Merge branch 'openvinotoolkit:master' into softmax_cross_entropy_loss…

…_for_onnx_fe
openvinotoolkit · Feb 8, 2025 · 4242b3c · 4242b3c
2 parents 9a44591 + 06c3f20
commit 4242b3c
Show file tree

Hide file tree

Showing 6 changed files with 35 additions and 6 deletions.
diff --git a/src/frontends/onnx/frontend/include/openvino/frontend/onnx/node_context.hpp b/src/frontends/onnx/frontend/include/openvino/frontend/onnx/node_context.hpp
@@ -21,6 +21,8 @@ class ONNX_FRONTEND_API NodeContext : public ov::frontend::NodeContext {
 
     Output<ov::Node> get_input(int port_idx) const override;
 
+    const std::string& get_name() const override;
+
     ov::Any get_attribute_as_any(const std::string& name) const override;
 
 protected:

diff --git a/src/frontends/onnx/frontend/src/node_context.cpp b/src/frontends/onnx/frontend/src/node_context.cpp
@@ -15,6 +15,10 @@ ov::Output<ov::Node> ov::frontend::onnx::NodeContext::get_input(int port_idx) co
     return m_inputs.at(port_idx);
 }
 
+const std::string& ov::frontend::onnx::NodeContext::get_name() const {
+    return m_context.get_name();
+}
+
 ov::Any ov::frontend::onnx::NodeContext::get_attribute_as_any(const std::string& name) const {
     try {
         return m_context.get_attribute_value<ov::Any>(name);

diff --git a/src/frontends/pytorch/src/op_table.cpp b/src/frontends/pytorch/src/op_table.cpp
@@ -882,6 +882,7 @@ const std::unordered_map<std::string, CreatorFunction> get_supported_ops_fx() {
         {"aten.hardtanh.default", op::translate_hardtanh},
         {"aten.hardtanh_.default", op::inplace_op<op::translate_hardtanh>},
         {"aten.index.Tensor", op::translate_index_fx},
+        {"aten._unsafe_index.Tensor", op::translate_index_fx},
         {"aten.index_select.default", op::translate_index_select},
         {"aten.isfinite.default", op::translate_1to1_match_1_inputs<opset10::IsFinite>},
         {"aten.isinf.default", op::translate_1to1_match_1_inputs<opset10::IsInf>},

diff --git a/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp b/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp
@@ -1077,7 +1077,7 @@ void Transformations::MainSnippets(void) {
         const auto is_bf16 = (in_type0 == ov::element::bf16 && in_type1 == ov::element::bf16) ||
                              ((in_type0 == element::f32 && in_type1 == ov::element::f32 &&
                                config.inferencePrecision == ov::element::bf16));
-        const auto is_int8 = in_type0 == ov::element::i8;
+        const auto is_int8 = (in_type0 == element::i8 || in_type0 == element::u8) && (in_type1 == element::i8);
         if (matmul->get_transpose_a()) {
             return false;
         }
@@ -1096,7 +1096,7 @@ void Transformations::MainSnippets(void) {
         if (is_fp16) {
             return dnnl::impl::cpu::x64::mayiuse(dnnl::impl::cpu::x64::avx512_core_amx_fp16);
         }
-        return true;
+        return false;
     };
     auto is_unsupported_parallel_work_amount = [&](const std::shared_ptr<const ov::Node>& n,
                                                    const ov::PartialShape& shape) {

diff --git a/src/plugins/intel_npu/src/plugin/npuw/just_sync_infer_request.cpp b/src/plugins/intel_npu/src/plugin/npuw/just_sync_infer_request.cpp
@@ -733,9 +733,10 @@ void ov::npuw::JustInferRequest::unsafe_infer(std::size_t real_idx) {
             // Collect spatial inputs for this offset
             for (auto&& param : spatial.params) {
                 const auto& iport = comp_model_desc.compiled_model->inputs()[param.idx];
-                r->set_tensor(
-                    iport,
-                    ov::npuw::util::view(m_spatial_io[real_idx].inputs.at(param.idx), param.dim, offset, spatial.nway));
+                const auto& iview = ov::npuw::util::view(m_spatial_io[real_idx].inputs.at(param.idx),
+                                                         param.dim, offset,
+                                                         spatial.nway);
+                r->set_tensor(iport, iview);
             }  // for(params)
 
             // Now set the spatial outputs

diff --git a/src/plugins/intel_npu/src/plugin/npuw/util.cpp b/src/plugins/intel_npu/src/plugin/npuw/util.cpp
@@ -360,11 +360,32 @@ ov::SoPtr<ov::ITensor> ov::npuw::util::view(const ov::SoPtr<ov::ITensor>& src,
                                             std::size_t offset,
                                             std::size_t len) {
     const auto& shape = src->get_shape();
+    NPUW_ASSERT(dim < shape.size());
     View view_start = View(shape.size(), 0u);
     View view_end = shape;
     view_start[dim] = offset;
     view_end[dim] = offset + len;
-    return ov::npuw::util::view(src, view_start, view_end);
+
+    auto ret_view = ov::npuw::util::view(src, view_start, view_end);
+
+    // Check if a tensor view can be faked as "continuous"
+    // FIXME: This trick should be removed after strided tensor
+    // checks are relaxed
+    if (std::all_of(shape.begin(), shape.begin() + dim, [](std::size_t d) { return d == 1u; })) {
+        // If all dimensions up to the sub-ranged dimension are 1s,
+        // This tensor can be faked as continuous
+        const auto type = src->get_element_type();
+        const auto view_shape = ret_view->get_shape();
+        ov::Strides fake_strides(shape.size());
+        fake_strides.back() = type.size();
+        std::transform(view_shape.crbegin(),
+                       view_shape.crend() - 1,
+                       fake_strides.rbegin(),
+                       fake_strides.rbegin() + 1,
+                       std::multiplies<size_t>());
+        return ov::get_tensor_impl(ov::Tensor(type, view_shape, ret_view->data(), fake_strides));
+    }
+    return ret_view;
 }
 
 template <typename InT>