[PATCH] D23131: AMDGPU: Fix an interaction between WQM and polygon stippling

[Nicolai Hähnle via llvm-commits]

nhaehnle created this revision.
nhaehnle added reviewers: arsenm, tstellarAMD, mareko.
nhaehnle added a subscriber: llvm-commits.
Herald added subscribers: kzhuravl, arsenm.

This fixes a rare bug in polygon stippling with non-monolithic pixel shaders.

The underlying problem is as follows: the prolog part contains the polygon
stippling sequence, i.e. a kill. The main part then enables WQM based on the
_reduced_ exec mask, effectively undoing most of the polygon stippling.

Since we cannot know whether polygon stippling will be used, the main part
of a non-monolithic shader must always return to exact mode to fix this
problem.

https://reviews.llvm.org/D23131

Files:
  lib/Target/AMDGPU/SIWholeQuadMode.cpp
  test/CodeGen/AMDGPU/wqm.ll

Index: test/CodeGen/AMDGPU/wqm.ll
===================================================================
--- test/CodeGen/AMDGPU/wqm.ll
+++ test/CodeGen/AMDGPU/wqm.ll
@@ -17,17 +17,18 @@
 ;CHECK-LABEL: {{^}}test2:
 ;CHECK-NEXT: ; %main_body
 ;CHECK-NEXT: s_wqm_b64 exec, exec
-;CHECK: image_sample
 ;CHECK-NOT: exec
-;CHECK: _load_dword v0,
-define amdgpu_ps float @test2(<8 x i32> inreg %rsrc, <4 x i32> inreg %sampler, float addrspace(1)* inreg %ptr, <4 x i32> %c) {
+define amdgpu_ps void @test2(<8 x i32> inreg %rsrc, <4 x i32> inreg %sampler, float addrspace(1)* inreg %ptr, <4 x i32> %c) {
 main_body:
   %c.1 = call <4 x float> @llvm.SI.image.sample.v4i32(<4 x i32> %c, <8 x i32> %rsrc, <4 x i32> %sampler, i32 15, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0)
   %c.2 = bitcast <4 x float> %c.1 to <4 x i32>
   %c.3 = extractelement <4 x i32> %c.2, i32 0
   %gep = getelementptr float, float addrspace(1)* %ptr, i32 %c.3
   %data = load float, float addrspace(1)* %gep
-  ret float %data
+
+  call void @llvm.SI.export(i32 15, i32 1, i32 1, i32 0, i32 1, float %data, float undef, float undef, float undef)
+
+  ret void
 }
 
 ; ... but disabled for stores (and, in this simple case, not re-enabled).
@@ -414,6 +415,23 @@
   ret void
 }
 
+; Must return to exact at the end of a non-void returning shader,
+; otherwise the EXEC mask exported by the epilog will be wrong. This is true
+; even if the shader has no kills, because a kill could have happened in a
+; previous shader fragment.
+;
+; CHECK-LABEL: {{^}}test_nonvoid_return:
+; CHECK: s_mov_b64 [[LIVE:s\[[0-9]+:[0-9]+\]]], exec
+; CHECK: s_wqm_b64 exec, exec
+;
+; CHECK: s_and_b64 exec, exec, [[LIVE]]
+; CHECK-NOT: exec
+define amdgpu_ps <4 x float> @test_nonvoid_return() nounwind {
+  %tex = call <4 x float> @llvm.SI.image.sample.v4i32(<4 x i32> undef, <8 x i32> undef, <4 x i32> undef, i32 15, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0)
+  %tex.i = bitcast <4 x float> %tex to <4 x i32>
+  %dtex = call <4 x float> @llvm.SI.image.sample.v4i32(<4 x i32> %tex.i, <8 x i32> undef, <4 x i32> undef, i32 15, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0)
+  ret <4 x float> %dtex
+}
 
 declare void @llvm.amdgcn.image.store.v4i32(<4 x float>, <4 x i32>, <8 x i32>, i32, i1, i1, i1, i1) #1
 declare void @llvm.amdgcn.buffer.store.f32(float, <4 x i32>, i32, i32, i1, i1) #1
Index: lib/Target/AMDGPU/SIWholeQuadMode.cpp
===================================================================
--- lib/Target/AMDGPU/SIWholeQuadMode.cpp
+++ lib/Target/AMDGPU/SIWholeQuadMode.cpp
@@ -220,8 +220,9 @@
       GlobalFlags |= Flags;
     }
 
-    if (WQMOutputs && MBB.succ_empty()) {
-      // This is a prolog shader. Make sure we go back to exact mode at the end.
+    // Go back to exact mode at the end of non-monolithic shader parts.
+    if (MBB.succ_empty() &&
+        !MF.getInfo<SIMachineFunctionInfo>()->returnsVoid()) {
       Blocks[&MBB].OutNeeds = StateExact;
       Worklist.push_back(&MBB);
       GlobalFlags |= StateExact;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D23131.66697.patch
Type: text/x-patch
Size: 3005 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160803/a578f7f3/attachment.bin>