ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-11 17:00:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

fixed wait

Browse Source
This commit is contained in:
Jing Zhang
2019-04-03 15:30:19 -05:00
parent 0de4286a4f
commit b188c0d243
1 changed files with 8 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
src/include/inline_asm.hpp
View File

@@ -5,41 +5,36 @@ typedef float Float4 __attribute__((ext_vector_type(4)));
extern "C" __attribute__((address_space(3))) void* __to_local(void* p)[[hc]];
inline __device__ void lgkmcnt(int cnt){
#if 1
if(cnt == 0) {
asm volatile("\n \
s_waitcnt lgkmcnt(0) \n \
"::);
}
if(cnt == 1) {
else if(cnt == 1) {
asm volatile("\n \
s_waitcnt lgkmcnt(1) \n \
"::);
}
if(cnt == 2) {
else if(cnt == 2) {
asm volatile("\n \
s_waitcnt lgkmcnt(2) \n \
"::);
}
if(cnt == 3) {
else if(cnt == 3) {
asm volatile("\n \
s_waitcnt lgkmcnt(3) \n \
"::);
}
if(cnt == 4) {
else if(cnt == 4) {
asm volatile("\n \
s_waitcnt lgkmcnt(4) \n \
"::);
}
if(cnt == 5) {
asm volatile("\n \
s_waitcnt lgkmcnt(5) \n \
"::);
}
if(cnt == 6) {
asm volatile("\n \
s_waitcnt lgkmcnt(6) \n \
"::);
else {
assert(0);
}
#endif
}
inline __device__ void outerProduct1x4(const float *a, const float *b, float *c) {