Skip to content

[CPU] Add ACL concat executor path #34043

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
aobolensk wants to merge 11 commits into
base: master
Choose a base branch
from
+499 −4
Open

[CPU] Add ACL concat executor path #34043

Show file tree
Hide file tree
Changes from 10 commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
2864388
[CPU] Add ACL concat executor path
aobolensk Feb 10, 2026
e8da409
refactor
aobolensk Feb 11, 2026
b437e81
fix tests
aobolensk Feb 11, 2026
dcabc53
tidy
aobolensk Feb 11, 2026
7a98077
Address review comments
aobolensk Feb 18, 2026
7fa7b36
fix arm concat issue
aobolensk Feb 18, 2026
1854cea
ref
aobolensk Feb 18, 2026
07fad56
tidy
aobolensk Feb 18, 2026
fb9dd83
tidy arm
aobolensk Feb 19, 2026
d02ecb5
tidy fin
aobolensk Feb 19, 2026
b05dbc0
Address review comments
aobolensk Feb 25, 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
128 changes: 125 additions & 3 deletions src/plugins/intel_cpu/src/nodes/concat.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,11 @@
#include "graph_context.h"
#include "memory_desc/cpu_memory_desc.h"
#include "node.h"
#include "nodes/executors/executor.hpp"
#include "nodes/executors/executor_config.hpp"
#include "nodes/executors/executor_factory.hpp"
#include "nodes/executors/implementations.hpp"
#include "nodes/executors/memory_arguments.hpp"
#include "nodes/node_config.h"
#include "onednn/dnnl.h"
#include "openvino/core/except.hpp"
Expand Down Expand Up @@ -92,6 +97,7 @@ Concat::Concat(const std::shared_ptr<ov::Node>& op, const GraphContext::CPtr& co
}
CPU_NODE_ASSERT(axis < static_cast<int64_t>(inRank) && axis >= 0, "has invalid value of axis parameter: ", axis);
this->axis = axis;
m_attrs.axis = axis;
}

void Concat::getSupportedDescriptors() {
Expand Down Expand Up @@ -242,12 +248,49 @@ void Concat::initSupportedPrimitiveDescriptors() {
// Optimized inplace case
for (auto refPdIndex : pdIndexesToReuse) {
auto config = supportedPrimitiveDescriptors[refPdIndex].getConfig();
;
for (auto& inConf : config.inConfs) {
inConf.inPlace(0);
}
supportedPrimitiveDescriptors.emplace_back(config, impl_desc_type::unknown);
}

const auto& concatImplementations = getImplementations<ConcatAttrs>();
Copy link
Contributor

@EgorDuplensky EgorDuplensky Feb 24, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could you please clarify why we access implementation list here?
In general, we just need to create and use a factory, not directly the implementation list.
Also, after refactoring, 'getSupportedDescriptors()' becomes empty, and most of the logic is moved to 'initSupportedPrimitiveDescriptors' and 'createPrimitive'

Copy link
Contributor Author

@aobolensk aobolensk Feb 25, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done

if (!concatImplementations.empty()) {
const auto& creatorsMap = BlockedDescCreator::getCommonCreators();
auto pushExecutorDesc = [&](LayoutType layoutType) {
NodeConfig nodeConfig;
nodeConfig.outConfs.resize(1);
nodeConfig.inConfs.resize(getParentEdges().size());

MemoryDescArgs descs;
descs.reserve(getParentEdges().size() + 1);
for (size_t i = 0; i < getParentEdges().size(); ++i) {
auto srcDesc = creatorsMap.at(layoutType)->createSharedDesc(inputPrecision, getInputShapeAtPort(i));
nodeConfig.inConfs[i].setMemDesc(srcDesc);
nodeConfig.inConfs[i].inPlace(-1);
nodeConfig.inConfs[i].constant(false);
descs[ARG_SRC + i] = srcDesc;
}

auto dstDesc = creatorsMap.at(layoutType)->createSharedDesc(outputPrecision, getOutputShapeAtPort(0));
nodeConfig.outConfs[0].setMemDesc(dstDesc);
nodeConfig.outConfs[0].inPlace(-1);
nodeConfig.outConfs[0].constant(false);
descs[ARG_DST] = dstDesc;

const executor::Config<ConcatAttrs> config{descs, m_attrs};
const bool supported =
std::any_of(concatImplementations.begin(), concatImplementations.end(), [&](const auto& impl) {
return impl.supports(config, memoryFormatFilter);
});
if (supported) {
supportedPrimitiveDescriptors.emplace_back(nodeConfig, impl_desc_type::undef);
}
};

pushExecutorDesc(LayoutType::ncsp);
pushExecutorDesc(LayoutType::nspc);
}
}

void Concat::selectOptimalPrimitiveDescriptor() {
Expand Down Expand Up @@ -354,14 +397,27 @@ void Concat::selectOptimalPrimitiveDescriptor() {
return;
}

// if there are more than one PD with similar data layouts - select the optimized one
for (auto indx : canSelectPrimitive) {
if (supportedPrimitiveDescriptors[indx].getImplementationType() == impl_desc_type::undef) {
selectPrimitiveDescriptorByIndex(static_cast<int>(indx));
return;
}
}

for (auto indx : canSelectPrimitive) {
if (supportedPrimitiveDescriptors[indx].getImplementationType() == impl_desc_type::unknown) {
selectPrimitiveDescriptorByIndex(static_cast<int>(indx));
return;
}
}

for (auto indx : canSelectPrimitive) {
if (supportedPrimitiveDescriptors[indx].getImplementationType() == impl_desc_type::ref) {
selectPrimitiveDescriptorByIndex(static_cast<int>(indx));
return;
}
}

// if there are no matching data layouts, select first optimized implementation
for (size_t i = 0; i < supportedPrimitiveDescriptors.size(); i++) {
if (canBeInPlace && supportedPrimitiveDescriptors[i].getImplementationType() == impl_desc_type::unknown) {
Expand Down Expand Up @@ -389,10 +445,29 @@ void Concat::prepareParams() {
return;
}

auto* selectedPd = getSelectedPrimitiveDescriptor();
CPU_NODE_ASSERT(selectedPd, "Preferable primitive descriptor is not set.");

if (useExecutor && m_executor) {
for (size_t i = 0; i < getParentEdges().size(); ++i) {
m_memory[ARG_SRC + i] = getSrcMemoryAtPort(i);
}
m_memory[ARG_DST] = getDstMemoryAtPort(0);

if (m_executor->update(m_memory)) {
selectedPd->setImplementationType(m_executor->implType());
return;
}

// Fallback to oneDNN/ref concat when executor update is not applicable for runtime shapes.
useExecutor = false;
m_executor.reset();
selectedPd->setImplementationType(impl_desc_type::ref);
}

const auto& dstMemPtr = getDstMemoryAtPort(0);
CPU_NODE_ASSERT(dstMemPtr && dstMemPtr->isDefined(), "Destination memory is undefined.");
auto dstMemDesc = dstMemPtr->getDescWithType<BlockedMemoryDesc>();
CPU_NODE_ASSERT(getSelectedPrimitiveDescriptor(), "Preferable primitive descriptor is not set.");

const auto& outputStrides = dstMemDesc->getStrides();
size_t curConcatOffset = 0;
Expand Down Expand Up @@ -502,6 +577,46 @@ size_t Concat::inverseOrder(const VectorDims& order, size_t axis) {
return -1;
}

void Concat::createPrimitive() {
auto* selectedPd = getSelectedPrimitiveDescriptor();
CPU_NODE_ASSERT(selectedPd, "Preferable primitive descriptor is not set.");

if (!isInPlace()) {
m_memory.clear();
m_memory.reserve(getParentEdges().size() + 1);
for (size_t i = 0; i < getParentEdges().size(); ++i) {
m_memory[ARG_SRC + i] = getSrcMemoryAtPort(i);
}
m_memory[ARG_DST] = getDstMemoryAtPort(0);

useExecutor = selectedPd->getImplementationType() == impl_desc_type::undef && !canOptimizeNspc;
m_executor.reset();

if (useExecutor) {
MemoryDescArgs descs;
descs.reserve(m_memory.size());
for (const auto& [arg, mem] : m_memory) {
descs[arg] = mem->getDescPtr();
}

try {
auto executionContext = std::make_shared<ExecutorContext>(context, getImplPriority());
auto factory = std::make_shared<ExecutorFactory<ConcatAttrs>>(m_attrs,
executionContext,
descs,
memoryFormatFilter);
m_executor = factory->make(m_memory);
selectedPd->setImplementationType(m_executor->implType());
} catch (...) {
useExecutor = false;
m_executor.reset();
}
}
}

Node::createPrimitive();
}

void Concat::initOptimalPrimitiveDescriptor() {
auto* selected_pd = getSelectedPrimitiveDescriptor();
CPU_NODE_ASSERT(selected_pd, "Preferable primitive descriptor is not set.");
Expand All @@ -525,6 +640,8 @@ void Concat::initOptimalPrimitiveDescriptor() {
}
}

useExecutor = selected_pd->getImplementationType() == impl_desc_type::undef;

// block layout may have axis greater than rank, disable ref_concat
auto* primDesc = getSelectedPrimitiveDescriptor();
auto* memDesc = primDesc->getConfig().outConfs[0].getMemDesc()->as<BlockedMemoryDesc>();
Expand Down Expand Up @@ -558,6 +675,11 @@ void Concat::execute(const dnnl::stream& strm) {
return;
}

if (useExecutor && m_executor) {
m_executor->execute(m_memory);
return;
}

if (canOptimize1DCase) {
exec1DCase();
return;
Expand Down
8 changes: 8 additions & 0 deletions src/plugins/intel_cpu/src/nodes/concat.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,9 @@
#include "edge.h"
#include "graph_context.h"
#include "node.h"
#include "nodes/executors/concat.hpp"
#include "nodes/executors/executor.hpp"
#include "nodes/executors/memory_arguments.hpp"
#include "openvino/core/node.hpp"
#include "openvino/core/type/element_type.hpp"

Expand Down Expand Up @@ -41,6 +44,7 @@ class Concat : public Node {
[[nodiscard]] bool isExecutable() const override;
[[nodiscard]] bool needPrepareParams() const override;
void prepareParams() override;
void createPrimitive() override;
// TODO: Move to base Node class when more nodes support fuse convert
bool supportConvertFusion() const {
return supportFuseConvert;
Expand Down Expand Up @@ -70,6 +74,10 @@ class Concat : public Node {
bool doFuseConvert = false; // whether to perform FP16 to FP32 conversion
static constexpr size_t MAX_RANK_REF = 6;
dnnl::primitive prim;
bool useExecutor = false;
ConcatAttrs m_attrs;
MemoryArgs m_memory;
ExecutorPtr m_executor = nullptr;
};

} // namespace ov::intel_cpu::node
Loading
Loading