[CPU] Fix ScaledDotProductAttention build failure on ubuntu18 (openvi…

…notoolkit#21123)
ababushk · Nov 17, 2023 · e4311aa · e4311aa
1 parent 6bdc159
commit e4311aa
Show file tree

Hide file tree

Showing 3 changed files with 7 additions and 17 deletions.
diff --git a/src/plugins/intel_cpu/src/nodes/scaled_attn.cpp b/src/plugins/intel_cpu/src/nodes/scaled_attn.cpp
@@ -753,7 +753,6 @@ void ScaledDotProductAttention::execute(dnnl::stream strm) {
 }
 
 bool ScaledDotProductAttention::isSupportedOperation(const std::shared_ptr<const ngraph::Node>& op, std::string& errorMessage) noexcept {
-#if defined(OPENVINO_ARCH_X86_64)
     try {
         if (!std::dynamic_pointer_cast<const ov::op::v13::ScaledDotProductAttention>(op)) {
             errorMessage = "Only ScaledDotProductAttention operation are supported";
@@ -774,10 +773,6 @@ bool ScaledDotProductAttention::isSupportedOperation(const std::shared_ptr<const
         return false;
     }
     return true;
-#else
-    // current optimization is not suitable for ARM
-    return false;
-#endif
 }
 
 }  // namespace node

diff --git a/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp b/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp
@@ -433,7 +433,7 @@ void Transformations::PreLpt(const std::vector<ov::element::Type>& defaultPrecis
     CPU_SET_CALLBACK_COMMON(manager, nmsCallback, ov::pass::ConvertNMS9ToNMSIEInternal);
     CPU_SET_CALLBACK_COMMON(manager, nmsCallback, ov::pass::ConvertMulticlassNmsToMulticlassNmsIE);
     CPU_SET_CALLBACK_COMMON(manager, nmsCallback, ov::pass::ConvertMatrixNmsToMatrixNmsIE);
-    CPU_SET_CALLBACK_COMMON(manager,
+    CPU_SET_CALLBACK_X64(manager,
         [](const_node_ptr &node) -> bool {
             std::string errorMsg;
             return node::ScaledDotProductAttention::isSupportedOperation(node, errorMsg);
@@ -470,7 +470,6 @@ void Transformations::PreLpt(const std::vector<ov::element::Type>& defaultPrecis
     CPU_DISABLE_PASS_COMMON(manager, ov::pass::ConvertTopK11ToTopK3);
     CPU_DISABLE_PASS_COMMON(manager, ov::pass::HSwishDecomposition);
     CPU_DISABLE_PASS_COMMON(manager, ov::pass::MatMulConstTransposesExtraction);
-    CPU_DISABLE_PASS_X64(manager, ov::pass::ScaledDotProductAttentionDecomposition);
     CPU_DISABLE_PASS_X64(manager, ov::pass::HSigmoidDecomposition);
 
     CPU_DISABLE_PASS_X64(manager, ov::pass::ReduceL1Decomposition);

diff --git a/src/plugins/intel_cpu/src/utils/plain_tensor.hpp b/src/plugins/intel_cpu/src/utils/plain_tensor.hpp
@@ -109,13 +109,16 @@ struct precision_of<float16> {
 };
 
 #define PLAINTENSOR_RANK_MAX 8
-struct PlainTensorBase {
+
+template <typename DT>
+struct PlainTensor {
     size_t m_strides[PLAINTENSOR_RANK_MAX];
     size_t m_dims[PLAINTENSOR_RANK_MAX];
     size_t m_rank = 0;
     void* m_ptr = nullptr;
     size_t m_capacity = 0;
     bool with_storage = false;
+    MemoryPtr m_mem;        // hold memory ptr reference
 
     operator bool() const {
         return static_cast<bool>(m_ptr);
@@ -135,13 +138,6 @@ struct PlainTensorBase {
         assert(i < m_rank);
         return m_strides[i];
     }
-    virtual ov::element::Type get_precision(void) = 0;
-    virtual void reset(MemoryPtr mem) = 0;
-};
-
-template <typename DT>
-struct PlainTensor : public PlainTensorBase {
-    MemoryPtr m_mem;        // hold memory ptr reference
     PlainTensor(MemoryPtr mem) {
         reset(mem);
     }
@@ -172,14 +168,14 @@ struct PlainTensor : public PlainTensorBase {
         }
     }
 
-    void reset(MemoryPtr mem) override {
+    void reset(MemoryPtr mem) {
         assert_dt<DT>(mem->getDesc().getPrecision());
         m_mem = mem;
         // this reshape_to() can do reshape w/o additional cost
         resize(mem->getStaticDims(), reinterpret_cast<DT*>(mem->getData()));
     }
 
-    ov::element::Type get_precision(void) override {
+    ov::element::Type get_precision(void) {
         return precision_of<DT>::value;
     }