %splat.splat53 = shufflevector <4 x float> %splat.splatinsert52, <4 x float> undef, <4 x i32> zeroinitializer
>  %add55 = add i32 %i28.2162, %.pre-phi184
>  %arrayidx56 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %cachedData, i32 %add55
>  %65 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx56, align 16, !tbaa !14
>  %66 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat53, <4 x float> %65, <4 x float> %blurredPixel.2163)
>  %inc59 = add nuw nsw i32 %i28.2162, 1
>  %exitcond = icmp eq i32 %inc59, %width
>  br i1 %exitcond, label %for.end60, label %for.body50
>
>for.end60:                                        ; preds = %for.body50, %for.cond48.preheader
>  %blurredPixel.2.lcssa = phi <4 x float> [ %blurredPixel.0.lcssa, %for.cond48.preheader ], [ %66, %for.body50 ]
>  %mul1.i.i134 = mul i32 %37, %columns
>  %mul4.i.i135 = add i32 %mul1.i.i134, %0
>  %add.i.i136 = mul i32 %mul4.i.i135, %number_channels
>  %add.ptr.i137 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i136
>  %and.i.i138 = and i32 %channel, 1
>  %cmp.i.i139 = icmp eq i32 %and.i.i138, 0
>  br i1 %cmp.i.i139, label %if.end.i.i142, label %if.then.i.i140
>
>if.then.i.i140:                                   ; preds = %for.end60
>  %67 = load float, float addrspace(1)* %add.ptr.i137, align 4, !tbaa !12
>  br label %if.end.i.i142
>
>if.end.i.i142:                                    ; preds = %if.then.i.i140, %for.end60
>  %68 = phi float [ 0.000000e+00, %for.end60 ], [ %67, %if.then.i.i140 ]
>  %cmp1.i.i141 = icmp ugt i32 %number_channels, 2
>  br i1 %cmp1.i.i141, label %if.then2.i.i145, label %if.end13.i.i151
>
>if.then2.i.i145:                                  ; preds = %if.end.i.i142
>  %and3.i.i143 = and i32 %channel, 2
>  %cmp4.i.i144 = icmp eq i32 %and3.i.i143, 0
>  br i1 %cmp4.i.i144, label %if.end7.i.i149, label %if.then5.i.i146
>
>if.then5.i.i146:                                  ; preds = %if.then2.i.i145
>  %add.ptr.i31.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i137, i32 1
>  %69 = load float, float addrspace(1)* %add.ptr.i31.i.i, align 4, !tbaa !12
>  br label %if.end7.i.i149
>
>if.end7.i.i149:                                   ; preds = %if.then5.i.i146, %if.then2.i.i145
>  %70 = phi float [ 0.000000e+00, %if.then2.i.i145 ], [ %69, %if.then5.i.i146 ]
>  %and8.i.i147 = and i32 %channel, 4
>  %cmp9.i.i148 = icmp eq i32 %and8.i.i147, 0
>  br i1 %cmp9.i.i148, label %if.end13.i.i151, label %if.then10.i.i150
>
>if.then10.i.i150:                                 ; preds = %if.end7.i.i149
>  %add.ptr.i30.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i137, i32 2
>  %71 = load float, float addrspace(1)* %add.ptr.i30.i.i, align 4, !tbaa !12
>  br label %if.end13.i.i151
>
>if.end13.i.i151:                                  ; preds = %if.then10.i.i150, %if.end7.i.i149, %if.end.i.i142
>  %72 = phi float [ %70, %if.end7.i.i149 ], [ %70, %if.then10.i.i150 ], [ 0.000000e+00, %if.end.i.i142 ]
>  %73 = phi float [ 0.000000e+00, %if.end7.i.i149 ], [ %71, %if.then10.i.i150 ], [ 0.000000e+00, %if.end.i.i142 ]
>  switch i32 %number_channels, label %ReadFloat4.exit [
>    i32 4, label %land.lhs.true.i.i154
>    i32 2, label %land.lhs.true.i.i154
>  ]
>
>land.lhs.true.i.i154:                             ; preds = %if.end13.i.i151, %if.end13.i.i151
>  %and16.i.i152 = and i32 %channel, 16
>  %cmp17.i.i153 = icmp eq i32 %and16.i.i152, 0
>  br i1 %cmp17.i.i153, label %ReadFloat4.exit, label %if.then18.i.i157
>
>if.then18.i.i157:                                 ; preds = %land.lhs.true.i.i154
>  %add.ptr.i.i.i155 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i137, i32 %number_channels
>  %add.ptr1.i.i.i156 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i155, i32 -1
>  %74 = load float, float addrspace(1)* %add.ptr1.i.i.i156, align 4, !tbaa !12
>  br label %ReadFloat4.exit
>
>ReadFloat4.exit:                                  ; preds = %if.end13.i.i151, %land.lhs.true.i.i154, %if.then18.i.i157
>  %75 = phi float [ 0.000000e+00, %if.end13.i.i151 ], [ 0.000000e+00, %land.lhs.true.i.i154 ], [ %74, %if.then18.i.i157 ]
>  %vecinit.i = insertelement <4 x float> undef, float %68, i32 0
>  %vecinit1.i = insertelement <4 x float> %vecinit.i, float %72, i32 1
>  %vecinit2.i = insertelement <4 x float> %vecinit1.i, float %73, i32 2
>  %vecinit3.i = insertelement <4 x float> %vecinit2.i, float %75, i32 3
>  %sub62 = fsub <4 x float> %vecinit3.i, %blurredPixel.2.lcssa
>  %mul63 = fmul float %threshold, 6.553500e+04
>  %mul64 = fmul <4 x float> %sub62, <float 2.000000e+00, float 2.000000e+00, float 2.000000e+00, float 2.000000e+00>
>  %call65 = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %mul64) #15
>  %splat.splatinsert66 = insertelement <4 x float> undef, float %mul63, i32 0
>  %splat.splat67 = shufflevector <4 x float> %splat.splatinsert66, <4 x float> undef, <4 x i32> zeroinitializer
>  %76 = fcmp olt <4 x float> %call65, %splat.splat67
>  %splat.splatinsert69 = insertelement <4 x float> undef, float %gain, i32 0
>  %splat.splat70 = shufflevector <4 x float> %splat.splatinsert69, <4 x float> undef, <4 x i32> zeroinitializer
>  %77 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %sub62, <4 x float> %splat.splat70, <4 x float> %vecinit3.i)
>  %.v = select <4 x i1> %76, <4 x float> %vecinit3.i, <4 x float> %77
>  %add.ptr.i = getelementptr inbounds float, float addrspace(1)* %filteredImage, i32 %add.i.i136
>  %78 = extractelement <4 x float> %.v, i32 0
>  %79 = extractelement <4 x float> %.v, i32 1
>  %80 = extractelement <4 x float> %.v, i32 2
>  %81 = extractelement <4 x float> %.v, i32 3
>  br i1 %cmp.i.i139, label %if.end.i.i, label %if.then.i.i
>
>if.then.i.i:                                      ; preds = %ReadFloat4.exit
>  %82 = fcmp ogt float %78, 6.553500e+04
>  %83 = fcmp olt float %78, 0.000000e+00
>  %84 = select i1 %83, float 0.000000e+00, float %78
>  %.op.i.i.i = fadd float %84, 5.000000e-01
>  %add.i.i.i = select i1 %82, float 6.553550e+04, float %.op.i.i.i
>  store float %add.i.i.i, float addrspace(1)* %add.ptr.i, align 4, !tbaa !12
>  br label %if.end.i.i
>
>if.end.i.i:                                       ; preds = %if.then.i.i, %ReadFloat4.exit
>  br i1 %cmp1.i.i141, label %if.then2.i.i, label %if.end13.i.i
>
>if.then2.i.i:                                     ; preds = %if.end.i.i
>  %and3.i.i = and i32 %channel, 2
>  %cmp4.i.i = icmp eq i32 %and3.i.i, 0
>  br i1 %cmp4.i.i, label %if.end7.i.i, label %if.then5.i.i
>
>if.then5.i.i:                                     ; preds = %if.then2.i.i
>  %add.ptr.i35.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 1
>  %85 = fcmp ogt float %79, 6.553500e+04
>  %86 = fcmp olt float %79, 0.000000e+00
>  %87 = select i1 %86, float 0.000000e+00, float %79
>  %.op.i36.i.i = fadd float %87, 5.000000e-01
>  %add.i37.i.i = select i1 %85, float 6.553550e+04, float %.op.i36.i.i
>  store float %add.i37.i.i, float addrspace(1)* %add.ptr.i35.i.i, align 4, !tbaa !12
>  br label %if.end7.i.i
>
>if.end7.i.i:                                      ; preds = %if.then5.i.i, %if.then2.i.i
>  %and8.i.i = and i32 %channel, 4
>  %cmp9.i.i = icmp eq i32 %and8.i.i, 0
>  br i1 %cmp9.i.i, label %if.end13.i.i, label %if.then10.i.i
>
>if.then10.i.i:                                    ; preds = %if.end7.i.i
>  %add.ptr.i32.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 2
>  %88 = fcmp ogt float %80, 6.553500e+04
>  %89 = fcmp olt float %80, 0.000000e+00
>  %90 = select i1 %89, float 0.000000e+00, float %80
>  %.op.i33.i.i = fadd float %90, 5.000000e-01
>  %add.i34.i.i = select i1 %88, float 6.553550e+04, float %.op.i33.i.i
>  store float %add.i34.i.i, float addrspace(1)* %add.ptr.i32.i.i, align 4, !tbaa !12
>  br label %if.end13.i.i
>
>if.end13.i.i:                                     ; preds = %if.then10.i.i, %if.end7.i.i, %if.end.i.i
>  switch i32 %number_channels, label %if.end72 [
>    i32 4, label %land.lhs.true.i.i
>    i32 2, label %land.lhs.true.i.i
>  ]
>
>land.lhs.true.i.i:                                ; preds = %if.end13.i.i, %if.end13.i.i
>  %and16.i.i = and i32 %channel, 16
>  %cmp17.i.i = icmp eq i32 %and16.i.i, 0
>  br i1 %cmp17.i.i, label %if.end72, label %if.then18.i.i
>
>if.then18.i.i:                                    ; preds = %land.lhs.true.i.i
>  %add.ptr.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 %number_channels
>  %add.ptr1.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i, i32 -1
>  %91 = fcmp ogt float %81, 6.553500e+04
>  %92 = fcmp olt float %81, 0.000000e+00
>  %93 = select i1 %92, float 0.000000e+00, float %81
>  %.op.i30.i.i = fadd float %93, 5.000000e-01
>  %add.i31.i.i = select i1 %91, float 6.553550e+04, float %.op.i30.i.i
>  store float %add.i31.i.i, float addrspace(1)* %add.ptr1.i.i.i, align 4, !tbaa !12
>  br label %if.end72
>
>if.end72:                                         ; preds = %if.then18.i.i, %land.lhs.true.i.i, %if.end13.i.i, %_Z21async_work_group_copyPU3AS3fPU3AS1Kfj9ocl_event.exit
>  ret void
>}
>
>; Function Attrs: nounwind readnone speculatable
>declare <4 x float> @llvm.fabs.v4f32(<4 x float>) #5
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @UnsharpMask(float addrspace(1)* nocapture readonly %image, i32 %number_channels, i32 %channel, float addrspace(4)* nocapture readonly %filter, i32 %width, i32 %columns, i32 %rows, <4 x float> addrspace(3)* nocapture %pixels, float %gain, float %threshold, float addrspace(1)* nocapture %filteredImage) local_unnamed_addr #4 !kernel_arg_addr_space !95 !kernel_arg_access_qual !16 !kernel_arg_type !96 !kernel_arg_base_type !97 !kernel_arg_type_qual !98 {
>entry:
>  %0 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %x.i.i = tail call i32 @llvm.r600.read.local.size.x() #13
>  %1 = mul i32 %x.i.i, %0
>  %2 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %3 = add i32 %1, %2
>  %4 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %5 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 4
>  %6 = bitcast i8 addrspace(7)* %5 to i32 addrspace(7)*
>  %7 = load i32, i32 addrspace(7)* %6, align 4, !tbaa !4
>  %8 = add i32 %3, %7
>  %9 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %y.i.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  %10 = mul i32 %y.i.i, %9
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %12 = add i32 %10, %11
>  %13 = getelementptr inbounds i8, i8 addrspace(7)* %4, i32 8
>  %14 = bitcast i8 addrspace(7)* %13 to i32 addrspace(7)*
>  %15 = load i32, i32 addrspace(7)* %14, align 4, !tbaa !4
>  %16 = add i32 %12, %15
>  %sub = add i32 %width, -1
>  %div = lshr i32 %sub, 1
>  %sub2 = sub i32 %16, %div
>  %add = add i32 %9, 1
>  %mul8 = mul i32 %add, %y.i.i
>  %add9 = add i32 %div, %mul8
>  %cmp357 = icmp slt i32 %sub2, %add9
>  br i1 %cmp357, label %while.body.lr.ph, label %while.end80
>
>while.body.lr.ph:                                 ; preds = %entry
>  %mul14 = shl i32 %rows, 1
>  %sub15 = add i32 %mul14, -1
>  %sub20 = sub i32 %8, %div
>  %cmp23345 = icmp ugt i32 %width, 7
>  %mul35 = shl i32 %columns, 1
>  %sub36 = add i32 %mul35, -1
>  %and.i.i304 = and i32 %channel, 1
>  %cmp.i.i305 = icmp eq i32 %and.i.i304, 0
>  %cmp1.i.i307 = icmp ugt i32 %number_channels, 2
>  %and3.i.i309 = and i32 %channel, 2
>  %cmp4.i.i310 = icmp eq i32 %and3.i.i309, 0
>  %and8.i.i314 = and i32 %channel, 4
>  %cmp9.i.i315 = icmp eq i32 %and8.i.i314, 0
>  %and16.i.i320 = and i32 %channel, 16
>  %cmp17.i.i321 = icmp eq i32 %and16.i.i320, 0
>  %sub5243 = sub i32 %div, %10
>  %17 = add i32 %2, %7
>  %18 = add i32 %17, %1
>  %19 = add i32 %width, -8
>  %20 = and i32 %19, -8
>  %21 = add i32 %18, %20
>  %22 = add i32 %21, 8
>  %23 = sub i32 %22, %div
>  %24 = add i32 %20, 8
>  br label %while.body
>
>while.body:                                       ; preds = %while.body.lr.ph, %while.end71
>  %row.0358 = phi i32 [ %sub2, %while.body.lr.ph ], [ %add79, %while.end71 ]
>  %cmp10 = icmp slt i32 %row.0358, 0
>  %sub11 = sub nsw i32 0, %row.0358
>  %cond = select i1 %cmp10, i32 %sub11, i32 %row.0358
>  %cmp12 = icmp ult i32 %cond, %rows
>  %sub16 = sub i32 %sub15, %cond
>  %spec.select = select i1 %cmp12, i32 %cond, i32 %sub16
>  br i1 %cmp23345, label %for.cond.preheader.lr.ph, label %while.cond46.preheader
>
>for.cond.preheader.lr.ph:                         ; preds = %while.body
>  %mul1.i.i300 = mul i32 %spec.select, %columns
>  br label %for.cond.preheader
>
>while.cond46.preheader:                           ; preds = %ReadFloat4.exit330.7, %while.body
>  %i.0.lcssa = phi i32 [ 0, %while.body ], [ %24, %ReadFloat4.exit330.7 ]
>  %ix.0.lcssa = phi i32 [ %sub20, %while.body ], [ %23, %ReadFloat4.exit330.7 ]
>  %value.0.lcssa = phi <4 x float> [ zeroinitializer, %while.body ], [ %171, %ReadFloat4.exit330.7 ]
>  %cmp47352 = icmp ult i32 %i.0.lcssa, %width
>  br i1 %cmp47352, label %while.body48.lr.ph, label %while.end71
>
>while.body48.lr.ph:                               ; preds = %while.cond46.preheader
>  %mul1.i.i269 = mul i32 %spec.select, %columns
>  br label %while.body48
>
>for.cond.preheader:                               ; preds = %for.cond.preheader.lr.ph, %ReadFloat4.exit330.7
>  %value.0348 = phi <4 x float> [ zeroinitializer, %for.cond.preheader.lr.ph ], [ %171, %ReadFloat4.exit330.7 ]
>  %ix.0347 = phi i32 [ %sub20, %for.cond.preheader.lr.ph ], [ %add44, %ReadFloat4.exit330.7 ]
>  %i.0346 = phi i32 [ 0, %for.cond.preheader.lr.ph ], [ %add45, %ReadFloat4.exit330.7 ]
>  %cmp27 = icmp slt i32 %ix.0347, 0
>  %sub29 = sub nsw i32 0, %ix.0347
>  %cond32 = select i1 %cmp27, i32 %sub29, i32 %ix.0347
>  %cmp33 = icmp ult i32 %cond32, %columns
>  %sub37 = sub i32 %sub36, %cond32
>  %spec.select331 = select i1 %cmp33, i32 %cond32, i32 %sub37
>  %arrayidx = getelementptr inbounds float, float addrspace(4)* %filter, i32 %i.0346
>  %25 = load float, float addrspace(4)* %arrayidx, align 4, !tbaa !12
>  %splat.splatinsert = insertelement <4 x float> undef, float %25, i32 0
>  %splat.splat = shufflevector <4 x float> %splat.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul4.i.i301 = add i32 %spec.select331, %mul1.i.i300
>  %add.i.i302 = mul i32 %mul4.i.i301, %number_channels
>  %add.ptr.i303 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i302
>  br i1 %cmp.i.i305, label %if.end.i.i308, label %if.then.i.i306
>
>if.then.i.i306:                                   ; preds = %for.cond.preheader
>  %26 = load float, float addrspace(1)* %add.ptr.i303, align 4, !tbaa !12
>  br label %if.end.i.i308
>
>if.end.i.i308:                                    ; preds = %if.then.i.i306, %for.cond.preheader
>  %27 = phi float [ 0.000000e+00, %for.cond.preheader ], [ %26, %if.then.i.i306 ]
>  br i1 %cmp1.i.i307, label %if.then2.i.i311, label %if.end13.i.i319
>
>if.then2.i.i311:                                  ; preds = %if.end.i.i308
>  br i1 %cmp4.i.i310, label %if.end7.i.i316, label %if.then5.i.i313
>
>if.then5.i.i313:                                  ; preds = %if.then2.i.i311
>  %add.ptr.i31.i.i312 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303, i32 1
>  %28 = load float, float addrspace(1)* %add.ptr.i31.i.i312, align 4, !tbaa !12
>  br label %if.end7.i.i316
>
>if.end7.i.i316:                                   ; preds = %if.then5.i.i313, %if.then2.i.i311
>  %29 = phi float [ 0.000000e+00, %if.then2.i.i311 ], [ %28, %if.then5.i.i313 ]
>  br i1 %cmp9.i.i315, label %if.end13.i.i319, label %if.then10.i.i318
>
>if.then10.i.i318:                                 ; preds = %if.end7.i.i316
>  %add.ptr.i30.i.i317 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303, i32 2
>  %30 = load float, float addrspace(1)* %add.ptr.i30.i.i317, align 4, !tbaa !12
>  br label %if.end13.i.i319
>
>if.end13.i.i319:                                  ; preds = %if.then10.i.i318, %if.end7.i.i316, %if.end.i.i308
>  %31 = phi float [ %29, %if.end7.i.i316 ], [ %29, %if.then10.i.i318 ], [ 0.000000e+00, %if.end.i.i308 ]
>  %32 = phi float [ 0.000000e+00, %if.end7.i.i316 ], [ %30, %if.then10.i.i318 ], [ 0.000000e+00, %if.end.i.i308 ]
>  switch i32 %number_channels, label %ReadFloat4.exit330 [
>    i32 4, label %land.lhs.true.i.i322
>    i32 2, label %land.lhs.true.i.i322
>  ]
>
>land.lhs.true.i.i322:                             ; preds = %if.end13.i.i319, %if.end13.i.i319
>  br i1 %cmp17.i.i321, label %ReadFloat4.exit330, label %if.then18.i.i325
>
>if.then18.i.i325:                                 ; preds = %land.lhs.true.i.i322
>  %add.ptr.i.i.i323 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303, i32 %number_channels
>  %add.ptr1.i.i.i324 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i323, i32 -1
>  %33 = load float, float addrspace(1)* %add.ptr1.i.i.i324, align 4, !tbaa !12
>  br label %ReadFloat4.exit330
>
>ReadFloat4.exit330:                               ; preds = %if.end13.i.i319, %land.lhs.true.i.i322, %if.then18.i.i325
>  %34 = phi float [ 0.000000e+00, %if.end13.i.i319 ], [ 0.000000e+00, %land.lhs.true.i.i322 ], [ %33, %if.then18.i.i325 ]
>  %vecinit.i326 = insertelement <4 x float> undef, float %27, i32 0
>  %vecinit1.i327 = insertelement <4 x float> %vecinit.i326, float %31, i32 1
>  %vecinit2.i328 = insertelement <4 x float> %vecinit1.i327, float %32, i32 2
>  %vecinit3.i329 = insertelement <4 x float> %vecinit2.i328, float %34, i32 3
>  %35 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat, <4 x float> %vecinit3.i329, <4 x float> %value.0348)
>  %add26.1 = add nsw i32 %ix.0347, 1
>  %cmp27.1 = icmp slt i32 %ix.0347, -1
>  %sub29.1 = xor i32 %ix.0347, -1
>  %cond32.1 = select i1 %cmp27.1, i32 %sub29.1, i32 %add26.1
>  %cmp33.1 = icmp ult i32 %cond32.1, %columns
>  %sub37.1 = sub i32 %sub36, %cond32.1
>  %spec.select331.1 = select i1 %cmp33.1, i32 %cond32.1, i32 %sub37.1
>  %add41.1 = or i32 %i.0346, 1
>  %arrayidx.1 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add41.1
>  %36 = load float, float addrspace(4)* %arrayidx.1, align 4, !tbaa !12
>  %splat.splatinsert.1 = insertelement <4 x float> undef, float %36, i32 0
>  %splat.splat.1 = shufflevector <4 x float> %splat.splatinsert.1, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul4.i.i301.1 = add i32 %spec.select331.1, %mul1.i.i300
>  %add.i.i302.1 = mul i32 %mul4.i.i301.1, %number_channels
>  %add.ptr.i303.1 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i302.1
>  br i1 %cmp.i.i305, label %if.end.i.i308.1, label %if.then.i.i306.1
>
>while.body48:                                     ; preds = %ReadFloat4.exit299, %while.body48.lr.ph
>  %value.2355 = phi <4 x float> [ %value.0.lcssa, %while.body48.lr.ph ], [ %47, %ReadFloat4.exit299 ]
>  %ix.1354 = phi i32 [ %ix.0.lcssa, %while.body48.lr.ph ], [ %inc70, %ReadFloat4.exit299 ]
>  %i.1353 = phi i32 [ %i.0.lcssa, %while.body48.lr.ph ], [ %inc69, %ReadFloat4.exit299 ]
>  %cmp50 = icmp slt i32 %ix.1354, 0
>  %sub52 = sub nsw i32 0, %ix.1354
>  %cond55 = select i1 %cmp50, i32 %sub52, i32 %ix.1354
>  %cmp56 = icmp ult i32 %cond55, %columns
>  %sub60 = sub i32 %sub36, %cond55
>  %spec.select332 = select i1 %cmp56, i32 %cond55, i32 %sub60
>  %arrayidx64 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %i.1353
>  %37 = load float, float addrspace(4)* %arrayidx64, align 4, !tbaa !12
>  %splat.splatinsert65 = insertelement <4 x float> undef, float %37, i32 0
>  %splat.splat66 = shufflevector <4 x float> %splat.splatinsert65, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul4.i.i270 = add i32 %spec.select332, %mul1.i.i269
>  %add.i.i271 = mul i32 %mul4.i.i270, %number_channels
>  %add.ptr.i272 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i271
>  br i1 %cmp.i.i305, label %if.end.i.i277, label %if.then.i.i275
>
>if.then.i.i275:                                   ; preds = %while.body48
>  %38 = load float, float addrspace(1)* %add.ptr.i272, align 4, !tbaa !12
>  br label %if.end.i.i277
>
>if.end.i.i277:                                    ; preds = %if.then.i.i275, %while.body48
>  %39 = phi float [ 0.000000e+00, %while.body48 ], [ %38, %if.then.i.i275 ]
>  br i1 %cmp1.i.i307, label %if.then2.i.i280, label %if.end13.i.i288
>
>if.then2.i.i280:                                  ; preds = %if.end.i.i277
>  br i1 %cmp4.i.i310, label %if.end7.i.i285, label %if.then5.i.i282
>
>if.then5.i.i282:                                  ; preds = %if.then2.i.i280
>  %add.ptr.i31.i.i281 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i272, i32 1
>  %40 = load float, float addrspace(1)* %add.ptr.i31.i.i281, align 4, !tbaa !12
>  br label %if.end7.i.i285
>
>if.end7.i.i285:                                   ; preds = %if.then5.i.i282, %if.then2.i.i280
>  %41 = phi float [ 0.000000e+00, %if.then2.i.i280 ], [ %40, %if.then5.i.i282 ]
>  br i1 %cmp9.i.i315, label %if.end13.i.i288, label %if.then10.i.i287
>
>if.then10.i.i287:                                 ; preds = %if.end7.i.i285
>  %add.ptr.i30.i.i286 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i272, i32 2
>  %42 = load float, float addrspace(1)* %add.ptr.i30.i.i286, align 4, !tbaa !12
>  br label %if.end13.i.i288
>
>if.end13.i.i288:                                  ; preds = %if.then10.i.i287, %if.end7.i.i285, %if.end.i.i277
>  %43 = phi float [ %41, %if.end7.i.i285 ], [ %41, %if.then10.i.i287 ], [ 0.000000e+00, %if.end.i.i277 ]
>  %44 = phi float [ 0.000000e+00, %if.end7.i.i285 ], [ %42, %if.then10.i.i287 ], [ 0.000000e+00, %if.end.i.i277 ]
>  switch i32 %number_channels, label %ReadFloat4.exit299 [
>    i32 4, label %land.lhs.true.i.i291
>    i32 2, label %land.lhs.true.i.i291
>  ]
>
>land.lhs.true.i.i291:                             ; preds = %if.end13.i.i288, %if.end13.i.i288
>  br i1 %cmp17.i.i321, label %ReadFloat4.exit299, label %if.then18.i.i294
>
>if.then18.i.i294:                                 ; preds = %land.lhs.true.i.i291
>  %add.ptr.i.i.i292 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i272, i32 %number_channels
>  %add.ptr1.i.i.i293 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i292, i32 -1
>  %45 = load float, float addrspace(1)* %add.ptr1.i.i.i293, align 4, !tbaa !12
>  br label %ReadFloat4.exit299
>
>ReadFloat4.exit299:                               ; preds = %if.end13.i.i288, %land.lhs.true.i.i291, %if.then18.i.i294
>  %46 = phi float [ 0.000000e+00, %if.end13.i.i288 ], [ 0.000000e+00, %land.lhs.true.i.i291 ], [ %45, %if.then18.i.i294 ]
>  %vecinit.i295 = insertelement <4 x float> undef, float %39, i32 0
>  %vecinit1.i296 = insertelement <4 x float> %vecinit.i295, float %43, i32 1
>  %vecinit2.i297 = insertelement <4 x float> %vecinit1.i296, float %44, i32 2
>  %vecinit3.i298 = insertelement <4 x float> %vecinit2.i297, float %46, i32 3
>  %47 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat66, <4 x float> %vecinit3.i298, <4 x float> %value.2355)
>  %inc69 = add nuw nsw i32 %i.1353, 1
>  %inc70 = add nsw i32 %ix.1354, 1
>  %exitcond364 = icmp eq i32 %inc69, %width
>  br i1 %exitcond364, label %while.end71, label %while.body48
>
>while.end71:                                      ; preds = %ReadFloat4.exit299, %while.cond46.preheader
>  %value.2.lcssa = phi <4 x float> [ %value.0.lcssa, %while.cond46.preheader ], [ %47, %ReadFloat4.exit299 ]
>  %sub72 = add i32 %sub5243, %row.0358
>  %mul74 = mul i32 %sub72, %x.i.i
>  %add76 = add i32 %mul74, %2
>  %arrayidx77 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add76
>  store <4 x float> %value.2.lcssa, <4 x float> addrspace(3)* %arrayidx77, align 16, !tbaa !14
>  %add79 = add i32 %row.0358, %y.i.i
>  %cmp = icmp slt i32 %add79, %add9
>  br i1 %cmp, label %while.body, label %while.end80
>
>while.end80:                                      ; preds = %while.end71, %entry
>  tail call void @llvm.r600.group.barrier() #13
>  %cmp88338 = icmp ugt i32 %width, 7
>  br i1 %cmp88338, label %for.cond91.preheader, label %while.cond109.preheader
>
>for.cond91.preheader:                             ; preds = %while.end80, %for.cond91.preheader
>  %i85.0340 = phi i32 [ %add107, %for.cond91.preheader ], [ 0, %while.end80 ]
>  %value84.0339 = phi <4 x float> [ %64, %for.cond91.preheader ], [ zeroinitializer, %while.end80 ]
>  %arrayidx95 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %i85.0340
>  %48 = load float, float addrspace(4)* %arrayidx95, align 4, !tbaa !12
>  %splat.splatinsert96 = insertelement <4 x float> undef, float %48, i32 0
>  %splat.splat97 = shufflevector <4 x float> %splat.splatinsert96, <4 x float> undef, <4 x i32> zeroinitializer
>  %add98 = add nuw nsw i32 %i85.0340, %11
>  %mul100 = mul nsw i32 %add98, %x.i.i
>  %add101 = add nsw i32 %mul100, %2
>  %arrayidx102 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add101
>  %49 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx102, align 16, !tbaa !14
>  %50 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat97, <4 x float> %49, <4 x float> %value84.0339)
>  %add99.1 = add nuw nsw i32 %add98, 1
>  %mul100.1 = mul nsw i32 %add99.1, %x.i.i
>  %add101.1 = add nsw i32 %mul100.1, %2
>  %arrayidx102.1 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add101.1
>  %51 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx102.1, align 16, !tbaa !14
>  %52 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat97, <4 x float> %51, <4 x float> %50)
>  %add99.2 = add nuw nsw i32 %add98, 2
>  %mul100.2 = mul nsw i32 %add99.2, %x.i.i
>  %add101.2 = add nsw i32 %mul100.2, %2
>  %arrayidx102.2 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add101.2
>  %53 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx102.2, align 16, !tbaa !14
>  %54 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat97, <4 x float> %53, <4 x float> %52)
>  %add99.3 = add nuw nsw i32 %add98, 3
>  %mul100.3 = mul nsw i32 %add99.3, %x.i.i
>  %add101.3 = add nsw i32 %mul100.3, %2
>  %arrayidx102.3 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add101.3
>  %55 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx102.3, align 16, !tbaa !14
>  %56 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat97, <4 x float> %55, <4 x float> %54)
>  %add99.4 = add nuw nsw i32 %add98, 4
>  %mul100.4 = mul nsw i32 %add99.4, %x.i.i
>  %add101.4 = add nsw i32 %mul100.4, %2
>  %arrayidx102.4 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add101.4
>  %57 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx102.4, align 16, !tbaa !14
>  %58 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat97, <4 x float> %57, <4 x float> %56)
>  %add99.5 = add nuw nsw i32 %add98, 5
>  %mul100.5 = mul nsw i32 %add99.5, %x.i.i
>  %add101.5 = add nsw i32 %mul100.5, %2
>  %arrayidx102.5 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add101.5
>  %59 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx102.5, align 16, !tbaa !14
>  %60 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat97, <4 x float> %59, <4 x float> %58)
>  %add99.6 = add nuw nsw i32 %add98, 6
>  %mul100.6 = mul nsw i32 %add99.6, %x.i.i
>  %add101.6 = add nsw i32 %mul100.6, %2
>  %arrayidx102.6 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add101.6
>  %61 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx102.6, align 16, !tbaa !14
>  %62 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat97, <4 x float> %61, <4 x float> %60)
>  %add99.7 = add nuw nsw i32 %add98, 7
>  %mul100.7 = mul nsw i32 %add99.7, %x.i.i
>  %add101.7 = add nsw i32 %mul100.7, %2
>  %arrayidx102.7 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add101.7
>  %63 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx102.7, align 16, !tbaa !14
>  %64 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat97, <4 x float> %63, <4 x float> %62)
>  %add107 = add nuw nsw i32 %i85.0340, 8
>  %add87 = or i32 %add107, 7
>  %cmp88 = icmp ult i32 %add87, %width
>  br i1 %cmp88, label %for.cond91.preheader, label %while.cond109.preheader.loopexit
>
>while.cond109.preheader.loopexit:                 ; preds = %for.cond91.preheader
>  %65 = and i32 %width, -8
>  br label %while.cond109.preheader
>
>while.cond109.preheader:                          ; preds = %while.cond109.preheader.loopexit, %while.end80
>  %value84.0.lcssa = phi <4 x float> [ zeroinitializer, %while.end80 ], [ %64, %while.cond109.preheader.loopexit ]
>  %i85.0.lcssa = phi i32 [ 0, %while.end80 ], [ %65, %while.cond109.preheader.loopexit ]
>  %cmp110333 = icmp ult i32 %i85.0.lcssa, %width
>  br i1 %cmp110333, label %while.body111, label %while.end121
>
>while.body111:                                    ; preds = %while.cond109.preheader, %while.body111
>  %i85.1335 = phi i32 [ %inc120, %while.body111 ], [ %i85.0.lcssa, %while.cond109.preheader ]
>  %value84.2334 = phi <4 x float> [ %68, %while.body111 ], [ %value84.0.lcssa, %while.cond109.preheader ]
>  %arrayidx112 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %i85.1335
>  %66 = load float, float addrspace(4)* %arrayidx112, align 4, !tbaa !12
>  %splat.splatinsert113 = insertelement <4 x float> undef, float %66, i32 0
>  %splat.splat114 = shufflevector <4 x float> %splat.splatinsert113, <4 x float> undef, <4 x i32> zeroinitializer
>  %add115 = add nuw nsw i32 %i85.1335, %11
>  %mul116 = mul nsw i32 %add115, %x.i.i
>  %add117 = add nsw i32 %mul116, %2
>  %arrayidx118 = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %pixels, i32 %add117
>  %67 = load <4 x float>, <4 x float> addrspace(3)* %arrayidx118, align 16, !tbaa !14
>  %68 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat114, <4 x float> %67, <4 x float> %value84.2334)
>  %inc120 = add nuw nsw i32 %i85.1335, 1
>  %exitcond = icmp eq i32 %inc120, %width
>  br i1 %exitcond, label %while.end121, label %while.body111
>
>while.end121:                                     ; preds = %while.body111, %while.cond109.preheader
>  %value84.2.lcssa = phi <4 x float> [ %value84.0.lcssa, %while.cond109.preheader ], [ %68, %while.body111 ]
>  %cmp122 = icmp ult i32 %8, %columns
>  %cmp123 = icmp ult i32 %16, %rows
>  %or.cond = and i1 %cmp122, %cmp123
>  br i1 %or.cond, label %if.then, label %if.end
>
>if.then:                                          ; preds = %while.end121
>  %mul1.i.i244 = mul i32 %16, %columns
>  %mul4.i.i245 = add i32 %mul1.i.i244, %8
>  %add.i.i246 = mul i32 %mul4.i.i245, %number_channels
>  %add.ptr.i247 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i246
>  %and.i.i248 = and i32 %channel, 1
>  %cmp.i.i249 = icmp eq i32 %and.i.i248, 0
>  br i1 %cmp.i.i249, label %if.end.i.i252, label %if.then.i.i250
>
>if.then.i.i250:                                   ; preds = %if.then
>  %69 = load float, float addrspace(1)* %add.ptr.i247, align 4, !tbaa !12
>  br label %if.end.i.i252
>
>if.end.i.i252:                                    ; preds = %if.then.i.i250, %if.then
>  %70 = phi float [ 0.000000e+00, %if.then ], [ %69, %if.then.i.i250 ]
>  %cmp1.i.i251 = icmp ugt i32 %number_channels, 2
>  br i1 %cmp1.i.i251, label %if.then2.i.i255, label %if.end13.i.i261
>
>if.then2.i.i255:                                  ; preds = %if.end.i.i252
>  %and3.i.i253 = and i32 %channel, 2
>  %cmp4.i.i254 = icmp eq i32 %and3.i.i253, 0
>  br i1 %cmp4.i.i254, label %if.end7.i.i259, label %if.then5.i.i256
>
>if.then5.i.i256:                                  ; preds = %if.then2.i.i255
>  %add.ptr.i31.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i247, i32 1
>  %71 = load float, float addrspace(1)* %add.ptr.i31.i.i, align 4, !tbaa !12
>  br label %if.end7.i.i259
>
>if.end7.i.i259:                                   ; preds = %if.then5.i.i256, %if.then2.i.i255
>  %72 = phi float [ 0.000000e+00, %if.then2.i.i255 ], [ %71, %if.then5.i.i256 ]
>  %and8.i.i257 = and i32 %channel, 4
>  %cmp9.i.i258 = icmp eq i32 %and8.i.i257, 0
>  br i1 %cmp9.i.i258, label %if.end13.i.i261, label %if.then10.i.i260
>
>if.then10.i.i260:                                 ; preds = %if.end7.i.i259
>  %add.ptr.i30.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i247, i32 2
>  %73 = load float, float addrspace(1)* %add.ptr.i30.i.i, align 4, !tbaa !12
>  br label %if.end13.i.i261
>
>if.end13.i.i261:                                  ; preds = %if.then10.i.i260, %if.end7.i.i259, %if.end.i.i252
>  %74 = phi float [ %72, %if.end7.i.i259 ], [ %72, %if.then10.i.i260 ], [ 0.000000e+00, %if.end.i.i252 ]
>  %75 = phi float [ 0.000000e+00, %if.end7.i.i259 ], [ %73, %if.then10.i.i260 ], [ 0.000000e+00, %if.end.i.i252 ]
>  switch i32 %number_channels, label %ReadFloat4.exit [
>    i32 4, label %land.lhs.true.i.i264
>    i32 2, label %land.lhs.true.i.i264
>  ]
>
>land.lhs.true.i.i264:                             ; preds = %if.end13.i.i261, %if.end13.i.i261
>  %and16.i.i262 = and i32 %channel, 16
>  %cmp17.i.i263 = icmp eq i32 %and16.i.i262, 0
>  br i1 %cmp17.i.i263, label %ReadFloat4.exit, label %if.then18.i.i267
>
>if.then18.i.i267:                                 ; preds = %land.lhs.true.i.i264
>  %add.ptr.i.i.i265 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i247, i32 %number_channels
>  %add.ptr1.i.i.i266 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i265, i32 -1
>  %76 = load float, float addrspace(1)* %add.ptr1.i.i.i266, align 4, !tbaa !12
>  br label %ReadFloat4.exit
>
>ReadFloat4.exit:                                  ; preds = %if.end13.i.i261, %land.lhs.true.i.i264, %if.then18.i.i267
>  %77 = phi float [ 0.000000e+00, %if.end13.i.i261 ], [ 0.000000e+00, %land.lhs.true.i.i264 ], [ %76, %if.then18.i.i267 ]
>  %vecinit.i = insertelement <4 x float> undef, float %70, i32 0
>  %vecinit1.i = insertelement <4 x float> %vecinit.i, float %74, i32 1
>  %vecinit2.i = insertelement <4 x float> %vecinit1.i, float %75, i32 2
>  %vecinit3.i = insertelement <4 x float> %vecinit2.i, float %77, i32 3
>  %sub125 = fsub <4 x float> %vecinit3.i, %value84.2.lcssa
>  %mul126 = fmul float %threshold, 6.553500e+04
>  %mul127 = fmul <4 x float> %sub125, <float 2.000000e+00, float 2.000000e+00, float 2.000000e+00, float 2.000000e+00>
>  %call128 = tail call <4 x float> @llvm.fabs.v4f32(<4 x float> %mul127) #15
>  %splat.splatinsert129 = insertelement <4 x float> undef, float %mul126, i32 0
>  %splat.splat130 = shufflevector <4 x float> %splat.splatinsert129, <4 x float> undef, <4 x i32> zeroinitializer
>  %78 = fcmp olt <4 x float> %call128, %splat.splat130
>  %splat.splatinsert132 = insertelement <4 x float> undef, float %gain, i32 0
>  %splat.splat133 = shufflevector <4 x float> %splat.splatinsert132, <4 x float> undef, <4 x i32> zeroinitializer
>  %79 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %sub125, <4 x float> %splat.splat133, <4 x float> %vecinit3.i)
>  %.v = select <4 x i1> %78, <4 x float> %vecinit3.i, <4 x float> %79
>  %add.ptr.i = getelementptr inbounds float, float addrspace(1)* %filteredImage, i32 %add.i.i246
>  %80 = extractelement <4 x float> %.v, i32 0
>  %81 = extractelement <4 x float> %.v, i32 1
>  %82 = extractelement <4 x float> %.v, i32 2
>  %83 = extractelement <4 x float> %.v, i32 3
>  br i1 %cmp.i.i249, label %if.end.i.i, label %if.then.i.i
>
>if.then.i.i:                                      ; preds = %ReadFloat4.exit
>  %84 = fcmp ogt float %80, 6.553500e+04
>  %85 = fcmp olt float %80, 0.000000e+00
>  %86 = select i1 %85, float 0.000000e+00, float %80
>  %.op.i.i.i = fadd float %86, 5.000000e-01
>  %add.i.i.i = select i1 %84, float 6.553550e+04, float %.op.i.i.i
>  store float %add.i.i.i, float addrspace(1)* %add.ptr.i, align 4, !tbaa !12
>  br label %if.end.i.i
>
>if.end.i.i:                                       ; preds = %if.then.i.i, %ReadFloat4.exit
>  br i1 %cmp1.i.i251, label %if.then2.i.i, label %if.end13.i.i
>
>if.then2.i.i:                                     ; preds = %if.end.i.i
>  %and3.i.i = and i32 %channel, 2
>  %cmp4.i.i = icmp eq i32 %and3.i.i, 0
>  br i1 %cmp4.i.i, label %if.end7.i.i, label %if.then5.i.i
>
>if.then5.i.i:                                     ; preds = %if.then2.i.i
>  %add.ptr.i35.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 1
>  %87 = fcmp ogt float %81, 6.553500e+04
>  %88 = fcmp olt float %81, 0.000000e+00
>  %89 = select i1 %88, float 0.000000e+00, float %81
>  %.op.i36.i.i = fadd float %89, 5.000000e-01
>  %add.i37.i.i = select i1 %87, float 6.553550e+04, float %.op.i36.i.i
>  store float %add.i37.i.i, float addrspace(1)* %add.ptr.i35.i.i, align 4, !tbaa !12
>  br label %if.end7.i.i
>
>if.end7.i.i:                                      ; preds = %if.then5.i.i, %if.then2.i.i
>  %and8.i.i = and i32 %channel, 4
>  %cmp9.i.i = icmp eq i32 %and8.i.i, 0
>  br i1 %cmp9.i.i, label %if.end13.i.i, label %if.then10.i.i
>
>if.then10.i.i:                                    ; preds = %if.end7.i.i
>  %add.ptr.i32.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 2
>  %90 = fcmp ogt float %82, 6.553500e+04
>  %91 = fcmp olt float %82, 0.000000e+00
>  %92 = select i1 %91, float 0.000000e+00, float %82
>  %.op.i33.i.i = fadd float %92, 5.000000e-01
>  %add.i34.i.i = select i1 %90, float 6.553550e+04, float %.op.i33.i.i
>  store float %add.i34.i.i, float addrspace(1)* %add.ptr.i32.i.i, align 4, !tbaa !12
>  br label %if.end13.i.i
>
>if.end13.i.i:                                     ; preds = %if.then10.i.i, %if.end7.i.i, %if.end.i.i
>  switch i32 %number_channels, label %if.end [
>    i32 4, label %land.lhs.true.i.i
>    i32 2, label %land.lhs.true.i.i
>  ]
>
>land.lhs.true.i.i:                                ; preds = %if.end13.i.i, %if.end13.i.i
>  %and16.i.i = and i32 %channel, 16
>  %cmp17.i.i = icmp eq i32 %and16.i.i, 0
>  br i1 %cmp17.i.i, label %if.end, label %if.then18.i.i
>
>if.then18.i.i:                                    ; preds = %land.lhs.true.i.i
>  %add.ptr.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i, i32 %number_channels
>  %add.ptr1.i.i.i = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i, i32 -1
>  %93 = fcmp ogt float %83, 6.553500e+04
>  %94 = fcmp olt float %83, 0.000000e+00
>  %95 = select i1 %94, float 0.000000e+00, float %83
>  %.op.i30.i.i = fadd float %95, 5.000000e-01
>  %add.i31.i.i = select i1 %93, float 6.553550e+04, float %.op.i30.i.i
>  store float %add.i31.i.i, float addrspace(1)* %add.ptr1.i.i.i, align 4, !tbaa !12
>  br label %if.end
>
>if.end:                                           ; preds = %if.then18.i.i, %land.lhs.true.i.i, %if.end13.i.i, %while.end121
>  ret void
>
>if.then.i.i306.1:                                 ; preds = %ReadFloat4.exit330
>  %96 = load float, float addrspace(1)* %add.ptr.i303.1, align 4, !tbaa !12
>  br label %if.end.i.i308.1
>
>if.end.i.i308.1:                                  ; preds = %if.then.i.i306.1, %ReadFloat4.exit330
>  %97 = phi float [ 0.000000e+00, %ReadFloat4.exit330 ], [ %96, %if.then.i.i306.1 ]
>  br i1 %cmp1.i.i307, label %if.then2.i.i311.1, label %if.end13.i.i319.1
>
>if.then2.i.i311.1:                                ; preds = %if.end.i.i308.1
>  br i1 %cmp4.i.i310, label %if.end7.i.i316.1, label %if.then5.i.i313.1
>
>if.then5.i.i313.1:                                ; preds = %if.then2.i.i311.1
>  %add.ptr.i31.i.i312.1 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.1, i32 1
>  %98 = load float, float addrspace(1)* %add.ptr.i31.i.i312.1, align 4, !tbaa !12
>  br label %if.end7.i.i316.1
>
>if.end7.i.i316.1:                                 ; preds = %if.then5.i.i313.1, %if.then2.i.i311.1
>  %99 = phi float [ 0.000000e+00, %if.then2.i.i311.1 ], [ %98, %if.then5.i.i313.1 ]
>  br i1 %cmp9.i.i315, label %if.end13.i.i319.1, label %if.then10.i.i318.1
>
>if.then10.i.i318.1:                               ; preds = %if.end7.i.i316.1
>  %add.ptr.i30.i.i317.1 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.1, i32 2
>  %100 = load float, float addrspace(1)* %add.ptr.i30.i.i317.1, align 4, !tbaa !12
>  br label %if.end13.i.i319.1
>
>if.end13.i.i319.1:                                ; preds = %if.then10.i.i318.1, %if.end7.i.i316.1, %if.end.i.i308.1
>  %101 = phi float [ %99, %if.end7.i.i316.1 ], [ %99, %if.then10.i.i318.1 ], [ 0.000000e+00, %if.end.i.i308.1 ]
>  %102 = phi float [ 0.000000e+00, %if.end7.i.i316.1 ], [ %100, %if.then10.i.i318.1 ], [ 0.000000e+00, %if.end.i.i308.1 ]
>  switch i32 %number_channels, label %ReadFloat4.exit330.1 [
>    i32 4, label %land.lhs.true.i.i322.1
>    i32 2, label %land.lhs.true.i.i322.1
>  ]
>
>land.lhs.true.i.i322.1:                           ; preds = %if.end13.i.i319.1, %if.end13.i.i319.1
>  br i1 %cmp17.i.i321, label %ReadFloat4.exit330.1, label %if.then18.i.i325.1
>
>if.then18.i.i325.1:                               ; preds = %land.lhs.true.i.i322.1
>  %add.ptr.i.i.i323.1 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.1, i32 %number_channels
>  %add.ptr1.i.i.i324.1 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i323.1, i32 -1
>  %103 = load float, float addrspace(1)* %add.ptr1.i.i.i324.1, align 4, !tbaa !12
>  br label %ReadFloat4.exit330.1
>
>ReadFloat4.exit330.1:                             ; preds = %if.then18.i.i325.1, %land.lhs.true.i.i322.1, %if.end13.i.i319.1
>  %104 = phi float [ 0.000000e+00, %if.end13.i.i319.1 ], [ 0.000000e+00, %land.lhs.true.i.i322.1 ], [ %103, %if.then18.i.i325.1 ]
>  %vecinit.i326.1 = insertelement <4 x float> undef, float %97, i32 0
>  %vecinit1.i327.1 = insertelement <4 x float> %vecinit.i326.1, float %101, i32 1
>  %vecinit2.i328.1 = insertelement <4 x float> %vecinit1.i327.1, float %102, i32 2
>  %vecinit3.i329.1 = insertelement <4 x float> %vecinit2.i328.1, float %104, i32 3
>  %105 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.1, <4 x float> %vecinit3.i329.1, <4 x float> %35)
>  %add26.2 = add nsw i32 %ix.0347, 2
>  %cmp27.2 = icmp slt i32 %ix.0347, -2
>  %sub29.2 = sub i32 -2, %ix.0347
>  %cond32.2 = select i1 %cmp27.2, i32 %sub29.2, i32 %add26.2
>  %cmp33.2 = icmp ult i32 %cond32.2, %columns
>  %sub37.2 = sub i32 %sub36, %cond32.2
>  %spec.select331.2 = select i1 %cmp33.2, i32 %cond32.2, i32 %sub37.2
>  %add41.2 = or i32 %i.0346, 2
>  %arrayidx.2 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add41.2
>  %106 = load float, float addrspace(4)* %arrayidx.2, align 4, !tbaa !12
>  %splat.splatinsert.2 = insertelement <4 x float> undef, float %106, i32 0
>  %splat.splat.2 = shufflevector <4 x float> %splat.splatinsert.2, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul4.i.i301.2 = add i32 %spec.select331.2, %mul1.i.i300
>  %add.i.i302.2 = mul i32 %mul4.i.i301.2, %number_channels
>  %add.ptr.i303.2 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i302.2
>  br i1 %cmp.i.i305, label %if.end.i.i308.2, label %if.then.i.i306.2
>
>if.then.i.i306.2:                                 ; preds = %ReadFloat4.exit330.1
>  %107 = load float, float addrspace(1)* %add.ptr.i303.2, align 4, !tbaa !12
>  br label %if.end.i.i308.2
>
>if.end.i.i308.2:                                  ; preds = %if.then.i.i306.2, %ReadFloat4.exit330.1
>  %108 = phi float [ 0.000000e+00, %ReadFloat4.exit330.1 ], [ %107, %if.then.i.i306.2 ]
>  br i1 %cmp1.i.i307, label %if.then2.i.i311.2, label %if.end13.i.i319.2
>
>if.then2.i.i311.2:                                ; preds = %if.end.i.i308.2
>  br i1 %cmp4.i.i310, label %if.end7.i.i316.2, label %if.then5.i.i313.2
>
>if.then5.i.i313.2:                                ; preds = %if.then2.i.i311.2
>  %add.ptr.i31.i.i312.2 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.2, i32 1
>  %109 = load float, float addrspace(1)* %add.ptr.i31.i.i312.2, align 4, !tbaa !12
>  br label %if.end7.i.i316.2
>
>if.end7.i.i316.2:                                 ; preds = %if.then5.i.i313.2, %if.then2.i.i311.2
>  %110 = phi float [ 0.000000e+00, %if.then2.i.i311.2 ], [ %109, %if.then5.i.i313.2 ]
>  br i1 %cmp9.i.i315, label %if.end13.i.i319.2, label %if.then10.i.i318.2
>
>if.then10.i.i318.2:                               ; preds = %if.end7.i.i316.2
>  %add.ptr.i30.i.i317.2 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.2, i32 2
>  %111 = load float, float addrspace(1)* %add.ptr.i30.i.i317.2, align 4, !tbaa !12
>  br label %if.end13.i.i319.2
>
>if.end13.i.i319.2:                                ; preds = %if.then10.i.i318.2, %if.end7.i.i316.2, %if.end.i.i308.2
>  %112 = phi float [ %110, %if.end7.i.i316.2 ], [ %110, %if.then10.i.i318.2 ], [ 0.000000e+00, %if.end.i.i308.2 ]
>  %113 = phi float [ 0.000000e+00, %if.end7.i.i316.2 ], [ %111, %if.then10.i.i318.2 ], [ 0.000000e+00, %if.end.i.i308.2 ]
>  switch i32 %number_channels, label %ReadFloat4.exit330.2 [
>    i32 4, label %land.lhs.true.i.i322.2
>    i32 2, label %land.lhs.true.i.i322.2
>  ]
>
>land.lhs.true.i.i322.2:                           ; preds = %if.end13.i.i319.2, %if.end13.i.i319.2
>  br i1 %cmp17.i.i321, label %ReadFloat4.exit330.2, label %if.then18.i.i325.2
>
>if.then18.i.i325.2:                               ; preds = %land.lhs.true.i.i322.2
>  %add.ptr.i.i.i323.2 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.2, i32 %number_channels
>  %add.ptr1.i.i.i324.2 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i323.2, i32 -1
>  %114 = load float, float addrspace(1)* %add.ptr1.i.i.i324.2, align 4, !tbaa !12
>  br label %ReadFloat4.exit330.2
>
>ReadFloat4.exit330.2:                             ; preds = %if.then18.i.i325.2, %land.lhs.true.i.i322.2, %if.end13.i.i319.2
>  %115 = phi float [ 0.000000e+00, %if.end13.i.i319.2 ], [ 0.000000e+00, %land.lhs.true.i.i322.2 ], [ %114, %if.then18.i.i325.2 ]
>  %vecinit.i326.2 = insertelement <4 x float> undef, float %108, i32 0
>  %vecinit1.i327.2 = insertelement <4 x float> %vecinit.i326.2, float %112, i32 1
>  %vecinit2.i328.2 = insertelement <4 x float> %vecinit1.i327.2, float %113, i32 2
>  %vecinit3.i329.2 = insertelement <4 x float> %vecinit2.i328.2, float %115, i32 3
>  %116 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.2, <4 x float> %vecinit3.i329.2, <4 x float> %105)
>  %add26.3 = add nsw i32 %ix.0347, 3
>  %cmp27.3 = icmp slt i32 %ix.0347, -3
>  %sub29.3 = sub i32 -3, %ix.0347
>  %cond32.3 = select i1 %cmp27.3, i32 %sub29.3, i32 %add26.3
>  %cmp33.3 = icmp ult i32 %cond32.3, %columns
>  %sub37.3 = sub i32 %sub36, %cond32.3
>  %spec.select331.3 = select i1 %cmp33.3, i32 %cond32.3, i32 %sub37.3
>  %add41.3 = or i32 %i.0346, 3
>  %arrayidx.3 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add41.3
>  %117 = load float, float addrspace(4)* %arrayidx.3, align 4, !tbaa !12
>  %splat.splatinsert.3 = insertelement <4 x float> undef, float %117, i32 0
>  %splat.splat.3 = shufflevector <4 x float> %splat.splatinsert.3, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul4.i.i301.3 = add i32 %spec.select331.3, %mul1.i.i300
>  %add.i.i302.3 = mul i32 %mul4.i.i301.3, %number_channels
>  %add.ptr.i303.3 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i302.3
>  br i1 %cmp.i.i305, label %if.end.i.i308.3, label %if.then.i.i306.3
>
>if.then.i.i306.3:                                 ; preds = %ReadFloat4.exit330.2
>  %118 = load float, float addrspace(1)* %add.ptr.i303.3, align 4, !tbaa !12
>  br label %if.end.i.i308.3
>
>if.end.i.i308.3:                                  ; preds = %if.then.i.i306.3, %ReadFloat4.exit330.2
>  %119 = phi float [ 0.000000e+00, %ReadFloat4.exit330.2 ], [ %118, %if.then.i.i306.3 ]
>  br i1 %cmp1.i.i307, label %if.then2.i.i311.3, label %if.end13.i.i319.3
>
>if.then2.i.i311.3:                                ; preds = %if.end.i.i308.3
>  br i1 %cmp4.i.i310, label %if.end7.i.i316.3, label %if.then5.i.i313.3
>
>if.then5.i.i313.3:                                ; preds = %if.then2.i.i311.3
>  %add.ptr.i31.i.i312.3 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.3, i32 1
>  %120 = load float, float addrspace(1)* %add.ptr.i31.i.i312.3, align 4, !tbaa !12
>  br label %if.end7.i.i316.3
>
>if.end7.i.i316.3:                                 ; preds = %if.then5.i.i313.3, %if.then2.i.i311.3
>  %121 = phi float [ 0.000000e+00, %if.then2.i.i311.3 ], [ %120, %if.then5.i.i313.3 ]
>  br i1 %cmp9.i.i315, label %if.end13.i.i319.3, label %if.then10.i.i318.3
>
>if.then10.i.i318.3:                               ; preds = %if.end7.i.i316.3
>  %add.ptr.i30.i.i317.3 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.3, i32 2
>  %122 = load float, float addrspace(1)* %add.ptr.i30.i.i317.3, align 4, !tbaa !12
>  br label %if.end13.i.i319.3
>
>if.end13.i.i319.3:                                ; preds = %if.then10.i.i318.3, %if.end7.i.i316.3, %if.end.i.i308.3
>  %123 = phi float [ %121, %if.end7.i.i316.3 ], [ %121, %if.then10.i.i318.3 ], [ 0.000000e+00, %if.end.i.i308.3 ]
>  %124 = phi float [ 0.000000e+00, %if.end7.i.i316.3 ], [ %122, %if.then10.i.i318.3 ], [ 0.000000e+00, %if.end.i.i308.3 ]
>  switch i32 %number_channels, label %ReadFloat4.exit330.3 [
>    i32 4, label %land.lhs.true.i.i322.3
>    i32 2, label %land.lhs.true.i.i322.3
>  ]
>
>land.lhs.true.i.i322.3:                           ; preds = %if.end13.i.i319.3, %if.end13.i.i319.3
>  br i1 %cmp17.i.i321, label %ReadFloat4.exit330.3, label %if.then18.i.i325.3
>
>if.then18.i.i325.3:                               ; preds = %land.lhs.true.i.i322.3
>  %add.ptr.i.i.i323.3 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.3, i32 %number_channels
>  %add.ptr1.i.i.i324.3 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i323.3, i32 -1
>  %125 = load float, float addrspace(1)* %add.ptr1.i.i.i324.3, align 4, !tbaa !12
>  br label %ReadFloat4.exit330.3
>
>ReadFloat4.exit330.3:                             ; preds = %if.then18.i.i325.3, %land.lhs.true.i.i322.3, %if.end13.i.i319.3
>  %126 = phi float [ 0.000000e+00, %if.end13.i.i319.3 ], [ 0.000000e+00, %land.lhs.true.i.i322.3 ], [ %125, %if.then18.i.i325.3 ]
>  %vecinit.i326.3 = insertelement <4 x float> undef, float %119, i32 0
>  %vecinit1.i327.3 = insertelement <4 x float> %vecinit.i326.3, float %123, i32 1
>  %vecinit2.i328.3 = insertelement <4 x float> %vecinit1.i327.3, float %124, i32 2
>  %vecinit3.i329.3 = insertelement <4 x float> %vecinit2.i328.3, float %126, i32 3
>  %127 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.3, <4 x float> %vecinit3.i329.3, <4 x float> %116)
>  %add26.4 = add nsw i32 %ix.0347, 4
>  %cmp27.4 = icmp slt i32 %ix.0347, -4
>  %sub29.4 = sub i32 -4, %ix.0347
>  %cond32.4 = select i1 %cmp27.4, i32 %sub29.4, i32 %add26.4
>  %cmp33.4 = icmp ult i32 %cond32.4, %columns
>  %sub37.4 = sub i32 %sub36, %cond32.4
>  %spec.select331.4 = select i1 %cmp33.4, i32 %cond32.4, i32 %sub37.4
>  %add41.4 = or i32 %i.0346, 4
>  %arrayidx.4 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add41.4
>  %128 = load float, float addrspace(4)* %arrayidx.4, align 4, !tbaa !12
>  %splat.splatinsert.4 = insertelement <4 x float> undef, float %128, i32 0
>  %splat.splat.4 = shufflevector <4 x float> %splat.splatinsert.4, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul4.i.i301.4 = add i32 %spec.select331.4, %mul1.i.i300
>  %add.i.i302.4 = mul i32 %mul4.i.i301.4, %number_channels
>  %add.ptr.i303.4 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i302.4
>  br i1 %cmp.i.i305, label %if.end.i.i308.4, label %if.then.i.i306.4
>
>if.then.i.i306.4:                                 ; preds = %ReadFloat4.exit330.3
>  %129 = load float, float addrspace(1)* %add.ptr.i303.4, align 4, !tbaa !12
>  br label %if.end.i.i308.4
>
>if.end.i.i308.4:                                  ; preds = %if.then.i.i306.4, %ReadFloat4.exit330.3
>  %130 = phi float [ 0.000000e+00, %ReadFloat4.exit330.3 ], [ %129, %if.then.i.i306.4 ]
>  br i1 %cmp1.i.i307, label %if.then2.i.i311.4, label %if.end13.i.i319.4
>
>if.then2.i.i311.4:                                ; preds = %if.end.i.i308.4
>  br i1 %cmp4.i.i310, label %if.end7.i.i316.4, label %if.then5.i.i313.4
>
>if.then5.i.i313.4:                                ; preds = %if.then2.i.i311.4
>  %add.ptr.i31.i.i312.4 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.4, i32 1
>  %131 = load float, float addrspace(1)* %add.ptr.i31.i.i312.4, align 4, !tbaa !12
>  br label %if.end7.i.i316.4
>
>if.end7.i.i316.4:                                 ; preds = %if.then5.i.i313.4, %if.then2.i.i311.4
>  %132 = phi float [ 0.000000e+00, %if.then2.i.i311.4 ], [ %131, %if.then5.i.i313.4 ]
>  br i1 %cmp9.i.i315, label %if.end13.i.i319.4, label %if.then10.i.i318.4
>
>if.then10.i.i318.4:                               ; preds = %if.end7.i.i316.4
>  %add.ptr.i30.i.i317.4 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.4, i32 2
>  %133 = load float, float addrspace(1)* %add.ptr.i30.i.i317.4, align 4, !tbaa !12
>  br label %if.end13.i.i319.4
>
>if.end13.i.i319.4:                                ; preds = %if.then10.i.i318.4, %if.end7.i.i316.4, %if.end.i.i308.4
>  %134 = phi float [ %132, %if.end7.i.i316.4 ], [ %132, %if.then10.i.i318.4 ], [ 0.000000e+00, %if.end.i.i308.4 ]
>  %135 = phi float [ 0.000000e+00, %if.end7.i.i316.4 ], [ %133, %if.then10.i.i318.4 ], [ 0.000000e+00, %if.end.i.i308.4 ]
>  switch i32 %number_channels, label %ReadFloat4.exit330.4 [
>    i32 4, label %land.lhs.true.i.i322.4
>    i32 2, label %land.lhs.true.i.i322.4
>  ]
>
>land.lhs.true.i.i322.4:                           ; preds = %if.end13.i.i319.4, %if.end13.i.i319.4
>  br i1 %cmp17.i.i321, label %ReadFloat4.exit330.4, label %if.then18.i.i325.4
>
>if.then18.i.i325.4:                               ; preds = %land.lhs.true.i.i322.4
>  %add.ptr.i.i.i323.4 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.4, i32 %number_channels
>  %add.ptr1.i.i.i324.4 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i323.4, i32 -1
>  %136 = load float, float addrspace(1)* %add.ptr1.i.i.i324.4, align 4, !tbaa !12
>  br label %ReadFloat4.exit330.4
>
>ReadFloat4.exit330.4:                             ; preds = %if.then18.i.i325.4, %land.lhs.true.i.i322.4, %if.end13.i.i319.4
>  %137 = phi float [ 0.000000e+00, %if.end13.i.i319.4 ], [ 0.000000e+00, %land.lhs.true.i.i322.4 ], [ %136, %if.then18.i.i325.4 ]
>  %vecinit.i326.4 = insertelement <4 x float> undef, float %130, i32 0
>  %vecinit1.i327.4 = insertelement <4 x float> %vecinit.i326.4, float %134, i32 1
>  %vecinit2.i328.4 = insertelement <4 x float> %vecinit1.i327.4, float %135, i32 2
>  %vecinit3.i329.4 = insertelement <4 x float> %vecinit2.i328.4, float %137, i32 3
>  %138 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.4, <4 x float> %vecinit3.i329.4, <4 x float> %127)
>  %add26.5 = add nsw i32 %ix.0347, 5
>  %cmp27.5 = icmp slt i32 %ix.0347, -5
>  %sub29.5 = sub i32 -5, %ix.0347
>  %cond32.5 = select i1 %cmp27.5, i32 %sub29.5, i32 %add26.5
>  %cmp33.5 = icmp ult i32 %cond32.5, %columns
>  %sub37.5 = sub i32 %sub36, %cond32.5
>  %spec.select331.5 = select i1 %cmp33.5, i32 %cond32.5, i32 %sub37.5
>  %add41.5 = or i32 %i.0346, 5
>  %arrayidx.5 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add41.5
>  %139 = load float, float addrspace(4)* %arrayidx.5, align 4, !tbaa !12
>  %splat.splatinsert.5 = insertelement <4 x float> undef, float %139, i32 0
>  %splat.splat.5 = shufflevector <4 x float> %splat.splatinsert.5, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul4.i.i301.5 = add i32 %spec.select331.5, %mul1.i.i300
>  %add.i.i302.5 = mul i32 %mul4.i.i301.5, %number_channels
>  %add.ptr.i303.5 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i302.5
>  br i1 %cmp.i.i305, label %if.end.i.i308.5, label %if.then.i.i306.5
>
>if.then.i.i306.5:                                 ; preds = %ReadFloat4.exit330.4
>  %140 = load float, float addrspace(1)* %add.ptr.i303.5, align 4, !tbaa !12
>  br label %if.end.i.i308.5
>
>if.end.i.i308.5:                                  ; preds = %if.then.i.i306.5, %ReadFloat4.exit330.4
>  %141 = phi float [ 0.000000e+00, %ReadFloat4.exit330.4 ], [ %140, %if.then.i.i306.5 ]
>  br i1 %cmp1.i.i307, label %if.then2.i.i311.5, label %if.end13.i.i319.5
>
>if.then2.i.i311.5:                                ; preds = %if.end.i.i308.5
>  br i1 %cmp4.i.i310, label %if.end7.i.i316.5, label %if.then5.i.i313.5
>
>if.then5.i.i313.5:                                ; preds = %if.then2.i.i311.5
>  %add.ptr.i31.i.i312.5 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.5, i32 1
>  %142 = load float, float addrspace(1)* %add.ptr.i31.i.i312.5, align 4, !tbaa !12
>  br label %if.end7.i.i316.5
>
>if.end7.i.i316.5:                                 ; preds = %if.then5.i.i313.5, %if.then2.i.i311.5
>  %143 = phi float [ 0.000000e+00, %if.then2.i.i311.5 ], [ %142, %if.then5.i.i313.5 ]
>  br i1 %cmp9.i.i315, label %if.end13.i.i319.5, label %if.then10.i.i318.5
>
>if.then10.i.i318.5:                               ; preds = %if.end7.i.i316.5
>  %add.ptr.i30.i.i317.5 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.5, i32 2
>  %144 = load float, float addrspace(1)* %add.ptr.i30.i.i317.5, align 4, !tbaa !12
>  br label %if.end13.i.i319.5
>
>if.end13.i.i319.5:                                ; preds = %if.then10.i.i318.5, %if.end7.i.i316.5, %if.end.i.i308.5
>  %145 = phi float [ %143, %if.end7.i.i316.5 ], [ %143, %if.then10.i.i318.5 ], [ 0.000000e+00, %if.end.i.i308.5 ]
>  %146 = phi float [ 0.000000e+00, %if.end7.i.i316.5 ], [ %144, %if.then10.i.i318.5 ], [ 0.000000e+00, %if.end.i.i308.5 ]
>  switch i32 %number_channels, label %ReadFloat4.exit330.5 [
>    i32 4, label %land.lhs.true.i.i322.5
>    i32 2, label %land.lhs.true.i.i322.5
>  ]
>
>land.lhs.true.i.i322.5:                           ; preds = %if.end13.i.i319.5, %if.end13.i.i319.5
>  br i1 %cmp17.i.i321, label %ReadFloat4.exit330.5, label %if.then18.i.i325.5
>
>if.then18.i.i325.5:                               ; preds = %land.lhs.true.i.i322.5
>  %add.ptr.i.i.i323.5 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.5, i32 %number_channels
>  %add.ptr1.i.i.i324.5 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i323.5, i32 -1
>  %147 = load float, float addrspace(1)* %add.ptr1.i.i.i324.5, align 4, !tbaa !12
>  br label %ReadFloat4.exit330.5
>
>ReadFloat4.exit330.5:                             ; preds = %if.then18.i.i325.5, %land.lhs.true.i.i322.5, %if.end13.i.i319.5
>  %148 = phi float [ 0.000000e+00, %if.end13.i.i319.5 ], [ 0.000000e+00, %land.lhs.true.i.i322.5 ], [ %147, %if.then18.i.i325.5 ]
>  %vecinit.i326.5 = insertelement <4 x float> undef, float %141, i32 0
>  %vecinit1.i327.5 = insertelement <4 x float> %vecinit.i326.5, float %145, i32 1
>  %vecinit2.i328.5 = insertelement <4 x float> %vecinit1.i327.5, float %146, i32 2
>  %vecinit3.i329.5 = insertelement <4 x float> %vecinit2.i328.5, float %148, i32 3
>  %149 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.5, <4 x float> %vecinit3.i329.5, <4 x float> %138)
>  %add26.6 = add nsw i32 %ix.0347, 6
>  %cmp27.6 = icmp slt i32 %ix.0347, -6
>  %sub29.6 = sub i32 -6, %ix.0347
>  %cond32.6 = select i1 %cmp27.6, i32 %sub29.6, i32 %add26.6
>  %cmp33.6 = icmp ult i32 %cond32.6, %columns
>  %sub37.6 = sub i32 %sub36, %cond32.6
>  %spec.select331.6 = select i1 %cmp33.6, i32 %cond32.6, i32 %sub37.6
>  %add41.6 = or i32 %i.0346, 6
>  %arrayidx.6 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add41.6
>  %150 = load float, float addrspace(4)* %arrayidx.6, align 4, !tbaa !12
>  %splat.splatinsert.6 = insertelement <4 x float> undef, float %150, i32 0
>  %splat.splat.6 = shufflevector <4 x float> %splat.splatinsert.6, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul4.i.i301.6 = add i32 %spec.select331.6, %mul1.i.i300
>  %add.i.i302.6 = mul i32 %mul4.i.i301.6, %number_channels
>  %add.ptr.i303.6 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i302.6
>  br i1 %cmp.i.i305, label %if.end.i.i308.6, label %if.then.i.i306.6
>
>if.then.i.i306.6:                                 ; preds = %ReadFloat4.exit330.5
>  %151 = load float, float addrspace(1)* %add.ptr.i303.6, align 4, !tbaa !12
>  br label %if.end.i.i308.6
>
>if.end.i.i308.6:                                  ; preds = %if.then.i.i306.6, %ReadFloat4.exit330.5
>  %152 = phi float [ 0.000000e+00, %ReadFloat4.exit330.5 ], [ %151, %if.then.i.i306.6 ]
>  br i1 %cmp1.i.i307, label %if.then2.i.i311.6, label %if.end13.i.i319.6
>
>if.then2.i.i311.6:                                ; preds = %if.end.i.i308.6
>  br i1 %cmp4.i.i310, label %if.end7.i.i316.6, label %if.then5.i.i313.6
>
>if.then5.i.i313.6:                                ; preds = %if.then2.i.i311.6
>  %add.ptr.i31.i.i312.6 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.6, i32 1
>  %153 = load float, float addrspace(1)* %add.ptr.i31.i.i312.6, align 4, !tbaa !12
>  br label %if.end7.i.i316.6
>
>if.end7.i.i316.6:                                 ; preds = %if.then5.i.i313.6, %if.then2.i.i311.6
>  %154 = phi float [ 0.000000e+00, %if.then2.i.i311.6 ], [ %153, %if.then5.i.i313.6 ]
>  br i1 %cmp9.i.i315, label %if.end13.i.i319.6, label %if.then10.i.i318.6
>
>if.then10.i.i318.6:                               ; preds = %if.end7.i.i316.6
>  %add.ptr.i30.i.i317.6 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.6, i32 2
>  %155 = load float, float addrspace(1)* %add.ptr.i30.i.i317.6, align 4, !tbaa !12
>  br label %if.end13.i.i319.6
>
>if.end13.i.i319.6:                                ; preds = %if.then10.i.i318.6, %if.end7.i.i316.6, %if.end.i.i308.6
>  %156 = phi float [ %154, %if.end7.i.i316.6 ], [ %154, %if.then10.i.i318.6 ], [ 0.000000e+00, %if.end.i.i308.6 ]
>  %157 = phi float [ 0.000000e+00, %if.end7.i.i316.6 ], [ %155, %if.then10.i.i318.6 ], [ 0.000000e+00, %if.end.i.i308.6 ]
>  switch i32 %number_channels, label %ReadFloat4.exit330.6 [
>    i32 4, label %land.lhs.true.i.i322.6
>    i32 2, label %land.lhs.true.i.i322.6
>  ]
>
>land.lhs.true.i.i322.6:                           ; preds = %if.end13.i.i319.6, %if.end13.i.i319.6
>  br i1 %cmp17.i.i321, label %ReadFloat4.exit330.6, label %if.then18.i.i325.6
>
>if.then18.i.i325.6:                               ; preds = %land.lhs.true.i.i322.6
>  %add.ptr.i.i.i323.6 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.6, i32 %number_channels
>  %add.ptr1.i.i.i324.6 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i323.6, i32 -1
>  %158 = load float, float addrspace(1)* %add.ptr1.i.i.i324.6, align 4, !tbaa !12
>  br label %ReadFloat4.exit330.6
>
>ReadFloat4.exit330.6:                             ; preds = %if.then18.i.i325.6, %land.lhs.true.i.i322.6, %if.end13.i.i319.6
>  %159 = phi float [ 0.000000e+00, %if.end13.i.i319.6 ], [ 0.000000e+00, %land.lhs.true.i.i322.6 ], [ %158, %if.then18.i.i325.6 ]
>  %vecinit.i326.6 = insertelement <4 x float> undef, float %152, i32 0
>  %vecinit1.i327.6 = insertelement <4 x float> %vecinit.i326.6, float %156, i32 1
>  %vecinit2.i328.6 = insertelement <4 x float> %vecinit1.i327.6, float %157, i32 2
>  %vecinit3.i329.6 = insertelement <4 x float> %vecinit2.i328.6, float %159, i32 3
>  %160 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.6, <4 x float> %vecinit3.i329.6, <4 x float> %149)
>  %add26.7 = add nsw i32 %ix.0347, 7
>  %cmp27.7 = icmp slt i32 %ix.0347, -7
>  %sub29.7 = sub i32 -7, %ix.0347
>  %cond32.7 = select i1 %cmp27.7, i32 %sub29.7, i32 %add26.7
>  %cmp33.7 = icmp ult i32 %cond32.7, %columns
>  %sub37.7 = sub i32 %sub36, %cond32.7
>  %spec.select331.7 = select i1 %cmp33.7, i32 %cond32.7, i32 %sub37.7
>  %add41.7 = or i32 %i.0346, 7
>  %arrayidx.7 = getelementptr inbounds float, float addrspace(4)* %filter, i32 %add41.7
>  %161 = load float, float addrspace(4)* %arrayidx.7, align 4, !tbaa !12
>  %splat.splatinsert.7 = insertelement <4 x float> undef, float %161, i32 0
>  %splat.splat.7 = shufflevector <4 x float> %splat.splatinsert.7, <4 x float> undef, <4 x i32> zeroinitializer
>  %mul4.i.i301.7 = add i32 %spec.select331.7, %mul1.i.i300
>  %add.i.i302.7 = mul i32 %mul4.i.i301.7, %number_channels
>  %add.ptr.i303.7 = getelementptr inbounds float, float addrspace(1)* %image, i32 %add.i.i302.7
>  br i1 %cmp.i.i305, label %if.end.i.i308.7, label %if.then.i.i306.7
>
>if.then.i.i306.7:                                 ; preds = %ReadFloat4.exit330.6
>  %162 = load float, float addrspace(1)* %add.ptr.i303.7, align 4, !tbaa !12
>  br label %if.end.i.i308.7
>
>if.end.i.i308.7:                                  ; preds = %if.then.i.i306.7, %ReadFloat4.exit330.6
>  %163 = phi float [ 0.000000e+00, %ReadFloat4.exit330.6 ], [ %162, %if.then.i.i306.7 ]
>  br i1 %cmp1.i.i307, label %if.then2.i.i311.7, label %if.end13.i.i319.7
>
>if.then2.i.i311.7:                                ; preds = %if.end.i.i308.7
>  br i1 %cmp4.i.i310, label %if.end7.i.i316.7, label %if.then5.i.i313.7
>
>if.then5.i.i313.7:                                ; preds = %if.then2.i.i311.7
>  %add.ptr.i31.i.i312.7 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.7, i32 1
>  %164 = load float, float addrspace(1)* %add.ptr.i31.i.i312.7, align 4, !tbaa !12
>  br label %if.end7.i.i316.7
>
>if.end7.i.i316.7:                                 ; preds = %if.then5.i.i313.7, %if.then2.i.i311.7
>  %165 = phi float [ 0.000000e+00, %if.then2.i.i311.7 ], [ %164, %if.then5.i.i313.7 ]
>  br i1 %cmp9.i.i315, label %if.end13.i.i319.7, label %if.then10.i.i318.7
>
>if.then10.i.i318.7:                               ; preds = %if.end7.i.i316.7
>  %add.ptr.i30.i.i317.7 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.7, i32 2
>  %166 = load float, float addrspace(1)* %add.ptr.i30.i.i317.7, align 4, !tbaa !12
>  br label %if.end13.i.i319.7
>
>if.end13.i.i319.7:                                ; preds = %if.then10.i.i318.7, %if.end7.i.i316.7, %if.end.i.i308.7
>  %167 = phi float [ %165, %if.end7.i.i316.7 ], [ %165, %if.then10.i.i318.7 ], [ 0.000000e+00, %if.end.i.i308.7 ]
>  %168 = phi float [ 0.000000e+00, %if.end7.i.i316.7 ], [ %166, %if.then10.i.i318.7 ], [ 0.000000e+00, %if.end.i.i308.7 ]
>  switch i32 %number_channels, label %ReadFloat4.exit330.7 [
>    i32 4, label %land.lhs.true.i.i322.7
>    i32 2, label %land.lhs.true.i.i322.7
>  ]
>
>land.lhs.true.i.i322.7:                           ; preds = %if.end13.i.i319.7, %if.end13.i.i319.7
>  br i1 %cmp17.i.i321, label %ReadFloat4.exit330.7, label %if.then18.i.i325.7
>
>if.then18.i.i325.7:                               ; preds = %land.lhs.true.i.i322.7
>  %add.ptr.i.i.i323.7 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i303.7, i32 %number_channels
>  %add.ptr1.i.i.i324.7 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i.i.i323.7, i32 -1
>  %169 = load float, float addrspace(1)* %add.ptr1.i.i.i324.7, align 4, !tbaa !12
>  br label %ReadFloat4.exit330.7
>
>ReadFloat4.exit330.7:                             ; preds = %if.then18.i.i325.7, %land.lhs.true.i.i322.7, %if.end13.i.i319.7
>  %170 = phi float [ 0.000000e+00, %if.end13.i.i319.7 ], [ 0.000000e+00, %land.lhs.true.i.i322.7 ], [ %169, %if.then18.i.i325.7 ]
>  %vecinit.i326.7 = insertelement <4 x float> undef, float %163, i32 0
>  %vecinit1.i327.7 = insertelement <4 x float> %vecinit.i326.7, float %167, i32 1
>  %vecinit2.i328.7 = insertelement <4 x float> %vecinit1.i327.7, float %168, i32 2
>  %vecinit3.i329.7 = insertelement <4 x float> %vecinit2.i328.7, float %170, i32 3
>  %171 = tail call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %splat.splat.7, <4 x float> %vecinit3.i329.7, <4 x float> %160)
>  %add44 = add nsw i32 %ix.0347, 8
>  %add45 = add nuw nsw i32 %i.0346, 8
>  %add22 = or i32 %add45, 7
>  %cmp23 = icmp ult i32 %add22, %width
>  br i1 %cmp23, label %for.cond.preheader, label %while.cond46.preheader
>}
>
>; Function Attrs: convergent nounwind
>define amdgpu_kernel void @WaveletDenoise(float addrspace(1)* nocapture readonly %srcImage, float addrspace(1)* nocapture %dstImage, i32 %number_channels, i32 %max_channels, float %threshold, i32 %passes, i32 %imageWidth, i32 %imageHeight) local_unnamed_addr #9 !kernel_arg_addr_space !99 !kernel_arg_access_qual !86 !kernel_arg_type !100 !kernel_arg_base_type !100 !kernel_arg_type_qual !101 !reqd_work_group_size !102 {
>entry:
>  %stage = alloca [48 x float], align 4, addrspace(5)
>  %tmp = alloca [16 x float], align 4, addrspace(5)
>  %accum = alloca [16 x float], align 4, addrspace(5)
>  %sub = add nsw i32 %passes, -1
>  %shl.mask = and i32 %sub, 31
>  %shl = shl i32 1, %shl.mask
>  %0 = bitcast [48 x float] addrspace(5)* %stage to i8 addrspace(5)*
>  call void @llvm.lifetime.start.p5i8(i64 192, i8 addrspace(5)* %0) #13
>  %1 = tail call i32 @llvm.r600.read.tgid.x() #13
>  %2 = tail call i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #13
>  %3 = getelementptr inbounds i8, i8 addrspace(7)* %2, i32 4
>  %4 = bitcast i8 addrspace(7)* %3 to i32 addrspace(7)*
>  %5 = load i32, i32 addrspace(7)* %4, align 4, !tbaa !4
>  %div = lshr i32 %5, 6
>  %add = add i32 %div, %1
>  %mul = shl nsw i32 %shl, 1
>  %sub2 = sub nsw i32 64, %mul
>  %mul3 = mul i32 %add, %sub2
>  %sub4 = sub i32 %mul3, %shl
>  %6 = tail call i32 @llvm.r600.read.tidig.x() #13, !range !3
>  %add6 = add i32 %sub4, %6
>  %7 = tail call i32 @llvm.r600.read.tgid.y() #13
>  %8 = getelementptr inbounds i8, i8 addrspace(7)* %2, i32 8
>  %9 = bitcast i8 addrspace(7)* %8 to i32 addrspace(7)*
>  %10 = load i32, i32 addrspace(7)* %9, align 4, !tbaa !4
>  %div9 = lshr i32 %10, 2
>  %add10 = add i32 %div9, %7
>  %mul13 = mul i32 %add10, %sub2
>  %sub14 = sub i32 %mul13, %shl
>  %11 = tail call i32 @llvm.r600.read.tidig.y() #13, !range !3
>  %cmp446 = icmp ult i32 %11, 64
>  br i1 %cmp446, label %for.body.lr.ph, label %for.cond39.preheader
>
>for.body.lr.ph:                                   ; preds = %entry
>  %cmp.i419 = icmp slt i32 %add6, 0
>  %sub.i420 = sub nsw i32 0, %add6
>  %cond.i421 = select i1 %cmp.i419, i32 %sub.i420, i32 %add6
>  %cmp.i414 = icmp slt i32 %cond.i421, %imageWidth
>  %mul.i415 = shl i32 %imageWidth, 1
>  %sub.i416 = xor i32 %cond.i421, -1
>  %sub1.i417 = add i32 %mul.i415, %sub.i416
>  %spec.select.i418 = select i1 %cmp.i414, i32 %cond.i421, i32 %sub1.i417
>  %mul.i = shl i32 %imageHeight, 1
>  %cmp26444 = icmp eq i32 %max_channels, 0
>  %y.i406 = tail call i32 @llvm.r600.read.local.size.y() #13
>  br label %for.body
>
>for.cond39.preheader:                             ; preds = %for.cond.cleanup27, %entry
>  %cmp40442 = icmp eq i32 %max_channels, 0
>  br i1 %cmp40442, label %for.cond.cleanup41, label %for.cond45.preheader.lr.ph
>
>for.cond45.preheader.lr.ph:                       ; preds = %for.cond39.preheader
>  %12 = bitcast [16 x float] addrspace(5)* %tmp to i8 addrspace(5)*
>  %13 = bitcast [16 x float] addrspace(5)* %accum to i8 addrspace(5)*
>  %y.i405 = tail call i32 @llvm.r600.read.local.size.y() #13
>  %cmp72438 = icmp sgt i32 %passes, 0
>  %arrayidx67 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 0
>  %arrayidx67.1 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 1
>  %arrayidx67.2 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 2
>  %arrayidx67.3 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 3
>  %arrayidx67.4 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 4
>  %arrayidx67.5 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 5
>  %arrayidx67.6 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 6
>  %arrayidx67.7 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 7
>  %arrayidx67.8 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 8
>  %arrayidx67.9 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 9
>  %arrayidx67.10 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 10
>  %arrayidx67.11 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 11
>  %arrayidx67.12 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 12
>  %arrayidx67.13 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 13
>  %arrayidx67.14 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 14
>  %arrayidx67.15 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 15
>  br label %for.cond45.preheader
>
>for.body:                                         ; preds = %for.body.lr.ph, %for.cond.cleanup27
>  %i.0447 = phi i32 [ %11, %for.body.lr.ph ], [ %add36, %for.cond.cleanup27 ]
>  %add19 = add nsw i32 %i.0447, %sub14
>  %cmp.i411 = icmp slt i32 %add19, 0
>  %sub.i412 = sub nsw i32 0, %add19
>  %cond.i413 = select i1 %cmp.i411, i32 %sub.i412, i32 %add19
>  %cmp.i407 = icmp slt i32 %cond.i413, %imageHeight
>  %sub.i408 = xor i32 %cond.i413, -1
>  %sub1.i409 = add i32 %mul.i, %sub.i408
>  %spec.select.i410 = select i1 %cmp.i407, i32 %cond.i413, i32 %sub1.i409
>  %mul22 = mul i32 %spec.select.i410, %imageWidth
>  %mul18387 = add i32 %mul22, %spec.select.i418
>  %add24 = mul i32 %mul18387, %number_channels
>  br i1 %cmp26444, label %for.cond.cleanup27, label %for.body28.lr.ph
>
>for.body28.lr.ph:                                 ; preds = %for.body
>  %div30 = sdiv i32 %i.0447, 4
>  br label %for.body28
>
>for.cond.cleanup27:                               ; preds = %for.body28, %for.body
>  %add36 = add i32 %i.0447, %y.i406
>  %cmp = icmp slt i32 %add36, 64
>  br i1 %cmp, label %for.body, label %for.cond39.preheader
>
>for.body28:                                       ; preds = %for.body28, %for.body28.lr.ph
>  %channel.0445 = phi i32 [ 0, %for.body28.lr.ph ], [ %inc, %for.body28 ]
>  %add29 = add nsw i32 %channel.0445, %add24
>  %arrayidx = getelementptr inbounds float, float addrspace(1)* %srcImage, i32 %add29
>  %14 = bitcast float addrspace(1)* %arrayidx to i32 addrspace(1)*
>  %15 = load i32, i32 addrspace(1)* %14, align 4, !tbaa !12
>  %mul31 = shl i32 %channel.0445, 4
>  %add32 = add nsw i32 %mul31, %div30
>  %arrayidx33 = getelementptr inbounds [48 x float], [48 x float] addrspace(5)* %stage, i32 0, i32 %add32
>  %16 = bitcast float addrspace(5)* %arrayidx33 to i32 addrspace(5)*
>  store i32 %15, i32 addrspace(5)* %16, align 4, !tbaa !12
>  %inc = add nuw nsw i32 %channel.0445, 1
>  %exitcond453 = icmp eq i32 %inc, %max_channels
>  br i1 %exitcond453, label %for.cond.cleanup27, label %for.body28
>
>for.cond45.preheader:                             ; preds = %for.cond.cleanup197, %for.cond45.preheader.lr.ph
>  %channel38.0443 = phi i32 [ 0, %for.cond45.preheader.lr.ph ], [ %inc211, %for.cond.cleanup197 ]
>  br i1 %cmp446, label %for.body48.lr.ph, label %for.cond.cleanup47
>
>for.body48.lr.ph:                                 ; preds = %for.cond45.preheader
>  %mul50 = shl i32 %channel38.0443, 4
>  br label %for.body48
>
>for.cond.cleanup41:                               ; preds = %for.cond.cleanup197, %for.cond39.preheader
>  %cmp214 = icmp ult i32 %6, %shl
>  br i1 %cmp214, label %if.end265, label %land.lhs.true
>
>for.cond.cleanup47:                               ; preds = %for.body48, %for.cond45.preheader
>  call void @llvm.lifetime.start.p5i8(i64 64, i8 addrspace(5)* %12) #13
>  call void @llvm.lifetime.start.p5i8(i64 64, i8 addrspace(5)* %13) #13
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.1, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.2, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.3, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.4, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.5, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.6, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.7, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.8, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.9, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.10, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.11, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.12, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.13, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.14, align 4, !tbaa !12
>  store float 0.000000e+00, float addrspace(5)* %arrayidx67.15, align 4, !tbaa !12
>  br i1 %cmp72438, label %for.body74, label %for.cond195.preheader
>
>for.body48:                                       ; preds = %for.body48.lr.ph, %for.body48
>  %i43.0428 = phi i32 [ %11, %for.body48.lr.ph ], [ %add60, %for.body48 ]
>  %div49 = sdiv i32 %i43.0428, 4
>  %add51 = add nsw i32 %div49, %mul50
>  %arrayidx52 = getelementptr inbounds [48 x float], [48 x float] addrspace(5)* %stage, i32 0, i32 %add51
>  %17 = bitcast float addrspace(5)* %arrayidx52 to i32 addrspace(5)*
>  %18 = load i32, i32 addrspace(5)* %17, align 4, !tbaa !12
>  %mul55 = shl i32 %i43.0428, 6
>  %add56 = add i32 %mul55, %6
>  %arrayidx57 = getelementptr inbounds [4096 x float], [4096 x float] addrspace(3)* @WaveletDenoise.buffer, i32 0, i32 %add56
>  %19 = bitcast float addrspace(3)* %arrayidx57 to i32 addrspace(3)*
>  store i32 %18, i32 addrspace(3)* %19, align 4, !tbaa !12
>  %add60 = add i32 %i43.0428, %y.i405
>  %cmp46 = icmp slt i32 %add60, 64
>  br i1 %cmp46, label %for.body48, label %for.cond.cleanup47
>
>for.cond195.preheader:                            ; preds = %if.end189, %for.cond.cleanup47
>  br i1 %cmp446, label %for.body198.lr.ph, label %for.cond.cleanup197
>
>for.body198.lr.ph:                                ; preds = %for.cond195.preheader
>  %mul203 = shl i32 %channel38.0443, 4
>  br label %for.body198
>
>for.body74:                                       ; preds = %for.cond.cleanup47, %if.end189
>  %pass.0439 = phi i32 [ %inc191, %if.end189 ], [ 0, %for.cond.cleanup47 ]
>  %shl.mask75 = and i32 %pass.0439, 31
>  %shl76 = shl i32 1, %shl.mask75
>  %arrayidx78 = getelementptr [8 x float], [8 x float] addrspace(4)* @WaveletDenoise.noise, i32 0, i32 %pass.0439
>  %20 = load float, float addrspace(4)* %arrayidx78, align 4
>  %mul79 = fmul float %20, %threshold
>  br i1 %cmp446, label %for.body85.lr.ph, label %for.cond.cleanup84
>
>for.body85.lr.ph:                                 ; preds = %for.body74
>  %cmp87 = icmp eq i32 %pass.0439, 0
>  %sub95 = sub nsw i32 %6, %shl76
>  %cmp.i402 = icmp slt i32 %sub95, 0
>  %sub.i403 = sub nsw i32 0, %sub95
>  %cond.i404 = select i1 %cmp.i402, i32 %sub.i403, i32 %sub95
>  %add99 = add nsw i32 %shl76, %6
>  %cmp.i398 = icmp slt i32 %add99, 64
>  %sub1.i400 = sub i32 127, %add99
>  %spec.select.i401 = select i1 %cmp.i398, i32 %add99, i32 %sub1.i400
>  br label %for.body85
>
>for.cond.cleanup84:                               ; preds = %if.end, %for.body74
>  %cmp446.pr = phi i1 [ false, %for.body74 ], [ %cmp446, %if.end ]
>  tail call void @llvm.r600.group.barrier() #13
>  br i1 %cmp446.pr, label %for.body116.lr.ph, label %for.cond.cleanup115
>
>for.body116.lr.ph:                                ; preds = %for.cond.cleanup84
>  %sub138 = fsub float -0.000000e+00, %mul79
>  br label %for.body116
>
>for.body85:                                       ; preds = %for.body85.lr.ph, %if.end
>  %i80.0431 = phi i32 [ %11, %for.body85.lr.ph ], [ %add109, %if.end ]
>  %mul86 = shl nsw i32 %i80.0431, 6
>  br i1 %cmp87, label %if.then, label %for.body85.if.end_crit_edge
>
>for.body85.if.end_crit_edge:                      ; preds = %for.body85
>  %.pre = sdiv i32 %i80.0431, 4
>  %.pre454 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %tmp, i32 0, i32 %.pre
>  %.pre455 = add nsw i32 %mul86, %6
>  %.pre456 = getelementptr inbounds [4096 x float], [4096 x float] addrspace(3)* @WaveletDenoise.buffer, i32 0, i32 %.pre455
>  br label %if.end
>
>if.then:                                          ; preds = %for.body85
>  %add88 = add nsw i32 %mul86, %6
>  %arrayidx89 = getelementptr inbounds [4096 x float], [4096 x float] addrspace(3)* @WaveletDenoise.buffer, i32 0, i32 %add88
>  %21 = bitcast float addrspace(3)* %arrayidx89 to i32 addrspace(3)*
>  %22 = load i32, i32 addrspace(3)* %21, align 4, !tbaa !12
>  %div90 = sdiv i32 %i80.0431, 4
>  %arrayidx91 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %tmp, i32 0, i32 %div90
>  %23 = bitcast float addrspace(5)* %arrayidx91 to i32 addrspace(5)*
>  store i32 %22, i32 addrspace(5)* %23, align 4, !tbaa !12
>  br label %if.end
>
>if.end:                                           ; preds = %for.body85.if.end_crit_edge, %if.then
>  %arrayidx106.pre-phi = phi float addrspace(3)* [ %.pre456, %for.body85.if.end_crit_edge ], [ %arrayidx89, %if.then ]
>  %arrayidx93.pre-phi = phi float addrspace(5)* [ %.pre454, %for.body85.if.end_crit_edge ], [ %arrayidx91, %if.then ]
>  %24 = load float, float addrspace(5)* %arrayidx93.pre-phi, align 4, !tbaa !12
>  %add97 = add nsw i32 %mul86, %cond.i404
>  %arrayidx98 = getelementptr inbounds [4096 x float], [4096 x float] addrspace(3)* @WaveletDenoise.buffer, i32 0, i32 %add97
>  %25 = load float, float addrspace(3)* %arrayidx98, align 4, !tbaa !12
>  %add101 = add nsw i32 %mul86, %spec.select.i401
>  %arrayidx102 = getelementptr inbounds [4096 x float], [4096 x float] addrspace(3)* @WaveletDenoise.buffer, i32 0, i32 %add101
>  %26 = load float, float addrspace(3)* %arrayidx102, align 4, !tbaa !12
>  %add103 = fadd float %25, %26
>  %mul104 = fmul float %add103, 2.500000e-01
>  %27 = tail call float @llvm.fmuladd.f32(float %24, float 5.000000e-01, float %mul104)
>  tail call void @llvm.r600.group.barrier() #13
>  store float %27, float addrspace(3)* %arrayidx106.pre-phi, align 4, !tbaa !12
>  %add109 = add i32 %i80.0431, %y.i405
>  %cmp83 = icmp slt i32 %add109, 64
>  br i1 %cmp83, label %for.body85, label %for.cond.cleanup84
>
>for.cond.cleanup115:                              ; preds = %if.end147, %for.cond.cleanup84
>  tail call void @llvm.r600.group.barrier() #13
>  %cmp156 = icmp slt i32 %pass.0439, %sub
>  br i1 %cmp156, label %for.cond160.preheader, label %for.cond176.preheader
>
>for.cond176.preheader:                            ; preds = %for.cond.cleanup115
>  br i1 %cmp446, label %for.body179, label %if.end189
>
>for.cond160.preheader:                            ; preds = %for.cond.cleanup115
>  br i1 %cmp446, label %for.body163, label %if.end189
>
>for.body116:                                      ; preds = %for.body116.lr.ph, %if.end147
>  %i111.0433 = phi i32 [ %11, %for.body116.lr.ph ], [ %add153, %if.end147 ]
>  %mul117 = shl i32 %i111.0433, 6
>  %add118 = add nsw i32 %mul117, %6
>  %arrayidx119 = getelementptr inbounds [4096 x float], [4096 x float] addrspace(3)* @WaveletDenoise.buffer, i32 0, i32 %add118
>  %28 = load float, float addrspace(3)* %arrayidx119, align 4, !tbaa !12
>  %sub121 = sub nsw i32 %i111.0433, %shl76
>  %cmp.i395 = icmp slt i32 %sub121, 0
>  %sub.i396 = sub nsw i32 0, %sub121
>  %cond.i = select i1 %cmp.i395, i32 %sub.i396, i32 %sub121
>  %mul123 = shl i32 %cond.i, 6
>  %add124 = add nsw i32 %mul123, %6
>  %arrayidx125 = getelementptr inbounds [4096 x float], [4096 x float] addrspace(3)* @WaveletDenoise.buffer, i32 0, i32 %add124
>  %29 = load float, float addrspace(3)* %arrayidx125, align 4, !tbaa !12
>  %add126 = add nsw i32 %i111.0433, %shl76
>  %cmp.i = icmp slt i32 %add126, 64
>  %sub1.i = sub i32 127, %add126
>  %spec.select.i = select i1 %cmp.i, i32 %add126, i32 %sub1.i
>  %mul128 = shl i32 %spec.select.i, 6
>  %add129 = add nsw i32 %mul128, %6
>  %arrayidx130 = getelementptr inbounds [4096 x float], [4096 x float] addrspace(3)* @WaveletDenoise.buffer, i32 0, i32 %add129
>  %30 = load float, float addrspace(3)* %arrayidx130, align 4, !tbaa !12
>  %add131 = fadd float %29, %30
>  %mul132 = fmul float %add131, 2.500000e-01
>  %31 = tail call float @llvm.fmuladd.f32(float %28, float 5.000000e-01, float %mul132)
>  %div133 = sdiv i32 %i111.0433, 4
>  %arrayidx134 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %tmp, i32 0, i32 %div133
>  %32 = load float, float addrspace(5)* %arrayidx134, align 4, !tbaa !12
>  %sub135 = fsub float %32, %31
>  store float %31, float addrspace(5)* %arrayidx134, align 4, !tbaa !12
>  %cmp139 = fcmp olt float %sub135, %sub138
>  br i1 %cmp139, label %if.then140, label %if.else
>
>if.then140:                                       ; preds = %for.body116
>  %add141 = fadd float %mul79, %sub135
>  br label %if.end147
>
>if.else:                                          ; preds = %for.body116
>  %cmp142 = fcmp ogt float %sub135, %mul79
>  br i1 %cmp142, label %if.then143, label %if.end147
>
>if.then143:                                       ; preds = %if.else
>  %sub144 = fsub float %sub135, %mul79
>  br label %if.end147
>
>if.end147:                                        ; preds = %if.else, %if.then143, %if.then140
>  %delta.0 = phi float [ %add141, %if.then140 ], [ %sub144, %if.then143 ], [ 0.000000e+00, %if.else ]
>  %arrayidx149 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 %div133
>  %33 = load float, float addrspace(5)* %arrayidx149, align 4, !tbaa !12
>  %add150 = fadd float %delta.0, %33
>  store float %add150, float addrspace(5)* %arrayidx149, align 4, !tbaa !12
>  %add153 = add i32 %i111.0433, %y.i405
>  %cmp114 = icmp slt i32 %add153, 64
>  br i1 %cmp114, label %for.body116, label %for.cond.cleanup115
>
>for.body163:                                      ; preds = %for.cond160.preheader, %for.body163
>  %i158.0437 = phi i32 [ %add171, %for.body163 ], [ %11, %for.cond160.preheader ]
>  %div164 = sdiv i32 %i158.0437, 4
>  %arrayidx165 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %tmp, i32 0, i32 %div164
>  %34 = bitcast float addrspace(5)* %arrayidx165 to i32 addrspace(5)*
>  %35 = load i32, i32 addrspace(5)* %34, align 4, !tbaa !12
>  %mul166 = shl i32 %i158.0437, 6
>  %add167 = add nsw i32 %mul166, %6
>  %arrayidx168 = getelementptr inbounds [4096 x float], [4096 x float] addrspace(3)* @WaveletDenoise.buffer, i32 0, i32 %add167
>  %36 = bitcast float addrspace(3)* %arrayidx168 to i32 addrspace(3)*
>  store i32 %35, i32 addrspace(3)* %36, align 4, !tbaa !12
>  %add171 = add i32 %i158.0437, %y.i405
>  %cmp161 = icmp slt i32 %add171, 64
>  br i1 %cmp161, label %for.body163, label %if.end189
>
>for.body179:                                      ; preds = %for.cond176.preheader, %for.body179
>  %i174.0435 = phi i32 [ %add187, %for.body179 ], [ %11, %for.cond176.preheader ]
>  %div180 = sdiv i32 %i174.0435, 4
>  %arrayidx181 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %tmp, i32 0, i32 %div180
>  %37 = load float, float addrspace(5)* %arrayidx181, align 4, !tbaa !12
>  %arrayidx183 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 %div180
>  %38 = load float, float addrspace(5)* %arrayidx183, align 4, !tbaa !12
>  %add184 = fadd float %37, %38
>  store float %add184, float addrspace(5)* %arrayidx183, align 4, !tbaa !12
>  %add187 = add i32 %i174.0435, %y.i405
>  %cmp177 = icmp slt i32 %add187, 64
>  br i1 %cmp177, label %for.body179, label %if.end189
>
>if.end189:                                        ; preds = %for.body179, %for.body163, %for.cond176.preheader, %for.cond160.preheader
>  tail call void @llvm.r600.group.barrier() #13
>  %inc191 = add nuw nsw i32 %pass.0439, 1
>  %exitcond451 = icmp eq i32 %inc191, %passes
>  br i1 %exitcond451, label %for.cond195.preheader, label %for.body74
>
>for.cond.cleanup197:                              ; preds = %for.body198, %for.cond195.preheader
>  tail call void @llvm.r600.group.barrier() #13
>  call void @llvm.lifetime.end.p5i8(i64 64, i8 addrspace(5)* %13) #13
>  call void @llvm.lifetime.end.p5i8(i64 64, i8 addrspace(5)* %12) #13
>  %inc211 = add nuw nsw i32 %channel38.0443, 1
>  %exitcond452 = icmp eq i32 %inc211, %max_channels
>  br i1 %exitcond452, label %for.cond.cleanup41, label %for.cond45.preheader
>
>for.body198:                                      ; preds = %for.body198.lr.ph, %for.body198
>  %i193.0441 = phi i32 [ %11, %for.body198.lr.ph ], [ %add208, %for.body198 ]
>  %div199 = sdiv i32 %i193.0441, 4
>  %arrayidx200 = getelementptr inbounds [16 x float], [16 x float] addrspace(5)* %accum, i32 0, i32 %div199
>  %39 = load float, float addrspace(5)* %arrayidx200, align 4, !tbaa !12
>  %40 = fcmp ogt float %39, 6.553500e+04
>  %41 = fcmp olt float %39, 0.000000e+00
>  %42 = select i1 %41, float 0.000000e+00, float %39
>  %.op.i = fadd float %42, 5.000000e-01
>  %add.i = select i1 %40, float 6.553550e+04, float %.op.i
>  %add204 = add nsw i32 %div199, %mul203
>  %arrayidx205 = getelementptr inbounds [48 x float], [48 x float] addrspace(5)* %stage, i32 0, i32 %add204
>  store float %add.i, float addrspace(5)* %arrayidx205, align 4, !tbaa !12
>  %add208 = add i32 %i193.0441, %y.i405
>  %cmp196 = icmp slt i32 %add208, 64
>  br i1 %cmp196, label %for.body198, label %for.cond.cleanup197
>
>land.lhs.true:                                    ; preds = %for.cond.cleanup41
>  %sub216 = sub nsw i32 64, %shl
>  %cmp217 = icmp ult i32 %6, %sub216
>  %cmp219 = icmp sgt i32 %add6, -1
>  %or.cond = and i1 %cmp217, %cmp219
>  %cmp221 = icmp ult i32 %add6, %imageWidth
>  %or.cond388 = and i1 %cmp221, %or.cond
>  %or.cond448 = and i1 %or.cond388, %cmp446
>  br i1 %or.cond448, label %for.body228.lr.ph, label %if.end265
>
>for.body228.lr.ph:                                ; preds = %land.lhs.true
>  %y.i = tail call i32 @llvm.r600.read.local.size.y() #13
>  br label %for.body228
>
>for.body228:                                      ; preds = %for.body228.lr.ph, %for.inc261
>  %i223.0426 = phi i32 [ %11, %for.body228.lr.ph ], [ %add263, %for.inc261 ]
>  %cmp229 = icmp sge i32 %i223.0426, %shl
>  %cmp232 = icmp slt i32 %i223.0426, %sub216
>  %or.cond389 = and i1 %cmp229, %cmp232
>  br i1 %or.cond389, label %land.lhs.true233, label %for.inc261
>
>land.lhs.true233:                                 ; preds = %for.body228
>  %add234 = add nsw i32 %i223.0426, %sub14
>  %cmp235 = icmp slt i32 %add234, 0
>  %cmp238 = icmp uge i32 %add234, %imageHeight
>  %or.cond390.not = or i1 %cmp238, %cmp235
>  %brmerge = or i1 %or.cond390.not, %cmp40442
>  br i1 %brmerge, label %for.inc261, label %for.body250.lr.ph
>
>for.body250.lr.ph:                                ; preds = %land.lhs.true233
>  %mul244 = mul i32 %add234, %imageWidth
>  %div251 = sdiv i32 %i223.0426, 4
>  %reass.add = add i32 %mul244, %add6
>  %reass.mul = mul i32 %reass.add, %number_channels
>  br label %for.body250
>
>for.body250:                                      ; preds = %for.body250, %for.body250.lr.ph
>  %channel246.0424 = phi i32 [ 0, %for.body250.lr.ph ], [ %inc258, %for.body250 ]
>  %mul252 = shl i32 %channel246.0424, 4
>  %add253 = add nsw i32 %mul252, %div251
>  %arrayidx254 = getelementptr inbounds [48 x float], [48 x float] addrspace(5)* %stage, i32 0, i32 %add253
>  %43 = bitcast float addrspace(5)* %arrayidx254 to i32 addrspace(5)*
>  %44 = load i32, i32 addrspace(5)* %43, align 4, !tbaa !12
>  %add255 = add i32 %reass.mul, %channel246.0424
>  %arrayidx256 = getelementptr inbounds float, float addrspace(1)* %dstImage, i32 %add255
>  %45 = bitcast float addrspace(1)* %arrayidx256 to i32 addrspace(1)*
>  store i32 %44, i32 addrspace(1)* %45, align 4, !tbaa !12
>  %inc258 = add nuw nsw i32 %channel246.0424, 1
>  %exitcond = icmp eq i32 %inc258, %max_channels
>  br i1 %exitcond, label %for.inc261, label %for.body250
>
>for.inc261:                                       ; preds = %for.body250, %land.lhs.true233, %for.body228
>  %add263 = add i32 %i223.0426, %y.i
>  %cmp226 = icmp slt i32 %add263, 64
>  br i1 %cmp226, label %for.body228, label %if.end265
>
>if.end265:                                        ; preds = %for.inc261, %for.cond.cleanup41, %land.lhs.true
>  call void @llvm.lifetime.end.p5i8(i64 192, i8 addrspace(5)* %0) #13
>  ret void
>}
>
>; Function Attrs: nounwind readnone speculatable
>declare float @llvm.floor.f32(float) #5
>
>; Function Attrs: nounwind readnone speculatable
>declare float @llvm.fabs.f32(float) #5
>
>; Function Attrs: convergent inlinehint nounwind
>declare float @applyResizeFilter(float, i32, float addrspace(1)*) local_unnamed_addr #10
>
>; Function Attrs: convergent nounwind
>declare void @llvm.r600.group.barrier() #11
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i8 addrspace(7)* @llvm.r600.implicitarg.ptr() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.tgid.x() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.tgid.y() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.global.size.x() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.global.size.y() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.tidig.x() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.tidig.y() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.tidig.z() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.local.size.x() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.local.size.y() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.r600.read.local.size.z() #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare float @llvm.sqrt.f32(float) #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare float @llvm.copysign.f32(float, float) #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare float @llvm.trunc.f32(float) #12
>
>; Function Attrs: convergent nounwind readnone speculatable
>declare i32 @llvm.ctlz.i32(i32, i1) #12
>
>attributes #0 = { norecurse nounwind readnone "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="juniper" "target-features"="-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #1 = { argmemonly nounwind }
>attributes #2 = { convergent nounwind readonly "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="juniper" "target-features"="-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #3 = { norecurse nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="juniper" "target-features"="-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #4 = { convergent nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="juniper" "target-features"="-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #5 = { nounwind readnone speculatable }
>attributes #6 = { nounwind readonly "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="juniper" "target-features"="-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #7 = { nounwind readnone "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="juniper" "target-features"="-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #8 = { convergent nounwind readnone "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="juniper" "target-features"="-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #9 = { convergent nounwind "amdgpu-flat-work-group-size"="256,256" "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="juniper" "target-features"="-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #10 = { convergent inlinehint nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="juniper" "target-features"="-fp32-denormals" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #11 = { convergent nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #12 = { convergent nounwind readnone speculatable "correctly-rounded-divide-sqrt-fp-math"="false" "less-precise-fpmad"="true" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
>attributes #13 = { nounwind }
>attributes #14 = { convergent nobuiltin nounwind }
>attributes #15 = { convergent }
>attributes #16 = { nounwind readnone }
>attributes #17 = { convergent nounwind }
>
>!llvm.module.flags = !{!0}
>!opencl.ocl.version = !{!1}
>!llvm.ident = !{!2}
>
>!0 = !{i32 1, !"wchar_size", i32 4}
>!1 = !{i32 1, i32 1}
>!2 = !{!"clang version 7.0.0 (trunk 325439)"}
>!3 = !{i32 0, i32 1024}
>!4 = !{!5, !5, i64 0}
>!5 = !{!"int", !6, i64 0}
>!6 = !{!"omnipotent char", !7, i64 0}
>!7 = !{!"Simple C/C++ TBAA"}
>!8 = !{!9, !5, i64 0}
>!9 = !{!"", !5, i64 0, !5, i64 4}
>!10 = !{!9, !5, i64 4}
>!11 = !{float 2.500000e+00}
>!12 = !{!13, !13, i64 0}
>!13 = !{!"float", !6, i64 0}
>!14 = !{!6, !6, i64 0}
>!15 = !{i32 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1}
>!16 = !{!"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none"}
>!17 = !{!"float*", !"uint", !"ChannelType", !"uint", !"uint", !"NoiseType", !"float", !"uint", !"uint", !"uint", !"float*"}
>!18 = !{!"const", !"", !"", !"", !"", !"", !"", !"", !"", !"", !""}
>!19 = !{i32 1, i32 0, i32 0, i32 2, i32 0, i32 0, i32 0, i32 3, i32 1}
>!20 = !{!"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none"}
>!21 = !{!"float*", !"uint", !"ChannelType", !"float*", !"uint", !"uint", !"uint", !"float4*", !"float4*"}
>!22 = !{!"float*", !"uint", !"ChannelType", !"float*", !"uint", !"uint", !"uint", !"float __attribute__((ext_vector_type(4)))*", !"float __attribute__((ext_vector_type(4)))*"}
>!23 = !{!"const", !"", !"", !"const", !"", !"", !"", !"", !""}
>!24 = !{!"float4*", !"uint", !"ChannelType", !"float*", !"uint", !"uint", !"uint", !"float4*", !"float*"}
>!25 = !{!"float __attribute__((ext_vector_type(4)))*", !"uint", !"ChannelType", !"float*", !"uint", !"uint", !"uint", !"float __attribute__((ext_vector_type(4)))*", !"float*"}
>!26 = !{i32 1, i32 0, i32 0}
>!27 = !{!"none", !"none", !"none"}
>!28 = !{!"float*", !"uint", !"int"}
>!29 = !{!"", !"", !""}
>!30 = !{i32 1, i32 0, i32 0, i32 0, i32 1}
>!31 = !{!"none", !"none", !"none", !"none", !"none"}
>!32 = !{!"float4*", !"ChannelType", !"uint", !"uint", !"uint4*"}
>!33 = !{!"float __attribute__((ext_vector_type(4)))*", !"ChannelType", !"uint", !"uint", !"uint __attribute__((ext_vector_type(4)))*"}
>!34 = !{!"restrict", !"", !"", !"", !"restrict"}
>!35 = !{i32 1, i32 0, i32 1, i32 0, i32 0}
>!36 = !{!"float4*", !"ChannelType", !"float4*", !"float4", !"float4"}
>!37 = !{!"float __attribute__((ext_vector_type(4)))*", !"ChannelType", !"float __attribute__((ext_vector_type(4)))*", !"float __attribute__((ext_vector_type(4)))", !"float __attribute__((ext_vector_type(4)))"}
>!38 = !{!"restrict", !"", !"restrict", !"", !""}
>!39 = !{i32 1, i32 1, i32 0, i32 0, i32 2, i32 0, i32 0, i32 0, i32 0, i32 3, i32 3}
>!40 = !{!"float4*", !"float4*", !"uint", !"uint", !"float*", !"uint", !"uint", !"uint", !"ChannelType", !"float4*", !"float*"}
>!41 = !{!"float __attribute__((ext_vector_type(4)))*", !"float __attribute__((ext_vector_type(4)))*", !"uint", !"uint", !"float*", !"uint", !"uint", !"uint", !"ChannelType", !"float __attribute__((ext_vector_type(4)))*", !"float*"}
>!42 = !{!"const", !"", !"", !"", !"const", !"", !"", !"", !"", !"", !""}
>!43 = !{i32 1, i32 1, i32 0, i32 0, i32 2, i32 0, i32 0, i32 0, i32 0}
>!44 = !{!"float4*", !"float4*", !"uint", !"uint", !"float*", !"uint", !"uint", !"uint", !"ChannelType"}
>!45 = !{!"float __attribute__((ext_vector_type(4)))*", !"float __attribute__((ext_vector_type(4)))*", !"uint", !"uint", !"float*", !"uint", !"uint", !"uint", !"ChannelType"}
>!46 = !{!"const", !"", !"", !"", !"const", !"", !"", !"", !""}
>!47 = !{i32 1, i32 1, i32 0, i32 0, i32 0, i32 0, i32 0}
>!48 = !{!"none", !"none", !"none", !"none", !"none", !"none", !"none"}
>!49 = !{!"float4*", !"float4*", !"uint", !"uint", !"int2", !"int", !"int"}
>!50 = !{!"float __attribute__((ext_vector_type(4)))*", !"float __attribute__((ext_vector_type(4)))*", !"uint", !"uint", !"int __attribute__((ext_vector_type(2)))", !"int", !"int"}
>!51 = !{!"const", !"", !"", !"", !"", !"", !""}
>!52 = !{i32 0, i32 33}
>!53 = !{i32 1, i32 0, i32 0, i32 0, i32 0, i32 2}
>!54 = !{!"none", !"none", !"none", !"none", !"none", !"none"}
>!55 = !{!"float*", !"uint", !"ChannelType", !"MagickFunction", !"uint", !"float*"}
>!56 = !{!"", !"", !"", !"", !"", !"const"}
>!57 = !{i32 1, i32 0, i32 0, i32 0}
>!58 = !{!"none", !"none", !"none", !"none"}
>!59 = !{!"float*", !"int", !"uint", !"uint"}
>!60 = !{!"", !"", !"", !""}
>!61 = !{i32 1, i32 1, i32 1, i32 0, i32 0, i32 0}
>!62 = !{!"float4*", !"float4*", !"float*", !"int", !"int", !"int"}
>!63 = !{!"float __attribute__((ext_vector_type(4)))*", !"float __attribute__((ext_vector_type(4)))*", !"float*", !"int", !"int", !"int"}
>!64 = !{!"", !"", !"", !"", !"", !""}
>!65 = !{i32 1, i32 1, i32 1, i32 0, i32 0, i32 0, i32 0}
>!66 = !{!"float4*", !"float4*", !"float*", !"int", !"float", !"int", !"int"}
>!67 = !{!"float __attribute__((ext_vector_type(4)))*", !"float __attribute__((ext_vector_type(4)))*", !"float*", !"int", !"float", !"int", !"int"}
>!68 = !{!"", !"", !"", !"", !"", !"", !""}
>!69 = !{i32 1, i32 0, i32 0, i32 0, i32 0}
>!70 = !{!"float4*", !"float", !"float", !"float", !"int"}
>!71 = !{!"float __attribute__((ext_vector_type(4)))*", !"float", !"float", !"float", !"int"}
>!72 = !{!"", !"", !"", !"", !""}
>!73 = !{i32 1, i32 1, i32 0, i32 0, i32 1, i32 0, i32 1, i32 0, i32 0, i32 0}
>!74 = !{!"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none"}
>!75 = !{!"float4*", !"float4*", !"uint", !"uint", !"float*", !"uint", !"int2*", !"float4", !"ChannelType", !"uint"}
>!76 = !{!"float __attribute__((ext_vector_type(4)))*", !"float __attribute__((ext_vector_type(4)))*", !"uint", !"uint", !"float*", !"uint", !"int __attribute__((ext_vector_type(2)))*", !"float __attribute__((ext_vector_type(4)))", !"ChannelType", !"uint"}
>!77 = !{!"const", !"", !"", !"", !"const", !"", !"const", !"", !"", !""}
>!78 = !{i32 1, i32 0, i32 0, i32 0, i32 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 0, i32 3, i32 3, i32 3}
>!79 = !{!"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none"}
>!80 = !{!"float*", !"uint", !"uint", !"uint", !"float*", !"uint", !"uint", !"float", !"int", !"int", !"float*", !"float", !"float", !"float", !"float", !"float*", !"int", !"uint", !"uint", !"float4*", !"float*", !"float*"}
>!81 = !{!"float*", !"uint", !"uint", !"uint", !"float*", !"uint", !"uint", !"float", !"int", !"int", !"float*", !"float", !"float", !"float", !"float", !"float*", !"int", !"uint", !"uint", !"float __attribute__((ext_vector_type(4)))*", !"float*", !"float*"}
>!82 = !{!"const", !"", !"", !"", !"", !"", !"", !"", !"", !"", !"const", !"", !"", !"", !"", !"", !"", !"", !"", !"", !"", !""}
>!83 = !{i32 256, i32 1, i32 1}
>!84 = !{i32 1, i32 256, i32 1}
>!85 = !{i32 1, i32 0, i32 0, i32 0, i32 2, i32 2, i32 0, i32 1}
>!86 = !{!"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none"}
>!87 = !{!"float*", !"uint", !"uint", !"float2", !"float*", !"float*", !"uint", !"float*"}
>!88 = !{!"float*", !"uint", !"uint", !"float __attribute__((ext_vector_type(2)))", !"float*", !"float*", !"uint", !"float*"}
>!89 = !{!"const", !"", !"", !"", !"const", !"const", !"", !""}
>!90 = !{i32 1, i32 1, i32 0, i32 0, i32 0, i32 0, i32 3, i32 3, i32 1, i32 0, i32 0, i32 0, i32 1}
>!91 = !{!"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none", !"none"}
>!92 = !{!"float*", !"float4*", !"uint", !"ChannelType", !"uint", !"uint", !"float4*", !"float*", !"float*", !"uint", !"float", !"float", !"float*"}
>!93 = !{!"float*", !"float __attribute__((ext_vector_type(4)))*", !"uint", !"ChannelType", !"uint", !"uint", !"float __attribute__((ext_vector_type(4)))*", !"float*", !"float*", !"uint", !"float", !"float", !"float*"}
>!94 = !{!"const", !"const", !"", !"", !"", !"", !"", !"", !"const", !"", !"", !"", !""}
>!95 = !{i32 1, i32 0, i32 0, i32 2, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 1}
>!96 = !{!"float*", !"uint", !"ChannelType", !"float*", !"uint", !"uint", !"uint", !"float4*", !"float", !"float", !"float*"}
>!97 = !{!"float*", !"uint", !"ChannelType", !"float*", !"uint", !"uint", !"uint", !"float __attribute__((ext_vector_type(4)))*", !"float", !"float", !"float*"}
>!98 = !{!"const", !"", !"", !"const", !"", !"", !"", !"", !"", !"", !""}
>!99 = !{i32 1, i32 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0}
>!100 = !{!"float*", !"float*", !"uint", !"uint", !"float", !"int", !"uint", !"uint"}
>!101 = !{!"", !"", !"", !"", !"", !"", !"", !""}
>!102 = !{i32 64, i32 4, i32 1}

Actions: View

Attachments on bug 105200: 137519 | 137571 | 137572 | 137573

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.