[llvm] r363561 - AMDGPU: Ignore subtarget for InferAddressSpaces

[Matt Arsenault via llvm-commits]

Author: arsenm
Date: Mon Jun 17 07:13:24 2019
New Revision: 363561

URL: http://llvm.org/viewvc/llvm-project?rev=363561&view=rev
Log:
AMDGPU: Ignore subtarget for InferAddressSpaces

Even if the target doesn't have flat instructions, addrspace(0) is
still flat. It just happens to not work.

Added:
    llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/no-flat-addrspace.ll
Modified:
    llvm/trunk/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.h

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.h?rev=363561&r1=363560&r2=363561&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.h (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.h Mon Jun 17 07:13:24 2019
@@ -180,8 +180,7 @@ public:
     // don't use flat addressing.
     if (IsGraphicsShader)
       return -1;
-    return ST->hasFlatAddressSpace() ?
-      AMDGPUAS::FLAT_ADDRESS : AMDGPUAS::UNKNOWN_ADDRESS_SPACE;
+    return AMDGPUAS::FLAT_ADDRESS;
   }
 
   unsigned getVectorSplitCost() { return 0; }

Added: llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/no-flat-addrspace.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/no-flat-addrspace.ll?rev=363561&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/no-flat-addrspace.ll (added)
+++ llvm/trunk/test/Transforms/InferAddressSpaces/AMDGPU/no-flat-addrspace.ll Mon Jun 17 07:13:24 2019
@@ -0,0 +1,13 @@
+; RUN: opt -S -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -infer-address-spaces %s | FileCheck %s
+
+; Make sure addrspace(0) is still treated as flat on targets without
+; flat instructions. It's still flat, it just doesn't work.
+
+; CHECK-LABEL: @load_flat_from_global(
+; CHECK-NEXT: %tmp1 = load float, float addrspace(1)* %ptr
+; CHECK-NEXT: ret float %tmp1
+define float @load_flat_from_global(float addrspace(1)*%ptr) #0 {
+  %tmp0 = addrspacecast float addrspace(1)* %ptr to float*
+  %tmp1 = load float, float* %tmp0
+  ret float %tmp1
+}