[llvm] [AMDGPU] Lower mem intrinsics on LDS pointers in AMDGPUSwLowerLDS (PR #202552)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Tue Jun 9 02:22:19 PDT 2026
================
@@ -0,0 +1,107 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt < %s -passes=amdgpu-sw-lower-lds -S -mtriple=amdgcn-amd-amdhsa | FileCheck %s
+
+; Test to check that memset/memcpy/memmove on LDS pointers are lowered to the
+; global memory buffer that backs the lowered LDS, rather than left targeting
+; the LDS address space.
+
+ at lds_1 = internal addrspace(3) global [16 x i8] poison, align 4
+ at lds_2 = internal addrspace(3) global [16 x i8] poison, align 4
+
+define amdgpu_kernel void @memintrinsic_kernel(ptr addrspace(1) %src) sanitize_address {
+; CHECK-LABEL: define amdgpu_kernel void @memintrinsic_kernel(
+; CHECK-SAME: ptr addrspace(1) [[SRC:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[WID:.*]]:
+; CHECK-NEXT: [[TMP0:%.*]] = call i32 @llvm.amdgcn.workitem.id.x()
+; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.amdgcn.workitem.id.y()
+; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.amdgcn.workitem.id.z()
+; CHECK-NEXT: [[TMP3:%.*]] = or i32 [[TMP0]], [[TMP1]]
+; CHECK-NEXT: [[TMP4:%.*]] = or i32 [[TMP3]], [[TMP2]]
+; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i32 [[TMP4]], 0
+; CHECK-NEXT: br i1 [[TMP5]], label %[[MALLOC:.*]], label %[[BB20:.*]]
+; CHECK: [[MALLOC]]:
+; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_MEMINTRINSIC_KERNEL_MD_TYPE:%.*]], ptr addrspace(1) @llvm.amdgcn.sw.lds.memintrinsic_kernel.md, i32 0, i32 2, i32 0), align 4
+; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_MEMINTRINSIC_KERNEL_MD_TYPE]], ptr addrspace(1) @llvm.amdgcn.sw.lds.memintrinsic_kernel.md, i32 0, i32 2, i32 2), align 4
+; CHECK-NEXT: [[TMP8:%.*]] = add i32 [[TMP6]], [[TMP7]]
+; CHECK-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64
+; CHECK-NEXT: [[TMP10:%.*]] = call ptr @llvm.returnaddress.p0(i32 0)
+; CHECK-NEXT: [[TMP11:%.*]] = ptrtoint ptr [[TMP10]] to i64
+; CHECK-NEXT: [[TMP12:%.*]] = call i64 @__asan_malloc_impl(i64 [[TMP9]], i64 [[TMP11]])
+; CHECK-NEXT: [[TMP13:%.*]] = inttoptr i64 [[TMP12]] to ptr addrspace(1)
+; CHECK-NEXT: store ptr addrspace(1) [[TMP13]], ptr addrspace(3) @llvm.amdgcn.sw.lds.memintrinsic_kernel, align 8
+; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP13]], i64 8
+; CHECK-NEXT: [[TMP15:%.*]] = ptrtoint ptr addrspace(1) [[TMP14]] to i64
+; CHECK-NEXT: call void @__asan_poison_region(i64 [[TMP15]], i64 24)
+; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP13]], i64 48
+; CHECK-NEXT: [[TMP17:%.*]] = ptrtoint ptr addrspace(1) [[TMP16]] to i64
+; CHECK-NEXT: call void @__asan_poison_region(i64 [[TMP17]], i64 16)
+; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP13]], i64 80
+; CHECK-NEXT: [[TMP19:%.*]] = ptrtoint ptr addrspace(1) [[TMP18]] to i64
+; CHECK-NEXT: call void @__asan_poison_region(i64 [[TMP19]], i64 16)
+; CHECK-NEXT: br label %[[BB20]]
+; CHECK: [[BB20]]:
+; CHECK-NEXT: [[XYZCOND:%.*]] = phi i1 [ false, %[[WID]] ], [ true, %[[MALLOC]] ]
+; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()
+; CHECK-NEXT: [[TMP21:%.*]] = load ptr addrspace(1), ptr addrspace(3) @llvm.amdgcn.sw.lds.memintrinsic_kernel, align 8
+; CHECK-NEXT: [[TMP22:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_MEMINTRINSIC_KERNEL_MD_TYPE]], ptr addrspace(1) @llvm.amdgcn.sw.lds.memintrinsic_kernel.md, i32 0, i32 1, i32 0), align 4
+; CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds i8, ptr addrspace(3) @llvm.amdgcn.sw.lds.memintrinsic_kernel, i32 [[TMP22]]
+; CHECK-NEXT: [[TMP24:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_MEMINTRINSIC_KERNEL_MD_TYPE]], ptr addrspace(1) @llvm.amdgcn.sw.lds.memintrinsic_kernel.md, i32 0, i32 2, i32 0), align 4
+; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds i8, ptr addrspace(3) @llvm.amdgcn.sw.lds.memintrinsic_kernel, i32 [[TMP24]]
+; CHECK-NEXT: [[A:%.*]] = getelementptr [16 x i8], ptr addrspace(3) [[TMP23]], i32 0, i32 0
+; CHECK-NEXT: [[B:%.*]] = getelementptr [16 x i8], ptr addrspace(3) [[TMP25]], i32 0, i32 0
+; CHECK-NEXT: [[TMP26:%.*]] = ptrtoint ptr addrspace(3) [[A]] to i32
+; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP26]]
+; CHECK-NEXT: call void @llvm.memset.p1.i32(ptr addrspace(1) [[TMP27]], i8 0, i32 16, i1 false)
+; CHECK-NEXT: [[TMP28:%.*]] = ptrtoint ptr addrspace(3) [[A]] to i32
+; CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP28]]
+; CHECK-NEXT: call void @llvm.memcpy.p1.p1.i32(ptr addrspace(1) [[TMP29]], ptr addrspace(1) [[SRC]], i32 16, i1 false)
+; CHECK-NEXT: [[TMP30:%.*]] = ptrtoint ptr addrspace(3) [[A]] to i32
+; CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP30]]
+; CHECK-NEXT: [[TMP32:%.*]] = ptrtoint ptr addrspace(3) [[B]] to i32
+; CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP32]]
+; CHECK-NEXT: call void @llvm.memmove.p1.p1.i32(ptr addrspace(1) [[TMP31]], ptr addrspace(1) [[TMP33]], i32 16, i1 false)
+; CHECK-NEXT: [[TMP42:%.*]] = ptrtoint ptr addrspace(3) [[A]] to i32
+; CHECK-NEXT: [[TMP43:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP42]]
+; CHECK-NEXT: call void @llvm.memset.element.unordered.atomic.p1.i32(ptr addrspace(1) align 4 [[TMP43]], i8 0, i32 16, i32 4)
+; CHECK-NEXT: [[TMP44:%.*]] = ptrtoint ptr addrspace(3) [[A]] to i32
+; CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP44]]
+; CHECK-NEXT: call void @llvm.memcpy.element.unordered.atomic.p1.p1.i32(ptr addrspace(1) align 4 [[TMP37]], ptr addrspace(1) align 4 [[SRC]], i32 16, i32 4)
+; CHECK-NEXT: [[TMP38:%.*]] = ptrtoint ptr addrspace(3) [[A]] to i32
+; CHECK-NEXT: [[TMP39:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP38]]
+; CHECK-NEXT: [[TMP40:%.*]] = ptrtoint ptr addrspace(3) [[B]] to i32
+; CHECK-NEXT: [[TMP41:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP21]], i32 [[TMP40]]
+; CHECK-NEXT: call void @llvm.memmove.element.unordered.atomic.p1.p1.i32(ptr addrspace(1) align 4 [[TMP39]], ptr addrspace(1) align 4 [[TMP41]], i32 16, i32 4)
+; CHECK-NEXT: br label %[[CONDFREE:.*]]
+; CHECK: [[CONDFREE]]:
+; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()
+; CHECK-NEXT: br i1 [[XYZCOND]], label %[[FREE:.*]], label %[[END:.*]]
+; CHECK: [[FREE]]:
+; CHECK-NEXT: [[TMP34:%.*]] = call ptr @llvm.returnaddress.p0(i32 0)
+; CHECK-NEXT: [[TMP35:%.*]] = ptrtoint ptr [[TMP34]] to i64
+; CHECK-NEXT: [[TMP36:%.*]] = ptrtoint ptr addrspace(1) [[TMP21]] to i64
+; CHECK-NEXT: call void @__asan_free_impl(i64 [[TMP36]], i64 [[TMP35]])
+; CHECK-NEXT: br label %[[END]]
+; CHECK: [[END]]:
+; CHECK-NEXT: ret void
+;
+ %a = getelementptr [16 x i8], ptr addrspace(3) @lds_1, i32 0, i32 0
+ %b = getelementptr [16 x i8], ptr addrspace(3) @lds_2, i32 0, i32 0
+ call void @llvm.memset.p3.i32(ptr addrspace(3) %a, i8 0, i32 16, i1 false)
----------------
arsenm wrote:
I'd prefer one function per case
https://github.com/llvm/llvm-project/pull/202552
More information about the llvm-commits
mailing list