AMDGPU/SI: Fix implemenation of isSourceOfDivergence() for graphics shaders

[• tstellarAMD]

Summary

The analysis of shader inputs was completely wrong. We were passing the
wrong index to AttributeSet::hasAttribute() and the logic for which
inputs where in SGPRs was wrong too.

[arsenm]

LGTM

lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp

133 ↗

(On Diff #43134)

Typo marekd

test/Analysis/DivergenceAnalysis/AMDGPU/kernel-args.ll

3 ↗

(On Diff #43134)

This can be the -mtriple option to opt

test/Analysis/DivergenceAnalysis/AMDGPU/lit.local.cfg

1 ↗

(On Diff #43134)

This should be AMDGPU

llvm/trunk/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp

 static bool isArgPassedInSGPR(const Argument *A) {
   const Function *F = A->getParent();
   unsigned ShaderType = AMDGPU::getShaderType(*F);
 
   // Arguments to compute shaders are never a source of divergence.
   if (ShaderType == ShaderType::COMPUTE)
     return true;
 
-  // For non-compute shaders, the inreg attribute is used to mark inputs,
-  // which pre-loaded into SGPRs.
-  if (F->getAttributes().hasAttribute(A->getArgNo(), Attribute::InReg))
+  // For non-compute shaders, SGPR inputs are marked with either inreg or byval.
+  if (F->getAttributes().hasAttribute(A->getArgNo() + 1, Attribute::InReg) ||
+      F->getAttributes().hasAttribute(A->getArgNo() + 1, Attribute::ByVal))
     return true;
 
-  // For non-compute shaders, 32-bit values are pre-loaded into vgprs, all
-  // other value types use SGPRS.
-  return !A->getType()->isIntegerTy(32) && !A->getType()->isFloatTy();
+  // Everything else is in VGPRs.
+  return false;
 }
 
 ///
 /// \returns true if the result of the value could potentially be
 /// different across workitems in a wavefront.
 bool AMDGPUTTIImpl::isSourceOfDivergence(const Value *V) const {
 
   if (const Argument *A = dyn_cast<Argument>(V))

llvm/trunk/test/Analysis/DivergenceAnalysis/AMDGPU/kernel-args.ll

+; RUN: opt %s -mtriple amdgcn-- -analyze -divergence | FileCheck %s
+
+; CHECK: DIVERGENT:
+; CHECK-NOT: %arg0
+; CHECK-NOT: %arg1
+; CHECK-NOT; %arg2
+; CHECK: <2 x i32> %arg3
+; CHECK: DIVERGENT:  <3 x i32> %arg4
+; CHECK: DIVERGENT:  float %arg5
+; CHECK: DIVERGENT:  i32 %arg6
+
+define void @main([4 x <16 x i8>] addrspace(2)* byval %arg0, float inreg %arg1, i32 inreg %arg2, <2 x i32> %arg3, <3 x i32> %arg4, float %arg5, i32 %arg6) #0 {
+  ret void
+}
+
+attributes #0 = { "ShaderType"="0" }

llvm/trunk/test/Analysis/DivergenceAnalysis/AMDGPU/lit.local.cfg

+if not 'AMDGPU' in config.root.targets:
+    config.unsupported = True