Skip to content

[CPU] Fix dnnl::mamtul executor for FullyConnected node #32356

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
wants to merge 1 commit into
base: master
Choose a base branch
from
+27 −22
Open

[CPU] Fix dnnl::mamtul executor for FullyConnected node #32356

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
27 changes: 15 additions & 12 deletions src/plugins/intel_cpu/src/nodes/executors/dnnl/dnnl_matmul_primitive.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,7 @@
#include "memory_desc/dnnl_blocked_memory_desc.h"
#include "memory_desc/dnnl_memory_desc.h"
#include "nodes/executors/dnnl/dnnl_aliases.hpp"
#include "nodes/executors/dnnl/dnnl_fullyconnected_primitive.hpp"
#include "nodes/executors/dnnl/dnnl_shape_agnostic_data.hpp"
#include "nodes/executors/executor.hpp"
#include "nodes/executors/fullyconnected_config.hpp"
Expand Down Expand Up @@ -306,7 +307,8 @@ static primitive_desc createPrimitiveDesc(const dnnl::memory::desc& inputDesc,
[[maybe_unused]] const bool useSparseWeights,
const bool useWeightsDecompression,
const bool fcSemantic) {
if (defaultImplType == impl_desc_type::undef) {
// priority-based implementation selection if implementation type is not specified
if (defaultImplType == impl_desc_type::undef && !fcSemantic) {
struct PrimitiveDescWithPriority {
dnnl::primitive_desc prim_desc;
size_t priority = 0UL;
Expand Down Expand Up @@ -484,28 +486,29 @@ bool DnnlMatMulPrimitive::useWeightsDecompressionImpl(const ov::element::Type in
return (any_of(inputType, f32, bf16, f16) && any_of(weightsType, u8, i8, u4, i4));
}

DnnlShapeAgnosticDataPtr DnnlMatMulPrimitive::createShapeAgnosticData(const FCAttrs& fcAttrs,
const MemoryArgs& memory,
const ExecutorContext::CPtr& context,
const bool cacheWeights) {
MatMulAttrs attrs;
attrs.postOps = fcAttrs.postOps;
attrs.weightsNonTransposed = fcAttrs.weightsNonTransposed;

return createShapeAgnosticData(attrs, memory, context, cacheWeights);
}

DnnlShapeAgnosticDataPtr DnnlMatMulPrimitive::createShapeAgnosticData(const MatMulAttrs& attrs,
const MemoryArgs& memory,
const ExecutorContext::CPtr& context,
const bool cacheWeights) {
if (attrs.fcSemantic) {
FCAttrs fcAttrs{attrs.withBias,
attrs.weightsNonTransposed,
false,
0,
true,
ov::intel_cpu::Config::ModelType::Unknown,
attrs.postOps};
return DnnlFCPrimitive::createShapeAgnosticData(fcAttrs, memory, context, cacheWeights);
}

DEBUG_LOG("Creating shape agnostic data");
auto srcDesc = memory.at(ARG_SRC)->getDescPtr();
auto weiDesc = memory.at(ARG_WEI)->getDescPtr();
auto dstDesc = memory.at(ARG_DST)->getDescPtr();
const auto& biasDesc = memory.at(ARG_BIAS)->getDescPtr();

const auto useWeightsDecompression = useWeightsDecompressionImpl(srcDesc->getPrecision(), weiDesc->getPrecision());

const auto postOpData =
createPrimitiveAttrs(attrs, memory, context, useWeightsDecompression, attrs.weightsNonTransposed);

Expand Down
5 changes: 0 additions & 5 deletions src/plugins/intel_cpu/src/nodes/executors/dnnl/dnnl_matmul_primitive.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,11 +73,6 @@ class DnnlMatMulPrimitive {
const ExecutorContext::CPtr& context,
bool cacheWeights);

static DnnlShapeAgnosticDataPtr createShapeAgnosticData(const FCAttrs& fcAttrs,
const MemoryArgs& memory,
const ExecutorContext::CPtr& context,
bool cacheWeights);

static DnnlMemoryDescPtr makeTransposedWeightDescriptor(const DnnlMemoryDescPtr& srcDesc,
const DnnlMemoryDescPtr& dstDesc,
const MatMulAttrs& attrs);
Expand Down
17 changes: 12 additions & 5 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 @@ -422,11 +422,18 @@ const std::vector<ExecutorImplementation<FCAttrs>>& getImplementations() {
[](const FCAttrs& attrs,
const MemoryArgs& memory,
const ExecutorContext::CPtr& context) -> ExecutorPtr {
MatMulAttrs matMulAttrs{false,
false};
matMulAttrs.postOps = attrs.postOps;
matMulAttrs.transposeB = attrs.weightsNonTransposed;
matMulAttrs.constantWeights = true;
MatMulAttrs matMulAttrs {
false,
false,
attrs.withBias,
attrs.weightsNonTransposed,
attrs.sparseWeights,
true,
true,
attrs.dynamicQuantizationGroupSize,
{},
attrs.postOps
};

return std::make_shared<
DnnlExecutor<DnnlMatMulPrimitive, MatMulAttrs, DnnlShapeAgnosticData,
Expand Down
Loading