Files
TheRock/compiler/amd-llvm/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomicrmw-integer-ops-0-to-add-0.ll
T
derek.holloway 7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
Flatten subprojects into a single base monorepo
2026-09-02 16:34:47 -07:00

375 lines
20 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 4
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes='require<libcall-lowering-info>,atomic-expand' -mcpu=gfx803 %s | FileCheck -check-prefixes=CHECK,GFX803 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes='require<libcall-lowering-info>,atomic-expand' -mcpu=gfx900 %s | FileCheck -check-prefixes=CHECK,GFX900 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes='require<libcall-lowering-info>,atomic-expand' -mcpu=gfx90a %s | FileCheck -check-prefixes=CHECK,GFX90A %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes='require<libcall-lowering-info>,atomic-expand' -mcpu=gfx1030 %s | FileCheck -check-prefixes=CHECK,GFX10 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes='require<libcall-lowering-info>,atomic-expand' -mcpu=gfx1100 %s | FileCheck -check-prefixes=CHECK,GFX11 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes='require<libcall-lowering-info>,atomic-expand' -mcpu=gfx942 %s | FileCheck -check-prefixes=CHECK,GFX942 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes='require<libcall-lowering-info>,atomic-expand' -mcpu=gfx1200 %s | FileCheck -check-prefixes=CHECK,GFX12 %s
; Test that system scoped atomicrmw or 0 is transformed to add 0.
; Transform to add
define i32 @test_atomicrmw_or_0_global_system(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_global_system(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !foo.md [[META0:![0-9]+]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 0 seq_cst, !foo.md !0
ret i32 %res
}
; Transform to add
define i32 @test_atomicrmw_or_0_global_one_as(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_global_one_as(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 syncscope("one-as") seq_cst, align 4
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 0 syncscope("one-as") seq_cst
ret i32 %res
}
; Transform to add
define i32 @test_atomicrmw_or_0_flat_system(ptr %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_flat_system(
; CHECK-SAME: ptr [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]])
; CHECK-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]]
; CHECK: atomicrmw.private:
; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5)
; CHECK-NEXT: [[LOADED_PRIVATE:%.*]] = load i32, ptr addrspace(5) [[TMP1]], align 4
; CHECK-NEXT: [[NEW:%.*]] = add i32 [[LOADED_PRIVATE]], 0
; CHECK-NEXT: store i32 [[NEW]], ptr addrspace(5) [[TMP1]], align 4
; CHECK-NEXT: br label [[ATOMICRMW_PHI:%.*]]
; CHECK: atomicrmw.global:
; CHECK-NEXT: [[TMP2:%.*]] = atomicrmw add ptr [[PTR]], i32 0 seq_cst, align 4, !noalias.addrspace [[META1:![0-9]+]]
; CHECK-NEXT: br label [[ATOMICRMW_PHI]]
; CHECK: atomicrmw.phi:
; CHECK-NEXT: [[RES1:%.*]] = phi i32 [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ]
; CHECK-NEXT: br label [[ATOMICRMW_END:%.*]]
; CHECK: atomicrmw.end:
; CHECK-NEXT: ret i32 [[RES1]]
;
%res = atomicrmw or ptr %ptr, i32 0 seq_cst
ret i32 %res
}
; Transform to add
define i32 @test_atomicrmw_or_0_as999_system(ptr addrspace(999) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_as999_system(
; CHECK-SAME: ptr addrspace(999) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(999) [[PTR]], i32 0 seq_cst, align 4
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(999) %ptr, i32 0 seq_cst
ret i32 %res
}
; Leave as-is, only system scope should be changed.
define i32 @test_atomicrmw_or_0_global_agent(ptr addrspace(1) %ptr) {
; GFX803-LABEL: define i32 @test_atomicrmw_or_0_global_agent(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; GFX803-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4
; GFX803-NEXT: ret i32 [[RES]]
;
; GFX900-LABEL: define i32 @test_atomicrmw_or_0_global_agent(
; GFX900-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; GFX900-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4
; GFX900-NEXT: ret i32 [[RES]]
;
; GFX90A-LABEL: define i32 @test_atomicrmw_or_0_global_agent(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; GFX90A-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4
; GFX90A-NEXT: ret i32 [[RES]]
;
; GFX10-LABEL: define i32 @test_atomicrmw_or_0_global_agent(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; GFX10-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4
; GFX10-NEXT: ret i32 [[RES]]
;
; GFX11-LABEL: define i32 @test_atomicrmw_or_0_global_agent(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; GFX11-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4
; GFX11-NEXT: ret i32 [[RES]]
;
; GFX942-LABEL: define i32 @test_atomicrmw_or_0_global_agent(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; GFX942-NEXT: [[RES:%.*]] = atomicrmw or ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4
; GFX942-NEXT: ret i32 [[RES]]
;
; GFX12-LABEL: define i32 @test_atomicrmw_or_0_global_agent(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; GFX12-NEXT: [[RES:%.*]] = atomicrmw or ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4
; GFX12-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst
ret i32 %res
}
; Leave as-is, LDS atomics aren't relevant.
define i32 @test_atomicrmw_or_0_local(ptr addrspace(3) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_local(
; CHECK-SAME: ptr addrspace(3) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw or ptr addrspace(3) [[PTR]], i32 0 seq_cst, align 4
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(3) %ptr, i32 0 seq_cst
ret i32 %res
}
; Leave non-0 values alone.
define i32 @test_atomicrmw_or_1_global_system(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_1_global_system(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(1) [[PTR]], align 4
; CHECK-NEXT: br label [[ATOMICRMW_START:%.*]]
; CHECK: atomicrmw.start:
; CHECK-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT: [[NEW:%.*]] = or i32 [[LOADED]], 1
; CHECK-NEXT: [[TMP2:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[NEW]] seq_cst seq_cst, align 4
; CHECK-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP2]], 1
; CHECK-NEXT: [[RES]] = extractvalue { i32, i1 } [[TMP2]], 0
; CHECK-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK: atomicrmw.end:
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 1 seq_cst
ret i32 %res
}
define i32 @test_atomicrmw_or_var_global_system(ptr addrspace(1) %ptr, i32 %val) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_var_global_system(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]], i32 [[VAL:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(1) [[PTR]], align 4
; CHECK-NEXT: br label [[ATOMICRMW_START:%.*]]
; CHECK: atomicrmw.start:
; CHECK-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT: [[NEW:%.*]] = or i32 [[LOADED]], [[VAL]]
; CHECK-NEXT: [[TMP2:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[NEW]] seq_cst seq_cst, align 4
; CHECK-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP2]], 1
; CHECK-NEXT: [[RES]] = extractvalue { i32, i1 } [[TMP2]], 0
; CHECK-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK: atomicrmw.end:
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 %val seq_cst
ret i32 %res
}
; Leave as-is
define i32 @test_atomicrmw_add_0_global_system(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_add_0_global_system(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw add ptr addrspace(1) %ptr, i32 0 seq_cst
ret i32 %res
}
; Transform to add
define i32 @test_atomicrmw_sub_0_global_system(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_sub_0_global_system(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !foo.md [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw sub ptr addrspace(1) %ptr, i32 0 seq_cst, !foo.md !0
ret i32 %res
}
; Transform to add
define i32 @test_atomicrmw_xor_0_global_system(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_xor_0_global_system(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw add ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !foo.md [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw xor ptr addrspace(1) %ptr, i32 0 seq_cst, !foo.md !0
ret i32 %res
}
define i32 @test_atomicrmw_or_0_global_system__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_global_system__amdgpu_no_fine_grained_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw or ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 0 seq_cst, !amdgpu.no.fine.grained.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_or_0_global_system__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_global_system__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw or ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 0 seq_cst, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_or_0_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw or ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 0 seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_xor_0_global_system__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_xor_0_global_system__amdgpu_no_fine_grained_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw xor ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw xor ptr addrspace(1) %ptr, i32 0 seq_cst, !amdgpu.no.fine.grained.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_xor_0_global_system__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_xor_0_global_system__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw xor ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw xor ptr addrspace(1) %ptr, i32 0 seq_cst, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_xor_0_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_xor_0_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw xor ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw xor ptr addrspace(1) %ptr, i32 0 seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_sub_0_global_system__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_sub_0_global_system__amdgpu_no_fine_grained_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw sub ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw sub ptr addrspace(1) %ptr, i32 0 seq_cst, !amdgpu.no.fine.grained.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_sub_0_global_system__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_sub_0_global_system__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw sub ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw sub ptr addrspace(1) %ptr, i32 0 seq_cst, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_sub_0_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_sub_0_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw sub ptr addrspace(1) [[PTR]], i32 0 seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw sub ptr addrspace(1) %ptr, i32 0 seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_or_0_global_agent__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_global_agent__amdgpu_no_fine_grained_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw or ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_or_0_global_agent__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_global_agent__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw or ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_or_0_global_agent__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_or_0_global_agent__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw or ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw or ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_xor_0_global_agent__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_xor_0_global_agent__amdgpu_no_fine_grained_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw xor ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw xor ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_xor_0_global_agent__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_xor_0_global_agent__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw xor ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw xor ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_xor_0_global_agent__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_xor_0_global_agent__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw xor ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw xor ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_sub_0_global_agent__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_sub_0_global_agent__amdgpu_no_fine_grained_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw sub ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw sub ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_sub_0_global_agent__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_sub_0_global_agent__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw sub ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw sub ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst, !amdgpu.no.remote.memory !0
ret i32 %res
}
define i32 @test_atomicrmw_sub_0_global_agent__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define i32 @test_atomicrmw_sub_0_global_agent__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RES:%.*]] = atomicrmw sub ptr addrspace(1) [[PTR]], i32 0 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RES]]
;
%res = atomicrmw sub ptr addrspace(1) %ptr, i32 0 syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
ret i32 %res
}
!0 = !{}