Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Snippets][CPU] Enabled dynamic INT8,BF16,FP16 MHA tokenization #28276

Merged
merged 1 commit into from
Jan 9, 2025
Merged

[Snippets][CPU] Enabled dynamic INT8,BF16,FP16 MHA tokenization #28276

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
24 changes: 6 additions & 18 deletions src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1064,29 +1064,17 @@ void Transformations::MainSnippets(void) {
((in_type0 == element::f32 && in_type1 == ov::element::f32 &&
config.inferencePrecision == ov::element::bf16));
const auto is_int8 = in_type0 == ov::element::i8;
if (is_fp32)
return true;
// Only FP32 dynamic MHA is supported
if (matmul->is_dynamic())
return false;
if (matmul->get_transpose_a())
return false;
// [150842] The execution of Brgemm INT8/BF16/FP16 on AMX platforms depends on the value of "K % VNNIFactor".
// For more details, please teake a look at the ticket 150842
if (dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_amx)) {
const auto& b_shape = matmul->get_input_partial_shape(1);
const auto K = matmul->get_transpose_b() ? *b_shape.rbegin() : *++b_shape.rbegin();
const size_t brgemm_vnni_factor_for_real16 = 2; // 4/2(size in term of byte for bf16/fp16)
if (is_bf16 || is_fp16)
return K.is_static() && (K.get_length() % brgemm_vnni_factor_for_real16 == 0);
if (is_int8)
return K.is_static();
}
if (is_fp32)
return dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx2);
if (is_int8)
return dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_vnni) ||
return dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_amx) ||
dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_vnni) ||
IvanNovoselov marked this conversation as resolved.
Show resolved Hide resolved
dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx2_vnni);
if (is_bf16)
return dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_bf16);
return dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_amx) ||
dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_bf16);
if (is_fp16)
return dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_amx_fp16);
return true;
Expand Down
Loading