diff options
| author | Michalis Spyrou <michalis.spyrou@arm.com> | 2020-02-21 18:58:38 +0000 |
|---|---|---|
| committer | Michalis Spyrou <michalis.spyrou@arm.com> | 2020-02-27 14:38:09 +0000 |
| commit | 4de2d59419bbbeefccdaccb563d3157d4c88db68 (patch) | |
| tree | 9fed5f3f69a1e7b07789bc5a685ee9e8787dd3e0 /src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp | |
| parent | ee3e35c0bc19c43a222cda614b4332085a5f144d (diff) | |
| download | ComputeLibrary-4de2d59419bbbeefccdaccb563d3157d4c88db68.tar.gz | |
COMPMID-3147: Remove padding from NEDepthConvertLayerKernel
Change-Id: I0e388e9898ea82a7a1af16a987890696bb489547
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2781
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp')
| -rw-r--r-- | src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp | 1073 |
1 files changed, 741 insertions, 332 deletions
diff --git a/src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp b/src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp index f5fb9c09aa..f824f7ac58 100644 --- a/src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp +++ b/src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp @@ -31,6 +31,7 @@ #include "arm_compute/core/NEON/NEMath.h" #include "arm_compute/core/TensorInfo.h" #include "arm_compute/core/Validate.h" +#include "arm_compute/core/utils/misc/SaturateCast.h" #include <arm_neon.h> @@ -90,21 +91,6 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, C return Status{}; } - -std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output) -{ - constexpr unsigned int num_elems_processed_per_iteration = 16; - - Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration)); - - AccessWindowHorizontal input_access(input, 0, num_elems_processed_per_iteration); - AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration); - bool window_changed = update_window_and_padding(win, input_access, output_access); - output_access.set_valid_region(win, output->valid_region()); - - Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{}; - return std::make_pair(err, win); -} } // namespace NEDepthConvertLayerKernel::NEDepthConvertLayerKernel() @@ -127,16 +113,17 @@ void NEDepthConvertLayerKernel::configure(const ITensor *input, ITensor *output, ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), policy, shift)); // Configure kernel window - auto win_config = validate_and_configure_window(input->info(), output->info()); - ARM_COMPUTE_ERROR_THROW_ON(win_config.first); - ICPPKernel::configure(win_config.second); + Window win = calculate_max_window(*input->info(), Steps()); + Coordinates coord; + coord.set_num_dimensions(output->info()->num_dimensions()); + output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape())); + + ICPPKernel::configure(win); } Status NEDepthConvertLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, ConvertPolicy policy, uint32_t shift) { ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, policy, shift)); - ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first); - return Status{}; } @@ -148,8 +135,15 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info ARM_COMPUTE_ERROR_ON_NULLPTR(_input, _output); ARM_COMPUTE_ERROR_ON(_input == _output); - Iterator input(_input, window); - Iterator output(_output, window); + const auto window_start_x = static_cast<int>(window.x().start()); + const auto window_end_x = static_cast<int>(window.x().end()); + const int window_step_x = 16; + + Window win{ window }; + win.set(Window::DimX, Window::Dimension(0, 1, 1)); + + Iterator input(_input, win); + Iterator output(_output, win); switch(_input->info()->data_type()) { @@ -162,20 +156,33 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::S16: { /* Up-conversion QASYMM8_SIGNED -> S16 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int8x16_t texels_s8 = vld1q_s8(reinterpret_cast<int8_t *>(input.ptr())); + const auto input_ptr = reinterpret_cast<const int8_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int16_t *>(output.ptr()); + int x = window_start_x; - const int16x8x2_t texels = + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int8x16_t texels_s8 = vld1q_s8(input_ptr + x); + + const int16x8x2_t texels = { - vshlq_s16(vmovl_s8(vget_low_s8(texels_s8)), b), - vshlq_s16(vmovl_s8(vget_high_s8(texels_s8)), b) - } - }; + { + vshlq_s16(vmovl_s8(vget_low_s8(texels_s8)), b), + vshlq_s16(vmovl_s8(vget_high_s8(texels_s8)), b) + } + }; - vst1q_s16(reinterpret_cast<int16_t *>(output.ptr()), texels.val[0]); - vst1q_s16(reinterpret_cast<int16_t *>(output.ptr()) + 8, texels.val[1]); + vst1q_s16(output_ptr + x, texels.val[0]); + vst1q_s16(output_ptr + x + 8, texels.val[1]); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<int16_t>(*(input_ptr + x) << _shift); + } }, input, output); break; @@ -183,22 +190,35 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::S32: { /* Up-conversion QASYMM8_SIGNED -> S32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int8x16_t texels_s8 = vld1q_s8(reinterpret_cast<int8_t *>(input.ptr())); + const auto input_ptr = reinterpret_cast<const int8_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int32_t *>(output.ptr()); + int x = window_start_x; - const int16x8x2_t texels = + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int8x16_t texels_s8 = vld1q_s8(input_ptr + x); + + const int16x8x2_t texels = { - vshlq_s16(vmovl_s8(vget_low_s8(texels_s8)), b), - vshlq_s16(vmovl_s8(vget_high_s8(texels_s8)), b) - } - }; + { + vshlq_s16(vmovl_s8(vget_low_s8(texels_s8)), b), + vshlq_s16(vmovl_s8(vget_high_s8(texels_s8)), b) + } + }; - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()), vmovl_s16(vget_low_s16(texels.val[0]))); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 4, vmovl_s16(vget_high_s16(texels.val[0]))); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 8, vmovl_s16(vget_low_s16(texels.val[1]))); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 12, vmovl_s16(vget_high_s16(texels.val[1]))); + vst1q_s32(output_ptr + x, vmovl_s16(vget_low_s16(texels.val[0]))); + vst1q_s32(output_ptr + x + 4, vmovl_s16(vget_high_s16(texels.val[0]))); + vst1q_s32(output_ptr + x + 8, vmovl_s16(vget_low_s16(texels.val[1]))); + vst1q_s32(output_ptr + x + 12, vmovl_s16(vget_high_s16(texels.val[1]))); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<int32_t>(*(input_ptr + x) << _shift); + } }, input, output); break; @@ -206,21 +226,34 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::F32: { /* Up-conversion QASYMM8_SIGNED -> F32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int8x16_t texels_s8 = vld1q_s8(reinterpret_cast<int8_t *>(input.ptr())); + const auto input_ptr = reinterpret_cast<const int8_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<float *>(output.ptr()); - const int16x8x2_t texels = + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int8x16_t texels_s8 = vld1q_s8(reinterpret_cast<int8_t *>(input.ptr())); + + const int16x8x2_t texels = { - vshlq_s16(vmovl_s8(vget_low_s8(texels_s8)), b), - vshlq_s16(vmovl_s8(vget_high_s8(texels_s8)), b) - } - }; - vst1q_f32(reinterpret_cast<float *>(output.ptr()), vcvtq_f32_s32(vmovl_s16(vget_low_s16(texels.val[0])))); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 4, vcvtq_f32_s32(vmovl_s16(vget_high_s16(texels.val[0])))); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 8, vcvtq_f32_s32(vmovl_s16(vget_low_s16(texels.val[1])))); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 12, vcvtq_f32_s32(vmovl_s16(vget_high_s16(texels.val[1])))); + { + vshlq_s16(vmovl_s8(vget_low_s8(texels_s8)), b), + vshlq_s16(vmovl_s8(vget_high_s8(texels_s8)), b) + } + }; + vst1q_f32(output_ptr + x, vcvtq_f32_s32(vmovl_s16(vget_low_s16(texels.val[0])))); + vst1q_f32(output_ptr + x + 4, vcvtq_f32_s32(vmovl_s16(vget_high_s16(texels.val[0])))); + vst1q_f32(output_ptr + x + 8, vcvtq_f32_s32(vmovl_s16(vget_low_s16(texels.val[1])))); + vst1q_f32(output_ptr + x + 12, vcvtq_f32_s32(vmovl_s16(vget_high_s16(texels.val[1])))); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<float>(*(input_ptr + x) << _shift); + } }, input, output); break; @@ -229,19 +262,32 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::F16: { /* Up-conversion QASYMM8_SIGNED -> F16 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int8x16_t texels_s8 = vld1q_s8(reinterpret_cast<int8_t *>(input.ptr())); + const auto input_ptr = reinterpret_cast<const int8_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<float16_t *>(output.ptr()); + int x = window_start_x; - const int16x8x2_t texels = + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int8x16_t texels_s8 = vld1q_s8(input_ptr + x); + + const int16x8x2_t texels = { - vshlq_s16(vmovl_s8(vget_low_s8(texels_s8)), b), - vshlq_s16(vmovl_s8(vget_high_s8(texels_s8)), b) - } - }; - vst1q_f16(reinterpret_cast<float16_t *>(output.ptr()), vcvtq_f16_s16(texels.val[0])); - vst1q_f16(reinterpret_cast<float16_t *>(output.ptr()) + 8, vcvtq_f16_s16(texels.val[1])); + { + vshlq_s16(vmovl_s8(vget_low_s8(texels_s8)), b), + vshlq_s16(vmovl_s8(vget_high_s8(texels_s8)), b) + } + }; + vst1q_f16(output_ptr + x, vcvtq_f16_s16(texels.val[0])); + vst1q_f16(output_ptr + x + 8, vcvtq_f16_s16(texels.val[1])); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<float16_t>(*(input_ptr + x) << _shift); + } }, input, output); break; @@ -264,20 +310,34 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::S16: { /* Up-conversion U8 -> S16 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const uint8x16_t texels_u8 = vld1q_u8(input.ptr()); + const auto input_ptr = reinterpret_cast<const uint8_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int16_t *>(output.ptr()); - const int16x8x2_t texels = + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const uint8x16_t texels_u8 = vld1q_u8(input_ptr + x); + + const int16x8x2_t texels = { - vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), - vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) - } - }; + { + vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), + vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) + } + }; + + vst1q_s16(output_ptr + x, texels.val[0]); + vst1q_s16(output_ptr + x + 8, texels.val[1]); + } - vst1q_s16(reinterpret_cast<int16_t *>(output.ptr()), texels.val[0]); - vst1q_s16(reinterpret_cast<int16_t *>(output.ptr()) + 8, texels.val[1]); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + auto in = static_cast<int32_t>(*(input_ptr + x)); + *(output_ptr + x) = in << _shift; + } }, input, output); break; @@ -285,22 +345,36 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::S32: { /* Up-conversion U8 -> S32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const uint8x16_t texels_u8 = vld1q_u8(input.ptr()); + const auto input_ptr = reinterpret_cast<const uint8_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int32_t *>(output.ptr()); - const int16x8x2_t texels = + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const uint8x16_t texels_u8 = vld1q_u8(input_ptr + x); + + const int16x8x2_t texels = { - vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), - vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) - } - }; + { + vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), + vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) + } + }; + + vst1q_s32(output_ptr + x, vmovl_s16(vget_low_s16(texels.val[0]))); + vst1q_s32(output_ptr + x + 4, vmovl_s16(vget_high_s16(texels.val[0]))); + vst1q_s32(output_ptr + x + 8, vmovl_s16(vget_low_s16(texels.val[1]))); + vst1q_s32(output_ptr + x + 12, vmovl_s16(vget_high_s16(texels.val[1]))); + } - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()), vmovl_s16(vget_low_s16(texels.val[0]))); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 4, vmovl_s16(vget_high_s16(texels.val[0]))); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 8, vmovl_s16(vget_low_s16(texels.val[1]))); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 12, vmovl_s16(vget_high_s16(texels.val[1]))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + auto in = static_cast<uint32_t>(*(input_ptr + x)); + *(output_ptr + x) = in << _shift; + } }, input, output); break; @@ -308,21 +382,35 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::F32: { /* Up-conversion U8 -> F32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const uint8x16_t texels_u8 = vld1q_u8(input.ptr()); + const auto input_ptr = reinterpret_cast<const uint8_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<float *>(output.ptr()); - const int16x8x2_t texels = + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const uint8x16_t texels_u8 = vld1q_u8(input_ptr + x); + + const int16x8x2_t texels = { - vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), - vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) - } - }; - vst1q_f32(reinterpret_cast<float *>(output.ptr()), vcvtq_f32_s32(vmovl_s16(vget_low_s16(texels.val[0])))); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 4, vcvtq_f32_s32(vmovl_s16(vget_high_s16(texels.val[0])))); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 8, vcvtq_f32_s32(vmovl_s16(vget_low_s16(texels.val[1])))); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 12, vcvtq_f32_s32(vmovl_s16(vget_high_s16(texels.val[1])))); + { + vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), + vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) + } + }; + vst1q_f32(output_ptr + x, vcvtq_f32_s32(vmovl_s16(vget_low_s16(texels.val[0])))); + vst1q_f32(output_ptr + x + 4, vcvtq_f32_s32(vmovl_s16(vget_high_s16(texels.val[0])))); + vst1q_f32(output_ptr + x + 8, vcvtq_f32_s32(vmovl_s16(vget_low_s16(texels.val[1])))); + vst1q_f32(output_ptr + x + 12, vcvtq_f32_s32(vmovl_s16(vget_high_s16(texels.val[1])))); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + auto in = static_cast<uint32_t>(*(input_ptr + x)); + *(output_ptr + x) = static_cast<float>(in << _shift); + } }, input, output); break; @@ -331,42 +419,67 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::F16: { /* Up-conversion U8 -> F16 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const uint8x16_t texels_u8 = vld1q_u8(input.ptr()); + const auto input_ptr = reinterpret_cast<const uint8_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<float16_t *>(output.ptr()); - const int16x8x2_t texels = + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const uint8x16_t texels_u8 = vld1q_u8(input_ptr + x); + + const int16x8x2_t texels = { - vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), - vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) - } - }; - vst1q_f16(reinterpret_cast<float16_t *>(output.ptr()), vcvtq_f16_s16(texels.val[0])); - vst1q_f16(reinterpret_cast<float16_t *>(output.ptr()) + 8, vcvtq_f16_s16(texels.val[1])); + { + vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), + vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) + } + }; + vst1q_f16(output_ptr + x, vcvtq_f16_s16(texels.val[0])); + vst1q_f16(output_ptr + x + 8, vcvtq_f16_s16(texels.val[1])); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<float16_t>(*(input_ptr + x) << _shift); + } }, input, output); break; } #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC - case DataType::U16: { /* Up-conversion U8 -> U16 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const uint8x16_t texels_u8 = vld1q_u8(input.ptr()); + const auto input_ptr = reinterpret_cast<const uint8_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint16_t *>(output.ptr()); - const uint16x8x2_t texels = + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const uint8x16_t texels_u8 = vld1q_u8(input_ptr + x); + + const uint16x8x2_t texels = { - vshlq_u16(vmovl_u8(vget_low_u8(texels_u8)), b), - vshlq_u16(vmovl_u8(vget_high_u8(texels_u8)), b) - } - }; + { + vshlq_u16(vmovl_u8(vget_low_u8(texels_u8)), b), + vshlq_u16(vmovl_u8(vget_high_u8(texels_u8)), b) + } + }; - vst1q_u16(reinterpret_cast<uint16_t *>(output.ptr()), texels.val[0]); - vst1q_u16(reinterpret_cast<uint16_t *>(output.ptr()) + 8, texels.val[1]); + vst1q_u16(output_ptr + x, texels.val[0]); + vst1q_u16(output_ptr + x + 8, texels.val[1]); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<uint16_t>(*(input_ptr + x)) << _shift; + } }, input, output); break; @@ -387,33 +500,59 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info /* Down-conversion S16 -> QASYMM8_SIGNED */ if(ConvertPolicy::SATURATE == _policy) { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const int16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int16x8x2_t texels = { - vqshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr())), b), - vqshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr()) + 8), b) - } - }; + { + vqshlq_s16(vld1q_s16(input_ptr + x), b), + vqshlq_s16(vld1q_s16(input_ptr + x + 8), b) + } + }; + + vst1q_s8(output_ptr + x, vcombine_s8(vqmovn_s16(texels.val[0]), vqmovn_s16(texels.val[1]))); + } - vst1q_s8(reinterpret_cast<int8_t *>(output.ptr()), vcombine_s8(vqmovn_s16(texels.val[0]), vqmovn_s16(texels.val[1]))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = utils::cast::saturate_cast<int8_t>(*(input_ptr + x) >> _shift); + } }, input, output); } else { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const int16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int16x8x2_t texels = { - vshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr())), b), - vshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr()) + 8), b) - } - }; + { + vshlq_s16(vld1q_s16(input_ptr + x), b), + vshlq_s16(vld1q_s16(input_ptr + x + 8), b) + } + }; + + vst1q_s8(output_ptr + x, vcombine_s8(vmovn_s16(texels.val[0]), vmovn_s16(texels.val[1]))); + } - vst1q_s8(reinterpret_cast<int8_t *>(output.ptr()), vcombine_s8(vmovn_s16(texels.val[0]), vmovn_s16(texels.val[1]))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<int8_t>(*(input_ptr + x) >> _shift); + } }, input, output); } @@ -426,34 +565,60 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info /* Down-conversion S16 -> U8 */ if(ConvertPolicy::SATURATE == _policy) { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const int16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int16x8x2_t texels = { - vqshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr())), b), - vqshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr()) + 8), b) - } - }; + { + vqshlq_s16(vld1q_s16(input_ptr + x), b), + vqshlq_s16(vld1q_s16(input_ptr + x + 8), b) + } + }; + + vst1q_u8(output_ptr + x, vcombine_u8(vqmovun_s16(texels.val[0]), vqmovun_s16(texels.val[1]))); + } - vst1q_u8(output.ptr(), vcombine_u8(vqmovun_s16(texels.val[0]), vqmovun_s16(texels.val[1]))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = utils::cast::saturate_cast<uint8_t>(*(input_ptr + x) >> _shift); + } }, input, output); } else { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const int16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int16x8x2_t texels = { - vshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr())), b), - vshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr()) + 8), b) - } - }; + { + vshlq_s16(vld1q_s16(input_ptr + x), b), + vshlq_s16(vld1q_s16(input_ptr + x + 8), b) + } + }; + + vst1q_u8(output_ptr + x, vcombine_u8(vmovn_u16(vreinterpretq_u16_s16(texels.val[0])), + vmovn_u16(vreinterpretq_u16_s16(texels.val[1])))); + } - vst1q_u8(output.ptr(), vcombine_u8(vmovn_u16(vreinterpretq_u16_s16(texels.val[0])), - vmovn_u16(vreinterpretq_u16_s16(texels.val[1])))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<uint8_t>(*(input_ptr + x) >> _shift); + } }, input, output); } @@ -464,30 +629,43 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info const int32x4_t b = vdupq_n_s32(_shift); /* Up-conversion S16 -> S32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const int16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int32_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int16x8x2_t texels = { - vld1q_s16(reinterpret_cast<int16_t *>(input.ptr())), - vld1q_s16(reinterpret_cast<int16_t *>(input.ptr()) + 8) - } - }; + { + vld1q_s16(input_ptr + x), + vld1q_s16(input_ptr + x + 8) + } + }; - const int32x4x4_t texels_s32 = - { + const int32x4x4_t texels_s32 = { - vshlq_s32(vmovl_s16(vget_low_s16(texels.val[0])), b), - vshlq_s32(vmovl_s16(vget_high_s16(texels.val[0])), b), - vshlq_s32(vmovl_s16(vget_low_s16(texels.val[1])), b), - vshlq_s32(vmovl_s16(vget_high_s16(texels.val[1])), b) - } - }; + { + vshlq_s32(vmovl_s16(vget_low_s16(texels.val[0])), b), + vshlq_s32(vmovl_s16(vget_high_s16(texels.val[0])), b), + vshlq_s32(vmovl_s16(vget_low_s16(texels.val[1])), b), + vshlq_s32(vmovl_s16(vget_high_s16(texels.val[1])), b) + } + }; - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()), texels_s32.val[0]); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 4, texels_s32.val[1]); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 8, texels_s32.val[2]); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 12, texels_s32.val[3]); + vst1q_s32(output_ptr + x, texels_s32.val[0]); + vst1q_s32(output_ptr + x + 4, texels_s32.val[1]); + vst1q_s32(output_ptr + x + 8, texels_s32.val[2]); + vst1q_s32(output_ptr + x + 12, texels_s32.val[3]); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<int32_t>(*(input_ptr + x) << _shift); + } }, input, output); break; @@ -508,33 +686,60 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info /* Down-conversion U16 -> U8 */ if(ConvertPolicy::SATURATE == _policy) { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const uint16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const uint16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const uint16x8x2_t texels = { - vqshlq_u16(vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr())), b), - vqshlq_u16(vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr()) + 8), b) - } - }; + { + vqshlq_u16(vld1q_u16(input_ptr + x), b), + vqshlq_u16(vld1q_u16(input_ptr + x + 8), b) + } + }; + + vst1q_u8(output_ptr + x, vcombine_u8(vqmovn_u16(texels.val[0]), vqmovn_u16(texels.val[1]))); + } - vst1q_u8(output.ptr(), vcombine_u8(vqmovn_u16(texels.val[0]), vqmovn_u16(texels.val[1]))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = utils::cast::saturate_cast<uint8_t>(*(input_ptr + x) >> _shift); + } }, input, output); } else { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const uint16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const uint16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const uint16x8x2_t texels = { - vshlq_u16(vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr())), b), - vshlq_u16(vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr()) + 8), b) - } - }; + { + vshlq_u16(vld1q_u16(input_ptr + x), b), + vshlq_u16(vld1q_u16(input_ptr + x + 8), b) + } + }; + + vst1q_u8(output_ptr + x, vcombine_u8(vmovn_u16(texels.val[0]), vmovn_u16(texels.val[1]))); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<uint8_t>(*(input_ptr + x) >> _shift); + } - vst1q_u8(output.ptr(), vcombine_u8(vmovn_u16(texels.val[0]), vmovn_u16(texels.val[1]))); }, input, output); } @@ -545,20 +750,33 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info const int32x4_t b = vdupq_n_s32(_shift); /* Up-conversion U16 -> U32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const uint16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const uint16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint32_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const uint16x8x2_t texels = { - vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr())), - vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr()) + 8) - } - }; + { + vld1q_u16(input_ptr + x), + vld1q_u16(input_ptr + x + 8) + } + }; + + vst1q_u32(output_ptr + x, vshlq_u32(vmovl_u16(vget_low_u16(texels.val[0])), b)); + vst1q_u32(output_ptr + x + 4, vshlq_u32(vmovl_u16(vget_high_u16(texels.val[0])), b)); + vst1q_u32(output_ptr + x + 8, vshlq_u32(vmovl_u16(vget_low_u16(texels.val[1])), b)); + vst1q_u32(output_ptr + x + 12, vshlq_u32(vmovl_u16(vget_high_u16(texels.val[1])), b)); + } + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<uint32_t>(*(input_ptr + x) << _shift); + } - vst1q_u32(reinterpret_cast<uint32_t *>(output.ptr()), vshlq_u32(vmovl_u16(vget_low_u16(texels.val[0])), b)); - vst1q_u32(reinterpret_cast<uint32_t *>(output.ptr()) + 4, vshlq_u32(vmovl_u16(vget_high_u16(texels.val[0])), b)); - vst1q_u32(reinterpret_cast<uint32_t *>(output.ptr()) + 8, vshlq_u32(vmovl_u16(vget_low_u16(texels.val[1])), b)); - vst1q_u32(reinterpret_cast<uint32_t *>(output.ptr()) + 12, vshlq_u32(vmovl_u16(vget_high_u16(texels.val[1])), b)); }, input, output); break; @@ -574,20 +792,34 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info { case DataType::QASYMM8_SIGNED: { - const float16x8_t scale = vdupq_n_f16(1 << _shift); + const float16_t scale_s = 1 << _shift; + const float16x8_t scale = vdupq_n_f16(scale_s); /* Up-conversion F16 -> QASYMM8_SIGNED */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const float16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const float16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const float16x8x2_t texels = { - vmulq_f16(vld1q_f16(reinterpret_cast<float16_t *>(input.ptr())), scale), - vmulq_f16(vld1q_f16(reinterpret_cast<float16_t *>(input.ptr()) + 8), scale), - } - }; + { + vmulq_f16(vld1q_f16(input_ptr + x), scale), + vmulq_f16(vld1q_f16(input_ptr + x + 8), scale), + } + }; + + vst1q_s8(output_ptr + x, vcombine_s8(vqmovn_s16(vcvtq_s16_f16(texels.val[0])), vqmovn_s16(vcvtq_s16_f16(texels.val[1])))); + } - vst1q_s8(reinterpret_cast<int8_t *>(output.ptr()), vcombine_s8(vqmovn_s16(vcvtq_s16_f16(texels.val[0])), vqmovn_s16(vcvtq_s16_f16(texels.val[1])))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<int8_t>(*(input_ptr + x) * scale_s); + } }, input, output); break; @@ -595,66 +827,108 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::QASYMM8: case DataType::U8: { - const float16x8_t scale = vdupq_n_f16(1 << _shift); + const float16_t scale_s = 1 << _shift; + const float16x8_t scale = vdupq_n_f16(scale_s); /* Up-conversion F16 -> U8 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const float16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const float16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const float16x8x2_t texels = { - vmulq_f16(vld1q_f16(reinterpret_cast<float16_t *>(input.ptr())), scale), - vmulq_f16(vld1q_f16(reinterpret_cast<float16_t *>(input.ptr()) + 8), scale), - } - }; + { + vmulq_f16(vld1q_f16(input_ptr + x), scale), + vmulq_f16(vld1q_f16(input_ptr + x + 8), scale), + } + }; + + vst1q_u8(output_ptr + x, vcombine_u8(vqmovun_s16(vcvtq_s16_f16(texels.val[0])), vqmovun_s16(vcvtq_s16_f16(texels.val[1])))); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<uint8_t>(*(input_ptr + x) * scale_s); + } - vst1q_u8(reinterpret_cast<uint8_t *>(output.ptr()), vcombine_u8(vqmovun_s16(vcvtq_s16_f16(texels.val[0])), vqmovun_s16(vcvtq_s16_f16(texels.val[1])))); }, input, output); break; } case DataType::F32: { - const float32x4_t scale = vdupq_n_f32(1 << _shift); + const float scale_s = 1 << _shift; + const float32x4_t scale = vdupq_n_f32(scale_s); /* Up-conversion F16 -> F32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const float16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const float16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<float *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const float16x8x2_t texels = { - vld1q_f16(reinterpret_cast<float16_t *>(input.ptr())), - vld1q_f16(reinterpret_cast<float16_t *>(input.ptr()) + 8) - } - }; - - vst1q_f32(reinterpret_cast<float *>(output.ptr()), vmulq_f32(vcvt_f32_f16(vget_low_f16(texels.val[0])), scale)); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 4, vmulq_f32(vcvt_f32_f16(vget_high_f16(texels.val[0])), scale)); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 8, vmulq_f32(vcvt_f32_f16(vget_low_f16(texels.val[1])), scale)); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 12, vmulq_f32(vcvt_f32_f16(vget_high_f16(texels.val[1])), scale)); + { + vld1q_f16(input_ptr + x), + vld1q_f16(input_ptr + x + 8) + } + }; + vst1q_f32(output_ptr + x, vmulq_f32(vcvt_f32_f16(vget_low_f16(texels.val[0])), scale)); + vst1q_f32(output_ptr + x + 4, vmulq_f32(vcvt_f32_f16(vget_high_f16(texels.val[0])), scale)); + vst1q_f32(output_ptr + x + 8, vmulq_f32(vcvt_f32_f16(vget_low_f16(texels.val[1])), scale)); + vst1q_f32(output_ptr + x + 12, vmulq_f32(vcvt_f32_f16(vget_high_f16(texels.val[1])), scale)); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<float>(*(input_ptr + x) * scale_s); + } }, input, output); break; } case DataType::S32: { - const float32x4_t scale = vdupq_n_f32(1 << _shift); + const float scale_s = 1 << _shift; + const float32x4_t scale = vdupq_n_f32(scale_s); /* Up-conversion F16 -> S32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const float16x8x2_t texels = + const auto input_ptr = reinterpret_cast<const float16_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int32_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const float16x8x2_t texels = { - vld1q_f16(reinterpret_cast<float16_t *>(input.ptr())), - vld1q_f16(reinterpret_cast<float16_t *>(input.ptr()) + 8) - } - }; + { + vld1q_f16(input_ptr + x), + vld1q_f16(input_ptr + x + 8) + } + }; + + vst1q_s32(output_ptr + x, vcvtq_s32_f32(vmulq_f32(vcvt_f32_f16(vget_low_f16(texels.val[0])), scale))); + vst1q_s32(output_ptr + x + 4, vcvtq_s32_f32(vmulq_f32(vcvt_f32_f16(vget_high_f16(texels.val[0])), scale))); + vst1q_s32(output_ptr + x + 8, vcvtq_s32_f32(vmulq_f32(vcvt_f32_f16(vget_low_f16(texels.val[1])), scale))); + vst1q_s32(output_ptr + x + 12, vcvtq_s32_f32(vmulq_f32(vcvt_f32_f16(vget_high_f16(texels.val[1])), scale))); + } - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()), vcvtq_s32_f32(vmulq_f32(vcvt_f32_f16(vget_low_f16(texels.val[0])), scale))); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 4, vcvtq_s32_f32(vmulq_f32(vcvt_f32_f16(vget_high_f16(texels.val[0])), scale))); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 8, vcvtq_s32_f32(vmulq_f32(vcvt_f32_f16(vget_low_f16(texels.val[1])), scale))); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 12, vcvtq_s32_f32(vmulq_f32(vcvt_f32_f16(vget_high_f16(texels.val[1])), scale))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<int32_t>(*(input_ptr + x) * scale_s); + } }, input, output); break; @@ -670,23 +944,37 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC case DataType::F16: { - const float32x4_t scale = vdupq_n_f32(1.f / (1 << _shift)); + const float scale_s = 1.f / (1 << _shift); + const float32x4_t scale = vdupq_n_f32(scale_s); /* Down-conversion F32 -> F16 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const float32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const float *>(input.ptr()); + const auto output_ptr = reinterpret_cast<float16_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const float32x4x4_t texels = { - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr())), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 4), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 8), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 12), scale) - } - }; + { + vmulq_f32(vld1q_f32(input_ptr + x), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 4), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 8), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 12), scale) + } + }; - vst1q_f16(reinterpret_cast<float16_t *>(output.ptr()), vcombine_f16(vcvt_f16_f32(texels.val[0]), vcvt_f16_f32(texels.val[1]))); - vst1q_f16(reinterpret_cast<float16_t *>(output.ptr()) + 8, vcombine_f16(vcvt_f16_f32(texels.val[2]), vcvt_f16_f32(texels.val[3]))); + vst1q_f16(output_ptr + x, vcombine_f16(vcvt_f16_f32(texels.val[0]), vcvt_f16_f32(texels.val[1]))); + vst1q_f16(output_ptr + x + 8, vcombine_f16(vcvt_f16_f32(texels.val[2]), vcvt_f16_f32(texels.val[3]))); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<float16_t>(*(input_ptr + x) * scale_s); + } }, input, output); break; @@ -694,25 +982,39 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ case DataType::S32: { - const float32x4_t scale = vdupq_n_f32(1.f / (1 << _shift)); + const float scale_s = 1.f / (1 << _shift); + const float32x4_t scale = vdupq_n_f32(scale_s); /* Conversion F32 -> S32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const float32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const float *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int32_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const float32x4x4_t texels = { - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr())), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 4), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 8), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 12), scale), - } - }; + { + vmulq_f32(vld1q_f32(input_ptr + x), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 4), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 8), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 12), scale), + } + }; - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()), vcvtq_s32_f32(texels.val[0])); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 4, vcvtq_s32_f32(texels.val[1])); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 8, vcvtq_s32_f32(texels.val[2])); - vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 12, vcvtq_s32_f32(texels.val[3])); + vst1q_s32(output_ptr + x, vcvtq_s32_f32(texels.val[0])); + vst1q_s32(output_ptr + x + 4, vcvtq_s32_f32(texels.val[1])); + vst1q_s32(output_ptr + x + 8, vcvtq_s32_f32(texels.val[2])); + vst1q_s32(output_ptr + x + 12, vcvtq_s32_f32(texels.val[3])); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<int32_t>(*(input_ptr + x) * scale_s); + } }, input, output); break; @@ -720,46 +1022,73 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info case DataType::QASYMM8: case DataType::U8: { - const float32x4_t scale = vdupq_n_f32(1.f / (1 << _shift)); + const float scale_s = 1.f / (1 << _shift); + const float32x4_t scale = vdupq_n_f32(scale_s); /* Down-conversion F32 -> U8 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const float32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const float *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const float32x4x4_t texels = { - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr())), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 4), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 8), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 12), scale), - } - }; + { + vmulq_f32(vld1q_f32(input_ptr + x), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 4), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 8), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 12), scale), + } + }; + + vst1_u8(output_ptr + x, vqmovn_u16(vcombine_u16(vqmovun_s32(vcvtq_s32_f32(texels.val[0])), vqmovun_s32(vcvtq_s32_f32(texels.val[1]))))); + vst1_u8(output_ptr + x + 8, vqmovn_u16(vcombine_u16(vqmovun_s32(vcvtq_s32_f32(texels.val[2])), vqmovun_s32(vcvtq_s32_f32(texels.val[3]))))); + } - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()), vqmovn_u16(vcombine_u16(vqmovun_s32(vcvtq_s32_f32(texels.val[0])), vqmovun_s32(vcvtq_s32_f32(texels.val[1]))))); - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()) + 8, vqmovn_u16(vcombine_u16(vqmovun_s32(vcvtq_s32_f32(texels.val[2])), vqmovun_s32(vcvtq_s32_f32(texels.val[3]))))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = utils::cast::saturate_cast<uint8_t>(*(input_ptr + x) * scale_s); + } }, input, output); break; } case DataType::QASYMM8_SIGNED: { - const float32x4_t scale = vdupq_n_f32(1.f / (1 << _shift)); + const float scale_s = 1.f / (1 << _shift); + const float32x4_t scale = vdupq_n_f32(scale_s); /* Down-conversion F32 -> QASYMM8_SIGNED */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const float32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const float *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const float32x4x4_t texels = { - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr())), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 4), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 8), scale), - vmulq_f32(vld1q_f32(reinterpret_cast<float *>(input.ptr()) + 12), scale), - } - }; + { + vmulq_f32(vld1q_f32(input_ptr + x), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 4), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 8), scale), + vmulq_f32(vld1q_f32(input_ptr + x + 12), scale), + } + }; - vst1_s8(reinterpret_cast<int8_t *>(output.ptr()), vqmovn_s16(vcombine_s16(vqmovn_s32(vcvtq_s32_f32(texels.val[0])), vqmovn_s32(vcvtq_s32_f32(texels.val[1]))))); - vst1_s8(reinterpret_cast<int8_t *>(output.ptr()) + 8, vqmovn_s16(vcombine_s16(vqmovn_s32(vcvtq_s32_f32(texels.val[2])), vqmovn_s32(vcvtq_s32_f32(texels.val[3]))))); + vst1_s8(output_ptr + x, vqmovn_s16(vcombine_s16(vqmovn_s32(vcvtq_s32_f32(texels.val[0])), vqmovn_s32(vcvtq_s32_f32(texels.val[1]))))); + vst1_s8(output_ptr + x + 8, vqmovn_s16(vcombine_s16(vqmovn_s32(vcvtq_s32_f32(texels.val[2])), vqmovn_s32(vcvtq_s32_f32(texels.val[3]))))); + } + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = utils::cast::saturate_cast<int8_t>(*(input_ptr + x) * scale_s); + } }, input, output); break; @@ -776,23 +1105,37 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC case DataType::F16: { - const float32x4_t scale = vdupq_n_f32(1.f / (1 << _shift)); + const float scale_s = 1.f / (1 << _shift); + const float32x4_t scale = vdupq_n_f32(scale_s); /* Down-conversion S32 -> F16 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const float32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const int32_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<float16_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const float32x4x4_t texels = { - vmulq_f32(vcvtq_f32_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()))), scale), - vmulq_f32(vcvtq_f32_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 4)), scale), - vmulq_f32(vcvtq_f32_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 8)), scale), - vmulq_f32(vcvtq_f32_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 12)), scale) - } - }; + { + vmulq_f32(vcvtq_f32_s32(vld1q_s32(input_ptr + x)), scale), + vmulq_f32(vcvtq_f32_s32(vld1q_s32(input_ptr + x + 4)), scale), + vmulq_f32(vcvtq_f32_s32(vld1q_s32(input_ptr + x + 8)), scale), + vmulq_f32(vcvtq_f32_s32(vld1q_s32(input_ptr + x + 12)), scale) + } + }; - vst1q_f16(reinterpret_cast<float16_t *>(output.ptr()), vcombine_f16(vcvt_f16_f32(texels.val[0]), vcvt_f16_f32(texels.val[1]))); - vst1q_f16(reinterpret_cast<float16_t *>(output.ptr()) + 8, vcombine_f16(vcvt_f16_f32(texels.val[2]), vcvt_f16_f32(texels.val[3]))); + vst1q_f16(output_ptr + x, vcombine_f16(vcvt_f16_f32(texels.val[0]), vcvt_f16_f32(texels.val[1]))); + vst1q_f16(output_ptr + x + 8, vcombine_f16(vcvt_f16_f32(texels.val[2]), vcvt_f16_f32(texels.val[3]))); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<int8_t>(*(input_ptr + x) * scale_s); + } }, input, output); break; @@ -800,25 +1143,39 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ case DataType::F32: { - const int32x4_t scale = vdupq_n_s32(1.f / (1 << _shift)); + const int scale_s = 1.f / (1 << _shift); + const int32x4_t scale = vdupq_n_s32(scale_s); /* Conversion S32 -> F32 */ - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const int32_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<float *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int32x4x4_t texels = { - vmulq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr())), scale), - vmulq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 4), scale), - vmulq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 8), scale), - vmulq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 12), scale), - } - }; + { + vmulq_s32(vld1q_s32(input_ptr + x), scale), + vmulq_s32(vld1q_s32(input_ptr + x + 4), scale), + vmulq_s32(vld1q_s32(input_ptr + x + 8), scale), + vmulq_s32(vld1q_s32(input_ptr + x + 12), scale), + } + }; + + vst1q_f32(output_ptr + x, vcvtq_f32_s32(texels.val[0])); + vst1q_f32(output_ptr + x + 4, vcvtq_f32_s32(texels.val[1])); + vst1q_f32(output_ptr + x + 8, vcvtq_f32_s32(texels.val[2])); + vst1q_f32(output_ptr + x + 12, vcvtq_f32_s32(texels.val[3])); + } - vst1q_f32(reinterpret_cast<float *>(output.ptr()), vcvtq_f32_s32(texels.val[0])); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 4, vcvtq_f32_s32(texels.val[1])); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 8, vcvtq_f32_s32(texels.val[2])); - vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 12, vcvtq_f32_s32(texels.val[3])); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<float>(*(input_ptr + x) * scale_s); + } }, input, output); break; @@ -830,38 +1187,64 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info /* Down-conversion S32 -> QASYMM8_SIGNED */ if(ConvertPolicy::SATURATE == _policy) { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const int32_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int32x4x4_t texels = { - vqshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr())), b), - vqshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 4), b), - vqshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 8), b), - vqshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 12), b) - } - }; - vst1_s8(reinterpret_cast<int8_t *>(output.ptr()), vqmovn_s16(vcombine_s16(vqmovn_s32(texels.val[0]), vqmovn_s32(texels.val[1])))); - vst1_s8(reinterpret_cast<int8_t *>(output.ptr()) + 8, vqmovn_s16(vcombine_s16(vqmovn_s32(texels.val[2]), vqmovn_s32(texels.val[3])))); + { + vqshlq_s32(vld1q_s32(input_ptr + x), b), + vqshlq_s32(vld1q_s32(input_ptr + x + 4), b), + vqshlq_s32(vld1q_s32(input_ptr + x + 8), b), + vqshlq_s32(vld1q_s32(input_ptr + x + 12), b) + } + }; + vst1_s8(output_ptr + x, vqmovn_s16(vcombine_s16(vqmovn_s32(texels.val[0]), vqmovn_s32(texels.val[1])))); + vst1_s8(output_ptr + x + 8, vqmovn_s16(vcombine_s16(vqmovn_s32(texels.val[2]), vqmovn_s32(texels.val[3])))); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = utils::cast::saturate_cast<int8_t>(*(input_ptr + x) >> _shift); + } }, input, output); } else { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const int32_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<int8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int32x4x4_t texels = { - vshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr())), b), - vshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 4), b), - vshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 8), b), - vshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 12), b) - } - }; + { + vshlq_s32(vld1q_s32(input_ptr + x), b), + vshlq_s32(vld1q_s32(input_ptr + x + 4), b), + vshlq_s32(vld1q_s32(input_ptr + x + 8), b), + vshlq_s32(vld1q_s32(input_ptr + x + 12), b) + } + }; + + vst1_s8(output_ptr + x, vmovn_s16(vcombine_s16(vmovn_s32(texels.val[0]), vmovn_s32(texels.val[1])))); + vst1_s8(output_ptr + x + 8, vmovn_s16(vcombine_s16(vmovn_s32(texels.val[2]), vmovn_s32(texels.val[3])))); + } - vst1_s8(reinterpret_cast<int8_t *>(output.ptr()), vmovn_s16(vcombine_s16(vmovn_s32(texels.val[0]), vmovn_s32(texels.val[1])))); - vst1_s8(reinterpret_cast<int8_t *>(output.ptr()) + 8, vmovn_s16(vcombine_s16(vmovn_s32(texels.val[2]), vmovn_s32(texels.val[3])))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<int8_t>(*(input_ptr + x) >> _shift); + } }, input, output); } @@ -875,38 +1258,64 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info /* Down-conversion S32 -> U8 */ if(ConvertPolicy::SATURATE == _policy) { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const int32_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int32x4x4_t texels = { - vqshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr())), b), - vqshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 4), b), - vqshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 8), b), - vqshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 12), b) - } - }; - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()), vqmovn_u16(vcombine_u16(vqmovun_s32(texels.val[0]), vqmovun_s32(texels.val[1])))); - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()) + 8, vqmovn_u16(vcombine_u16(vqmovun_s32(texels.val[2]), vqmovun_s32(texels.val[3])))); + { + vqshlq_s32(vld1q_s32(input_ptr + x), b), + vqshlq_s32(vld1q_s32(input_ptr + x + 4), b), + vqshlq_s32(vld1q_s32(input_ptr + x + 8), b), + vqshlq_s32(vld1q_s32(input_ptr + x + 12), b) + } + }; + vst1_u8(output_ptr + x, vqmovn_u16(vcombine_u16(vqmovun_s32(texels.val[0]), vqmovun_s32(texels.val[1])))); + vst1_u8(output_ptr + x + 8, vqmovn_u16(vcombine_u16(vqmovun_s32(texels.val[2]), vqmovun_s32(texels.val[3])))); + } + + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = utils::cast::saturate_cast<uint8_t>(*(input_ptr + x) >> _shift); + } }, input, output); } else { - execute_window_loop(window, [&](const Coordinates &) + execute_window_loop(win, [&](const Coordinates &) { - const int32x4x4_t texels = + const auto input_ptr = reinterpret_cast<const int32_t *>(input.ptr()); + const auto output_ptr = reinterpret_cast<uint8_t *>(output.ptr()); + + int x = window_start_x; + for(; x <= (window_end_x - window_step_x); x += window_step_x) { + const int32x4x4_t texels = { - vshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr())), b), - vshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 4), b), - vshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 8), b), - vshlq_s32(vld1q_s32(reinterpret_cast<int32_t *>(input.ptr()) + 12), b) - } - }; + { + vshlq_s32(vld1q_s32(input_ptr + x), b), + vshlq_s32(vld1q_s32(input_ptr + x + 4), b), + vshlq_s32(vld1q_s32(input_ptr + x + 8), b), + vshlq_s32(vld1q_s32(input_ptr + x + 12), b) + } + }; + + vst1_u8(output_ptr + x, vmovn_u16(vcombine_u16(vmovn_u32(vreinterpretq_u32_s32(texels.val[0])), vmovn_u32(vreinterpretq_u32_s32(texels.val[1]))))); + vst1_u8(output_ptr + x + 8, vmovn_u16(vcombine_u16(vmovn_u32(vreinterpretq_u32_s32(texels.val[2])), vmovn_u32(vreinterpretq_u32_s32(texels.val[3]))))); + } - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()), vmovn_u16(vcombine_u16(vmovn_u32(vreinterpretq_u32_s32(texels.val[0])), vmovn_u32(vreinterpretq_u32_s32(texels.val[1]))))); - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()) + 8, vmovn_u16(vcombine_u16(vmovn_u32(vreinterpretq_u32_s32(texels.val[2])), vmovn_u32(vreinterpretq_u32_s32(texels.val[3]))))); + // Compute left-over elements + for(; x < window_end_x; ++x) + { + *(output_ptr + x) = static_cast<uint8_t>(*(input_ptr + x) >> _shift); + } }, input, output); } |