biondizzle/DeepGEMM
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Polish get_best_configs modeling. (#158)

Browse Source
This commit is contained in:
zhonghui-J
2025-08-14 16:50:21 +08:00
committed by GitHub
parent 6d3717d541
commit 3254b758e2
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
csrc/jit_kernels/heuristics/common.hpp
View File

@@ -194,7 +194,9 @@ static GemmConfig get_best_config(const GemmType& gemm_type, const KernelType& k
// Case 2: same `block_n`, smaller `block_m` (wasted)
success |= block_n == best_block_n and block_m < best_block_m;
// Case 3: different for both `block_m` and `block_n`, larger `block_n` is better
success |= block_m != best_block_m and block_n > best_block_n;
// NOTES: don't pick `block_m/block_n` larger than shape `m/n` in this case
success |= block_m != best_block_m and block_n > best_block_n
and block_n <= n and block_m <= m;
}
}