diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_optimize.c b/src/mesa/drivers/dri/r300/compiler/radeon_optimize.c index c15a9b1..d055930 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_optimize.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_optimize.c @@ -626,6 +626,21 @@ static void presub_replace_add(struct peephole_state *s, inst->U.I.SrcReg[src_index].Index = presub_opcode; } +static int is_presub_candidate(struct rc_instruction * inst) +{ + const struct rc_opcode_info * info = rc_get_opcode_info(inst->U.I.Opcode); + unsigned int i; + + if (inst->U.I.PreSub.Opcode != RC_PRESUB_NONE || inst->U.I.SaturateMode) + return 0; + + for(i = 0; i < info->NumSrcRegs; i++) { + if (src_reads_dst_mask(inst->U.I.SrcReg[i], inst->U.I.DstReg)) + return 0; + } + return 1; +} + static int peephole_add_presub_add( struct radeon_compiler * c, struct rc_instruction * inst_add) @@ -635,10 +650,7 @@ static int peephole_add_presub_add( unsigned int i; struct peephole_state s; - if (inst_add->U.I.PreSub.Opcode != RC_PRESUB_NONE) - return 0; - - if (inst_add->U.I.SaturateMode) + if (!is_presub_candidate(inst_add)) return 0; if (inst_add->U.I.SrcReg[0].Swizzle != inst_add->U.I.SrcReg[1].Swizzle) @@ -705,10 +717,7 @@ static int peephole_add_presub_inv( unsigned int i, swz, mask; struct peephole_state s; - if (inst_add->U.I.PreSub.Opcode != RC_PRESUB_NONE) - return 0; - - if (inst_add->U.I.SaturateMode) + if (!is_presub_candidate(inst_add)) return 0; mask = inst_add->U.I.DstReg.WriteMask;