Bug 86618 - [NV96] neg modifiers not working in MIN and MAX operations
Summary: [NV96] neg modifiers not working in MIN and MAX operations
Status: RESOLVED FIXED
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/DRI/nouveau (show other bugs)
Version: git
Hardware: Other All
: medium normal
Assignee: Nouveau Project
QA Contact:
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2014-11-23 11:17 UTC by Grzegorz Wójcik
Modified: 2014-11-23 20:48 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Attachments

Description Grzegorz Wójcik 2014-11-23 11:17:05 UTC
Neg modifier for source in MIN and MAX ops is ignored.
Tested on git 89b9ef9.
I tested it only on NV96.

shader with:
float x = -sin(uv.x);
float d = max(0.0, x);
generates:
EMIT: presin f32 $r0 $r0 (8)
EMIT: sin f32 $r0 $r0 (8)
EMIT: mov u32 $r1 0x00000000 (8)
EMIT: max f32 $r2 neg $r0 $r1 (8)

Shader output is identical to version without unary minus.
Only difference is lack of neg modifier.

After disabling modifiers shader works as expected:

--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp
@@ -85,8 +85,8 @@ static const struct opProperties _initProps[] =
    { OP_ADD,    0x3, 0x0, 0x0, 0x8, 0x2, 0x1, 0x1, 0x2 },
    { OP_SUB,    0x3, 0x0, 0x0, 0x0, 0x2, 0x1, 0x1, 0x2 },
    { OP_MUL,    0x3, 0x0, 0x0, 0x0, 0x2, 0x1, 0x1, 0x2 },
-   { OP_MAX,    0x3, 0x3, 0x0, 0x0, 0x2, 0x1, 0x1, 0x0 },
-   { OP_MIN,    0x3, 0x3, 0x0, 0x0, 0x2, 0x1, 0x1, 0x0 },
+   { OP_MAX,    0x0, 0x3, 0x0, 0x0, 0x2, 0x1, 0x1, 0x0 },
+   { OP_MIN,    0x0, 0x3, 0x0, 0x0, 0x2, 0x1, 0x1, 0x0 },
    { OP_MAD,    0x7, 0x0, 0x0, 0x0, 0x6, 0x1, 0x1, 0x0 }, // special constraint
    { OP_ABS,    0x0, 0x0, 0x0, 0x0, 0x0, 0x1, 0x1, 0x0 },
    { OP_NEG,    0x0, 0x1, 0x0, 0x0, 0x0, 0x1, 0x1, 0x0 },

After this change:
EMIT: presin f32 $r0 $r0 (8)
EMIT: sin f32 $r0 $r0 (8)
EMIT: mov u32 $r1 0x00000000 (8)
EMIT: neg f32 $r0 $r0 (8)
EMIT: max f32 $r2 $r0 $r1 (8)
Comment 1 Ilia Mirkin 2014-11-23 17:20:41 UTC
Great find! However the min/max ops can actually take the neg modifiers (on either arg), but the emit logic didn't have that hooked up somehow.

Does http://patchwork.freedesktop.org/patch/37356/ resolve your issue?
Comment 2 Grzegorz Wójcik 2014-11-23 18:23:53 UTC
(In reply to Ilia Mirkin from comment #1)
Yes, this patch solved my issue.
Comment 3 Ilia Mirkin 2014-11-23 20:48:21 UTC
Pushed as 7d07083c upstream, should be backported to 10.3 and 10.4 releases.


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.