| // Copyright (c) Facebook, Inc. and its affiliates. |
| // All rights reserved. |
| // |
| // Copyright 2019 Google LLC |
| // |
| // This source code is licensed under the BSD-style license found in the |
| // LICENSE file in the root directory of this source tree. |
| |
| #include <cmath> |
| #include <cstddef> |
| #include <cstdlib> |
| |
| #include <gtest/gtest.h> |
| |
| #include <xnnpack/common.h> |
| |
| #include <xnnpack/requantization-stubs.h> |
| #include "requantization-tester.h" |
| |
| |
| /* |
| * Precise scalar implementation using unsigned 32-bit arithmetics. |
| */ |
| |
| TEST(PRECISE__SCALAR_UNSIGNED32, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__scalar_unsigned32); |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED32, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__scalar_unsigned32); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED32, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_precise__scalar_unsigned32); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED32, divide_by_po2_with_rounding_down) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingDown(xnn_requantize_precise__scalar_unsigned32); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED32, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_precise__scalar_unsigned32); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED32, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_precise__scalar_unsigned32); |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED32, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesPrecise(xnn_requantize_precise__scalar_unsigned32); |
| } |
| |
| |
| /* |
| * Precise scalar implementation using unsigned 64-bit arithmetics. |
| */ |
| |
| TEST(PRECISE__SCALAR_UNSIGNED64, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__scalar_unsigned64); |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED64, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__scalar_unsigned64); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED64, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_precise__scalar_unsigned64); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED64, divide_by_po2_with_rounding_down) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingDown(xnn_requantize_precise__scalar_unsigned64); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED64, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_precise__scalar_unsigned64); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED64, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_precise__scalar_unsigned64); |
| } |
| |
| TEST(PRECISE__SCALAR_UNSIGNED64, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesPrecise(xnn_requantize_precise__scalar_unsigned64); |
| } |
| |
| |
| /* |
| * Precise scalar implementation using signed 64-bit arithmetics. |
| */ |
| |
| TEST(PRECISE__SCALAR_SIGNED64, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__scalar_signed64); |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_SIGNED64, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__scalar_signed64); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_SIGNED64, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_precise__scalar_signed64); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_SIGNED64, divide_by_po2_with_rounding_down) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingDown(xnn_requantize_precise__scalar_signed64); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_SIGNED64, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_precise__scalar_signed64); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SCALAR_SIGNED64, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_precise__scalar_signed64); |
| } |
| |
| TEST(PRECISE__SCALAR_SIGNED64, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesPrecise(xnn_requantize_precise__scalar_signed64); |
| } |
| |
| |
| /* |
| * FP32-based scalar implementation using lrintf function. |
| */ |
| |
| TEST(FP32__SCALAR_LRINTF, random_cases) { |
| RequantizationTester() |
| .iterations(1000) |
| .TestRandomCasesApproximate(xnn_requantize_fp32__scalar_lrintf); |
| } |
| |
| |
| /* |
| * FP32-based scalar implementation using magic trick for FP32->INT32 conversion. |
| */ |
| |
| TEST(FP32__SCALAR_MAGIC, random_cases) { |
| RequantizationTester() |
| .iterations(1000) |
| .TestRandomCasesApproximate(xnn_requantize_fp32__scalar_magic); |
| } |
| |
| |
| /* |
| * Q31-based scalar implementation. |
| */ |
| |
| TEST(Q31__SCALAR, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__scalar); |
| } |
| } |
| |
| TEST(Q31__SCALAR, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__scalar); |
| } |
| } |
| } |
| |
| TEST(Q31__SCALAR, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_q31__scalar); |
| } |
| } |
| } |
| |
| /* No rounding down Test - it fails because of upward bias in multiplication */ |
| |
| TEST(Q31__SCALAR, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_q31__scalar); |
| } |
| } |
| } |
| |
| TEST(Q31__SCALAR, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_q31__scalar); |
| } |
| |
| TEST(Q31__SCALAR, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_q31__scalar); |
| } |
| |
| TEST(Q31__SCALAR, random_match_gemmlowp) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesAgainstReference(xnn_requantize_q31__scalar, xnn_requantize_gemmlowp__scalar); |
| } |
| |
| |
| /* |
| * Scalar implementation from gemmlowp. |
| */ |
| |
| TEST(GEMMLOWP__SCALAR, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_gemmlowp__scalar); |
| } |
| |
| |
| #if !XNN_ARCH_ASMJS && !XNN_ARCH_WASM && !XNN_COMPILER_MSVC && !XNN_COMPILER_ICC |
| /* |
| * Precise PSIMD implementation using unsigned 32-bit arithmetics. |
| */ |
| |
| TEST(PRECISE__PSIMD, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__psimd); |
| } |
| } |
| |
| TEST(PRECISE__PSIMD, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__psimd); |
| } |
| } |
| } |
| |
| TEST(PRECISE__PSIMD, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_precise__psimd); |
| } |
| } |
| } |
| |
| TEST(PRECISE__PSIMD, divide_by_po2_with_rounding_down) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingDown(xnn_requantize_precise__psimd); |
| } |
| } |
| } |
| |
| TEST(PRECISE__PSIMD, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_precise__psimd); |
| } |
| } |
| } |
| |
| TEST(PRECISE__PSIMD, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_precise__psimd); |
| } |
| |
| TEST(PRECISE__PSIMD, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesPrecise(xnn_requantize_precise__psimd); |
| } |
| |
| |
| /* |
| * FP32-based PSIMD implementation using magic trick for FP32->INT32 conversion. |
| */ |
| |
| TEST(FP32__PSIMD, random_cases) { |
| RequantizationTester() |
| .iterations(1000) |
| .TestRandomCasesApproximate(xnn_requantize_fp32__psimd); |
| } |
| #endif // !XNN_ARCH_ASMJS && !XNN_ARCH_WASM && !XNN_COMPILER_MSVC && !XNN_COMPILER_ICC |
| |
| |
| #if XNN_ARCH_X86 || XNN_ARCH_X86_64 |
| /* |
| * Precise SSE2 implementation using floating-point shuffle. |
| */ |
| |
| TEST(PRECISE__SSE2, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__sse2); |
| } |
| } |
| |
| TEST(PRECISE__SSE2, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__sse2); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSE2, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_precise__sse2); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSE2, divide_by_po2_with_rounding_down) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingDown(xnn_requantize_precise__sse2); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSE2, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_precise__sse2); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSE2, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_precise__sse2); |
| } |
| |
| TEST(PRECISE__SSE2, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesPrecise(xnn_requantize_precise__sse2); |
| } |
| |
| |
| /* |
| * Precise SSSE3 implementation using floating-point shuffle. |
| */ |
| |
| TEST(PRECISE__SSSE3, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__ssse3); |
| } |
| } |
| |
| TEST(PRECISE__SSSE3, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__ssse3); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSSE3, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_precise__ssse3); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSSE3, divide_by_po2_with_rounding_down) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingDown(xnn_requantize_precise__ssse3); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSSE3, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_precise__ssse3); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSSE3, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_precise__ssse3); |
| } |
| |
| TEST(PRECISE__SSSE3, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesPrecise(xnn_requantize_precise__ssse3); |
| } |
| |
| |
| /* |
| * Precise SSE4.1 implementation using static blend instruction. |
| */ |
| |
| TEST(PRECISE__SSE4, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__sse4); |
| } |
| } |
| |
| TEST(PRECISE__SSE4, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__sse4); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSE4, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_precise__sse4); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSE4, divide_by_po2_with_rounding_down) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingDown(xnn_requantize_precise__sse4); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSE4, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_precise__sse4); |
| } |
| } |
| } |
| |
| TEST(PRECISE__SSE4, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_precise__sse4); |
| } |
| |
| TEST(PRECISE__SSE4, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesPrecise(xnn_requantize_precise__sse4); |
| } |
| |
| |
| /* |
| * FP32-based x86 SSE2 implementation. |
| */ |
| |
| TEST(FP32__SSE2, random_cases) { |
| RequantizationTester() |
| .iterations(1000) |
| .TestRandomCasesApproximate(xnn_requantize_fp32__sse2); |
| } |
| |
| |
| /* |
| * Q31-based x86 SSE2 implementation. |
| */ |
| |
| TEST(Q31__SSE2, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__sse2); |
| } |
| } |
| |
| TEST(Q31__SSE2, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__sse2); |
| } |
| } |
| } |
| |
| TEST(Q31__SSE2, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_q31__sse2); |
| } |
| } |
| } |
| |
| /* No rounding down Test - it fails because of upward bias in multiplication */ |
| |
| TEST(Q31__SSE2, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_q31__sse2); |
| } |
| } |
| } |
| |
| TEST(Q31__SSE2, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_q31__sse2); |
| } |
| |
| TEST(Q31__SSE2, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_q31__sse2); |
| } |
| |
| TEST(Q31__SSE2, random_match_gemmlowp) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesAgainstReference(xnn_requantize_q31__sse2, xnn_requantize_gemmlowp__sse2); |
| } |
| |
| |
| /* |
| * Q31-based x86 SSSE3 implementation. |
| */ |
| |
| TEST(Q31__SSSE3, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__ssse3); |
| } |
| } |
| |
| TEST(Q31__SSSE3, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__ssse3); |
| } |
| } |
| } |
| |
| TEST(Q31__SSSE3, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_q31__ssse3); |
| } |
| } |
| } |
| |
| /* No rounding down Test - it fails because of upward bias in multiplication */ |
| |
| TEST(Q31__SSSE3, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_q31__ssse3); |
| } |
| } |
| } |
| |
| TEST(Q31__SSSE3, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_q31__ssse3); |
| } |
| |
| TEST(Q31__SSSE3, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_q31__ssse3); |
| } |
| |
| TEST(Q31__SSSE3, random_match_gemmlowp) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesAgainstReference(xnn_requantize_q31__ssse3, xnn_requantize_gemmlowp__ssse3); |
| } |
| |
| |
| /* |
| * Q31-based x86 SSE4 implementation. |
| */ |
| |
| TEST(Q31__SSE4, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__sse4); |
| } |
| } |
| |
| TEST(Q31__SSE4, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__sse4); |
| } |
| } |
| } |
| |
| TEST(Q31__SSE4, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_q31__sse4); |
| } |
| } |
| } |
| |
| /* No rounding down Test - it fails because of upward bias in multiplication */ |
| |
| TEST(Q31__SSE4, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_q31__sse4); |
| } |
| } |
| } |
| |
| TEST(Q31__SSE4, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_q31__sse4); |
| } |
| |
| TEST(Q31__SSE4, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_q31__sse4); |
| } |
| |
| TEST(Q31__SSE4, random_match_gemmlowp) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesAgainstReference(xnn_requantize_q31__sse4, xnn_requantize_gemmlowp__sse4); |
| } |
| |
| |
| /* |
| * x86 SSE2 implementation from gemmlowp. |
| */ |
| |
| TEST(GEMMLOWP__SSE2, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_gemmlowp__sse2); |
| } |
| } |
| |
| TEST(GEMMLOWP__SSE2, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_gemmlowp__sse2); |
| } |
| } |
| } |
| |
| TEST(GEMMLOWP__SSE2, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_gemmlowp__sse2); |
| } |
| } |
| } |
| |
| /* No rounding down Test - it fails because of upward bias in multiplication */ |
| |
| TEST(GEMMLOWP__SSE2, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_gemmlowp__sse2); |
| } |
| } |
| } |
| |
| TEST(GEMMLOWP__SSE2, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_gemmlowp__sse2); |
| } |
| |
| TEST(GEMMLOWP__SSE2, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_gemmlowp__sse2); |
| } |
| |
| |
| /* |
| * x86 SSSE3 implementation from gemmlowp. |
| */ |
| |
| TEST(GEMMLOWP__SSSE3, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_gemmlowp__ssse3); |
| } |
| } |
| |
| TEST(GEMMLOWP__SSSE3, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_gemmlowp__ssse3); |
| } |
| } |
| } |
| |
| TEST(GEMMLOWP__SSSE3, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_gemmlowp__ssse3); |
| } |
| } |
| } |
| |
| /* No rounding down Test - it fails because of upward bias in multiplication */ |
| |
| TEST(GEMMLOWP__SSSE3, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_gemmlowp__ssse3); |
| } |
| } |
| } |
| |
| TEST(GEMMLOWP__SSSE3, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_gemmlowp__ssse3); |
| } |
| |
| TEST(GEMMLOWP__SSSE3, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_gemmlowp__ssse3); |
| } |
| |
| |
| /* |
| * x86 SSE4 implementation from gemmlowp. |
| */ |
| |
| TEST(GEMMLOWP__SSE4, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_gemmlowp__sse4); |
| } |
| } |
| |
| TEST(GEMMLOWP__SSE4, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_gemmlowp__sse4); |
| } |
| } |
| } |
| |
| TEST(GEMMLOWP__SSE4, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_gemmlowp__sse4); |
| } |
| } |
| } |
| |
| /* No rounding down Test - it fails because of upward bias in multiplication */ |
| |
| TEST(GEMMLOWP__SSE4, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_gemmlowp__sse4); |
| } |
| } |
| } |
| |
| TEST(GEMMLOWP__SSE4, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_gemmlowp__sse4); |
| } |
| |
| TEST(GEMMLOWP__SSE4, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_gemmlowp__sse4); |
| } |
| #endif // XNN_ARCH_X86 || XNN_ARCH_X86_64 |
| |
| #if XNN_ARCH_ARM || XNN_ARCH_ARM64 |
| /* |
| * Precise ARM NEON implementation. |
| */ |
| |
| TEST(PRECISE__NEON, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__neon); |
| } |
| } |
| |
| TEST(PRECISE__NEON, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_precise__neon); |
| } |
| } |
| } |
| |
| TEST(PRECISE__NEON, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_precise__neon); |
| } |
| } |
| } |
| |
| TEST(PRECISE__NEON, divide_by_po2_with_rounding_down) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingDown(xnn_requantize_precise__neon); |
| } |
| } |
| } |
| |
| TEST(PRECISE__NEON, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_precise__neon); |
| } |
| } |
| } |
| |
| TEST(PRECISE__NEON, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_precise__neon); |
| } |
| |
| TEST(PRECISE__NEON, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesPrecise(xnn_requantize_precise__neon); |
| } |
| |
| |
| /* |
| * FP32-based ARM NEON implementation. |
| */ |
| |
| TEST(FP32__NEON, random_cases) { |
| RequantizationTester() |
| .iterations(1000) |
| .TestRandomCasesApproximate(xnn_requantize_fp32__neon); |
| } |
| |
| |
| /* |
| * Q31-based ARM NEON implementation. |
| */ |
| |
| TEST(Q31__NEON, exact_divide_by_po2) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__neon); |
| } |
| } |
| |
| TEST(Q31__NEON, exact_divide_by_po2_with_zero_point) { |
| for (int32_t zero_point = 1; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestExactDivideByPO2(xnn_requantize_q31__neon); |
| } |
| } |
| } |
| |
| TEST(Q31__NEON, divide_by_po2_with_rounding_up) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingUp(xnn_requantize_q31__neon); |
| } |
| } |
| } |
| |
| /* No rounding down Test - it fails because of upward bias in multiplication */ |
| |
| TEST(Q31__NEON, divide_by_po2_with_rounding_away) { |
| for (int32_t zero_point = 0; zero_point < 256; zero_point++) { |
| for (uint32_t s = 1; s < 32; s++) { |
| RequantizationTester() |
| .zero_point(zero_point) |
| .s(s) |
| .TestDivideByPO2WithRoundingAway(xnn_requantize_q31__neon); |
| } |
| } |
| } |
| |
| TEST(Q31__NEON, special_cases) { |
| RequantizationTester() |
| .TestSpecialCases(xnn_requantize_q31__neon); |
| } |
| |
| TEST(Q31__NEON, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_q31__neon); |
| } |
| |
| TEST(Q31__NEON, random_match_gemmlowp) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesAgainstReference(xnn_requantize_q31__neon, xnn_requantize_gemmlowp__neon); |
| } |
| |
| |
| /* |
| * ARM NEON implementation from gemmlowp. |
| */ |
| |
| TEST(GEMMLOWP__NEON, random_cases) { |
| RequantizationTester() |
| .iterations(100) |
| .TestRandomCasesApproximate(xnn_requantize_gemmlowp__neon); |
| } |
| #endif // XNN_ARCH_ARM || XNN_ARCH_ARM64 |
| |
| #if XNN_ARCH_WASMSIMD |
| /* |
| * FP32-based ARM NEON implementation. |
| */ |
| |
| TEST(FP32__WASMSIMD, random_cases) { |
| RequantizationTester() |
| .iterations(1000) |
| .TestRandomCasesApproximate(xnn_requantize_fp32__wasmsimd); |
| } |
| #endif // XNN_ARCH_WASMSIMD |