Skip to content
This repository has been archived by the owner on Aug 30, 2024. It is now read-only.

disable MHA_AVX2 #173

Merged
merged 2 commits into from
Mar 18, 2024
Merged

disable MHA_AVX2 #173

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f2e9171
disable MHA
yuchengliu1 Mar 13, 2024
9982a81
disable mha_avx2
yuchengliu1 Mar 15, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion neural_speed/core/layers/mha_dense.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,9 @@ bool bestla_reordered_attn_fp32_support(const attn_shape_t* params) {
// TODO(Yi): check K V's layout
if (_cd->AMX_BF16()) return true;
#endif
return !_cd->AVX512F() || _cd->AVX2(); // use avx2 and f16c on avx2 platforms
// use avx2 and f16c on avx2 platforms
// todo: check avx2 mha on sever
return !_cd->AVX512F() && _cd->AVX2();
yuchengliu1 marked this conversation as resolved.
Show resolved Hide resolved
}
// kv cache sizes in bytes per layer per batch per beam for;
void bestla_reordered_attn_fp32_batch_kv_info(const kv_shape_t* params, kv_cache_info_t* out) {
Expand Down
Loading