Skip to content

[FORK][FEATURE][x64]Extend InnerProduct source dynamic quantization to BF16 #310

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

Merged
maxnick merged 1 commit into from
May 28, 2026
Merged

[FORK][FEATURE][x64]Extend InnerProduct source dynamic quantization to BF16 #310

Show file tree
Hide file tree
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
11 changes: 11 additions & 0 deletions src/cpu/cpu_inner_product_list.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -134,13 +134,19 @@ const std::map<pk_dt_impl_key_t, std::vector<impl_list_item_t>> &impl_list_map()
CPU_INSTANCE(ref_inner_product_fwd_t)
nullptr,
}},
// VNNI instance handles the BF16 source dynamic-quantization path:
// BF16 src is quantized to s8 and dispatched to the VNNI s8 GEMM.
// Plain BF16 (no src dyn-quant) is rejected by init_conf and falls
// through to avx512_core_bf16.
{{forward, bf16, u8, f32}, {
CPU_INSTANCE_AMX(brgemm_inner_product_fwd_t, avx512_core_amx)
CPU_INSTANCE_AVX512(brgemm_inner_product_fwd_t, avx512_core_vnni)
CPU_INSTANCE_AVX512(brgemm_inner_product_fwd_t, avx512_core_bf16)
nullptr,
}},
{{forward, bf16, u8, bf16}, {
CPU_INSTANCE_AMX(brgemm_inner_product_fwd_t, avx512_core_amx)
CPU_INSTANCE_AVX512(brgemm_inner_product_fwd_t, avx512_core_vnni)
CPU_INSTANCE_AVX512(brgemm_inner_product_fwd_t, avx512_core_bf16)
nullptr,
}},
Expand Down Expand Up @@ -184,13 +190,18 @@ const std::map<pk_dt_impl_key_t, std::vector<impl_list_item_t>> &impl_list_map()
CPU_INSTANCE_AVX512(brgemm_inner_product_fwd_t, avx512_core_bf16)
nullptr,
}},
// VNNI instance here serves the BF16 source dynamic-quantization path
// (BF16 src -> s8 -> VNNI GEMM). See the {bf16, u8, *} entries above
// for the dispatch details.
{{forward, bf16, u4, f32}, {
CPU_INSTANCE_AMX(brgemm_inner_product_fwd_t, avx512_core_amx)
CPU_INSTANCE_AVX512(brgemm_inner_product_fwd_t, avx512_core_vnni)
CPU_INSTANCE_AVX512(brgemm_inner_product_fwd_t, avx512_core_bf16)
nullptr,
}},
{{forward, bf16, u4, bf16}, {
CPU_INSTANCE_AMX(brgemm_inner_product_fwd_t, avx512_core_amx)
CPU_INSTANCE_AVX512(brgemm_inner_product_fwd_t, avx512_core_vnni)
CPU_INSTANCE_AVX512(brgemm_inner_product_fwd_t, avx512_core_bf16)
nullptr,
}},
Expand Down
Loading
Loading