[clang] [Clang][AMDGPU] Permit language address spaces for AMDGPU globals (PR #66205)

Joseph Huber via cfe-commits cfe-commits at lists.llvm.org
Wed Sep 13 06:00:23 PDT 2023


https://github.com/jhuber6 created https://github.com/llvm/llvm-project/pull/66205:

Summary:
Currently, there is an assertion that prevents us from emitting an
AMDGPU global with a non-target specific address space (i.e. numerical
attribute). I'm unsure what the original intentions of this assertion
were, but we should be able to use OpenCL address spaces when compiling
directly to AMDGPU from C++. This is permitted on NVPTX so I'm unsure
what this assertion is guarding. The patch simply removes the assertion
and adds a test to ensure that these emit the expected address spaces.

Fixes https://github.com/llvm/llvm-project/issues/65069


>From 9447db0049e01be5c936bcd5f0f704f122282a13 Mon Sep 17 00:00:00 2001
From: Joseph Huber <jhuber6 at vols.utk.edu>
Date: Wed, 13 Sep 2023 07:57:47 -0500
Subject: [PATCH] [Clang][AMDGPU] Permit language address spaces for AMDGPU
 globals

Summary:
Currently, there is an assertion that prevents us from emitting an
AMDGPU global with a non-target specific address space (i.e. numerical
attribute). I'm unsure what the original intentions of this assertion
were, but we should be able to use OpenCL address spaces when compiling
directly to AMDGPU from C++. This is permitted on NVPTX so I'm unsure
what this assertion is guarding. The patch simply removes the assertion
and adds a test to ensure that these emit the expected address spaces.

Fixes https://github.com/llvm/llvm-project/issues/65069
---
 clang/lib/CodeGen/Targets/AMDGPU.cpp         |  1 -
 clang/test/CodeGen/amdgpu-address-spaces.cpp | 31 ++++++++++++++++++++
 2 files changed, 31 insertions(+), 1 deletion(-)
 create mode 100644 clang/test/CodeGen/amdgpu-address-spaces.cpp

diff --git a/clang/lib/CodeGen/Targets/AMDGPU.cpp b/clang/lib/CodeGen/Targets/AMDGPU.cpp
index c168bd4b7c7cc15..5aca4e540752524 100644
--- a/clang/lib/CodeGen/Targets/AMDGPU.cpp
+++ b/clang/lib/CodeGen/Targets/AMDGPU.cpp
@@ -438,7 +438,6 @@ AMDGPUTargetCodeGenInfo::getGlobalVarAddressSpace(CodeGenModule &CGM,
     return DefaultGlobalAS;
 
   LangAS AddrSpace = D->getType().getAddressSpace();
-  assert(AddrSpace == LangAS::Default || isTargetAddressSpace(AddrSpace));
   if (AddrSpace != LangAS::Default)
     return AddrSpace;
 
diff --git a/clang/test/CodeGen/amdgpu-address-spaces.cpp b/clang/test/CodeGen/amdgpu-address-spaces.cpp
new file mode 100644
index 000000000000000..e247cbb911added
--- /dev/null
+++ b/clang/test/CodeGen/amdgpu-address-spaces.cpp
@@ -0,0 +1,31 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-globals --version 3
+// RUN: %clang_cc1 -cc1 -triple amdgcn-amd-amdhsa -emit-llvm %s -o - | FileCheck %s
+
+int [[clang::opencl_global]] a = 0;
+int [[clang::opencl_generic]] b = 0;
+int [[clang::opencl_constant]] c = 0;
+[[clang::loader_uninitialized]] int [[clang::opencl_local]] d;
+[[clang::loader_uninitialized]] int [[clang::opencl_private]] e;
+
+//.
+// CHECK: @a = addrspace(1) global i32 0, align 4
+// CHECK: @b = global i32 0, align 4
+// CHECK: @c = addrspace(4) constant i32 0, align 4
+// CHECK: @d = addrspace(3) global i32 undef, align 4
+// CHECK: @e = addrspace(5) global i32 undef, align 4
+//.
+// CHECK-LABEL: define dso_local amdgpu_kernel void @_Z3foov(
+// CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    store i32 1, ptr addrspace(1) @a, align 4
+// CHECK-NEXT:    store i32 1, ptr @b, align 4
+// CHECK-NEXT:    store i32 1, ptr addrspace(3) @d, align 4
+// CHECK-NEXT:    store i32 1, ptr addrspace(5) @e, align 4
+// CHECK-NEXT:    ret void
+//
+[[clang::amdgpu_kernel]] void foo() {
+  a = 1;
+  b = 1;
+  d = 1;
+  e = 1;
+}



More information about the cfe-commits mailing list