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

[BUILD] Move code generation to python from CMake (#1360)

Browse Source 
* Use generate.py for func generation

* Convert AddUnroll.cmake to bash
This commit is contained in:
Bertan Dogancay authored and GitHub committed 2024-10-03 10:21:19 -04:00
1 parent 038517b169
commit 2dd10c8f17
7 files changed
+366 -768
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/include/device.h
+2 -2
View File
@@ -552,7 +552,7 @@ inline bool ncclNvlsSupported(int devRedOp, int type) {
// Map the rowIdx to funcIdx
extern int const ncclDevFuncRowToId[];
// `ncclFuncIndex()` needs to be in sync with 'ALL_COLLS' in Generate.cmake
// `ncclDevFuncId()` needs to be in sync with 'ALL_COLLS' in generate.py
inline int ncclDevFuncId(int coll, int devRedOp, int type, int algo, int proto) {
int row = 0;
do {
@@ -568,7 +568,7 @@ inline int ncclDevFuncId(int coll, int devRedOp, int type, int algo, int proto)
row += (((algo * NCCL_NUM_PROTOCOLS + proto) * ncclNumDevRedOps + devRedOp) * ncclNumTypes + type) - NCCL_NUM_FLOATS * (algo * NCCL_NUM_PROTOCOLS + proto);
break;
}
row += (NCCL_NUM_ALGORITHMS - 2) * NCCL_NUM_PROTOCOLS * (ncclNumDevRedOps * ncclNumTypes - NCCL_NUM_FLOATS);
row += (NCCL_NUM_ALGORITHMS - 4) * NCCL_NUM_PROTOCOLS * (ncclNumDevRedOps * ncclNumTypes - NCCL_NUM_FLOATS);
// RING / SIMPLE / Sum / int8_t
if (coll == ncclFuncAllToAllPivot) break;