AMDGPU: llvm.SI.fs.constant is a source of divergence

[nhaehnle]

Summary

This intrinsic is used to get flat-shaded fragment shader inputs. Those are
uniform across a primitive, but a fragment shader wave may process pixels from
multiple primitives (as indicated by the prim_mask), and so that's where
divergence can arise.

[arsenm]

LGTM

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

static bool isIntrinsicSourceOfDivergence(const TargetIntrinsicInfo *TII,
   }
 
   StringRef Name = I->getCalledFunction()->getName();
   switch (TII->lookupName((const char *)Name.bytes_begin(), Name.size())) {
   default:
     return false;
   case AMDGPUIntrinsic::SI_tid:
   case AMDGPUIntrinsic::SI_fs_interp:
+  case AMDGPUIntrinsic::SI_fs_constant:
     return true;
   }
 }
 
 static bool isArgPassedInSGPR(const Argument *A) {
   const Function *F = A->getParent();
 
   // Arguments to compute shaders are never a source of divergence.

llvm/trunk/test/Analysis/DivergenceAnalysis/AMDGPU/interp-intrinsics.ll

+; RUN: opt  -mtriple amdgcn--- -analyze -divergence %s | FileCheck %s
+
+; CHECK-LABEL: 'fs_interp'
+; CHECK: DIVERGENT: %v = call float @llvm.SI.fs.interp(
+define amdgpu_ps void @fs_interp(i32 inreg %prim_mask, <2 x i32> %interp_param) #1 {
+  %v = call float @llvm.SI.fs.interp(i32 0, i32 0, i32 %prim_mask, <2 x i32> %interp_param)
+  store volatile float %v, float addrspace(1)* undef
+  ret void
+}
+
+; CHECK-LABEL: 'fs_constant'
+; CHECK: DIVERGENT: %v = call float @llvm.SI.fs.constant(
+define amdgpu_ps void @fs_constant(i32 inreg %prim_mask, <2 x i32> %interp_param) #1 {
+  %v = call float @llvm.SI.fs.constant(i32 0, i32 0, i32 %prim_mask)
+  store volatile float %v, float addrspace(1)* undef
+  ret void
+}
+
+declare float @llvm.SI.fs.interp(i32, i32, i32, <2 x i32>) #0
+declare float @llvm.SI.fs.constant(i32, i32, i32) #0
+
+attributes #0 = { nounwind readnone }