Commit 191d3dbb authored by Tom Stellard's avatar Tom Stellard
Browse files

Merging r264214:

------------------------------------------------------------------------
r264214 | Matthew.Arsenault | 2016-03-23 16:17:29 -0700 (Wed, 23 Mar 2016) | 2 lines

AMDGPU: Promote alloca should skip volatiles

------------------------------------------------------------------------

llvm-svn: 271729
parent de723c05
Loading
Loading
Loading
Loading
+13 −0
Original line number Diff line number Diff line
@@ -301,9 +301,22 @@ static bool collectUsesWithPtrTypes(Value *Val, std::vector<Value*> &WorkList) {
      return false;

    if (StoreInst *SI = dyn_cast_or_null<StoreInst>(UseInst)) {
      if (SI->isVolatile())
        return false;

      // Reject if the stored value is not the pointer operand.
      if (SI->getPointerOperand() != Val)
        return false;
    } else if (LoadInst *LI = dyn_cast_or_null<LoadInst>(UseInst)) {
      if (LI->isVolatile())
        return false;
    } else if (AtomicRMWInst *RMW = dyn_cast_or_null<AtomicRMWInst>(UseInst)) {
      if (RMW->isVolatile())
        return false;
    } else if (AtomicCmpXchgInst *CAS
               = dyn_cast_or_null<AtomicCmpXchgInst>(UseInst)) {
      if (CAS->isVolatile())
        return false;
    }

    if (!User->getType()->isPointerTy())
+26 −0
Original line number Diff line number Diff line
; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -amdgpu-promote-alloca < %s | FileCheck %s

; CHECK-LABEL: @volatile_load(
; CHECK: alloca [5 x i32]
; CHECK load volatile i32, i32*
define void @volatile_load(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {
entry:
  %stack = alloca [5 x i32], align 4
  %tmp = load i32, i32 addrspace(1)* %in, align 4
  %arrayidx1 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 %tmp
  %load = load volatile i32, i32* %arrayidx1
  store i32 %load, i32 addrspace(1)* %out
 ret void
}

; CHECK-LABEL: @volatile_store(
; CHECK: alloca [5 x i32]
; CHECK store volatile i32 %tmp, i32*
define void @volatile_store(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {
entry:
  %stack = alloca [5 x i32], align 4
  %tmp = load i32, i32 addrspace(1)* %in, align 4
  %arrayidx1 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 %tmp
  store volatile i32 %tmp, i32* %arrayidx1
 ret void
}