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
llvm-svn: 346267
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
index fe9e4ca0..ec7ea2b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
@@ -70,6 +70,11 @@
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 @@
if (tryPromoteAllocaToVector(&I))
return true; // Promoted to vector.
+ if (DisablePromoteAllocaToLDS)
+ return false;
+
const Function &ContainingFunction = *I.getParent()->getParent();
CallingConv::ID CC = ContainingFunction.getCallingConv();
diff --git a/llvm/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll b/llvm/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll
index ebef612..8d12a72 100644
--- a/llvm/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll
+++ b/llvm/test/CodeGen/AMDGPU/promote-alloca-to-lds-icmp.ll
@@ -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