doc/doxygen-3.6/volk__16i__branch__4__state__8__a_8h_source.html

00001 #ifndef INCLUDED_volk_16i_branch_4_state_8_a_H
00002 #define INCLUDED_volk_16i_branch_4_state_8_a_H
00003
00004
00005 #include<inttypes.h>
00006 #include<stdio.h>
00007
00008
00009
00010
00011 #ifdef LV_HAVE_SSSE3
00012
00013 #include<xmmintrin.h>
00014 #include<emmintrin.h>
00015 #include<tmmintrin.h>
00016
00017 static inline  void volk_16i_branch_4_state_8_a_ssse3(short* target,  short* src0, char** permuters, short* cntl2, short* cntl3, short* scalars) {
00018
00019
00020   __m128i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10, xmm11;
00021
00022   __m128i *p_target, *p_src0, *p_cntl2, *p_cntl3, *p_scalars;
00023
00024
00025
00026   p_target = (__m128i*)target;
00027   p_src0 = (__m128i*)src0;
00028   p_cntl2 = (__m128i*)cntl2;
00029   p_cntl3 = (__m128i*)cntl3;
00030   p_scalars = (__m128i*)scalars;
00031
00032   int i = 0;
00033
00034   int bound = 1;
00035
00036
00037   xmm0 = _mm_load_si128(p_scalars);
00038
00039   xmm1 = _mm_shufflelo_epi16(xmm0, 0);
00040   xmm2 = _mm_shufflelo_epi16(xmm0, 0x55);
00041   xmm3 = _mm_shufflelo_epi16(xmm0, 0xaa);
00042   xmm4 = _mm_shufflelo_epi16(xmm0, 0xff);
00043
00044   xmm1 = _mm_shuffle_epi32(xmm1, 0x00);
00045   xmm2 = _mm_shuffle_epi32(xmm2, 0x00);
00046   xmm3 = _mm_shuffle_epi32(xmm3, 0x00);
00047   xmm4 = _mm_shuffle_epi32(xmm4, 0x00);
00048
00049   xmm0 = _mm_load_si128((__m128i*)permuters[0]);
00050   xmm6 = _mm_load_si128((__m128i*)permuters[1]);
00051   xmm8 = _mm_load_si128((__m128i*)permuters[2]);
00052   xmm10 = _mm_load_si128((__m128i*)permuters[3]);
00053
00054   for(; i < bound; ++i) {
00055
00056     xmm5 = _mm_load_si128(p_src0);
00057
00058
00059
00060
00061
00062
00063
00064
00065
00066     xmm0 = _mm_shuffle_epi8(xmm5, xmm0);
00067     xmm6 = _mm_shuffle_epi8(xmm5, xmm6);
00068     xmm8 = _mm_shuffle_epi8(xmm5, xmm8);
00069     xmm10 = _mm_shuffle_epi8(xmm5, xmm10);
00070
00071     p_src0 += 4;
00072
00073
00074     xmm5 = _mm_add_epi16(xmm1, xmm2);
00075
00076     xmm6 = _mm_add_epi16(xmm2, xmm6);
00077     xmm8 = _mm_add_epi16(xmm1, xmm8);
00078
00079
00080     xmm7 = _mm_load_si128(p_cntl2);
00081     xmm9 = _mm_load_si128(p_cntl3);
00082
00083     xmm0 = _mm_add_epi16(xmm5, xmm0);
00084
00085
00086     xmm7 = _mm_and_si128(xmm7, xmm3);
00087     xmm9 = _mm_and_si128(xmm9, xmm4);
00088
00089     xmm5 = _mm_load_si128(&p_cntl2[1]);
00090     xmm11 = _mm_load_si128(&p_cntl3[1]);
00091
00092     xmm7 = _mm_add_epi16(xmm7, xmm9);
00093
00094     xmm5 = _mm_and_si128(xmm5, xmm3);
00095     xmm11 = _mm_and_si128(xmm11, xmm4);
00096
00097     xmm0 = _mm_add_epi16(xmm0, xmm7);
00098
00099
00100
00101     xmm7 = _mm_load_si128(&p_cntl2[2]);
00102     xmm9 = _mm_load_si128(&p_cntl3[2]);
00103
00104     xmm5 = _mm_add_epi16(xmm5, xmm11);
00105
00106     xmm7 = _mm_and_si128(xmm7, xmm3);
00107     xmm9 = _mm_and_si128(xmm9, xmm4);
00108
00109     xmm6 = _mm_add_epi16(xmm6, xmm5);
00110
00111
00112     xmm5 = _mm_load_si128(&p_cntl2[3]);
00113     xmm11 = _mm_load_si128(&p_cntl3[3]);
00114
00115     xmm7 = _mm_add_epi16(xmm7, xmm9);
00116
00117     xmm5 = _mm_and_si128(xmm5, xmm3);
00118     xmm11 = _mm_and_si128(xmm11, xmm4);
00119
00120     xmm8 = _mm_add_epi16(xmm8, xmm7);
00121
00122     xmm5 = _mm_add_epi16(xmm5, xmm11);
00123
00124     _mm_store_si128(p_target, xmm0);
00125     _mm_store_si128(&p_target[1], xmm6);
00126
00127     xmm10 = _mm_add_epi16(xmm5, xmm10);
00128
00129     _mm_store_si128(&p_target[2], xmm8);
00130
00131     _mm_store_si128(&p_target[3], xmm10);
00132
00133     p_target += 3;
00134   }
00135 }
00136
00137
00138 #endif /*LV_HAVE_SSEs*/
00139
00140 #ifdef LV_HAVE_GENERIC
00141 static inline  void volk_16i_branch_4_state_8_a_generic(short* target,  short* src0, char** permuters, short* cntl2, short* cntl3, short* scalars) {
00142         int i = 0;
00143
00144         int bound = 4;
00145
00146         for(; i < bound; ++i) {
00147           target[i* 8] = src0[((char)permuters[i][0])/2]
00148             + ((i + 1)%2  * scalars[0])
00149             + (((i >> 1)^1) * scalars[1])
00150             + (cntl2[i * 8] & scalars[2])
00151             + (cntl3[i * 8] & scalars[3]);
00152           target[i* 8 + 1] = src0[((char)permuters[i][1 * 2])/2]
00153             + ((i + 1)%2  * scalars[0])
00154             + (((i >> 1)^1) * scalars[1])
00155             + (cntl2[i * 8 + 1] & scalars[2])
00156             + (cntl3[i * 8 + 1] & scalars[3]);
00157           target[i* 8 + 2] = src0[((char)permuters[i][2 * 2])/2]
00158             + ((i + 1)%2  * scalars[0])
00159             + (((i >> 1)^1) * scalars[1])
00160             + (cntl2[i * 8 + 2] & scalars[2])
00161             + (cntl3[i * 8 + 2] & scalars[3]);
00162           target[i* 8 + 3] = src0[((char)permuters[i][3 * 2])/2]
00163             + ((i + 1)%2  * scalars[0])
00164             + (((i >> 1)^1) * scalars[1])
00165             + (cntl2[i * 8 + 3] & scalars[2])
00166             + (cntl3[i * 8 + 3] & scalars[3]);
00167           target[i* 8 + 4] = src0[((char)permuters[i][4 * 2])/2]
00168             + ((i + 1)%2  * scalars[0])
00169             + (((i >> 1)^1) * scalars[1])
00170             + (cntl2[i * 8 + 4] & scalars[2])
00171             + (cntl3[i * 8 + 4] & scalars[3]);
00172           target[i* 8 + 5] = src0[((char)permuters[i][5 * 2])/2]
00173             + ((i + 1)%2  * scalars[0])
00174             + (((i >> 1)^1) * scalars[1])
00175             + (cntl2[i * 8 + 5] & scalars[2])
00176             + (cntl3[i * 8 + 5] & scalars[3]);
00177           target[i* 8 + 6] = src0[((char)permuters[i][6 * 2])/2]
00178             + ((i + 1)%2  * scalars[0])
00179             + (((i >> 1)^1) * scalars[1])
00180             + (cntl2[i * 8 + 6] & scalars[2])
00181             + (cntl3[i * 8 + 6] & scalars[3]);
00182           target[i* 8 + 7] = src0[((char)permuters[i][7 * 2])/2]
00183             + ((i + 1)%2  * scalars[0])
00184             + (((i >> 1)^1) * scalars[1])
00185             + (cntl2[i * 8 + 7] & scalars[2])
00186             + (cntl3[i * 8 + 7] & scalars[3]);
00187
00188         }
00189 }
00190
00191 #endif /*LV_HAVE_GENERIC*/
00192
00193
00194 #endif /*INCLUDED_volk_16i_branch_4_state_8_a_H*/