diff options
author | Gunes Bayir <gunes.bayir@arm.com> | 2022-09-11 15:59:19 +0100 |
---|---|---|
committer | Gunes Bayir <gunes.bayir@arm.com> | 2022-09-16 12:38:59 +0000 |
commit | c4f2743951473f8d97f5a43767fdbb31a4df967c (patch) | |
tree | 26c49b1af8113bb169931f3af5e502904d455a25 /tests/validation/reference/Scale.cpp | |
parent | 0d05b6690fe69c57f63ca43d59b551f074613062 (diff) | |
download | ComputeLibrary-c4f2743951473f8d97f5a43767fdbb31a4df967c.tar.gz |
Optimize Quantized/Integer Bilinear Scale for Neon™
This patch introduces several performance optimizations regarding the Bilinear Scale operator with REPLICATE Border mode. Changes apply only to NHWC.
This patch
- Reduces the memory footprint by disabling precomputation of indices and weights when they're not used
- Rewrites the kernels for QASYMM8/QASYMM8_SIGNED/U8(Uint8)
- Adds S8(Int8) Bilinear Scale for Border mode REPLICATE
- Removes Bilinear Scale SVE kernels for Quantized and Integer types and adjust the heuristics to choose the Neon™ implementation
- Adds new test cases where the input and output of the Bilinear Scale operator have different quantization scale and offset
Resolves: COMPMID-5453, COMPMID-5454
Change-Id: I3d251e76e0c6978fd5a0a1795ec62ab536bec93c
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8250
Reviewed-by: SiCong Li <sicong.li@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'tests/validation/reference/Scale.cpp')
-rw-r--r-- | tests/validation/reference/Scale.cpp | 20 |
1 files changed, 10 insertions, 10 deletions
diff --git a/tests/validation/reference/Scale.cpp b/tests/validation/reference/Scale.cpp index 71e98fd776..2f429cb29b 100644 --- a/tests/validation/reference/Scale.cpp +++ b/tests/validation/reference/Scale.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2020 Arm Limited. + * Copyright (c) 2017-2020, 2022 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -25,7 +25,6 @@ #include "Scale.h" #include "Utils.h" -#include "arm_compute/core/utils/misc/Utility.h" #include "src/core/utils/ScaleUtils.h" #include "support/Rounding.h" @@ -183,14 +182,15 @@ SimpleTensor<T> scale_core(const SimpleTensor<T> &in, float scale_x, float scale template <typename T> SimpleTensor<T> scale(const SimpleTensor<T> &src, float scale_x, float scale_y, InterpolationPolicy policy, BorderMode border_mode, T constant_border_value, - SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners) + SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners, QuantizationInfo output_quantization_info) { + ARM_COMPUTE_UNUSED(output_quantization_info); return scale_core<T>(src, scale_x, scale_y, policy, border_mode, constant_border_value, sampling_policy, ceil_policy_scale, align_corners); } template <> SimpleTensor<uint8_t> scale(const SimpleTensor<uint8_t> &src, float scale_x, float scale_y, InterpolationPolicy policy, BorderMode border_mode, uint8_t constant_border_value, - SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners) + SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners, QuantizationInfo output_quantization_info) { SimpleTensor<uint8_t> dst; if(src.quantization_info().uniform().scale != 0.f) @@ -198,7 +198,7 @@ SimpleTensor<uint8_t> scale(const SimpleTensor<uint8_t> &src, float scale_x, flo SimpleTensor<float> src_tmp = convert_from_asymmetric(src); float constant_border_value_f = dequantize_qasymm8(constant_border_value, src.quantization_info()); SimpleTensor<float> dst_tmp = scale_core<float>(src_tmp, scale_x, scale_y, policy, border_mode, constant_border_value_f, sampling_policy, ceil_policy_scale, align_corners); - dst = convert_to_asymmetric<uint8_t>(dst_tmp, src.quantization_info()); + dst = convert_to_asymmetric<uint8_t>(dst_tmp, output_quantization_info); } else { @@ -209,7 +209,7 @@ SimpleTensor<uint8_t> scale(const SimpleTensor<uint8_t> &src, float scale_x, flo template <> SimpleTensor<int8_t> scale(const SimpleTensor<int8_t> &src, float scale_x, float scale_y, InterpolationPolicy policy, BorderMode border_mode, int8_t constant_border_value, - SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners) + SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners, QuantizationInfo output_quantization_info) { SimpleTensor<int8_t> dst; if(src.quantization_info().uniform().scale != 0.f) @@ -217,7 +217,7 @@ SimpleTensor<int8_t> scale(const SimpleTensor<int8_t> &src, float scale_x, float SimpleTensor<float> src_tmp = convert_from_asymmetric(src); float constant_border_value_f = dequantize_qasymm8_signed(constant_border_value, src.quantization_info()); SimpleTensor<float> dst_tmp = scale_core<float>(src_tmp, scale_x, scale_y, policy, border_mode, constant_border_value_f, sampling_policy, ceil_policy_scale, align_corners); - dst = convert_to_asymmetric<int8_t>(dst_tmp, src.quantization_info()); + dst = convert_to_asymmetric<int8_t>(dst_tmp, output_quantization_info); } else { @@ -227,11 +227,11 @@ SimpleTensor<int8_t> scale(const SimpleTensor<int8_t> &src, float scale_x, float } template SimpleTensor<int16_t> scale(const SimpleTensor<int16_t> &src, float scale_x, float scale_y, InterpolationPolicy policy, BorderMode border_mode, int16_t constant_border_value, - SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners); + SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners, QuantizationInfo output_quantization_info); template SimpleTensor<half> scale(const SimpleTensor<half> &src, float scale_x, float scale_y, InterpolationPolicy policy, BorderMode border_mode, half constant_border_value, - SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners); + SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners, QuantizationInfo output_quantization_info); template SimpleTensor<float> scale(const SimpleTensor<float> &src, float scale_x, float scale_y, InterpolationPolicy policy, BorderMode border_mode, float constant_border_value, - SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners); + SamplingPolicy sampling_policy, bool ceil_policy_scale, bool align_corners, QuantizationInfo output_quantization_info); } // namespace reference } // namespace validation } // namespace test |