clang/test/OpenMP/amdgpu-unsafe-fp-atomics.cpp

// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple amdgcn-amd-amdhsa -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-is-target-device -o - | FileCheck -check-prefix=DEFAULT %s
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple amdgcn-amd-amdhsa -fopenmp-targets=amdgcn-amd-amdhsa -munsafe-fp-atomics -emit-llvm %s -fopenmp-is-target-device -o - | FileCheck -check-prefix=UNSAFE-FP-ATOMICS %s

#pragma omp declare target

float fv, fx;
double dv, dx;

// DEFAULT-LABEL: define hidden void @_Z15atomic_fadd_f32v(
// DEFAULT-SAME: ) #[[ATTR0:[0-9]+]] {
// DEFAULT-NEXT:  [[ENTRY:.*:]]
// DEFAULT-NEXT:    [[TMP0:%.*]] = load float, ptr addrspacecast (ptr addrspace(1) @fv to ptr), align 4
// DEFAULT-NEXT:    [[TMP1:%.*]] = atomicrmw fadd ptr addrspacecast (ptr addrspace(1) @fx to ptr), float [[TMP0]] monotonic, align 4, !amdgpu.no.fine.grained.memory [[META5:![0-9]+]], !amdgpu.no.remote.memory [[META5]]
// DEFAULT-NEXT:    [[ADD:%.*]] = fadd float [[TMP1]], [[TMP0]]
// DEFAULT-NEXT:    store float [[ADD]], ptr addrspacecast (ptr addrspace(1) @fv to ptr), align 4
// DEFAULT-NEXT:    ret void
//
// UNSAFE-FP-ATOMICS-LABEL: define hidden void @_Z15atomic_fadd_f32v(
// UNSAFE-FP-ATOMICS-SAME: ) #[[ATTR0:[0-9]+]] {
// UNSAFE-FP-ATOMICS-NEXT:  [[ENTRY:.*:]]
// UNSAFE-FP-ATOMICS-NEXT:    [[TMP0:%.*]] = load float, ptr addrspacecast (ptr addrspace(1) @fv to ptr), align 4
// UNSAFE-FP-ATOMICS-NEXT:    [[TMP1:%.*]] = atomicrmw fadd ptr addrspacecast (ptr addrspace(1) @fx to ptr), float [[TMP0]] monotonic, align 4, !amdgpu.no.fine.grained.memory [[META5:![0-9]+]], !amdgpu.no.remote.memory [[META5]], !amdgpu.ignore.denormal.mode [[META5]]
// UNSAFE-FP-ATOMICS-NEXT:    [[ADD:%.*]] = fadd float [[TMP1]], [[TMP0]]
// UNSAFE-FP-ATOMICS-NEXT:    store float [[ADD]], ptr addrspacecast (ptr addrspace(1) @fv to ptr), align 4
// UNSAFE-FP-ATOMICS-NEXT:    ret void
//
void atomic_fadd_f32() {
#pragma omp atomic capture
  fv = fx = fx + fv;
}

// DEFAULT-LABEL: define hidden void @_Z15atomic_fadd_f64v(
// DEFAULT-SAME: ) #[[ATTR0]] {
// DEFAULT-NEXT:  [[ENTRY:.*:]]
// DEFAULT-NEXT:    [[TMP0:%.*]] = load double, ptr addrspacecast (ptr addrspace(1) @dv to ptr), align 8
// DEFAULT-NEXT:    [[TMP1:%.*]] = atomicrmw fadd ptr addrspacecast (ptr addrspace(1) @dx to ptr), double [[TMP0]] monotonic, align 8, !amdgpu.no.fine.grained.memory [[META5]], !amdgpu.no.remote.memory [[META5]]
// DEFAULT-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], [[TMP0]]
// DEFAULT-NEXT:    store double [[ADD]], ptr addrspacecast (ptr addrspace(1) @dv to ptr), align 8
// DEFAULT-NEXT:    ret void
//
// UNSAFE-FP-ATOMICS-LABEL: define hidden void @_Z15atomic_fadd_f64v(
// UNSAFE-FP-ATOMICS-SAME: ) #[[ATTR0]] {
// UNSAFE-FP-ATOMICS-NEXT:  [[ENTRY:.*:]]
// UNSAFE-FP-ATOMICS-NEXT:    [[TMP0:%.*]] = load double, ptr addrspacecast (ptr addrspace(1) @dv to ptr), align 8
// UNSAFE-FP-ATOMICS-NEXT:    [[TMP1:%.*]] = atomicrmw fadd ptr addrspacecast (ptr addrspace(1) @dx to ptr), double [[TMP0]] monotonic, align 8, !amdgpu.no.fine.grained.memory [[META5]], !amdgpu.no.remote.memory [[META5]]
// UNSAFE-FP-ATOMICS-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], [[TMP0]]
// UNSAFE-FP-ATOMICS-NEXT:    store double [[ADD]], ptr addrspacecast (ptr addrspace(1) @dv to ptr), align 8
// UNSAFE-FP-ATOMICS-NEXT:    ret void
//
void atomic_fadd_f64() {
#pragma omp atomic capture
  dv = dx = dx + dv;
}

#pragma omp end declare target
//.
// DEFAULT: [[META5]] = !{}
//.
// UNSAFE-FP-ATOMICS: [[META5]] = !{}
//.