[CPU Plugin 2.0] replace ie::precision with ov::element (openvinotool…

…kit#21041) * [CPU Plugin 2.0] replace ie::precision with ov::element Signed-off-by: Zhai, Xuejun <xuejun.zhai@intel.com> * [CPU Plugin 2.0] fix MatmulWeightsDecompression.CompareWithRefs test error Signed-off-by: Zhai, Xuejun <xuejun.zhai@intel.com> * [CPU Plugin 2.0] fix compiler error Signed-off-by: Zhai, Xuejun <xuejun.zhai@intel.com> * [CPU Plugin 2.0] fix compiler error Signed-off-by: Zhai, Xuejun <xuejun.zhai@intel.com> * [CPU Plugin 2.0] fix review comment Signed-off-by: Zhai, Xuejun <xuejun.zhai@intel.com> --------- Signed-off-by: Zhai, Xuejun <xuejun.zhai@intel.com>
syaifulnizamiphone7 · Nov 15, 2023 · 66a8600 · 66a8600
1 parent 0180078
commit 66a8600
Show file tree

Hide file tree

Showing 238 changed files with 2,383 additions and 2,280 deletions.
diff --git a/src/plugins/intel_cpu/src/cpu_memory.cpp b/src/plugins/intel_cpu/src/cpu_memory.cpp
@@ -49,7 +49,7 @@ namespace {
         if (!ftz) {
             return;
         }
-        if (src.getDesc().getPrecision() != Precision::FP32 || dst.getDesc().getPrecision() == Precision::BF16) {
+        if (src.getDesc().getPrecision() != ov::element::f32 || dst.getDesc().getPrecision() == ov::element::bf16) {
             return;
         }
         size_t offset = 0;

diff --git a/src/plugins/intel_cpu/src/cpu_memory.h b/src/plugins/intel_cpu/src/cpu_memory.h
@@ -17,7 +17,6 @@
 #include <functional>
 #include <memory>
 #include <vector>
-#include <ie_precision.hpp>
 
 /**
  * @file contains a concept classes to work with memory/tensor/blob abstractions on plugin level.
@@ -187,7 +186,7 @@ class IMemory {
     //oneDNN specifics for backward compatibility
     virtual dnnl::memory getPrimitive() const = 0;
     dnnl::memory::data_type getDataType() const {
-        return DnnlExtensionUtils::IEPrecisionToDataType(getDesc().getPrecision());
+        return DnnlExtensionUtils::ElementTypeToDataType(getDesc().getPrecision());
     }
 
     virtual void nullify() = 0;

diff --git a/src/plugins/intel_cpu/src/cpu_tensor.cpp b/src/plugins/intel_cpu/src/cpu_tensor.cpp
@@ -17,7 +17,7 @@ Tensor::Tensor(MemoryPtr memptr) : m_memptr{memptr} {
     auto memdesc = m_memptr->getDescPtr();
     OPENVINO_ASSERT(memdesc->hasLayoutType(LayoutType::ncsp), "intel_cpu::Tensor only supports memory with ncsp layout.");
 
-    m_element_type = InferenceEngine::details::convertPrecision(memdesc->getPrecision());
+    m_element_type = memdesc->getPrecision();
 }
 
 void Tensor::set_shape(ov::Shape new_shape) {

diff --git a/src/plugins/intel_cpu/src/dnnl_extension_utils.cpp b/src/plugins/intel_cpu/src/dnnl_extension_utils.cpp
@@ -42,63 +42,63 @@ uint8_t DnnlExtensionUtils::sizeOfDataType(dnnl::memory::data_type dataType) {
     }
 }
 
-memory::data_type DnnlExtensionUtils::IEPrecisionToDataType(const InferenceEngine::Precision& prec) {
-    switch (prec) {
-        case InferenceEngine::Precision::FP32:
+dnnl::memory::data_type DnnlExtensionUtils::ElementTypeToDataType(const ov::element::Type& elementType) {
+    switch (elementType) {
+        case ov::element::f32:
             return memory::data_type::f32;
-        case InferenceEngine::Precision::I32:
+        case ov::element::i32:
             return memory::data_type::s32;
-        case InferenceEngine::Precision::BF16:
+        case ov::element::bf16:
             return memory::data_type::bf16;
-        case InferenceEngine::Precision::I8:
+        case ov::element::i8:
             return memory::data_type::s8;
-        case InferenceEngine::Precision::U8:
-        case InferenceEngine::Precision::BOOL:
+        case ov::element::u8:
+        case ov::element::boolean:
             return memory::data_type::u8;
-        case InferenceEngine::Precision::BIN:
+        case ov::element::u1:
             return memory::data_type::bin;
-        case InferenceEngine::Precision::FP16:
+        case ov::element::f16:
             return memory::data_type::f16;
-        case InferenceEngine::Precision::NF4:
+        case ov::element::nf4:
             return memory::data_type::nf4;
-        case InferenceEngine::Precision::I4:
+        case ov::element::i4:
             return memory::data_type::s4;
-        case InferenceEngine::Precision::U4:
+        case ov::element::u4:
             return memory::data_type::u4;
-        case InferenceEngine::Precision::UNSPECIFIED:
+        case ov::element::undefined:
             return memory::data_type::undef;
         default: {
-            OPENVINO_THROW("The plugin does not support ", prec.name());
+            OPENVINO_THROW("The plugin does not support ", elementType.to_string());
         }
     }
 }
 
-InferenceEngine::Precision DnnlExtensionUtils::DataTypeToIEPrecision(memory::data_type dataType) {
+ov::element::Type DnnlExtensionUtils::DataTypeToElementType(const dnnl::memory::data_type& dataType) {
     switch (dataType) {
         case memory::data_type::f32:
-            return InferenceEngine::Precision::FP32;
+            return ov::element::f32;
         case memory::data_type::s32:
-            return InferenceEngine::Precision::I32;
+            return ov::element::i32;
         case memory::data_type::bf16:
-            return InferenceEngine::Precision::BF16;
+            return ov::element::bf16;
         case memory::data_type::s8:
-            return InferenceEngine::Precision::I8;
+            return ov::element::i8;
         case memory::data_type::u8:
-            return InferenceEngine::Precision::U8;
+            return ov::element::u8;
         case memory::data_type::bin:
-            return InferenceEngine::Precision::BIN;
+            return ov::element::u1;
         case memory::data_type::f16:
-            return InferenceEngine::Precision::FP16;
+            return ov::element::f16;
         case memory::data_type::f64:
-            return InferenceEngine::Precision::FP64;
+            return ov::element::f64;
         case memory::data_type::nf4:
-            return InferenceEngine::Precision::NF4;
+            return ov::element::nf4;
         case memory::data_type::s4:
-            return InferenceEngine::Precision::I4;
+            return ov::element::i4;
         case memory::data_type::u4:
-            return InferenceEngine::Precision::U4;
+            return ov::element::u4;
         case memory::data_type::undef:
-            return InferenceEngine::Precision::UNSPECIFIED;
+            return ov::element::undefined;
         default: {
             OPENVINO_THROW("Unsupported data type.");
         }

diff --git a/src/plugins/intel_cpu/src/dnnl_extension_utils.h b/src/plugins/intel_cpu/src/dnnl_extension_utils.h
@@ -23,8 +23,8 @@ class DnnlMemoryDesc;
 class DnnlExtensionUtils {
 public:
     static uint8_t sizeOfDataType(dnnl::memory::data_type dataType);
-    static dnnl::memory::data_type IEPrecisionToDataType(const InferenceEngine::Precision& prec);
-    static InferenceEngine::Precision DataTypeToIEPrecision(dnnl::memory::data_type dataType);
+    static dnnl::memory::data_type ElementTypeToDataType(const ov::element::Type& elementType);
+    static ov::element::Type DataTypeToElementType(const dnnl::memory::data_type& dataType);
     static Dim convertToDim(const dnnl::memory::dim &dim);
     static dnnl::memory::dim convertToDnnlDim(const Dim &dim);
     static VectorDims convertToVectorDims(const dnnl::memory::dims& dims);

diff --git a/src/plugins/intel_cpu/src/dnnl_postops_composer.cpp b/src/plugins/intel_cpu/src/dnnl_postops_composer.cpp
@@ -58,7 +58,7 @@ void DnnlPostOpsComposer::updateWeiScales() {
     DEBUG_LOG("Set weight scales mask ", "DNNL_ARG: ", DNNL_ARG_WEIGHTS, " mask: ", wei_scale_mask);
     attr.set_scales_mask(DNNL_ARG_WEIGHTS, wei_scale_mask);
 
-    DnnlBlockedMemoryDesc memoryDesc(InferenceEngine::Precision::FP32, Shape({wei_scale_values.size()}));
+    DnnlBlockedMemoryDesc memoryDesc(ov::element::f32, Shape({wei_scale_values.size()}));
     auto mem = std::make_shared<Memory>(engine, memoryDesc);
     memcpy(mem->getData(), wei_scale_values.data(), wei_scale_values.size() * sizeof(float));
     args[DNNL_ARG_ATTR_SCALES | DNNL_ARG_WEIGHTS] = mem;
@@ -71,7 +71,7 @@ void DnnlPostOpsComposer::updateDestScales() {
     DEBUG_LOG("Set dest scale mask ", "DNNL_ARG: ", DNNL_ARG_DST, " mask: ", 0);
     attr.set_scales_mask(DNNL_ARG_DST, 0);
 
-    DnnlBlockedMemoryDesc memoryDesc(InferenceEngine::Precision::FP32, Shape({1}));
+    DnnlBlockedMemoryDesc memoryDesc(ov::element::f32, Shape({1}));
     auto mem = std::make_shared<Memory>(engine, memoryDesc);
     memcpy(mem->getData(), &dst_scale_val, sizeof(float));
     args[DNNL_ARG_ATTR_SCALES | DNNL_ARG_DST] = mem;
@@ -86,7 +86,7 @@ void DnnlPostOpsComposer::appendBinary(const dnnl::algorithm alg, const std::vec
 
     DEBUG_LOG("Append binary post op with algorithm: ", convert_to_c(alg));
 
-    DnnlBlockedMemoryDesc memoryDesc(InferenceEngine::Precision::FP32, Shape(*pdims));
+    DnnlBlockedMemoryDesc memoryDesc(ov::element::f32, Shape(*pdims));
     ops.append_binary(alg, memoryDesc.getDnnlDesc());
 
     // copy the data as args
@@ -259,7 +259,7 @@ MemoryPtr DnnlPostOpsComposer::prepackDecompressionParams(const MemoryCPtr& para
 
     if (needTranspose) {
         VectorDims dnnlShape = {shape[0], shape[1]};
-        DnnlBlockedMemoryDesc memoryDesc(InferenceEngine::Precision::FP32, Shape(dnnlShape));
+        DnnlBlockedMemoryDesc memoryDesc(ov::element::f32, Shape(dnnlShape));
         mem = std::make_shared<Memory>(engine, memoryDesc);
         auto memory_buf = static_cast<float*>(mem->getData());
 
@@ -271,7 +271,7 @@ MemoryPtr DnnlPostOpsComposer::prepackDecompressionParams(const MemoryCPtr& para
         }
     } else {
         VectorDims dnnlShape = {shape[shape.size() - 1], shape[0]};
-        DnnlBlockedMemoryDesc memoryDesc(InferenceEngine::Precision::FP32, Shape(dnnlShape));
+        DnnlBlockedMemoryDesc memoryDesc(ov::element::f32, Shape(dnnlShape));
         mem = std::make_shared<Memory>(engine, memoryDesc);
         auto memory_buf = static_cast<float*>(mem->getData());
         const size_t elements_count = std::accumulate(shape.begin(), shape.end(), 1, std::multiplies<size_t>());

diff --git a/src/plugins/intel_cpu/src/emitters/x64/jit_bf16_emitters.hpp b/src/plugins/intel_cpu/src/emitters/x64/jit_bf16_emitters.hpp
@@ -12,7 +12,7 @@ namespace intel_cpu {
 class jit_uni_vcvtneps2bf16 : public jit_emitter {
 public:
     jit_uni_vcvtneps2bf16(dnnl::impl::cpu::x64::jit_generator* host, dnnl::impl::cpu::x64::cpu_isa_t host_isa,
-        InferenceEngine::Precision exec_prc = InferenceEngine::Precision::BF16) : jit_emitter(host, host_isa, exec_prc) {
+        ov::element::Type exec_prc = ov::element::bf16) : jit_emitter(host, host_isa, exec_prc) {
         if (!dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_bf16))
             prepare_table();
     }

diff --git a/src/plugins/intel_cpu/src/emitters/x64/jit_conversion_emitters.cpp b/src/plugins/intel_cpu/src/emitters/x64/jit_conversion_emitters.cpp
@@ -17,7 +17,7 @@ using namespace Xbyak;
 namespace ov {
 namespace intel_cpu {
 
-jit_convert_emitter::jit_convert_emitter(jit_generator *host, cpu_isa_t host_isa, const std::shared_ptr<ov::Node>& node, Precision exec_prc)
+jit_convert_emitter::jit_convert_emitter(jit_generator *host, cpu_isa_t host_isa, const std::shared_ptr<ov::Node>& node, ov::element::Type exec_prc)
 : jit_emitter(host, host_isa, exec_prc) {
     input_type = node->get_input_element_type(0);
     output_type = node->get_output_element_type(0);
@@ -58,7 +58,7 @@ void jit_convert_emitter::float2bfloat(const std::vector<size_t> &in_vec_idxs, c
 }
 
 jit_convert_truncation_emitter::jit_convert_truncation_emitter(jit_generator *host, cpu_isa_t host_isa,
-                                                               const std::shared_ptr<ov::Node>& node, Precision exec_prc)
+                                                               const std::shared_ptr<ov::Node>& node, ov::element::Type exec_prc)
         : jit_convert_emitter(host, host_isa, node, exec_prc) {
     prepare_table();
 }
@@ -193,7 +193,7 @@ void jit_convert_truncation_emitter::dword2int8(const std::vector<size_t> &in_ve
 }
 
 jit_convert_saturation_emitter::jit_convert_saturation_emitter(jit_generator *host, cpu_isa_t host_isa,
-                                                               const std::shared_ptr<ov::Node>& node, Precision exec_prc)
+                                                               const std::shared_ptr<ov::Node>& node, ov::element::Type exec_prc)
     : jit_convert_emitter(host, host_isa, node, exec_prc) {
 }
 

diff --git a/src/plugins/intel_cpu/src/emitters/x64/jit_conversion_emitters.hpp b/src/plugins/intel_cpu/src/emitters/x64/jit_conversion_emitters.hpp
@@ -14,7 +14,7 @@ namespace intel_cpu {
 class jit_convert_emitter : public jit_emitter {
 public:
     jit_convert_emitter(dnnl::impl::cpu::x64::jit_generator *host, dnnl::impl::cpu::x64::cpu_isa_t host_isa,
-                        const std::shared_ptr<ov::Node>& n, InferenceEngine::Precision exec_prc = InferenceEngine::Precision::FP32);
+                        const std::shared_ptr<ov::Node>& n, ov::element::Type exec_prc = ov::element::f32);
 
     size_t get_inputs_num() const override;
 
@@ -47,7 +47,7 @@ class jit_convert_emitter : public jit_emitter {
 class jit_convert_truncation_emitter : public jit_convert_emitter {
 public:
     jit_convert_truncation_emitter(dnnl::impl::cpu::x64::jit_generator *host, dnnl::impl::cpu::x64::cpu_isa_t host_isa,
-                                   const std::shared_ptr<ov::Node>& n, InferenceEngine::Precision exec_prc = InferenceEngine::Precision::FP32);
+                                   const std::shared_ptr<ov::Node>& n, ov::element::Type exec_prc = ov::element::f32);
 
 private:
     void emit_impl(const std::vector<size_t>& in, const std::vector<size_t>& out) const override;
@@ -68,7 +68,7 @@ class jit_convert_truncation_emitter : public jit_convert_emitter {
 class jit_convert_saturation_emitter : public jit_convert_emitter {
 public:
     jit_convert_saturation_emitter(dnnl::impl::cpu::x64::jit_generator *host, dnnl::impl::cpu::x64::cpu_isa_t host_isa,
-                                   const std::shared_ptr<ov::Node>& n, InferenceEngine::Precision exec_prc = InferenceEngine::Precision::FP32);
+                                   const std::shared_ptr<ov::Node>& n, ov::element::Type exec_prc = ov::element::f32);
 
 private:
     void emit_impl(const std::vector<size_t>& in, const std::vector<size_t>& out) const override;

diff --git a/src/plugins/intel_cpu/src/emitters/x64/jit_dnnl_emitters.cpp b/src/plugins/intel_cpu/src/emitters/x64/jit_dnnl_emitters.cpp
@@ -17,7 +17,7 @@ std::set<std::vector<element::Type>> jit_dnnl_emitter::get_supported_precisions(
     return {{element::f32}};
 }
 
-jit_dnnl_emitter::jit_dnnl_emitter(jit_generator *host, cpu_isa_t host_isa, const std::shared_ptr<ov::Node>& node, InferenceEngine::Precision exec_prc)
+jit_dnnl_emitter::jit_dnnl_emitter(jit_generator *host, cpu_isa_t host_isa, const std::shared_ptr<ov::Node>& node, ov::element::Type exec_prc)
     : jit_emitter(host, host_isa, exec_prc) {
 
     kind = dnnl_eltwise_tanh;
@@ -29,7 +29,7 @@ jit_dnnl_emitter::jit_dnnl_emitter(jit_generator *host, cpu_isa_t host_isa, cons
 
 jit_dnnl_emitter::jit_dnnl_emitter(jit_generator *host, cpu_isa_t host_isa,
                                    dnnl_alg_kind_t algKind, float alpha, float beta,
-                                   InferenceEngine::Precision exec_prc)
+                                   ov::element::Type exec_prc)
     : jit_emitter(host, host_isa, exec_prc), kind(algKind), alpha(alpha), beta(beta) {
 
     set_injector();
@@ -85,7 +85,7 @@ void jit_dnnl_emitter::emit_data() const {
 
 jit_dnnl_aux_emitter::jit_dnnl_aux_emitter(jit_generator *host, cpu_isa_t host_isa,
                                            dnnl_alg_kind_t algKind, float inpAlpha, float inpBeta,
-                                           InferenceEngine::Precision exec_prc)
+                                           ov::element::Type exec_prc)
     : jit_dnnl_emitter(host, host_isa, algKind, inpAlpha, inpBeta, exec_prc) {
 }
 

diff --git a/src/plugins/intel_cpu/src/emitters/x64/jit_dnnl_emitters.hpp b/src/plugins/intel_cpu/src/emitters/x64/jit_dnnl_emitters.hpp
@@ -25,9 +25,9 @@ class jit_dnnl_emitter : public jit_emitter {
 protected:
     jit_dnnl_emitter(dnnl::impl::cpu::x64::jit_generator *host, dnnl::impl::cpu::x64::cpu_isa_t host_isa,
                        dnnl_alg_kind_t algKind, float inpAlpha, float inpBeta,
-                       InferenceEngine::Precision exec_prc = InferenceEngine::Precision::FP32);
+                       ov::element::Type exec_prc = ov::element::f32);
     jit_dnnl_emitter(dnnl::impl::cpu::x64::jit_generator *host, dnnl::impl::cpu::x64::cpu_isa_t host_isa, const std::shared_ptr<ov::Node>& n,
-                       InferenceEngine::Precision exec_prc = InferenceEngine::Precision::FP32);
+                       ov::element::Type exec_prc = ov::element::f32);
     void set_injector();
 
     dnnl_alg_kind_t kind {dnnl_alg_kind_undef};
@@ -46,7 +46,7 @@ class jit_dnnl_aux_emitter : public jit_dnnl_emitter {
 public:
     jit_dnnl_aux_emitter(dnnl::impl::cpu::x64::jit_generator *host, dnnl::impl::cpu::x64::cpu_isa_t host_isa,
                            dnnl_alg_kind_t algKind, float inpAlpha, float inpBeta,
-                           InferenceEngine::Precision exec_prc = InferenceEngine::Precision::FP32);
+                           ov::element::Type exec_prc = ov::element::f32);
 
 private:
 };