1 // RUN: %clang_cc1 %s -emit-llvm -o - -triple=amdgcn-amd-amdhsa \
2 // RUN:   -fcuda-is-device -target-cpu gfx906 -fnative-half-type \
3 // RUN:   -fnative-half-arguments-and-returns | FileCheck %s
4 
5 // REQUIRES: amdgpu-registered-target
6 
7 #include "Inputs/cuda.h"
8 #include <stdatomic.h>
9 
10 __device__ float ffp1(float *p) {
11   // CHECK-LABEL: @_Z4ffp1Pf
12   // CHECK: atomicrmw fadd float* {{.*}} monotonic
13   return __atomic_fetch_add(p, 1.0f, memory_order_relaxed);
14 }
15 
16 __device__ double ffp2(double *p) {
17   // CHECK-LABEL: @_Z4ffp2Pd
18   // CHECK: atomicrmw fsub double* {{.*}} monotonic
19   return __atomic_fetch_sub(p, 1.0, memory_order_relaxed);
20 }
21 
22 // long double is the same as double for amdgcn.
23 __device__ long double ffp3(long double *p) {
24   // CHECK-LABEL: @_Z4ffp3Pe
25   // CHECK: atomicrmw fsub double* {{.*}} monotonic
26   return __atomic_fetch_sub(p, 1.0L, memory_order_relaxed);
27 }
28 
29 __device__ double ffp4(double *p, float f) {
30   // CHECK-LABEL: @_Z4ffp4Pdf
31   // CHECK: fpext float {{.*}} to double
32   // CHECK: atomicrmw fsub double* {{.*}} monotonic
33   return __atomic_fetch_sub(p, f, memory_order_relaxed);
34 }
35 
36 __device__ double ffp5(double *p, int i) {
37   // CHECK-LABEL: @_Z4ffp5Pdi
38   // CHECK: sitofp i32 {{.*}} to double
39   // CHECK: atomicrmw fsub double* {{.*}} monotonic
40   return __atomic_fetch_sub(p, i, memory_order_relaxed);
41 }
42