[GPU] Unified FillCLKernelData's parameter that affects adding SHAPE_INFO to arguments during kernel generating

Author: sshlyapn

src/plugins/intel_gpu/src/kernel_selector/kernels/activation/activation_kernel_base.cpp

@@ -109,7 +109,7 @@ KernelsData ActivationKernelBase::GetCommonKernelsData(const Params& params) con
     auto& kernel = kd.kernels[0];
     FillCLKernelData(kernel, dispatchData, params.engineInfo, kernelName, jit, entry_point,
                      EXE_MODE_DEFAULT, false, false, 1,
-                     GetFusedPrimitiveInputsCount(params), 1, newParams.outputs[0].is_dynamic());
+                     GetFusedPrimitiveInputsCount(params), 1, newParams.is_shape_agnostic);
 
     if (!newParams.inputActivationParams.empty()) {
         kernel.params.arguments.push_back({ArgumentDescriptor::Types::SLOPE, 0});

src/plugins/intel_gpu/src/kernel_selector/kernels/arg_max_min/arg_max_min_kernel_axis.cpp

@@ -165,7 +165,7 @@ KernelsData ArgMaxMinKernelAxis::GetKernelsData(const Params& params) const {
                      1,
                      GetFusedPrimitiveInputsCount(params),
                      orgParams.use_multiple_outputs ? 2 : 1,
-                     is_dynamic);
+                     orgParams.is_shape_agnostic);
 
     if (orgParams.has_second_output && !orgParams.use_multiple_outputs)
         kernel.params.arguments.push_back({ArgumentDescriptor::Types::INPUT, 1});

src/plugins/intel_gpu/src/kernel_selector/kernels/arg_max_min/arg_max_min_kernel_base.cpp

@@ -79,7 +79,7 @@ KernelsData ArgMaxMinKernelBase::GetCommonKernelsData(const Params& params) cons
                      (uint32_t)orgParams.inputs.size(),
                      GetFusedPrimitiveInputsCount(params),
                      (uint32_t)orgParams.outputs.size(),
-                     orgParams.has_dynamic_tensors());
+                     orgParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/batch_to_space/batch_to_space_kernel_base.cpp

@@ -114,7 +114,7 @@ KernelsData BatchToSpaceKernelBase::GetCommonKernelsData(const Params& params) c
 
     FillCLKernelData(kernel, dispatchData, params.engineInfo, kernelName, jit, entry_point,
                      "", false, false, static_cast<int>(newParams.inputs.size()),
-                     GetFusedPrimitiveInputsCount(params), 1, newParams.has_dynamic_tensors());
+                     GetFusedPrimitiveInputsCount(params), 1, newParams.is_shape_agnostic);
 
     return { kd };
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/beam_table_update/beam_table_update_kernel_ref.cpp

@@ -53,7 +53,7 @@ KernelsData BeamTableUpdateKernelRef::GetKernelsData(const Params& params) const
                      static_cast<int>(kernel_params.inputs.size()),
                      GetFusedPrimitiveInputsCount(kernel_params),
                      static_cast<int>(kernel_params.outputs.size()),
-                     kernel_params.outputs[0].is_dynamic());
+                     kernel_params.is_shape_agnostic);
 
     ScalarDescriptor is_state_set;
     is_state_set.t = ScalarDescriptor::Types::UINT8;

src/plugins/intel_gpu/src/kernel_selector/kernels/border/border_kernel_base.cpp

@@ -104,7 +104,7 @@ KernelsData BorderKernelBase::GetCommonKernelsData(const Params& params) const {
                      (uint32_t)prim_params.inputs.size(),
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     prim_params.outputs[0].is_dynamic());
+                     prim_params.is_shape_agnostic);
 
     return {k_data};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/broadcast/broadcast_kernel_base.cpp

@@ -146,7 +146,7 @@ KernelsData BroadcastKernelBase::GetCommonKernelsData(const Params& params) cons
                      1,
                      0,
                      1,
-                     prim_params.inputs[0].is_dynamic() || prim_params.outputs[0].is_dynamic());
+                     prim_params.is_shape_agnostic);
 
     return {k_data};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/convolution/convolution_kernel_base.cpp

@@ -232,7 +232,7 @@ KernelsData ConvolutionKernelBase::GetCommonKernelsData(const Params& params,
                      true,
                      !newParams.bias.empty(),
                      1, 0, 1,
-                     newParams.inputs[0].is_dynamic() || newParams.outputs[0].is_dynamic());
+                     newParams.is_shape_agnostic);
 
     if (newParams.deformable_mode) {
         kernel.params.arguments.push_back({ArgumentDescriptor::Types::INPUT, 1});

src/plugins/intel_gpu/src/kernel_selector/kernels/convolution/convolution_kernel_bfyx_os_iyx_osv16.h

@@ -36,7 +36,7 @@ class ConvolutionKernel_bfyx_os_iyx_osv16 : public ConvolutionKernelBase {
     bool NeedPaddedInput() const override { return true; }
     DispatchData SetDefault(const convolution_params& arg, int autoTuneIndex = -1) const override;
     size_t GetSubGroupSize(const convolution_params& params) const {
-        if (params.engineInfo.computeUnitsCount <= 24) {
+        if (params.engineInfo.computeUnitsCount <= 24 && !params.is_shape_agnostic) {
             // Smaller # EU tends to be computation bounds.
             // In such case, using larger worksize will result in larger computational inefficiency
             // w.r.t the unalined output feature

src/plugins/intel_gpu/src/kernel_selector/kernels/cum_sum/cum_sum_kernel_base.cpp

@@ -98,7 +98,7 @@ KernelsData CumSumKernelBase::GetCommonKernelsData(const Params& params) const {
                      "", false, false, 1,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     newParams.outputs[0].is_dynamic());
+                     newParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/eltwise/eltwise_kernel_base.cpp

@@ -757,7 +757,7 @@ KernelsData EltwiseKernelBase::GetCommonKernelsData(const Params& params) const
 
     kernel.params.workGroups.global = dispatchData.gws;
     kernel.params.workGroups.local = dispatchData.lws;
-    bool is_dynamic = newParams.has_dynamic_tensors();
+    bool is_dynamic = newParams.is_shape_agnostic;
     kernel.params.arguments = GetArgsDesc((uint32_t)newParams.inputs.size(),
                                    false,
                                    false,

src/plugins/intel_gpu/src/kernel_selector/kernels/fully_connected/fully_connected_kernel_base.cpp

@@ -146,7 +146,7 @@ KernelsData FullyConnectedKernelBase::GetCommonKernelsData(const Params &params,
                      inputs_count,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     orgParams.outputs[0].is_dynamic());
+                     orgParams.is_shape_agnostic);
 
     // TODO Pass estimated time only through DispatchData
     kd.autoTuneIndex = autoTuneIndex;

src/plugins/intel_gpu/src/kernel_selector/kernels/gather/gather_elements_kernel_ref.cpp

@@ -178,7 +178,7 @@ KernelsData GatherElementsKernelRef::GetKernelsData(const Params& params) const
 
     auto& kernel = kd.kernels[0];
     FillCLKernelData(kernel, dispatchData, params.engineInfo, kernelName, jit, entry_point,
-                     "", false, false, 2, GetFusedPrimitiveInputsCount(params), 1, newParams.has_dynamic_tensors());
+                     "", false, false, 2, GetFusedPrimitiveInputsCount(params), 1, newParams.is_shape_agnostic);
     return { kd };
 }
 

src/plugins/intel_gpu/src/kernel_selector/kernels/gather/gather_kernel_ref.cpp

@@ -385,7 +385,7 @@ KernelsData GatherKernelRef::GetKernelsData(const Params& params) const {
                      inputs_count,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     newParams.has_dynamic_tensors());
+                     newParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/gather/gather_nd_kernel_ref.cpp

@@ -212,7 +212,7 @@ KernelsData GatherNDKernelRef::GetKernelsData(const Params& params) const {
                      2,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     newParams.has_dynamic_tensors());
+                     newParams.is_shape_agnostic);
 
     return { kd };
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/gemm/gemm_kernel_base.cpp

@@ -278,7 +278,7 @@ KernelsData GemmKernelBase::GetCommonKernelsData(const Params& params) const {
                      (uint32_t)prim_params.inputs.size(),
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     prim_params.has_dynamic_tensors());
+                     prim_params.is_shape_agnostic);
 
     return {k_data};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/mvn/mvn_kernel_base.cpp

@@ -90,7 +90,7 @@ KernelsData MVNKernelBase::GetCommonKernelsData(const Params& params) const {
                      1,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     orgParams.outputs[0].is_dynamic());
+                     orgParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/non_zero/count_nonzero_kernel_ref.cpp

@@ -102,7 +102,7 @@ KernelsData CountNonzeroKernelRef::GetKernelsData(const Params& params) const {
                      1,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     newParams.inputs[0].is_dynamic());
+                     newParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/non_zero/gather_nonzero_kernel_ref.cpp

@@ -105,7 +105,7 @@ KernelsData GatherNonzeroKernelRef::GetKernelsData(const Params& params) const {
                      2,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     newParams.outputs[0].is_dynamic());
+                     newParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/permute/permute_kernel_base.cpp

@@ -69,7 +69,7 @@ KernelsData PermuteKernelBase::GetKernelsData(const Params& params) const {
                      1,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     newParams.outputs[0].is_dynamic());
+                     newParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/reduce/reduce_kernel_base.cpp

@@ -271,7 +271,7 @@ KernelsData ReduceKernelBase::GetCommonKernelsData(const Params& p) const {
                      1,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     params.inputs[0].is_dynamic());
+                     params.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/reorder/reorder_kernel_base.cpp

@@ -259,9 +259,9 @@ KernelsData ReorderKernelBase::GetCommonKernelsData(const reorder_params& params
                      1,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     newParams.outputs[0].is_dynamic());
+                     newParams.is_shape_agnostic);
 
-    kernel.params.arguments = GetArgsDesc(1, false, false, GetFusedPrimitiveInputsCount(params), 1, newParams.outputs[0].is_dynamic());
+    kernel.params.arguments = GetArgsDesc(1, false, false, GetFusedPrimitiveInputsCount(params), 1, newParams.is_shape_agnostic);
     if (newParams.mode == MeanSubtractMode::IN_BUFFER) {
         kernel.params.arguments.push_back({ArgumentDescriptor::Types::BIAS, 0});
     }

src/plugins/intel_gpu/src/kernel_selector/kernels/rms/rms_kernel_base.cpp

@@ -78,7 +78,7 @@ KernelsData RMSKernelBase::GetCommonKernelsData(const Params& params) const {
                      2,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     orgParams.outputs[0].is_dynamic());
+                     orgParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/scatter_update/scatter_nd_update_kernel_ref.cpp

@@ -222,7 +222,7 @@ KernelsData ScatterNDUpdateKernelRef::GetKernelsData(const Params& params) const
         clKernelData& kernel = kd.kernels[i];
 
         FillCLKernelData(kernel, dispatchData, params.engineInfo, kernelName, jit, entry_point,
-                         "", false, false, inputs_number, GetFusedPrimitiveInputsCount(params), 1, newParams.has_dynamic_tensors());
+                         "", false, false, inputs_number, GetFusedPrimitiveInputsCount(params), 1, newParams.is_shape_agnostic);
     }
 
     return {kd};

src/plugins/intel_gpu/src/kernel_selector/kernels/scatter_update/scatter_update_kernel_ref.cpp

@@ -351,7 +351,7 @@ KernelsData ScatterUpdateKernelRef::GetKernelsData(const Params& params) const {
         clKernelData& kernel = kd.kernels[i - start_with_iteration];
 
         FillCLKernelData(kernel, dispatchData, params.engineInfo, kernelName, jit, entry_point,
-                         "", false, false, 3, GetFusedPrimitiveInputsCount(params), 1, newParams.has_dynamic_tensors());
+                         "", false, false, 3, GetFusedPrimitiveInputsCount(params), 1, newParams.is_shape_agnostic);
     }
 
     return {kd};

src/plugins/intel_gpu/src/kernel_selector/kernels/select/select_kernel_base.cpp

@@ -142,7 +142,7 @@ KernelsData SelectKernelBase::GetCommonKernelsData(const Params& params) const {
                      (uint32_t)newParams.inputs.size(),
                      0,
                      1,
-                     newParams.outputs[0].is_dynamic());
+                     newParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/shape_of/shape_of_kernel_ref.cpp

@@ -60,7 +60,7 @@ KernelsData ShapeOfKernelRef::GetKernelsData(const Params &params) const {
     GetUpdateDispatchDataFunc(kernel_data);
 
     FillCLKernelData(clKernelData, dispatch_data, params.engineInfo, kernelName, jit, entry_point, EXE_MODE_DEFAULT,
-                     false, false, 0, 0, 1, derived_params.inputs[0].is_dynamic());
+                     false, false, 0, 0, 1, derived_params.is_shape_agnostic);
     return kernels_data;
 }
 

src/plugins/intel_gpu/src/kernel_selector/kernels/softmax/softmax_kernel_base.cpp

@@ -69,7 +69,7 @@ KernelsData SoftmaxKernelBase::GetCommonKernelsData(const Params& params) const
                      1,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     orgParams.outputs[0].is_dynamic());
+                     orgParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/space_to_batch/space_to_batch_kernel_base.cpp

@@ -114,7 +114,7 @@ KernelsData SpaceToBatchKernelBase::GetCommonKernelsData(const Params& params) c
 
     FillCLKernelData(kernel, dispatchData, params.engineInfo, kernelName, jit, entry_point,
                      "", false, false, static_cast<int>(newParams.inputs.size()),
-                     GetFusedPrimitiveInputsCount(params), 1, newParams.has_dynamic_tensors());
+                     GetFusedPrimitiveInputsCount(params), 1, newParams.is_shape_agnostic);
 
     return { kd };
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/strided_slice/strided_slice_kernel_ref.cpp

@@ -252,7 +252,7 @@ KernelsData StridedSliceKernelRef::GetKernelsData(const Params& params) const {
 
     FillCLKernelData(kernel, dispatchData, params.engineInfo, kernelName, jit, entry_point,
                      "", false, false, static_cast<int>(newParams.inputs.size()),
-                     0, 1, newParams.has_dynamic_tensors());
+                     0, 1, newParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/swiglu/swiglu_kernel_ref.cpp

@@ -86,7 +86,7 @@ KernelsData SwiGLUKernelRef::GetKernelsData(const Params& params) const {
                      1,
                      GetFusedPrimitiveInputsCount(params),
                      1,
-                     orgParams.has_dynamic_tensors());
+                     orgParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/tile/tile_kernel_ref.cpp

@@ -77,7 +77,7 @@ KernelsData TileKernelRef::GetKernelsData(const Params& params) const {
     auto& kernel = kd.kernels[0];
 
     FillCLKernelData(kernel, dispatchData, params.engineInfo, kernelName, jit, entry_point,
-                     EXE_MODE_DEFAULT, false, false, 1, 0, 1, newParams.has_dynamic_tensors());
+                     EXE_MODE_DEFAULT, false, false, 1, 0, 1, newParams.is_shape_agnostic);
 
     return {kd};
 }

src/plugins/intel_gpu/src/kernel_selector/kernels/unique/unique_kernel_ref.cpp

@@ -165,7 +165,7 @@ KernelsData UniqueCountKernelRef::GetKernelsData(const Params& params) const {
                      static_cast<int>(kernel_params.inputs.size()),
                      GetFusedPrimitiveInputsCount(kernel_params),
                      static_cast<int>(kernel_params.outputs.size()),
-                     kernel_params.inputs.front().is_dynamic());
+                     kernel_params.is_shape_agnostic);
 
     // Additional buffer to save intermediate algorithm results
     kernel.params.arguments.push_back({ArgumentDescriptor::Types::INTERNAL_BUFFER, 0});
@@ -276,7 +276,7 @@ KernelsData UniqueGatherKernelRef::GetKernelsData(const Params& params) const {
                      static_cast<int>(kernel_params.inputs.size()),
                      GetFusedPrimitiveInputsCount(kernel_params),
                      static_cast<int>(kernel_params.outputs.size()),
-                     kernel_params.outputs.front().is_dynamic());
+                     kernel_params.is_shape_agnostic);
 
     return {kernel_data};
 }