[AMDGPU][clang] Fix __builtin_nontemporal_store() failure on AMDGPU
authorskc7 <Krishna.Sankisa@amd.com>
Wed, 1 Dec 2021 06:12:57 +0000 (06:12 +0000)
committerskc7 <Krishna.Sankisa@amd.com>
Thu, 2 Dec 2021 05:53:25 +0000 (05:53 +0000)
Reviewed By: yaxunl, sameerds

Differential Revision: https://reviews.llvm.org/D114849

clang/lib/CodeGen/CGBuiltin.cpp
clang/test/CodeGenOpenCL/amdgcn-non-temporal-store.cl [new file with mode: 0644]

index 849423c8b9baeb10112926c94b5348253c322057..5d6df59cc4059ef6da6d847a6ba2b901cdec2a9d 100644 (file)
@@ -170,8 +170,9 @@ static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) {
 
   // Convert the type of the pointer to a pointer to the stored type.
   Val = CGF.EmitToMemory(Val, E->getArg(0)->getType());
+  unsigned SrcAddrSpace = Address->getType()->getPointerAddressSpace();
   Value *BC = CGF.Builder.CreateBitCast(
-      Address, llvm::PointerType::getUnqual(Val->getType()), "cast");
+      Address, llvm::PointerType::get(Val->getType(), SrcAddrSpace), "cast");
   LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType());
   LV.setNontemporal(true);
   CGF.EmitStoreOfScalar(Val, LV, false);
diff --git a/clang/test/CodeGenOpenCL/amdgcn-non-temporal-store.cl b/clang/test/CodeGenOpenCL/amdgcn-non-temporal-store.cl
new file mode 100644 (file)
index 0000000..539d857
--- /dev/null
@@ -0,0 +1,8 @@
+// REQUIRES: amdgpu-registered-target
+// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -S -emit-llvm -o - %s | FileCheck %s
+// CHECK-LABEL: @test_non_temporal_store_kernel
+// CHECK: store i32 0, i32 addrspace(1)* %{{.*}}, align 4, !tbaa !{{.*}}, !nontemporal {{.*}}
+
+kernel void test_non_temporal_store_kernel(global unsigned int* io) {
+  __builtin_nontemporal_store(0, io);
+}