[llvm] r346267 - AMDGPU: Add an option -disable-promote-alloca-to-lds

[Yaxun Liu via llvm-commits]

Author: yaxunl
Date: Tue Nov  6 13:28:17 2018
New Revision: 346267

URL: http://llvm.org/viewvc/llvm-project?rev=346267&view=rev
Log:
AMDGPU: Add an option -disable-promote-alloca-to-lds

Add this option for debugging and providing workaround.

By default it is off so no behavior change in backend.

Differential Revision: https://reviews.llvm.org/D54158

Modified:
    llvm/trunk/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
    llvm/trunk/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp?rev=346267&r1=346266&r2=346267&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp Tue Nov  6 13:28:17 2018
@@ -70,6 +70,11 @@ static cl::opt<bool> DisablePromoteAlloc
   cl::desc("Disable promote alloca to vector"),
   cl::init(false));
 
+static cl::opt<bool> DisablePromoteAllocaToLDS(
+  "disable-promote-alloca-to-lds",
+  cl::desc("Disable promote alloca to LDS"),
+  cl::init(false));
+
 // FIXME: This can create globals so should be a module pass.
 class AMDGPUPromoteAlloca : public FunctionPass {
 private:
@@ -706,6 +711,9 @@ bool AMDGPUPromoteAlloca::handleAlloca(A
   if (tryPromoteAllocaToVector(&I))
     return true; // Promoted to vector.
 
+  if (DisablePromoteAllocaToLDS)
+    return false;
+
   const Function &ContainingFunction = *I.getParent()->getParent();
   CallingConv::ID CC = ContainingFunction.getCallingConv();
 

Modified: llvm/trunk/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll?rev=346267&r1=346266&r2=346267&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll Tue Nov  6 13:28:17 2018
@@ -1,8 +1,11 @@
 ; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck %s
+; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca -disable-promote-alloca-to-lds< %s | FileCheck -check-prefix=NOLDS %s
 
 ; This normally would be fixed by instcombine to be compare to the GEP
 ; indices
 
+; NOLDS-NOT: addrspace(3)
+
 ; CHECK-LABEL: @lds_promoted_alloca_icmp_same_derived_pointer(
 ; CHECK: [[ARRAYGEP:%[0-9]+]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_same_derived_pointer.alloca, i32 0, i32 %{{[0-9]+}}
 ; CHECK: %ptr0 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[ARRAYGEP]], i32 0, i32 %a