AMDGPU: mark llvm.amdgcn.image.atomic.* as a source of divergence

Summary:
When multiple threads perform an atomic op with the same arguments, they
will usually see different return values.

Reviewers: arsenm, tstellarAMD

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D18101

llvm-svn: 263440

2 files changed, 135 insertions, 0 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
index d1bb40aff9d..540ca5f399f 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
@@ -126,6 +126,19 @@ static bool isIntrinsicSourceOfDivergence(const TargetIntrinsicInfo *TII,
   case Intrinsic::r600_read_tidig_x:
   case Intrinsic::r600_read_tidig_y:
   case Intrinsic::r600_read_tidig_z:
+  case Intrinsic::amdgcn_image_atomic_swap:
+  case Intrinsic::amdgcn_image_atomic_add:
+  case Intrinsic::amdgcn_image_atomic_sub:
+  case Intrinsic::amdgcn_image_atomic_smin:
+  case Intrinsic::amdgcn_image_atomic_umin:
+  case Intrinsic::amdgcn_image_atomic_smax:
+  case Intrinsic::amdgcn_image_atomic_umax:
+  case Intrinsic::amdgcn_image_atomic_and:
+  case Intrinsic::amdgcn_image_atomic_or:
+  case Intrinsic::amdgcn_image_atomic_xor:
+  case Intrinsic::amdgcn_image_atomic_inc:
+  case Intrinsic::amdgcn_image_atomic_dec:
+  case Intrinsic::amdgcn_image_atomic_cmpswap:
     return true;
   }
 
diff --git a/llvm/test/Analysis/DivergenceAnalysis/AMDGPU/llvm.amdgcn.image.atomic.ll b/llvm/test/Analysis/DivergenceAnalysis/AMDGPU/llvm.amdgcn.image.atomic.ll
new file mode 100644
index 00000000000..31102c9306d
--- /dev/null
+++ b/llvm/test/Analysis/DivergenceAnalysis/AMDGPU/llvm.amdgcn.image.atomic.ll
@@ -0,0 +1,122 @@
+;RUN: opt -mtriple=amdgcn-mesa-mesa3d -analyze -divergence %s | FileCheck %s
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.swap.i32(
+define float @image_atomic_swap(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.swap.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.add.i32(
+define float @image_atomic_add(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.add.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.sub.i32(
+define float @image_atomic_sub(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.sub.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.smin.i32(
+define float @image_atomic_smin(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.smin.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.umin.i32(
+define float @image_atomic_umin(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.umin.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.smax.i32(
+define float @image_atomic_smax(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.smax.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.umax.i32(
+define float @image_atomic_umax(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.umax.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.and.i32(
+define float @image_atomic_and(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.and.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.or.i32(
+define float @image_atomic_or(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.or.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.xor.i32(
+define float @image_atomic_xor(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.xor.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.inc.i32(
+define float @image_atomic_inc(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.inc.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.dec.i32(
+define float @image_atomic_dec(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.dec.i32(i32 %data, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+;CHECK: DIVERGENT: %orig = call i32 @llvm.amdgcn.image.atomic.cmpswap.i32(
+define float @image_atomic_cmpswap(<8 x i32> inreg %rsrc, i32 inreg %addr, i32 inreg %data, i32 inreg %cmp) #0 {
+main_body:
+  %orig = call i32 @llvm.amdgcn.image.atomic.cmpswap.i32(i32 %data, i32 %cmp, i32 %addr, <8 x i32> %rsrc, i1 0, i1 0, i1 0)
+  %r = bitcast i32 %orig to float
+  ret float %r
+}
+
+declare i32 @llvm.amdgcn.image.atomic.swap.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.add.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.sub.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.smin.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.umin.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.smax.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.umax.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.and.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.or.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.xor.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.inc.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.dec.i32(i32, i32, <8 x i32>, i1, i1, i1) #1
+declare i32 @llvm.amdgcn.image.atomic.cmpswap.i32(i32, i32, i32, <8 x i32>,i1, i1, i1) #1
+
+attributes #0 = { "ShaderType"="0" }
+attributes #1 = { nounwind }