From 62d600fe8c0afba81cc5f5dd315eb6dcc04f90b8 Mon Sep 17 00:00:00 2001 From: Radu Salavat Date: Mon, 15 Apr 2024 14:42:07 +0000 Subject: Move s32 to f32 conversion in reference layers from quantization to dequantization Signed-off-by: Radu Salavat Change-Id: Ib17946b526d35deeca94b5d2f163b92101e313c4 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11420 Benchmark: Arm Jenkins Tested-by: Arm Jenkins Reviewed-by: Gunes Bayir Comments-Addressed: Arm Jenkins --- tests/validation/fixtures/GEMMLowpFixture.h | 12 +++++++----- tests/validation/reference/DequantizationLayer.cpp | 9 ++++++++- tests/validation/reference/QuantizationLayer.cpp | 10 ---------- 3 files changed, 15 insertions(+), 16 deletions(-) (limited to 'tests') diff --git a/tests/validation/fixtures/GEMMLowpFixture.h b/tests/validation/fixtures/GEMMLowpFixture.h index 11a491faa7..6b7cbba92e 100644 --- a/tests/validation/fixtures/GEMMLowpFixture.h +++ b/tests/validation/fixtures/GEMMLowpFixture.h @@ -31,7 +31,7 @@ #include "tests/validation/Validation.h" #include "tests/validation/reference/GEMMLowp.h" #include "tests/validation/reference/ArithmeticOperations.h" -#include "tests/validation/reference/QuantizationLayer.h" +#include "tests/validation/reference/DequantizationLayer.h" #include #include @@ -485,7 +485,7 @@ public: const auto b_qinfo = QuantizationInfo(5.0f / 255, b_offset); TensorFillInfo finfo; _target = compute_target(shape_a, shape_b, shape_output, a_qinfo, b_qinfo, finfo, accumulate, dynamic_qinfo); - _reference = compute_reference(shape_a, shape_b, shape_output, a_qinfo, b_qinfo, finfo, accumulate); + _reference = compute_reference(shape_a, shape_b, shape_output, a_qinfo, b_qinfo, finfo, accumulate, dynamic_qinfo); } protected: @@ -495,14 +495,16 @@ protected: return compute_gemmlowp_target(shape_a, shape_b, shape_output, a_qinfo, b_qinfo, output_qinfo, DataType::QASYMM8_SIGNED, DataType::QASYMM8_SIGNED, GEMMLowpOutputStageInfo(), false, finfo, accumulate, dynamic_qinfo, DataType::F32); } - SimpleTensor compute_reference(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &shape_output, const QuantizationInfo& a_qinfo, const QuantizationInfo& b_qinfo, const TensorFillInfo& finfo, bool accumulate) + SimpleTensor compute_reference(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &shape_output, const QuantizationInfo& a_qinfo, const QuantizationInfo& b_qinfo, const TensorFillInfo& finfo, bool accumulate, const bool dynamic_qinfo) { + QuantizationInfo s32_ref_output_quant_info = QuantizationInfo(a_qinfo.uniform().scale * b_qinfo.uniform().scale, 0, dynamic_qinfo); + SimpleTensor s32_ref_output = compute_gemmlowp_reference(shape_a, shape_b, shape_output, a_qinfo, b_qinfo, DataType::QASYMM8_SIGNED, DataType::QASYMM8_SIGNED, finfo); + s32_ref_output.quantization_info(s32_ref_output_quant_info); SimpleTensor f32_ref_output(s32_ref_output.shape(), DataType::F32); - QuantizationInfo dst_quant_info = QuantizationInfo(a_qinfo.uniform().scale * b_qinfo.uniform().scale, 0); - f32_ref_output = reference::quantization_layer(s32_ref_output, DataType::F32, dst_quant_info); + f32_ref_output = reference::dequantization_layer(s32_ref_output); if (accumulate) { diff --git a/tests/validation/reference/DequantizationLayer.cpp b/tests/validation/reference/DequantizationLayer.cpp index 64a89aa6a0..67d69c2c38 100644 --- a/tests/validation/reference/DequantizationLayer.cpp +++ b/tests/validation/reference/DequantizationLayer.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2020 Arm Limited. + * Copyright (c) 2017-2020, 2024 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -59,6 +59,12 @@ TOut dequantize(int16_t val, const UniformQuantizationInfo qinfo, DataType dt) ARM_COMPUTE_UNUSED(dt); return static_cast(dequantize_qsymm16(val, qinfo)); } +template +TOut dequantize(int32_t val, const UniformQuantizationInfo qinfo, DataType dt) +{ + ARM_COMPUTE_UNUSED(dt); + return static_cast(dequantize_s32(val, qinfo)); +} } // namespace template SimpleTensor dequantization_layer(const SimpleTensor &src) @@ -115,6 +121,7 @@ template SimpleTensor dequantization_layer(const SimpleTensor &src template SimpleTensor dequantization_layer(const SimpleTensor &src); template SimpleTensor dequantization_layer(const SimpleTensor &src); template SimpleTensor dequantization_layer(const SimpleTensor &src); +template SimpleTensor dequantization_layer(const SimpleTensor &src); } // namespace reference } // namespace validation } // namespace test diff --git a/tests/validation/reference/QuantizationLayer.cpp b/tests/validation/reference/QuantizationLayer.cpp index b76263bf95..ad7ba7ac43 100644 --- a/tests/validation/reference/QuantizationLayer.cpp +++ b/tests/validation/reference/QuantizationLayer.cpp @@ -80,15 +80,6 @@ SimpleTensor quantization_layer(const SimpleTensor &src, DataType out dst[i] = quantize_qasymm16((src[i]), qinfo, rounding_policy); } break; - case DataType::F32: -#if defined(_OPENMP) - #pragma omp parallel for -#endif /* _OPENMP */ - for(int i = 0; i < src.num_elements(); ++i) - { - dst[i] = dequantize_s32((src[i]), qinfo); - } - break; default: ARM_COMPUTE_ERROR("Unsupported output data type"); } @@ -136,7 +127,6 @@ template SimpleTensor quantization_layer(const SimpleTensor &src, template SimpleTensor quantization_layer(const SimpleTensor &src, DataType output_data_type, const QuantizationInfo &quantization_info); template SimpleTensor quantization_layer(const SimpleTensor &src, DataType output_data_type, const QuantizationInfo &quantization_info); template SimpleTensor quantization_layer(const SimpleTensor &src, DataType output_data_type, const QuantizationInfo &quantization_info); -template SimpleTensor quantization_layer(const SimpleTensor &src, DataType output_data_type, const QuantizationInfo &quantization_info); } // namespace reference } // namespace validation } // namespace test -- cgit v1.2.1