1// Copyright 2020 Google LLC 2// 3// This source code is licensed under the BSD-style license found in the 4// LICENSE file in the root directory of this source tree. 5 6$assert BATCH_TILE % 8 == 0 7$assert BATCH_TILE >= 8 8$ABC = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ" 9#include <assert.h> 10 11#include <immintrin.h> 12 13#include <xnnpack/common.h> 14#include <xnnpack/vunary.h> 15 16 17void xnn_f32_vlrelu_ukernel__avx_x${BATCH_TILE}( 18 size_t n, 19 const float* x, 20 float* y, 21 const union xnn_f32_lrelu_params params[restrict XNN_MIN_ELEMENTS(1)]) 22{ 23 assert(n != 0); 24 assert(n % sizeof(float) == 0); 25 26 const __m256 vslope = _mm256_load_ps(params->avx.slope); 27 for (; n >= ${BATCH_TILE} * sizeof(float); n -= ${BATCH_TILE} * sizeof(float)) { 28 const __m256 vx${ABC[0:8]} = _mm256_loadu_ps(x); 29 $for N in range(8, BATCH_TILE, 8): 30 const __m256 vx${ABC[N:N+8]} = _mm256_loadu_ps(x + ${N}); 31 x += ${BATCH_TILE}; 32 33 $for N in range(0, BATCH_TILE, 8): 34 __m256 vacc${ABC[N:N+8]} = _mm256_mul_ps(vx${ABC[N:N+8]}, vslope); 35 36 $for N in range(0, BATCH_TILE, 8): 37 vacc${ABC[N:N+8]} = _mm256_blendv_ps(vx${ABC[N:N+8]}, vacc${ABC[N:N+8]}, vx${ABC[N:N+8]}); 38 39 _mm256_storeu_ps(y, vacc${ABC[0:8]}); 40 $for N in range(8, BATCH_TILE, 8): 41 _mm256_storeu_ps(y + ${N}, vacc${ABC[N:N+8]}); 42 y += ${BATCH_TILE}; 43 } 44 $if BATCH_TILE > 8: 45 for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) { 46 const __m256 vx = _mm256_loadu_ps(x); 47 x += 8; 48 __m256 vacc = _mm256_mul_ps(vx, vslope); 49 vacc = _mm256_blendv_ps(vx, vacc, vx); 50 _mm256_storeu_ps(y, vacc); 51 y += 8; 52 } 53 if XNN_UNLIKELY(n != 0) { 54 assert(n >= 1 * sizeof(float)); 55 assert(n <= 7 * sizeof(float)); 56 const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) ¶ms->avx.mask_table[7] - n)); 57 58 const __m256 vx = _mm256_maskload_ps(x, vmask); 59 __m256 vacc = _mm256_mul_ps(vx, vslope); 60 vacc = _mm256_blendv_ps(vx, vacc, vx); 61 62 __m128 vacc_lo = _mm256_castps256_ps128(vacc); 63 if (n & (4 * sizeof(float))) { 64 _mm_storeu_ps(y, vacc_lo); 65 vacc_lo = _mm256_extractf128_ps(vacc, 1); 66 y += 4; 67 } 68 if (n & (2 * sizeof(float))) { 69 _mm_storel_pi((__m64*) y, vacc_lo); 70 vacc_lo = _mm_movehl_ps(vacc_lo, vacc_lo); 71 y += 2; 72 } 73 if (n & (1 * sizeof(float))) { 74 _mm_store_ss(y, vacc_lo); 75 } 76 } 77} 78