AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix build with hip-clang.

Browse Source 
- Add necessary function attribute for HIP programming model.
- Explicitly include hsa headers.


[ROCm/rccl commit: 9369f8d75d]
This commit is contained in:
Michael LIAO
2019-08-15 12:15:30 -04:00
parent c920272a9e
commit f4a240065f
10 changed files with 38 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/collectives/device/reduce_kernel.h
+8 -8
View File
@@ -19,7 +19,7 @@ struct FuncNull {
}
};
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
//we really don't need any specializations and we don't need
//to break things into uint32_t
@@ -180,7 +180,7 @@ static __device__ uint32_t addChar4(const uint32_t x, const uint32_t y) {
template<>
struct FuncSum<int8_t> {
__device__ uint32_t operator()(const uint32_t x, const uint32_t y) const {
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
#else
#if (__CUDA_ARCH__ >= 300) && (__CUDA_ARCH__ < 500)
int32_t rv, z=0;
@@ -198,7 +198,7 @@ struct FuncSum<int8_t> {
template<>
struct FuncSum<uint8_t> {
__device__ uint32_t operator()(const uint32_t x, const uint32_t y) const {
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
#else
#if (__CUDA_ARCH__ >= 300) && (__CUDA_ARCH__ < 500)
int32_t rv, z=0;
@@ -250,7 +250,7 @@ template<>
struct FuncMax<int8_t> {
union converter { uint32_t storage; char4 a; };
__device__ uint32_t operator()(const uint32_t x, const uint32_t y) const {
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
#else
#if (__CUDA_ARCH__ >= 300) && (__CUDA_ARCH__ < 500)
int32_t rv, z=0;
@@ -276,7 +276,7 @@ template<>
struct FuncMax<uint8_t> {
union converter { uint32_t storage; uchar4 a; };
__device__ uint32_t operator()(const uint32_t x, const uint32_t y) const {
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
#else
#if (__CUDA_ARCH__ >= 300) && (__CUDA_ARCH__ < 500)
int32_t rv, z=0;
@@ -303,7 +303,7 @@ template<>
struct FuncMin<int8_t> {
union converter { uint32_t storage; char4 a; };
__device__ uint32_t operator()(const uint32_t x, const uint32_t y) const {
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
#else
#if (__CUDA_ARCH__ >= 300) && (__CUDA_ARCH__ < 500)
int32_t rv, z=0;
@@ -329,7 +329,7 @@ template<>
struct FuncMin<uint8_t> {
union converter { uint32_t storage; uchar4 a; };
__device__ uint32_t operator()(const uint32_t x, const uint32_t y) const {
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
#else
#if (__CUDA_ARCH__ >= 300) && (__CUDA_ARCH__ < 500)
int32_t rv, z=0;
@@ -436,6 +436,6 @@ struct FuncMin<half> {
}
};
#endif // defined(__HIP_PLATFORM_HCC__) || defined(__HCC__)
#endif // defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
#endif // REDUCE_KERNEL_H_