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

add reduce/broadcast algo/proto selection table for multi-node gfx940 (#1889)

Browse Source
This commit is contained in:
isaki001
2025-09-10 14:25:23 -05:00
committed by GitHub
parent c2bccf9156
commit 9c36439354
4 changed files with 9 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graph/tuning.cc
+4
View File
@@ -349,6 +349,10 @@ static struct tuningModel tuning_model_5 {
{/*LL (min/max/factor/thread_threshold)*/ {0, 98304, 1, 16}, /*LL64/128 (min/max/factor/thread_threshold)*/ {98304, 5046272, 1, 64}},
/*AllReduce*/
{/*LL (min/max/factor/thread_threshold)*/ {0, 1048576, 1, 0},/*LL64/128 (min/max/factor/thread_threshold)*/ {1048576, 9437184, 3145728, 0}},
/*Reduce*/
{/*LL (min/max/factor/thread_threshold)*/ {0, 4096, 1, 0},/*LL64/128 (min/max/factor/thread_threshold)*/ {4096, 16777216, 1, 0}},
/*Broadcast*/
{/*LL (min/max/factor/thread_threshold)*/ {0, 8192, 1, 0},/*LL64/128 (min/max/factor/thread_threshold)*/ {8192, 33554432, 1, 0}},
},
};