[AMDGPU] Do not annotate features for graphics

[sebastian-ne]

Summary

SITargetLowering::LowerFormalArguments asserts that none of these
features are used for graphics calling conventions, so
AnnotateKernelFeatures should not add them.

[arsenm]

llvm/test/CodeGen/AMDGPU/pal-simple-indirect-call.ll

3

The IR checks look missing except for the label

[sebastian-ne]

llvm/test/CodeGen/AMDGPU/pal-simple-indirect-call.ll

3

This test is copied from simple-indirect-call.ll (it can’t be in the same file because amdhsa does not allow amdgpu_cs functions). The one check line still checks that there are no attributes added, but I can just remove the opt test if that is not worth it.

[madhur13490]

llvm/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp

406–408

Please put a comment about why you "continue" when particular CC is met.

llvm/test/CodeGen/AMDGPU/pal-simple-indirect-call.ll

3

Since this patch makes changes in the pass, I think opt check is required. llc line from simple-indirect-call.ll made sure that the test compiles fine. It's up to you if you want the similar assurance.

[sebastian-ne]

Add comment

[arsenm]

llvm/test/CodeGen/AMDGPU/pal-simple-indirect-call.ll

3

Can you add a comment indicating there should be no attribute on the function? This isn't the most obvious test

[sebastian-ne]

Good point, added comments to the test.

llvm/lib/Target/AMDGPU/AMDGPUAnnotateKernelFeatures.cpp

for (CallGraphNode *I : SCC) {
     if (I->getNumReferences())
       NodeList.push_back(I);
     else {
       processUniformWorkGroupAttribute();
       NodeList.clear();
     }
 
     Function *F = I->getFunction();
-    // Add feature attributes
-    if (!F || F->isDeclaration())
+    // Ignore functions with graphics calling conventions, these are currently
+    // not allowed to have kernel arguments.
+    if (!F || F->isDeclaration() || AMDGPU::isGraphics(F->getCallingConv()))
       continue;
+    // Add feature attributes
     Changed |= addFeatureAttributes(*F);
   }
 
   return Changed;
 }
 
 bool AMDGPUAnnotateKernelFeatures::doInitialization(CallGraph &CG) {
   auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();

llvm/test/CodeGen/AMDGPU/pal-simple-indirect-call.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: opt -S -mtriple=amdgcn-amd-amdpal -amdgpu-annotate-kernel-features  %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX9 %s
+
+target datalayout = "A5"
+
+define amdgpu_cs void @test_simple_indirect_call() {
+; GCN-LABEL: define amdgpu_cs void @test_simple_indirect_call() {
+; GFX9-LABEL: test_simple_indirect_call:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_getpc_b64 s[36:37]
+; GFX9-NEXT:    s_mov_b32 s36, s0
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_load_dwordx4 s[36:39], s[36:37], 0x10
+; GFX9-NEXT:    s_mov_b32 s32, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_add_u32 s36, s36, s0
+; GFX9-NEXT:    s_addc_u32 s37, s37, 0
+; GFX9-NEXT:    s_mov_b64 s[0:1], s[36:37]
+; GFX9-NEXT:    s_mov_b64 s[2:3], s[38:39]
+; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
+; GFX9-NEXT:    s_endpgm
+
+  %pc = call i64 @llvm.amdgcn.s.getpc()
+  %fun = inttoptr i64 %pc to void()*
+  call amdgpu_gfx void %fun()
+  ret void
+}
+
+; Function Attrs: nounwind readnone speculatable willreturn
+declare i64 @llvm.amdgcn.s.getpc() #0
+
+attributes #0 = { nounwind readnone speculatable willreturn }