pitch_avx.c (4507B)
1 /* Copyright (c) 2023 Amazon */ 2 /* 3 Redistribution and use in source and binary forms, with or without 4 modification, are permitted provided that the following conditions 5 are met: 6 7 - Redistributions of source code must retain the above copyright 8 notice, this list of conditions and the following disclaimer. 9 10 - Redistributions in binary form must reproduce the above copyright 11 notice, this list of conditions and the following disclaimer in the 12 documentation and/or other materials provided with the distribution. 13 14 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 15 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 16 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 17 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER 18 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 19 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 20 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 21 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 22 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 23 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 24 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27 #ifdef HAVE_CONFIG_H 28 #include "config.h" 29 #endif 30 31 32 #include <immintrin.h> 33 #include "x86cpu.h" 34 #include "pitch.h" 35 36 #if defined(OPUS_X86_MAY_HAVE_AVX2) && !defined(FIXED_POINT) 37 38 /* Like the "regular" xcorr_kernel(), but computes 8 results at a time. */ 39 static void xcorr_kernel_avx(const float *x, const float *y, float sum[8], int len) 40 { 41 __m256 xsum0, xsum1, xsum2, xsum3, xsum4, xsum5, xsum6, xsum7; 42 xsum7 = xsum6 = xsum5 = xsum4 = xsum3 = xsum2 = xsum1 = xsum0 = _mm256_setzero_ps(); 43 int i; 44 __m256 x0; 45 /* Compute 8 inner products using partial sums. */ 46 for (i=0;i<len-7;i+=8) 47 { 48 x0 = _mm256_loadu_ps(x+i); 49 xsum0 = _mm256_fmadd_ps(x0, _mm256_loadu_ps(y+i ), xsum0); 50 xsum1 = _mm256_fmadd_ps(x0, _mm256_loadu_ps(y+i+1), xsum1); 51 xsum2 = _mm256_fmadd_ps(x0, _mm256_loadu_ps(y+i+2), xsum2); 52 xsum3 = _mm256_fmadd_ps(x0, _mm256_loadu_ps(y+i+3), xsum3); 53 xsum4 = _mm256_fmadd_ps(x0, _mm256_loadu_ps(y+i+4), xsum4); 54 xsum5 = _mm256_fmadd_ps(x0, _mm256_loadu_ps(y+i+5), xsum5); 55 xsum6 = _mm256_fmadd_ps(x0, _mm256_loadu_ps(y+i+6), xsum6); 56 xsum7 = _mm256_fmadd_ps(x0, _mm256_loadu_ps(y+i+7), xsum7); 57 } 58 if (i != len) { 59 static const int mask[15] = {-1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 0, 0, 0, 0}; 60 __m256i m; 61 m = _mm256_loadu_si256((__m256i*)(void*)(mask + 7+i-len)); 62 x0 = _mm256_maskload_ps(x+i, m); 63 xsum0 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i , m), xsum0); 64 xsum1 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i+1, m), xsum1); 65 xsum2 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i+2, m), xsum2); 66 xsum3 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i+3, m), xsum3); 67 xsum4 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i+4, m), xsum4); 68 xsum5 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i+5, m), xsum5); 69 xsum6 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i+6, m), xsum6); 70 xsum7 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i+7, m), xsum7); 71 } 72 /* 8 horizontal adds. */ 73 /* Compute [0 4] [1 5] [2 6] [3 7] */ 74 xsum0 = _mm256_add_ps(_mm256_permute2f128_ps(xsum0, xsum4, 2<<4), _mm256_permute2f128_ps(xsum0, xsum4, 1 | (3<<4))); 75 xsum1 = _mm256_add_ps(_mm256_permute2f128_ps(xsum1, xsum5, 2<<4), _mm256_permute2f128_ps(xsum1, xsum5, 1 | (3<<4))); 76 xsum2 = _mm256_add_ps(_mm256_permute2f128_ps(xsum2, xsum6, 2<<4), _mm256_permute2f128_ps(xsum2, xsum6, 1 | (3<<4))); 77 xsum3 = _mm256_add_ps(_mm256_permute2f128_ps(xsum3, xsum7, 2<<4), _mm256_permute2f128_ps(xsum3, xsum7, 1 | (3<<4))); 78 /* Compute [0 1 4 5] [2 3 6 7] */ 79 xsum0 = _mm256_hadd_ps(xsum0, xsum1); 80 xsum1 = _mm256_hadd_ps(xsum2, xsum3); 81 /* Compute [0 1 2 3 4 5 6 7] */ 82 xsum0 = _mm256_hadd_ps(xsum0, xsum1); 83 _mm256_storeu_ps(sum, xsum0); 84 } 85 86 void celt_pitch_xcorr_avx2(const float *_x, const float *_y, float *xcorr, int len, int max_pitch, int arch) 87 { 88 int i; 89 celt_assert(max_pitch>0); 90 (void)arch; 91 for (i=0;i<max_pitch-7;i+=8) 92 { 93 xcorr_kernel_avx(_x, _y+i, &xcorr[i], len); 94 } 95 for (;i<max_pitch;i++) 96 { 97 xcorr[i] = celt_inner_prod(_x, _y+i, len, arch); 98 } 99 } 100 101 #endif