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

modified team based to_all -> reduce

Browse Source
This commit is contained in:
Yiltan Hassan Temucin
2024-10-29 06:32:13 -07:00
committed by Yiltan Temucin
parent 75ece02048
commit 997eb69b5a
14 changed files with 52 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/context.hpp
+2 -4
View File
@@ -192,8 +192,7 @@ class Context {
long* pSync); // NOLINT(runtime/int)
template <typename T, ROC_SHMEM_OP Op>
__device__ void to_all(roc_shmem_team_t team, T* dest, const T* source,
int nreduce);
__device__ int reduce(roc_shmem_team_t team, T* dest, const T* source, int nreduce);
template <typename T>
__device__ void put(T* dest, const T* source, size_t nelems, int pe);
@@ -361,8 +360,7 @@ class Context {
long* pSync); // NOLINT(runtime/int)
template <typename T, ROC_SHMEM_OP Op>
__host__ void to_all(roc_shmem_team_t team, T* dest, const T* source,
int nreduce);
__host__ int reduce(roc_shmem_team_t team, T* dest, const T* source, int nreduce);
template <typename T>
__host__ void wait_until(T *ivars, int cmp, T val);