35 lines · plain
1// RUN: %clang_cc1 "-triple" "nvptx-nvidia-cuda" "-target-feature" "+ptx86" "-target-cpu" "sm_100a" -emit-llvm -fcuda-is-device -o - %s | FileCheck %s2// RUN: %clang_cc1 "-triple" "nvptx64-nvidia-cuda" "-target-feature" "+ptx86" "-target-cpu" "sm_100a" -emit-llvm -fcuda-is-device -o - %s | FileCheck %s3 4// CHECK: define{{.*}} void @_Z6kernelPf(ptr noundef %out_f)5__attribute__((global)) void kernel(float* out_f) {6 float a = 3.0;7 int i = 0;8 9 out_f[i++] = __nvvm_redux_sync_fmin(a, 0xFF);10 // CHECK: call contract float @llvm.nvvm.redux.sync.fmin11 12 out_f[i++] = __nvvm_redux_sync_fmin_abs(a, 0xFF);13 // CHECK: call contract float @llvm.nvvm.redux.sync.fmin.abs14 15 out_f[i++] = __nvvm_redux_sync_fmin_NaN(a, 0xF0);16 // CHECK: call contract float @llvm.nvvm.redux.sync.fmin.NaN17 18 out_f[i++] = __nvvm_redux_sync_fmin_abs_NaN(a, 0x0F);19 // CHECK: call contract float @llvm.nvvm.redux.sync.fmin.abs.NaN20 21 out_f[i++] = __nvvm_redux_sync_fmax(a, 0xFF);22 // CHECK: call contract float @llvm.nvvm.redux.sync.fmax23 24 out_f[i++] = __nvvm_redux_sync_fmax_abs(a, 0x01);25 // CHECK: call contract float @llvm.nvvm.redux.sync.fmax.abs26 27 out_f[i++] = __nvvm_redux_sync_fmax_NaN(a, 0xF1);28 // CHECK: call contract float @llvm.nvvm.redux.sync.fmax.NaN29 30 out_f[i++] = __nvvm_redux_sync_fmax_abs_NaN(a, 0x10);31 // CHECK: call contract float @llvm.nvvm.redux.sync.fmax.abs.NaN32 33 // CHECK: ret void34}35