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

Update collective APIs naming (#77)

Browse Source 
* Update the naming convention for collective APIs to ensure consistency across the interface.

* Move all collective API declarations to rocshmem_COLL.hpp

* The following APIs were updated as part of this change:
  - `barrier`
  - `barrier_all`
  - `sync`
  - `sync_all`
  - `all_to_all`
  - `broadcast`
  - `fcollect`
  - `all_reduce`

* Update header file generation code for collective APIs

[ROCm/rocshmem commit: 68421895d6]
This commit is contained in:
Avinash Kethineedi
2025-04-10 12:14:47 -05:00
committed by GitHub
parent 5b22ddd1ff
commit 41d5d739e2
15 changed files with 334 additions and 334 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocshmem/utils/header_files_gen/COLL.py
+4 -4
View File
@@ -43,7 +43,7 @@ types = [
def alltoall_api(T, TNAME):
return (
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_wg_alltoall(\n"
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_alltoall_wg(\n"
f" rocshmem_ctx_t ctx, rocshmem_team_t team, {T} *dest,\n"
f" const {T} *source, int nelems);\n\n"
)
@@ -75,7 +75,7 @@ def generate_alltoall_api():
def broadcast_api(T, TNAME):
return (
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_wg_broadcast(\n"
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_broadcast_wg(\n"
f" rocshmem_ctx_t ctx, rocshmem_team_t team, {T} *dest,\n"
f" const {T} *source, int nelems, int pe_root);\n"
f"__host__ void rocshmem_ctx_{TNAME}_broadcast(\n"
@@ -120,7 +120,7 @@ def generate_broadcast_api():
def fcollect_api(T, TNAME):
return (
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_wg_fcollect(\n"
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_fcollect_wg(\n"
f" rocshmem_ctx_t ctx, rocshmem_team_t team, {T} *dest,\n"
f" const {T} *source, int nelems);\n\n"
)
@@ -152,7 +152,7 @@ def generate_fcollect_api():
def reduction_api(T, TNAME, Op_API):
return (
f"__device__ ATTR_NO_INLINE int rocshmem_ctx_{TNAME}_{Op_API}_wg_reduce(\n"
f"__device__ ATTR_NO_INLINE int rocshmem_ctx_{TNAME}_{Op_API}_reduce_wg(\n"
f" rocshmem_ctx_t ctx, rocshmem_team_t team, {T} *dest, const {T} *source,\n"
f" int nreduce);\n"
f"__host__ int rocshmem_ctx_{TNAME}_{Op_API}_reduce(\n"