Skip to content

[CPU] FullyConnected acceleration for BF16 compressed weights #24633

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 5 commits into from
Jul 1, 2024
Merged

[CPU] FullyConnected acceleration for BF16 compressed weights #24633

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
ea73407
[CPU] FullyConnected acceleration for BF16 compressed weights
May 22, 2024
aea404f
Merge remote-tracking branch 'origin/master' into feature/bf16_weight…
maxnick Jun 27, 2024
76b50e0
Fix oneDNN bf16 load-store
maxnick Jun 28, 2024
890ec7e
More safe check for bf16 conversion in brgemm
maxnick Jul 1, 2024
1f39b19
Merge remote-tracking branch 'origin/master' into feature/bf16_weight…
maxnick Jul 1, 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
2 changes: 1 addition & 1 deletion src/plugins/intel_cpu/src/nodes/executors/dnnl/dnnl_fullyconnected_primitive.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,7 +122,7 @@ bool DnnlFCPrimitive::useWeightsDecompressionImpl(const ov::element::Type inputT
// f16c kernel saves memory footprint with additional decompression computational overhead
// which is only meaningful on LLM with small batch-size.
// TODO: fall-back to use f32 weights on large batch-size
if (inputType == f32 && weightsType == f16)
if (inputType == f32 && one_of(weightsType, f16, bf16))
return true;
}
}
Expand Down
2 changes: 1 addition & 1 deletion src/plugins/intel_cpu/thirdparty/onednn
Open in desktop
Submodule onednn updated 7 files
+5 −0 src/cpu/cpu_inner_product_list.cpp
+2 −3 src/cpu/x64/brgemm/brgemm_utils.cpp
+8 −4 src/cpu/x64/brgemm/jit_brgemm_kernel.cpp
+4 −4 src/cpu/x64/jit_brgemm_inner_product.hpp
+4 −4 src/cpu/x64/jit_brgemm_inner_product_utils.cpp
+5 −0 src/cpu/x64/jit_brgemm_weights_decompression_kernel.cpp
+1 −1 src/cpu/x64/jit_uni_reorder.cpp