[GPU] Enable output transposed gemm for onednn

openvinotoolkit · Nov 28, 2024 · e9a9996 · e9a9996
1 parent 79493c2
commit e9a9996
Show file tree

Hide file tree

Showing 2 changed files with 101 additions and 2 deletions.
diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/gemm_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/gemm_onednn.cpp
@@ -186,8 +186,15 @@ struct gemm_onednn : typed_primitive_onednn_impl<gemm> {
             if (ret) {
                 tag = convert_data_format(transposed_format);
                 dnnl::memory::dims original_dims = dims;
-                for (size_t i = 0; i < original_dims.size(); ++i) {
-                    dims[i] = original_dims[order[i]];
+                if (is_input) {
+                    for (size_t i = 0; i < original_dims.size(); ++i) {
+                        dims[i] = original_dims[order[i]];
+                    }
+                } else {
+                    // Get non-transposed dims for output dims
+                    for (size_t i = 0; i < original_dims.size(); ++i) {
+                        dims[order[i]] = original_dims[i];
+                    }
                 }
             } else {
                 std::ostringstream ostream;

diff --git a/...el_gpu/tests/functional/shared_tests_instances/subgraph_tests/transpose_matmul_fusion.cpp b/...el_gpu/tests/functional/shared_tests_instances/subgraph_tests/transpose_matmul_fusion.cpp
@@ -96,3 +96,95 @@ TEST_P(TransposeMatMulFusionOnGPU, CompareWithRefs){
 };
 
 }  // namespace
+
+
+//=================================================================================
+// Transpose + MatMul + Transpose pattern fusion (TransposeMatMulTransposeMatcher)
+//=================================================================================
+namespace ov {
+namespace test {
+
+using MatMulTransposeFusionParams = std::tuple<ov::PartialShape,  // input A shapes
+                                        ov::PartialShape,         // input B shapes
+                                        ov::PartialShape>;        // input C shapes
+class MatMulTransposeFusionOnGPU: public testing::WithParamInterface<MatMulTransposeFusionParams>,
+                     virtual public ov::test::SubgraphBaseTest {
+public:
+    static std::string getTestCaseName(testing::TestParamInfo<MatMulTransposeFusionParams> obj) {
+        ov::PartialShape input0;
+        ov::PartialShape input1;
+        ov::PartialShape input2;
+
+        std::tie(input0, input1, input2) = obj.param;
+
+        std::ostringstream result;
+        result << "device=(" << std::string(utils::DEVICE_GPU) << ")_";
+        result << ov::test::utils::partialShape2str({input0}) << "_";
+        result << ov::test::utils::partialShape2str({input1}) << "_";
+        result << ov::test::utils::partialShape2str({input2}) << "_";
+        return result.str();
+    }
+protected:
+    void SetUp() override {
+        targetDevice = ov::test::utils::DEVICE_GPU;
+
+        ov::PartialShape shape1;
+        ov::PartialShape shape2;
+        ov::PartialShape shape3;
+
+        std::tie(shape1, shape2, shape3) = GetParam();
+
+        InputShape input_shape1 = {shape1, {shape1.get_shape()}};
+        InputShape input_shape2 = {shape2, {shape2.get_shape()}};
+        InputShape input_shape3 = {shape3, {shape3.get_shape()}};
+        init_input_shapes({input_shape1, input_shape2, input_shape3});
+
+        const auto param1 = std::make_shared<ov::op::v0::Parameter>(ov::element::f16, shape1);
+        const auto param2 = std::make_shared<ov::op::v0::Parameter>(ov::element::f16, shape2);
+        const auto param3 = std::make_shared<ov::op::v0::Parameter>(ov::element::f16, shape3);
+
+        auto input2_shape = shape2.get_shape();
+
+        //input0
+        const auto input0_order = ov::op::v0::Constant::create(ov::element::i32, Shape{4}, {1, 0, 2, 3});
+        const auto input0_transpose = std::make_shared<ov::op::v1::Transpose>(param1, input0_order);
+        const auto input0_shape_pattern = ov::op::v0::Constant::create(ov::element::i32, Shape{4}, input2_shape);
+        const auto input0_reshape = std::make_shared<ov::op::v1::Reshape>(input0_transpose, input0_shape_pattern, false);
+
+        //input1
+        const auto input1_order = ov::op::v0::Constant::create(ov::element::i32, Shape{4}, {0, 1, 3, 2});
+        const auto input1_transpose = std::make_shared<ov::op::v1::Transpose>(param2, input1_order);
+
+        // matmul & softmax
+        const auto matmul1 = std::make_shared<ov::op::v0::MatMul>(input0_reshape, input1_transpose, false, false);
+        const auto softmax = std::make_shared<ov::op::v8::Softmax>(matmul1, -1);
+
+        // input3
+        const auto input3_transpose = std::make_shared<ov::op::v1::Transpose>(param3, input0_order);
+        const auto input3_shape_pattern = ov::op::v0::Constant::create(ov::element::i32, Shape{4}, input2_shape);
+        const auto input3_reshape = std::make_shared<ov::op::v1::Reshape>(input3_transpose, input3_shape_pattern, false);
+
+        // target matmul
+        const auto matmul2 = std::make_shared<ov::op::v0::MatMul>(softmax, input3_reshape, false, false);
+        const auto order = ov::op::v0::Constant::create(ov::element::i32, Shape{4}, {2, 0, 1, 3});
+        const auto transpose = std::make_shared<ov::op::v1::Transpose>(matmul2, order);
+
+        function = std::make_shared<ov::Model>(transpose, ov::ParameterVector{param1, param2, param3});
+    }
+};
+
+
+}  // namespace test
+}  // namespace ov
+
+
+namespace {
+INSTANTIATE_TEST_SUITE_P(smoke_MatMulTransposeFusion, MatMulTransposeFusionOnGPU,
+                         ::testing::Values(
+                            MatMulTransposeFusionParams({3, 8, 16, 1}, {2, 4, 3, 16}, {3, 8, 16, 1})),
+                         MatMulTransposeFusionOnGPU::getTestCaseName);
+
+TEST_P(MatMulTransposeFusionOnGPU, CompareWithRefs){
+    run();
+};
+}  // namespace