// Auto-generated file. Do not edit! // Template: src/f16-dwconv/multipass-fma3.c.in // Generator: tools/xngen // // Copyright 2022 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 #include #include #include #include "xnnpack/dwconv.h" #include "xnnpack/intrinsics-polyfill.h" #include "xnnpack/math.h" void xnn_f16_dwconv_minmax_ukernel_8f8m9l32c8s4r__fma3( size_t channels, size_t output_width, const xnn_float16** input, const xnn_float16* weights, xnn_float16* output, intptr_t input_stride, size_t output_increment, size_t input_offset, const xnn_float16* zero, size_t kernel_size, xnn_float16* buffer, const union xnn_f16_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(channels != 0); assert(output_width != 0); assert(kernel_size > 8); const __m256 vmin = _mm256_cvtph_ps(_mm_set1_epi16(*(const uint16_t*) ¶ms->scalar.min)); const __m256 vmax = _mm256_cvtph_ps(_mm_set1_epi16(*(const uint16_t*) ¶ms->scalar.max)); XNN_FORCE_REALIZATION(vmin); XNN_FORCE_REALIZATION(vmax); do { const xnn_float16* w = weights; // First pass to process 8 inputs. { uint16_t* b = (uint16_t*) buffer; const uint16_t* i0 = (const uint16_t*) input[0]; assert(i0 != NULL); if XNN_UNPREDICTABLE(i0 != (const uint16_t*) zero) { i0 = (const uint16_t*) ((uintptr_t) i0 + input_offset); } const uint16_t* i1 = (const uint16_t*) input[1]; assert(i1 != NULL); if XNN_UNPREDICTABLE(i1 != (const uint16_t*) zero) { i1 = (const uint16_t*) ((uintptr_t) i1 + input_offset); } const uint16_t* i2 = (const uint16_t*) input[2]; assert(i2 != NULL); if XNN_UNPREDICTABLE(i2 != (const uint16_t*) zero) { i2 = (const uint16_t*) ((uintptr_t) i2 + input_offset); } const uint16_t* i3 = (const uint16_t*) input[3]; assert(i3 != NULL); if XNN_UNPREDICTABLE(i3 != (const uint16_t*) zero) { i3 = (const uint16_t*) ((uintptr_t) i3 + input_offset); } const uint16_t* i4 = (const uint16_t*) input[4]; assert(i4 != NULL); if XNN_UNPREDICTABLE(i4 != (const uint16_t*) zero) { i4 = (const uint16_t*) ((uintptr_t) i4 + input_offset); } const uint16_t* i5 = (const uint16_t*) input[5]; assert(i5 != NULL); if XNN_UNPREDICTABLE(i5 != (const uint16_t*) zero) { i5 = (const uint16_t*) ((uintptr_t) i5 + input_offset); } const uint16_t* i6 = (const uint16_t*) input[6]; assert(i6 != NULL); if XNN_UNPREDICTABLE(i6 != (const uint16_t*) zero) { i6 = (const uint16_t*) ((uintptr_t) i6 + input_offset); } const uint16_t* i7 = (const uint16_t*) input[7]; assert(i7 != NULL); if XNN_UNPREDICTABLE(i7 != (const uint16_t*) zero) { i7 = (const uint16_t*) ((uintptr_t) i7 + input_offset); } input += 8; // Process c channels and write to buffer. size_t c = round_up_po2(channels, 4); for (; c >= 32; c -= 32) { __m256 vacc01234567p0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) w)); __m256 vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 8))); __m256 vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 16))); __m256 vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 24))); const __m256 vi0x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0))); const __m256 vi0x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0 + 8))); const __m256 vi0xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m256 vi0xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0 + 24))); i0 += 32; const __m256 vk0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 32))); const __m256 vk0x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 40))); const __m256 vk0xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); const __m256 vk0xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 56))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x01234567, vk0x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x89ABCDEF, vk0x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0xGHIJKLMN, vk0xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0xOPQRSTUV, vk0xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi1x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1))); const __m256 vi1x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1 + 8))); const __m256 vi1xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m256 vi1xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1 + 24))); i1 += 32; const __m256 vk1x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 64))); const __m256 vk1x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 72))); const __m256 vk1xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 80))); const __m256 vk1xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 88))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x01234567, vk1x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x89ABCDEF, vk1x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1xGHIJKLMN, vk1xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1xOPQRSTUV, vk1xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi2x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2))); const __m256 vi2x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2 + 8))); const __m256 vi2xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m256 vi2xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2 + 24))); i2 += 32; const __m256 vk2x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 96))); const __m256 vk2x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 104))); const __m256 vk2xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 112))); const __m256 vk2xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 120))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x01234567, vk2x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x89ABCDEF, vk2x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2xGHIJKLMN, vk2xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2xOPQRSTUV, vk2xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi3x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3))); const __m256 vi3x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3 + 8))); const __m256 vi3xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3 + 16))); const __m256 vi3xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3 + 24))); i3 += 32; const __m256 vk3x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 128))); const __m256 vk3x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 136))); const __m256 vk3xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 144))); const __m256 vk3xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 152))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x01234567, vk3x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x89ABCDEF, vk3x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3xGHIJKLMN, vk3xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3xOPQRSTUV, vk3xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi4x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4))); const __m256 vi4x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4 + 8))); const __m256 vi4xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4 + 16))); const __m256 vi4xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4 + 24))); i4 += 32; const __m256 vk4x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 160))); const __m256 vk4x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 168))); const __m256 vk4xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 176))); const __m256 vk4xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 184))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x01234567, vk4x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x89ABCDEF, vk4x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4xGHIJKLMN, vk4xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4xOPQRSTUV, vk4xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5))); const __m256 vi5x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5 + 8))); const __m256 vi5xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5 + 16))); const __m256 vi5xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5 + 24))); i5 += 32; const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 192))); const __m256 vk5x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 200))); const __m256 vk5xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 208))); const __m256 vk5xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 216))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x89ABCDEF, vk5x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5xGHIJKLMN, vk5xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5xOPQRSTUV, vk5xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi6x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6))); const __m256 vi6x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6 + 8))); const __m256 vi6xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6 + 16))); const __m256 vi6xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6 + 24))); i6 += 32; const __m256 vk6x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 224))); const __m256 vk6x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 232))); const __m256 vk6xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 240))); const __m256 vk6xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 248))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x01234567, vk6x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x89ABCDEF, vk6x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6xGHIJKLMN, vk6xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6xOPQRSTUV, vk6xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi7x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7))); const __m256 vi7x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7 + 8))); const __m256 vi7xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7 + 16))); const __m256 vi7xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7 + 24))); i7 += 32; const __m256 vk7x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 256))); const __m256 vk7x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 264))); const __m256 vk7xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 272))); const __m256 vk7xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 280))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x01234567, vk7x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x89ABCDEF, vk7x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7xGHIJKLMN, vk7xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7xOPQRSTUV, vk7xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); w += 288; _mm_store_si128((__m128i*) b, _mm256_cvtps_ph(vacc01234567p0, _MM_FROUND_TO_NEAREST_INT)); _mm_store_si128((__m128i*) (b + 8), _mm256_cvtps_ph(vacc89ABCDEFp0, _MM_FROUND_TO_NEAREST_INT)); _mm_store_si128((__m128i*) (b + 16), _mm256_cvtps_ph(vaccGHIJKLMNp0, _MM_FROUND_TO_NEAREST_INT)); _mm_store_si128((__m128i*) (b + 24), _mm256_cvtps_ph(vaccOPQRSTUVp0, _MM_FROUND_TO_NEAREST_INT)); b += 32; } for (; c >= 8; c -= 8) { __m256 vacc01234567p0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) w)); const __m256 vi0x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0))); i0 += 8; const __m256 vk0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 8))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x01234567, vk0x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi1x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1))); i1 += 8; const __m256 vk1x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 16))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x01234567, vk1x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi2x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2))); i2 += 8; const __m256 vk2x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 24))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x01234567, vk2x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi3x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3))); i3 += 8; const __m256 vk3x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 32))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x01234567, vk3x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi4x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4))); i4 += 8; const __m256 vk4x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 40))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x01234567, vk4x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5))); i5 += 8; const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi6x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6))); i6 += 8; const __m256 vk6x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 56))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x01234567, vk6x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi7x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7))); i7 += 8; const __m256 vk7x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 64))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x01234567, vk7x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); w += 72; _mm_store_si128((__m128i*) b, _mm256_cvtps_ph(vacc01234567p0, _MM_FROUND_TO_NEAREST_INT)); b += 8; } if (c != 0) { assert(c >= 1); assert(c <= 7); __m256 vacc01234567p0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) w)); const __m256 vi0x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i0)); const __m256 vk0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 8))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x01234567, vk0x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi1x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i1)); const __m256 vk1x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 16))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x01234567, vk1x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi2x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i2)); const __m256 vk2x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 24))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x01234567, vk2x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi3x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i3)); const __m256 vk3x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 32))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x01234567, vk3x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi4x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i4)); const __m256 vk4x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 40))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x01234567, vk4x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i5)); const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi6x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i6)); const __m256 vk6x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 56))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x01234567, vk6x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi7x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i7)); const __m256 vk7x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 64))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x01234567, vk7x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); w += 72; _mm_store_si128((__m128i*) b, _mm256_cvtps_ph(vacc01234567p0, _MM_FROUND_TO_NEAREST_INT)); } } // Middle pass to process 8 inputs in each iteration. for (size_t ks = kernel_size - 8; ks > 9; ks -= 8) { uint16_t* b = (uint16_t*) buffer; const uint16_t* i0 = (const uint16_t*) input[0]; assert(i0 != NULL); if XNN_UNPREDICTABLE(i0 != (const uint16_t*) zero) { i0 = (const uint16_t*) ((uintptr_t) i0 + input_offset); } const uint16_t* i1 = (const uint16_t*) input[1]; assert(i1 != NULL); if XNN_UNPREDICTABLE(i1 != (const uint16_t*) zero) { i1 = (const uint16_t*) ((uintptr_t) i1 + input_offset); } const uint16_t* i2 = (const uint16_t*) input[2]; assert(i2 != NULL); if XNN_UNPREDICTABLE(i2 != (const uint16_t*) zero) { i2 = (const uint16_t*) ((uintptr_t) i2 + input_offset); } const uint16_t* i3 = (const uint16_t*) input[3]; assert(i3 != NULL); if XNN_UNPREDICTABLE(i3 != (const uint16_t*) zero) { i3 = (const uint16_t*) ((uintptr_t) i3 + input_offset); } const uint16_t* i4 = (const uint16_t*) input[4]; assert(i4 != NULL); if XNN_UNPREDICTABLE(i4 != (const uint16_t*) zero) { i4 = (const uint16_t*) ((uintptr_t) i4 + input_offset); } const uint16_t* i5 = (const uint16_t*) input[5]; assert(i5 != NULL); if XNN_UNPREDICTABLE(i5 != (const uint16_t*) zero) { i5 = (const uint16_t*) ((uintptr_t) i5 + input_offset); } const uint16_t* i6 = (const uint16_t*) input[6]; assert(i6 != NULL); if XNN_UNPREDICTABLE(i6 != (const uint16_t*) zero) { i6 = (const uint16_t*) ((uintptr_t) i6 + input_offset); } const uint16_t* i7 = (const uint16_t*) input[7]; assert(i7 != NULL); if XNN_UNPREDICTABLE(i7 != (const uint16_t*) zero) { i7 = (const uint16_t*) ((uintptr_t) i7 + input_offset); } input += 8; size_t c = round_up_po2(channels, 4); for (; c >= 32; c -= 32) { __m256 vacc01234567p0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b))); __m256 vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b + 8))); __m256 vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b + 16))); __m256 vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b + 24))); const __m256 vi0x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0))); const __m256 vi0x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0 + 8))); const __m256 vi0xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m256 vi0xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0 + 24))); i0 += 32; const __m256 vk0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w))); const __m256 vk0x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 8))); const __m256 vk0xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 16))); const __m256 vk0xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 24))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x01234567, vk0x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x89ABCDEF, vk0x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0xGHIJKLMN, vk0xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0xOPQRSTUV, vk0xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi1x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1))); const __m256 vi1x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1 + 8))); const __m256 vi1xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m256 vi1xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1 + 24))); i1 += 32; const __m256 vk1x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 32))); const __m256 vk1x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 40))); const __m256 vk1xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); const __m256 vk1xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 56))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x01234567, vk1x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x89ABCDEF, vk1x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1xGHIJKLMN, vk1xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1xOPQRSTUV, vk1xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi2x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2))); const __m256 vi2x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2 + 8))); const __m256 vi2xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m256 vi2xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2 + 24))); i2 += 32; const __m256 vk2x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 64))); const __m256 vk2x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 72))); const __m256 vk2xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 80))); const __m256 vk2xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 88))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x01234567, vk2x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x89ABCDEF, vk2x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2xGHIJKLMN, vk2xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2xOPQRSTUV, vk2xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi3x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3))); const __m256 vi3x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3 + 8))); const __m256 vi3xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3 + 16))); const __m256 vi3xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3 + 24))); i3 += 32; const __m256 vk3x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 96))); const __m256 vk3x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 104))); const __m256 vk3xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 112))); const __m256 vk3xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 120))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x01234567, vk3x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x89ABCDEF, vk3x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3xGHIJKLMN, vk3xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3xOPQRSTUV, vk3xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi4x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4))); const __m256 vi4x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4 + 8))); const __m256 vi4xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4 + 16))); const __m256 vi4xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4 + 24))); i4 += 32; const __m256 vk4x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 128))); const __m256 vk4x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 136))); const __m256 vk4xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 144))); const __m256 vk4xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 152))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x01234567, vk4x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x89ABCDEF, vk4x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4xGHIJKLMN, vk4xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4xOPQRSTUV, vk4xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5))); const __m256 vi5x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5 + 8))); const __m256 vi5xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5 + 16))); const __m256 vi5xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5 + 24))); i5 += 32; const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 160))); const __m256 vk5x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 168))); const __m256 vk5xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 176))); const __m256 vk5xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 184))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x89ABCDEF, vk5x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5xGHIJKLMN, vk5xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5xOPQRSTUV, vk5xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi6x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6))); const __m256 vi6x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6 + 8))); const __m256 vi6xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6 + 16))); const __m256 vi6xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6 + 24))); i6 += 32; const __m256 vk6x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 192))); const __m256 vk6x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 200))); const __m256 vk6xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 208))); const __m256 vk6xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 216))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x01234567, vk6x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x89ABCDEF, vk6x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6xGHIJKLMN, vk6xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6xOPQRSTUV, vk6xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi7x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7))); const __m256 vi7x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7 + 8))); const __m256 vi7xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7 + 16))); const __m256 vi7xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7 + 24))); i7 += 32; const __m256 vk7x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 224))); const __m256 vk7x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 232))); const __m256 vk7xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 240))); const __m256 vk7xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 248))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x01234567, vk7x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x89ABCDEF, vk7x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7xGHIJKLMN, vk7xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7xOPQRSTUV, vk7xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); w += 256; _mm_store_si128((__m128i*) b, _mm256_cvtps_ph(vacc01234567p0, _MM_FROUND_TO_NEAREST_INT)); _mm_store_si128((__m128i*) (b + 8), _mm256_cvtps_ph(vacc89ABCDEFp0, _MM_FROUND_TO_NEAREST_INT)); _mm_store_si128((__m128i*) (b + 16), _mm256_cvtps_ph(vaccGHIJKLMNp0, _MM_FROUND_TO_NEAREST_INT)); _mm_store_si128((__m128i*) (b + 24), _mm256_cvtps_ph(vaccOPQRSTUVp0, _MM_FROUND_TO_NEAREST_INT)); b += 32; } for (; c >= 8; c -= 8) { __m256 vacc01234567p0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b))); const __m256 vi0x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0))); i0 += 8; const __m256 vk0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x01234567, vk0x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi1x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1))); i1 += 8; const __m256 vk1x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 8))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x01234567, vk1x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi2x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2))); i2 += 8; const __m256 vk2x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 16))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x01234567, vk2x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi3x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3))); i3 += 8; const __m256 vk3x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 24))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x01234567, vk3x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi4x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4))); i4 += 8; const __m256 vk4x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 32))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x01234567, vk4x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5))); i5 += 8; const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 40))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi6x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6))); i6 += 8; const __m256 vk6x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x01234567, vk6x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi7x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7))); i7 += 8; const __m256 vk7x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 56))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x01234567, vk7x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); w += 64; _mm_store_si128((__m128i*) b, _mm256_cvtps_ph(vacc01234567p0, _MM_FROUND_TO_NEAREST_INT)); b += 8; } if (c != 0) { assert(c >= 1); assert(c <= 7); __m256 vacc01234567p0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b))); const __m256 vi0x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i0)); const __m256 vk0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x01234567, vk0x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi1x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i1)); const __m256 vk1x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 8))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x01234567, vk1x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi2x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i2)); const __m256 vk2x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 16))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x01234567, vk2x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi3x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i3)); const __m256 vk3x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 24))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x01234567, vk3x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi4x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i4)); const __m256 vk4x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 32))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x01234567, vk4x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i5)); const __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 40))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi6x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i6)); const __m256 vk6x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x01234567, vk6x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi7x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i7)); const __m256 vk7x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 56))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x01234567, vk7x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); w += 64; _mm_store_si128((__m128i*) b, _mm256_cvtps_ph(vacc01234567p0, _MM_FROUND_TO_NEAREST_INT)); } } // Last pass to process up to 9 inputs. { uint16_t* b = (uint16_t*) buffer; const uint16_t* i0 = (const uint16_t*) input[0]; assert(i0 != NULL); if XNN_UNPREDICTABLE(i0 != (const uint16_t*) zero) { i0 = (const uint16_t*) ((uintptr_t) i0 + input_offset); } const uint16_t* i1 = (const uint16_t*) input[1]; assert(i1 != NULL); if XNN_UNPREDICTABLE(i1 != (const uint16_t*) zero) { i1 = (const uint16_t*) ((uintptr_t) i1 + input_offset); } const uint16_t* i2 = (const uint16_t*) input[2]; assert(i2 != NULL); if XNN_UNPREDICTABLE(i2 != (const uint16_t*) zero) { i2 = (const uint16_t*) ((uintptr_t) i2 + input_offset); } const uint16_t* i3 = (const uint16_t*) input[3]; assert(i3 != NULL); if XNN_UNPREDICTABLE(i3 != (const uint16_t*) zero) { i3 = (const uint16_t*) ((uintptr_t) i3 + input_offset); } const uint16_t* i4 = (const uint16_t*) input[4]; assert(i4 != NULL); if XNN_UNPREDICTABLE(i4 != (const uint16_t*) zero) { i4 = (const uint16_t*) ((uintptr_t) i4 + input_offset); } const uint16_t* i5 = (const uint16_t*) input[5]; assert(i5 != NULL); if XNN_UNPREDICTABLE(i5 != (const uint16_t*) zero) { i5 = (const uint16_t*) ((uintptr_t) i5 + input_offset); } const uint16_t* i6 = (const uint16_t*) input[6]; assert(i6 != NULL); if XNN_UNPREDICTABLE(i6 != (const uint16_t*) zero) { i6 = (const uint16_t*) ((uintptr_t) i6 + input_offset); } const uint16_t* i7 = (const uint16_t*) input[7]; assert(i7 != NULL); if XNN_UNPREDICTABLE(i7 != (const uint16_t*) zero) { i7 = (const uint16_t*) ((uintptr_t) i7 + input_offset); } const uint16_t* i8 = (const uint16_t*) input[8]; assert(i8 != NULL); if XNN_UNPREDICTABLE(i8 != (const uint16_t*) zero) { i8 = (const uint16_t*) ((uintptr_t) i8 + input_offset); } size_t c = channels; for (; c >= 32; c -= 32) { __m256 vacc01234567p0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b))); __m256 vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b + 8))); __m256 vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b + 16))); __m256 vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b + 24))); b += 32; const __m256 vi0x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0))); const __m256 vi0x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0 + 8))); const __m256 vi0xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m256 vi0xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0 + 24))); i0 += 32; __m256 vk0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w))); __m256 vk0x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 8))); __m256 vk0xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 16))); __m256 vk0xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 24))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x01234567, vk0x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x89ABCDEF, vk0x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0xGHIJKLMN, vk0xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0xOPQRSTUV, vk0xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi1x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1))); const __m256 vi1x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1 + 8))); const __m256 vi1xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m256 vi1xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1 + 24))); i1 += 32; __m256 vk1x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 32))); __m256 vk1x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 40))); __m256 vk1xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); __m256 vk1xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 56))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x01234567, vk1x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x89ABCDEF, vk1x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1xGHIJKLMN, vk1xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1xOPQRSTUV, vk1xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi2x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2))); const __m256 vi2x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2 + 8))); const __m256 vi2xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m256 vi2xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2 + 24))); i2 += 32; __m256 vk2x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 64))); __m256 vk2x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 72))); __m256 vk2xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 80))); __m256 vk2xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 88))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x01234567, vk2x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x89ABCDEF, vk2x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2xGHIJKLMN, vk2xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2xOPQRSTUV, vk2xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi3x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3))); const __m256 vi3x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3 + 8))); const __m256 vi3xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3 + 16))); const __m256 vi3xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3 + 24))); i3 += 32; __m256 vk3x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 96))); __m256 vk3x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 104))); __m256 vk3xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 112))); __m256 vk3xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 120))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x01234567, vk3x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x89ABCDEF, vk3x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3xGHIJKLMN, vk3xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3xOPQRSTUV, vk3xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi4x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4))); const __m256 vi4x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4 + 8))); const __m256 vi4xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4 + 16))); const __m256 vi4xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4 + 24))); i4 += 32; __m256 vk4x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 128))); __m256 vk4x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 136))); __m256 vk4xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 144))); __m256 vk4xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 152))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x01234567, vk4x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x89ABCDEF, vk4x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4xGHIJKLMN, vk4xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4xOPQRSTUV, vk4xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5))); const __m256 vi5x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5 + 8))); const __m256 vi5xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5 + 16))); const __m256 vi5xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5 + 24))); i5 += 32; __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 160))); __m256 vk5x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 168))); __m256 vk5xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 176))); __m256 vk5xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 184))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x89ABCDEF, vk5x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5xGHIJKLMN, vk5xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5xOPQRSTUV, vk5xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi6x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6))); const __m256 vi6x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6 + 8))); const __m256 vi6xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6 + 16))); const __m256 vi6xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6 + 24))); i6 += 32; __m256 vk6x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 192))); __m256 vk6x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 200))); __m256 vk6xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 208))); __m256 vk6xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 216))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x01234567, vk6x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x89ABCDEF, vk6x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6xGHIJKLMN, vk6xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6xOPQRSTUV, vk6xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi7x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7))); const __m256 vi7x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7 + 8))); const __m256 vi7xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7 + 16))); const __m256 vi7xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7 + 24))); i7 += 32; __m256 vk7x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 224))); __m256 vk7x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 232))); __m256 vk7xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 240))); __m256 vk7xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 248))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x01234567, vk7x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x89ABCDEF, vk7x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7xGHIJKLMN, vk7xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7xOPQRSTUV, vk7xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi8x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i8))); const __m256 vi8x89ABCDEF = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i8 + 8))); const __m256 vi8xGHIJKLMN = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i8 + 16))); const __m256 vi8xOPQRSTUV = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i8 + 24))); i8 += 32; __m256 vk8x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 256))); __m256 vk8x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 264))); __m256 vk8xGHIJKLMN = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 272))); __m256 vk8xOPQRSTUV = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 280))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi8x01234567, vk8x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); vacc89ABCDEFp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi8x89ABCDEF, vk8x89ABCDEF, vacc89ABCDEFp0), _MM_FROUND_TO_NEAREST_INT)); vaccGHIJKLMNp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi8xGHIJKLMN, vk8xGHIJKLMN, vaccGHIJKLMNp0), _MM_FROUND_TO_NEAREST_INT)); vaccOPQRSTUVp0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi8xOPQRSTUV, vk8xOPQRSTUV, vaccOPQRSTUVp0), _MM_FROUND_TO_NEAREST_INT)); w += 288; __m256 vacc01234567 = _mm256_max_ps(vacc01234567p0, vmin); __m256 vacc89ABCDEF = _mm256_max_ps(vacc89ABCDEFp0, vmin); __m256 vaccGHIJKLMN = _mm256_max_ps(vaccGHIJKLMNp0, vmin); __m256 vaccOPQRSTUV = _mm256_max_ps(vaccOPQRSTUVp0, vmin); vacc01234567 = _mm256_min_ps(vacc01234567, vmax); vacc89ABCDEF = _mm256_min_ps(vacc89ABCDEF, vmax); vaccGHIJKLMN = _mm256_min_ps(vaccGHIJKLMN, vmax); vaccOPQRSTUV = _mm256_min_ps(vaccOPQRSTUV, vmax); _mm_storeu_si128((__m128i*) output, _mm256_cvtps_ph(vacc01234567, _MM_FROUND_TO_NEAREST_INT)); _mm_storeu_si128((__m128i*) ((uint16_t*) output + 8), _mm256_cvtps_ph(vacc89ABCDEF, _MM_FROUND_TO_NEAREST_INT)); _mm_storeu_si128((__m128i*) ((uint16_t*) output + 16), _mm256_cvtps_ph(vaccGHIJKLMN, _MM_FROUND_TO_NEAREST_INT)); _mm_storeu_si128((__m128i*) ((uint16_t*) output + 24), _mm256_cvtps_ph(vaccOPQRSTUV, _MM_FROUND_TO_NEAREST_INT)); output = (xnn_float16*) output + 32; } for (; c >= 8; c -= 8) { __m256 vacc01234567p0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b))); b += 8; const __m256 vi0x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i0))); i0 += 8; __m256 vk0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x01234567, vk0x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi1x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i1))); i1 += 8; __m256 vk1x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 8))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x01234567, vk1x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi2x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i2))); i2 += 8; __m256 vk2x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 16))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x01234567, vk2x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi3x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i3))); i3 += 8; __m256 vk3x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 24))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x01234567, vk3x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi4x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i4))); i4 += 8; __m256 vk4x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 32))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x01234567, vk4x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i5))); i5 += 8; __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 40))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi6x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i6))); i6 += 8; __m256 vk6x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x01234567, vk6x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi7x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i7))); i7 += 8; __m256 vk7x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 56))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x01234567, vk7x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi8x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) (i8))); i8 += 8; __m256 vk8x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 64))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi8x01234567, vk8x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); w += 72; __m256 vacc01234567 = _mm256_max_ps(vacc01234567p0, vmin); vacc01234567 = _mm256_min_ps(vacc01234567, vmax); _mm_storeu_si128((__m128i*) output, _mm256_cvtps_ph(vacc01234567, _MM_FROUND_TO_NEAREST_INT)); output = (xnn_float16*) output + 8; } if XNN_UNLIKELY(c != 0) { assert(c >= 1); assert(c <= 7); __m256 vacc01234567p0 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (b))); const __m256 vi0x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i0)); __m256 vk0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi0x01234567, vk0x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi1x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i1)); __m256 vk1x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 8))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi1x01234567, vk1x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi2x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i2)); __m256 vk2x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 16))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi2x01234567, vk2x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi3x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i3)); __m256 vk3x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 24))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi3x01234567, vk3x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi4x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i4)); __m256 vk4x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 32))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi4x01234567, vk4x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi5x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i5)); __m256 vk5x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 40))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi5x01234567, vk5x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi6x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i6)); __m256 vk6x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 48))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi6x01234567, vk6x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi7x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i7)); __m256 vk7x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 56))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi7x01234567, vk7x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); const __m256 vi8x01234567 = _mm256_cvtph_ps(_mm_loadu_si128((const __m128i*) i8)); __m256 vk8x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) (w + 64))); vacc01234567p0 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(vi8x01234567, vk8x01234567, vacc01234567p0), _MM_FROUND_TO_NEAREST_INT)); __m256 vacc01234567 = _mm256_max_ps(vacc01234567p0, vmin); vacc01234567 = _mm256_min_ps(vacc01234567, vmax); __m128i vh01234567 = _mm256_cvtps_ph(vacc01234567, _MM_FROUND_TO_NEAREST_INT); if (c & 4) { _mm_storel_epi64((__m128i*) output, vh01234567); vh01234567 = _mm_unpackhi_epi64(vh01234567, vh01234567); output = (xnn_float16*) output + 4; } if (c & 2) { _mm_storeu_si32(output, vh01234567); vh01234567 = _mm_srli_epi64(vh01234567, 32); output = (xnn_float16*) output + 2; } if (c & 1) { *((uint16_t*) output) = (uint16_t) _mm_extract_epi16(vh01234567, 0); output = (xnn_float16*) output + 1; } } } input = (const xnn_float16**) ((uintptr_t) input + input_stride); output = (xnn_float16*) ((uintptr_t) output + output_increment); } while (--output_width != 0); }