// Copyright 2021 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 #include #include #include #include #include #include void xnn_f16_f32_vcvt_ukernel__avx512skx_x16( size_t n, const void* input, float* output, const union xnn_f16_f32_cvt_params params[restrict XNN_MIN_ELEMENTS(1)]) { assert(n != 0); assert(n % sizeof(uint16_t) == 0); assert(input != NULL); assert(output != NULL); const uint16_t* i = (const uint16_t*) input; for (; n >= 16 * sizeof(uint16_t); n -= 16 * sizeof(uint16_t)) { const __m512 vacc = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i*) i)); i += 16; _mm512_storeu_ps(output, vacc); output += 16; } if XNN_UNLIKELY(n != 0) { assert(n >= 1 * sizeof(uint16_t)); assert(n <= 15 * sizeof(uint16_t)); // Prepare mask for valid 32-bit elements (depends on n). n >>= 1 /* log2(sizeof(uint16_t)) */; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << n) - UINT32_C(1))); const __m512 vacc = _mm512_cvtph_ps(_mm256_maskz_loadu_epi16(vmask, i)); _mm512_mask_storeu_ps(output, vmask, vacc); } } void xnn_f32_f16_vcvt_ukernel__avx512skx_x16( size_t n, const float* input, void* output, const union xnn_f32_f16_cvt_params params[restrict XNN_MIN_ELEMENTS(1)]) { assert(n != 0); assert(n % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); uint16_t* o = (uint16_t*) output; for (; n >= 16 * sizeof(float); n -= 16 * sizeof(float)) { const __m512 vf = _mm512_loadu_ps(input); input += 16; _mm256_storeu_si256((__m256i*) o, _mm512_cvtps_ph(vf, _MM_FROUND_NO_EXC)); o += 16; } if XNN_UNLIKELY(n != 0) { assert(n >= 1 * sizeof(float)); assert(n <= 15 * sizeof(float)); // Prepare mask for valid elements (depends on n). n >>= 2 /* log2(sizeof(float)) */; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << n) - UINT32_C(1))); const __m512 vf = _mm512_maskz_loadu_ps(vmask, input); const __m256i vh = _mm512_cvtps_ph(vf, _MM_FROUND_NO_EXC); _mm256_mask_storeu_epi16(o, vmask, vh); } } void xnn_f32_qs8_vcvt_ukernel__avx512skx_x128( size_t n, const float* x, int8_t* y, const union xnn_f32_qs8_cvt_params params[restrict XNN_MIN_ELEMENTS(1)]) { assert(n != 0); assert(n % sizeof(float) == 0); assert(x != NULL); assert(y != NULL); const __m512 vscale = _mm512_load_ps(params->avx2.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->avx512.output_zero_point); const __m512i vshuffle512_mask = _mm512_load_si512(params->avx512.shuffle512_mask); const __m512i voutput_min = _mm512_load_si512(params->avx512.output_min); for (; n >= 128 * sizeof(float); n -= 128 * sizeof(float)) { __m512 vx0123 = _mm512_loadu_ps(x); __m512 vx4567 = _mm512_loadu_ps(x + 16); __m512 vx89AB = _mm512_loadu_ps(x + 32); __m512 vxCDEF = _mm512_loadu_ps(x + 48); __m512 vxGHIJ = _mm512_loadu_ps(x + 64); __m512 vxKLMN = _mm512_loadu_ps(x + 80); __m512 vxOPQR = _mm512_loadu_ps(x + 96); __m512 vxSTUV = _mm512_loadu_ps(x + 112); x += 128; vx0123 = _mm512_mul_ps(vx0123, vscale); vx4567 = _mm512_mul_ps(vx4567, vscale); vx89AB = _mm512_mul_ps(vx89AB, vscale); vxCDEF = _mm512_mul_ps(vxCDEF, vscale); vxGHIJ = _mm512_mul_ps(vxGHIJ, vscale); vxKLMN = _mm512_mul_ps(vxKLMN, vscale); vxOPQR = _mm512_mul_ps(vxOPQR, vscale); vxSTUV = _mm512_mul_ps(vxSTUV, vscale); vx0123 = _mm512_min_ps(vx0123, voutput_max_less_zero_point); vx4567 = _mm512_min_ps(vx4567, voutput_max_less_zero_point); vx89AB = _mm512_min_ps(vx89AB, voutput_max_less_zero_point); vxCDEF = _mm512_min_ps(vxCDEF, voutput_max_less_zero_point); vxGHIJ = _mm512_min_ps(vxGHIJ, voutput_max_less_zero_point); vxKLMN = _mm512_min_ps(vxKLMN, voutput_max_less_zero_point); vxOPQR = _mm512_min_ps(vxOPQR, voutput_max_less_zero_point); vxSTUV = _mm512_min_ps(vxSTUV, voutput_max_less_zero_point); const __m512i vacc0123 = _mm512_cvtps_epi32(vx0123); const __m512i vacc4567 = _mm512_cvtps_epi32(vx4567); const __m512i vacc89AB = _mm512_cvtps_epi32(vx89AB); const __m512i vaccCDEF = _mm512_cvtps_epi32(vxCDEF); const __m512i vaccGHIJ = _mm512_cvtps_epi32(vxGHIJ); const __m512i vaccKLMN = _mm512_cvtps_epi32(vxKLMN); const __m512i vaccOPQR = _mm512_cvtps_epi32(vxOPQR); const __m512i vaccSTUV = _mm512_cvtps_epi32(vxSTUV); __m512i vacc04152637 = _mm512_packs_epi32(vacc0123, vacc4567); __m512i vacc8C9DAEBF = _mm512_packs_epi32(vacc89AB, vaccCDEF); __m512i vaccGKHLIMJN = _mm512_packs_epi32(vaccGHIJ, vaccKLMN); __m512i vaccOSPTQURV = _mm512_packs_epi32(vaccOPQR, vaccSTUV); vacc04152637 = _mm512_adds_epi16(vacc04152637, voutput_zero_point); vacc8C9DAEBF = _mm512_adds_epi16(vacc8C9DAEBF, voutput_zero_point); vaccGKHLIMJN = _mm512_adds_epi16(vaccGKHLIMJN, voutput_zero_point); vaccOSPTQURV = _mm512_adds_epi16(vaccOSPTQURV, voutput_zero_point); __m512i vy048C159D26AE37BF = _mm512_packs_epi16(vacc04152637, vacc8C9DAEBF); __m512i vyGKOSHLPTIMQUJNRV = _mm512_packs_epi16(vaccGKHLIMJN, vaccOSPTQURV); vy048C159D26AE37BF = _mm512_max_epi8(vy048C159D26AE37BF, voutput_min); vyGKOSHLPTIMQUJNRV = _mm512_max_epi8(vyGKOSHLPTIMQUJNRV, voutput_min); const __m512i vy0123456789ABCDEF = _mm512_permutexvar_epi32(vshuffle512_mask, vy048C159D26AE37BF); const __m512i vyGHIJKLMNOPQRSTUV = _mm512_permutexvar_epi32(vshuffle512_mask, vyGKOSHLPTIMQUJNRV); _mm512_storeu_si512(y, vy0123456789ABCDEF); _mm512_storeu_si512(y + 64, vyGHIJKLMNOPQRSTUV); y += 128; } for (; n >= 16 * sizeof(float); n -= 16 * sizeof(float)) { __m512 vx0123 = _mm512_loadu_ps(x); vx0123 = _mm512_mul_ps(vx0123, vscale); vx0123 = _mm512_min_ps(vx0123, voutput_max_less_zero_point); x += 16; const __m512i vacc0123 = _mm512_cvtps_epi32(vx0123); __m256i vacc0213 = _mm256_packs_epi32(_mm512_castsi512_si256(vacc0123), _mm512_extracti32x8_epi32(vacc0123, 1)); vacc0213 = _mm256_adds_epi16(vacc0213, _mm512_castsi512_si256(voutput_zero_point)); const __m128i vy0213 = _mm_packs_epi16(_mm256_castsi256_si128(vacc0213), _mm256_extracti128_si256(vacc0213, 1)); __m128i vy0123 = _mm_shuffle_epi32(vy0213, _MM_SHUFFLE(3, 1, 2, 0)); vy0123 = _mm_max_epi8(vy0123, _mm512_castsi512_si128(voutput_min)); _mm_storeu_si128((__m128i*) y, vy0123); y += 16; } if XNN_UNLIKELY(n != 0) { assert(n >= 1 * sizeof(float)); assert(n <= 15 * sizeof(float)); // Prepare mask for valid elements (depends on n). n >>= 2 /* log2(sizeof(float)) */; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << n) - UINT32_C(1))); __m512 vx0123 = _mm512_maskz_loadu_ps(vmask, x); vx0123 = _mm512_mul_ps(vx0123, vscale); vx0123 = _mm512_min_ps(vx0123, voutput_max_less_zero_point); const __m512i vacc0123 = _mm512_cvtps_epi32(vx0123); __m256i vacc0213 = _mm256_packs_epi32(_mm512_castsi512_si256(vacc0123), _mm512_extracti32x8_epi32(vacc0123, 1)); vacc0213 = _mm256_adds_epi16(vacc0213, _mm512_castsi512_si256(voutput_zero_point)); const __m128i vy0213 = _mm_packs_epi16(_mm256_castsi256_si128(vacc0213), _mm256_extracti128_si256(vacc0213, 1)); __m128i vy0123 = _mm_shuffle_epi32(vy0213, _MM_SHUFFLE(3, 1, 2, 0)); vy0123 = _mm_max_epi8(vy0123, _mm512_castsi512_si128(voutput_min)); _mm_mask_storeu_epi8(y, vmask, vy0123); } } void xnn_f32_qu8_vcvt_ukernel__avx512skx_x128( size_t n, const float* x, uint8_t* y, const union xnn_f32_qu8_cvt_params params[restrict XNN_MIN_ELEMENTS(1)]) { assert(n != 0); assert(n % sizeof(float) == 0); assert(x != NULL); assert(y != NULL); const __m512 vscale = _mm512_load_ps(params->avx2.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->avx512.output_zero_point); const __m512i vshuffle512_mask = _mm512_load_si512(params->avx512.shuffle512_mask); const __m512i voutput_min = _mm512_load_si512(params->avx512.output_min); for (; n >= 128 * sizeof(float); n -= 128 * sizeof(float)) { __m512 vx0123 = _mm512_loadu_ps(x); __m512 vx4567 = _mm512_loadu_ps(x + 16); __m512 vx89AB = _mm512_loadu_ps(x + 32); __m512 vxCDEF = _mm512_loadu_ps(x + 48); __m512 vxGHIJ = _mm512_loadu_ps(x + 64); __m512 vxKLMN = _mm512_loadu_ps(x + 80); __m512 vxOPQR = _mm512_loadu_ps(x + 96); __m512 vxSTUV = _mm512_loadu_ps(x + 112); x += 128; vx0123 = _mm512_mul_ps(vx0123, vscale); vx4567 = _mm512_mul_ps(vx4567, vscale); vx89AB = _mm512_mul_ps(vx89AB, vscale); vxCDEF = _mm512_mul_ps(vxCDEF, vscale); vxGHIJ = _mm512_mul_ps(vxGHIJ, vscale); vxKLMN = _mm512_mul_ps(vxKLMN, vscale); vxOPQR = _mm512_mul_ps(vxOPQR, vscale); vxSTUV = _mm512_mul_ps(vxSTUV, vscale); vx0123 = _mm512_min_ps(vx0123, voutput_max_less_zero_point); vx4567 = _mm512_min_ps(vx4567, voutput_max_less_zero_point); vx89AB = _mm512_min_ps(vx89AB, voutput_max_less_zero_point); vxCDEF = _mm512_min_ps(vxCDEF, voutput_max_less_zero_point); vxGHIJ = _mm512_min_ps(vxGHIJ, voutput_max_less_zero_point); vxKLMN = _mm512_min_ps(vxKLMN, voutput_max_less_zero_point); vxOPQR = _mm512_min_ps(vxOPQR, voutput_max_less_zero_point); vxSTUV = _mm512_min_ps(vxSTUV, voutput_max_less_zero_point); const __m512i vacc0123 = _mm512_cvtps_epi32(vx0123); const __m512i vacc4567 = _mm512_cvtps_epi32(vx4567); const __m512i vacc89AB = _mm512_cvtps_epi32(vx89AB); const __m512i vaccCDEF = _mm512_cvtps_epi32(vxCDEF); const __m512i vaccGHIJ = _mm512_cvtps_epi32(vxGHIJ); const __m512i vaccKLMN = _mm512_cvtps_epi32(vxKLMN); const __m512i vaccOPQR = _mm512_cvtps_epi32(vxOPQR); const __m512i vaccSTUV = _mm512_cvtps_epi32(vxSTUV); __m512i vacc04152637 = _mm512_packs_epi32(vacc0123, vacc4567); __m512i vacc8C9DAEBF = _mm512_packs_epi32(vacc89AB, vaccCDEF); __m512i vaccGKHLIMJN = _mm512_packs_epi32(vaccGHIJ, vaccKLMN); __m512i vaccOSPTQURV = _mm512_packs_epi32(vaccOPQR, vaccSTUV); vacc04152637 = _mm512_adds_epi16(vacc04152637, voutput_zero_point); vacc8C9DAEBF = _mm512_adds_epi16(vacc8C9DAEBF, voutput_zero_point); vaccGKHLIMJN = _mm512_adds_epi16(vaccGKHLIMJN, voutput_zero_point); vaccOSPTQURV = _mm512_adds_epi16(vaccOSPTQURV, voutput_zero_point); __m512i vy048C159D26AE37BF = _mm512_packus_epi16(vacc04152637, vacc8C9DAEBF); __m512i vyGKOSHLPTIMQUJNRV = _mm512_packus_epi16(vaccGKHLIMJN, vaccOSPTQURV); vy048C159D26AE37BF = _mm512_max_epu8(vy048C159D26AE37BF, voutput_min); vyGKOSHLPTIMQUJNRV = _mm512_max_epu8(vyGKOSHLPTIMQUJNRV, voutput_min); const __m512i vy0123456789ABCDEF = _mm512_permutexvar_epi32(vshuffle512_mask, vy048C159D26AE37BF); const __m512i vyGHIJKLMNOPQRSTUV = _mm512_permutexvar_epi32(vshuffle512_mask, vyGKOSHLPTIMQUJNRV); _mm512_storeu_si512(y, vy0123456789ABCDEF); _mm512_storeu_si512(y + 64, vyGHIJKLMNOPQRSTUV); y += 128; } for (; n >= 16 * sizeof(float); n -= 16 * sizeof(float)) { __m512 vx0123 = _mm512_loadu_ps(x); vx0123 = _mm512_mul_ps(vx0123, vscale); vx0123 = _mm512_min_ps(vx0123, voutput_max_less_zero_point); x += 16; const __m512i vacc0123 = _mm512_cvtps_epi32(vx0123); __m256i vacc0213 = _mm256_packs_epi32(_mm512_castsi512_si256(vacc0123), _mm512_extracti32x8_epi32(vacc0123, 1)); vacc0213 = _mm256_adds_epi16(vacc0213, _mm512_castsi512_si256(voutput_zero_point)); const __m128i vy0213 = _mm_packus_epi16(_mm256_castsi256_si128(vacc0213), _mm256_extracti128_si256(vacc0213, 1)); __m128i vy0123 = _mm_shuffle_epi32(vy0213, _MM_SHUFFLE(3, 1, 2, 0)); vy0123 = _mm_max_epu8(vy0123, _mm512_castsi512_si128(voutput_min)); _mm_storeu_si128((__m128i*) y, vy0123); y += 16; } if XNN_UNLIKELY(n != 0) { assert(n >= 1 * sizeof(float)); assert(n <= 15 * sizeof(float)); // Prepare mask for valid elements (depends on n). n >>= 2 /* log2(sizeof(float)) */; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << n) - UINT32_C(1))); __m512 vx0123 = _mm512_maskz_loadu_ps(vmask, x); vx0123 = _mm512_mul_ps(vx0123, vscale); vx0123 = _mm512_min_ps(vx0123, voutput_max_less_zero_point); const __m512i vacc0123 = _mm512_cvtps_epi32(vx0123); __m256i vacc0213 = _mm256_packs_epi32(_mm512_castsi512_si256(vacc0123), _mm512_extracti32x8_epi32(vacc0123, 1)); vacc0213 = _mm256_adds_epi16(vacc0213, _mm512_castsi512_si256(voutput_zero_point)); const __m128i vy0213 = _mm_packus_epi16(_mm256_castsi256_si128(vacc0213), _mm256_extracti128_si256(vacc0213, 1)); __m128i vy0123 = _mm_shuffle_epi32(vy0213, _MM_SHUFFLE(3, 1, 2, 0)); vy0123 = _mm_max_epu8(vy0123, _mm512_castsi512_si128(voutput_min)); _mm_mask_storeu_epi8(y, vmask, vy0123); } } void xnn_qc8_dwconv_minmax_fp32_ukernel_up32x25__avx512skx_mul32( 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_qc8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_MSAN { assert(channels != 0); assert(output_width != 0); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m256i voutput_min = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_min); const __m256i vpermute_mask = _mm256_set_epi32(7, 3, 5, 1, 6, 2, 4, 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); } const int8_t* i9 = input[9]; assert(i9 != NULL); if XNN_UNPREDICTABLE(i9 != zero) { i9 = (const int8_t*) ((uintptr_t) i9 + input_offset); } const int8_t* i10 = input[10]; assert(i10 != NULL); if XNN_UNPREDICTABLE(i10 != zero) { i10 = (const int8_t*) ((uintptr_t) i10 + input_offset); } const int8_t* i11 = input[11]; assert(i11 != NULL); if XNN_UNPREDICTABLE(i11 != zero) { i11 = (const int8_t*) ((uintptr_t) i11 + input_offset); } const int8_t* i12 = input[12]; assert(i12 != NULL); if XNN_UNPREDICTABLE(i12 != zero) { i12 = (const int8_t*) ((uintptr_t) i12 + input_offset); } const int8_t* i13 = input[13]; assert(i13 != NULL); if XNN_UNPREDICTABLE(i13 != zero) { i13 = (const int8_t*) ((uintptr_t) i13 + input_offset); } const int8_t* i14 = input[14]; assert(i14 != NULL); if XNN_UNPREDICTABLE(i14 != zero) { i14 = (const int8_t*) ((uintptr_t) i14 + input_offset); } const int8_t* i15 = input[15]; assert(i15 != NULL); if XNN_UNPREDICTABLE(i15 != zero) { i15 = (const int8_t*) ((uintptr_t) i15 + input_offset); } const int8_t* i16 = input[16]; assert(i16 != NULL); if XNN_UNPREDICTABLE(i16 != zero) { i16 = (const int8_t*) ((uintptr_t) i16 + input_offset); } const int8_t* i17 = input[17]; assert(i17 != NULL); if XNN_UNPREDICTABLE(i17 != zero) { i17 = (const int8_t*) ((uintptr_t) i17 + input_offset); } const int8_t* i18 = input[18]; assert(i18 != NULL); if XNN_UNPREDICTABLE(i18 != zero) { i18 = (const int8_t*) ((uintptr_t) i18 + input_offset); } const int8_t* i19 = input[19]; assert(i19 != NULL); if XNN_UNPREDICTABLE(i19 != zero) { i19 = (const int8_t*) ((uintptr_t) i19 + input_offset); } const int8_t* i20 = input[20]; assert(i20 != NULL); if XNN_UNPREDICTABLE(i20 != zero) { i20 = (const int8_t*) ((uintptr_t) i20 + input_offset); } const int8_t* i21 = input[21]; assert(i21 != NULL); if XNN_UNPREDICTABLE(i21 != zero) { i21 = (const int8_t*) ((uintptr_t) i21 + input_offset); } const int8_t* i22 = input[22]; assert(i22 != NULL); if XNN_UNPREDICTABLE(i22 != zero) { i22 = (const int8_t*) ((uintptr_t) i22 + input_offset); } const int8_t* i23 = input[23]; assert(i23 != NULL); if XNN_UNPREDICTABLE(i23 != zero) { i23 = (const int8_t*) ((uintptr_t) i23 + input_offset); } const int8_t* i24 = input[24]; assert(i24 != NULL); if XNN_UNPREDICTABLE(i24 != zero) { i24 = (const int8_t*) ((uintptr_t) i24 + input_offset); } input = (const int8_t**) ((uintptr_t) input + input_stride); size_t c = channels; const void* w = weights; for (; c >= 32; c -= 32) { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); __m512i vaccGHIJKLMNOPQRSTUV = _mm512_loadu_si512((const void*) ((uintptr_t) w + 16 * sizeof(int32_t))); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 0 * sizeof(int8_t)))); const __m512i vi0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m512i vk0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 16 * sizeof(int8_t)))); i0 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi0xGHIJKLMNOPQRSTUV, vk0xGHIJKLMNOPQRSTUV)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 32 * sizeof(int8_t)))); const __m512i vi1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m512i vk1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 48 * sizeof(int8_t)))); i1 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi1xGHIJKLMNOPQRSTUV, vk1xGHIJKLMNOPQRSTUV)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 64 * sizeof(int8_t)))); const __m512i vi2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m512i vk2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 80 * sizeof(int8_t)))); i2 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi2xGHIJKLMNOPQRSTUV, vk2xGHIJKLMNOPQRSTUV)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 96 * sizeof(int8_t)))); const __m512i vi3xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i3 + 16))); const __m512i vk3xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 112 * sizeof(int8_t)))); i3 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi3xGHIJKLMNOPQRSTUV, vk3xGHIJKLMNOPQRSTUV)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 128 * sizeof(int8_t)))); const __m512i vi4xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i4 + 16))); const __m512i vk4xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 144 * sizeof(int8_t)))); i4 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi4xGHIJKLMNOPQRSTUV, vk4xGHIJKLMNOPQRSTUV)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 160 * sizeof(int8_t)))); const __m512i vi5xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i5 + 16))); const __m512i vk5xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 176 * sizeof(int8_t)))); i5 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi5xGHIJKLMNOPQRSTUV, vk5xGHIJKLMNOPQRSTUV)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 192 * sizeof(int8_t)))); const __m512i vi6xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i6 + 16))); const __m512i vk6xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 208 * sizeof(int8_t)))); i6 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi6xGHIJKLMNOPQRSTUV, vk6xGHIJKLMNOPQRSTUV)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 224 * sizeof(int8_t)))); const __m512i vi7xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i7 + 16))); const __m512i vk7xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 240 * sizeof(int8_t)))); i7 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi7xGHIJKLMNOPQRSTUV, vk7xGHIJKLMNOPQRSTUV)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 256 * sizeof(int8_t)))); const __m512i vi8xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i8 + 16))); const __m512i vk8xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 272 * sizeof(int8_t)))); i8 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi8xGHIJKLMNOPQRSTUV, vk8xGHIJKLMNOPQRSTUV)); const __m512i vi9x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i9)); const __m512i vk9x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 288 * sizeof(int8_t)))); const __m512i vi9xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i9 + 16))); const __m512i vk9xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 304 * sizeof(int8_t)))); i9 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi9x0123456789ABCDEF, vk9x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi9xGHIJKLMNOPQRSTUV, vk9xGHIJKLMNOPQRSTUV)); const __m512i vi10x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i10)); const __m512i vk10x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 320 * sizeof(int8_t)))); const __m512i vi10xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i10 + 16))); const __m512i vk10xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 336 * sizeof(int8_t)))); i10 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi10x0123456789ABCDEF, vk10x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi10xGHIJKLMNOPQRSTUV, vk10xGHIJKLMNOPQRSTUV)); const __m512i vi11x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i11)); const __m512i vk11x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 352 * sizeof(int8_t)))); const __m512i vi11xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i11 + 16))); const __m512i vk11xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 368 * sizeof(int8_t)))); i11 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi11x0123456789ABCDEF, vk11x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi11xGHIJKLMNOPQRSTUV, vk11xGHIJKLMNOPQRSTUV)); const __m512i vi12x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i12)); const __m512i vk12x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 384 * sizeof(int8_t)))); const __m512i vi12xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i12 + 16))); const __m512i vk12xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 400 * sizeof(int8_t)))); i12 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi12x0123456789ABCDEF, vk12x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi12xGHIJKLMNOPQRSTUV, vk12xGHIJKLMNOPQRSTUV)); const __m512i vi13x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i13)); const __m512i vk13x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 416 * sizeof(int8_t)))); const __m512i vi13xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i13 + 16))); const __m512i vk13xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 432 * sizeof(int8_t)))); i13 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi13x0123456789ABCDEF, vk13x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi13xGHIJKLMNOPQRSTUV, vk13xGHIJKLMNOPQRSTUV)); const __m512i vi14x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i14)); const __m512i vk14x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 448 * sizeof(int8_t)))); const __m512i vi14xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i14 + 16))); const __m512i vk14xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 464 * sizeof(int8_t)))); i14 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi14x0123456789ABCDEF, vk14x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi14xGHIJKLMNOPQRSTUV, vk14xGHIJKLMNOPQRSTUV)); const __m512i vi15x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i15)); const __m512i vk15x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 480 * sizeof(int8_t)))); const __m512i vi15xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i15 + 16))); const __m512i vk15xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 496 * sizeof(int8_t)))); i15 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi15x0123456789ABCDEF, vk15x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi15xGHIJKLMNOPQRSTUV, vk15xGHIJKLMNOPQRSTUV)); const __m512i vi16x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i16)); const __m512i vk16x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 512 * sizeof(int8_t)))); const __m512i vi16xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i16 + 16))); const __m512i vk16xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 528 * sizeof(int8_t)))); i16 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi16x0123456789ABCDEF, vk16x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi16xGHIJKLMNOPQRSTUV, vk16xGHIJKLMNOPQRSTUV)); const __m512i vi17x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i17)); const __m512i vk17x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 544 * sizeof(int8_t)))); const __m512i vi17xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i17 + 16))); const __m512i vk17xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 560 * sizeof(int8_t)))); i17 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi17x0123456789ABCDEF, vk17x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi17xGHIJKLMNOPQRSTUV, vk17xGHIJKLMNOPQRSTUV)); const __m512i vi18x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i18)); const __m512i vk18x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 576 * sizeof(int8_t)))); const __m512i vi18xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i18 + 16))); const __m512i vk18xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 592 * sizeof(int8_t)))); i18 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi18x0123456789ABCDEF, vk18x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi18xGHIJKLMNOPQRSTUV, vk18xGHIJKLMNOPQRSTUV)); const __m512i vi19x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i19)); const __m512i vk19x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 608 * sizeof(int8_t)))); const __m512i vi19xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i19 + 16))); const __m512i vk19xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 624 * sizeof(int8_t)))); i19 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi19x0123456789ABCDEF, vk19x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi19xGHIJKLMNOPQRSTUV, vk19xGHIJKLMNOPQRSTUV)); const __m512i vi20x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i20)); const __m512i vk20x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 640 * sizeof(int8_t)))); const __m512i vi20xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i20 + 16))); const __m512i vk20xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 656 * sizeof(int8_t)))); i20 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi20x0123456789ABCDEF, vk20x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi20xGHIJKLMNOPQRSTUV, vk20xGHIJKLMNOPQRSTUV)); const __m512i vi21x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i21)); const __m512i vk21x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 672 * sizeof(int8_t)))); const __m512i vi21xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i21 + 16))); const __m512i vk21xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 688 * sizeof(int8_t)))); i21 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi21x0123456789ABCDEF, vk21x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi21xGHIJKLMNOPQRSTUV, vk21xGHIJKLMNOPQRSTUV)); const __m512i vi22x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i22)); const __m512i vk22x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 704 * sizeof(int8_t)))); const __m512i vi22xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i22 + 16))); const __m512i vk22xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 720 * sizeof(int8_t)))); i22 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi22x0123456789ABCDEF, vk22x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi22xGHIJKLMNOPQRSTUV, vk22xGHIJKLMNOPQRSTUV)); const __m512i vi23x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i23)); const __m512i vk23x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 736 * sizeof(int8_t)))); const __m512i vi23xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i23 + 16))); const __m512i vk23xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 752 * sizeof(int8_t)))); i23 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi23x0123456789ABCDEF, vk23x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi23xGHIJKLMNOPQRSTUV, vk23xGHIJKLMNOPQRSTUV)); const __m512i vi24x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i24)); const __m512i vk24x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 768 * sizeof(int8_t)))); const __m512i vi24xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i24 + 16))); const __m512i vk24xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 784 * sizeof(int8_t)))); i24 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi24x0123456789ABCDEF, vk24x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi24xGHIJKLMNOPQRSTUV, vk24xGHIJKLMNOPQRSTUV)); w = (const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 800 * sizeof(int8_t)); __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); __m512 vscaledGHIJKLMNOPQRSTUV = _mm512_cvtepi32_ps(vaccGHIJKLMNOPQRSTUV); const __m512 vscale0123456789ABCDEF = _mm512_loadu_ps(w); const __m512 vscaleGHIJKLMNOPQRSTUV = _mm512_loadu_ps((const void*) ((uintptr_t) w + 16 * sizeof(float))); w = (const void*) ((uintptr_t) w + 32 * sizeof(float)); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale0123456789ABCDEF); vscaledGHIJKLMNOPQRSTUV = _mm512_mul_ps(vscaledGHIJKLMNOPQRSTUV, vscaleGHIJKLMNOPQRSTUV); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vscaledGHIJKLMNOPQRSTUV = _mm512_min_ps(vscaledGHIJKLMNOPQRSTUV, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); vaccGHIJKLMNOPQRSTUV = _mm512_cvtps_epi32(vscaledGHIJKLMNOPQRSTUV); __m512i vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV = _mm512_adds_epi16(_mm512_packs_epi32(vacc0123456789ABCDEF, vaccGHIJKLMNOPQRSTUV), voutput_zero_point); __m256i voutGHIJOPQRKLMNSTUV = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vaccGHIJKLMNOPQRSTUV), _mm512_extracti32x8_epi32(vaccGHIJKLMNOPQRSTUV, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m256i vout0123GHIJ4567KLMN = _mm512_castsi512_si256(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV); const __m256i vout89ABOPQRCDEFSTUV = _mm512_extracti32x8_epi32(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV, 1); const __m256i vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV = _mm256_packs_epi16(vout0123GHIJ4567KLMN, vout89ABOPQRCDEFSTUV); __m256i vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_permutevar8x32_epi32(vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV, vpermute_mask); const __m128i voutGHIJOPQR = _mm256_castsi256_si128(voutGHIJOPQRKLMNSTUV); const __m128i voutKLMNSTUV = _mm256_extracti128_si256(voutGHIJOPQRKLMNSTUV, 1); __m128i voutGHIJKLMNOPQRSTUV = _mm_shuffle_epi32(_mm_packs_epi16(voutGHIJOPQR, voutKLMNSTUV), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_max_epi8(vout0123456789ABCDEFGHIJKLMNOPQRSTUV, voutput_min); voutGHIJKLMNOPQRSTUV = _mm_max_epi8(voutGHIJKLMNOPQRSTUV, _mm256_castsi256_si128(voutput_min)); _mm256_storeu_si256((__m256i*) output, vout0123456789ABCDEFGHIJKLMNOPQRSTUV); _mm_storeu_si128((__m128i*) (output + 16), voutGHIJKLMNOPQRSTUV); output += 32; } if XNN_UNLIKELY(c != 0) { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << (c & 15)) - UINT32_C(1))); const int8_t* k = (const int8_t*) ((uintptr_t) w + 32 * sizeof(int32_t)); do { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) k)); i0 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 32))); i1 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 64))); i2 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 96))); i3 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 128))); i4 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 160))); i5 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 192))); i6 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 224))); i7 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 256))); i8 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); const __m512i vi9x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i9)); const __m512i vk9x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 288))); i9 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi9x0123456789ABCDEF, vk9x0123456789ABCDEF)); const __m512i vi10x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i10)); const __m512i vk10x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 320))); i10 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi10x0123456789ABCDEF, vk10x0123456789ABCDEF)); const __m512i vi11x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i11)); const __m512i vk11x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 352))); i11 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi11x0123456789ABCDEF, vk11x0123456789ABCDEF)); const __m512i vi12x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i12)); const __m512i vk12x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 384))); i12 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi12x0123456789ABCDEF, vk12x0123456789ABCDEF)); const __m512i vi13x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i13)); const __m512i vk13x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 416))); i13 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi13x0123456789ABCDEF, vk13x0123456789ABCDEF)); const __m512i vi14x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i14)); const __m512i vk14x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 448))); i14 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi14x0123456789ABCDEF, vk14x0123456789ABCDEF)); const __m512i vi15x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i15)); const __m512i vk15x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 480))); i15 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi15x0123456789ABCDEF, vk15x0123456789ABCDEF)); const __m512i vi16x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i16)); const __m512i vk16x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 512))); i16 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi16x0123456789ABCDEF, vk16x0123456789ABCDEF)); const __m512i vi17x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i17)); const __m512i vk17x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 544))); i17 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi17x0123456789ABCDEF, vk17x0123456789ABCDEF)); const __m512i vi18x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i18)); const __m512i vk18x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 576))); i18 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi18x0123456789ABCDEF, vk18x0123456789ABCDEF)); const __m512i vi19x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i19)); const __m512i vk19x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 608))); i19 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi19x0123456789ABCDEF, vk19x0123456789ABCDEF)); const __m512i vi20x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i20)); const __m512i vk20x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 640))); i20 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi20x0123456789ABCDEF, vk20x0123456789ABCDEF)); const __m512i vi21x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i21)); const __m512i vk21x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 672))); i21 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi21x0123456789ABCDEF, vk21x0123456789ABCDEF)); const __m512i vi22x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i22)); const __m512i vk22x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 704))); i22 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi22x0123456789ABCDEF, vk22x0123456789ABCDEF)); const __m512i vi23x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i23)); const __m512i vk23x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 736))); i23 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi23x0123456789ABCDEF, vk23x0123456789ABCDEF)); const __m512i vi24x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i24)); const __m512i vk24x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 768))); i24 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi24x0123456789ABCDEF, vk24x0123456789ABCDEF)); k += 16; __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); const __m512 vscale0123456789ABCDEF = _mm512_loadu_ps((const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 800 * sizeof(int8_t))); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale0123456789ABCDEF); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); w = (const void*) ((uintptr_t) w + 16 * sizeof(int32_t)); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m128i vout012389AB = _mm256_castsi256_si128(vout012389AB4567CDEF); const __m128i vout4567CDEF = _mm256_extracti128_si256(vout012389AB4567CDEF, 1); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packs_epi16(vout012389AB, vout4567CDEF), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, _mm256_castsi256_si128(voutput_min)); if XNN_LIKELY(c >= 16) { _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; c -= 16; } else { _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); output = (int8_t*) ((uintptr_t) output + c); c = 0; } } while (c != 0); } output = (int8_t*) ((uintptr_t) output + output_increment); } while (--output_width != 0); } void xnn_qc8_dwconv_minmax_fp32_ukernel_up32x3__avx512skx_mul32( 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_qc8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_MSAN { assert(channels != 0); assert(output_width != 0); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m256i voutput_min = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_min); const __m256i vpermute_mask = _mm256_set_epi32(7, 3, 5, 1, 6, 2, 4, 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); } input = (const int8_t**) ((uintptr_t) input + input_stride); size_t c = channels; const void* w = weights; for (; c >= 32; c -= 32) { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); __m512i vaccGHIJKLMNOPQRSTUV = _mm512_loadu_si512((const void*) ((uintptr_t) w + 16 * sizeof(int32_t))); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 0 * sizeof(int8_t)))); const __m512i vi0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m512i vk0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 16 * sizeof(int8_t)))); i0 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi0xGHIJKLMNOPQRSTUV, vk0xGHIJKLMNOPQRSTUV)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 32 * sizeof(int8_t)))); const __m512i vi1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m512i vk1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 48 * sizeof(int8_t)))); i1 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi1xGHIJKLMNOPQRSTUV, vk1xGHIJKLMNOPQRSTUV)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 64 * sizeof(int8_t)))); const __m512i vi2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m512i vk2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 80 * sizeof(int8_t)))); i2 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi2xGHIJKLMNOPQRSTUV, vk2xGHIJKLMNOPQRSTUV)); w = (const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 96 * sizeof(int8_t)); __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); __m512 vscaledGHIJKLMNOPQRSTUV = _mm512_cvtepi32_ps(vaccGHIJKLMNOPQRSTUV); const __m512 vscale0123456789ABCDEF = _mm512_loadu_ps(w); const __m512 vscaleGHIJKLMNOPQRSTUV = _mm512_loadu_ps((const void*) ((uintptr_t) w + 16 * sizeof(float))); w = (const void*) ((uintptr_t) w + 32 * sizeof(float)); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale0123456789ABCDEF); vscaledGHIJKLMNOPQRSTUV = _mm512_mul_ps(vscaledGHIJKLMNOPQRSTUV, vscaleGHIJKLMNOPQRSTUV); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vscaledGHIJKLMNOPQRSTUV = _mm512_min_ps(vscaledGHIJKLMNOPQRSTUV, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); vaccGHIJKLMNOPQRSTUV = _mm512_cvtps_epi32(vscaledGHIJKLMNOPQRSTUV); __m512i vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV = _mm512_adds_epi16(_mm512_packs_epi32(vacc0123456789ABCDEF, vaccGHIJKLMNOPQRSTUV), voutput_zero_point); __m256i voutGHIJOPQRKLMNSTUV = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vaccGHIJKLMNOPQRSTUV), _mm512_extracti32x8_epi32(vaccGHIJKLMNOPQRSTUV, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m256i vout0123GHIJ4567KLMN = _mm512_castsi512_si256(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV); const __m256i vout89ABOPQRCDEFSTUV = _mm512_extracti32x8_epi32(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV, 1); const __m256i vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV = _mm256_packs_epi16(vout0123GHIJ4567KLMN, vout89ABOPQRCDEFSTUV); __m256i vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_permutevar8x32_epi32(vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV, vpermute_mask); const __m128i voutGHIJOPQR = _mm256_castsi256_si128(voutGHIJOPQRKLMNSTUV); const __m128i voutKLMNSTUV = _mm256_extracti128_si256(voutGHIJOPQRKLMNSTUV, 1); __m128i voutGHIJKLMNOPQRSTUV = _mm_shuffle_epi32(_mm_packs_epi16(voutGHIJOPQR, voutKLMNSTUV), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_max_epi8(vout0123456789ABCDEFGHIJKLMNOPQRSTUV, voutput_min); voutGHIJKLMNOPQRSTUV = _mm_max_epi8(voutGHIJKLMNOPQRSTUV, _mm256_castsi256_si128(voutput_min)); _mm256_storeu_si256((__m256i*) output, vout0123456789ABCDEFGHIJKLMNOPQRSTUV); _mm_storeu_si128((__m128i*) (output + 16), voutGHIJKLMNOPQRSTUV); output += 32; } if XNN_UNLIKELY(c != 0) { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << (c & 15)) - UINT32_C(1))); const int8_t* k = (const int8_t*) ((uintptr_t) w + 32 * sizeof(int32_t)); do { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) k)); i0 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 32))); i1 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 64))); i2 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); k += 16; __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); const __m512 vscale0123456789ABCDEF = _mm512_loadu_ps((const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 96 * sizeof(int8_t))); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale0123456789ABCDEF); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); w = (const void*) ((uintptr_t) w + 16 * sizeof(int32_t)); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m128i vout012389AB = _mm256_castsi256_si128(vout012389AB4567CDEF); const __m128i vout4567CDEF = _mm256_extracti128_si256(vout012389AB4567CDEF, 1); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packs_epi16(vout012389AB, vout4567CDEF), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, _mm256_castsi256_si128(voutput_min)); if XNN_LIKELY(c >= 16) { _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; c -= 16; } else { _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); output = (int8_t*) ((uintptr_t) output + c); c = 0; } } while (c != 0); } output = (int8_t*) ((uintptr_t) output + output_increment); } while (--output_width != 0); } void xnn_qc8_dwconv_minmax_fp32_ukernel_up32x9__avx512skx_mul32( 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_qc8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_MSAN { assert(channels != 0); assert(output_width != 0); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m256i voutput_min = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_min); const __m256i vpermute_mask = _mm256_set_epi32(7, 3, 5, 1, 6, 2, 4, 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) { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); __m512i vaccGHIJKLMNOPQRSTUV = _mm512_loadu_si512((const void*) ((uintptr_t) w + 16 * sizeof(int32_t))); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 0 * sizeof(int8_t)))); const __m512i vi0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m512i vk0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 16 * sizeof(int8_t)))); i0 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi0xGHIJKLMNOPQRSTUV, vk0xGHIJKLMNOPQRSTUV)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 32 * sizeof(int8_t)))); const __m512i vi1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m512i vk1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 48 * sizeof(int8_t)))); i1 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi1xGHIJKLMNOPQRSTUV, vk1xGHIJKLMNOPQRSTUV)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 64 * sizeof(int8_t)))); const __m512i vi2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m512i vk2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 80 * sizeof(int8_t)))); i2 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi2xGHIJKLMNOPQRSTUV, vk2xGHIJKLMNOPQRSTUV)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 96 * sizeof(int8_t)))); const __m512i vi3xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i3 + 16))); const __m512i vk3xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 112 * sizeof(int8_t)))); i3 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi3xGHIJKLMNOPQRSTUV, vk3xGHIJKLMNOPQRSTUV)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 128 * sizeof(int8_t)))); const __m512i vi4xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i4 + 16))); const __m512i vk4xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 144 * sizeof(int8_t)))); i4 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi4xGHIJKLMNOPQRSTUV, vk4xGHIJKLMNOPQRSTUV)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 160 * sizeof(int8_t)))); const __m512i vi5xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i5 + 16))); const __m512i vk5xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 176 * sizeof(int8_t)))); i5 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi5xGHIJKLMNOPQRSTUV, vk5xGHIJKLMNOPQRSTUV)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 192 * sizeof(int8_t)))); const __m512i vi6xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i6 + 16))); const __m512i vk6xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 208 * sizeof(int8_t)))); i6 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi6xGHIJKLMNOPQRSTUV, vk6xGHIJKLMNOPQRSTUV)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 224 * sizeof(int8_t)))); const __m512i vi7xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i7 + 16))); const __m512i vk7xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 240 * sizeof(int8_t)))); i7 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi7xGHIJKLMNOPQRSTUV, vk7xGHIJKLMNOPQRSTUV)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 256 * sizeof(int8_t)))); const __m512i vi8xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i8 + 16))); const __m512i vk8xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 272 * sizeof(int8_t)))); i8 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi8xGHIJKLMNOPQRSTUV, vk8xGHIJKLMNOPQRSTUV)); w = (const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 288 * sizeof(int8_t)); __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); __m512 vscaledGHIJKLMNOPQRSTUV = _mm512_cvtepi32_ps(vaccGHIJKLMNOPQRSTUV); const __m512 vscale0123456789ABCDEF = _mm512_loadu_ps(w); const __m512 vscaleGHIJKLMNOPQRSTUV = _mm512_loadu_ps((const void*) ((uintptr_t) w + 16 * sizeof(float))); w = (const void*) ((uintptr_t) w + 32 * sizeof(float)); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale0123456789ABCDEF); vscaledGHIJKLMNOPQRSTUV = _mm512_mul_ps(vscaledGHIJKLMNOPQRSTUV, vscaleGHIJKLMNOPQRSTUV); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vscaledGHIJKLMNOPQRSTUV = _mm512_min_ps(vscaledGHIJKLMNOPQRSTUV, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); vaccGHIJKLMNOPQRSTUV = _mm512_cvtps_epi32(vscaledGHIJKLMNOPQRSTUV); __m512i vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV = _mm512_adds_epi16(_mm512_packs_epi32(vacc0123456789ABCDEF, vaccGHIJKLMNOPQRSTUV), voutput_zero_point); __m256i voutGHIJOPQRKLMNSTUV = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vaccGHIJKLMNOPQRSTUV), _mm512_extracti32x8_epi32(vaccGHIJKLMNOPQRSTUV, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m256i vout0123GHIJ4567KLMN = _mm512_castsi512_si256(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV); const __m256i vout89ABOPQRCDEFSTUV = _mm512_extracti32x8_epi32(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV, 1); const __m256i vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV = _mm256_packs_epi16(vout0123GHIJ4567KLMN, vout89ABOPQRCDEFSTUV); __m256i vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_permutevar8x32_epi32(vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV, vpermute_mask); const __m128i voutGHIJOPQR = _mm256_castsi256_si128(voutGHIJOPQRKLMNSTUV); const __m128i voutKLMNSTUV = _mm256_extracti128_si256(voutGHIJOPQRKLMNSTUV, 1); __m128i voutGHIJKLMNOPQRSTUV = _mm_shuffle_epi32(_mm_packs_epi16(voutGHIJOPQR, voutKLMNSTUV), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_max_epi8(vout0123456789ABCDEFGHIJKLMNOPQRSTUV, voutput_min); voutGHIJKLMNOPQRSTUV = _mm_max_epi8(voutGHIJKLMNOPQRSTUV, _mm256_castsi256_si128(voutput_min)); _mm256_storeu_si256((__m256i*) output, vout0123456789ABCDEFGHIJKLMNOPQRSTUV); _mm_storeu_si128((__m128i*) (output + 16), voutGHIJKLMNOPQRSTUV); output += 32; } if XNN_UNLIKELY(c != 0) { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << (c & 15)) - UINT32_C(1))); const int8_t* k = (const int8_t*) ((uintptr_t) w + 32 * sizeof(int32_t)); do { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) k)); i0 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 32))); i1 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 64))); i2 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 96))); i3 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 128))); i4 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 160))); i5 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 192))); i6 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 224))); i7 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 256))); i8 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); k += 16; __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); const __m512 vscale0123456789ABCDEF = _mm512_loadu_ps((const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 288 * sizeof(int8_t))); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale0123456789ABCDEF); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); w = (const void*) ((uintptr_t) w + 16 * sizeof(int32_t)); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m128i vout012389AB = _mm256_castsi256_si128(vout012389AB4567CDEF); const __m128i vout4567CDEF = _mm256_extracti128_si256(vout012389AB4567CDEF, 1); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packs_epi16(vout012389AB, vout4567CDEF), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, _mm256_castsi256_si128(voutput_min)); if XNN_LIKELY(c >= 16) { _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; c -= 16; } else { _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); output = (int8_t*) ((uintptr_t) output + c); c = 0; } } while (c != 0); } output = (int8_t*) ((uintptr_t) output + output_increment); } while (--output_width != 0); } void xnn_qc8_gemm_minmax_fp32_ukernel_1x16c8__avx512skx( size_t mr, size_t nc, size_t kc, const int8_t* restrict a, size_t a_stride, const void* restrict w, int8_t* restrict c, size_t cm_stride, size_t cn_stride, const union xnn_qc8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 1); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(int8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); const int8_t* a0 = a; int8_t* c0 = c; const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); w = (const void*) ((const int32_t*) w + 16); size_t k = 0; while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i vb0123 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) w)); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); const __m512i vb4567 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 32))); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); const __m512i vb89AB = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 64))); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); const __m512i vbCDEF = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 96))); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); w = (const void*) ((const int8_t*) w + 128); k += 8 * sizeof(int8_t); } const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); const __m512 vscale012345678ABCDEF = _mm512_load_ps(w); w = (const void*) ((const float*) w + 16); const __m512 vscale084C195D2A6E3B7F = _mm512_permutexvar_ps(_mm512_set_epi32(15, 7, 11, 3, 14, 6, 10, 2, 13, 5, 9, 1, 12, 4, 8, 0), vscale012345678ABCDEF); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); const __m256i vacc0x084C2A6E195D3B7F = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0x084C195D2A6E3B7F), _mm512_extracti32x8_epi32(vacc0x084C195D2A6E3B7F, 1)), voutput_zero_point); const __m128i vout0x084C2A6E195D3B7F = _mm_packs_epi16(_mm256_castsi256_si128(vacc0x084C2A6E195D3B7F), _mm256_extracti128_si256(vacc0x084C2A6E195D3B7F, 1)); __m128i vout0x0123456789ABCDEF = _mm_shuffle_epi8(vout0x084C2A6E195D3B7F, _mm_set_epi8(15, 7, 11, 3, 13, 5, 9, 1, 14, 6, 10, 2, 12, 4, 8, 0)); vout0x0123456789ABCDEF = _mm_max_epi8(vout0x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c0, vout0x0123456789ABCDEF); a0 = (const int8_t*) ((uintptr_t) a0 - k); c0 = (int8_t*) ((uintptr_t) c0 + cn_stride); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT32_C(1) << nc) - UINT32_C(1))); _mm_mask_storeu_epi8(c0, vmask, vout0x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qc8_gemm_minmax_fp32_ukernel_4x16c8__avx512skx( size_t mr, size_t nc, size_t kc, const int8_t* restrict a, size_t a_stride, const void* restrict w, int8_t* restrict c, size_t cm_stride, size_t cn_stride, const union xnn_qc8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 4); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(int8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); const int8_t* a0 = a; int8_t* c0 = c; const int8_t* a1 = (const int8_t*) ((uintptr_t) a0 + a_stride); int8_t* c1 = (int8_t*) ((uintptr_t) c0 + cm_stride); if XNN_UNPREDICTABLE(mr < 2) { a1 = a0; c1 = c0; } const int8_t* a2 = (const int8_t*) ((uintptr_t) a1 + a_stride); int8_t* c2 = (int8_t*) ((uintptr_t) c1 + cm_stride); if XNN_UNPREDICTABLE(mr <= 2) { a2 = a1; c2 = c1; } const int8_t* a3 = (const int8_t*) ((uintptr_t) a2 + a_stride); int8_t* c3 = (int8_t*) ((uintptr_t) c2 + cm_stride); if XNN_UNPREDICTABLE(mr != 4) { a3 = a2; c3 = c2; } const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m512i voutput_min = _mm512_load_si512(params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); __m512i vacc1x0123 = vacc0x0123; __m512i vacc1x4567 = vacc0x4567; __m512i vacc1x89AB = vacc0x89AB; __m512i vacc1xCDEF = vacc0xCDEF; __m512i vacc2x0123 = vacc0x0123; __m512i vacc2x4567 = vacc0x4567; __m512i vacc2x89AB = vacc0x89AB; __m512i vacc2xCDEF = vacc0xCDEF; __m512i vacc3x0123 = vacc0x0123; __m512i vacc3x4567 = vacc0x4567; __m512i vacc3x89AB = vacc0x89AB; __m512i vacc3xCDEF = vacc0xCDEF; w = (const void*) ((const int32_t*) w + 16); size_t k = 0; while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i va1 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a1))); a1 += 8; const __m512i va2 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a2))); a2 += 8; const __m512i va3 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a3))); a3 += 8; const __m512i vb0123 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) w)); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); vacc1x0123 = _mm512_add_epi32(vacc1x0123, _mm512_madd_epi16(va1, vb0123)); vacc2x0123 = _mm512_add_epi32(vacc2x0123, _mm512_madd_epi16(va2, vb0123)); vacc3x0123 = _mm512_add_epi32(vacc3x0123, _mm512_madd_epi16(va3, vb0123)); const __m512i vb4567 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 32))); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); vacc1x4567 = _mm512_add_epi32(vacc1x4567, _mm512_madd_epi16(va1, vb4567)); vacc2x4567 = _mm512_add_epi32(vacc2x4567, _mm512_madd_epi16(va2, vb4567)); vacc3x4567 = _mm512_add_epi32(vacc3x4567, _mm512_madd_epi16(va3, vb4567)); const __m512i vb89AB = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 64))); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); vacc1x89AB = _mm512_add_epi32(vacc1x89AB, _mm512_madd_epi16(va1, vb89AB)); vacc2x89AB = _mm512_add_epi32(vacc2x89AB, _mm512_madd_epi16(va2, vb89AB)); vacc3x89AB = _mm512_add_epi32(vacc3x89AB, _mm512_madd_epi16(va3, vb89AB)); const __m512i vbCDEF = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 96))); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); vacc1xCDEF = _mm512_add_epi32(vacc1xCDEF, _mm512_madd_epi16(va1, vbCDEF)); vacc2xCDEF = _mm512_add_epi32(vacc2xCDEF, _mm512_madd_epi16(va2, vbCDEF)); vacc3xCDEF = _mm512_add_epi32(vacc3xCDEF, _mm512_madd_epi16(va3, vbCDEF)); w = (const void*) ((const int8_t*) w + 128); k += 8 * sizeof(int8_t); } const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); const __m512i vacc1x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x0123, vacc1x4567), _mm512_unpackhi_epi32(vacc1x0123, vacc1x4567)); const __m512i vacc1x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x89AB, vacc1xCDEF), _mm512_unpackhi_epi32(vacc1x89AB, vacc1xCDEF)); const __m512i vacc2x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x0123, vacc2x4567), _mm512_unpackhi_epi32(vacc2x0123, vacc2x4567)); const __m512i vacc2x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x89AB, vacc2xCDEF), _mm512_unpackhi_epi32(vacc2x89AB, vacc2xCDEF)); const __m512i vacc3x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x0123, vacc3x4567), _mm512_unpackhi_epi32(vacc3x0123, vacc3x4567)); const __m512i vacc3x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x89AB, vacc3xCDEF), _mm512_unpackhi_epi32(vacc3x89AB, vacc3xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512i vacc1x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x04152637, vacc1x8C9DAEBF), _mm512_unpackhi_epi32(vacc1x04152637, vacc1x8C9DAEBF)); __m512i vacc2x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x04152637, vacc2x8C9DAEBF), _mm512_unpackhi_epi32(vacc2x04152637, vacc2x8C9DAEBF)); __m512i vacc3x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x04152637, vacc3x8C9DAEBF), _mm512_unpackhi_epi32(vacc3x04152637, vacc3x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); __m512 vscaled1x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc1x084C195D2A6E3B7F); __m512 vscaled2x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc2x084C195D2A6E3B7F); __m512 vscaled3x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc3x084C195D2A6E3B7F); const __m512 vscale012345678ABCDEF = _mm512_load_ps(w); w = (const void*) ((const float*) w + 16); const __m512 vscale084C195D2A6E3B7F = _mm512_permutexvar_ps(_mm512_set_epi32(15, 7, 11, 3, 14, 6, 10, 2, 13, 5, 9, 1, 12, 4, 8, 0), vscale012345678ABCDEF); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled1x084C195D2A6E3B7F = _mm512_mul_ps(vscaled1x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled2x084C195D2A6E3B7F = _mm512_mul_ps(vscaled2x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled3x084C195D2A6E3B7F = _mm512_mul_ps(vscaled3x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled1x084C195D2A6E3B7F = _mm512_min_ps(vscaled1x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled2x084C195D2A6E3B7F = _mm512_min_ps(vscaled2x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled3x084C195D2A6E3B7F = _mm512_min_ps(vscaled3x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); vacc1x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled1x084C195D2A6E3B7F); vacc2x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled2x084C195D2A6E3B7F); vacc3x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled3x084C195D2A6E3B7F); const __m512i vacc01x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc0x084C195D2A6E3B7F, vacc1x084C195D2A6E3B7F), voutput_zero_point); const __m512i vacc23x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc2x084C195D2A6E3B7F, vacc3x084C195D2A6E3B7F), voutput_zero_point); __m512i vout0123x084Cx195Dx2A6Ex3B7F = _mm512_packs_epi16(vacc01x084Cx195Dx2A6Ex3B7F, vacc23x084Cx195Dx2A6Ex3B7F); vout0123x084Cx195Dx2A6Ex3B7F = _mm512_permutexvar_epi32(_mm512_set_epi32(15, 11, 7, 3, 14, 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0), vout0123x084Cx195Dx2A6Ex3B7F); __m512i vout0123x0123456789ABCDEF = _mm512_shuffle_epi8(vout0123x084Cx195Dx2A6Ex3B7F, _mm512_set_epi8(15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0)); vout0123x0123456789ABCDEF = _mm512_max_epi8(vout0123x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c0, _mm512_castsi512_si128(vout0123x0123456789ABCDEF)); _mm_storeu_si128((__m128i*) c1, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 1)); _mm_storeu_si128((__m128i*) c2, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 2)); _mm_storeu_si128((__m128i*) c3, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 3)); a0 = (const int8_t*) ((uintptr_t) a0 - k); a1 = (const int8_t*) ((uintptr_t) a1 - k); a2 = (const int8_t*) ((uintptr_t) a2 - k); a3 = (const int8_t*) ((uintptr_t) a3 - k); c0 = (int8_t*) ((uintptr_t) c0 + cn_stride); c1 = (int8_t*) ((uintptr_t) c1 + cn_stride); c2 = (int8_t*) ((uintptr_t) c2 + cn_stride); c3 = (int8_t*) ((uintptr_t) c3 + cn_stride); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT32_C(1) << nc) - UINT32_C(1))); _mm512_mask_storeu_epi8(c0, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftli_mask64(vmask, 16); _mm512_mask_storeu_epi8(c1 - 16, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftli_mask64(vmask, 16); _mm512_mask_storeu_epi8(c2 - 32, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftli_mask64(vmask, 16); _mm512_mask_storeu_epi8(c3 - 48, vmask, vout0123x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qc8_igemm_minmax_fp32_ukernel_1x16c8__avx512skx( size_t mr, size_t nc, size_t kc, size_t ks, const int8_t** restrict a, const void* restrict w, int8_t* restrict c, size_t cm_stride, size_t cn_stride, size_t a_offset, const int8_t* zero, const union xnn_qc8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 1); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(int8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); int8_t* c0 = c; const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); w = (const void*) ((const int32_t*) w + 16); size_t p = ks; do { const int8_t* restrict a0 = a[0]; if XNN_UNPREDICTABLE(a0 != zero) { a0 = (const int8_t*) ((uintptr_t) a0 + a_offset); } a += 1; size_t k = 0; while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i vb0123 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) w)); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); const __m512i vb4567 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 32))); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); const __m512i vb89AB = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 64))); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); const __m512i vbCDEF = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 96))); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); w = (const void*) ((const int8_t*) w + 128); k += 8 * sizeof(int8_t); } p -= 1 * sizeof(void*); } while (p != 0); const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); const __m512 vscale012345678ABCDEF = _mm512_load_ps(w); w = (const void*) ((const float*) w + 16); const __m512 vscale084C195D2A6E3B7F = _mm512_permutexvar_ps(_mm512_set_epi32(15, 7, 11, 3, 14, 6, 10, 2, 13, 5, 9, 1, 12, 4, 8, 0), vscale012345678ABCDEF); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); const __m256i vacc0x084C2A6E195D3B7F = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0x084C195D2A6E3B7F), _mm512_extracti32x8_epi32(vacc0x084C195D2A6E3B7F, 1)), voutput_zero_point); const __m128i vout0x084C2A6E195D3B7F = _mm_packs_epi16(_mm256_castsi256_si128(vacc0x084C2A6E195D3B7F), _mm256_extracti128_si256(vacc0x084C2A6E195D3B7F, 1)); __m128i vout0x0123456789ABCDEF = _mm_shuffle_epi8(vout0x084C2A6E195D3B7F, _mm_set_epi8(15, 7, 11, 3, 13, 5, 9, 1, 14, 6, 10, 2, 12, 4, 8, 0)); vout0x0123456789ABCDEF = _mm_max_epi8(vout0x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c0, vout0x0123456789ABCDEF); c0 = (int8_t*) ((uintptr_t) c0 + cn_stride); a = (const int8_t**restrict) ((uintptr_t) a - ks); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT32_C(1) << nc) - UINT32_C(1))); _mm_mask_storeu_epi8(c0, vmask, vout0x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qc8_igemm_minmax_fp32_ukernel_4x16c8__avx512skx( size_t mr, size_t nc, size_t kc, size_t ks, const int8_t** restrict a, const void* restrict w, int8_t* restrict c, size_t cm_stride, size_t cn_stride, size_t a_offset, const int8_t* zero, const union xnn_qc8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 4); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(int8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); int8_t* c0 = c; int8_t* c1 = (int8_t*) ((uintptr_t) c0 + cm_stride); if XNN_UNPREDICTABLE(mr < 2) { c1 = c0; } int8_t* c2 = (int8_t*) ((uintptr_t) c1 + cm_stride); if XNN_UNPREDICTABLE(mr <= 2) { c2 = c1; } int8_t* c3 = (int8_t*) ((uintptr_t) c2 + cm_stride); if XNN_UNPREDICTABLE(mr != 4) { c3 = c2; } const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m512i voutput_min = _mm512_load_si512(params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); __m512i vacc1x0123 = vacc0x0123; __m512i vacc1x4567 = vacc0x4567; __m512i vacc1x89AB = vacc0x89AB; __m512i vacc1xCDEF = vacc0xCDEF; __m512i vacc2x0123 = vacc0x0123; __m512i vacc2x4567 = vacc0x4567; __m512i vacc2x89AB = vacc0x89AB; __m512i vacc2xCDEF = vacc0xCDEF; __m512i vacc3x0123 = vacc0x0123; __m512i vacc3x4567 = vacc0x4567; __m512i vacc3x89AB = vacc0x89AB; __m512i vacc3xCDEF = vacc0xCDEF; w = (const void*) ((const int32_t*) w + 16); size_t p = ks; do { const int8_t* restrict a0 = a[0]; if XNN_UNPREDICTABLE(a0 != zero) { a0 = (const int8_t*) ((uintptr_t) a0 + a_offset); } const int8_t* restrict a1 = a[1]; if XNN_UNPREDICTABLE(a1 != zero) { a1 = (const int8_t*) ((uintptr_t) a1 + a_offset); } const int8_t* restrict a2 = a[2]; if XNN_UNPREDICTABLE(a2 != zero) { a2 = (const int8_t*) ((uintptr_t) a2 + a_offset); } const int8_t* restrict a3 = a[3]; if XNN_UNPREDICTABLE(a3 != zero) { a3 = (const int8_t*) ((uintptr_t) a3 + a_offset); } a += 4; size_t k = 0; while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i va1 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a1))); a1 += 8; const __m512i va2 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a2))); a2 += 8; const __m512i va3 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a3))); a3 += 8; const __m512i vb0123 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) w)); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); vacc1x0123 = _mm512_add_epi32(vacc1x0123, _mm512_madd_epi16(va1, vb0123)); vacc2x0123 = _mm512_add_epi32(vacc2x0123, _mm512_madd_epi16(va2, vb0123)); vacc3x0123 = _mm512_add_epi32(vacc3x0123, _mm512_madd_epi16(va3, vb0123)); const __m512i vb4567 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 32))); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); vacc1x4567 = _mm512_add_epi32(vacc1x4567, _mm512_madd_epi16(va1, vb4567)); vacc2x4567 = _mm512_add_epi32(vacc2x4567, _mm512_madd_epi16(va2, vb4567)); vacc3x4567 = _mm512_add_epi32(vacc3x4567, _mm512_madd_epi16(va3, vb4567)); const __m512i vb89AB = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 64))); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); vacc1x89AB = _mm512_add_epi32(vacc1x89AB, _mm512_madd_epi16(va1, vb89AB)); vacc2x89AB = _mm512_add_epi32(vacc2x89AB, _mm512_madd_epi16(va2, vb89AB)); vacc3x89AB = _mm512_add_epi32(vacc3x89AB, _mm512_madd_epi16(va3, vb89AB)); const __m512i vbCDEF = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 96))); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); vacc1xCDEF = _mm512_add_epi32(vacc1xCDEF, _mm512_madd_epi16(va1, vbCDEF)); vacc2xCDEF = _mm512_add_epi32(vacc2xCDEF, _mm512_madd_epi16(va2, vbCDEF)); vacc3xCDEF = _mm512_add_epi32(vacc3xCDEF, _mm512_madd_epi16(va3, vbCDEF)); w = (const void*) ((const int8_t*) w + 128); k += 8 * sizeof(int8_t); } p -= 4 * sizeof(void*); } while (p != 0); const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); const __m512i vacc1x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x0123, vacc1x4567), _mm512_unpackhi_epi32(vacc1x0123, vacc1x4567)); const __m512i vacc1x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x89AB, vacc1xCDEF), _mm512_unpackhi_epi32(vacc1x89AB, vacc1xCDEF)); const __m512i vacc2x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x0123, vacc2x4567), _mm512_unpackhi_epi32(vacc2x0123, vacc2x4567)); const __m512i vacc2x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x89AB, vacc2xCDEF), _mm512_unpackhi_epi32(vacc2x89AB, vacc2xCDEF)); const __m512i vacc3x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x0123, vacc3x4567), _mm512_unpackhi_epi32(vacc3x0123, vacc3x4567)); const __m512i vacc3x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x89AB, vacc3xCDEF), _mm512_unpackhi_epi32(vacc3x89AB, vacc3xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512i vacc1x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x04152637, vacc1x8C9DAEBF), _mm512_unpackhi_epi32(vacc1x04152637, vacc1x8C9DAEBF)); __m512i vacc2x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x04152637, vacc2x8C9DAEBF), _mm512_unpackhi_epi32(vacc2x04152637, vacc2x8C9DAEBF)); __m512i vacc3x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x04152637, vacc3x8C9DAEBF), _mm512_unpackhi_epi32(vacc3x04152637, vacc3x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); __m512 vscaled1x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc1x084C195D2A6E3B7F); __m512 vscaled2x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc2x084C195D2A6E3B7F); __m512 vscaled3x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc3x084C195D2A6E3B7F); const __m512 vscale012345678ABCDEF = _mm512_load_ps(w); w = (const void*) ((const float*) w + 16); const __m512 vscale084C195D2A6E3B7F = _mm512_permutexvar_ps(_mm512_set_epi32(15, 7, 11, 3, 14, 6, 10, 2, 13, 5, 9, 1, 12, 4, 8, 0), vscale012345678ABCDEF); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled1x084C195D2A6E3B7F = _mm512_mul_ps(vscaled1x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled2x084C195D2A6E3B7F = _mm512_mul_ps(vscaled2x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled3x084C195D2A6E3B7F = _mm512_mul_ps(vscaled3x084C195D2A6E3B7F, vscale084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled1x084C195D2A6E3B7F = _mm512_min_ps(vscaled1x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled2x084C195D2A6E3B7F = _mm512_min_ps(vscaled2x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled3x084C195D2A6E3B7F = _mm512_min_ps(vscaled3x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); vacc1x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled1x084C195D2A6E3B7F); vacc2x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled2x084C195D2A6E3B7F); vacc3x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled3x084C195D2A6E3B7F); const __m512i vacc01x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc0x084C195D2A6E3B7F, vacc1x084C195D2A6E3B7F), voutput_zero_point); const __m512i vacc23x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc2x084C195D2A6E3B7F, vacc3x084C195D2A6E3B7F), voutput_zero_point); __m512i vout0123x084Cx195Dx2A6Ex3B7F = _mm512_packs_epi16(vacc01x084Cx195Dx2A6Ex3B7F, vacc23x084Cx195Dx2A6Ex3B7F); vout0123x084Cx195Dx2A6Ex3B7F = _mm512_permutexvar_epi32(_mm512_set_epi32(15, 11, 7, 3, 14, 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0), vout0123x084Cx195Dx2A6Ex3B7F); __m512i vout0123x0123456789ABCDEF = _mm512_shuffle_epi8(vout0123x084Cx195Dx2A6Ex3B7F, _mm512_set_epi8(15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0)); vout0123x0123456789ABCDEF = _mm512_max_epi8(vout0123x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c3, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 3)); _mm_storeu_si128((__m128i*) c2, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 2)); _mm_storeu_si128((__m128i*) c1, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 1)); _mm_storeu_si128((__m128i*) c0, _mm512_castsi512_si128(vout0123x0123456789ABCDEF)); c3 = (int8_t*) ((uintptr_t) c3 + cn_stride); c2 = (int8_t*) ((uintptr_t) c2 + cn_stride); c1 = (int8_t*) ((uintptr_t) c1 + cn_stride); c0 = (int8_t*) ((uintptr_t) c0 + cn_stride); a = (const int8_t**restrict) ((uintptr_t) a - ks); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT64_C(1) << (nc + 48)) - (UINT64_C(1) << 48))); _mm512_mask_storeu_epi8(c3 - 48, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftri_mask64(vmask, 16); _mm512_mask_storeu_epi8(c2 - 32, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftri_mask64(vmask, 16); _mm512_mask_storeu_epi8(c1 - 16, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftri_mask64(vmask, 16); _mm512_mask_storeu_epi8(c0, vmask, vout0123x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qs8_dwconv_minmax_fp32_ukernel_up32x25__avx512skx_mul32( 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_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_MSAN { assert(channels != 0); assert(output_width != 0); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m256i voutput_min = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_min); const __m256i vpermute_mask = _mm256_set_epi32(7, 3, 5, 1, 6, 2, 4, 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); } const int8_t* i9 = input[9]; assert(i9 != NULL); if XNN_UNPREDICTABLE(i9 != zero) { i9 = (const int8_t*) ((uintptr_t) i9 + input_offset); } const int8_t* i10 = input[10]; assert(i10 != NULL); if XNN_UNPREDICTABLE(i10 != zero) { i10 = (const int8_t*) ((uintptr_t) i10 + input_offset); } const int8_t* i11 = input[11]; assert(i11 != NULL); if XNN_UNPREDICTABLE(i11 != zero) { i11 = (const int8_t*) ((uintptr_t) i11 + input_offset); } const int8_t* i12 = input[12]; assert(i12 != NULL); if XNN_UNPREDICTABLE(i12 != zero) { i12 = (const int8_t*) ((uintptr_t) i12 + input_offset); } const int8_t* i13 = input[13]; assert(i13 != NULL); if XNN_UNPREDICTABLE(i13 != zero) { i13 = (const int8_t*) ((uintptr_t) i13 + input_offset); } const int8_t* i14 = input[14]; assert(i14 != NULL); if XNN_UNPREDICTABLE(i14 != zero) { i14 = (const int8_t*) ((uintptr_t) i14 + input_offset); } const int8_t* i15 = input[15]; assert(i15 != NULL); if XNN_UNPREDICTABLE(i15 != zero) { i15 = (const int8_t*) ((uintptr_t) i15 + input_offset); } const int8_t* i16 = input[16]; assert(i16 != NULL); if XNN_UNPREDICTABLE(i16 != zero) { i16 = (const int8_t*) ((uintptr_t) i16 + input_offset); } const int8_t* i17 = input[17]; assert(i17 != NULL); if XNN_UNPREDICTABLE(i17 != zero) { i17 = (const int8_t*) ((uintptr_t) i17 + input_offset); } const int8_t* i18 = input[18]; assert(i18 != NULL); if XNN_UNPREDICTABLE(i18 != zero) { i18 = (const int8_t*) ((uintptr_t) i18 + input_offset); } const int8_t* i19 = input[19]; assert(i19 != NULL); if XNN_UNPREDICTABLE(i19 != zero) { i19 = (const int8_t*) ((uintptr_t) i19 + input_offset); } const int8_t* i20 = input[20]; assert(i20 != NULL); if XNN_UNPREDICTABLE(i20 != zero) { i20 = (const int8_t*) ((uintptr_t) i20 + input_offset); } const int8_t* i21 = input[21]; assert(i21 != NULL); if XNN_UNPREDICTABLE(i21 != zero) { i21 = (const int8_t*) ((uintptr_t) i21 + input_offset); } const int8_t* i22 = input[22]; assert(i22 != NULL); if XNN_UNPREDICTABLE(i22 != zero) { i22 = (const int8_t*) ((uintptr_t) i22 + input_offset); } const int8_t* i23 = input[23]; assert(i23 != NULL); if XNN_UNPREDICTABLE(i23 != zero) { i23 = (const int8_t*) ((uintptr_t) i23 + input_offset); } const int8_t* i24 = input[24]; assert(i24 != NULL); if XNN_UNPREDICTABLE(i24 != zero) { i24 = (const int8_t*) ((uintptr_t) i24 + input_offset); } input = (const int8_t**) ((uintptr_t) input + input_stride); size_t c = channels; const void* w = weights; for (; c >= 32; c -= 32) { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); __m512i vaccGHIJKLMNOPQRSTUV = _mm512_loadu_si512((const void*) ((uintptr_t) w + 16 * sizeof(int32_t))); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 0 * sizeof(int8_t)))); const __m512i vi0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m512i vk0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 16 * sizeof(int8_t)))); i0 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi0xGHIJKLMNOPQRSTUV, vk0xGHIJKLMNOPQRSTUV)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 32 * sizeof(int8_t)))); const __m512i vi1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m512i vk1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 48 * sizeof(int8_t)))); i1 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi1xGHIJKLMNOPQRSTUV, vk1xGHIJKLMNOPQRSTUV)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 64 * sizeof(int8_t)))); const __m512i vi2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m512i vk2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 80 * sizeof(int8_t)))); i2 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi2xGHIJKLMNOPQRSTUV, vk2xGHIJKLMNOPQRSTUV)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 96 * sizeof(int8_t)))); const __m512i vi3xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i3 + 16))); const __m512i vk3xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 112 * sizeof(int8_t)))); i3 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi3xGHIJKLMNOPQRSTUV, vk3xGHIJKLMNOPQRSTUV)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 128 * sizeof(int8_t)))); const __m512i vi4xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i4 + 16))); const __m512i vk4xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 144 * sizeof(int8_t)))); i4 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi4xGHIJKLMNOPQRSTUV, vk4xGHIJKLMNOPQRSTUV)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 160 * sizeof(int8_t)))); const __m512i vi5xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i5 + 16))); const __m512i vk5xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 176 * sizeof(int8_t)))); i5 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi5xGHIJKLMNOPQRSTUV, vk5xGHIJKLMNOPQRSTUV)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 192 * sizeof(int8_t)))); const __m512i vi6xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i6 + 16))); const __m512i vk6xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 208 * sizeof(int8_t)))); i6 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi6xGHIJKLMNOPQRSTUV, vk6xGHIJKLMNOPQRSTUV)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 224 * sizeof(int8_t)))); const __m512i vi7xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i7 + 16))); const __m512i vk7xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 240 * sizeof(int8_t)))); i7 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi7xGHIJKLMNOPQRSTUV, vk7xGHIJKLMNOPQRSTUV)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 256 * sizeof(int8_t)))); const __m512i vi8xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i8 + 16))); const __m512i vk8xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 272 * sizeof(int8_t)))); i8 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi8xGHIJKLMNOPQRSTUV, vk8xGHIJKLMNOPQRSTUV)); const __m512i vi9x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i9)); const __m512i vk9x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 288 * sizeof(int8_t)))); const __m512i vi9xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i9 + 16))); const __m512i vk9xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 304 * sizeof(int8_t)))); i9 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi9x0123456789ABCDEF, vk9x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi9xGHIJKLMNOPQRSTUV, vk9xGHIJKLMNOPQRSTUV)); const __m512i vi10x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i10)); const __m512i vk10x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 320 * sizeof(int8_t)))); const __m512i vi10xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i10 + 16))); const __m512i vk10xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 336 * sizeof(int8_t)))); i10 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi10x0123456789ABCDEF, vk10x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi10xGHIJKLMNOPQRSTUV, vk10xGHIJKLMNOPQRSTUV)); const __m512i vi11x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i11)); const __m512i vk11x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 352 * sizeof(int8_t)))); const __m512i vi11xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i11 + 16))); const __m512i vk11xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 368 * sizeof(int8_t)))); i11 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi11x0123456789ABCDEF, vk11x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi11xGHIJKLMNOPQRSTUV, vk11xGHIJKLMNOPQRSTUV)); const __m512i vi12x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i12)); const __m512i vk12x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 384 * sizeof(int8_t)))); const __m512i vi12xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i12 + 16))); const __m512i vk12xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 400 * sizeof(int8_t)))); i12 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi12x0123456789ABCDEF, vk12x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi12xGHIJKLMNOPQRSTUV, vk12xGHIJKLMNOPQRSTUV)); const __m512i vi13x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i13)); const __m512i vk13x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 416 * sizeof(int8_t)))); const __m512i vi13xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i13 + 16))); const __m512i vk13xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 432 * sizeof(int8_t)))); i13 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi13x0123456789ABCDEF, vk13x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi13xGHIJKLMNOPQRSTUV, vk13xGHIJKLMNOPQRSTUV)); const __m512i vi14x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i14)); const __m512i vk14x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 448 * sizeof(int8_t)))); const __m512i vi14xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i14 + 16))); const __m512i vk14xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 464 * sizeof(int8_t)))); i14 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi14x0123456789ABCDEF, vk14x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi14xGHIJKLMNOPQRSTUV, vk14xGHIJKLMNOPQRSTUV)); const __m512i vi15x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i15)); const __m512i vk15x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 480 * sizeof(int8_t)))); const __m512i vi15xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i15 + 16))); const __m512i vk15xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 496 * sizeof(int8_t)))); i15 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi15x0123456789ABCDEF, vk15x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi15xGHIJKLMNOPQRSTUV, vk15xGHIJKLMNOPQRSTUV)); const __m512i vi16x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i16)); const __m512i vk16x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 512 * sizeof(int8_t)))); const __m512i vi16xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i16 + 16))); const __m512i vk16xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 528 * sizeof(int8_t)))); i16 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi16x0123456789ABCDEF, vk16x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi16xGHIJKLMNOPQRSTUV, vk16xGHIJKLMNOPQRSTUV)); const __m512i vi17x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i17)); const __m512i vk17x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 544 * sizeof(int8_t)))); const __m512i vi17xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i17 + 16))); const __m512i vk17xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 560 * sizeof(int8_t)))); i17 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi17x0123456789ABCDEF, vk17x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi17xGHIJKLMNOPQRSTUV, vk17xGHIJKLMNOPQRSTUV)); const __m512i vi18x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i18)); const __m512i vk18x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 576 * sizeof(int8_t)))); const __m512i vi18xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i18 + 16))); const __m512i vk18xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 592 * sizeof(int8_t)))); i18 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi18x0123456789ABCDEF, vk18x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi18xGHIJKLMNOPQRSTUV, vk18xGHIJKLMNOPQRSTUV)); const __m512i vi19x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i19)); const __m512i vk19x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 608 * sizeof(int8_t)))); const __m512i vi19xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i19 + 16))); const __m512i vk19xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 624 * sizeof(int8_t)))); i19 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi19x0123456789ABCDEF, vk19x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi19xGHIJKLMNOPQRSTUV, vk19xGHIJKLMNOPQRSTUV)); const __m512i vi20x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i20)); const __m512i vk20x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 640 * sizeof(int8_t)))); const __m512i vi20xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i20 + 16))); const __m512i vk20xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 656 * sizeof(int8_t)))); i20 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi20x0123456789ABCDEF, vk20x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi20xGHIJKLMNOPQRSTUV, vk20xGHIJKLMNOPQRSTUV)); const __m512i vi21x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i21)); const __m512i vk21x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 672 * sizeof(int8_t)))); const __m512i vi21xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i21 + 16))); const __m512i vk21xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 688 * sizeof(int8_t)))); i21 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi21x0123456789ABCDEF, vk21x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi21xGHIJKLMNOPQRSTUV, vk21xGHIJKLMNOPQRSTUV)); const __m512i vi22x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i22)); const __m512i vk22x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 704 * sizeof(int8_t)))); const __m512i vi22xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i22 + 16))); const __m512i vk22xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 720 * sizeof(int8_t)))); i22 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi22x0123456789ABCDEF, vk22x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi22xGHIJKLMNOPQRSTUV, vk22xGHIJKLMNOPQRSTUV)); const __m512i vi23x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i23)); const __m512i vk23x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 736 * sizeof(int8_t)))); const __m512i vi23xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i23 + 16))); const __m512i vk23xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 752 * sizeof(int8_t)))); i23 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi23x0123456789ABCDEF, vk23x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi23xGHIJKLMNOPQRSTUV, vk23xGHIJKLMNOPQRSTUV)); const __m512i vi24x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i24)); const __m512i vk24x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 768 * sizeof(int8_t)))); const __m512i vi24xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i24 + 16))); const __m512i vk24xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 784 * sizeof(int8_t)))); i24 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi24x0123456789ABCDEF, vk24x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi24xGHIJKLMNOPQRSTUV, vk24xGHIJKLMNOPQRSTUV)); w = (const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 800 * sizeof(int8_t)); __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); __m512 vscaledGHIJKLMNOPQRSTUV = _mm512_cvtepi32_ps(vaccGHIJKLMNOPQRSTUV); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale); vscaledGHIJKLMNOPQRSTUV = _mm512_mul_ps(vscaledGHIJKLMNOPQRSTUV, vscale); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vscaledGHIJKLMNOPQRSTUV = _mm512_min_ps(vscaledGHIJKLMNOPQRSTUV, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); vaccGHIJKLMNOPQRSTUV = _mm512_cvtps_epi32(vscaledGHIJKLMNOPQRSTUV); __m512i vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV = _mm512_adds_epi16(_mm512_packs_epi32(vacc0123456789ABCDEF, vaccGHIJKLMNOPQRSTUV), voutput_zero_point); __m256i voutGHIJOPQRKLMNSTUV = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vaccGHIJKLMNOPQRSTUV), _mm512_extracti32x8_epi32(vaccGHIJKLMNOPQRSTUV, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m256i vout0123GHIJ4567KLMN = _mm512_castsi512_si256(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV); const __m256i vout89ABOPQRCDEFSTUV = _mm512_extracti32x8_epi32(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV, 1); const __m256i vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV = _mm256_packs_epi16(vout0123GHIJ4567KLMN, vout89ABOPQRCDEFSTUV); __m256i vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_permutevar8x32_epi32(vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV, vpermute_mask); const __m128i voutGHIJOPQR = _mm256_castsi256_si128(voutGHIJOPQRKLMNSTUV); const __m128i voutKLMNSTUV = _mm256_extracti128_si256(voutGHIJOPQRKLMNSTUV, 1); __m128i voutGHIJKLMNOPQRSTUV = _mm_shuffle_epi32(_mm_packs_epi16(voutGHIJOPQR, voutKLMNSTUV), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_max_epi8(vout0123456789ABCDEFGHIJKLMNOPQRSTUV, voutput_min); voutGHIJKLMNOPQRSTUV = _mm_max_epi8(voutGHIJKLMNOPQRSTUV, _mm256_castsi256_si128(voutput_min)); _mm256_storeu_si256((__m256i*) output, vout0123456789ABCDEFGHIJKLMNOPQRSTUV); _mm_storeu_si128((__m128i*) (output + 16), voutGHIJKLMNOPQRSTUV); output += 32; } if XNN_UNLIKELY(c != 0) { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << (c & 15)) - UINT32_C(1))); const int8_t* k = (const int8_t*) ((uintptr_t) w + 32 * sizeof(int32_t)); do { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) k)); i0 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 32))); i1 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 64))); i2 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 96))); i3 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 128))); i4 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 160))); i5 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 192))); i6 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 224))); i7 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 256))); i8 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); const __m512i vi9x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i9)); const __m512i vk9x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 288))); i9 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi9x0123456789ABCDEF, vk9x0123456789ABCDEF)); const __m512i vi10x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i10)); const __m512i vk10x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 320))); i10 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi10x0123456789ABCDEF, vk10x0123456789ABCDEF)); const __m512i vi11x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i11)); const __m512i vk11x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 352))); i11 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi11x0123456789ABCDEF, vk11x0123456789ABCDEF)); const __m512i vi12x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i12)); const __m512i vk12x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 384))); i12 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi12x0123456789ABCDEF, vk12x0123456789ABCDEF)); const __m512i vi13x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i13)); const __m512i vk13x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 416))); i13 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi13x0123456789ABCDEF, vk13x0123456789ABCDEF)); const __m512i vi14x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i14)); const __m512i vk14x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 448))); i14 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi14x0123456789ABCDEF, vk14x0123456789ABCDEF)); const __m512i vi15x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i15)); const __m512i vk15x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 480))); i15 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi15x0123456789ABCDEF, vk15x0123456789ABCDEF)); const __m512i vi16x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i16)); const __m512i vk16x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 512))); i16 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi16x0123456789ABCDEF, vk16x0123456789ABCDEF)); const __m512i vi17x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i17)); const __m512i vk17x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 544))); i17 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi17x0123456789ABCDEF, vk17x0123456789ABCDEF)); const __m512i vi18x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i18)); const __m512i vk18x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 576))); i18 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi18x0123456789ABCDEF, vk18x0123456789ABCDEF)); const __m512i vi19x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i19)); const __m512i vk19x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 608))); i19 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi19x0123456789ABCDEF, vk19x0123456789ABCDEF)); const __m512i vi20x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i20)); const __m512i vk20x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 640))); i20 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi20x0123456789ABCDEF, vk20x0123456789ABCDEF)); const __m512i vi21x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i21)); const __m512i vk21x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 672))); i21 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi21x0123456789ABCDEF, vk21x0123456789ABCDEF)); const __m512i vi22x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i22)); const __m512i vk22x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 704))); i22 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi22x0123456789ABCDEF, vk22x0123456789ABCDEF)); const __m512i vi23x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i23)); const __m512i vk23x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 736))); i23 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi23x0123456789ABCDEF, vk23x0123456789ABCDEF)); const __m512i vi24x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i24)); const __m512i vk24x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 768))); i24 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi24x0123456789ABCDEF, vk24x0123456789ABCDEF)); k += 16; __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); w = (const void*) ((uintptr_t) w + 16 * sizeof(int32_t)); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m128i vout012389AB = _mm256_castsi256_si128(vout012389AB4567CDEF); const __m128i vout4567CDEF = _mm256_extracti128_si256(vout012389AB4567CDEF, 1); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packs_epi16(vout012389AB, vout4567CDEF), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, _mm256_castsi256_si128(voutput_min)); if XNN_LIKELY(c >= 16) { _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; c -= 16; } else { _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); output = (int8_t*) ((uintptr_t) output + c); c = 0; } } while (c != 0); } output = (int8_t*) ((uintptr_t) output + output_increment); } while (--output_width != 0); } void xnn_qs8_dwconv_minmax_fp32_ukernel_up32x9__avx512skx_mul32( 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_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_MSAN { assert(channels != 0); assert(output_width != 0); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m256i voutput_min = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_min); const __m256i vpermute_mask = _mm256_set_epi32(7, 3, 5, 1, 6, 2, 4, 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) { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); __m512i vaccGHIJKLMNOPQRSTUV = _mm512_loadu_si512((const void*) ((uintptr_t) w + 16 * sizeof(int32_t))); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 0 * sizeof(int8_t)))); const __m512i vi0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m512i vk0xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 16 * sizeof(int8_t)))); i0 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi0xGHIJKLMNOPQRSTUV, vk0xGHIJKLMNOPQRSTUV)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 32 * sizeof(int8_t)))); const __m512i vi1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m512i vk1xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 48 * sizeof(int8_t)))); i1 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi1xGHIJKLMNOPQRSTUV, vk1xGHIJKLMNOPQRSTUV)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 64 * sizeof(int8_t)))); const __m512i vi2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m512i vk2xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 80 * sizeof(int8_t)))); i2 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi2xGHIJKLMNOPQRSTUV, vk2xGHIJKLMNOPQRSTUV)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 96 * sizeof(int8_t)))); const __m512i vi3xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i3 + 16))); const __m512i vk3xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 112 * sizeof(int8_t)))); i3 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi3xGHIJKLMNOPQRSTUV, vk3xGHIJKLMNOPQRSTUV)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 128 * sizeof(int8_t)))); const __m512i vi4xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i4 + 16))); const __m512i vk4xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 144 * sizeof(int8_t)))); i4 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi4xGHIJKLMNOPQRSTUV, vk4xGHIJKLMNOPQRSTUV)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 160 * sizeof(int8_t)))); const __m512i vi5xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i5 + 16))); const __m512i vk5xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 176 * sizeof(int8_t)))); i5 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi5xGHIJKLMNOPQRSTUV, vk5xGHIJKLMNOPQRSTUV)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 192 * sizeof(int8_t)))); const __m512i vi6xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i6 + 16))); const __m512i vk6xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 208 * sizeof(int8_t)))); i6 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi6xGHIJKLMNOPQRSTUV, vk6xGHIJKLMNOPQRSTUV)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 224 * sizeof(int8_t)))); const __m512i vi7xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i7 + 16))); const __m512i vk7xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 240 * sizeof(int8_t)))); i7 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi7xGHIJKLMNOPQRSTUV, vk7xGHIJKLMNOPQRSTUV)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 256 * sizeof(int8_t)))); const __m512i vi8xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (i8 + 16))); const __m512i vk8xGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 272 * sizeof(int8_t)))); i8 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi8xGHIJKLMNOPQRSTUV, vk8xGHIJKLMNOPQRSTUV)); w = (const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 288 * sizeof(int8_t)); __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); __m512 vscaledGHIJKLMNOPQRSTUV = _mm512_cvtepi32_ps(vaccGHIJKLMNOPQRSTUV); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale); vscaledGHIJKLMNOPQRSTUV = _mm512_mul_ps(vscaledGHIJKLMNOPQRSTUV, vscale); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vscaledGHIJKLMNOPQRSTUV = _mm512_min_ps(vscaledGHIJKLMNOPQRSTUV, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); vaccGHIJKLMNOPQRSTUV = _mm512_cvtps_epi32(vscaledGHIJKLMNOPQRSTUV); __m512i vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV = _mm512_adds_epi16(_mm512_packs_epi32(vacc0123456789ABCDEF, vaccGHIJKLMNOPQRSTUV), voutput_zero_point); __m256i voutGHIJOPQRKLMNSTUV = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vaccGHIJKLMNOPQRSTUV), _mm512_extracti32x8_epi32(vaccGHIJKLMNOPQRSTUV, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m256i vout0123GHIJ4567KLMN = _mm512_castsi512_si256(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV); const __m256i vout89ABOPQRCDEFSTUV = _mm512_extracti32x8_epi32(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV, 1); const __m256i vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV = _mm256_packs_epi16(vout0123GHIJ4567KLMN, vout89ABOPQRCDEFSTUV); __m256i vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_permutevar8x32_epi32(vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV, vpermute_mask); const __m128i voutGHIJOPQR = _mm256_castsi256_si128(voutGHIJOPQRKLMNSTUV); const __m128i voutKLMNSTUV = _mm256_extracti128_si256(voutGHIJOPQRKLMNSTUV, 1); __m128i voutGHIJKLMNOPQRSTUV = _mm_shuffle_epi32(_mm_packs_epi16(voutGHIJOPQR, voutKLMNSTUV), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_max_epi8(vout0123456789ABCDEFGHIJKLMNOPQRSTUV, voutput_min); voutGHIJKLMNOPQRSTUV = _mm_max_epi8(voutGHIJKLMNOPQRSTUV, _mm256_castsi256_si128(voutput_min)); _mm256_storeu_si256((__m256i*) output, vout0123456789ABCDEFGHIJKLMNOPQRSTUV); _mm_storeu_si128((__m128i*) (output + 16), voutGHIJKLMNOPQRSTUV); output += 32; } if XNN_UNLIKELY(c != 0) { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << (c & 15)) - UINT32_C(1))); const int8_t* k = (const int8_t*) ((uintptr_t) w + 32 * sizeof(int32_t)); do { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) k)); i0 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 32))); i1 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 64))); i2 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 96))); i3 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 128))); i4 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 160))); i5 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 192))); i6 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 224))); i7 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (k + 256))); i8 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); k += 16; __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); w = (const void*) ((uintptr_t) w + 16 * sizeof(int32_t)); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m128i vout012389AB = _mm256_castsi256_si128(vout012389AB4567CDEF); const __m128i vout4567CDEF = _mm256_extracti128_si256(vout012389AB4567CDEF, 1); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packs_epi16(vout012389AB, vout4567CDEF), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, _mm256_castsi256_si128(voutput_min)); if XNN_LIKELY(c >= 16) { _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; c -= 16; } else { _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); output = (int8_t*) ((uintptr_t) output + c); c = 0; } } while (c != 0); } output = (int8_t*) ((uintptr_t) output + output_increment); } while (--output_width != 0); } void xnn_qs8_f32_vcvt_ukernel__avx512skx_x32( size_t n, const int8_t* x, float* y, const union xnn_qs8_f32_cvt_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(n != 0); assert(n % sizeof(int8_t) == 0); assert(x != NULL); assert(y != NULL); const __m512i vminus_zero_point = _mm512_load_si512(params->avx512.minus_zero_point); const __m512 vscale = _mm512_load_ps(params->avx512.scale); for (; n >= 32 * sizeof(int8_t); n -= 32 * sizeof(int8_t)) { __m512i vx0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) x)); __m512i vxGHIJKLMNOPQRSTUV = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) (x + 16))); x += 32; vx0123456789ABCDEF = _mm512_add_epi32(vx0123456789ABCDEF, vminus_zero_point); vxGHIJKLMNOPQRSTUV = _mm512_add_epi32(vxGHIJKLMNOPQRSTUV, vminus_zero_point); __m512 vy0123456789ABCDEF = _mm512_cvtepi32_ps(vx0123456789ABCDEF); __m512 vyGHIJKLMNOPQRSTUV = _mm512_cvtepi32_ps(vxGHIJKLMNOPQRSTUV); vy0123456789ABCDEF = _mm512_mul_ps(vy0123456789ABCDEF, vscale); vyGHIJKLMNOPQRSTUV = _mm512_mul_ps(vyGHIJKLMNOPQRSTUV, vscale); _mm512_storeu_ps(y, vy0123456789ABCDEF); _mm512_storeu_ps(y + 16, vyGHIJKLMNOPQRSTUV); y += 32; } for (; n >= 16 * sizeof(int8_t); n -= 16 * sizeof(int8_t)) { __m512i vx = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) x)); vx = _mm512_add_epi32(vx, vminus_zero_point); x += 16; __m512 vy = _mm512_cvtepi32_ps(vx); vy = _mm512_mul_ps(vy, vscale); _mm512_storeu_ps(y, vy); y += 16; } if XNN_UNLIKELY(n != 0) { assert(n >= 1 * sizeof(int8_t)); assert(n <= 15 * sizeof(int8_t)); // Prepare mask for valid elements (depends on n). const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << n) - UINT32_C(1))); __m512i vx = _mm512_cvtepi8_epi32(_mm_maskz_loadu_epi8(vmask, x)); vx = _mm512_add_epi32(vx, vminus_zero_point); __m512 vy = _mm512_cvtepi32_ps(vx); vy = _mm512_mul_ps(vy, vscale); _mm512_mask_storeu_ps(y, vmask, vy); } } void xnn_qs8_gemm_minmax_fp32_ukernel_1x16c8__avx512skx( size_t mr, size_t nc, size_t kc, const int8_t* restrict a, size_t a_stride, const void* restrict w, int8_t* restrict c, size_t cm_stride, size_t cn_stride, const union xnn_qs8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 1); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(int8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); const int8_t* a0 = a; int8_t* c0 = c; const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); w = (const void*) ((const int32_t*) w + 16); size_t k = 0; while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i vb0123 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) w)); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); const __m512i vb4567 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 32))); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); const __m512i vb89AB = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 64))); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); const __m512i vbCDEF = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 96))); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); w = (const void*) ((const int8_t*) w + 128); k += 8 * sizeof(int8_t); } const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); const __m256i vacc0x084C2A6E195D3B7F = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0x084C195D2A6E3B7F), _mm512_extracti32x8_epi32(vacc0x084C195D2A6E3B7F, 1)), voutput_zero_point); const __m128i vout0x084C2A6E195D3B7F = _mm_packs_epi16(_mm256_castsi256_si128(vacc0x084C2A6E195D3B7F), _mm256_extracti128_si256(vacc0x084C2A6E195D3B7F, 1)); __m128i vout0x0123456789ABCDEF = _mm_shuffle_epi8(vout0x084C2A6E195D3B7F, _mm_set_epi8(15, 7, 11, 3, 13, 5, 9, 1, 14, 6, 10, 2, 12, 4, 8, 0)); vout0x0123456789ABCDEF = _mm_max_epi8(vout0x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c0, vout0x0123456789ABCDEF); a0 = (const int8_t*) ((uintptr_t) a0 - k); c0 = (int8_t*) ((uintptr_t) c0 + cn_stride); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT32_C(1) << nc) - UINT32_C(1))); _mm_mask_storeu_epi8(c0, vmask, vout0x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qs8_gemm_minmax_fp32_ukernel_4x16c8__avx512skx( size_t mr, size_t nc, size_t kc, const int8_t* restrict a, size_t a_stride, const void* restrict w, int8_t* restrict c, size_t cm_stride, size_t cn_stride, const union xnn_qs8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 4); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(int8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); const int8_t* a0 = a; int8_t* c0 = c; const int8_t* a1 = (const int8_t*) ((uintptr_t) a0 + a_stride); int8_t* c1 = (int8_t*) ((uintptr_t) c0 + cm_stride); if XNN_UNPREDICTABLE(mr < 2) { a1 = a0; c1 = c0; } const int8_t* a2 = (const int8_t*) ((uintptr_t) a1 + a_stride); int8_t* c2 = (int8_t*) ((uintptr_t) c1 + cm_stride); if XNN_UNPREDICTABLE(mr <= 2) { a2 = a1; c2 = c1; } const int8_t* a3 = (const int8_t*) ((uintptr_t) a2 + a_stride); int8_t* c3 = (int8_t*) ((uintptr_t) c2 + cm_stride); if XNN_UNPREDICTABLE(mr != 4) { a3 = a2; c3 = c2; } const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m512i voutput_min = _mm512_load_si512(params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); __m512i vacc1x0123 = vacc0x0123; __m512i vacc1x4567 = vacc0x4567; __m512i vacc1x89AB = vacc0x89AB; __m512i vacc1xCDEF = vacc0xCDEF; __m512i vacc2x0123 = vacc0x0123; __m512i vacc2x4567 = vacc0x4567; __m512i vacc2x89AB = vacc0x89AB; __m512i vacc2xCDEF = vacc0xCDEF; __m512i vacc3x0123 = vacc0x0123; __m512i vacc3x4567 = vacc0x4567; __m512i vacc3x89AB = vacc0x89AB; __m512i vacc3xCDEF = vacc0xCDEF; w = (const void*) ((const int32_t*) w + 16); size_t k = 0; while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i va1 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a1))); a1 += 8; const __m512i va2 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a2))); a2 += 8; const __m512i va3 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a3))); a3 += 8; const __m512i vb0123 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) w)); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); vacc1x0123 = _mm512_add_epi32(vacc1x0123, _mm512_madd_epi16(va1, vb0123)); vacc2x0123 = _mm512_add_epi32(vacc2x0123, _mm512_madd_epi16(va2, vb0123)); vacc3x0123 = _mm512_add_epi32(vacc3x0123, _mm512_madd_epi16(va3, vb0123)); const __m512i vb4567 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 32))); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); vacc1x4567 = _mm512_add_epi32(vacc1x4567, _mm512_madd_epi16(va1, vb4567)); vacc2x4567 = _mm512_add_epi32(vacc2x4567, _mm512_madd_epi16(va2, vb4567)); vacc3x4567 = _mm512_add_epi32(vacc3x4567, _mm512_madd_epi16(va3, vb4567)); const __m512i vb89AB = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 64))); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); vacc1x89AB = _mm512_add_epi32(vacc1x89AB, _mm512_madd_epi16(va1, vb89AB)); vacc2x89AB = _mm512_add_epi32(vacc2x89AB, _mm512_madd_epi16(va2, vb89AB)); vacc3x89AB = _mm512_add_epi32(vacc3x89AB, _mm512_madd_epi16(va3, vb89AB)); const __m512i vbCDEF = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 96))); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); vacc1xCDEF = _mm512_add_epi32(vacc1xCDEF, _mm512_madd_epi16(va1, vbCDEF)); vacc2xCDEF = _mm512_add_epi32(vacc2xCDEF, _mm512_madd_epi16(va2, vbCDEF)); vacc3xCDEF = _mm512_add_epi32(vacc3xCDEF, _mm512_madd_epi16(va3, vbCDEF)); w = (const void*) ((const int8_t*) w + 128); k += 8 * sizeof(int8_t); } const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); const __m512i vacc1x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x0123, vacc1x4567), _mm512_unpackhi_epi32(vacc1x0123, vacc1x4567)); const __m512i vacc1x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x89AB, vacc1xCDEF), _mm512_unpackhi_epi32(vacc1x89AB, vacc1xCDEF)); const __m512i vacc2x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x0123, vacc2x4567), _mm512_unpackhi_epi32(vacc2x0123, vacc2x4567)); const __m512i vacc2x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x89AB, vacc2xCDEF), _mm512_unpackhi_epi32(vacc2x89AB, vacc2xCDEF)); const __m512i vacc3x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x0123, vacc3x4567), _mm512_unpackhi_epi32(vacc3x0123, vacc3x4567)); const __m512i vacc3x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x89AB, vacc3xCDEF), _mm512_unpackhi_epi32(vacc3x89AB, vacc3xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512i vacc1x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x04152637, vacc1x8C9DAEBF), _mm512_unpackhi_epi32(vacc1x04152637, vacc1x8C9DAEBF)); __m512i vacc2x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x04152637, vacc2x8C9DAEBF), _mm512_unpackhi_epi32(vacc2x04152637, vacc2x8C9DAEBF)); __m512i vacc3x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x04152637, vacc3x8C9DAEBF), _mm512_unpackhi_epi32(vacc3x04152637, vacc3x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); __m512 vscaled1x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc1x084C195D2A6E3B7F); __m512 vscaled2x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc2x084C195D2A6E3B7F); __m512 vscaled3x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc3x084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale); vscaled1x084C195D2A6E3B7F = _mm512_mul_ps(vscaled1x084C195D2A6E3B7F, vscale); vscaled2x084C195D2A6E3B7F = _mm512_mul_ps(vscaled2x084C195D2A6E3B7F, vscale); vscaled3x084C195D2A6E3B7F = _mm512_mul_ps(vscaled3x084C195D2A6E3B7F, vscale); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled1x084C195D2A6E3B7F = _mm512_min_ps(vscaled1x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled2x084C195D2A6E3B7F = _mm512_min_ps(vscaled2x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled3x084C195D2A6E3B7F = _mm512_min_ps(vscaled3x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); vacc1x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled1x084C195D2A6E3B7F); vacc2x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled2x084C195D2A6E3B7F); vacc3x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled3x084C195D2A6E3B7F); const __m512i vacc01x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc0x084C195D2A6E3B7F, vacc1x084C195D2A6E3B7F), voutput_zero_point); const __m512i vacc23x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc2x084C195D2A6E3B7F, vacc3x084C195D2A6E3B7F), voutput_zero_point); __m512i vout0123x084Cx195Dx2A6Ex3B7F = _mm512_packs_epi16(vacc01x084Cx195Dx2A6Ex3B7F, vacc23x084Cx195Dx2A6Ex3B7F); vout0123x084Cx195Dx2A6Ex3B7F = _mm512_permutexvar_epi32(_mm512_set_epi32(15, 11, 7, 3, 14, 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0), vout0123x084Cx195Dx2A6Ex3B7F); __m512i vout0123x0123456789ABCDEF = _mm512_shuffle_epi8(vout0123x084Cx195Dx2A6Ex3B7F, _mm512_set_epi8(15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0)); vout0123x0123456789ABCDEF = _mm512_max_epi8(vout0123x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c0, _mm512_castsi512_si128(vout0123x0123456789ABCDEF)); _mm_storeu_si128((__m128i*) c1, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 1)); _mm_storeu_si128((__m128i*) c2, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 2)); _mm_storeu_si128((__m128i*) c3, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 3)); a0 = (const int8_t*) ((uintptr_t) a0 - k); a1 = (const int8_t*) ((uintptr_t) a1 - k); a2 = (const int8_t*) ((uintptr_t) a2 - k); a3 = (const int8_t*) ((uintptr_t) a3 - k); c0 = (int8_t*) ((uintptr_t) c0 + cn_stride); c1 = (int8_t*) ((uintptr_t) c1 + cn_stride); c2 = (int8_t*) ((uintptr_t) c2 + cn_stride); c3 = (int8_t*) ((uintptr_t) c3 + cn_stride); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT32_C(1) << nc) - UINT32_C(1))); _mm512_mask_storeu_epi8(c0, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftli_mask64(vmask, 16); _mm512_mask_storeu_epi8(c1 - 16, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftli_mask64(vmask, 16); _mm512_mask_storeu_epi8(c2 - 32, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftli_mask64(vmask, 16); _mm512_mask_storeu_epi8(c3 - 48, vmask, vout0123x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qs8_igemm_minmax_fp32_ukernel_1x16c8__avx512skx( size_t mr, size_t nc, size_t kc, size_t ks, const int8_t** restrict a, const void* restrict w, int8_t* restrict c, size_t cm_stride, size_t cn_stride, size_t a_offset, const int8_t* zero, const union xnn_qs8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 1); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(int8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); int8_t* c0 = c; const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); w = (const void*) ((const int32_t*) w + 16); size_t p = ks; do { const int8_t* restrict a0 = a[0]; if XNN_UNPREDICTABLE(a0 != zero) { a0 = (const int8_t*) ((uintptr_t) a0 + a_offset); } a += 1; size_t k = 0; while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i vb0123 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) w)); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); const __m512i vb4567 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 32))); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); const __m512i vb89AB = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 64))); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); const __m512i vbCDEF = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 96))); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); w = (const void*) ((const int8_t*) w + 128); k += 8 * sizeof(int8_t); } p -= 1 * sizeof(void*); } while (p != 0); const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); const __m256i vacc0x084C2A6E195D3B7F = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0x084C195D2A6E3B7F), _mm512_extracti32x8_epi32(vacc0x084C195D2A6E3B7F, 1)), voutput_zero_point); const __m128i vout0x084C2A6E195D3B7F = _mm_packs_epi16(_mm256_castsi256_si128(vacc0x084C2A6E195D3B7F), _mm256_extracti128_si256(vacc0x084C2A6E195D3B7F, 1)); __m128i vout0x0123456789ABCDEF = _mm_shuffle_epi8(vout0x084C2A6E195D3B7F, _mm_set_epi8(15, 7, 11, 3, 13, 5, 9, 1, 14, 6, 10, 2, 12, 4, 8, 0)); vout0x0123456789ABCDEF = _mm_max_epi8(vout0x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c0, vout0x0123456789ABCDEF); c0 = (int8_t*) ((uintptr_t) c0 + cn_stride); a = (const int8_t**restrict) ((uintptr_t) a - ks); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT32_C(1) << nc) - UINT32_C(1))); _mm_mask_storeu_epi8(c0, vmask, vout0x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qs8_igemm_minmax_fp32_ukernel_4x16c8__avx512skx( size_t mr, size_t nc, size_t kc, size_t ks, const int8_t** restrict a, const void* restrict w, int8_t* restrict c, size_t cm_stride, size_t cn_stride, size_t a_offset, const int8_t* zero, const union xnn_qs8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 4); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(int8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); int8_t* c0 = c; int8_t* c1 = (int8_t*) ((uintptr_t) c0 + cm_stride); if XNN_UNPREDICTABLE(mr < 2) { c1 = c0; } int8_t* c2 = (int8_t*) ((uintptr_t) c1 + cm_stride); if XNN_UNPREDICTABLE(mr <= 2) { c2 = c1; } int8_t* c3 = (int8_t*) ((uintptr_t) c2 + cm_stride); if XNN_UNPREDICTABLE(mr != 4) { c3 = c2; } const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m512i voutput_min = _mm512_load_si512(params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); __m512i vacc1x0123 = vacc0x0123; __m512i vacc1x4567 = vacc0x4567; __m512i vacc1x89AB = vacc0x89AB; __m512i vacc1xCDEF = vacc0xCDEF; __m512i vacc2x0123 = vacc0x0123; __m512i vacc2x4567 = vacc0x4567; __m512i vacc2x89AB = vacc0x89AB; __m512i vacc2xCDEF = vacc0xCDEF; __m512i vacc3x0123 = vacc0x0123; __m512i vacc3x4567 = vacc0x4567; __m512i vacc3x89AB = vacc0x89AB; __m512i vacc3xCDEF = vacc0xCDEF; w = (const void*) ((const int32_t*) w + 16); size_t p = ks; do { const int8_t* restrict a0 = a[0]; if XNN_UNPREDICTABLE(a0 != zero) { a0 = (const int8_t*) ((uintptr_t) a0 + a_offset); } const int8_t* restrict a1 = a[1]; if XNN_UNPREDICTABLE(a1 != zero) { a1 = (const int8_t*) ((uintptr_t) a1 + a_offset); } const int8_t* restrict a2 = a[2]; if XNN_UNPREDICTABLE(a2 != zero) { a2 = (const int8_t*) ((uintptr_t) a2 + a_offset); } const int8_t* restrict a3 = a[3]; if XNN_UNPREDICTABLE(a3 != zero) { a3 = (const int8_t*) ((uintptr_t) a3 + a_offset); } a += 4; size_t k = 0; while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i va1 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a1))); a1 += 8; const __m512i va2 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a2))); a2 += 8; const __m512i va3 = _mm512_broadcast_i32x4(_mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) a3))); a3 += 8; const __m512i vb0123 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) w)); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); vacc1x0123 = _mm512_add_epi32(vacc1x0123, _mm512_madd_epi16(va1, vb0123)); vacc2x0123 = _mm512_add_epi32(vacc2x0123, _mm512_madd_epi16(va2, vb0123)); vacc3x0123 = _mm512_add_epi32(vacc3x0123, _mm512_madd_epi16(va3, vb0123)); const __m512i vb4567 = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 32))); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); vacc1x4567 = _mm512_add_epi32(vacc1x4567, _mm512_madd_epi16(va1, vb4567)); vacc2x4567 = _mm512_add_epi32(vacc2x4567, _mm512_madd_epi16(va2, vb4567)); vacc3x4567 = _mm512_add_epi32(vacc3x4567, _mm512_madd_epi16(va3, vb4567)); const __m512i vb89AB = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 64))); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); vacc1x89AB = _mm512_add_epi32(vacc1x89AB, _mm512_madd_epi16(va1, vb89AB)); vacc2x89AB = _mm512_add_epi32(vacc2x89AB, _mm512_madd_epi16(va2, vb89AB)); vacc3x89AB = _mm512_add_epi32(vacc3x89AB, _mm512_madd_epi16(va3, vb89AB)); const __m512i vbCDEF = _mm512_cvtepi8_epi16(_mm256_load_si256((const __m256i*) ((const int8_t*) w + 96))); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); vacc1xCDEF = _mm512_add_epi32(vacc1xCDEF, _mm512_madd_epi16(va1, vbCDEF)); vacc2xCDEF = _mm512_add_epi32(vacc2xCDEF, _mm512_madd_epi16(va2, vbCDEF)); vacc3xCDEF = _mm512_add_epi32(vacc3xCDEF, _mm512_madd_epi16(va3, vbCDEF)); w = (const void*) ((const int8_t*) w + 128); k += 8 * sizeof(int8_t); } p -= 4 * sizeof(void*); } while (p != 0); const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); const __m512i vacc1x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x0123, vacc1x4567), _mm512_unpackhi_epi32(vacc1x0123, vacc1x4567)); const __m512i vacc1x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x89AB, vacc1xCDEF), _mm512_unpackhi_epi32(vacc1x89AB, vacc1xCDEF)); const __m512i vacc2x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x0123, vacc2x4567), _mm512_unpackhi_epi32(vacc2x0123, vacc2x4567)); const __m512i vacc2x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x89AB, vacc2xCDEF), _mm512_unpackhi_epi32(vacc2x89AB, vacc2xCDEF)); const __m512i vacc3x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x0123, vacc3x4567), _mm512_unpackhi_epi32(vacc3x0123, vacc3x4567)); const __m512i vacc3x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x89AB, vacc3xCDEF), _mm512_unpackhi_epi32(vacc3x89AB, vacc3xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512i vacc1x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x04152637, vacc1x8C9DAEBF), _mm512_unpackhi_epi32(vacc1x04152637, vacc1x8C9DAEBF)); __m512i vacc2x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x04152637, vacc2x8C9DAEBF), _mm512_unpackhi_epi32(vacc2x04152637, vacc2x8C9DAEBF)); __m512i vacc3x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x04152637, vacc3x8C9DAEBF), _mm512_unpackhi_epi32(vacc3x04152637, vacc3x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); __m512 vscaled1x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc1x084C195D2A6E3B7F); __m512 vscaled2x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc2x084C195D2A6E3B7F); __m512 vscaled3x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc3x084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale); vscaled1x084C195D2A6E3B7F = _mm512_mul_ps(vscaled1x084C195D2A6E3B7F, vscale); vscaled2x084C195D2A6E3B7F = _mm512_mul_ps(vscaled2x084C195D2A6E3B7F, vscale); vscaled3x084C195D2A6E3B7F = _mm512_mul_ps(vscaled3x084C195D2A6E3B7F, vscale); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled1x084C195D2A6E3B7F = _mm512_min_ps(vscaled1x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled2x084C195D2A6E3B7F = _mm512_min_ps(vscaled2x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled3x084C195D2A6E3B7F = _mm512_min_ps(vscaled3x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); vacc1x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled1x084C195D2A6E3B7F); vacc2x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled2x084C195D2A6E3B7F); vacc3x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled3x084C195D2A6E3B7F); const __m512i vacc01x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc0x084C195D2A6E3B7F, vacc1x084C195D2A6E3B7F), voutput_zero_point); const __m512i vacc23x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc2x084C195D2A6E3B7F, vacc3x084C195D2A6E3B7F), voutput_zero_point); __m512i vout0123x084Cx195Dx2A6Ex3B7F = _mm512_packs_epi16(vacc01x084Cx195Dx2A6Ex3B7F, vacc23x084Cx195Dx2A6Ex3B7F); vout0123x084Cx195Dx2A6Ex3B7F = _mm512_permutexvar_epi32(_mm512_set_epi32(15, 11, 7, 3, 14, 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0), vout0123x084Cx195Dx2A6Ex3B7F); __m512i vout0123x0123456789ABCDEF = _mm512_shuffle_epi8(vout0123x084Cx195Dx2A6Ex3B7F, _mm512_set_epi8(15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0)); vout0123x0123456789ABCDEF = _mm512_max_epi8(vout0123x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c3, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 3)); _mm_storeu_si128((__m128i*) c2, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 2)); _mm_storeu_si128((__m128i*) c1, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 1)); _mm_storeu_si128((__m128i*) c0, _mm512_castsi512_si128(vout0123x0123456789ABCDEF)); c3 = (int8_t*) ((uintptr_t) c3 + cn_stride); c2 = (int8_t*) ((uintptr_t) c2 + cn_stride); c1 = (int8_t*) ((uintptr_t) c1 + cn_stride); c0 = (int8_t*) ((uintptr_t) c0 + cn_stride); a = (const int8_t**restrict) ((uintptr_t) a - ks); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT64_C(1) << (nc + 48)) - (UINT64_C(1) << 48))); _mm512_mask_storeu_epi8(c3 - 48, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftri_mask64(vmask, 16); _mm512_mask_storeu_epi8(c2 - 32, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftri_mask64(vmask, 16); _mm512_mask_storeu_epi8(c1 - 16, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftri_mask64(vmask, 16); _mm512_mask_storeu_epi8(c0, vmask, vout0123x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qs8_vadd_minmax_ukernel__avx512skx_mul32_ld128_x16( size_t n, const int8_t* input_a, const int8_t* input_b, int8_t* output, const union xnn_qs8_add_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) { const __m512i vbias = _mm512_load_si512(params->avx512.bias); const __m512i va_multiplier = _mm512_load_si512(params->avx512.a_multiplier); const __m512i vb_multiplier = _mm512_load_si512(params->avx512.b_multiplier); const __m128i vshift = _mm_load_si128((const __m128i*) params->avx512.shift); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->avx512.output_min); const __m128i voutput_max = _mm_load_si128((const __m128i*) params->avx512.output_max); for (; n >= 16 * sizeof(int8_t); n -= 16 * sizeof(int8_t)) { const __m512i va0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) input_a)); const __m512i vb0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) input_b)); input_a += 16; input_b += 16; __m512i vacc0123456789ABCDEF = _mm512_add_epi32(vbias, _mm512_mullo_epi32(va0123456789ABCDEF, va_multiplier)); vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vb0123456789ABCDEF, vb_multiplier)); vacc0123456789ABCDEF = _mm512_sra_epi32(vacc0123456789ABCDEF, vshift); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), 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)); vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, voutput_min); vout0123456789ABCDEF = _mm_min_epi8(vout0123456789ABCDEF, voutput_max); _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; } if XNN_UNLIKELY(n != 0) { { const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << n) - UINT32_C(1))); const __m512i va0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_maskz_loadu_epi8(vmask, input_a)); const __m512i vb0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_maskz_loadu_epi8(vmask, input_b)); __m512i vacc0123456789ABCDEF = _mm512_add_epi32(vbias, _mm512_mullo_epi32(va0123456789ABCDEF, va_multiplier)); vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vb0123456789ABCDEF, vb_multiplier)); vacc0123456789ABCDEF = _mm512_sra_epi32(vacc0123456789ABCDEF, vshift); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), 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)); vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, voutput_min); vout0123456789ABCDEF = _mm_min_epi8(vout0123456789ABCDEF, voutput_max); _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); } } } void xnn_qs8_vaddc_minmax_ukernel__avx512skx_mul32_ld128_x16( size_t n, const int8_t* input_a, const int8_t* input_b, int8_t* output, const union xnn_qs8_add_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) { const __m512i va_multiplier = _mm512_load_si512(params->avx512.a_multiplier); const __m128i vshift = _mm_load_si128((const __m128i*) params->avx512.shift); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->avx512.output_min); const __m128i voutput_max = _mm_load_si128((const __m128i*) params->avx512.output_max); const __m512i vbias = _mm512_add_epi32( _mm512_broadcastd_epi32(_mm_cvtsi32_si128(params->avx512.b_multiplier[0] * (int32_t) *input_b)), _mm512_load_si512(params->avx512.bias)); for (; n >= 16 * sizeof(int8_t); n -= 16 * sizeof(int8_t)) { const __m512i va0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i*) input_a)); input_a += 16; __m512i vacc0123456789ABCDEF = _mm512_add_epi32(vbias, _mm512_mullo_epi32(va0123456789ABCDEF, va_multiplier)); vacc0123456789ABCDEF = _mm512_sra_epi32(vacc0123456789ABCDEF, vshift); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), 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)); vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, voutput_min); vout0123456789ABCDEF = _mm_min_epi8(vout0123456789ABCDEF, voutput_max); _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; } if XNN_UNLIKELY(n != 0) { { const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << n) - UINT32_C(1))); const __m512i va0123456789ABCDEF = _mm512_cvtepi8_epi32(_mm_maskz_loadu_epi8(vmask, input_a)); __m512i vacc0123456789ABCDEF = _mm512_add_epi32(vbias, _mm512_mullo_epi32(va0123456789ABCDEF, va_multiplier)); vacc0123456789ABCDEF = _mm512_sra_epi32(vacc0123456789ABCDEF, vshift); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), 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)); vout0123456789ABCDEF = _mm_max_epi8(vout0123456789ABCDEF, voutput_min); vout0123456789ABCDEF = _mm_min_epi8(vout0123456789ABCDEF, voutput_max); _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); } } } void xnn_qu8_dwconv_minmax_fp32_ukernel_up32x25__avx512skx_mul32( size_t channels, size_t output_width, const uint8_t** input, const void* weights, uint8_t* output, size_t input_stride, size_t output_increment, size_t input_offset, const uint8_t* zero, const union xnn_qu8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_MSAN { assert(channels != 0); assert(output_width != 0); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m256i voutput_min = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_min); const __m256i vpermute_mask = _mm256_set_epi32(7, 3, 5, 1, 6, 2, 4, 0); const __m512i vk_zero_point = _mm512_cvtepu16_epi32(_mm256_load_si256((const __m256i*) params->fp32_avx512.kernel_zero_point)); do { const uint8_t* i0 = input[0]; assert(i0 != NULL); if XNN_UNPREDICTABLE(i0 != zero) { i0 = (const uint8_t*) ((uintptr_t) i0 + input_offset); } const uint8_t* i1 = input[1]; assert(i1 != NULL); if XNN_UNPREDICTABLE(i1 != zero) { i1 = (const uint8_t*) ((uintptr_t) i1 + input_offset); } const uint8_t* i2 = input[2]; assert(i2 != NULL); if XNN_UNPREDICTABLE(i2 != zero) { i2 = (const uint8_t*) ((uintptr_t) i2 + input_offset); } const uint8_t* i3 = input[3]; assert(i3 != NULL); if XNN_UNPREDICTABLE(i3 != zero) { i3 = (const uint8_t*) ((uintptr_t) i3 + input_offset); } const uint8_t* i4 = input[4]; assert(i4 != NULL); if XNN_UNPREDICTABLE(i4 != zero) { i4 = (const uint8_t*) ((uintptr_t) i4 + input_offset); } const uint8_t* i5 = input[5]; assert(i5 != NULL); if XNN_UNPREDICTABLE(i5 != zero) { i5 = (const uint8_t*) ((uintptr_t) i5 + input_offset); } const uint8_t* i6 = input[6]; assert(i6 != NULL); if XNN_UNPREDICTABLE(i6 != zero) { i6 = (const uint8_t*) ((uintptr_t) i6 + input_offset); } const uint8_t* i7 = input[7]; assert(i7 != NULL); if XNN_UNPREDICTABLE(i7 != zero) { i7 = (const uint8_t*) ((uintptr_t) i7 + input_offset); } const uint8_t* i8 = input[8]; assert(i8 != NULL); if XNN_UNPREDICTABLE(i8 != zero) { i8 = (const uint8_t*) ((uintptr_t) i8 + input_offset); } const uint8_t* i9 = input[9]; assert(i9 != NULL); if XNN_UNPREDICTABLE(i9 != zero) { i9 = (const uint8_t*) ((uintptr_t) i9 + input_offset); } const uint8_t* i10 = input[10]; assert(i10 != NULL); if XNN_UNPREDICTABLE(i10 != zero) { i10 = (const uint8_t*) ((uintptr_t) i10 + input_offset); } const uint8_t* i11 = input[11]; assert(i11 != NULL); if XNN_UNPREDICTABLE(i11 != zero) { i11 = (const uint8_t*) ((uintptr_t) i11 + input_offset); } const uint8_t* i12 = input[12]; assert(i12 != NULL); if XNN_UNPREDICTABLE(i12 != zero) { i12 = (const uint8_t*) ((uintptr_t) i12 + input_offset); } const uint8_t* i13 = input[13]; assert(i13 != NULL); if XNN_UNPREDICTABLE(i13 != zero) { i13 = (const uint8_t*) ((uintptr_t) i13 + input_offset); } const uint8_t* i14 = input[14]; assert(i14 != NULL); if XNN_UNPREDICTABLE(i14 != zero) { i14 = (const uint8_t*) ((uintptr_t) i14 + input_offset); } const uint8_t* i15 = input[15]; assert(i15 != NULL); if XNN_UNPREDICTABLE(i15 != zero) { i15 = (const uint8_t*) ((uintptr_t) i15 + input_offset); } const uint8_t* i16 = input[16]; assert(i16 != NULL); if XNN_UNPREDICTABLE(i16 != zero) { i16 = (const uint8_t*) ((uintptr_t) i16 + input_offset); } const uint8_t* i17 = input[17]; assert(i17 != NULL); if XNN_UNPREDICTABLE(i17 != zero) { i17 = (const uint8_t*) ((uintptr_t) i17 + input_offset); } const uint8_t* i18 = input[18]; assert(i18 != NULL); if XNN_UNPREDICTABLE(i18 != zero) { i18 = (const uint8_t*) ((uintptr_t) i18 + input_offset); } const uint8_t* i19 = input[19]; assert(i19 != NULL); if XNN_UNPREDICTABLE(i19 != zero) { i19 = (const uint8_t*) ((uintptr_t) i19 + input_offset); } const uint8_t* i20 = input[20]; assert(i20 != NULL); if XNN_UNPREDICTABLE(i20 != zero) { i20 = (const uint8_t*) ((uintptr_t) i20 + input_offset); } const uint8_t* i21 = input[21]; assert(i21 != NULL); if XNN_UNPREDICTABLE(i21 != zero) { i21 = (const uint8_t*) ((uintptr_t) i21 + input_offset); } const uint8_t* i22 = input[22]; assert(i22 != NULL); if XNN_UNPREDICTABLE(i22 != zero) { i22 = (const uint8_t*) ((uintptr_t) i22 + input_offset); } const uint8_t* i23 = input[23]; assert(i23 != NULL); if XNN_UNPREDICTABLE(i23 != zero) { i23 = (const uint8_t*) ((uintptr_t) i23 + input_offset); } const uint8_t* i24 = input[24]; assert(i24 != NULL); if XNN_UNPREDICTABLE(i24 != zero) { i24 = (const uint8_t*) ((uintptr_t) i24 + input_offset); } input = (const uint8_t**) ((uintptr_t) input + input_stride); size_t c = channels; const void* w = weights; for (; c >= 32; c -= 32) { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); __m512i vaccGHIJKLMNOPQRSTUV = _mm512_loadu_si512((const void*) ((uintptr_t) w + 16 * sizeof(int32_t))); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 0 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi0xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m512i vk0xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 16 * sizeof(uint8_t)))), vk_zero_point); i0 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi0xGHIJKLMNOPQRSTUV, vk0xGHIJKLMNOPQRSTUV)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 32 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi1xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m512i vk1xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 48 * sizeof(uint8_t)))), vk_zero_point); i1 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi1xGHIJKLMNOPQRSTUV, vk1xGHIJKLMNOPQRSTUV)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 64 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi2xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m512i vk2xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 80 * sizeof(uint8_t)))), vk_zero_point); i2 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi2xGHIJKLMNOPQRSTUV, vk2xGHIJKLMNOPQRSTUV)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 96 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi3xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i3 + 16))); const __m512i vk3xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 112 * sizeof(uint8_t)))), vk_zero_point); i3 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi3xGHIJKLMNOPQRSTUV, vk3xGHIJKLMNOPQRSTUV)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 128 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi4xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i4 + 16))); const __m512i vk4xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 144 * sizeof(uint8_t)))), vk_zero_point); i4 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi4xGHIJKLMNOPQRSTUV, vk4xGHIJKLMNOPQRSTUV)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 160 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi5xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i5 + 16))); const __m512i vk5xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 176 * sizeof(uint8_t)))), vk_zero_point); i5 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi5xGHIJKLMNOPQRSTUV, vk5xGHIJKLMNOPQRSTUV)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 192 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi6xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i6 + 16))); const __m512i vk6xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 208 * sizeof(uint8_t)))), vk_zero_point); i6 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi6xGHIJKLMNOPQRSTUV, vk6xGHIJKLMNOPQRSTUV)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 224 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi7xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i7 + 16))); const __m512i vk7xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 240 * sizeof(uint8_t)))), vk_zero_point); i7 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi7xGHIJKLMNOPQRSTUV, vk7xGHIJKLMNOPQRSTUV)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 256 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi8xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i8 + 16))); const __m512i vk8xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 272 * sizeof(uint8_t)))), vk_zero_point); i8 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi8xGHIJKLMNOPQRSTUV, vk8xGHIJKLMNOPQRSTUV)); const __m512i vi9x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i9)); const __m512i vk9x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 288 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi9xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i9 + 16))); const __m512i vk9xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 304 * sizeof(uint8_t)))), vk_zero_point); i9 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi9x0123456789ABCDEF, vk9x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi9xGHIJKLMNOPQRSTUV, vk9xGHIJKLMNOPQRSTUV)); const __m512i vi10x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i10)); const __m512i vk10x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 320 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi10xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i10 + 16))); const __m512i vk10xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 336 * sizeof(uint8_t)))), vk_zero_point); i10 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi10x0123456789ABCDEF, vk10x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi10xGHIJKLMNOPQRSTUV, vk10xGHIJKLMNOPQRSTUV)); const __m512i vi11x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i11)); const __m512i vk11x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 352 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi11xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i11 + 16))); const __m512i vk11xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 368 * sizeof(uint8_t)))), vk_zero_point); i11 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi11x0123456789ABCDEF, vk11x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi11xGHIJKLMNOPQRSTUV, vk11xGHIJKLMNOPQRSTUV)); const __m512i vi12x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i12)); const __m512i vk12x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 384 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi12xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i12 + 16))); const __m512i vk12xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 400 * sizeof(uint8_t)))), vk_zero_point); i12 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi12x0123456789ABCDEF, vk12x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi12xGHIJKLMNOPQRSTUV, vk12xGHIJKLMNOPQRSTUV)); const __m512i vi13x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i13)); const __m512i vk13x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 416 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi13xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i13 + 16))); const __m512i vk13xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 432 * sizeof(uint8_t)))), vk_zero_point); i13 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi13x0123456789ABCDEF, vk13x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi13xGHIJKLMNOPQRSTUV, vk13xGHIJKLMNOPQRSTUV)); const __m512i vi14x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i14)); const __m512i vk14x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 448 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi14xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i14 + 16))); const __m512i vk14xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 464 * sizeof(uint8_t)))), vk_zero_point); i14 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi14x0123456789ABCDEF, vk14x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi14xGHIJKLMNOPQRSTUV, vk14xGHIJKLMNOPQRSTUV)); const __m512i vi15x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i15)); const __m512i vk15x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 480 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi15xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i15 + 16))); const __m512i vk15xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 496 * sizeof(uint8_t)))), vk_zero_point); i15 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi15x0123456789ABCDEF, vk15x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi15xGHIJKLMNOPQRSTUV, vk15xGHIJKLMNOPQRSTUV)); const __m512i vi16x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i16)); const __m512i vk16x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 512 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi16xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i16 + 16))); const __m512i vk16xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 528 * sizeof(uint8_t)))), vk_zero_point); i16 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi16x0123456789ABCDEF, vk16x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi16xGHIJKLMNOPQRSTUV, vk16xGHIJKLMNOPQRSTUV)); const __m512i vi17x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i17)); const __m512i vk17x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 544 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi17xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i17 + 16))); const __m512i vk17xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 560 * sizeof(uint8_t)))), vk_zero_point); i17 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi17x0123456789ABCDEF, vk17x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi17xGHIJKLMNOPQRSTUV, vk17xGHIJKLMNOPQRSTUV)); const __m512i vi18x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i18)); const __m512i vk18x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 576 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi18xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i18 + 16))); const __m512i vk18xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 592 * sizeof(uint8_t)))), vk_zero_point); i18 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi18x0123456789ABCDEF, vk18x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi18xGHIJKLMNOPQRSTUV, vk18xGHIJKLMNOPQRSTUV)); const __m512i vi19x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i19)); const __m512i vk19x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 608 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi19xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i19 + 16))); const __m512i vk19xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 624 * sizeof(uint8_t)))), vk_zero_point); i19 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi19x0123456789ABCDEF, vk19x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi19xGHIJKLMNOPQRSTUV, vk19xGHIJKLMNOPQRSTUV)); const __m512i vi20x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i20)); const __m512i vk20x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 640 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi20xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i20 + 16))); const __m512i vk20xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 656 * sizeof(uint8_t)))), vk_zero_point); i20 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi20x0123456789ABCDEF, vk20x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi20xGHIJKLMNOPQRSTUV, vk20xGHIJKLMNOPQRSTUV)); const __m512i vi21x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i21)); const __m512i vk21x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 672 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi21xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i21 + 16))); const __m512i vk21xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 688 * sizeof(uint8_t)))), vk_zero_point); i21 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi21x0123456789ABCDEF, vk21x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi21xGHIJKLMNOPQRSTUV, vk21xGHIJKLMNOPQRSTUV)); const __m512i vi22x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i22)); const __m512i vk22x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 704 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi22xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i22 + 16))); const __m512i vk22xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 720 * sizeof(uint8_t)))), vk_zero_point); i22 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi22x0123456789ABCDEF, vk22x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi22xGHIJKLMNOPQRSTUV, vk22xGHIJKLMNOPQRSTUV)); const __m512i vi23x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i23)); const __m512i vk23x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 736 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi23xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i23 + 16))); const __m512i vk23xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 752 * sizeof(uint8_t)))), vk_zero_point); i23 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi23x0123456789ABCDEF, vk23x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi23xGHIJKLMNOPQRSTUV, vk23xGHIJKLMNOPQRSTUV)); const __m512i vi24x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i24)); const __m512i vk24x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 768 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi24xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i24 + 16))); const __m512i vk24xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 784 * sizeof(uint8_t)))), vk_zero_point); i24 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi24x0123456789ABCDEF, vk24x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi24xGHIJKLMNOPQRSTUV, vk24xGHIJKLMNOPQRSTUV)); w = (const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 800 * sizeof(uint8_t)); __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); __m512 vscaledGHIJKLMNOPQRSTUV = _mm512_cvtepi32_ps(vaccGHIJKLMNOPQRSTUV); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale); vscaledGHIJKLMNOPQRSTUV = _mm512_mul_ps(vscaledGHIJKLMNOPQRSTUV, vscale); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vscaledGHIJKLMNOPQRSTUV = _mm512_min_ps(vscaledGHIJKLMNOPQRSTUV, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); vaccGHIJKLMNOPQRSTUV = _mm512_cvtps_epi32(vscaledGHIJKLMNOPQRSTUV); __m512i vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV = _mm512_adds_epi16(_mm512_packs_epi32(vacc0123456789ABCDEF, vaccGHIJKLMNOPQRSTUV), voutput_zero_point); __m256i voutGHIJOPQRKLMNSTUV = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vaccGHIJKLMNOPQRSTUV), _mm512_extracti32x8_epi32(vaccGHIJKLMNOPQRSTUV, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m256i vout0123GHIJ4567KLMN = _mm512_castsi512_si256(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV); const __m256i vout89ABOPQRCDEFSTUV = _mm512_extracti32x8_epi32(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV, 1); const __m256i vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV = _mm256_packus_epi16(vout0123GHIJ4567KLMN, vout89ABOPQRCDEFSTUV); __m256i vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_permutevar8x32_epi32(vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV, vpermute_mask); const __m128i voutGHIJOPQR = _mm256_castsi256_si128(voutGHIJOPQRKLMNSTUV); const __m128i voutKLMNSTUV = _mm256_extracti128_si256(voutGHIJOPQRKLMNSTUV, 1); __m128i voutGHIJKLMNOPQRSTUV = _mm_shuffle_epi32(_mm_packus_epi16(voutGHIJOPQR, voutKLMNSTUV), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_max_epu8(vout0123456789ABCDEFGHIJKLMNOPQRSTUV, voutput_min); voutGHIJKLMNOPQRSTUV = _mm_max_epu8(voutGHIJKLMNOPQRSTUV, _mm256_castsi256_si128(voutput_min)); _mm256_storeu_si256((__m256i*) output, vout0123456789ABCDEFGHIJKLMNOPQRSTUV); _mm_storeu_si128((__m128i*) (output + 16), voutGHIJKLMNOPQRSTUV); output += 32; } if XNN_UNLIKELY(c != 0) { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << (c & 15)) - UINT32_C(1))); const uint8_t* k = (const uint8_t*) ((uintptr_t) w + 32 * sizeof(int32_t)); do { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) k)), vk_zero_point); i0 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 32))), vk_zero_point); i1 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 64))), vk_zero_point); i2 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 96))), vk_zero_point); i3 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 128))), vk_zero_point); i4 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 160))), vk_zero_point); i5 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 192))), vk_zero_point); i6 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 224))), vk_zero_point); i7 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 256))), vk_zero_point); i8 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); const __m512i vi9x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i9)); const __m512i vk9x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 288))), vk_zero_point); i9 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi9x0123456789ABCDEF, vk9x0123456789ABCDEF)); const __m512i vi10x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i10)); const __m512i vk10x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 320))), vk_zero_point); i10 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi10x0123456789ABCDEF, vk10x0123456789ABCDEF)); const __m512i vi11x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i11)); const __m512i vk11x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 352))), vk_zero_point); i11 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi11x0123456789ABCDEF, vk11x0123456789ABCDEF)); const __m512i vi12x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i12)); const __m512i vk12x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 384))), vk_zero_point); i12 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi12x0123456789ABCDEF, vk12x0123456789ABCDEF)); const __m512i vi13x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i13)); const __m512i vk13x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 416))), vk_zero_point); i13 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi13x0123456789ABCDEF, vk13x0123456789ABCDEF)); const __m512i vi14x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i14)); const __m512i vk14x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 448))), vk_zero_point); i14 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi14x0123456789ABCDEF, vk14x0123456789ABCDEF)); const __m512i vi15x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i15)); const __m512i vk15x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 480))), vk_zero_point); i15 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi15x0123456789ABCDEF, vk15x0123456789ABCDEF)); const __m512i vi16x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i16)); const __m512i vk16x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 512))), vk_zero_point); i16 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi16x0123456789ABCDEF, vk16x0123456789ABCDEF)); const __m512i vi17x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i17)); const __m512i vk17x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 544))), vk_zero_point); i17 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi17x0123456789ABCDEF, vk17x0123456789ABCDEF)); const __m512i vi18x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i18)); const __m512i vk18x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 576))), vk_zero_point); i18 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi18x0123456789ABCDEF, vk18x0123456789ABCDEF)); const __m512i vi19x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i19)); const __m512i vk19x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 608))), vk_zero_point); i19 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi19x0123456789ABCDEF, vk19x0123456789ABCDEF)); const __m512i vi20x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i20)); const __m512i vk20x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 640))), vk_zero_point); i20 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi20x0123456789ABCDEF, vk20x0123456789ABCDEF)); const __m512i vi21x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i21)); const __m512i vk21x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 672))), vk_zero_point); i21 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi21x0123456789ABCDEF, vk21x0123456789ABCDEF)); const __m512i vi22x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i22)); const __m512i vk22x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 704))), vk_zero_point); i22 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi22x0123456789ABCDEF, vk22x0123456789ABCDEF)); const __m512i vi23x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i23)); const __m512i vk23x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 736))), vk_zero_point); i23 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi23x0123456789ABCDEF, vk23x0123456789ABCDEF)); const __m512i vi24x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i24)); const __m512i vk24x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 768))), vk_zero_point); i24 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi24x0123456789ABCDEF, vk24x0123456789ABCDEF)); k += 16; __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); w = (const void*) ((uintptr_t) w + 16 * sizeof(int32_t)); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m128i vout012389AB = _mm256_castsi256_si128(vout012389AB4567CDEF); const __m128i vout4567CDEF = _mm256_extracti128_si256(vout012389AB4567CDEF, 1); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packus_epi16(vout012389AB, vout4567CDEF), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epu8(vout0123456789ABCDEF, _mm256_castsi256_si128(voutput_min)); if XNN_LIKELY(c >= 16) { _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; c -= 16; } else { _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); output = (uint8_t*) ((uintptr_t) output + c); c = 0; } } while (c != 0); } output = (uint8_t*) ((uintptr_t) output + output_increment); } while (--output_width != 0); } void xnn_qu8_dwconv_minmax_fp32_ukernel_up32x9__avx512skx_mul32( size_t channels, size_t output_width, const uint8_t** input, const void* weights, uint8_t* output, size_t input_stride, size_t output_increment, size_t input_offset, const uint8_t* zero, const union xnn_qu8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_MSAN { assert(channels != 0); assert(output_width != 0); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m256i voutput_min = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_min); const __m256i vpermute_mask = _mm256_set_epi32(7, 3, 5, 1, 6, 2, 4, 0); const __m512i vk_zero_point = _mm512_cvtepu16_epi32(_mm256_load_si256((const __m256i*) params->fp32_avx512.kernel_zero_point)); do { const uint8_t* i0 = input[0]; assert(i0 != NULL); if XNN_UNPREDICTABLE(i0 != zero) { i0 = (const uint8_t*) ((uintptr_t) i0 + input_offset); } const uint8_t* i1 = input[1]; assert(i1 != NULL); if XNN_UNPREDICTABLE(i1 != zero) { i1 = (const uint8_t*) ((uintptr_t) i1 + input_offset); } const uint8_t* i2 = input[2]; assert(i2 != NULL); if XNN_UNPREDICTABLE(i2 != zero) { i2 = (const uint8_t*) ((uintptr_t) i2 + input_offset); } const uint8_t* i3 = input[3]; assert(i3 != NULL); if XNN_UNPREDICTABLE(i3 != zero) { i3 = (const uint8_t*) ((uintptr_t) i3 + input_offset); } const uint8_t* i4 = input[4]; assert(i4 != NULL); if XNN_UNPREDICTABLE(i4 != zero) { i4 = (const uint8_t*) ((uintptr_t) i4 + input_offset); } const uint8_t* i5 = input[5]; assert(i5 != NULL); if XNN_UNPREDICTABLE(i5 != zero) { i5 = (const uint8_t*) ((uintptr_t) i5 + input_offset); } const uint8_t* i6 = input[6]; assert(i6 != NULL); if XNN_UNPREDICTABLE(i6 != zero) { i6 = (const uint8_t*) ((uintptr_t) i6 + input_offset); } const uint8_t* i7 = input[7]; assert(i7 != NULL); if XNN_UNPREDICTABLE(i7 != zero) { i7 = (const uint8_t*) ((uintptr_t) i7 + input_offset); } const uint8_t* i8 = input[8]; assert(i8 != NULL); if XNN_UNPREDICTABLE(i8 != zero) { i8 = (const uint8_t*) ((uintptr_t) i8 + input_offset); } input = (const uint8_t**) ((uintptr_t) input + input_stride); size_t c = channels; const void* w = weights; for (; c >= 32; c -= 32) { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); __m512i vaccGHIJKLMNOPQRSTUV = _mm512_loadu_si512((const void*) ((uintptr_t) w + 16 * sizeof(int32_t))); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 0 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi0xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i0 + 16))); const __m512i vk0xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 16 * sizeof(uint8_t)))), vk_zero_point); i0 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi0xGHIJKLMNOPQRSTUV, vk0xGHIJKLMNOPQRSTUV)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 32 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi1xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i1 + 16))); const __m512i vk1xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 48 * sizeof(uint8_t)))), vk_zero_point); i1 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi1xGHIJKLMNOPQRSTUV, vk1xGHIJKLMNOPQRSTUV)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 64 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi2xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i2 + 16))); const __m512i vk2xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 80 * sizeof(uint8_t)))), vk_zero_point); i2 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi2xGHIJKLMNOPQRSTUV, vk2xGHIJKLMNOPQRSTUV)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 96 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi3xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i3 + 16))); const __m512i vk3xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 112 * sizeof(uint8_t)))), vk_zero_point); i3 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi3xGHIJKLMNOPQRSTUV, vk3xGHIJKLMNOPQRSTUV)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 128 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi4xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i4 + 16))); const __m512i vk4xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 144 * sizeof(uint8_t)))), vk_zero_point); i4 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi4xGHIJKLMNOPQRSTUV, vk4xGHIJKLMNOPQRSTUV)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 160 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi5xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i5 + 16))); const __m512i vk5xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 176 * sizeof(uint8_t)))), vk_zero_point); i5 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi5xGHIJKLMNOPQRSTUV, vk5xGHIJKLMNOPQRSTUV)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 192 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi6xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i6 + 16))); const __m512i vk6xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 208 * sizeof(uint8_t)))), vk_zero_point); i6 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi6xGHIJKLMNOPQRSTUV, vk6xGHIJKLMNOPQRSTUV)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 224 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi7xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i7 + 16))); const __m512i vk7xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 240 * sizeof(uint8_t)))), vk_zero_point); i7 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi7xGHIJKLMNOPQRSTUV, vk7xGHIJKLMNOPQRSTUV)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 256 * sizeof(uint8_t)))), vk_zero_point); const __m512i vi8xGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (i8 + 16))); const __m512i vk8xGHIJKLMNOPQRSTUV = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_load_si128((const __m128i*) ((uintptr_t) w + 32 * sizeof(int32_t) + 272 * sizeof(uint8_t)))), vk_zero_point); i8 += 32; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); vaccGHIJKLMNOPQRSTUV = _mm512_add_epi32(vaccGHIJKLMNOPQRSTUV, _mm512_mullo_epi32(vi8xGHIJKLMNOPQRSTUV, vk8xGHIJKLMNOPQRSTUV)); w = (const void*) ((uintptr_t) w + 32 * sizeof(int32_t) + 288 * sizeof(uint8_t)); __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); __m512 vscaledGHIJKLMNOPQRSTUV = _mm512_cvtepi32_ps(vaccGHIJKLMNOPQRSTUV); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale); vscaledGHIJKLMNOPQRSTUV = _mm512_mul_ps(vscaledGHIJKLMNOPQRSTUV, vscale); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vscaledGHIJKLMNOPQRSTUV = _mm512_min_ps(vscaledGHIJKLMNOPQRSTUV, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); vaccGHIJKLMNOPQRSTUV = _mm512_cvtps_epi32(vscaledGHIJKLMNOPQRSTUV); __m512i vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV = _mm512_adds_epi16(_mm512_packs_epi32(vacc0123456789ABCDEF, vaccGHIJKLMNOPQRSTUV), voutput_zero_point); __m256i voutGHIJOPQRKLMNSTUV = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vaccGHIJKLMNOPQRSTUV), _mm512_extracti32x8_epi32(vaccGHIJKLMNOPQRSTUV, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m256i vout0123GHIJ4567KLMN = _mm512_castsi512_si256(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV); const __m256i vout89ABOPQRCDEFSTUV = _mm512_extracti32x8_epi32(vout0123GHIJ4567KLMN89ABOPQRCDEFSTUV, 1); const __m256i vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV = _mm256_packus_epi16(vout0123GHIJ4567KLMN, vout89ABOPQRCDEFSTUV); __m256i vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_permutevar8x32_epi32(vout0123GHIJ89ABOPQR4567KLMNCDEFSTUV, vpermute_mask); const __m128i voutGHIJOPQR = _mm256_castsi256_si128(voutGHIJOPQRKLMNSTUV); const __m128i voutKLMNSTUV = _mm256_extracti128_si256(voutGHIJOPQRKLMNSTUV, 1); __m128i voutGHIJKLMNOPQRSTUV = _mm_shuffle_epi32(_mm_packus_epi16(voutGHIJOPQR, voutKLMNSTUV), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEFGHIJKLMNOPQRSTUV = _mm256_max_epu8(vout0123456789ABCDEFGHIJKLMNOPQRSTUV, voutput_min); voutGHIJKLMNOPQRSTUV = _mm_max_epu8(voutGHIJKLMNOPQRSTUV, _mm256_castsi256_si128(voutput_min)); _mm256_storeu_si256((__m256i*) output, vout0123456789ABCDEFGHIJKLMNOPQRSTUV); _mm_storeu_si128((__m128i*) (output + 16), voutGHIJKLMNOPQRSTUV); output += 32; } if XNN_UNLIKELY(c != 0) { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << (c & 15)) - UINT32_C(1))); const uint8_t* k = (const uint8_t*) ((uintptr_t) w + 32 * sizeof(int32_t)); do { __m512i vacc0123456789ABCDEF = _mm512_loadu_si512(w); const __m512i vi0x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i0)); const __m512i vk0x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) k)), vk_zero_point); i0 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi0x0123456789ABCDEF, vk0x0123456789ABCDEF)); const __m512i vi1x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i1)); const __m512i vk1x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 32))), vk_zero_point); i1 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi1x0123456789ABCDEF, vk1x0123456789ABCDEF)); const __m512i vi2x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i2)); const __m512i vk2x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 64))), vk_zero_point); i2 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi2x0123456789ABCDEF, vk2x0123456789ABCDEF)); const __m512i vi3x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i3)); const __m512i vk3x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 96))), vk_zero_point); i3 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi3x0123456789ABCDEF, vk3x0123456789ABCDEF)); const __m512i vi4x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i4)); const __m512i vk4x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 128))), vk_zero_point); i4 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi4x0123456789ABCDEF, vk4x0123456789ABCDEF)); const __m512i vi5x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i5)); const __m512i vk5x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 160))), vk_zero_point); i5 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi5x0123456789ABCDEF, vk5x0123456789ABCDEF)); const __m512i vi6x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i6)); const __m512i vk6x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 192))), vk_zero_point); i6 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi6x0123456789ABCDEF, vk6x0123456789ABCDEF)); const __m512i vi7x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i7)); const __m512i vk7x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 224))), vk_zero_point); i7 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi7x0123456789ABCDEF, vk7x0123456789ABCDEF)); const __m512i vi8x0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) i8)); const __m512i vk8x0123456789ABCDEF = _mm512_sub_epi32(_mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (k + 256))), vk_zero_point); i8 += 16; vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vi8x0123456789ABCDEF, vk8x0123456789ABCDEF)); k += 16; __m512 vscaled0123456789ABCDEF = _mm512_cvtepi32_ps(vacc0123456789ABCDEF); vscaled0123456789ABCDEF = _mm512_mul_ps(vscaled0123456789ABCDEF, vscale); vscaled0123456789ABCDEF = _mm512_min_ps(vscaled0123456789ABCDEF, voutput_max_less_zero_point); vacc0123456789ABCDEF = _mm512_cvtps_epi32(vscaled0123456789ABCDEF); w = (const void*) ((uintptr_t) w + 16 * sizeof(int32_t)); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), _mm512_castsi512_si256(voutput_zero_point)); const __m128i vout012389AB = _mm256_castsi256_si128(vout012389AB4567CDEF); const __m128i vout4567CDEF = _mm256_extracti128_si256(vout012389AB4567CDEF, 1); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packus_epi16(vout012389AB, vout4567CDEF), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epu8(vout0123456789ABCDEF, _mm256_castsi256_si128(voutput_min)); if XNN_LIKELY(c >= 16) { _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; c -= 16; } else { _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); output = (uint8_t*) ((uintptr_t) output + c); c = 0; } } while (c != 0); } output = (uint8_t*) ((uintptr_t) output + output_increment); } while (--output_width != 0); } void xnn_qu8_f32_vcvt_ukernel__avx512skx_x32( size_t n, const uint8_t* x, float* y, const union xnn_qu8_f32_cvt_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(n != 0); assert(n % sizeof(uint8_t) == 0); assert(x != NULL); assert(y != NULL); const __m512i vminus_zero_point = _mm512_load_si512(params->avx512.minus_zero_point); const __m512 vscale = _mm512_load_ps(params->avx512.scale); for (; n >= 32 * sizeof(uint8_t); n -= 32 * sizeof(uint8_t)) { __m512i vx0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) x)); __m512i vxGHIJKLMNOPQRSTUV = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) (x + 16))); x += 32; vx0123456789ABCDEF = _mm512_add_epi32(vx0123456789ABCDEF, vminus_zero_point); vxGHIJKLMNOPQRSTUV = _mm512_add_epi32(vxGHIJKLMNOPQRSTUV, vminus_zero_point); __m512 vy0123456789ABCDEF = _mm512_cvtepi32_ps(vx0123456789ABCDEF); __m512 vyGHIJKLMNOPQRSTUV = _mm512_cvtepi32_ps(vxGHIJKLMNOPQRSTUV); vy0123456789ABCDEF = _mm512_mul_ps(vy0123456789ABCDEF, vscale); vyGHIJKLMNOPQRSTUV = _mm512_mul_ps(vyGHIJKLMNOPQRSTUV, vscale); _mm512_storeu_ps(y, vy0123456789ABCDEF); _mm512_storeu_ps(y + 16, vyGHIJKLMNOPQRSTUV); y += 32; } for (; n >= 16 * sizeof(uint8_t); n -= 16 * sizeof(uint8_t)) { __m512i vx = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) x)); vx = _mm512_add_epi32(vx, vminus_zero_point); x += 16; __m512 vy = _mm512_cvtepi32_ps(vx); vy = _mm512_mul_ps(vy, vscale); _mm512_storeu_ps(y, vy); y += 16; } if XNN_UNLIKELY(n != 0) { assert(n >= 1 * sizeof(uint8_t)); assert(n <= 15 * sizeof(uint8_t)); // Prepare mask for valid elements (depends on n). const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << n) - UINT32_C(1))); __m512i vx = _mm512_cvtepu8_epi32(_mm_maskz_loadu_epi8(vmask, x)); vx = _mm512_add_epi32(vx, vminus_zero_point); __m512 vy = _mm512_cvtepi32_ps(vx); vy = _mm512_mul_ps(vy, vscale); _mm512_mask_storeu_ps(y, vmask, vy); } } void xnn_qu8_gemm_minmax_fp32_ukernel_1x16c8__avx512skx( size_t mr, size_t nc, size_t kc, const uint8_t* restrict a, size_t a_stride, const void* restrict w, uint8_t* restrict c, size_t cm_stride, size_t cn_stride, const union xnn_qu8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 1); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(uint8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); const uint8_t* a0 = a; uint8_t* c0 = c; const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); w = (const void*) ((const int32_t*) w + 16); size_t k = 0; const __m512i vb_zero_point = _mm512_load_si512(params->fp32_avx512.kernel_zero_point); while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i vb0123 = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) w)), vb_zero_point); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); const __m512i vb4567 = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 32))), vb_zero_point); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); const __m512i vb89AB = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 64))), vb_zero_point); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); const __m512i vbCDEF = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 96))), vb_zero_point); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); w = (const void*) ((const uint8_t*) w + 128); k += 8 * sizeof(uint8_t); } const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); const __m256i vacc0x084C2A6E195D3B7F = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0x084C195D2A6E3B7F), _mm512_extracti32x8_epi32(vacc0x084C195D2A6E3B7F, 1)), voutput_zero_point); const __m128i vout0x084C2A6E195D3B7F = _mm_packus_epi16(_mm256_castsi256_si128(vacc0x084C2A6E195D3B7F), _mm256_extracti128_si256(vacc0x084C2A6E195D3B7F, 1)); __m128i vout0x0123456789ABCDEF = _mm_shuffle_epi8(vout0x084C2A6E195D3B7F, _mm_set_epi8(15, 7, 11, 3, 13, 5, 9, 1, 14, 6, 10, 2, 12, 4, 8, 0)); vout0x0123456789ABCDEF = _mm_max_epu8(vout0x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c0, vout0x0123456789ABCDEF); a0 = (const uint8_t*) ((uintptr_t) a0 - k); c0 = (uint8_t*) ((uintptr_t) c0 + cn_stride); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT32_C(1) << nc) - UINT32_C(1))); _mm_mask_storeu_epi8(c0, vmask, vout0x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qu8_gemm_minmax_fp32_ukernel_4x16c8__avx512skx( size_t mr, size_t nc, size_t kc, const uint8_t* restrict a, size_t a_stride, const void* restrict w, uint8_t* restrict c, size_t cm_stride, size_t cn_stride, const union xnn_qu8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 4); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(uint8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); const uint8_t* a0 = a; uint8_t* c0 = c; const uint8_t* a1 = (const uint8_t*) ((uintptr_t) a0 + a_stride); uint8_t* c1 = (uint8_t*) ((uintptr_t) c0 + cm_stride); if XNN_UNPREDICTABLE(mr < 2) { a1 = a0; c1 = c0; } const uint8_t* a2 = (const uint8_t*) ((uintptr_t) a1 + a_stride); uint8_t* c2 = (uint8_t*) ((uintptr_t) c1 + cm_stride); if XNN_UNPREDICTABLE(mr <= 2) { a2 = a1; c2 = c1; } const uint8_t* a3 = (const uint8_t*) ((uintptr_t) a2 + a_stride); uint8_t* c3 = (uint8_t*) ((uintptr_t) c2 + cm_stride); if XNN_UNPREDICTABLE(mr != 4) { a3 = a2; c3 = c2; } const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m512i voutput_min = _mm512_load_si512(params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); __m512i vacc1x0123 = vacc0x0123; __m512i vacc1x4567 = vacc0x4567; __m512i vacc1x89AB = vacc0x89AB; __m512i vacc1xCDEF = vacc0xCDEF; __m512i vacc2x0123 = vacc0x0123; __m512i vacc2x4567 = vacc0x4567; __m512i vacc2x89AB = vacc0x89AB; __m512i vacc2xCDEF = vacc0xCDEF; __m512i vacc3x0123 = vacc0x0123; __m512i vacc3x4567 = vacc0x4567; __m512i vacc3x89AB = vacc0x89AB; __m512i vacc3xCDEF = vacc0xCDEF; w = (const void*) ((const int32_t*) w + 16); size_t k = 0; const __m512i vb_zero_point = _mm512_load_si512(params->fp32_avx512.kernel_zero_point); while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i va1 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a1))); a1 += 8; const __m512i va2 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a2))); a2 += 8; const __m512i va3 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a3))); a3 += 8; const __m512i vb0123 = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) w)), vb_zero_point); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); vacc1x0123 = _mm512_add_epi32(vacc1x0123, _mm512_madd_epi16(va1, vb0123)); vacc2x0123 = _mm512_add_epi32(vacc2x0123, _mm512_madd_epi16(va2, vb0123)); vacc3x0123 = _mm512_add_epi32(vacc3x0123, _mm512_madd_epi16(va3, vb0123)); const __m512i vb4567 = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 32))), vb_zero_point); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); vacc1x4567 = _mm512_add_epi32(vacc1x4567, _mm512_madd_epi16(va1, vb4567)); vacc2x4567 = _mm512_add_epi32(vacc2x4567, _mm512_madd_epi16(va2, vb4567)); vacc3x4567 = _mm512_add_epi32(vacc3x4567, _mm512_madd_epi16(va3, vb4567)); const __m512i vb89AB = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 64))), vb_zero_point); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); vacc1x89AB = _mm512_add_epi32(vacc1x89AB, _mm512_madd_epi16(va1, vb89AB)); vacc2x89AB = _mm512_add_epi32(vacc2x89AB, _mm512_madd_epi16(va2, vb89AB)); vacc3x89AB = _mm512_add_epi32(vacc3x89AB, _mm512_madd_epi16(va3, vb89AB)); const __m512i vbCDEF = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 96))), vb_zero_point); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); vacc1xCDEF = _mm512_add_epi32(vacc1xCDEF, _mm512_madd_epi16(va1, vbCDEF)); vacc2xCDEF = _mm512_add_epi32(vacc2xCDEF, _mm512_madd_epi16(va2, vbCDEF)); vacc3xCDEF = _mm512_add_epi32(vacc3xCDEF, _mm512_madd_epi16(va3, vbCDEF)); w = (const void*) ((const uint8_t*) w + 128); k += 8 * sizeof(uint8_t); } const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); const __m512i vacc1x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x0123, vacc1x4567), _mm512_unpackhi_epi32(vacc1x0123, vacc1x4567)); const __m512i vacc1x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x89AB, vacc1xCDEF), _mm512_unpackhi_epi32(vacc1x89AB, vacc1xCDEF)); const __m512i vacc2x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x0123, vacc2x4567), _mm512_unpackhi_epi32(vacc2x0123, vacc2x4567)); const __m512i vacc2x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x89AB, vacc2xCDEF), _mm512_unpackhi_epi32(vacc2x89AB, vacc2xCDEF)); const __m512i vacc3x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x0123, vacc3x4567), _mm512_unpackhi_epi32(vacc3x0123, vacc3x4567)); const __m512i vacc3x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x89AB, vacc3xCDEF), _mm512_unpackhi_epi32(vacc3x89AB, vacc3xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512i vacc1x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x04152637, vacc1x8C9DAEBF), _mm512_unpackhi_epi32(vacc1x04152637, vacc1x8C9DAEBF)); __m512i vacc2x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x04152637, vacc2x8C9DAEBF), _mm512_unpackhi_epi32(vacc2x04152637, vacc2x8C9DAEBF)); __m512i vacc3x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x04152637, vacc3x8C9DAEBF), _mm512_unpackhi_epi32(vacc3x04152637, vacc3x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); __m512 vscaled1x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc1x084C195D2A6E3B7F); __m512 vscaled2x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc2x084C195D2A6E3B7F); __m512 vscaled3x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc3x084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale); vscaled1x084C195D2A6E3B7F = _mm512_mul_ps(vscaled1x084C195D2A6E3B7F, vscale); vscaled2x084C195D2A6E3B7F = _mm512_mul_ps(vscaled2x084C195D2A6E3B7F, vscale); vscaled3x084C195D2A6E3B7F = _mm512_mul_ps(vscaled3x084C195D2A6E3B7F, vscale); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled1x084C195D2A6E3B7F = _mm512_min_ps(vscaled1x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled2x084C195D2A6E3B7F = _mm512_min_ps(vscaled2x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled3x084C195D2A6E3B7F = _mm512_min_ps(vscaled3x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); vacc1x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled1x084C195D2A6E3B7F); vacc2x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled2x084C195D2A6E3B7F); vacc3x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled3x084C195D2A6E3B7F); const __m512i vacc01x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc0x084C195D2A6E3B7F, vacc1x084C195D2A6E3B7F), voutput_zero_point); const __m512i vacc23x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc2x084C195D2A6E3B7F, vacc3x084C195D2A6E3B7F), voutput_zero_point); __m512i vout0123x084Cx195Dx2A6Ex3B7F = _mm512_packus_epi16(vacc01x084Cx195Dx2A6Ex3B7F, vacc23x084Cx195Dx2A6Ex3B7F); vout0123x084Cx195Dx2A6Ex3B7F = _mm512_permutexvar_epi32(_mm512_set_epi32(15, 11, 7, 3, 14, 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0), vout0123x084Cx195Dx2A6Ex3B7F); __m512i vout0123x0123456789ABCDEF = _mm512_shuffle_epi8(vout0123x084Cx195Dx2A6Ex3B7F, _mm512_set_epi8(15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0)); vout0123x0123456789ABCDEF = _mm512_max_epu8(vout0123x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c0, _mm512_castsi512_si128(vout0123x0123456789ABCDEF)); _mm_storeu_si128((__m128i*) c1, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 1)); _mm_storeu_si128((__m128i*) c2, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 2)); _mm_storeu_si128((__m128i*) c3, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 3)); a0 = (const uint8_t*) ((uintptr_t) a0 - k); a1 = (const uint8_t*) ((uintptr_t) a1 - k); a2 = (const uint8_t*) ((uintptr_t) a2 - k); a3 = (const uint8_t*) ((uintptr_t) a3 - k); c0 = (uint8_t*) ((uintptr_t) c0 + cn_stride); c1 = (uint8_t*) ((uintptr_t) c1 + cn_stride); c2 = (uint8_t*) ((uintptr_t) c2 + cn_stride); c3 = (uint8_t*) ((uintptr_t) c3 + cn_stride); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT32_C(1) << nc) - UINT32_C(1))); _mm512_mask_storeu_epi8(c0, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftli_mask64(vmask, 16); _mm512_mask_storeu_epi8(c1 - 16, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftli_mask64(vmask, 16); _mm512_mask_storeu_epi8(c2 - 32, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftli_mask64(vmask, 16); _mm512_mask_storeu_epi8(c3 - 48, vmask, vout0123x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qu8_igemm_minmax_fp32_ukernel_1x16c8__avx512skx( size_t mr, size_t nc, size_t kc, size_t ks, const uint8_t** restrict a, const void* restrict w, uint8_t* restrict c, size_t cm_stride, size_t cn_stride, size_t a_offset, const uint8_t* zero, const union xnn_qu8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 1); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(uint8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); uint8_t* c0 = c; const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->fp32_avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); w = (const void*) ((const int32_t*) w + 16); size_t p = ks; do { const uint8_t* restrict a0 = a[0]; if XNN_UNPREDICTABLE(a0 != zero) { a0 = (const uint8_t*) ((uintptr_t) a0 + a_offset); } a += 1; size_t k = 0; const __m512i vb_zero_point = _mm512_load_si512(params->fp32_avx512.kernel_zero_point); while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i vb0123 = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) w)), vb_zero_point); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); const __m512i vb4567 = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 32))), vb_zero_point); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); const __m512i vb89AB = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 64))), vb_zero_point); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); const __m512i vbCDEF = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 96))), vb_zero_point); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); w = (const void*) ((const uint8_t*) w + 128); k += 8 * sizeof(uint8_t); } p -= 1 * sizeof(void*); } while (p != 0); const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); const __m256i vacc0x084C2A6E195D3B7F = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0x084C195D2A6E3B7F), _mm512_extracti32x8_epi32(vacc0x084C195D2A6E3B7F, 1)), voutput_zero_point); const __m128i vout0x084C2A6E195D3B7F = _mm_packus_epi16(_mm256_castsi256_si128(vacc0x084C2A6E195D3B7F), _mm256_extracti128_si256(vacc0x084C2A6E195D3B7F, 1)); __m128i vout0x0123456789ABCDEF = _mm_shuffle_epi8(vout0x084C2A6E195D3B7F, _mm_set_epi8(15, 7, 11, 3, 13, 5, 9, 1, 14, 6, 10, 2, 12, 4, 8, 0)); vout0x0123456789ABCDEF = _mm_max_epu8(vout0x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c0, vout0x0123456789ABCDEF); c0 = (uint8_t*) ((uintptr_t) c0 + cn_stride); a = (const uint8_t**restrict) ((uintptr_t) a - ks); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT32_C(1) << nc) - UINT32_C(1))); _mm_mask_storeu_epi8(c0, vmask, vout0x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qu8_igemm_minmax_fp32_ukernel_4x16c8__avx512skx( size_t mr, size_t nc, size_t kc, size_t ks, const uint8_t** restrict a, const void* restrict w, uint8_t* restrict c, size_t cm_stride, size_t cn_stride, size_t a_offset, const uint8_t* zero, const union xnn_qu8_conv_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(mr != 0); assert(mr <= 4); assert(nc != 0); assert(kc != 0); assert(kc % sizeof(uint8_t) == 0); assert(a != NULL); assert(w != NULL); assert(c != NULL); kc = round_up_po2(kc, 8); uint8_t* c0 = c; uint8_t* c1 = (uint8_t*) ((uintptr_t) c0 + cm_stride); if XNN_UNPREDICTABLE(mr < 2) { c1 = c0; } uint8_t* c2 = (uint8_t*) ((uintptr_t) c1 + cm_stride); if XNN_UNPREDICTABLE(mr <= 2) { c2 = c1; } uint8_t* c3 = (uint8_t*) ((uintptr_t) c2 + cm_stride); if XNN_UNPREDICTABLE(mr != 4) { c3 = c2; } const __mmask16 vbias_mask = _cvtu32_mask16(0x1111); const __m512 vscale = _mm512_load_ps(params->fp32_avx512.scale); const __m512 voutput_max_less_zero_point = _mm512_load_ps(params->fp32_avx512.output_max_less_zero_point); const __m512i voutput_zero_point = _mm512_load_si512(params->fp32_avx512.output_zero_point); const __m512i voutput_min = _mm512_load_si512(params->fp32_avx512.output_min); do { __m512i vacc0x0123 = _mm512_maskz_expandloadu_epi32(vbias_mask, w); __m512i vacc0x4567 = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 4)); __m512i vacc0x89AB = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 8)); __m512i vacc0xCDEF = _mm512_maskz_expandloadu_epi32(vbias_mask, (const void*) ((const int32_t*) w + 12)); __m512i vacc1x0123 = vacc0x0123; __m512i vacc1x4567 = vacc0x4567; __m512i vacc1x89AB = vacc0x89AB; __m512i vacc1xCDEF = vacc0xCDEF; __m512i vacc2x0123 = vacc0x0123; __m512i vacc2x4567 = vacc0x4567; __m512i vacc2x89AB = vacc0x89AB; __m512i vacc2xCDEF = vacc0xCDEF; __m512i vacc3x0123 = vacc0x0123; __m512i vacc3x4567 = vacc0x4567; __m512i vacc3x89AB = vacc0x89AB; __m512i vacc3xCDEF = vacc0xCDEF; w = (const void*) ((const int32_t*) w + 16); size_t p = ks; do { const uint8_t* restrict a0 = a[0]; if XNN_UNPREDICTABLE(a0 != zero) { a0 = (const uint8_t*) ((uintptr_t) a0 + a_offset); } const uint8_t* restrict a1 = a[1]; if XNN_UNPREDICTABLE(a1 != zero) { a1 = (const uint8_t*) ((uintptr_t) a1 + a_offset); } const uint8_t* restrict a2 = a[2]; if XNN_UNPREDICTABLE(a2 != zero) { a2 = (const uint8_t*) ((uintptr_t) a2 + a_offset); } const uint8_t* restrict a3 = a[3]; if XNN_UNPREDICTABLE(a3 != zero) { a3 = (const uint8_t*) ((uintptr_t) a3 + a_offset); } a += 4; size_t k = 0; const __m512i vb_zero_point = _mm512_load_si512(params->fp32_avx512.kernel_zero_point); while (k < kc) { const __m512i va0 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a0))); a0 += 8; const __m512i va1 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a1))); a1 += 8; const __m512i va2 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a2))); a2 += 8; const __m512i va3 = _mm512_broadcast_i32x4(_mm_cvtepu8_epi16(_mm_loadl_epi64((const __m128i*) a3))); a3 += 8; const __m512i vb0123 = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) w)), vb_zero_point); vacc0x0123 = _mm512_add_epi32(vacc0x0123, _mm512_madd_epi16(va0, vb0123)); vacc1x0123 = _mm512_add_epi32(vacc1x0123, _mm512_madd_epi16(va1, vb0123)); vacc2x0123 = _mm512_add_epi32(vacc2x0123, _mm512_madd_epi16(va2, vb0123)); vacc3x0123 = _mm512_add_epi32(vacc3x0123, _mm512_madd_epi16(va3, vb0123)); const __m512i vb4567 = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 32))), vb_zero_point); vacc0x4567 = _mm512_add_epi32(vacc0x4567, _mm512_madd_epi16(va0, vb4567)); vacc1x4567 = _mm512_add_epi32(vacc1x4567, _mm512_madd_epi16(va1, vb4567)); vacc2x4567 = _mm512_add_epi32(vacc2x4567, _mm512_madd_epi16(va2, vb4567)); vacc3x4567 = _mm512_add_epi32(vacc3x4567, _mm512_madd_epi16(va3, vb4567)); const __m512i vb89AB = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 64))), vb_zero_point); vacc0x89AB = _mm512_add_epi32(vacc0x89AB, _mm512_madd_epi16(va0, vb89AB)); vacc1x89AB = _mm512_add_epi32(vacc1x89AB, _mm512_madd_epi16(va1, vb89AB)); vacc2x89AB = _mm512_add_epi32(vacc2x89AB, _mm512_madd_epi16(va2, vb89AB)); vacc3x89AB = _mm512_add_epi32(vacc3x89AB, _mm512_madd_epi16(va3, vb89AB)); const __m512i vbCDEF = _mm512_sub_epi16(_mm512_cvtepu8_epi16(_mm256_load_si256((const __m256i*) ((const uint8_t*) w + 96))), vb_zero_point); vacc0xCDEF = _mm512_add_epi32(vacc0xCDEF, _mm512_madd_epi16(va0, vbCDEF)); vacc1xCDEF = _mm512_add_epi32(vacc1xCDEF, _mm512_madd_epi16(va1, vbCDEF)); vacc2xCDEF = _mm512_add_epi32(vacc2xCDEF, _mm512_madd_epi16(va2, vbCDEF)); vacc3xCDEF = _mm512_add_epi32(vacc3xCDEF, _mm512_madd_epi16(va3, vbCDEF)); w = (const void*) ((const uint8_t*) w + 128); k += 8 * sizeof(uint8_t); } p -= 4 * sizeof(void*); } while (p != 0); const __m512i vacc0x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x0123, vacc0x4567), _mm512_unpackhi_epi32(vacc0x0123, vacc0x4567)); const __m512i vacc0x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x89AB, vacc0xCDEF), _mm512_unpackhi_epi32(vacc0x89AB, vacc0xCDEF)); const __m512i vacc1x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x0123, vacc1x4567), _mm512_unpackhi_epi32(vacc1x0123, vacc1x4567)); const __m512i vacc1x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x89AB, vacc1xCDEF), _mm512_unpackhi_epi32(vacc1x89AB, vacc1xCDEF)); const __m512i vacc2x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x0123, vacc2x4567), _mm512_unpackhi_epi32(vacc2x0123, vacc2x4567)); const __m512i vacc2x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x89AB, vacc2xCDEF), _mm512_unpackhi_epi32(vacc2x89AB, vacc2xCDEF)); const __m512i vacc3x04152637 = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x0123, vacc3x4567), _mm512_unpackhi_epi32(vacc3x0123, vacc3x4567)); const __m512i vacc3x8C9DAEBF = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x89AB, vacc3xCDEF), _mm512_unpackhi_epi32(vacc3x89AB, vacc3xCDEF)); __m512i vacc0x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc0x04152637, vacc0x8C9DAEBF), _mm512_unpackhi_epi32(vacc0x04152637, vacc0x8C9DAEBF)); __m512i vacc1x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc1x04152637, vacc1x8C9DAEBF), _mm512_unpackhi_epi32(vacc1x04152637, vacc1x8C9DAEBF)); __m512i vacc2x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc2x04152637, vacc2x8C9DAEBF), _mm512_unpackhi_epi32(vacc2x04152637, vacc2x8C9DAEBF)); __m512i vacc3x084C195D2A6E3B7F = _mm512_add_epi32(_mm512_unpacklo_epi32(vacc3x04152637, vacc3x8C9DAEBF), _mm512_unpackhi_epi32(vacc3x04152637, vacc3x8C9DAEBF)); __m512 vscaled0x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc0x084C195D2A6E3B7F); __m512 vscaled1x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc1x084C195D2A6E3B7F); __m512 vscaled2x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc2x084C195D2A6E3B7F); __m512 vscaled3x084C195D2A6E3B7F = _mm512_cvtepi32_ps(vacc3x084C195D2A6E3B7F); vscaled0x084C195D2A6E3B7F = _mm512_mul_ps(vscaled0x084C195D2A6E3B7F, vscale); vscaled1x084C195D2A6E3B7F = _mm512_mul_ps(vscaled1x084C195D2A6E3B7F, vscale); vscaled2x084C195D2A6E3B7F = _mm512_mul_ps(vscaled2x084C195D2A6E3B7F, vscale); vscaled3x084C195D2A6E3B7F = _mm512_mul_ps(vscaled3x084C195D2A6E3B7F, vscale); vscaled0x084C195D2A6E3B7F = _mm512_min_ps(vscaled0x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled1x084C195D2A6E3B7F = _mm512_min_ps(vscaled1x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled2x084C195D2A6E3B7F = _mm512_min_ps(vscaled2x084C195D2A6E3B7F, voutput_max_less_zero_point); vscaled3x084C195D2A6E3B7F = _mm512_min_ps(vscaled3x084C195D2A6E3B7F, voutput_max_less_zero_point); vacc0x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled0x084C195D2A6E3B7F); vacc1x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled1x084C195D2A6E3B7F); vacc2x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled2x084C195D2A6E3B7F); vacc3x084C195D2A6E3B7F = _mm512_cvtps_epi32(vscaled3x084C195D2A6E3B7F); const __m512i vacc01x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc0x084C195D2A6E3B7F, vacc1x084C195D2A6E3B7F), voutput_zero_point); const __m512i vacc23x084Cx195Dx2A6Ex3B7F = _mm512_adds_epi16(_mm512_packs_epi32(vacc2x084C195D2A6E3B7F, vacc3x084C195D2A6E3B7F), voutput_zero_point); __m512i vout0123x084Cx195Dx2A6Ex3B7F = _mm512_packus_epi16(vacc01x084Cx195Dx2A6Ex3B7F, vacc23x084Cx195Dx2A6Ex3B7F); vout0123x084Cx195Dx2A6Ex3B7F = _mm512_permutexvar_epi32(_mm512_set_epi32(15, 11, 7, 3, 14, 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0), vout0123x084Cx195Dx2A6Ex3B7F); __m512i vout0123x0123456789ABCDEF = _mm512_shuffle_epi8(vout0123x084Cx195Dx2A6Ex3B7F, _mm512_set_epi8(15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0, 15, 11, 7, 3, 13, 9, 5, 1, 14, 10, 6, 2, 12, 8, 4, 0)); vout0123x0123456789ABCDEF = _mm512_max_epu8(vout0123x0123456789ABCDEF, voutput_min); if (nc >= 16) { _mm_storeu_si128((__m128i*) c3, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 3)); _mm_storeu_si128((__m128i*) c2, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 2)); _mm_storeu_si128((__m128i*) c1, _mm512_extracti32x4_epi32(vout0123x0123456789ABCDEF, 1)); _mm_storeu_si128((__m128i*) c0, _mm512_castsi512_si128(vout0123x0123456789ABCDEF)); c3 = (uint8_t*) ((uintptr_t) c3 + cn_stride); c2 = (uint8_t*) ((uintptr_t) c2 + cn_stride); c1 = (uint8_t*) ((uintptr_t) c1 + cn_stride); c0 = (uint8_t*) ((uintptr_t) c0 + cn_stride); a = (const uint8_t**restrict) ((uintptr_t) a - ks); nc -= 16; } else { // Prepare mask for valid 8-bit elements (depends on nc). __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT64_C(1) << (nc + 48)) - (UINT64_C(1) << 48))); _mm512_mask_storeu_epi8(c3 - 48, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftri_mask64(vmask, 16); _mm512_mask_storeu_epi8(c2 - 32, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftri_mask64(vmask, 16); _mm512_mask_storeu_epi8(c1 - 16, vmask, vout0123x0123456789ABCDEF); vmask = _kshiftri_mask64(vmask, 16); _mm512_mask_storeu_epi8(c0, vmask, vout0123x0123456789ABCDEF); nc = 0; } } while (nc != 0); } void xnn_qu8_vadd_minmax_ukernel__avx512skx_mul32_ld128_x16( size_t n, const uint8_t* input_a, const uint8_t* input_b, uint8_t* output, const union xnn_qu8_add_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) { const __m512i vbias = _mm512_load_si512(params->avx512.bias); const __m512i va_multiplier = _mm512_load_si512(params->avx512.a_multiplier); const __m512i vb_multiplier = _mm512_load_si512(params->avx512.b_multiplier); const __m128i vshift = _mm_load_si128((const __m128i*) params->avx512.shift); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->avx512.output_min); const __m128i voutput_max = _mm_load_si128((const __m128i*) params->avx512.output_max); for (; n >= 16 * sizeof(uint8_t); n -= 16 * sizeof(uint8_t)) { const __m512i va0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) input_a)); const __m512i vb0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) input_b)); input_a += 16; input_b += 16; __m512i vacc0123456789ABCDEF = _mm512_add_epi32(vbias, _mm512_mullo_epi32(va0123456789ABCDEF, va_multiplier)); vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vb0123456789ABCDEF, vb_multiplier)); vacc0123456789ABCDEF = _mm512_sra_epi32(vacc0123456789ABCDEF, vshift); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), voutput_zero_point); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packus_epi16(_mm256_castsi256_si128(vout012389AB4567CDEF), _mm256_extracti128_si256(vout012389AB4567CDEF, 1)), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epu8(vout0123456789ABCDEF, voutput_min); vout0123456789ABCDEF = _mm_min_epu8(vout0123456789ABCDEF, voutput_max); _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; } if XNN_UNLIKELY(n != 0) { { const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << n) - UINT32_C(1))); const __m512i va0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_maskz_loadu_epi8(vmask, input_a)); const __m512i vb0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_maskz_loadu_epi8(vmask, input_b)); __m512i vacc0123456789ABCDEF = _mm512_add_epi32(vbias, _mm512_mullo_epi32(va0123456789ABCDEF, va_multiplier)); vacc0123456789ABCDEF = _mm512_add_epi32(vacc0123456789ABCDEF, _mm512_mullo_epi32(vb0123456789ABCDEF, vb_multiplier)); vacc0123456789ABCDEF = _mm512_sra_epi32(vacc0123456789ABCDEF, vshift); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), voutput_zero_point); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packus_epi16(_mm256_castsi256_si128(vout012389AB4567CDEF), _mm256_extracti128_si256(vout012389AB4567CDEF, 1)), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epu8(vout0123456789ABCDEF, voutput_min); vout0123456789ABCDEF = _mm_min_epu8(vout0123456789ABCDEF, voutput_max); _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); } } } void xnn_qu8_vaddc_minmax_ukernel__avx512skx_mul32_ld128_x16( size_t n, const uint8_t* input_a, const uint8_t* input_b, uint8_t* output, const union xnn_qu8_add_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) { const __m512i va_multiplier = _mm512_load_si512(params->avx512.a_multiplier); const __m128i vshift = _mm_load_si128((const __m128i*) params->avx512.shift); const __m256i voutput_zero_point = _mm256_load_si256((const __m256i*) params->avx512.output_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->avx512.output_min); const __m128i voutput_max = _mm_load_si128((const __m128i*) params->avx512.output_max); const __m512i vbias = _mm512_add_epi32( _mm512_broadcastd_epi32(_mm_cvtsi32_si128(params->avx512.b_multiplier[0] * (int32_t) *input_b)), _mm512_load_si512(params->avx512.bias)); for (; n >= 16 * sizeof(uint8_t); n -= 16 * sizeof(uint8_t)) { const __m512i va0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_loadu_si128((const __m128i*) input_a)); input_a += 16; __m512i vacc0123456789ABCDEF = _mm512_add_epi32(vbias, _mm512_mullo_epi32(va0123456789ABCDEF, va_multiplier)); vacc0123456789ABCDEF = _mm512_sra_epi32(vacc0123456789ABCDEF, vshift); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), voutput_zero_point); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packus_epi16(_mm256_castsi256_si128(vout012389AB4567CDEF), _mm256_extracti128_si256(vout012389AB4567CDEF, 1)), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epu8(vout0123456789ABCDEF, voutput_min); vout0123456789ABCDEF = _mm_min_epu8(vout0123456789ABCDEF, voutput_max); _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; } if XNN_UNLIKELY(n != 0) { { const __mmask16 vmask = _cvtu32_mask16((uint32_t) ((UINT32_C(1) << n) - UINT32_C(1))); const __m512i va0123456789ABCDEF = _mm512_cvtepu8_epi32(_mm_maskz_loadu_epi8(vmask, input_a)); __m512i vacc0123456789ABCDEF = _mm512_add_epi32(vbias, _mm512_mullo_epi32(va0123456789ABCDEF, va_multiplier)); vacc0123456789ABCDEF = _mm512_sra_epi32(vacc0123456789ABCDEF, vshift); __m256i vout012389AB4567CDEF = _mm256_adds_epi16(_mm256_packs_epi32(_mm512_castsi512_si256(vacc0123456789ABCDEF), _mm512_extracti32x8_epi32(vacc0123456789ABCDEF, 1)), voutput_zero_point); __m128i vout0123456789ABCDEF = _mm_shuffle_epi32(_mm_packus_epi16(_mm256_castsi256_si128(vout012389AB4567CDEF), _mm256_extracti128_si256(vout012389AB4567CDEF, 1)), _MM_SHUFFLE(3, 1, 2, 0)); vout0123456789ABCDEF = _mm_max_epu8(vout0123456789ABCDEF, voutput_min); vout0123456789ABCDEF = _mm_min_epu8(vout0123456789ABCDEF, voutput_max); _mm_mask_storeu_epi8(output, vmask, vout0123456789ABCDEF); } } } void xnn_x8_lut_ukernel__avx512skx_vpshufb_x64( size_t n, const uint8_t* x, uint8_t* y, const uint8_t t[restrict XNN_MIN_ELEMENTS(256)]) { assert(n != 0); assert(x != NULL); assert(y != NULL); const __m512i vt0 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) t)); const __m512i vt1 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 16))); const __m512i vt2 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 32))); const __m512i vt3 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 48))); const __m512i vt4 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 64))); const __m512i vt5 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 80))); const __m512i vt6 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 96))); const __m512i vt7 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 112))); const __m512i vt8 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 128))); const __m512i vt9 = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 144))); const __m512i vtA = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 160))); const __m512i vtB = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 176))); const __m512i vtC = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 192))); const __m512i vtD = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 208))); const __m512i vtE = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 224))); const __m512i vtF = _mm512_broadcast_i32x4(_mm_load_si128((const __m128i*) (t + 240))); const __m512i vtable0 = vt0; const __m512i vtable1 = _mm512_xor_si512(vt0, vt1); const __m512i vtable2 = _mm512_xor_si512(vt1, vt2); const __m512i vtable3 = _mm512_xor_si512(vt2, vt3); const __m512i vtable4 = _mm512_xor_si512(vt3, vt4); const __m512i vtable5 = _mm512_xor_si512(vt4, vt5); const __m512i vtable6 = _mm512_xor_si512(vt5, vt6); const __m512i vtable7 = _mm512_xor_si512(vt6, vt7); const __m512i vtable8 = _mm512_xor_si512(_mm512_xor_si512(vt7, vt8), vtable0); const __m512i vtable9 = _mm512_xor_si512(_mm512_xor_si512(vt8, vt9), vtable1); const __m512i vtableA = _mm512_xor_si512(_mm512_xor_si512(vt9, vtA), vtable2); const __m512i vtableB = _mm512_xor_si512(_mm512_xor_si512(vtA, vtB), vtable3); const __m512i vtableC = _mm512_xor_si512(_mm512_xor_si512(vtB, vtC), vtable4); const __m512i vtableD = _mm512_xor_si512(_mm512_xor_si512(vtC, vtD), vtable5); const __m512i vtableE = _mm512_xor_si512(_mm512_xor_si512(vtD, vtE), vtable6); const __m512i vtableF = _mm512_xor_si512(_mm512_xor_si512(vtE, vtF), vtable7); const __m512i voffset = _mm512_set1_epi8(16); for (; n >= 64 * sizeof(uint8_t); n -= 64 * sizeof(uint8_t)) { __m512i vx = _mm512_loadu_si512(x); x += 64; __m512i vy = _mm512_shuffle_epi8(vtable0, vx); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable1, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable2, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable3, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable4, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable5, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable6, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable7, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable8, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable9, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableA, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableB, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableC, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableD, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableE, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableF, vx)); _mm512_storeu_si512(y, vy); y += 64; } if XNN_UNLIKELY(n != 0) { assert(n < 64); const __mmask64 vmask = _cvtu64_mask64((uint64_t) ((UINT64_C(1) << n) - UINT64_C(1))); __m512i vx = _mm512_maskz_loadu_epi8(vmask, x); __m512i vy = _mm512_shuffle_epi8(vtable0, vx); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable1, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable2, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable3, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable4, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable5, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable6, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable7, vx)); vx = _mm512_sub_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable8, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtable9, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableA, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableB, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableC, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableD, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableE, vx)); vx = _mm512_subs_epi8(vx, voffset); vy = _mm512_xor_si512(vy, _mm512_shuffle_epi8(vtableF, vx)); _mm512_mask_storeu_epi8(y, vmask, vy); } }