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

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.

Reviewers: arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15608

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256082 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp b/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
index eb9e837d2914f99a10873756774d5abfe8be911d..c20a66e44817a2596ee03a08ecbe603cf5caef65 100644 (file)
--- a/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
+++ b/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
@@ -141,14 +141,13 @@ static bool isArgPassedInSGPR(const Argument *A) {
   if (ShaderType == ShaderType::COMPUTE)
     return true;
 
   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;
 
     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;

 }
 
 ///
 }
 
 ///

diff --git a/test/Analysis/DivergenceAnalysis/AMDGPU/kernel-args.ll b/test/Analysis/DivergenceAnalysis/AMDGPU/kernel-args.ll
new file mode 100644 (file)
index 0000000..63e6c5c
--- /dev/null
+++ b/test/Analysis/DivergenceAnalysis/AMDGPU/kernel-args.ll
@@ -0,0 +1,16 @@
+; 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" }

diff --git a/test/Analysis/DivergenceAnalysis/AMDGPU/lit.local.cfg b/test/Analysis/DivergenceAnalysis/AMDGPU/lit.local.cfg
new file mode 100644 (file)
index 0000000..2a665f0
--- /dev/null
+++ b/test/Analysis/DivergenceAnalysis/AMDGPU/lit.local.cfg
@@ -0,0 +1,2 @@
+if not 'AMDGPU' in config.root.targets:
+    config.unsupported = True