pitch_sse4_1.c (7815B)
1 /* Copyright (c) 2014, Cisco Systems, INC 2 Written by XiangMingZhu WeiZhou MinPeng YanWang 3 4 Redistribution and use in source and binary forms, with or without 5 modification, are permitted provided that the following conditions 6 are met: 7 8 - Redistributions of source code must retain the above copyright 9 notice, this list of conditions and the following disclaimer. 10 11 - Redistributions in binary form must reproduce the above copyright 12 notice, this list of conditions and the following disclaimer in the 13 documentation and/or other materials provided with the distribution. 14 15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 16 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 17 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 18 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER 19 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 20 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 22 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 23 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 24 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 25 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 #ifdef HAVE_CONFIG_H 29 #include "config.h" 30 #endif 31 32 #include <xmmintrin.h> 33 #include <emmintrin.h> 34 35 #include "macros.h" 36 #include "celt_lpc.h" 37 #include "stack_alloc.h" 38 #include "mathops.h" 39 #include "pitch.h" 40 41 #if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) 42 #include <smmintrin.h> 43 #include "x86cpu.h" 44 45 opus_val32 celt_inner_prod_sse4_1(const opus_val16 *x, const opus_val16 *y, 46 int N) 47 { 48 opus_int i, dataSize16; 49 opus_int32 sum; 50 __m128i inVec1_76543210, inVec1_FEDCBA98, acc1; 51 __m128i inVec2_76543210, inVec2_FEDCBA98, acc2; 52 __m128i inVec1_3210, inVec2_3210; 53 54 sum = 0; 55 dataSize16 = N & ~15; 56 57 acc1 = _mm_setzero_si128(); 58 acc2 = _mm_setzero_si128(); 59 60 for (i=0;i<dataSize16;i+=16) { 61 inVec1_76543210 = _mm_loadu_si128((__m128i *)(void*)(&x[i + 0])); 62 inVec2_76543210 = _mm_loadu_si128((__m128i *)(void*)(&y[i + 0])); 63 64 inVec1_FEDCBA98 = _mm_loadu_si128((__m128i *)(void*)(&x[i + 8])); 65 inVec2_FEDCBA98 = _mm_loadu_si128((__m128i *)(void*)(&y[i + 8])); 66 67 inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); 68 inVec1_FEDCBA98 = _mm_madd_epi16(inVec1_FEDCBA98, inVec2_FEDCBA98); 69 70 acc1 = _mm_add_epi32(acc1, inVec1_76543210); 71 acc2 = _mm_add_epi32(acc2, inVec1_FEDCBA98); 72 } 73 74 acc1 = _mm_add_epi32(acc1, acc2); 75 76 if (N - i >= 8) 77 { 78 inVec1_76543210 = _mm_loadu_si128((__m128i *)(void*)(&x[i + 0])); 79 inVec2_76543210 = _mm_loadu_si128((__m128i *)(void*)(&y[i + 0])); 80 81 inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); 82 83 acc1 = _mm_add_epi32(acc1, inVec1_76543210); 84 i += 8; 85 } 86 87 if (N - i >= 4) 88 { 89 inVec1_3210 = OP_CVTEPI16_EPI32_M64(&x[i + 0]); 90 inVec2_3210 = OP_CVTEPI16_EPI32_M64(&y[i + 0]); 91 92 inVec1_3210 = _mm_mullo_epi32(inVec1_3210, inVec2_3210); 93 94 acc1 = _mm_add_epi32(acc1, inVec1_3210); 95 i += 4; 96 } 97 98 acc1 = _mm_add_epi32(acc1, _mm_unpackhi_epi64(acc1, acc1)); 99 acc1 = _mm_add_epi32(acc1, _mm_shufflelo_epi16(acc1, 0x0E)); 100 101 sum += _mm_cvtsi128_si32(acc1); 102 103 for (;i<N;i++) 104 { 105 sum = silk_SMLABB(sum, x[i], y[i]); 106 } 107 108 return sum; 109 } 110 111 void xcorr_kernel_sse4_1(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[ 4 ], int len) 112 { 113 int j; 114 115 __m128i vecX, vecX0, vecX1, vecX2, vecX3; 116 __m128i vecY0, vecY1, vecY2, vecY3; 117 __m128i sum0, sum1, sum2, sum3, vecSum; 118 __m128i initSum; 119 120 #ifdef OPUS_CHECK_ASM 121 opus_val32 sum_c[4]; 122 for (j=0;j<4;j++) { 123 sum_c[j] = sum[j]; 124 } 125 xcorr_kernel_c(x, y, sum_c, len); 126 #endif 127 128 celt_assert(len >= 3); 129 130 sum0 = _mm_setzero_si128(); 131 sum1 = _mm_setzero_si128(); 132 sum2 = _mm_setzero_si128(); 133 sum3 = _mm_setzero_si128(); 134 135 for (j=0;j<(len-7);j+=8) 136 { 137 vecX = _mm_loadu_si128((__m128i *)(void*)(&x[j + 0])); 138 vecY0 = _mm_loadu_si128((__m128i *)(void*)(&y[j + 0])); 139 vecY1 = _mm_loadu_si128((__m128i *)(void*)(&y[j + 1])); 140 vecY2 = _mm_loadu_si128((__m128i *)(void*)(&y[j + 2])); 141 vecY3 = _mm_loadu_si128((__m128i *)(void*)(&y[j + 3])); 142 143 sum0 = _mm_add_epi32(sum0, _mm_madd_epi16(vecX, vecY0)); 144 sum1 = _mm_add_epi32(sum1, _mm_madd_epi16(vecX, vecY1)); 145 sum2 = _mm_add_epi32(sum2, _mm_madd_epi16(vecX, vecY2)); 146 sum3 = _mm_add_epi32(sum3, _mm_madd_epi16(vecX, vecY3)); 147 } 148 149 sum0 = _mm_add_epi32(sum0, _mm_unpackhi_epi64( sum0, sum0)); 150 sum0 = _mm_add_epi32(sum0, _mm_shufflelo_epi16( sum0, 0x0E)); 151 152 sum1 = _mm_add_epi32(sum1, _mm_unpackhi_epi64( sum1, sum1)); 153 sum1 = _mm_add_epi32(sum1, _mm_shufflelo_epi16( sum1, 0x0E)); 154 155 sum2 = _mm_add_epi32(sum2, _mm_unpackhi_epi64( sum2, sum2)); 156 sum2 = _mm_add_epi32(sum2, _mm_shufflelo_epi16( sum2, 0x0E)); 157 158 sum3 = _mm_add_epi32(sum3, _mm_unpackhi_epi64( sum3, sum3)); 159 sum3 = _mm_add_epi32(sum3, _mm_shufflelo_epi16( sum3, 0x0E)); 160 161 vecSum = _mm_unpacklo_epi64(_mm_unpacklo_epi32(sum0, sum1), 162 _mm_unpacklo_epi32(sum2, sum3)); 163 164 for (;j<(len-3);j+=4) 165 { 166 vecX = OP_CVTEPI16_EPI32_M64(&x[j + 0]); 167 vecX0 = _mm_shuffle_epi32(vecX, 0x00); 168 vecX1 = _mm_shuffle_epi32(vecX, 0x55); 169 vecX2 = _mm_shuffle_epi32(vecX, 0xaa); 170 vecX3 = _mm_shuffle_epi32(vecX, 0xff); 171 172 vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); 173 vecY1 = OP_CVTEPI16_EPI32_M64(&y[j + 1]); 174 vecY2 = OP_CVTEPI16_EPI32_M64(&y[j + 2]); 175 vecY3 = OP_CVTEPI16_EPI32_M64(&y[j + 3]); 176 177 sum0 = _mm_mullo_epi32(vecX0, vecY0); 178 sum1 = _mm_mullo_epi32(vecX1, vecY1); 179 sum2 = _mm_mullo_epi32(vecX2, vecY2); 180 sum3 = _mm_mullo_epi32(vecX3, vecY3); 181 182 sum0 = _mm_add_epi32(sum0, sum1); 183 sum2 = _mm_add_epi32(sum2, sum3); 184 vecSum = _mm_add_epi32(vecSum, sum0); 185 vecSum = _mm_add_epi32(vecSum, sum2); 186 } 187 188 vecX = OP_CVTEPI16_EPI32_M64(&x[len - 4]); 189 if (len - j == 3) 190 { 191 vecX0 = _mm_shuffle_epi32(vecX, 0x55); 192 vecX1 = _mm_shuffle_epi32(vecX, 0xaa); 193 vecX2 = _mm_shuffle_epi32(vecX, 0xff); 194 195 vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); 196 vecY1 = OP_CVTEPI16_EPI32_M64(&y[j + 1]); 197 vecY2 = OP_CVTEPI16_EPI32_M64(&y[j + 2]); 198 199 sum0 = _mm_mullo_epi32(vecX0, vecY0); 200 sum1 = _mm_mullo_epi32(vecX1, vecY1); 201 sum2 = _mm_mullo_epi32(vecX2, vecY2); 202 203 vecSum = _mm_add_epi32(vecSum, sum0); 204 vecSum = _mm_add_epi32(vecSum, sum1); 205 vecSum = _mm_add_epi32(vecSum, sum2); 206 } 207 else if (len - j == 2) 208 { 209 vecX0 = _mm_shuffle_epi32(vecX, 0xaa); 210 vecX1 = _mm_shuffle_epi32(vecX, 0xff); 211 212 vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); 213 vecY1 = OP_CVTEPI16_EPI32_M64(&y[j + 1]); 214 215 sum0 = _mm_mullo_epi32(vecX0, vecY0); 216 sum1 = _mm_mullo_epi32(vecX1, vecY1); 217 218 vecSum = _mm_add_epi32(vecSum, sum0); 219 vecSum = _mm_add_epi32(vecSum, sum1); 220 } 221 else if (len - j == 1) 222 { 223 vecX0 = _mm_shuffle_epi32(vecX, 0xff); 224 225 vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); 226 227 sum0 = _mm_mullo_epi32(vecX0, vecY0); 228 229 vecSum = _mm_add_epi32(vecSum, sum0); 230 } 231 232 initSum = _mm_loadu_si128((__m128i *)(void*)(&sum[0])); 233 initSum = _mm_add_epi32(initSum, vecSum); 234 _mm_storeu_si128((__m128i *)(void*)sum, initSum); 235 236 #ifdef OPUS_CHECK_ASM 237 celt_assert(!memcmp(sum_c, sum, sizeof(sum_c))); 238 #endif 239 } 240 #endif