// Auto-generated file. Do not edit! // Template: src/qs8-igemm/MRx8c8-avxvnni.c.in // Generator: tools/xngen // // Copyright 2024 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 "xnnpack/common.h" #include "xnnpack/gemm.h" #include "xnnpack/intrinsics-polyfill.h" #include "xnnpack/math.h" #include "xnnpack/unaligned.h" #include "xnnpack/prefetch.h" void xnn_qs8_qc8w_igemm_minmax_fp32_ukernel_1x8c8__avx256vnni_prfm( 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_qc8w_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 * sizeof(int8_t)); int8_t* c0 = c; const __m256i vsign_mask = _mm256_set1_epi8(0x80); XNN_FORCE_REALIZATION(vsign_mask); const __m256 voutput_max_less_zero_point = _mm256_set1_ps((int32_t) params->fp32_scalar.output_max - (int32_t) params->fp32_scalar.output_zero_point); const __m256i voutput_zero_point = _mm256_set1_epi32(params->fp32_scalar.output_zero_point); const __m128i voutput_min = _mm_set1_epi8(params->fp32_scalar.output_min); // XNN_FORCE_REALIZATION(voutput_max_less_zero_point); // XNN_FORCE_REALIZATION(voutput_zero_point); // XNN_FORCE_REALIZATION(voutput_min); do { __m256i vacc0x0123 = _mm256_cvtepu32_epi64(_mm_load_si128((const __m128i*) w)); __m256i vacc0x4567 = _mm256_cvtepu32_epi64(_mm_load_si128((const __m128i*) ((const int32_t*) w + 4))); __m256i vacc1x0x0123 = _mm256_setzero_si256(); __m256i vacc1x0x4567 = _mm256_setzero_si256(); w = (const int32_t*) w + 8; 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 = kc; while (k >= 16 * sizeof(int8_t)) { const __m256i va0x01234567 = _mm256_xor_si256(_mm256_set1_epi64x((int64_t) unaligned_load_u64(a0)), vsign_mask); const __m256i va0x89ABCDEF = _mm256_xor_si256(_mm256_set1_epi64x((int64_t) unaligned_load_u64(a0 + 8)), vsign_mask); a0 += 16; const __m256i vb01234567x0123 = _mm256_load_si256(w); const __m256i vb89ABCDEFx0123 = _mm256_load_si256((const __m256i*) ((const int8_t*) w + 32)); const __m256i vb01234567x4567 = _mm256_load_si256((const __m256i*) ((const int8_t*) w + 64)); const __m256i vb89ABCDEFx4567 = _mm256_load_si256((const __m256i*) ((const int8_t*) w + 96)); xnn_prefetch_to_l1((const int8_t*) w + 896); vacc0x0123 = _mm256_dpbusd_epi32(vacc0x0123, va0x01234567, vb01234567x0123); vacc0x4567 = _mm256_dpbusd_epi32(vacc0x4567, va0x01234567, vb89ABCDEFx0123); xnn_prefetch_to_l1((const int8_t*) w + 960); vacc1x0x0123 = _mm256_dpbusd_epi32(vacc1x0x0123, va0x89ABCDEF, vb01234567x4567); vacc1x0x4567 = _mm256_dpbusd_epi32(vacc1x0x4567, va0x89ABCDEF, vb89ABCDEFx4567); w = (const int8_t*) w + 128; k -= 16 * sizeof(int8_t); } if (k != 0) { const __m256i va0x01234567 = _mm256_xor_si256(_mm256_set1_epi64x((int64_t) unaligned_load_u64(a0)), vsign_mask); a0 += 8; const __m256i vb01234567x0123 = _mm256_load_si256(w); const __m256i vb89ABCDEFx0123 = _mm256_load_si256((const __m256i*) ((const int8_t*) w + 32)); vacc0x0123 = _mm256_dpbusd_epi32(vacc0x0123, va0x01234567, vb01234567x0123); vacc0x4567 = _mm256_dpbusd_epi32(vacc0x4567, va0x01234567, vb89ABCDEFx0123); xnn_prefetch_to_l1((const int8_t*) w + 960); w = (const int8_t*) w + 64; k -= 8 * sizeof(int8_t); } p -= 1 * sizeof(void*); } while (p != 0); vacc0x0123 = _mm256_add_epi32(vacc0x0123, vacc1x0x0123); vacc0x4567 = _mm256_add_epi32(vacc0x4567, vacc1x0x4567); // Add adjacent pairs const __m256i vsum0x02134657 = _mm256_hadd_epi32(vacc0x0123, vacc0x4567); __m256i vacc0x01234567 = _mm256_permute4x64_epi64(vsum0x02134657, _MM_SHUFFLE(3, 1, 2, 0)); __m256 vout0x01234567 = _mm256_cvtepi32_ps(vacc0x01234567); const __m256 vscale01234567 = _mm256_load_ps(w); w = (const float*) w + 8; vout0x01234567 = _mm256_mul_ps(vout0x01234567, vscale01234567); vout0x01234567 = _mm256_min_ps(vout0x01234567, voutput_max_less_zero_point); vacc0x01234567 = _mm256_cvtps_epi32(vout0x01234567); vacc0x01234567 = _mm256_add_epi32(vacc0x01234567, voutput_zero_point); __m128i voutb0x01234567 = _mm256_cvtsepi32_epi8(vacc0x01234567); voutb0x01234567 = _mm_max_epi8(voutb0x01234567, voutput_min); if (nc >= 8) { _mm_storel_epi64((__m128i*) c0, voutb0x01234567); c0 = (int8_t*) ((uintptr_t) c0 + cn_stride); a = (const int8_t**restrict) ((uintptr_t) a - ks); nc -= 8; } else { // Prepare mask for valid 8-bit elements (depends on nc). const __mmask16 vmask = _cvtu32_mask16((UINT32_C(1) << nc) - UINT32_C(1)); _mm_mask_storeu_epi8(c0, vmask, voutb0x01234567); nc = 0; } } while (nc != 0); }