cleanup and refactoring

openvinotoolkit · Sep 14, 2024 · d4c30c6 · d4c30c6
1 parent 7dd226e
commit d4c30c6
Show file tree

Hide file tree

Showing 7 changed files with 17 additions and 42 deletions.
diff --git a/src/plugins/intel_cpu/src/nodes/executors/acl/acl_fullyconnected.hpp b/src/plugins/intel_cpu/src/nodes/executors/acl/acl_fullyconnected.hpp
@@ -1,4 +1,4 @@
-// Copyright (C) 2018-2024 Intel Corporation
+// Copyright (C) 2024 Intel Corporation
 // SPDX-License-Identifier: Apache-2.0
 //
 

diff --git a/src/plugins/intel_cpu/src/nodes/executors/acl/acl_lowp_fullyconnected.cpp b/src/plugins/intel_cpu/src/nodes/executors/acl/acl_lowp_fullyconnected.cpp
@@ -4,19 +4,16 @@
 
 #include "acl_lowp_fullyconnected.hpp"
 
+#include "acl_weights.hpp"
 #include "arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h"
-
+#include "memory_desc/cpu_memory_desc_utils.h"
+#include "nodes/common/cpu_convert.h"
 #include "nodes/executors/acl/acl_utils.hpp"
 #include "nodes/executors/executor.hpp"
 #include "nodes/executors/memory_arguments.hpp"
-#include "utils/debug_capabilities.h"
 #include "nodes/executors/debug_messages.hpp"
 #include "nodes/executors/implementation_utils.hpp"
-#include "acl_weights.hpp"
-#include "acl_utils.hpp"
-
-#include "nodes/common/cpu_convert.h"
-#include "memory_desc/cpu_memory_desc_utils.h"
+#include "utils/debug_capabilities.h"
 
 namespace ov {
 namespace intel_cpu {
@@ -41,9 +38,7 @@ static void initFCAttrs(const FCAttrs &attrs,
                         arm_compute::GEMMInfo& fullyConnectedLayerInfo,
                         const PostOps &postOps) {
     aclTensorAttrs.hasLayoutTypeNHWC = memory.at(ARG_SRC)->getDescPtr()->hasLayoutType(LayoutType::nspc);
-    //fullyConnectedLayerInfo.weights_trained_layout = getAclDataLayoutByMemoryDesc(memory.at(ARG_WEI)->getDescPtr());
     aclfcAttrs.inputPrecision = memory.at(ARG_SRC)->getDescPtr()->getPrecision();
-    //fullyConnectedLayerInfo.transpose_weights = false;
     aclfcAttrs.weightsNonTransposed = attrs.weightsNonTransposed;
 
     if (!postOps.empty()) {
@@ -86,11 +81,6 @@ void ACLLowpFullyConnectedExecutor::updateTensorsShapes(ACLShapes& aclMemoryShap
 }
 
 arm_compute::Status ACLLowpFullyConnectedExecutor::validateTensorsInfo(const ACLInfos & aclMemoryInfos) {
-    // TODO: debug only
-    //const auto src0 = aclMemoryInfos[ACLArgs::ACL_SRC_0].get();
-    //const auto src1 = aclMemoryInfos[ACLArgs::ACL_WEI].get();
-    //const auto dst = aclMemoryInfos[ACLArgs::ACL_DST].get();
-
     auto &tensor_info = aclMemoryInfos[ACLArgs::ACL_SRC_0];
     if (dequantizationScales.empty()) {
         tensor_info->set_quantization_info(arm_compute::QuantizationInfo(1.f));
@@ -104,7 +94,7 @@ arm_compute::Status ACLLowpFullyConnectedExecutor::validateTensorsInfo(const ACL
     const auto matMulValid = arm_compute::NEGEMMLowpMatrixMultiplyCore::validate(
             aclMemoryInfos[ACLArgs::ACL_SRC_0].get(),
             aclMemoryInfos[ACLArgs::ACL_WEI].get(),
-            nullptr, //aclMemoryInfos[ACLArgs::ACL_BIAS].get(),
+            aclMemoryInfos[ACLArgs::ACL_BIAS].get(),
             aclMemoryInfos[ACLArgs::ACL_DST].get(),
             gemmInfo);
     return matMulValid;
@@ -126,7 +116,6 @@ ACLFunction ACLLowpFullyConnectedExecutor::configureFunction(const ACLTensors &
     return gemm;
 }
 
-// TODO: move to ACLLowpExecutor
 std::shared_ptr<arm_compute::TensorInfo> ACLLowpFullyConnectedExecutor::initTensorInfo(
         const arm_compute::TensorShape& tensorShape,
         const arm_compute::DataType& dataType,

diff --git a/src/plugins/intel_cpu/src/nodes/executors/acl/acl_weights.cpp b/src/plugins/intel_cpu/src/nodes/executors/acl/acl_weights.cpp
@@ -4,13 +4,11 @@
 
 #include "acl_fullyconnected.hpp"
 #include "acl_utils.hpp"
+#include "nodes/common/cpu_convert.h"
 #include "nodes/executors/executor.hpp"
 #include "nodes/executors/memory_arguments.hpp"
-#include "utils/debug_capabilities.h"
-#include "nodes/executors/debug_messages.hpp"
-#include "nodes/executors/implementation_utils.hpp"
-#include "nodes/common/cpu_convert.h"
 #include "memory_desc/cpu_memory_desc_utils.h"
+#include "utils/debug_capabilities.h"
 
 namespace ov {
 namespace intel_cpu {

diff --git a/src/plugins/intel_cpu/src/nodes/executors/fullyconnected_implementations.cpp b/src/plugins/intel_cpu/src/nodes/executors/fullyconnected_implementations.cpp
@@ -90,7 +90,7 @@ static const TypeMapping aclFCTypeMapping {
 
 static const TypeMapping aclLowpFCTypeMapping {
     // {src, wei, bia, dst}                  pt<src, wei, bias, dst>
-    {{_i8, _i8, _any, _f32},                 pt(just<i8>(), just<i8>(), just<i32>(), just<f32>())}
+    {{_i8, _i8, _any, _f32},                 pt(just<i8>(), just<i8>(), just<f32>(), just<f32>())}
 };
 
 static const MappingNotation dnnlConvolutionMappingNotation {
@@ -396,7 +396,6 @@ const std::vector<ExecutorImplementation<FCAttrs>>& getImplementations() {
             },
             // acceptsShapes
             [](const MemoryArgs& memory) -> bool {
-                // @todo create syntactic sugar (functor) for shape agnostic lambda
                 return true;
             },
             // create

diff --git a/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp b/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp
@@ -725,11 +725,7 @@ void Transformations::Lpt(const std::vector<ov::element::Type>& defaultPrecision
                 }),
             PrecisionsRestriction::create<ov::opset1::MatMul>({
                     {{0}, {ov::element::u8, ov::element::i8}},
-#if defined(OPENVINO_ARCH_ARM64)
-                    {{1}, {ov::element::u8, ov::element::i8}}
-#else
                     {{1}, {ov::element::i8}}
-#endif
                 }),
             PrecisionsRestriction::create<ov::opset5::LSTMSequence>({
                     {{0, 1}, {ov::element::u8}}

diff --git a/...functional/plugin/shared/include/low_precision_transformations/mat_mul_transformation.hpp b/...functional/plugin/shared/include/low_precision_transformations/mat_mul_transformation.hpp
@@ -19,10 +19,8 @@ class MatMulTransformationTestValues {
     ov::builder::subgraph::FakeQuantizeOnData fqOnData1;
     ov::Shape inputShape2;
     ov::builder::subgraph::FakeQuantizeOnData fqOnData2;
-    // TODO: remove, not used
     std::string expectedKernelName;
     std::string expectedRuntimePrecision;
-    bool requantization;
 };
 
 typedef std::tuple<

diff --git a/...sts/functional/plugin/shared/src/low_precision_transformations/mat_mul_transformation.cpp b/...sts/functional/plugin/shared/src/low_precision_transformations/mat_mul_transformation.cpp
@@ -27,11 +27,10 @@ std::string MatMulTransformation::getTestCaseName(const testing::TestParamInfo<M
     result <<
         precision << "_" <<
         targetDevice << "_" <<
-        "IN1=" << testValues.inputShape1 << "_" <<
+        testValues.inputShape1 << "_" <<
         testValues.fqOnData1 << "_" <<
-        "IN2=" << testValues.inputShape2 << "_" <<
-        testValues.fqOnData2 << "_" <<
-        testValues.requantization;
+        testValues.inputShape2 << "_" <<
+        testValues.fqOnData2;
 
     return result.str();
 }
@@ -46,12 +45,11 @@ void MatMulTransformation::SetUp() {
     init_input_shapes({ testValues.inputShape1, testValues.inputShape2 });
 
     function = ov::builder::subgraph::MatMulFunction::getOriginal(
-        ov::element::f16, // precision,
+        precision,
         testValues.inputShape1,
         testValues.fqOnData1,
         testValues.inputShape2,
-        testValues.fqOnData2,
-        testValues.requantization);
+        testValues.fqOnData2);
 
     ov::pass::InitNodeInfo().run_on_model(function);
 }
@@ -61,13 +59,10 @@ void MatMulTransformation::run() {
 
     LayerTransformation::run();
 
-    const auto& actualType = get_runtime_precision_by_type("MatMul");
-    const auto expected = std::get<3>(GetParam());
-    EXPECT_EQ(expected.expectedRuntimePrecision, actualType);
+    const auto params = std::get<3>(GetParam());
+    const auto actualType = get_runtime_precision(params.expectedKernelName);
 
-    const auto& actualPrimitiveType = get_property_by_type("MatMul", "primitiveType");
-    const auto expectedPrimitiveType = "gemm_acl_i8";
-    EXPECT_EQ(expectedPrimitiveType, actualPrimitiveType);
+    EXPECT_EQ(actualType, params.expectedRuntimePrecision);
 }
 
 TEST_P(MatMulTransformation, CompareWithRefImpl) {