| // Auto-generated file. Do not edit! |
| // Template: src/qs8-dwconv/unipass-avx2-mul16.c.in |
| // Generator: tools/xngen |
| // |
| // Copyright 2020 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 <assert.h> |
| |
| #include <immintrin.h> |
| |
| #include <xnnpack/dwconv.h> |
| |
| |
| void xnn_qs8_dwconv_minmax_gemmlowp_ukernel_up32x9__avx2_mul16( |
| size_t channels, |
| size_t output_width, |
| const int8_t** input, |
| const void* weights, |
| int8_t* output, |
| size_t input_stride, |
| size_t output_increment, |
| size_t input_offset, |
| const int8_t* zero, |
| const union xnn_qs8_gemm_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN |
| { |
| assert(channels != 0); |
| assert(output_width != 0); |
| |
| do { |
| const int8_t* i0 = input[0]; |
| assert(i0 != NULL); |
| if XNN_UNPREDICTABLE(i0 != zero) { |
| i0 = (const int8_t*) ((uintptr_t) i0 + input_offset); |
| } |
| const int8_t* i1 = input[1]; |
| assert(i1 != NULL); |
| if XNN_UNPREDICTABLE(i1 != zero) { |
| i1 = (const int8_t*) ((uintptr_t) i1 + input_offset); |
| } |
| const int8_t* i2 = input[2]; |
| assert(i2 != NULL); |
| if XNN_UNPREDICTABLE(i2 != zero) { |
| i2 = (const int8_t*) ((uintptr_t) i2 + input_offset); |
| } |
| const int8_t* i3 = input[3]; |
| assert(i3 != NULL); |
| if XNN_UNPREDICTABLE(i3 != zero) { |
| i3 = (const int8_t*) ((uintptr_t) i3 + input_offset); |
| } |
| const int8_t* i4 = input[4]; |
| assert(i4 != NULL); |
| if XNN_UNPREDICTABLE(i4 != zero) { |
| i4 = (const int8_t*) ((uintptr_t) i4 + input_offset); |
| } |
| const int8_t* i5 = input[5]; |
| assert(i5 != NULL); |
| if XNN_UNPREDICTABLE(i5 != zero) { |
| i5 = (const int8_t*) ((uintptr_t) i5 + input_offset); |
| } |
| const int8_t* i6 = input[6]; |
| assert(i6 != NULL); |
| if XNN_UNPREDICTABLE(i6 != zero) { |
| i6 = (const int8_t*) ((uintptr_t) i6 + input_offset); |
| } |
| const int8_t* i7 = input[7]; |
| assert(i7 != NULL); |
| if XNN_UNPREDICTABLE(i7 != zero) { |
| i7 = (const int8_t*) ((uintptr_t) i7 + input_offset); |
| } |
| const int8_t* i8 = input[8]; |
| assert(i8 != NULL); |
| if XNN_UNPREDICTABLE(i8 != zero) { |
| i8 = (const int8_t*) ((uintptr_t) i8 + input_offset); |
| } |
| input = (const int8_t**) ((uintptr_t) input + input_stride); |
| |
| size_t c = channels; |
| const void* w = weights; |
| for (; c >= 32; c -= 32) { |
| __m256i vacc01234567 = _mm256_loadu_si256((const __m256i*) w); |
| __m256i vacc89ABCDEF = _mm256_loadu_si256((const __m256i*) ((uintptr_t) w + 8 * sizeof(int32_t))); |
| __m256i vaccGHIJKLMN = _mm256_loadu_si256((const __m256i*) ((uintptr_t) w + 16 * sizeof(int32_t))); |
| __m256i vaccOPQRSTUV = _mm256_loadu_si256((const __m256i*) ((uintptr_t) w + 24 * sizeof(int32_t))); |
| |
| |
| const __m256i vi0x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i0)); |
| const __m256i vk0x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 0 * sizeof(int8_t)))); |
| const __m256i vi0xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (i0 + 16))); |
| const __m256i vk0xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 16 * sizeof(int8_t)))); |
| i0 += 32; |
| |
| const __m256i vprod0x0123456789ABCDEF = _mm256_mullo_epi16(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF); |
| const __m128i vprod0x89ABCDEF = _mm256_extracti128_si256(vprod0x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod0x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod0x89ABCDEF)); |
| const __m256i vprod0xGHIJKLMNOPQRSTUV = _mm256_mullo_epi16(vi0xGHIJKLMNOPQRSTUV, vk0xGHIJKLMNOPQRSTUV); |
| const __m128i vprod0xOPQRSTUV = _mm256_extracti128_si256(vprod0xGHIJKLMNOPQRSTUV, 1); |
| vaccGHIJKLMN = _mm256_add_epi32(vaccGHIJKLMN, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod0xGHIJKLMNOPQRSTUV))); |
| vaccOPQRSTUV = _mm256_add_epi32(vaccOPQRSTUV, _mm256_cvtepi16_epi32(vprod0xOPQRSTUV)); |
| |
| const __m256i vi1x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i1)); |
| const __m256i vk1x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 32 * sizeof(int8_t)))); |
| const __m256i vi1xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (i1 + 16))); |
| const __m256i vk1xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 48 * sizeof(int8_t)))); |
| i1 += 32; |
| |
| const __m256i vprod1x0123456789ABCDEF = _mm256_mullo_epi16(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF); |
| const __m128i vprod1x89ABCDEF = _mm256_extracti128_si256(vprod1x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod1x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod1x89ABCDEF)); |
| const __m256i vprod1xGHIJKLMNOPQRSTUV = _mm256_mullo_epi16(vi1xGHIJKLMNOPQRSTUV, vk1xGHIJKLMNOPQRSTUV); |
| const __m128i vprod1xOPQRSTUV = _mm256_extracti128_si256(vprod1xGHIJKLMNOPQRSTUV, 1); |
| vaccGHIJKLMN = _mm256_add_epi32(vaccGHIJKLMN, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod1xGHIJKLMNOPQRSTUV))); |
| vaccOPQRSTUV = _mm256_add_epi32(vaccOPQRSTUV, _mm256_cvtepi16_epi32(vprod1xOPQRSTUV)); |
| |
| const __m256i vi2x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i2)); |
| const __m256i vk2x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 64 * sizeof(int8_t)))); |
| const __m256i vi2xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (i2 + 16))); |
| const __m256i vk2xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 80 * sizeof(int8_t)))); |
| i2 += 32; |
| |
| const __m256i vprod2x0123456789ABCDEF = _mm256_mullo_epi16(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF); |
| const __m128i vprod2x89ABCDEF = _mm256_extracti128_si256(vprod2x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod2x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod2x89ABCDEF)); |
| const __m256i vprod2xGHIJKLMNOPQRSTUV = _mm256_mullo_epi16(vi2xGHIJKLMNOPQRSTUV, vk2xGHIJKLMNOPQRSTUV); |
| const __m128i vprod2xOPQRSTUV = _mm256_extracti128_si256(vprod2xGHIJKLMNOPQRSTUV, 1); |
| vaccGHIJKLMN = _mm256_add_epi32(vaccGHIJKLMN, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod2xGHIJKLMNOPQRSTUV))); |
| vaccOPQRSTUV = _mm256_add_epi32(vaccOPQRSTUV, _mm256_cvtepi16_epi32(vprod2xOPQRSTUV)); |
| |
| const __m256i vi3x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i3)); |
| const __m256i vk3x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 96 * sizeof(int8_t)))); |
| const __m256i vi3xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (i3 + 16))); |
| const __m256i vk3xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 112 * sizeof(int8_t)))); |
| i3 += 32; |
| |
| const __m256i vprod3x0123456789ABCDEF = _mm256_mullo_epi16(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF); |
| const __m128i vprod3x89ABCDEF = _mm256_extracti128_si256(vprod3x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod3x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod3x89ABCDEF)); |
| const __m256i vprod3xGHIJKLMNOPQRSTUV = _mm256_mullo_epi16(vi3xGHIJKLMNOPQRSTUV, vk3xGHIJKLMNOPQRSTUV); |
| const __m128i vprod3xOPQRSTUV = _mm256_extracti128_si256(vprod3xGHIJKLMNOPQRSTUV, 1); |
| vaccGHIJKLMN = _mm256_add_epi32(vaccGHIJKLMN, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod3xGHIJKLMNOPQRSTUV))); |
| vaccOPQRSTUV = _mm256_add_epi32(vaccOPQRSTUV, _mm256_cvtepi16_epi32(vprod3xOPQRSTUV)); |
| |
| const __m256i vi4x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i4)); |
| const __m256i vk4x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 128 * sizeof(int8_t)))); |
| const __m256i vi4xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (i4 + 16))); |
| const __m256i vk4xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 144 * sizeof(int8_t)))); |
| i4 += 32; |
| |
| const __m256i vprod4x0123456789ABCDEF = _mm256_mullo_epi16(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF); |
| const __m128i vprod4x89ABCDEF = _mm256_extracti128_si256(vprod4x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod4x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod4x89ABCDEF)); |
| const __m256i vprod4xGHIJKLMNOPQRSTUV = _mm256_mullo_epi16(vi4xGHIJKLMNOPQRSTUV, vk4xGHIJKLMNOPQRSTUV); |
| const __m128i vprod4xOPQRSTUV = _mm256_extracti128_si256(vprod4xGHIJKLMNOPQRSTUV, 1); |
| vaccGHIJKLMN = _mm256_add_epi32(vaccGHIJKLMN, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod4xGHIJKLMNOPQRSTUV))); |
| vaccOPQRSTUV = _mm256_add_epi32(vaccOPQRSTUV, _mm256_cvtepi16_epi32(vprod4xOPQRSTUV)); |
| |
| const __m256i vi5x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i5)); |
| const __m256i vk5x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 160 * sizeof(int8_t)))); |
| const __m256i vi5xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (i5 + 16))); |
| const __m256i vk5xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 176 * sizeof(int8_t)))); |
| i5 += 32; |
| |
| const __m256i vprod5x0123456789ABCDEF = _mm256_mullo_epi16(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF); |
| const __m128i vprod5x89ABCDEF = _mm256_extracti128_si256(vprod5x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod5x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod5x89ABCDEF)); |
| const __m256i vprod5xGHIJKLMNOPQRSTUV = _mm256_mullo_epi16(vi5xGHIJKLMNOPQRSTUV, vk5xGHIJKLMNOPQRSTUV); |
| const __m128i vprod5xOPQRSTUV = _mm256_extracti128_si256(vprod5xGHIJKLMNOPQRSTUV, 1); |
| vaccGHIJKLMN = _mm256_add_epi32(vaccGHIJKLMN, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod5xGHIJKLMNOPQRSTUV))); |
| vaccOPQRSTUV = _mm256_add_epi32(vaccOPQRSTUV, _mm256_cvtepi16_epi32(vprod5xOPQRSTUV)); |
| |
| const __m256i vi6x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i6)); |
| const __m256i vk6x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 192 * sizeof(int8_t)))); |
| const __m256i vi6xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (i6 + 16))); |
| const __m256i vk6xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 208 * sizeof(int8_t)))); |
| i6 += 32; |
| |
| const __m256i vprod6x0123456789ABCDEF = _mm256_mullo_epi16(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF); |
| const __m128i vprod6x89ABCDEF = _mm256_extracti128_si256(vprod6x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod6x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod6x89ABCDEF)); |
| const __m256i vprod6xGHIJKLMNOPQRSTUV = _mm256_mullo_epi16(vi6xGHIJKLMNOPQRSTUV, vk6xGHIJKLMNOPQRSTUV); |
| const __m128i vprod6xOPQRSTUV = _mm256_extracti128_si256(vprod6xGHIJKLMNOPQRSTUV, 1); |
| vaccGHIJKLMN = _mm256_add_epi32(vaccGHIJKLMN, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod6xGHIJKLMNOPQRSTUV))); |
| vaccOPQRSTUV = _mm256_add_epi32(vaccOPQRSTUV, _mm256_cvtepi16_epi32(vprod6xOPQRSTUV)); |
| |
| const __m256i vi7x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i7)); |
| const __m256i vk7x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 224 * sizeof(int8_t)))); |
| const __m256i vi7xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (i7 + 16))); |
| const __m256i vk7xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 240 * sizeof(int8_t)))); |
| i7 += 32; |
| |
| const __m256i vprod7x0123456789ABCDEF = _mm256_mullo_epi16(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF); |
| const __m128i vprod7x89ABCDEF = _mm256_extracti128_si256(vprod7x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod7x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod7x89ABCDEF)); |
| const __m256i vprod7xGHIJKLMNOPQRSTUV = _mm256_mullo_epi16(vi7xGHIJKLMNOPQRSTUV, vk7xGHIJKLMNOPQRSTUV); |
| const __m128i vprod7xOPQRSTUV = _mm256_extracti128_si256(vprod7xGHIJKLMNOPQRSTUV, 1); |
| vaccGHIJKLMN = _mm256_add_epi32(vaccGHIJKLMN, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod7xGHIJKLMNOPQRSTUV))); |
| vaccOPQRSTUV = _mm256_add_epi32(vaccOPQRSTUV, _mm256_cvtepi16_epi32(vprod7xOPQRSTUV)); |
| |
| const __m256i vi8x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i8)); |
| const __m256i vk8x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 256 * sizeof(int8_t)))); |
| const __m256i vi8xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (i8 + 16))); |
| const __m256i vk8xGHIJKLMNOPQRSTUV = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 272 * sizeof(int8_t)))); |
| i8 += 32; |
| |
| const __m256i vprod8x0123456789ABCDEF = _mm256_mullo_epi16(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF); |
| const __m128i vprod8x89ABCDEF = _mm256_extracti128_si256(vprod8x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod8x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod8x89ABCDEF)); |
| const __m256i vprod8xGHIJKLMNOPQRSTUV = _mm256_mullo_epi16(vi8xGHIJKLMNOPQRSTUV, vk8xGHIJKLMNOPQRSTUV); |
| const __m128i vprod8xOPQRSTUV = _mm256_extracti128_si256(vprod8xGHIJKLMNOPQRSTUV, 1); |
| vaccGHIJKLMN = _mm256_add_epi32(vaccGHIJKLMN, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod8xGHIJKLMNOPQRSTUV))); |
| vaccOPQRSTUV = _mm256_add_epi32(vaccOPQRSTUV, _mm256_cvtepi16_epi32(vprod8xOPQRSTUV)); |
| |
| w = (const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 288 * sizeof(int8_t)); |
| |
| const __m256i vmultiplier = _mm256_load_si256((const __m256i*) params->avx2.multiplier); |
| const __m256i vrounding = _mm256_load_si256((const __m256i*) params->avx2.rounding); |
| |
| const __m256i vacc1357 = _mm256_shuffle_epi32(vacc01234567, _MM_SHUFFLE(3, 3, 1, 1)); |
| const __m256i vacc9BDF = _mm256_shuffle_epi32(vacc89ABCDEF, _MM_SHUFFLE(3, 3, 1, 1)); |
| const __m256i vaccHJLN = _mm256_shuffle_epi32(vaccGHIJKLMN, _MM_SHUFFLE(3, 3, 1, 1)); |
| const __m256i vaccPRTV = _mm256_shuffle_epi32(vaccOPQRSTUV, _MM_SHUFFLE(3, 3, 1, 1)); |
| |
| const __m256i vprod0246 = _mm256_add_epi64(_mm256_mul_epi32(vacc01234567, vmultiplier), vrounding); |
| const __m256i vprod1357 = _mm256_add_epi64(_mm256_mul_epi32(vacc1357, vmultiplier), vrounding); |
| const __m256i vprod8ACE = _mm256_add_epi64(_mm256_mul_epi32(vacc89ABCDEF, vmultiplier), vrounding); |
| const __m256i vprod9BDF = _mm256_add_epi64(_mm256_mul_epi32(vacc9BDF, vmultiplier), vrounding); |
| const __m256i vprodGIKM = _mm256_add_epi64(_mm256_mul_epi32(vaccGHIJKLMN, vmultiplier), vrounding); |
| const __m256i vprodHJLN = _mm256_add_epi64(_mm256_mul_epi32(vaccHJLN, vmultiplier), vrounding); |
| const __m256i vprodOQSU = _mm256_add_epi64(_mm256_mul_epi32(vaccOPQRSTUV, vmultiplier), vrounding); |
| const __m256i vprodPRTV = _mm256_add_epi64(_mm256_mul_epi32(vaccPRTV, vmultiplier), vrounding); |
| |
| const __m256i vq31prod0246 = _mm256_srli_epi64(vprod0246, 31); |
| const __m256i vq31prod1357 = _mm256_add_epi64(vprod1357, vprod1357); |
| const __m256i vq31prod8ACE = _mm256_srli_epi64(vprod8ACE, 31); |
| const __m256i vq31prod9BDF = _mm256_add_epi64(vprod9BDF, vprod9BDF); |
| const __m256i vq31prodGIKM = _mm256_srli_epi64(vprodGIKM, 31); |
| const __m256i vq31prodHJLN = _mm256_add_epi64(vprodHJLN, vprodHJLN); |
| const __m256i vq31prodOQSU = _mm256_srli_epi64(vprodOQSU, 31); |
| const __m256i vq31prodPRTV = _mm256_add_epi64(vprodPRTV, vprodPRTV); |
| |
| const __m256i vq31prod01234567 = _mm256_blend_epi16(vq31prod0246, vq31prod1357, 0xCC); |
| const __m256i vq31prod89ABCDEF = _mm256_blend_epi16(vq31prod8ACE, vq31prod9BDF, 0xCC); |
| const __m256i vq31prodGHIJKLMN = _mm256_blend_epi16(vq31prodGIKM, vq31prodHJLN, 0xCC); |
| const __m256i vq31prodOPQRSTUV = _mm256_blend_epi16(vq31prodOQSU, vq31prodPRTV, 0xCC); |
| |
| const __m256i vremainder_mask = _mm256_load_si256((const __m256i*) params->avx2.remainder_mask); |
| const __m256i vrem01234567 = |
| _mm256_add_epi32(_mm256_and_si256(vq31prod01234567, vremainder_mask), _mm256_cmpgt_epi32(_mm256_setzero_si256(), vq31prod01234567)); |
| const __m256i vrem89ABCDEF = |
| _mm256_add_epi32(_mm256_and_si256(vq31prod89ABCDEF, vremainder_mask), _mm256_cmpgt_epi32(_mm256_setzero_si256(), vq31prod89ABCDEF)); |
| const __m256i vremGHIJKLMN = |
| _mm256_add_epi32(_mm256_and_si256(vq31prodGHIJKLMN, vremainder_mask), _mm256_cmpgt_epi32(_mm256_setzero_si256(), vq31prodGHIJKLMN)); |
| const __m256i vremOPQRSTUV = |
| _mm256_add_epi32(_mm256_and_si256(vq31prodOPQRSTUV, vremainder_mask), _mm256_cmpgt_epi32(_mm256_setzero_si256(), vq31prodOPQRSTUV)); |
| |
| const __m256i vremainder_threshold = _mm256_load_si256((const __m256i*) params->avx2.remainder_threshold); |
| const __m128i vshift = _mm_loadl_epi64((const __m128i*) params->avx2.shift); |
| vacc01234567 = |
| _mm256_sub_epi32(_mm256_sra_epi32(vq31prod01234567, vshift), _mm256_cmpgt_epi32(vrem01234567, vremainder_threshold)); |
| vacc89ABCDEF = |
| _mm256_sub_epi32(_mm256_sra_epi32(vq31prod89ABCDEF, vshift), _mm256_cmpgt_epi32(vrem89ABCDEF, vremainder_threshold)); |
| vaccGHIJKLMN = |
| _mm256_sub_epi32(_mm256_sra_epi32(vq31prodGHIJKLMN, vshift), _mm256_cmpgt_epi32(vremGHIJKLMN, vremainder_threshold)); |
| vaccOPQRSTUV = |
| _mm256_sub_epi32(_mm256_sra_epi32(vq31prodOPQRSTUV, vshift), _mm256_cmpgt_epi32(vremOPQRSTUV, vremainder_threshold)); |
| |
| const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->avx2.output_zero_point); |
| const __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(vacc01234567, vacc89ABCDEF), voutput_zero_point); |
| const __m256i voutGHIJOPQRKLMNSTUV = _mm256_adds_epi16(_mm256_packs_epi32(vaccGHIJKLMN, vaccOPQRSTUV), voutput_zero_point); |
| |
| __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packs_epi16(_mm256_castsi256_si128(vout012389AB4567CDEF), _mm256_extracti128_si256(vout012389AB4567CDEF, 1)), _MM_SHUFFLE(3, 1, 2, 0)); |
| __m128i voutGHIJKLMNOPQRSTUV = _mm_shuffle_epi32(_mm_packs_epi16(_mm256_castsi256_si128(voutGHIJOPQRKLMNSTUV), _mm256_extracti128_si256(voutGHIJOPQRKLMNSTUV, 1)), _MM_SHUFFLE(3, 1, 2, 0)); |
| |
| const __m128i voutput_min = _mm_load_si128((const __m128i*) params->avx2.output_min); |
| const __m128i voutput_max = _mm_load_si128((const __m128i*) params->avx2.output_max); |
| vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, voutput_min); |
| vout0123456789ABCDEF = _mm_min_epi8(vout0123456789ABCDEF, voutput_max); |
| voutGHIJKLMNOPQRSTUV = _mm_max_epi8(voutGHIJKLMNOPQRSTUV, voutput_min); |
| voutGHIJKLMNOPQRSTUV = _mm_min_epi8(voutGHIJKLMNOPQRSTUV, voutput_max); |
| |
| _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); |
| _mm_storeu_si128((__m128i*) (output + 16), voutGHIJKLMNOPQRSTUV); |
| output += 32; |
| } |
| if XNN_UNLIKELY(c != 0) { |
| const int8_t* k = (const int8_t*) ((uintptr_t) w + 32 * sizeof(int32_t)); |
| do { |
| __m256i vacc01234567 = _mm256_loadu_si256((const __m256i*) w); |
| __m256i vacc89ABCDEF = _mm256_loadu_si256((const __m256i*) ((uintptr_t) w + 8 * sizeof(int32_t))); |
| |
| |
| const __m256i vi0x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i0)); |
| const __m256i vk0x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) k)); |
| i0 += 16; |
| |
| const __m256i vprod0x0123456789ABCDEF = _mm256_mullo_epi16(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF); |
| const __m128i vprod0x89ABCDEF = _mm256_extracti128_si256(vprod0x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod0x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod0x89ABCDEF)); |
| |
| const __m256i vi1x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i1)); |
| const __m256i vk1x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (k + 32))); |
| i1 += 16; |
| |
| const __m256i vprod1x0123456789ABCDEF = _mm256_mullo_epi16(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF); |
| const __m128i vprod1x89ABCDEF = _mm256_extracti128_si256(vprod1x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod1x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod1x89ABCDEF)); |
| |
| const __m256i vi2x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i2)); |
| const __m256i vk2x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (k + 64))); |
| i2 += 16; |
| |
| const __m256i vprod2x0123456789ABCDEF = _mm256_mullo_epi16(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF); |
| const __m128i vprod2x89ABCDEF = _mm256_extracti128_si256(vprod2x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod2x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod2x89ABCDEF)); |
| |
| const __m256i vi3x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i3)); |
| const __m256i vk3x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (k + 96))); |
| i3 += 16; |
| |
| const __m256i vprod3x0123456789ABCDEF = _mm256_mullo_epi16(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF); |
| const __m128i vprod3x89ABCDEF = _mm256_extracti128_si256(vprod3x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod3x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod3x89ABCDEF)); |
| |
| const __m256i vi4x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i4)); |
| const __m256i vk4x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (k + 128))); |
| i4 += 16; |
| |
| const __m256i vprod4x0123456789ABCDEF = _mm256_mullo_epi16(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF); |
| const __m128i vprod4x89ABCDEF = _mm256_extracti128_si256(vprod4x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod4x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod4x89ABCDEF)); |
| |
| const __m256i vi5x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i5)); |
| const __m256i vk5x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (k + 160))); |
| i5 += 16; |
| |
| const __m256i vprod5x0123456789ABCDEF = _mm256_mullo_epi16(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF); |
| const __m128i vprod5x89ABCDEF = _mm256_extracti128_si256(vprod5x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod5x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod5x89ABCDEF)); |
| |
| const __m256i vi6x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i6)); |
| const __m256i vk6x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (k + 192))); |
| i6 += 16; |
| |
| const __m256i vprod6x0123456789ABCDEF = _mm256_mullo_epi16(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF); |
| const __m128i vprod6x89ABCDEF = _mm256_extracti128_si256(vprod6x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod6x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod6x89ABCDEF)); |
| |
| const __m256i vi7x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i7)); |
| const __m256i vk7x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (k + 224))); |
| i7 += 16; |
| |
| const __m256i vprod7x0123456789ABCDEF = _mm256_mullo_epi16(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF); |
| const __m128i vprod7x89ABCDEF = _mm256_extracti128_si256(vprod7x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod7x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod7x89ABCDEF)); |
| |
| const __m256i vi8x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) i8)); |
| const __m256i vk8x0123456789ABCDEF = _mm256_cvtepi8_epi16(_mm_loadu_si128((const __m128i*) (k + 256))); |
| i8 += 16; |
| |
| const __m256i vprod8x0123456789ABCDEF = _mm256_mullo_epi16(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF); |
| const __m128i vprod8x89ABCDEF = _mm256_extracti128_si256(vprod8x0123456789ABCDEF, 1); |
| vacc01234567 = _mm256_add_epi32(vacc01234567, _mm256_cvtepi16_epi32(_mm256_castsi256_si128(vprod8x0123456789ABCDEF))); |
| vacc89ABCDEF = _mm256_add_epi32(vacc89ABCDEF, _mm256_cvtepi16_epi32(vprod8x89ABCDEF)); |
| |
| w = (const void*) ((uintptr_t) w + 16 * sizeof(int32_t)); |
| k += 16; |
| |
| const __m256i vmultiplier = _mm256_load_si256((const __m256i*) params->avx2.multiplier); |
| const __m256i vrounding = _mm256_load_si256((const __m256i*) params->avx2.rounding); |
| |
| const __m256i vacc1357 = _mm256_shuffle_epi32(vacc01234567, _MM_SHUFFLE(3, 3, 1, 1)); |
| const __m256i vacc9BDF = _mm256_shuffle_epi32(vacc89ABCDEF, _MM_SHUFFLE(3, 3, 1, 1)); |
| |
| const __m256i vprod0246 = _mm256_add_epi64(_mm256_mul_epi32(vacc01234567, vmultiplier), vrounding); |
| const __m256i vprod1357 = _mm256_add_epi64(_mm256_mul_epi32(vacc1357, vmultiplier), vrounding); |
| const __m256i vprod8ACE = _mm256_add_epi64(_mm256_mul_epi32(vacc89ABCDEF, vmultiplier), vrounding); |
| const __m256i vprod9BDF = _mm256_add_epi64(_mm256_mul_epi32(vacc9BDF, vmultiplier), vrounding); |
| |
| const __m256i vq31prod0246 = _mm256_srli_epi64(vprod0246, 31); |
| const __m256i vq31prod1357 = _mm256_add_epi64(vprod1357, vprod1357); |
| const __m256i vq31prod8ACE = _mm256_srli_epi64(vprod8ACE, 31); |
| const __m256i vq31prod9BDF = _mm256_add_epi64(vprod9BDF, vprod9BDF); |
| |
| const __m256i vq31prod01234567 = _mm256_blend_epi16(vq31prod0246, vq31prod1357, 0xCC); |
| const __m256i vq31prod89ABCDEF = _mm256_blend_epi16(vq31prod8ACE, vq31prod9BDF, 0xCC); |
| |
| const __m256i vremainder_mask = _mm256_load_si256((const __m256i*) params->avx2.remainder_mask); |
| const __m256i vrem01234567 = |
| _mm256_add_epi32(_mm256_and_si256(vq31prod01234567, vremainder_mask), _mm256_cmpgt_epi32(_mm256_setzero_si256(), vq31prod01234567)); |
| const __m256i vrem89ABCDEF = |
| _mm256_add_epi32(_mm256_and_si256(vq31prod89ABCDEF, vremainder_mask), _mm256_cmpgt_epi32(_mm256_setzero_si256(), vq31prod89ABCDEF)); |
| |
| const __m256i vremainder_threshold = _mm256_load_si256((const __m256i*) params->avx2.remainder_threshold); |
| const __m128i vshift = _mm_loadl_epi64((const __m128i*) params->avx2.shift); |
| vacc01234567 = |
| _mm256_sub_epi32(_mm256_sra_epi32(vq31prod01234567, vshift), _mm256_cmpgt_epi32(vrem01234567, vremainder_threshold)); |
| vacc89ABCDEF = |
| _mm256_sub_epi32(_mm256_sra_epi32(vq31prod89ABCDEF, vshift), _mm256_cmpgt_epi32(vrem89ABCDEF, vremainder_threshold)); |
| |
| const __m128i voutput_zero_point = _mm_load_si128((const __m128i*) params->avx2.output_zero_point); |
| __m128i vout01234567 = _mm_adds_epi16(_mm_packs_epi32(_mm256_castsi256_si128(vacc01234567), _mm256_extracti128_si256(vacc01234567, 1)), voutput_zero_point); |
| __m128i vout89ABCDEF = _mm_adds_epi16(_mm_packs_epi32(_mm256_castsi256_si128(vacc89ABCDEF), _mm256_extracti128_si256(vacc89ABCDEF, 1)), voutput_zero_point); |
| |
| const __m128i voutput_min = _mm_load_si128((const __m128i*) params->avx2.output_min); |
| const __m128i voutput_max = _mm_load_si128((const __m128i*) params->avx2.output_max); |
| |
| __m128i vout0123456789ABCDEF = _mm_packs_epi16(vout01234567, vout89ABCDEF); |
| vout0123456789ABCDEF = _mm_min_epi8(_mm_max_epi8(vout0123456789ABCDEF, voutput_min), voutput_max); |
| |
| if XNN_LIKELY(c >= 16) { |
| _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); |
| output += 16; |
| c -= 16; |
| } else { |
| if (c & 8) { |
| _mm_storel_epi64((__m128i*) output, vout0123456789ABCDEF); |
| vout0123456789ABCDEF = _mm_unpackhi_epi64(vout0123456789ABCDEF, vout0123456789ABCDEF); |
| output += 8; |
| } |
| if (c & 4) { |
| *((uint32_t*) output) = (uint32_t) _mm_cvtsi128_si32(vout0123456789ABCDEF); |
| vout0123456789ABCDEF = _mm_srli_epi64(vout0123456789ABCDEF, 32); |
| output += 4; |
| } |
| if (c & 2) { |
| *((uint16_t*) output) = (uint16_t) _mm_extract_epi16(vout0123456789ABCDEF, 0); |
| vout0123456789ABCDEF = _mm_srli_epi32(vout0123456789ABCDEF, 16); |
| output += 2; |
| } |
| if (c & 1) { |
| *output = (int8_t) _mm_extract_epi8(vout0123456789ABCDEF, 0); |
| output += 1; |
| } |
| c = 0; |
| } |
| } while (c != 0); |
| } |
| |
| output = (int8_t*) ((uintptr_t) output + output_increment); |
| } while (--output_width != 0); |
| } |