diff options
Diffstat (limited to 'src/cpu/kernels/CpuTransposeKernel.cpp')
| -rw-r--r-- | src/cpu/kernels/CpuTransposeKernel.cpp | 819 |
1 files changed, 819 insertions, 0 deletions
diff --git a/src/cpu/kernels/CpuTransposeKernel.cpp b/src/cpu/kernels/CpuTransposeKernel.cpp new file mode 100644 index 0000000000..0f762ba041 --- /dev/null +++ b/src/cpu/kernels/CpuTransposeKernel.cpp @@ -0,0 +1,819 @@ +/* + * Copyright (c) 2021, 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. + */ +#include "src/cpu/kernels/CpuTransposeKernel.h" + +#include "arm_compute/core/Error.h" +#include "arm_compute/core/Helpers.h" +#include "arm_compute/core/ITensor.h" +#include "arm_compute/core/TensorInfo.h" +#include "arm_compute/core/Types.h" +#include "arm_compute/core/utils/misc/ShapeCalculator.h" +#include "arm_compute/core/Validate.h" + +#include "src/core/helpers/AutoConfiguration.h" +#include "src/core/helpers/WindowHelpers.h" + +#include <arm_neon.h> + +namespace arm_compute +{ +namespace cpu +{ +namespace kernels +{ +namespace +{ +unsigned int num_elems_processed(size_t element_size) +{ + switch (element_size) + { + case 1: + return 8; + case 2: + return 4; + case 4: +#ifdef __aarch64__ + return 8; +#else // __aarch64__ + return 4; +#endif // __aarch64__ + default: + break; + } + + ARM_COMPUTE_ERROR("Element size not supported"); +} + +void transpose_8bit_elements(const ITensor *in, ITensor *out, const Window &window) +{ + const int window_step_x = 8; + const int window_step_y = 8; + const int window_start_x = window.x().start(); + const int window_end_x = window.x().end(); + const int window_start_y = window.y().start(); + const int window_end_y = std::min(window.y().end(), static_cast<int>(in->info()->dimension(1))); + const int window_end_y_multiple_of = ((window_end_y - window_start_y) / window_step_y) * window_step_y; + const size_t input_stride_in_bytes = in->info()->strides_in_bytes()[1]; + const size_t output_stride_in_bytes = out->info()->strides_in_bytes()[1]; + + // Check if we need a left-over loop for the y dimension + bool left_over_loop_y = (((window_end_y - window_start_y) % window_step_y) != 0); + + Window window_in(window); + window_in.set(Window::DimX, Window::Dimension(0, 1, 1)); + if (left_over_loop_y) + { + // Check if window_end_y_multiple_of is greater than window_start_y + if (window_end_y_multiple_of > window_start_y) + { + window_in.set(Window::DimY, Window::Dimension(window_start_y, window_end_y_multiple_of, window_step_y)); + } + else + { + window_in.set(Window::DimY, Window::Dimension(0, 0, 1)); + } + } + + Window window_out(window); + window_out.set(Window::DimX, Window::Dimension(0, 0, 0)); + window_out.set(Window::DimY, Window::Dimension(0, 0, 0)); + + Iterator output(out, window_out); + + // Run the SIMD path if and only if the input is not a row-vector + if (in->info()->dimension(1) != 1) + { + Iterator input(in, window_in); + execute_window_loop( + window_in, + [&](const Coordinates &id) + { + // Compute 8x8 elements per iteration + int x = window_start_x; + for (; x <= (window_end_x - window_step_x); x += window_step_x) + { + const uint8x8_t row0 = + vld1_u8(reinterpret_cast<const uint8_t *>(input.ptr() + x + 0 * input_stride_in_bytes)); + const uint8x8_t row1 = + vld1_u8(reinterpret_cast<const uint8_t *>(input.ptr() + x + 1 * input_stride_in_bytes)); + const uint8x8_t row2 = + vld1_u8(reinterpret_cast<const uint8_t *>(input.ptr() + x + 2 * input_stride_in_bytes)); + const uint8x8_t row3 = + vld1_u8(reinterpret_cast<const uint8_t *>(input.ptr() + x + 3 * input_stride_in_bytes)); + const uint8x8_t row4 = + vld1_u8(reinterpret_cast<const uint8_t *>(input.ptr() + x + 4 * input_stride_in_bytes)); + const uint8x8_t row5 = + vld1_u8(reinterpret_cast<const uint8_t *>(input.ptr() + x + 5 * input_stride_in_bytes)); + const uint8x8_t row6 = + vld1_u8(reinterpret_cast<const uint8_t *>(input.ptr() + x + 6 * input_stride_in_bytes)); + const uint8x8_t row7 = + vld1_u8(reinterpret_cast<const uint8_t *>(input.ptr() + x + 7 * input_stride_in_bytes)); + + // Transpose 2x2 + const uint8x8x2_t k0_u8 = vtrn_u8(row0, row1); + const uint8x8x2_t k1_u8 = vtrn_u8(row2, row3); + const uint8x8x2_t k2_u8 = vtrn_u8(row4, row5); + const uint8x8x2_t k3_u8 = vtrn_u8(row6, row7); + + // Transpose 4x4 + const uint16x4x2_t k0_u16 = + vtrn_u16(vreinterpret_u16_u8(k0_u8.val[0]), vreinterpret_u16_u8(k1_u8.val[0])); + const uint16x4x2_t k1_u16 = + vtrn_u16(vreinterpret_u16_u8(k0_u8.val[1]), vreinterpret_u16_u8(k1_u8.val[1])); + const uint16x4x2_t k2_u16 = + vtrn_u16(vreinterpret_u16_u8(k2_u8.val[0]), vreinterpret_u16_u8(k3_u8.val[0])); + const uint16x4x2_t k3_u16 = + vtrn_u16(vreinterpret_u16_u8(k2_u8.val[1]), vreinterpret_u16_u8(k3_u8.val[1])); + + // Transpose 8x8 + const uint32x2x2_t k0_u32 = + vtrn_u32(vreinterpret_u32_u16(k0_u16.val[0]), vreinterpret_u32_u16(k2_u16.val[0])); + const uint32x2x2_t k1_u32 = + vtrn_u32(vreinterpret_u32_u16(k0_u16.val[1]), vreinterpret_u32_u16(k2_u16.val[1])); + const uint32x2x2_t k2_u32 = + vtrn_u32(vreinterpret_u32_u16(k1_u16.val[0]), vreinterpret_u32_u16(k3_u16.val[0])); + const uint32x2x2_t k3_u32 = + vtrn_u32(vreinterpret_u32_u16(k1_u16.val[1]), vreinterpret_u32_u16(k3_u16.val[1])); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint8_t) + x * output_stride_in_bytes; + + vst1_u8( + reinterpret_cast<uint8_t *>(output.ptr() + dst_offset_in_bytes + 0 * output_stride_in_bytes), + vreinterpret_u8_u16(vreinterpret_u16_u32(k0_u32.val[0]))); + vst1_u8( + reinterpret_cast<uint8_t *>(output.ptr() + dst_offset_in_bytes + 1 * output_stride_in_bytes), + vreinterpret_u8_u16(vreinterpret_u16_u32(k2_u32.val[0]))); + vst1_u8( + reinterpret_cast<uint8_t *>(output.ptr() + dst_offset_in_bytes + 2 * output_stride_in_bytes), + vreinterpret_u8_u16(vreinterpret_u16_u32(k1_u32.val[0]))); + vst1_u8( + reinterpret_cast<uint8_t *>(output.ptr() + dst_offset_in_bytes + 3 * output_stride_in_bytes), + vreinterpret_u8_u16(vreinterpret_u16_u32(k3_u32.val[0]))); + vst1_u8( + reinterpret_cast<uint8_t *>(output.ptr() + dst_offset_in_bytes + 4 * output_stride_in_bytes), + vreinterpret_u8_u16(vreinterpret_u16_u32(k0_u32.val[1]))); + vst1_u8( + reinterpret_cast<uint8_t *>(output.ptr() + dst_offset_in_bytes + 5 * output_stride_in_bytes), + vreinterpret_u8_u16(vreinterpret_u16_u32(k2_u32.val[1]))); + vst1_u8( + reinterpret_cast<uint8_t *>(output.ptr() + dst_offset_in_bytes + 6 * output_stride_in_bytes), + vreinterpret_u8_u16(vreinterpret_u16_u32(k1_u32.val[1]))); + vst1_u8( + reinterpret_cast<uint8_t *>(output.ptr() + dst_offset_in_bytes + 7 * output_stride_in_bytes), + vreinterpret_u8_u16(vreinterpret_u16_u32(k3_u32.val[1]))); + } + + // Compute left-over elements along the x dimension (1x8) + for (; x < window_end_x; ++x) + { + const uint8_t val0 = *(input.ptr() + x + 0 * input_stride_in_bytes); + const uint8_t val1 = *(input.ptr() + x + 1 * input_stride_in_bytes); + const uint8_t val2 = *(input.ptr() + x + 2 * input_stride_in_bytes); + const uint8_t val3 = *(input.ptr() + x + 3 * input_stride_in_bytes); + const uint8_t val4 = *(input.ptr() + x + 4 * input_stride_in_bytes); + const uint8_t val5 = *(input.ptr() + x + 5 * input_stride_in_bytes); + const uint8_t val6 = *(input.ptr() + x + 6 * input_stride_in_bytes); + const uint8_t val7 = *(input.ptr() + x + 7 * input_stride_in_bytes); + + uint8x8_t result = vdup_n_u8(0); + result = vset_lane_u8(val0, result, 0); + result = vset_lane_u8(val1, result, 1); + result = vset_lane_u8(val2, result, 2); + result = vset_lane_u8(val3, result, 3); + result = vset_lane_u8(val4, result, 4); + result = vset_lane_u8(val5, result, 5); + result = vset_lane_u8(val6, result, 6); + result = vset_lane_u8(val7, result, 7); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint8_t) + x * output_stride_in_bytes; + + vst1_u8(output.ptr() + dst_offset_in_bytes, result); + } + }, + input, output); + } + + if (left_over_loop_y) + { + window_in.set(Window::DimX, Window::Dimension(window.x().start(), window.x().end(), 1)); + window_in.set(Window::DimY, Window::Dimension(window_end_y_multiple_of, window_end_y, 1)); + + Iterator input(in, window_in); + Iterator output(out, window_out); + + // Compute left-over elements along the y dimension (1x1) + execute_window_loop( + window_in, + [&](const Coordinates &id) + { + const uint8_t val0 = *input.ptr(); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint8_t) + id.x() * output_stride_in_bytes; + + *(output.ptr() + dst_offset_in_bytes) = val0; + }, + input, output); + } +} + +void transpose_16bit_elements(const ITensor *in, ITensor *out, const Window &window) +{ + const int window_step_x = 4; + const int window_step_y = 4; + const int window_start_x = window.x().start(); + const int window_end_x = window.x().end(); + const int window_start_y = window.y().start(); + const int window_end_y = std::min(window.y().end(), static_cast<int>(in->info()->dimension(1))); + const int window_end_y_multiple_of = ((window_end_y - window_start_y) / window_step_y) * window_step_y; + const size_t input_stride_in_bytes = in->info()->strides_in_bytes()[1]; + const size_t output_stride_in_bytes = out->info()->strides_in_bytes()[1]; + + // Check if we need a left-over loop for the y dimension + bool left_over_loop_y = (((window_end_y - window_start_y) % window_step_y) != 0); + + Window window_in(window); + window_in.set(Window::DimX, Window::Dimension(0, 1, 1)); + if (left_over_loop_y) + { + // Check if window_end_y_multiple_of is greater than window_start_y + if (window_end_y_multiple_of > window_start_y) + { + window_in.set(Window::DimY, Window::Dimension(window_start_y, window_end_y_multiple_of, window_step_y)); + } + else + { + window_in.set(Window::DimY, Window::Dimension(0, 0, 1)); + } + } + + Window window_out(window); + window_out.set(Window::DimX, Window::Dimension(0, 0, 0)); + window_out.set(Window::DimY, Window::Dimension(0, 0, 0)); + + Iterator output(out, window_out); + + // Run the SIMD path if and only if the input is not a row-vector + if (in->info()->dimension(1) != 1) + { + Iterator input(in, window_in); + execute_window_loop( + window_in, + [&](const Coordinates &id) + { + // Compute 4x4 elements per iteration + int x = window_start_x; + for (; x <= (window_end_x - window_step_x); x += window_step_x) + { + const uint16x4_t row0 = + vld1_u16(reinterpret_cast<const uint16_t *>(input.ptr() + 0 * input_stride_in_bytes) + x); + const uint16x4_t row1 = + vld1_u16(reinterpret_cast<const uint16_t *>(input.ptr() + 1 * input_stride_in_bytes) + x); + const uint16x4_t row2 = + vld1_u16(reinterpret_cast<const uint16_t *>(input.ptr() + 2 * input_stride_in_bytes) + x); + const uint16x4_t row3 = + vld1_u16(reinterpret_cast<const uint16_t *>(input.ptr() + 3 * input_stride_in_bytes) + x); + + // Transpose 2x2 + const uint16x4x2_t k0_u16 = vtrn_u16(row0, row1); + const uint16x4x2_t k1_u16 = vtrn_u16(row2, row3); + + // Transpose 4x4 + const uint32x2x2_t k0_u32 = + vtrn_u32(vreinterpret_u32_u16(k0_u16.val[0]), vreinterpret_u32_u16(k1_u16.val[0])); + const uint32x2x2_t k1_u32 = + vtrn_u32(vreinterpret_u32_u16(k0_u16.val[1]), vreinterpret_u32_u16(k1_u16.val[1])); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint16_t) + x * output_stride_in_bytes; + + vst1_u16( + reinterpret_cast<uint16_t *>(output.ptr() + dst_offset_in_bytes + 0 * output_stride_in_bytes), + vreinterpret_u16_u32(k0_u32.val[0])); + vst1_u16( + reinterpret_cast<uint16_t *>(output.ptr() + dst_offset_in_bytes + 1 * output_stride_in_bytes), + vreinterpret_u16_u32(k1_u32.val[0])); + vst1_u16( + reinterpret_cast<uint16_t *>(output.ptr() + dst_offset_in_bytes + 2 * output_stride_in_bytes), + vreinterpret_u16_u32(k0_u32.val[1])); + vst1_u16( + reinterpret_cast<uint16_t *>(output.ptr() + dst_offset_in_bytes + 3 * output_stride_in_bytes), + vreinterpret_u16_u32(k1_u32.val[1])); + } + + // Compute left-over elements (1x4) + for (; x < window_end_x; ++x) + { + const uint16_t val0 = *(reinterpret_cast<uint16_t *>(input.ptr() + 0 * input_stride_in_bytes) + x); + const uint16_t val1 = *(reinterpret_cast<uint16_t *>(input.ptr() + 1 * input_stride_in_bytes) + x); + const uint16_t val2 = *(reinterpret_cast<uint16_t *>(input.ptr() + 2 * input_stride_in_bytes) + x); + const uint16_t val3 = *(reinterpret_cast<uint16_t *>(input.ptr() + 3 * input_stride_in_bytes) + x); + + uint16x4_t result = vdup_n_u16(0); + result = vset_lane_u16(val0, result, 0); + result = vset_lane_u16(val1, result, 1); + result = vset_lane_u16(val2, result, 2); + result = vset_lane_u16(val3, result, 3); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint16_t) + x * output_stride_in_bytes; + + vst1_u16(reinterpret_cast<uint16_t *>(output.ptr() + dst_offset_in_bytes), result); + } + }, + input, output); + } + + if (left_over_loop_y) + { + window_in.set(Window::DimX, Window::Dimension(window.x().start(), window.x().end(), 1)); + window_in.set(Window::DimY, Window::Dimension(window_end_y_multiple_of, window_end_y, 1)); + + Iterator input(in, window_in); + Iterator output(out, window_out); + + // Compute left-over elements along the y dimension (1x1) + execute_window_loop( + window_in, + [&](const Coordinates &id) + { + const uint16_t val0 = *(reinterpret_cast<uint16_t *>(input.ptr())); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint16_t) + id.x() * output_stride_in_bytes; + + *(reinterpret_cast<uint16_t *>(output.ptr() + dst_offset_in_bytes)) = val0; + }, + input, output); + } +} + +#ifdef __aarch64__ +inline uint32x4x2_t vld1q_u32_x2_(const uint32_t *ptr) +{ + // gcc-7 doesn't support vld1q_u32_x2 instruction + return {vld1q_u32(ptr), vld1q_u32(ptr + 4)}; +} + +inline void vst1q_u32_x2_(const uint32_t *ptr, const uint32x4x2_t &val) +{ + // gcc-7 doesn't support vst1q_u32_x2 instruction + vst1q_u32(const_cast<uint32_t *>(ptr), val.val[0]); + vst1q_u32(const_cast<uint32_t *>(ptr + 4), val.val[1]); +} + +void transpose_32bit_elements(const ITensor *in, ITensor *out, const Window &window) +{ + constexpr int window_step_x = 8; + constexpr int window_step_y = 8; + const int window_start_x = window.x().start(); + const int window_end_x = window.x().end(); + const int window_start_y = window.y().start(); + const int window_end_y = std::min(window.y().end(), static_cast<int>(in->info()->dimension(1))); + const int window_end_y_multiple_of = ((window_end_y - window_start_y) / window_step_y) * window_step_y; + const size_t input_stride_in_bytes = in->info()->strides_in_bytes()[1]; + const size_t output_stride_in_bytes = out->info()->strides_in_bytes()[1]; + + // Check if we need a left-over loop for the y dimension + bool left_over_loop_y = (((window_end_y - window_start_y) % window_step_y) != 0); + + Window window_in(window); + window_in.set(Window::DimX, Window::Dimension(0, 1, 1)); + if (left_over_loop_y) + { + // Check if window_end_y_multiple_of is greater than window_start_y + if (window_end_y_multiple_of > window_start_y) + { + window_in.set(Window::DimY, Window::Dimension(window_start_y, window_end_y_multiple_of, window_step_y)); + } + else + { + window_in.set(Window::DimY, Window::Dimension(0, 0, 1)); + } + } + + Window window_out(window); + window_out.set(Window::DimX, Window::Dimension(0, 0, 0)); + window_out.set(Window::DimY, Window::Dimension(0, 0, 0)); + + Iterator output(out, window_out); + + // Run the SIMD path if and only if the input is not a row-vector + if (in->info()->dimension(1) != 1) + { + Iterator input(in, window_in); + execute_window_loop( + window_in, + [&](const Coordinates &id) + { + // Compute 8x8 elements per iteration + int x = window_start_x; + for (; x <= (window_end_x - window_step_x); x += window_step_x) + { + // Load + const uint32x4x2_t row0 = + vld1q_u32_x2_(reinterpret_cast<const uint32_t *>(input.ptr() + 0 * input_stride_in_bytes) + x); + const uint32x4x2_t row1 = + vld1q_u32_x2_(reinterpret_cast<const uint32_t *>(input.ptr() + 1 * input_stride_in_bytes) + x); + const uint32x4x2_t row2 = + vld1q_u32_x2_(reinterpret_cast<const uint32_t *>(input.ptr() + 2 * input_stride_in_bytes) + x); + const uint32x4x2_t row3 = + vld1q_u32_x2_(reinterpret_cast<const uint32_t *>(input.ptr() + 3 * input_stride_in_bytes) + x); + const uint32x4x2_t row4 = + vld1q_u32_x2_(reinterpret_cast<const uint32_t *>(input.ptr() + 4 * input_stride_in_bytes) + x); + const uint32x4x2_t row5 = + vld1q_u32_x2_(reinterpret_cast<const uint32_t *>(input.ptr() + 5 * input_stride_in_bytes) + x); + const uint32x4x2_t row6 = + vld1q_u32_x2_(reinterpret_cast<const uint32_t *>(input.ptr() + 6 * input_stride_in_bytes) + x); + const uint32x4x2_t row7 = + vld1q_u32_x2_(reinterpret_cast<const uint32_t *>(input.ptr() + 7 * input_stride_in_bytes) + x); + + // Transpose 2x4 + const uint32x4x2_t k0_u32 = {vtrn1q_u32(row0.val[0], row1.val[0]), + vtrn2q_u32(row0.val[0], row1.val[0])}; + const uint32x4x2_t k1_u32 = {vtrn1q_u32(row0.val[1], row1.val[1]), + vtrn2q_u32(row0.val[1], row1.val[1])}; + const uint32x4x2_t k2_u32 = {vtrn1q_u32(row2.val[0], row3.val[0]), + vtrn2q_u32(row2.val[0], row3.val[0])}; + const uint32x4x2_t k3_u32 = {vtrn1q_u32(row2.val[1], row3.val[1]), + vtrn2q_u32(row2.val[1], row3.val[1])}; + const uint32x4x2_t k4_u32 = {vtrn1q_u32(row4.val[0], row5.val[0]), + vtrn2q_u32(row4.val[0], row5.val[0])}; + const uint32x4x2_t k5_u32 = {vtrn1q_u32(row4.val[1], row5.val[1]), + vtrn2q_u32(row4.val[1], row5.val[1])}; + const uint32x4x2_t k6_u32 = {vtrn1q_u32(row6.val[0], row7.val[0]), + vtrn2q_u32(row6.val[0], row7.val[0])}; + const uint32x4x2_t k7_u32 = {vtrn1q_u32(row6.val[1], row7.val[1]), + vtrn2q_u32(row6.val[1], row7.val[1])}; + + // Transpose 2x2 + const uint64x2x2_t k0_u64 = { + vtrn1q_u64(vreinterpretq_u64_u32(k0_u32.val[0]), vreinterpretq_u64_u32(k2_u32.val[0])), + vtrn2q_u64(vreinterpretq_u64_u32(k0_u32.val[0]), vreinterpretq_u64_u32(k2_u32.val[0]))}; + const uint64x2x2_t k1_u64 = { + vtrn1q_u64(vreinterpretq_u64_u32(k0_u32.val[1]), vreinterpretq_u64_u32(k2_u32.val[1])), + vtrn2q_u64(vreinterpretq_u64_u32(k0_u32.val[1]), vreinterpretq_u64_u32(k2_u32.val[1]))}; + const uint64x2x2_t k2_u64 = { + vtrn1q_u64(vreinterpretq_u64_u32(k1_u32.val[0]), vreinterpretq_u64_u32(k3_u32.val[0])), + vtrn2q_u64(vreinterpretq_u64_u32(k1_u32.val[0]), vreinterpretq_u64_u32(k3_u32.val[0]))}; + const uint64x2x2_t k3_u64 = { + vtrn1q_u64(vreinterpretq_u64_u32(k1_u32.val[1]), vreinterpretq_u64_u32(k3_u32.val[1])), + vtrn2q_u64(vreinterpretq_u64_u32(k1_u32.val[1]), vreinterpretq_u64_u32(k3_u32.val[1]))}; + const uint64x2x2_t k4_u64 = { + vtrn1q_u64(vreinterpretq_u64_u32(k4_u32.val[0]), vreinterpretq_u64_u32(k6_u32.val[0])), + vtrn2q_u64(vreinterpretq_u64_u32(k4_u32.val[0]), vreinterpretq_u64_u32(k6_u32.val[0]))}; + const uint64x2x2_t k5_u64 = { + vtrn1q_u64(vreinterpretq_u64_u32(k4_u32.val[1]), vreinterpretq_u64_u32(k6_u32.val[1])), + vtrn2q_u64(vreinterpretq_u64_u32(k4_u32.val[1]), vreinterpretq_u64_u32(k6_u32.val[1]))}; + const uint64x2x2_t k6_u64 = { + vtrn1q_u64(vreinterpretq_u64_u32(k5_u32.val[0]), vreinterpretq_u64_u32(k7_u32.val[0])), + vtrn2q_u64(vreinterpretq_u64_u32(k5_u32.val[0]), vreinterpretq_u64_u32(k7_u32.val[0]))}; + const uint64x2x2_t k7_u64 = { + vtrn1q_u64(vreinterpretq_u64_u32(k5_u32.val[1]), vreinterpretq_u64_u32(k7_u32.val[1])), + vtrn2q_u64(vreinterpretq_u64_u32(k5_u32.val[1]), vreinterpretq_u64_u32(k7_u32.val[1]))}; + + // Swap blocks + const uint32x4x2_t col0 = {vreinterpretq_u32_u64(k0_u64.val[0]), + vreinterpretq_u32_u64(k4_u64.val[0])}; + const uint32x4x2_t col1 = {vreinterpretq_u32_u64(k1_u64.val[0]), + vreinterpretq_u32_u64(k5_u64.val[0])}; + const uint32x4x2_t col2 = {vreinterpretq_u32_u64(k0_u64.val[1]), + vreinterpretq_u32_u64(k4_u64.val[1])}; + const uint32x4x2_t col3 = {vreinterpretq_u32_u64(k1_u64.val[1]), + vreinterpretq_u32_u64(k5_u64.val[1])}; + const uint32x4x2_t col4 = {vreinterpretq_u32_u64(k2_u64.val[0]), + vreinterpretq_u32_u64(k6_u64.val[0])}; + const uint32x4x2_t col5 = {vreinterpretq_u32_u64(k3_u64.val[0]), + vreinterpretq_u32_u64(k7_u64.val[0])}; + const uint32x4x2_t col6 = {vreinterpretq_u32_u64(k2_u64.val[1]), + vreinterpretq_u32_u64(k6_u64.val[1])}; + const uint32x4x2_t col7 = {vreinterpretq_u32_u64(k3_u64.val[1]), + vreinterpretq_u32_u64(k7_u64.val[1])}; + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint32_t) + x * output_stride_in_bytes; + + // Store + vst1q_u32_x2_( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 0 * output_stride_in_bytes), + col0); + vst1q_u32_x2_( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 1 * output_stride_in_bytes), + col1); + vst1q_u32_x2_( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 2 * output_stride_in_bytes), + col2); + vst1q_u32_x2_( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 3 * output_stride_in_bytes), + col3); + vst1q_u32_x2_( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 4 * output_stride_in_bytes), + col4); + vst1q_u32_x2_( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 5 * output_stride_in_bytes), + col5); + vst1q_u32_x2_( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 6 * output_stride_in_bytes), + col6); + vst1q_u32_x2_( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 7 * output_stride_in_bytes), + col7); + } + + // Compute left-over elements (8x1) + for (; x < window_end_x; ++x) + { + const uint32_t val0 = *(reinterpret_cast<uint32_t *>(input.ptr() + 0 * input_stride_in_bytes) + x); + const uint32_t val1 = *(reinterpret_cast<uint32_t *>(input.ptr() + 1 * input_stride_in_bytes) + x); + const uint32_t val2 = *(reinterpret_cast<uint32_t *>(input.ptr() + 2 * input_stride_in_bytes) + x); + const uint32_t val3 = *(reinterpret_cast<uint32_t *>(input.ptr() + 3 * input_stride_in_bytes) + x); + const uint32_t val4 = *(reinterpret_cast<uint32_t *>(input.ptr() + 4 * input_stride_in_bytes) + x); + const uint32_t val5 = *(reinterpret_cast<uint32_t *>(input.ptr() + 5 * input_stride_in_bytes) + x); + const uint32_t val6 = *(reinterpret_cast<uint32_t *>(input.ptr() + 6 * input_stride_in_bytes) + x); + const uint32_t val7 = *(reinterpret_cast<uint32_t *>(input.ptr() + 7 * input_stride_in_bytes) + x); + + uint32x4_t result0 = vdupq_n_u32(0); + uint32x4_t result1 = vdupq_n_u32(0); + result0 = vsetq_lane_u32(val0, result0, 0); + result0 = vsetq_lane_u32(val1, result0, 1); + result0 = vsetq_lane_u32(val2, result0, 2); + result0 = vsetq_lane_u32(val3, result0, 3); + result1 = vsetq_lane_u32(val4, result1, 0); + result1 = vsetq_lane_u32(val5, result1, 1); + result1 = vsetq_lane_u32(val6, result1, 2); + result1 = vsetq_lane_u32(val7, result1, 3); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint32_t) + x * output_stride_in_bytes; + + vst1q_u32_x2_(reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes), {result0, result1}); + } + }, + input, output); + } + + if (left_over_loop_y) + { + window_in.set(Window::DimX, Window::Dimension(window.x().start(), window.x().end(), 1)); + window_in.set(Window::DimY, Window::Dimension(window_end_y_multiple_of, window_end_y, 1)); + + Iterator input(in, window_in); + Iterator output(out, window_out); + + // Compute left-over elements along the y dimension (1x1) + execute_window_loop( + window_in, + [&](const Coordinates &id) + { + const uint32_t val0 = *(reinterpret_cast<uint32_t *>(input.ptr())); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint32_t) + id.x() * output_stride_in_bytes; + + *(reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes)) = val0; + }, + input, output); + } +} +#else // __aarch64__ +void transpose_32bit_elements(const ITensor *in, ITensor *out, const Window &window) +{ + const int window_step_x = 4; + const int window_step_y = 4; + const int window_start_x = window.x().start(); + const int window_end_x = window.x().end(); + const int window_start_y = window.y().start(); + const int window_end_y = std::min(window.y().end(), static_cast<int>(in->info()->dimension(1))); + const int window_end_y_multiple_of = ((window_end_y - window_start_y) / window_step_y) * window_step_y; + const size_t input_stride_in_bytes = in->info()->strides_in_bytes()[1]; + const size_t output_stride_in_bytes = out->info()->strides_in_bytes()[1]; + + // Check if we need a left-over loop for the y dimension + bool left_over_loop_y = (((window_end_y - window_start_y) % window_step_y) != 0); + + Window window_in(window); + window_in.set(Window::DimX, Window::Dimension(0, 1, 1)); + if (left_over_loop_y) + { + // Check if window_end_y_multiple_of is greater than window_start_y + if (window_end_y_multiple_of > window_start_y) + { + window_in.set(Window::DimY, Window::Dimension(window_start_y, window_end_y_multiple_of, window_step_y)); + } + else + { + window_in.set(Window::DimY, Window::Dimension(0, 0, 1)); + } + } + + Window window_out(window); + window_out.set(Window::DimX, Window::Dimension(0, 0, 0)); + window_out.set(Window::DimY, Window::Dimension(0, 0, 0)); + + Iterator output(out, window_out); + + // Run the SIMD path if and only if the input is not a row-vector + if (in->info()->dimension(1) != 1) + { + Iterator input(in, window_in); + execute_window_loop( + window_in, + [&](const Coordinates &id) + { + // Compute 4x4 elements per iteration + int x = window_start_x; + for (; x <= (window_end_x - window_step_x); x += window_step_x) + { + const uint32x4_t row0 = + vld1q_u32(reinterpret_cast<const uint32_t *>(input.ptr() + 0 * input_stride_in_bytes) + x); + const uint32x4_t row1 = + vld1q_u32(reinterpret_cast<const uint32_t *>(input.ptr() + 1 * input_stride_in_bytes) + x); + const uint32x4_t row2 = + vld1q_u32(reinterpret_cast<const uint32_t *>(input.ptr() + 2 * input_stride_in_bytes) + x); + const uint32x4_t row3 = + vld1q_u32(reinterpret_cast<const uint32_t *>(input.ptr() + 3 * input_stride_in_bytes) + x); + + // Transpose 2x2 + const uint32x2x2_t k0_u32 = vtrn_u32(vget_low_u32(row0), vget_low_u32(row1)); + const uint32x2x2_t k1_u32 = vtrn_u32(vget_high_u32(row2), vget_high_u32(row3)); + const uint32x2x2_t k2_u32 = vtrn_u32(vget_high_u32(row0), vget_high_u32(row1)); + const uint32x2x2_t k3_u32 = vtrn_u32(vget_low_u32(row2), vget_low_u32(row3)); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint32_t) + x * output_stride_in_bytes; + + // Swap block 01 with block 10 and store + vst1q_u32( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 0 * output_stride_in_bytes), + vcombine_u32(k0_u32.val[0], k3_u32.val[0])); + vst1q_u32( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 1 * output_stride_in_bytes), + vcombine_u32(k0_u32.val[1], k3_u32.val[1])); + vst1q_u32( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 2 * output_stride_in_bytes), + vcombine_u32(k2_u32.val[0], k1_u32.val[0])); + vst1q_u32( + reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes + 3 * output_stride_in_bytes), + vcombine_u32(k2_u32.val[1], k1_u32.val[1])); + } + + // Compute left-over elements (1x4) + for (; x < window_end_x; ++x) + { + const uint32_t val0 = *(reinterpret_cast<uint32_t *>(input.ptr() + 0 * input_stride_in_bytes) + x); + const uint32_t val1 = *(reinterpret_cast<uint32_t *>(input.ptr() + 1 * input_stride_in_bytes) + x); + const uint32_t val2 = *(reinterpret_cast<uint32_t *>(input.ptr() + 2 * input_stride_in_bytes) + x); + const uint32_t val3 = *(reinterpret_cast<uint32_t *>(input.ptr() + 3 * input_stride_in_bytes) + x); + + uint32x4_t result = vdupq_n_u32(0); + result = vsetq_lane_u32(val0, result, 0); + result = vsetq_lane_u32(val1, result, 1); + result = vsetq_lane_u32(val2, result, 2); + result = vsetq_lane_u32(val3, result, 3); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint32_t) + x * output_stride_in_bytes; + + vst1q_u32(reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes), result); + } + }, + input, output); + } + + if (left_over_loop_y) + { + window_in.set(Window::DimX, Window::Dimension(window.x().start(), window.x().end(), 1)); + window_in.set(Window::DimY, Window::Dimension(window_end_y_multiple_of, window_end_y, 1)); + + Iterator input(in, window_in); + Iterator output(out, window_out); + + // Compute left-over elements along the y dimension (1x1) + execute_window_loop( + window_in, + [&](const Coordinates &id) + { + const uint32_t val0 = *(reinterpret_cast<uint32_t *>(input.ptr())); + + // Compute destination address + const size_t dst_offset_in_bytes = id.y() * sizeof(uint32_t) + id.x() * output_stride_in_bytes; + + *(reinterpret_cast<uint32_t *>(output.ptr() + dst_offset_in_bytes)) = val0; + }, + input, output); + } +} +#endif // __aarch64__ +} // namespace + +void CpuTransposeKernel::configure(const ITensorInfo *src, ITensorInfo *dst) +{ + ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst); + + // Destination auto inizialitation if not yet initialized + const TensorShape dst_shape = misc::shape_calculator::compute_transposed_shape(*src); + auto_init_if_empty(*dst, src->clone()->set_tensor_shape(dst_shape)); + + // Explicitly set the tensor shape to preserve dimensions + dst->set_tensor_shape(dst_shape); + + // Perform validation step + ARM_COMPUTE_ERROR_THROW_ON(validate(src, dst)); + + // Note: This kernel performs 16 elements per iteration. + // However, since we use a left-over for loop on both dimensions (X and Y), we cannot have any read or write out of memory + // For this reason num_elems_processed_per_iteration_x is set to 1 + const unsigned int num_elems_processed_per_iteration_x = 1; + const unsigned int num_elems_processed_per_iteration_y = num_elems_processed(src->element_size()); + + // Configure kernel window + Window win = + calculate_max_window(*src, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y)); + + // The CpuTranspose doesn't need padding so update_window_and_padding() can be skipped + Coordinates coord; + coord.set_num_dimensions(dst->num_dimensions()); + dst->set_valid_region(ValidRegion(coord, dst->tensor_shape())); + + ICpuKernel::configure(win); +} + +Status CpuTransposeKernel::validate(const ITensorInfo *src, const ITensorInfo *dst) +{ + ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src); + //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use CPU FP16 instructions. + ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN); + + // Error if input is not 8 bit, 16bit or 32bit + ARM_COMPUTE_RETURN_ERROR_ON_MSG(src->element_size() != 1 && src->element_size() != 2 && src->element_size() != 4, + "Element size not supported"); + + // Validate configured destination + if (dst->total_size() != 0) + { + const TensorShape dst_shape = misc::shape_calculator::compute_transposed_shape(*src); + + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(dst->tensor_shape(), dst_shape); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst); + } + + return Status{}; +} + +void CpuTransposeKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) +{ + ARM_COMPUTE_UNUSED(info); + ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); + ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window); + + const auto src = tensors.get_const_tensor(TensorType::ACL_SRC); + auto dst = tensors.get_tensor(TensorType::ACL_DST); + + switch (src->info()->element_size()) + { + case 1: + transpose_8bit_elements(src, dst, window); + break; + case 2: + transpose_16bit_elements(src, dst, window); + break; + case 4: + transpose_32bit_elements(src, dst, window); + break; + default: + ARM_COMPUTE_ERROR("Element size not supported"); + break; + } +} + +const char *CpuTransposeKernel::name() const +{ + return "CpuTransposeKernel"; +} +} // namespace kernels +} // namespace cpu +} // namespace arm_compute |