From 47a396e3aae96f2dcad44f4e0d6cb6b87b368395 Mon Sep 17 00:00:00 2001 From: Gunes Bayir Date: Thu, 17 Aug 2023 11:04:02 +0100 Subject: Implement load/store API functions Add KernelWriter API functions for loading and storing tiles with and without dilations. Resolves: COMPMID-5791, COMPMID-6389 Change-Id: I9b1f5b2f081fa54e7bda488aac69ed8d43d1d35c Signed-off-by: Gunes Bayir Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10152 Tested-by: Arm Jenkins Reviewed-by: Jakub Sujak Reviewed-by: Viet-Hoa Do Comments-Addressed: Arm Jenkins Benchmark: Arm Jenkins --- .../tests/CLKernelWriterOpLoadStoreTest.h | 313 +++++++++++++++++++++ 1 file changed, 313 insertions(+) create mode 100644 compute_kernel_writer/validation/tests/CLKernelWriterOpLoadStoreTest.h (limited to 'compute_kernel_writer/validation/tests') diff --git a/compute_kernel_writer/validation/tests/CLKernelWriterOpLoadStoreTest.h b/compute_kernel_writer/validation/tests/CLKernelWriterOpLoadStoreTest.h new file mode 100644 index 0000000000..0f4afc8bf3 --- /dev/null +++ b/compute_kernel_writer/validation/tests/CLKernelWriterOpLoadStoreTest.h @@ -0,0 +1,313 @@ +/* + * Copyright (c) 2023 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. + */ + +#ifndef CKW_VALIDATION_TESTS_CLKERNELWRITEROPLOADSTORETEST_H +#define CKW_VALIDATION_TESTS_CLKERNELWRITEROPLOADSTORETEST_H + +#include "ckw/TileInfo.h" +#include "ckw/types/DataType.h" +#include "src/cl/CLKernelWriter.h" +#include "validation/tests/common/KernelWriterInterceptor.h" +#include "validation/tests/common/Common.h" + +#include "ckw/TensorSampler.h" +#include "ckw/types/MemoryOperation.h" +#include "ckw/types/TensorSamplerTypes.h" + +#include + +namespace ckw +{ + +class CLKernelWriterOpLoadStoreTest : public ITest +{ +private: + using AddressModeX = TensorSamplerAddressModeX; + using AddressModeY = TensorSamplerAddressModeY; + using AddressModeZ = TensorSamplerAddressModeZ; + using Format = TensorSamplerFormat; + using Storage = TensorStorageType; + + struct Coordinates + { + Coordinates(std::string x, std::string y, std::string z, std::string batch) + : x(x), y(y), z(z), batch(batch) + { + } + + std::string x; + std::string y; + std::string z; + std::string batch; + }; + + struct SamplerData + { + SamplerData(Format format, AddressModeX mode_x, AddressModeY mode_y, AddressModeZ mode_z) + : format(format), mode_x(mode_x), mode_y(mode_y), mode_z(mode_z) + { + } + + Format format; + AddressModeX mode_x; + AddressModeY mode_y; + AddressModeZ mode_z; + }; + + struct Dilations + { + Dilations(std::string dilation_x, std::string dilation_y) + : dilation_x(dilation_x), dilation_y(dilation_y) + { + } + + std::string dilation_x; + std::string dilation_y; + }; + + using CLKernelWriterOpLoadStoreConfig = std::tuple; + +public: + CLKernelWriterOpLoadStoreTest() + { + // Cases + const std::string load_fp_2x3_tile = R"_( +tile_0 = vload3(0, (__global float*)(G0__tensor_ptr + (x) * sizeof(float) + (y + 0) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (b) * G0__tensor_stride3)); +tile_1 = vload3(0, (__global float*)(G0__tensor_ptr + (x) * sizeof(float) + (y + 1) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (b) * G0__tensor_stride3)); +)_"; + const std::string load_half_2x4_tile_image_clamp_y = R"_( +tile_0 = read_imageh(G0__tensor_img2d, CLK_NORMALIZED_COORDS_FALSE | CLK_ADDRESS_CLAMP | CLK_FILTER_NEAREST, (int2)((x) >> 2, (y + 0 + (z) * G0__tensor_dim1 + (b) * G0__tensor_dim1 * G0__tensor_dim2))); +tile_1 = read_imageh(G0__tensor_img2d, CLK_NORMALIZED_COORDS_FALSE | CLK_ADDRESS_CLAMP | CLK_FILTER_NEAREST, (int2)((x) >> 2, (y + 1 + (z) * G0__tensor_dim1 + (b) * G0__tensor_dim1 * G0__tensor_dim2))); +)_"; + const std::string store_fp_2x3_tile = R"_( +vstore3(tile_0, 0, (__global float*)(G0__tensor_ptr + (x) * sizeof(float) + (y + 0) * G0__tensor_stride1 + (b) * G0__tensor_stride3)); +vstore3(tile_1, 0, (__global float*)(G0__tensor_ptr + (x) * sizeof(float) + (y + 1) * G0__tensor_stride1 + (b) * G0__tensor_stride3)); +)_"; + const std::string store_int8_4x4_y_dilation_batch_eq_0 = R"_( +vstore4(tile_0, 0, (__global char*)(G0__tensor_ptr + (1) * sizeof(char) + (y + 0 * y_dilation) * G0__tensor_stride1 + (z) * G0__tensor_stride2)); +vstore4(tile_1, 0, (__global char*)(G0__tensor_ptr + (1) * sizeof(char) + (y + 1 * y_dilation) * G0__tensor_stride1 + (z) * G0__tensor_stride2)); +vstore4(tile_2, 0, (__global char*)(G0__tensor_ptr + (1) * sizeof(char) + (y + 2 * y_dilation) * G0__tensor_stride1 + (z) * G0__tensor_stride2)); +vstore4(tile_3, 0, (__global char*)(G0__tensor_ptr + (1) * sizeof(char) + (y + 3 * y_dilation) * G0__tensor_stride1 + (z) * G0__tensor_stride2)); +)_"; + // tensor dimension is 10 + const std::string load_fp_2x3_tile_x_overlapping_min_y_eq_0_batch_eq_1 = R"_( +if(x > 0) +{ +tile_0 = vload3(0, (__global float*)(G0__tensor_ptr + (x) * sizeof(float) + (0 + 0) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (1) * G0__tensor_stride3)); +tile_1 = vload3(0, (__global float*)(G0__tensor_ptr + (x) * sizeof(float) + (0 + 1) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (1) * G0__tensor_stride3)); +} +else +{ +tile_0.s0 = *((__global float*)(G0__tensor_ptr + (x + 0) * sizeof(float) + (0 + 0) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (1) * G0__tensor_stride3)); +tile_1.s0 = *((__global float*)(G0__tensor_ptr + (x + 0) * sizeof(float) + (0 + 1) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (1) * G0__tensor_stride3)); +} +)_"; + const std::string store_fp_2x3_tile_x_overlapping_min_y_clamp_to_border_max_only = R"_( +if(x > 0) +{ +if(y + 0 < G0__tensor_dim1) +{ +vstore3(tile_0, 0, (__global float*)(G0__tensor_ptr + (x) * sizeof(float) + (y + 0) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (b) * G0__tensor_stride3)); +} +else +{ +tile_0 = 0.0f; +} +if(y + 1 < G0__tensor_dim1) +{ +vstore3(tile_1, 0, (__global float*)(G0__tensor_ptr + (x) * sizeof(float) + (y + 1) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (b) * G0__tensor_stride3)); +} +else +{ +tile_1 = 0.0f; +} +} +else +{ +if(y + 0 < G0__tensor_dim1) +{ +*((__global float*)(G0__tensor_ptr + (x + 0) * sizeof(float) + (y + 0) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (b) * G0__tensor_stride3)) = tile_0.s0; +} +else +{ +tile_0.s0 = 0.0f; +} +if(y + 1 < G0__tensor_dim1) +{ +*((__global float*)(G0__tensor_ptr + (x + 0) * sizeof(float) + (y + 1) * G0__tensor_stride1 + (z) * G0__tensor_stride2 + (b) * G0__tensor_stride3)) = tile_1.s0; +} +else +{ +tile_1.s0 = 0.0f; +} +} +)_"; + const std::string store_half_2x4_tile_x_image_y_dilation = R"_( +write_imageh(G0__tensor_img2d, (int2)((x) >> 2, (0 + 0 * y_dilation + (z) * G0__tensor_dim1 + (1) * G0__tensor_dim1 * G0__tensor_dim2)), tile_0); +write_imageh(G0__tensor_img2d, (int2)((x) >> 2, (0 + 1 * y_dilation + (z) * G0__tensor_dim1 + (1) * G0__tensor_dim1 * G0__tensor_dim2)), tile_1); +)_"; + + // Configs Bundled + _configs = { + // op, tile, storage, sampler, coordinates, dilation, expected + { + MemoryOperation::Load, + TileInfo(DataType::Fp32, 2, 3), + TensorStorageType::BufferUint8Ptr, + SamplerData(Format::Dim0_Dim1_Dim2, AddressModeX::None, AddressModeY::None, AddressModeZ::None), + Coordinates("x", "y", "z", "b"), + Dilations("1", "1"), + load_fp_2x3_tile + }, + { + MemoryOperation::Load, + TileInfo(DataType::Fp16, 2, 4), + TensorStorageType::Texture2dReadOnly, + SamplerData(Format::Dim0_Dim1_Dim2, AddressModeX::None, AddressModeY::ClampToBorderMaxOnly, AddressModeZ::None), + Coordinates("x", "y", "z", "b"), + Dilations("1", "1"), + load_half_2x4_tile_image_clamp_y + }, + { + MemoryOperation::Store, + TileInfo(DataType::Fp32, 2, 3), + TensorStorageType::BufferUint8Ptr, + SamplerData(Format::Dim0_Dim1xDim2_1,AddressModeX::None, AddressModeY::None, AddressModeZ::None), + Coordinates("x", "y", "z", "b"), + Dilations("1", "1"), + store_fp_2x3_tile + }, + { + MemoryOperation::Store, + TileInfo(DataType::Int8, 4, 4), + TensorStorageType::BufferUint8Ptr, + SamplerData(Format::Dim0_Dim1_Dim2, AddressModeX::None, AddressModeY::None, AddressModeZ::None), + Coordinates("1", "y", "z", "0"), + Dilations("1", "y_dilation"), + store_int8_4x4_y_dilation_batch_eq_0 + }, + { + MemoryOperation::Load, + TileInfo(DataType::Fp32, 2, 3), + TensorStorageType::BufferUint8Ptr, + SamplerData(Format::Dim0_Dim1_Dim2, AddressModeX::OverlappingMin, AddressModeY::None, AddressModeZ::None), + Coordinates("x", "0", "z", "1"), + Dilations("1", "1"), + load_fp_2x3_tile_x_overlapping_min_y_eq_0_batch_eq_1 + }, + { + MemoryOperation::Store, + TileInfo(DataType::Fp32, 2, 3), + TensorStorageType::BufferUint8Ptr, + SamplerData(Format::Dim0_Dim1_Dim2, AddressModeX::OverlappingMin, AddressModeY::ClampToBorderMaxOnly, AddressModeZ::None), + Coordinates("x", "y", "z", "b"), + Dilations("1", "1"), + store_fp_2x3_tile_x_overlapping_min_y_clamp_to_border_max_only + }, + { + MemoryOperation::Store, + TileInfo(DataType::Fp16, 2, 4), + TensorStorageType::Texture2dWriteOnly, + SamplerData(Format::Dim0_Dim1_Dim2, AddressModeX::None, AddressModeY::None, AddressModeZ::None), + Coordinates("x", "0", "z", "1"), + Dilations("1", "y_dilation"), + store_half_2x4_tile_x_image_y_dilation + } + }; + } + + bool run() override + { + bool all_tests_passed = true; + int32_t test_idx = 0; + + for(auto _config: _configs) + { + KernelWriterInterceptor writer; + + const MemoryOperation op = std::get<0>(_config); + const TileInfo tile_info = std::get<1>(_config); + const Storage storage = std::get<2>(_config); + const SamplerData sampler_data = std::get<3>(_config); + const Coordinates coord = std::get<4>(_config); + const Dilations dilations = std::get<5>(_config); + const std::string expected_code = std::get<6>(_config).substr(1); // ignore initial newline, which was added for convenience + + TileOperand tile_op = writer.declare_tile("tile", tile_info); + TileOperand x_op = writer.declare_tile(coord.x, TileInfo(DataType::Int32)); + TileOperand y_op = writer.declare_tile(coord.y, TileInfo(DataType::Int32)); + TileOperand z_op = writer.declare_tile(coord.z, TileInfo(DataType::Int32)); + TileOperand batch_op = writer.declare_tile(coord.batch, TileInfo(DataType::Int32)); + TileOperand dil_x_op = writer.declare_tile(dilations.dilation_x, TileInfo(DataType::Int32)); + TileOperand dil_y_op = writer.declare_tile(dilations.dilation_y, TileInfo(DataType::Int32)); + + TensorShape tensor_shape {10, 10, 10, 10}; + TensorInfo tensor_info(tile_info.data_type(), tensor_shape, TensorDataLayout::Nhwc, 0 /* id */); + TensorOperand tensor_op = writer.declare_tensor_argument("tensor", tensor_info); + TensorSampler sampler(storage, sampler_data.format, sampler_data.mode_x, sampler_data.mode_y, sampler_data.mode_z); + + const bool no_dilation = (dilations.dilation_x == "1" && dilations.dilation_y == "1"); + + writer.start_capture_code(); + if(op == MemoryOperation::Load) + { + if(no_dilation) + { + writer.op_load(tile_op, tensor_op, sampler, x_op, y_op, z_op, batch_op); + } + else + { + writer.op_load_dilated(tile_op, tensor_op, sampler, x_op, y_op, z_op, batch_op, dil_x_op, dil_y_op); + } + } + else + { + if(no_dilation) + { + writer.op_store(tensor_op, tile_op, sampler, x_op, y_op, z_op, batch_op); + } + else + { + writer.op_store_dilated(tensor_op, tile_op, sampler, x_op, y_op, z_op, batch_op, dil_x_op, dil_y_op); + } + } + + VALIDATE_TEST(writer.check_added_code(expected_code), all_tests_passed, test_idx++); + } + + return all_tests_passed; + } + + std::string name() override + { + return "CLKernelWriterOpLoadStoreTest"; + } + +private: + std::vector _configs {}; +}; + +} // namespace ckw + +#endif // CKW_VALIDATION_TESTS_CLKERNELWRITEROPLOADSTORETEST_H -- cgit v1.2.1