/* * Copyright (c) 2023-2024 Arm Limited. * * SPDX-License-Identifier: MIT * * Permission is hereby granted, free of charge, to any person obtaining a copy * of this software and associated documentation files (the "Software"), to * deal in the Software without restriction, including without limitation the * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or * sell copies of the Software, and to permit persons to whom the Software is * furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice shall be included in all * copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ #include "src/dynamic_fusion/sketch/gpu/ckw_driver/components/GpuCkwDepthwiseConv2d.h" #include "arm_compute/core/Error.h" #include "arm_compute/core/utils/helpers/AdjustVecSize.h" #include "arm_compute/core/Validate.h" #include "src/core/helpers/WindowHelpers.h" #include "src/dynamic_fusion/sketch/gpu/ckw_driver/components/utils/CkwHelper.h" #include "src/dynamic_fusion/sketch/gpu/ckw_driver/components/utils/type_converter/Common.h" #include "src/dynamic_fusion/sketch/gpu/ckw_driver/GpuCkwScopedKernelWriter.h" #include "src/dynamic_fusion/sketch/gpu/ckw_driver/GpuCkwVariableTable.h" #include "src/dynamic_fusion/sketch/gpu/GpuKernelArgument.h" #include "src/dynamic_fusion/sketch/gpu/GpuKernelComponentGroup.h" #include "compute_kernel_writer/include/ckw/KernelWriter.h" #include #include namespace arm_compute { namespace experimental { namespace dynamic_fusion { GpuCkwDepthwiseConv2d::GpuCkwDepthwiseConv2d(ComponentId id, const ArgumentPack &tensors, const Attributes &attributes, const Settings &settings) : IGpuCkwComponentDriver{id, tensors}, _src{}, _wei{}, _bia{}, _dst{}, _attributes{attributes}, _settings{settings} { _src = this->tensors().get_const_tensor(TensorType::ACL_SRC_0); _wei = this->tensors().get_const_tensor(TensorType::ACL_SRC_1); if (this->tensors().get_const_tensor(TensorType::ACL_SRC_2)) { _bia = this->tensors().get_const_tensor(TensorType::ACL_SRC_2); } _dst = this->tensors().get_const_tensor(TensorType::ACL_DST_0); ARM_COMPUTE_ERROR_ON_NULLPTR(_src, _wei, _bia, _dst); } void GpuCkwDepthwiseConv2d::write_component_code(const ComponentGroup &comp_group, GpuCkwVariableTable &vtable, GpuCkwScopedKernelWriter writer) const { // Data Layout is NHWC const uint32_t width_idx = get_data_layout_dimension_index(_wei->data_layout(), DataLayoutDimension::WIDTH); const uint32_t height_idx = get_data_layout_dimension_index(_wei->data_layout(), DataLayoutDimension::HEIGHT); /******************************************************************************** * 1 - Define tensors ********************************************************************************/ GpuCkwComponentArgument *src = vtable.declare_variable(comp_group, writer, _src, "src"); GpuCkwComponentArgument *wei = vtable.declare_variable(comp_group, writer, _wei, "wei"); GpuCkwComponentArgument *dst = vtable.declare_variable(comp_group, writer, _dst, "dst"); GpuCkwComponentArgument *bia = nullptr; const bool using_bias = _bia != nullptr; if (using_bias) { bia = vtable.declare_variable(comp_group, writer, _bia, "bia"); } /******************************************************************************** * 2 - Define CKW constants ********************************************************************************/ const auto dst_dt = to_ckw(_dst->data_type()); const auto kernel_height = static_cast(_wei->dimension(height_idx)); const auto kernel_width = static_cast(_wei->dimension(width_idx)); const auto src_w = static_cast(_src->dimension(width_idx)); const auto src_h = static_cast(_src->dimension(height_idx)); const auto dst_h = static_cast(_dst->dimension(height_idx)); const auto stride_x = static_cast(_attributes.stride().x()); const auto stride_y = static_cast(_attributes.stride().y()); const auto pad_x = static_cast(_attributes.pad().left); const auto pad_y = static_cast(_attributes.pad().top); const auto depth_multiplier = static_cast(_attributes.depth_multiplier()); const auto dilation_x = static_cast(_attributes.dilation().x()); const auto dilation_y = static_cast(_attributes.dilation().y()); const auto kernel_size = kernel_width * kernel_height; // CKW constants auto const_kernel_w_i32 = writer->declare_constant_tile(ckw::ConstantData({{kernel_width}}, ckw::DataType::Int32)); auto const_kernel_size_i32 = writer->declare_constant_tile(ckw::ConstantData({{kernel_size}}, ckw::DataType::Int32)); auto const_dst_h_i32 = writer->declare_constant_tile(ckw::ConstantData({{dst_h}}, ckw::DataType::Int32)); auto const_src_w_i32 = writer->declare_constant_tile(ckw::ConstantData({{src_w}}, ckw::DataType::Int32)); auto const_src_h_i32 = writer->declare_constant_tile(ckw::ConstantData({{src_h}}, ckw::DataType::Int32)); auto const_stride_x_i32 = writer->declare_constant_tile(ckw::ConstantData({{stride_x}}, ckw::DataType::Int32)); auto const_stride_y_i32 = writer->declare_constant_tile(ckw::ConstantData({{stride_y}}, ckw::DataType::Int32)); auto const_pad_x_i32 = writer->declare_constant_tile(ckw::ConstantData({{pad_x}}, ckw::DataType::Int32)); auto const_pad_y_i32 = writer->declare_constant_tile(ckw::ConstantData({{pad_y}}, ckw::DataType::Int32)); auto const_0_i32 = writer->declare_constant_tile(ckw::ConstantData({{0}}, ckw::DataType::Int32)); auto const_neg_1_i32 = writer->declare_constant_tile(ckw::ConstantData({{-1}}, ckw::DataType::Int32)); auto const_depth_multiplier_i32 = writer->declare_constant_tile(ckw::ConstantData({{depth_multiplier}}, ckw::DataType::Int32)); auto const_dilation_x_i32 = writer->declare_constant_tile(ckw::ConstantData({{dilation_x}}, ckw::DataType::Int32)); auto const_dilation_y_i32 = writer->declare_constant_tile(ckw::ConstantData({{dilation_y}}, ckw::DataType::Int32)); auto const_0_fp = writer->declare_constant_tile(ckw::ConstantData({{0.0f}}, dst_dt)); /******************************************************************************** * 3 - Define the compute block parameters and destination tile (if not root component) * Bind the tile to the tensor to share it among different components and * initialize the compute block parameters ********************************************************************************/ // The compute block parameters depend on the employed tensor format const auto root_window = comp_group.get_root_component()->ckw_component_driver()->get_window(); // Destination compute block size const int32_t dst_n0 = root_window.x().step(); const int32_t dst_m0 = root_window.y().step(); // Destination compute block size left-over const int32_t dst_n0_partial = _dst->dimension(0) % dst_n0; const int32_t dst_m0_partial = _dst->dimension(1) % dst_m0; // Shift-back for the overlapping-min strategy const int32_t dst_shift_back = (dst_n0 - dst_n0_partial) % dst_n0; const int32_t src_m0 = kernel_width + (dst_m0 - 1); const int32_t src_n0 = depth_multiplier > 1 ? 1 : dst_n0; const int32_t wei_m0 = kernel_width; const int32_t wei_n0 = dst_n0; ckw::TensorSampler sampler_dst; sampler_dst.format(ckw::TensorSamplerFormat::Dim0_Dim1_Dim2); if (dst_n0_partial == 0) { sampler_dst.address_mode_x(ckw::TensorSamplerAddressModeX::None); } else { sampler_dst.address_mode_x(ckw::TensorSamplerAddressModeX::OverlappingMin); } if (dst_m0_partial == 0) { sampler_dst.address_mode_y(ckw::TensorSamplerAddressModeY::None); } else { sampler_dst.address_mode_y(ckw::TensorSamplerAddressModeY::ClampToBorderMaxOnly); } sampler_dst.address_mode_z(ckw::TensorSamplerAddressModeZ::None); sampler_dst.storage(ckw::TensorStorageType::BufferUint8Ptr); // Declare destination tile auto tile_dst = writer->declare_tile("dst", ckw::TileInfo(dst_dt, dst_m0, dst_n0)); // Initialize the destination tile writer->op_assign(tile_dst, const_0_fp); // Bind tile to the tensor dst->init_virtual_tensor(tile_dst, sampler_dst); /******************************************************************************** * 4 - Define the compute block parameters CKW constants ********************************************************************************/ // Only now we can declare the N0 and M0 as constant auto const_dst_n0_i32 = writer->declare_constant_tile(ckw::ConstantData({{dst_n0}}, ckw::DataType::Int32)); auto const_dst_m0_i32 = writer->declare_constant_tile(ckw::ConstantData({{dst_m0}}, ckw::DataType::Int32)); auto const_shift_back_dst_n0_i32 = writer->declare_constant_tile(ckw::ConstantData({{dst_shift_back}}, ckw::DataType::Int32)); /******************************************************************************** * 5 - Define the sampler for the input tensors ********************************************************************************/ // SOURCE SAMPLER ckw::TensorSampler sampler_src; sampler_src.format(ckw::TensorSamplerFormat::Dim0_Dim1_Dim2); sampler_src.address_mode_x(ckw::TensorSamplerAddressModeX::None); sampler_src.address_mode_y(ckw::TensorSamplerAddressModeY::SkipLessThanZero); sampler_src.address_mode_z(ckw::TensorSamplerAddressModeZ::None); sampler_src.storage(ckw::TensorStorageType::BufferUint8Ptr); // WEIGHTS SAMPLER // We cannot have out-of-bounds accesses for the weights ckw::TensorSampler sampler_wei; sampler_wei.format(ckw::TensorSamplerFormat::Dim0_Dim1_Dim2); sampler_wei.address_mode_x(ckw::TensorSamplerAddressModeX::None); sampler_wei.address_mode_y(ckw::TensorSamplerAddressModeY::None); sampler_wei.address_mode_z(ckw::TensorSamplerAddressModeZ::None); if (_settings.export_weights_to_cl_image()) { sampler_wei.storage(ckw::TensorStorageType::Texture2dReadOnly); } else { sampler_wei.storage(ckw::TensorStorageType::BufferUint8Ptr); } // BIAS SAMPLER ckw::TensorSampler sampler_bia; sampler_bia.format(ckw::TensorSamplerFormat::Dim0_Dim1_Dim2); sampler_bia.address_mode_x(sampler_dst.address_mode_x()); sampler_bia.address_mode_y(ckw::TensorSamplerAddressModeY::None); sampler_bia.address_mode_z(ckw::TensorSamplerAddressModeZ::None); sampler_bia.storage(ckw::TensorStorageType::BufferUint8Ptr); /******************************************************************************** * 6 - Extra operations required before writing the main code (Optional) ********************************************************************************/ // Not required /******************************************************************************** * 7 - Get the coordinates of the destination tile ********************************************************************************/ auto tile_gid_0 = writer->declare_tile("gid_0", ckw::TileInfo(ckw::DataType::Int32)); auto tile_gid_1 = writer->declare_tile("gid_1", ckw::TileInfo(ckw::DataType::Int32)); auto tile_gid_2 = writer->declare_tile("gid_2", ckw::TileInfo(ckw::DataType::Int32)); writer->op_get_global_id(tile_gid_0, 0); writer->op_get_global_id(tile_gid_1, 1); writer->op_get_global_id(tile_gid_2, 2); auto tile_cout0 = writer->declare_tile("cout0", ckw::TileInfo(ckw::DataType::Int32)); // OFM auto tile_mout0 = writer->declare_tile("mout0", ckw::TileInfo(ckw::DataType::Int32)); // WIDTH auto tile_mout1 = writer->declare_tile("mout1", ckw::TileInfo(ckw::DataType::Int32)); // HEIGHT auto tile_bout0 = writer->declare_tile("bout0", ckw::TileInfo(ckw::DataType::Int32)); // BATCH SIZE IDX // Calculate coordinates get_coordinate_from_gws_overlapping_min(writer, tile_cout0, tile_gid_0, const_dst_n0_i32, const_shift_back_dst_n0_i32, const_0_i32); get_coordinate_from_gws(writer, tile_mout0, tile_gid_1, const_dst_m0_i32); writer->op_binary(tile_mout1, ckw::BinaryOp::Mod, tile_gid_2, const_dst_h_i32); writer->op_binary(tile_bout0, ckw::BinaryOp::Div, tile_gid_2, const_dst_h_i32); auto tile_src_ci = writer->declare_tile("src_ci", ckw::DataType::Int32); writer->op_binary(tile_src_ci, ckw::BinaryOp::Div, tile_cout0, const_depth_multiplier_i32); auto tile_src_xi = writer->declare_tile("src_xi", ckw::DataType::Int32); writer->op_binary(tile_src_xi, ckw::BinaryOp::Mul, tile_mout0, const_stride_x_i32); writer->op_binary(tile_src_xi, ckw::BinaryOp::Sub, tile_src_xi, const_pad_x_i32); auto tile_src_yi = writer->declare_tile("src_yi", ckw::DataType::Int32); writer->op_binary(tile_src_yi, ckw::BinaryOp::Mul, tile_mout1, const_stride_y_i32); writer->op_binary(tile_src_yi, ckw::BinaryOp::Sub, tile_src_yi, const_pad_y_i32); // Loop variables auto tile_yk = writer->declare_tile("yk", ckw::DataType::Int32); writer->op_assign(tile_yk, const_0_i32); // clang-format off writer->op_for_loop(tile_yk, ckw::BinaryOp::Less, const_kernel_size_i32, tile_yk, ckw::AssignmentOp::Increment, const_kernel_w_i32, [&]() { auto tile_src = writer->declare_tile("a", ckw::TileInfo(to_ckw(_src->data_type()), src_m0, src_n0)); auto tile_wei = writer->declare_tile("b", ckw::TileInfo(to_ckw(_wei->data_type()), wei_m0, wei_n0)); writer->op_assign(tile_src, const_0_fp); auto tile_x_gte_0 = writer->declare_tile("x_gte_0", ckw::TileInfo(ckw::DataType::Int32)); auto tile_y_gte_0 = writer->declare_tile("y_gte_0", ckw::TileInfo(ckw::DataType::Int32)); auto tile_x_lt_w = writer->declare_tile("x_lt_w", ckw::TileInfo(ckw::DataType::Int32)); auto tile_y_lt_h = writer->declare_tile("y_lt_h", ckw::TileInfo(ckw::DataType::Int32)); // Check if yi + yk * DILATION_Y is out-of-bound writer->op_binary(tile_y_gte_0, ckw::BinaryOp::GreaterEqual, tile_src_yi, const_0_i32); writer->op_binary(tile_y_lt_h, ckw::BinaryOp::Less, tile_src_yi, const_src_h_i32); auto tile_src_mi = writer->declare_tile("src_mi", ckw::TileInfo(ckw::DataType::Int32)); // Load src for(int32_t xk = 0; xk < src_m0; ++xk) { auto const_xk_i32 = writer->declare_constant_tile(ckw::ConstantData({{xk}}, ckw::DataType::Int32)); // xi + xk * DILATION_X writer->op_binary(tile_src_mi, ckw::BinaryOp::Mul, const_xk_i32, const_dilation_x_i32); writer->op_binary(tile_src_mi, ckw::BinaryOp::Add, tile_src_mi, tile_src_xi); // Check if xi + xk * DILATION_X is out-of-bound writer->op_binary(tile_x_gte_0, ckw::BinaryOp::GreaterEqual, tile_src_mi, const_0_i32); writer->op_binary(tile_x_lt_w, ckw::BinaryOp::Less, tile_src_mi, const_src_w_i32); // Set mi to -1 if we have out-of-bound memory accesses writer->op_ternary(tile_src_mi, ckw::TernaryOp::Select, const_neg_1_i32, tile_src_mi, tile_x_gte_0); writer->op_ternary(tile_src_mi, ckw::TernaryOp::Select, const_neg_1_i32, tile_src_mi, tile_x_lt_w); writer->op_ternary(tile_src_mi, ckw::TernaryOp::Select, const_neg_1_i32, tile_src_mi, tile_y_gte_0); writer->op_ternary(tile_src_mi, ckw::TernaryOp::Select, const_neg_1_i32, tile_src_mi, tile_y_lt_h); writer->op_load(tile_src.row(xk), src->tensor(), sampler_src, tile_src_ci, tile_src_mi, tile_src_yi, tile_bout0); } // Load wei writer->op_load(tile_wei, wei->tensor(), sampler_wei, tile_cout0, tile_yk, const_0_i32, const_0_i32); // Attention: MAC (Multiply-and-Accumulate) ternary operator is currently unsupported in CKW // Therefore, this part should be replaced with the MAC ternary operator when availabe auto tile_tmp = writer->declare_tile("tmp", ckw::TileInfo(to_ckw(_src->data_type()), 1, dst_n0)); for(int32_t m0 = 0; m0 < dst_m0; ++m0) { for(int32_t xk = 0; xk < kernel_width; ++xk) { auto tile_a = tile_src.row(m0 + xk); auto tile_b = tile_wei.row(xk); auto tile_c = tile_dst.row(m0); writer->op_binary(tile_tmp, ckw::BinaryOp::Mul, tile_a, tile_b); writer->op_binary(tile_c, ckw::BinaryOp::Add, tile_c, tile_tmp); } } writer->op_binary(tile_src_yi, ckw::BinaryOp::Add, tile_src_yi, const_dilation_y_i32); }); // clang-format on // Bias addition // NOTE: This operation will be removed from this kernel as the interface is standardized. The intended way of // performing bias addition is to fuse this convolution kernel with a following elementwise addition kernel. if (using_bias) { if (!bia->has_tile()) { auto tile_bia = writer->declare_tile("bia", ckw::TileInfo(to_ckw(_src->data_type()), 1, dst_n0)); writer->op_load(tile_bia, bia->tensor(), sampler_bia, tile_cout0, const_0_i32, const_0_i32, const_0_i32); bia->init_virtual_tensor(tile_bia, sampler_bia); } auto &tile_bia = bia->tile(); writer->op_binary(tile_dst, ckw::BinaryOp::Add, tile_dst, tile_bia); } ARM_COMPUTE_ERROR_ON_MSG(dst->has_tile() == false, "You must bind a tile before appending another component"); } Window GpuCkwDepthwiseConv2d::get_window() const { ARM_COMPUTE_ERROR_ON_MSG(_dst->tensor_shape().total_size() == 0U, "Destination tensor is not initialized"); TensorShape output_shape = _dst->tensor_shape(); Window win = calculate_max_window(output_shape, Steps(_settings.n0(), _settings.m0())); return win.collapse(win, Window::DimZ); } } // namespace dynamic_fusion } // namespace experimental } // namespace arm_compute