Skip to content

Refactored Kleidiai executor and Support for INT4 groupwise asym Quantization #37086

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

Open
hrushitfujitsu wants to merge 3 commits into
base: master
Choose a base branch
from
Open

Refactored Kleidiai executor and Support for INT4 groupwise asym Quantization #37086

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c57ac94
rebased
hrushitfujitsu Jul 28, 2026
ed40078
resolved comments
hrushitfujitsu Aug 4, 2026
8b2759f
fixed clang format issues and copilot comment
hrushitfujitsu Aug 6, 2026
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: 22 additions & 2 deletions src/plugins/intel_cpu/src/nodes/executors/fullyconnected_implementations.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,6 +106,20 @@ static const TypeMapping aclLowpFCTypeMapping {
{{_u8 | _i8, _i8, _any, _f32}, {bypass(), bypass(), use<3>(), bypass()}}
};

#if defined(OV_CPU_WITH_KLEIDIAI)
static const TypeMapping kleidiaiFCTypeMapping {
// {src, wei, bia, dst} pt<src, wei, bias, dst>
// f32 weights are used as-is by the F32 NEON MLA kernel
{{_f32, _f32, _any, _any}, {bypass(), bypass(), use<0>(), use<0>()}},
// Keep low-precision (compressed) weights intact so the dynamic-quant
// KleidiAI kernels (i8 / i4 / u4, incl. group asymmetric) receive the
// original weight precision instead of an upconverted f32 tensor.
{{_f32, _i8 | _i4 | _u4, _any, _any}, {bypass(), bypass(), use<0>(), use<0>()}},
// fallback
{{_any, _any, _any, _any}, {just<f32>(), just<f32>(), just<f32>(), just<f32>()}},
};
#endif

static const MappingNotation fcMappingNotation {
{ARG_SRC, 0},
{ARG_WEI, 1},
Expand Down Expand Up @@ -371,14 +385,20 @@ const std::vector<ExecutorImplementation<FCAttrs>>& getImplementations() {
VERIFY(noPostOps(config), UNSUPPORTED_POST_OPS);
VERIFY(noSparseDecompression(config), UNSUPPORTED_SPARSE_WEIGHTS);
VERIFY(all_of(f32, srcType(config), dstType(config)), UNSUPPORTED_SRC_PRECISIONS);
VERIFY(any_of(weiType(config), f32, i8, i4), UNSUPPORTED_WEI_PRECISIONS);
VERIFY(any_of(weiType(config), f32, i8, i4, u4), UNSUPPORTED_WEI_PRECISIONS);
VERIFY(implication(hasBias(config), biaType(config) == f32), UNSUPPORTED_SRC_PRECISIONS);
VERIFY(weiRank(config) == 2U, UNSUPPORTED_WEI_RANK);
VERIFY(MatMulKleidiAIExecutor::supports(config), UNSUPPORTED_BY_EXECUTOR);

return true;
},
HasNoOptimalConfig<FCAttrs>{},
// createOptimalConfig
[](const FCConfig& config) -> std::optional<executor::Config<FCAttrs>> {
return createOptimalConfigCommon(config,
kleidiaiFCTypeMapping,
dnnlFCLayoutConfig,
fcMappingNotation);
},
AcceptsAnyShape<FCAttrs>,
CreateDefault<MatMulKleidiAIExecutor, FCAttrs>{}
)
Expand Down
Loading
Loading