[GPU] Add more mixed type of bfyx to eltwise_blocked_opt (#27548)

Improve yolo_v5m peformance by using eltwise_blocked_opt for batchNormalization which help for platforms has XMX ### Tickets: - *cvs-152243* Signed-off-by: Chon Ming Lee <[email protected]>
openvinotoolkit · Nov 29, 2024 · 7b4e6af · 7b4e6af
1 parent e775b4c
commit 7b4e6af
Show file tree

Hide file tree

Showing 2 changed files with 25 additions and 3 deletions.
diff --git a/src/plugins/intel_gpu/src/kernel_selector/kernels/eltwise/eltwise_kernel_blocked_opt.cpp b/src/plugins/intel_gpu/src/kernel_selector/kernels/eltwise/eltwise_kernel_blocked_opt.cpp
@@ -112,7 +112,7 @@ bool EltwiseKernel_blocked_opt::Validate(const Params& params) const {
     }
 
     const auto vec_size = SelectVecSizeFromFormat(ewParams.outputs[0]);
-    const auto input0 = ewParams.inputs[0];
+    const auto& input0 = ewParams.inputs[0];
     const auto& output = ewParams.outputs[0];
     // Check that padding before features doesn't mis-align the blocks
     if (input0.Feature().pad.before % vec_size != 0 || output.Feature().pad.before % vec_size != 0)
@@ -137,11 +137,22 @@ bool EltwiseKernel_blocked_opt::Validate(const Params& params) const {
     };
 
     for (size_t i = 1; i < ewParams.inputs.size(); i++) {
-        if (ewParams.inputs[i].LogicalSize() == input0.LogicalSize() && !(compareTensors(ewParams.inputs[i], input0)))
+        const auto& input = ewParams.inputs[i];
+        if (input.LogicalSize() == input0.LogicalSize() && !(compareTensors(input, input0)))
             return false;
-        if (ewParams.inputs[i].Feature().pad.before % vec_size != 0) {
+        if (input.Feature().pad.before % vec_size != 0) {
             return false;
         }
+        if (input.GetLayout() == DataLayout::bfyx) {
+            bool is_valid = input.LogicalSize() == 1;           // Scalar value broadcast
+            is_valid |= input.LogicalSize() % vec_size == 0 &&  // Feature value broadcast
+                        input.LogicalSize() == input.Feature().v &&
+                        input.LogicalSize() == output.Feature().v &&
+                        GetInnerBatchBlockSize(input) == 1;
+            if (!is_valid) {
+                return false;
+            }
+        }
     }
 
     return true;
@@ -422,6 +433,7 @@ static inline int SelectVecSizeFromFormat(const DataTensor& tensor) {
 static inline int GetInnerBatchBlockSize(const DataTensor& tensor) {
     auto layout = tensor.GetLayout();
     switch (layout) {
+    case DataLayout::bfyx:
     case DataLayout::b_fs_yx_fsv4:
     case DataLayout::b_fs_yx_fsv16:
     case DataLayout::b_fs_zyx_fsv16:

diff --git a/src/plugins/intel_gpu/tests/unit/test_cases/eltwise_gpu_test.cpp b/src/plugins/intel_gpu/tests/unit/test_cases/eltwise_gpu_test.cpp
@@ -4711,6 +4711,11 @@ struct eltwise_layout_test_params {
 #define CASE_ELTWISE_TEST7  eltwise_mode::sum, {4, 5, 4, 1}, {4, 1, 4, 1}, format::bfyx, format::b_fs_yx_fsv16, "generic_eltwise_ref"
 #define CASE_ELTWISE_TEST8  eltwise_mode::sum, {4, 2, 4, 4}, {1, 1, 1, 1}, format::bfyx, format::b_fs_yx_fsv16, "generic_eltwise_ref"
 #define CASE_ELTWISE_TEST9  eltwise_mode::eq,  {4, 2, 4, 4}, {1, 1, 1, 1}, format::b_fs_yx_fsv16, format::bfyx, "generic_eltwise_ref"
+#define CASE_ELTWISE_TEST10 eltwise_mode::sum, {4, 8, 1, 1}, {1, 8, 1, 1}, format::b_fs_yx_fsv32, format::bfyx, "eltwise_blocked_opt"
+#define CASE_ELTWISE_TEST11 eltwise_mode::sum, {4, 8, 1, 1}, {1, 8, 1, 1}, format::b_fs_yx_fsv16, format::bfyx, "eltwise_blocked_opt"
+#define CASE_ELTWISE_TEST12 eltwise_mode::sum, {4, 16, 4, 4}, {1, 16, 1, 1}, format::b_fs_yx_fsv16, format::bfyx, "eltwise_blocked_opt"
+#define CASE_ELTWISE_TEST13 eltwise_mode::sum, {4, 7, 4, 4}, {1, 7, 1, 1}, format::b_fs_yx_fsv16, format::bfyx, "generic_eltwise_ref"
+#define CASE_ELTWISE_TEST14 eltwise_mode::sum, {1, 8, 1, 1}, {4, 8, 1, 1}, format::bfyx, format::b_fs_yx_fsv32, "generic_eltwise_ref"
 
 class eltwise_layout_test : public BaseEltwiseTest<eltwise_layout_test_params> {
 public:
@@ -4800,6 +4805,11 @@ INSTANTIATE_TEST_SUITE_P(eltwise, eltwise_test_mixed_layout,
                             eltwise_layout_test_params{CASE_ELTWISE_TEST7},
                             eltwise_layout_test_params{CASE_ELTWISE_TEST8},
                             eltwise_layout_test_params{CASE_ELTWISE_TEST9},
+                            eltwise_layout_test_params{CASE_ELTWISE_TEST10},
+                            eltwise_layout_test_params{CASE_ELTWISE_TEST11},
+                            eltwise_layout_test_params{CASE_ELTWISE_TEST12},
+                            eltwise_layout_test_params{CASE_ELTWISE_TEST13},
+                            eltwise_layout_test_params{CASE_ELTWISE_TEST14},
                         }));
 
 //