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

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
   2 // 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
   3 // 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
   4
   5 #pragma omp declare target
   6
   7 float fv, fx;
   8 double dv, dx;
   9
  10 // DEFAULT-LABEL: define hidden void @_Z15atomic_fadd_f32v(
  11 // DEFAULT-SAME: ) #[[ATTR0:[0-9]+]] {
  12 // DEFAULT-NEXT:  [[ENTRY:.*:]]
  13 // DEFAULT-NEXT:    [[TMP0:%.*]] = load float, ptr addrspacecast (ptr addrspace(1) @fv to ptr), align 4
  14 // DEFAULT-NEXT:    [[TMP1:%.*]] = atomicrmw fadd ptr addrspacecast (ptr addrspace(1) @fx to ptr), float [[TMP0]] monotonic, align 4
  15 // DEFAULT-NEXT:    [[ADD:%.*]] = fadd float [[TMP1]], [[TMP0]]
  16 // DEFAULT-NEXT:    store float [[ADD]], ptr addrspacecast (ptr addrspace(1) @fv to ptr), align 4
  17 // DEFAULT-NEXT:    ret void
  18 //
  19 // UNSAFE-FP-ATOMICS-LABEL: define hidden void @_Z15atomic_fadd_f32v(
  20 // UNSAFE-FP-ATOMICS-SAME: ) #[[ATTR0:[0-9]+]] {
  21 // UNSAFE-FP-ATOMICS-NEXT:  [[ENTRY:.*:]]
  22 // UNSAFE-FP-ATOMICS-NEXT:    [[TMP0:%.*]] = load float, ptr addrspacecast (ptr addrspace(1) @fv to ptr), align 4
  23 // 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.ignore.denormal.mode [[META5]]
  24 // UNSAFE-FP-ATOMICS-NEXT:    [[ADD:%.*]] = fadd float [[TMP1]], [[TMP0]]
  25 // UNSAFE-FP-ATOMICS-NEXT:    store float [[ADD]], ptr addrspacecast (ptr addrspace(1) @fv to ptr), align 4
  26 // UNSAFE-FP-ATOMICS-NEXT:    ret void
  27 //
  28 void atomic_fadd_f32() {
  29 #pragma omp atomic capture
  30   fv = fx = fx + fv;
  31 }
  32
  33 // DEFAULT-LABEL: define hidden void @_Z15atomic_fadd_f64v(
  34 // DEFAULT-SAME: ) #[[ATTR0]] {
  35 // DEFAULT-NEXT:  [[ENTRY:.*:]]
  36 // DEFAULT-NEXT:    [[TMP0:%.*]] = load double, ptr addrspacecast (ptr addrspace(1) @dv to ptr), align 8
  37 // DEFAULT-NEXT:    [[TMP1:%.*]] = atomicrmw fadd ptr addrspacecast (ptr addrspace(1) @dx to ptr), double [[TMP0]] monotonic, align 8
  38 // DEFAULT-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], [[TMP0]]
  39 // DEFAULT-NEXT:    store double [[ADD]], ptr addrspacecast (ptr addrspace(1) @dv to ptr), align 8
  40 // DEFAULT-NEXT:    ret void
  41 //
  42 // UNSAFE-FP-ATOMICS-LABEL: define hidden void @_Z15atomic_fadd_f64v(
  43 // UNSAFE-FP-ATOMICS-SAME: ) #[[ATTR0]] {
  44 // UNSAFE-FP-ATOMICS-NEXT:  [[ENTRY:.*:]]
  45 // UNSAFE-FP-ATOMICS-NEXT:    [[TMP0:%.*]] = load double, ptr addrspacecast (ptr addrspace(1) @dv to ptr), align 8
  46 // 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]]
  47 // UNSAFE-FP-ATOMICS-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], [[TMP0]]
  48 // UNSAFE-FP-ATOMICS-NEXT:    store double [[ADD]], ptr addrspacecast (ptr addrspace(1) @dv to ptr), align 8
  49 // UNSAFE-FP-ATOMICS-NEXT:    ret void
  50 //
  51 void atomic_fadd_f64() {
  52 #pragma omp atomic capture
  53   dv = dx = dx + dv;
  54 }
  55
  56 #pragma omp end declare target
  57 //.
  58 // UNSAFE-FP-ATOMICS: [[META5]] = !{}
  59 //.