lte_dfts.c 1.14 MB
Newer Older
ghaddab's avatar
ghaddab committed
1
/*******************************************************************************
2
    OpenAirInterface
ghaddab's avatar
ghaddab committed
3 4 5 6 7 8 9 10 11 12 13 14 15 16
    Copyright(c) 1999 - 2014 Eurecom

    OpenAirInterface is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation, either version 3 of the License, or
    (at your option) any later version.


    OpenAirInterface is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
17 18
    along with OpenAirInterface.The full GNU General Public License is
   included in this distribution in the file called "COPYING". If not,
ghaddab's avatar
ghaddab committed
19 20 21 22 23
   see <http://www.gnu.org/licenses/>.

  Contact Information
  OpenAirInterface Admin: openair_admin@eurecom.fr
  OpenAirInterface Tech : openair_tech@eurecom.fr
24
  OpenAirInterface Dev  : openair4g-devel@lists.eurecom.fr
25

ghaddab's avatar
ghaddab committed
26
  Address      : Eurecom, Campus SophiaTech, 450 Route des Chappes, CS 50193 - 06904 Biot Sophia Antipolis cedex, FRANCE
ghaddab's avatar
ghaddab committed
27 28

 *******************************************************************************/
29 30 31 32 33 34 35 36 37 38 39 40 41 42
#ifdef USER_MODE
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#endif

#include <stdint.h>

#ifndef MR_MAIN
#include "PHY/defs.h"
#include "PHY/extern.h"
#include "defs.h"
#else
#include "time_meas.h"
43
#include <math.h>
44 45 46 47 48 49 50

#define debug_msg
#define ONE_OVER_SQRT2_Q15 23170

#endif


51
#include "PHY/sse_intrin.h"
52

53 54
#define print_shorts(s,x) printf("%s %d,%d,%d,%d,%d,%d,%d,%d\n",s,(x)[0],(x)[1],(x)[2],(x)[3],(x)[4],(x)[5],(x)[6],(x)[7])
#define print_ints(s,x) printf("%s %d %d %d %d\n",s,(x)[0],(x)[1],(x)[2],(x)[3])
55

56 57 58
#ifdef AVX2
static int16_t conjugatedft2[16] __attribute__((aligned(32))) = {1,1,1,1,1,1,1,1,-1,1,-1,1,-1,1,-1,1,-1,1};
#endif
59

60
static int16_t conjugatedft[8] __attribute__((aligned(16))) = {-1,1,-1,1,-1,1,-1,1} ;
61 62 63

static short reflip[8]  __attribute__((aligned(16))) = {1,-1,1,-1,1,-1,1,-1};

64
#if defined(__x86_64__) || defined(__i386__)
65
static inline void cmac(__m128i a,__m128i b, __m128i *re32, __m128i *im32) __attribute__((always_inline));
66 67
static inline void cmac(__m128i a,__m128i b, __m128i *re32, __m128i *im32)
{
68 69 70 71 72 73 74

  __m128i cmac_tmp,cmac_tmp_re32,cmac_tmp_im32;

  cmac_tmp    = _mm_sign_epi16(b,*(__m128i*)reflip);
  cmac_tmp_re32  = _mm_madd_epi16(a,cmac_tmp);


knopp's avatar
 
knopp committed
75 76 77
  //  cmac_tmp    = _mm_shufflelo_epi16(b,_MM_SHUFFLE(2,3,0,1));
  //  cmac_tmp    = _mm_shufflehi_epi16(cmac_tmp,_MM_SHUFFLE(2,3,0,1));
  cmac_tmp = _mm_shuffle_epi8(b,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
78 79 80 81 82 83
  cmac_tmp_im32  = _mm_madd_epi16(cmac_tmp,a);

  *re32 = _mm_add_epi32(*re32,cmac_tmp_re32);
  *im32 = _mm_add_epi32(*im32,cmac_tmp_im32);
}

84 85 86 87 88 89 90 91 92 93 94 95 96
static inline void cmacc(__m128i a,__m128i b, __m128i *re32, __m128i *im32) __attribute__((always_inline));
static inline void cmacc(__m128i a,__m128i b, __m128i *re32, __m128i *im32)
{

  __m128i cmac_tmp,cmac_tmp_re32,cmac_tmp_im32;


  cmac_tmp_re32  = _mm_madd_epi16(a,b);


  cmac_tmp    = _mm_sign_epi16(b,*(__m128i*)reflip);
  //  cmac_tmp    = _mm_shufflelo_epi16(b,_MM_SHUFFLE(2,3,0,1));
  //  cmac_tmp    = _mm_shufflehi_epi16(cmac_tmp,_MM_SHUFFLE(2,3,0,1));
97
  cmac_tmp = _mm_shuffle_epi8(cmac_tmp,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
98 99 100 101 102 103
  cmac_tmp_im32  = _mm_madd_epi16(cmac_tmp,a);

  *re32 = _mm_add_epi32(*re32,cmac_tmp_re32);
  *im32 = _mm_add_epi32(*im32,cmac_tmp_im32);
}

104 105 106 107 108



static inline void cmult(__m128i a,__m128i b, __m128i *re32, __m128i *im32) __attribute__((always_inline));

109 110
static inline void cmult(__m128i a,__m128i b, __m128i *re32, __m128i *im32)
{
111 112 113 114 115

  register __m128i mmtmpb;

  mmtmpb    = _mm_sign_epi16(b,*(__m128i*)reflip);
  *re32     = _mm_madd_epi16(a,mmtmpb);
knopp's avatar
 
knopp committed
116 117 118
  //  mmtmpb    = _mm_shufflelo_epi16(b,_MM_SHUFFLE(2,3,0,1));
  //  mmtmpb    = _mm_shufflehi_epi16(mmtmpb,_MM_SHUFFLE(2,3,0,1));
  mmtmpb        = _mm_shuffle_epi8(b,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
119 120 121 122 123 124
  *im32  = _mm_madd_epi16(a,mmtmpb);

}

static inline void cmultc(__m128i a,__m128i b, __m128i *re32, __m128i *im32) __attribute__((always_inline));

125 126
static inline void cmultc(__m128i a,__m128i b, __m128i *re32, __m128i *im32)
{
127 128 129 130 131

  register __m128i mmtmpb;

  *re32     = _mm_madd_epi16(a,b);
  mmtmpb    = _mm_sign_epi16(b,*(__m128i*)reflip);
knopp's avatar
 
knopp committed
132
  mmtmpb    = _mm_shuffle_epi8(mmtmpb,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
133 134 135 136 137 138 139
  *im32  = _mm_madd_epi16(a,mmtmpb);

}


static inline __m128i cpack(__m128i xre,__m128i xim) __attribute__((always_inline));

140 141
static inline __m128i cpack(__m128i xre,__m128i xim)
{
142 143 144 145 146 147 148 149 150 151 152 153

  register __m128i cpack_tmp1,cpack_tmp2;

  cpack_tmp1 = _mm_unpacklo_epi32(xre,xim);
  cpack_tmp2 = _mm_unpackhi_epi32(xre,xim);
  return(_mm_packs_epi32(_mm_srai_epi32(cpack_tmp1,15),_mm_srai_epi32(cpack_tmp2,15)));

}


static inline void packed_cmult(__m128i a,__m128i b, __m128i *c) __attribute__((always_inline));

154 155
static inline void packed_cmult(__m128i a,__m128i b, __m128i *c)
{
156 157 158 159 160 161 162 163 164 165

  __m128i cre,cim;
  cmult(a,b,&cre,&cim);
  *c = cpack(cre,cim);

}


static inline void packed_cmultc(__m128i a,__m128i b, __m128i *c) __attribute__((always_inline));

166 167
static inline void packed_cmultc(__m128i a,__m128i b, __m128i *c)
{
168 169 170 171 172 173 174 175 176 177

  __m128i cre,cim;

  cmultc(a,b,&cre,&cim);
  *c = cpack(cre,cim);

}

static inline __m128i packed_cmult2(__m128i a,__m128i b,__m128i b2) __attribute__((always_inline));

178 179 180 181
static inline __m128i packed_cmult2(__m128i a,__m128i b,__m128i b2)
{


182 183 184 185 186 187 188 189 190
  register __m128i cre,cim;

  cre       = _mm_madd_epi16(a,b);
  cim       = _mm_madd_epi16(a,b2);

  return(cpack(cre,cim));

}

191 192 193 194
#elif defined (__arm__)
static inline void cmac(int16x8_t a,int16x8_t b, int32x4_t *re32, int32x4_t *im32) __attribute__((always_inline));
static inline void cmac(int16x8_t a,int16x8_t b, int32x4_t *re32, int32x4_t *im32)
{
195

196 197 198 199 200 201 202 203 204 205 206 207 208 209
  
  int32x4_t ab_re0,ab_re1,ab_im0,ab_im1;
  int16x8_t bflip = vrev32q_s16(b);
  int16x8_t bconj = vmulq_s16(b,*(int16x8_t *)reflip);

  ab_re0 = vmull_s16(((int16x4_t*)&a)[0],((int16x4_t*)&bconj)[0]);
  ab_re1 = vmull_s16(((int16x4_t*)&a)[1],((int16x4_t*)&bconj)[1]);
  ab_im0 = vmull_s16(((int16x4_t*)&a)[0],((int16x4_t*)&bflip)[0]);
  ab_im1 = vmull_s16(((int16x4_t*)&a)[1],((int16x4_t*)&bflip)[1]);
  *re32 = vqaddq_s32(*re32,vcombine_s32(vpadd_s32(((int32x2_t*)&ab_re0)[0],((int32x2_t*)&ab_re0)[1]),
					vpadd_s32(((int32x2_t*)&ab_re1)[0],((int32x2_t*)&ab_re1)[1])));
  *im32 = vqaddq_s32(*im32,vcombine_s32(vpadd_s32(((int32x2_t*)&ab_im0)[0],((int32x2_t*)&ab_im0)[1]),
					vpadd_s32(((int32x2_t*)&ab_im1)[0],((int32x2_t*)&ab_im1)[1])));
}
210

211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317
static inline void cmacc(int16x8_t a,int16x8_t b, int32x4_t *re32, int32x4_t *im32) __attribute__((always_inline));
static inline void cmacc(int16x8_t a,int16x8_t b, int32x4_t *re32, int32x4_t *im32)
{
  int32x4_t ab_re0,ab_re1,ab_im0,ab_im1;
  int16x8_t bconj = vmulq_s16(b,*(int16x8_t *)reflip);
  int16x8_t bflip = vrev32q_s16(bconj);

  ab_re0 = vmull_s16(((int16x4_t*)&a)[0],((int16x4_t*)&b)[0]);
  ab_re1 = vmull_s16(((int16x4_t*)&a)[1],((int16x4_t*)&b)[1]);
  ab_im0 = vmull_s16(((int16x4_t*)&a)[0],((int16x4_t*)&bflip)[0]);
  ab_im1 = vmull_s16(((int16x4_t*)&a)[1],((int16x4_t*)&bflip)[1]);
  *re32 = vqaddq_s32(*re32,vcombine_s32(vpadd_s32(((int32x2_t*)&ab_re0)[0],((int32x2_t*)&ab_re0)[1]),
					vpadd_s32(((int32x2_t*)&ab_re1)[0],((int32x2_t*)&ab_re1)[1])));
  *im32 = vqaddq_s32(*im32,vcombine_s32(vpadd_s32(((int32x2_t*)&ab_im0)[0],((int32x2_t*)&ab_im0)[1]),
					vpadd_s32(((int32x2_t*)&ab_im1)[0],((int32x2_t*)&ab_im1)[1])));

}

static inline void cmult(int16x8_t a,int16x8_t b, int32x4_t *re32, int32x4_t *im32) __attribute__((always_inline));
static inline void cmult(int16x8_t a,int16x8_t b, int32x4_t *re32, int32x4_t *im32)
{
  int32x4_t ab_re0,ab_re1,ab_im0,ab_im1;
  int16x8_t bflip = vrev32q_s16(b);
  int16x8_t bconj = vmulq_s16(b,*(int16x8_t *)reflip);
  int16x4_t al,ah,bcl,bch,bfl,bfh;
  int32x2_t abr0l,abr0h,abr1l,abr1h,abi0l,abi0h,abi1l,abi1h;

  al  = vget_low_s16(a);      ah = vget_high_s16(a);
  bcl = vget_low_s16(bconj);  bch = vget_high_s16(bconj);
  bfl = vget_low_s16(bflip);  bfh = vget_high_s16(bflip);

  ab_re0 = vmull_s16(al,bcl);
  ab_re1 = vmull_s16(ah,bch);
  ab_im0 = vmull_s16(al,bfl);
  ab_im1 = vmull_s16(ah,bfh);
  abr0l = vget_low_s32(ab_re0); abr0h = vget_high_s32(ab_re0);
  abr1l = vget_low_s32(ab_re1); abr1h = vget_high_s32(ab_re1);
  abi0l = vget_low_s32(ab_im0); abi0h = vget_high_s32(ab_im0);
  abi1l = vget_low_s32(ab_im1); abi1h = vget_high_s32(ab_im1);

  *re32 = vcombine_s32(vpadd_s32(abr0l,abr0h),
                       vpadd_s32(abr1l,abr1h));
  *im32 = vcombine_s32(vpadd_s32(abi0l,abi0h),
                       vpadd_s32(abi1l,abi1h));
}

static inline void cmultc(int16x8_t a,int16x8_t b, int32x4_t *re32, int32x4_t *im32) __attribute__((always_inline));

static inline void cmultc(int16x8_t a,int16x8_t b, int32x4_t *re32, int32x4_t *im32)
{
  int32x4_t ab_re0,ab_re1,ab_im0,ab_im1;
  int16x8_t bconj = vmulq_s16(b,*(int16x8_t *)reflip);
  int16x8_t bflip = vrev32q_s16(bconj);
  int16x4_t al,ah,bl,bh,bfl,bfh; 
  int32x2_t abr0l,abr0h,abr1l,abr1h,abi0l,abi0h,abi1l,abi1h;
  al  = vget_low_s16(a);     ah = vget_high_s16(a);
  bl  = vget_low_s16(b);     bh = vget_high_s16(b);
  bfl = vget_low_s16(bflip); bfh = vget_high_s16(bflip);

  ab_re0 = vmull_s16(al,bl);
  ab_re1 = vmull_s16(ah,bh);
  ab_im0 = vmull_s16(al,bfl);
  ab_im1 = vmull_s16(ah,bfh);

  abr0l = vget_low_s32(ab_re0); abr0h = vget_high_s32(ab_re0);
  abr1l = vget_low_s32(ab_re1); abr1h = vget_high_s32(ab_re1);
  abi0l = vget_low_s32(ab_im0); abi0h = vget_high_s32(ab_im0);
  abi1l = vget_low_s32(ab_im1); abi1h = vget_high_s32(ab_im1);

  *re32 = vcombine_s32(vpadd_s32(abr0l,abr0h),
		       vpadd_s32(abr1l,abr1h));
  *im32 = vcombine_s32(vpadd_s32(abi0l,abi0h),
		       vpadd_s32(abi1l,abi1h));

}


static inline int16x8_t cpack(int32x4_t xre,int32x4_t xim) __attribute__((always_inline));

static inline int16x8_t cpack(int32x4_t xre,int32x4_t xim)
{
  int32x4x2_t xtmp;

  xtmp = vzipq_s32(xre,xim);
  return(vcombine_s16(vqshrn_n_s32(xtmp.val[0],15),vqshrn_n_s32(xtmp.val[1],15)));

}


static inline void packed_cmult(int16x8_t a,int16x8_t b, int16x8_t *c) __attribute__((always_inline));

static inline void packed_cmult(int16x8_t a,int16x8_t b, int16x8_t *c)
{

  int32x4_t cre,cim;
  cmult(a,b,&cre,&cim);
  *c = cpack(cre,cim);

}


static inline void packed_cmultc(int16x8_t a,int16x8_t b, int16x8_t *c) __attribute__((always_inline));

static inline void packed_cmultc(int16x8_t a,int16x8_t b, int16x8_t *c)
{

  int32x4_t cre,cim;
318

319
  cmultc(a,b,&cre,&cim);
320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
  *c = cpack(cre,cim);

}

static inline int16x8_t packed_cmult2(int16x8_t a,int16x8_t b,  int16x8_t b2) __attribute__((always_inline));

static inline int16x8_t packed_cmult2(int16x8_t a,int16x8_t b,  int16x8_t b2)
{

  

  int32x4_t ab_re0,ab_re1,ab_im0,ab_im1,cre,cim;
  
  ab_re0 = vmull_s16(((int16x4_t*)&a)[0],((int16x4_t*)&b)[0]);
  ab_re1 = vmull_s16(((int16x4_t*)&a)[1],((int16x4_t*)&b)[1]);
  ab_im0 = vmull_s16(((int16x4_t*)&a)[0],((int16x4_t*)&b2)[0]);
  ab_im1 = vmull_s16(((int16x4_t*)&a)[1],((int16x4_t*)&b2)[1]);
  cre = vcombine_s32(vpadd_s32(((int32x2_t*)&ab_re0)[0],((int32x2_t*)&ab_re0)[1]),
		     vpadd_s32(((int32x2_t*)&ab_re1)[0],((int32x2_t*)&ab_re1)[1]));
  cim = vcombine_s32(vpadd_s32(((int32x2_t*)&ab_im0)[0],((int32x2_t*)&ab_im0)[1]),
		     vpadd_s32(((int32x2_t*)&ab_im1)[0],((int32x2_t*)&ab_im1)[1]));
341 342 343
  return(cpack(cre,cim));

}
344 345

#endif
346 347 348 349 350 351 352 353 354 355 356

static int16_t W0s[8]__attribute__((aligned(16))) = {32767,0,32767,0,32767,0,32767,0};

static int16_t W13s[8]__attribute__((aligned(16))) = {-16384,-28378,-16384,-28378,-16384,-28378,-16384,-28378};
static int16_t W23s[8]__attribute__((aligned(16))) = {-16384,28378,-16384,28378,-16384,28378,-16384,28378};

static int16_t W15s[8]__attribute__((aligned(16))) = {10126,-31163,10126,-31163,10126,-31163,10126,-31163};
static int16_t W25s[8]__attribute__((aligned(16))) = {-26509,-19260,-26509,-19260,-26509,-19260,-26509,-19260};
static int16_t W35s[8]__attribute__((aligned(16))) = {-26510,19260,-26510,19260,-26510,19260,-26510,19260};
static int16_t W45s[8]__attribute__((aligned(16))) = {10126,31163,10126,31163,10126,31163,10126,31163};

357
#if defined(__x86_64__) || defined(__i386__)
358 359 360 361 362 363 364
__m128i *W0 = (__m128i *)W0s;
__m128i *W13 = (__m128i *)W13s;
__m128i *W23 = (__m128i *)W23s;
__m128i *W15 = (__m128i *)W15s;
__m128i *W25 = (__m128i *)W25s;
__m128i *W35 = (__m128i *)W35s;
__m128i *W45 = (__m128i *)W45s;
365 366 367 368 369 370 371 372 373
#elif defined(__arm__)
int16x8_t *W0  = (int16x8_t *)W0s;
int16x8_t *W13 = (int16x8_t *)W13s;
int16x8_t *W23 = (int16x8_t *)W23s;
int16x8_t *W15 = (int16x8_t *)W15s;
int16x8_t *W25 = (int16x8_t *)W25s;
int16x8_t *W35 = (int16x8_t *)W35s;
int16x8_t *W45 = (int16x8_t *)W45s;
#endif
374
static int16_t dft_norm_table[16] = {9459,  //12
375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390
                                     6689,//24
                                     5461,//36
                                     4729,//48
                                     4230,//60
                                     23170,//72
                                     3344,//96
                                     3153,//108
                                     2991,//120
                                     18918,//sqrt(3),//144
                                     18918,//sqrt(3),//180
                                     16384,//2, //192
                                     18918,//sqrt(3), // 216
                                     16384,//2, //240
                                     18918,//sqrt(3), // 288
                                     14654
                                    }; //sqrt(5) //300
391 392


393
#if defined(__x86_64__) || defined(__i386__)
394 395
static inline void bfly2(__m128i *x0, __m128i *x1,__m128i *y0, __m128i *y1,__m128i *tw)__attribute__((always_inline));

396 397
static inline void bfly2(__m128i *x0, __m128i *x1,__m128i *y0, __m128i *y1,__m128i *tw)
{
398 399 400 401 402 403 404 405 406 407 408 409

  __m128i x0r_2,x0i_2,x1r_2,x1i_2,dy0r,dy1r,dy0i,dy1i;
  __m128i bfly2_tmp1,bfly2_tmp2;

  cmult(*(x0),*(W0),&x0r_2,&x0i_2);
  cmult(*(x1),*(tw),&x1r_2,&x1i_2);

  dy0r = _mm_srai_epi32(_mm_add_epi32(x0r_2,x1r_2),15);
  dy1r = _mm_srai_epi32(_mm_sub_epi32(x0r_2,x1r_2),15);
  dy0i = _mm_srai_epi32(_mm_add_epi32(x0i_2,x1i_2),15);
  //  printf("y0i %d\n",((int16_t *)y0i)[0]);
  dy1i = _mm_srai_epi32(_mm_sub_epi32(x0i_2,x1i_2),15);
410

411 412 413 414 415 416 417 418 419
  bfly2_tmp1 = _mm_unpacklo_epi32(dy0r,dy0i);
  bfly2_tmp2 = _mm_unpackhi_epi32(dy0r,dy0i);
  *y0 = _mm_packs_epi32(bfly2_tmp1,bfly2_tmp2);

  bfly2_tmp1 = _mm_unpacklo_epi32(dy1r,dy1i);
  bfly2_tmp2 = _mm_unpackhi_epi32(dy1r,dy1i);
  *y1 = _mm_packs_epi32(bfly2_tmp1,bfly2_tmp2);
}

420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444
#elif defined(__arm__)

static inline void bfly2(int16x8_t *x0, int16x8_t *x1,int16x8_t *y0, int16x8_t *y1,int16x8_t *tw)__attribute__((always_inline));

static inline void bfly2(int16x8_t *x0, int16x8_t *x1,int16x8_t *y0, int16x8_t *y1,int16x8_t *tw)
{

  int32x4_t x0r_2,x0i_2,x1r_2,x1i_2,dy0r,dy1r,dy0i,dy1i;

  cmult(*(x0),*(W0),&x0r_2,&x0i_2);
  cmult(*(x1),*(tw),&x1r_2,&x1i_2);

  dy0r = vqaddq_s32(x0r_2,x1r_2);
  dy1r = vqsubq_s32(x0r_2,x1r_2);
  dy0i = vqaddq_s32(x0i_2,x1i_2);
  dy1i = vqsubq_s32(x0i_2,x1i_2);

  *y0 = cpack(dy0r,dy0i);
  *y1 = cpack(dy1r,dy1i);
}


#endif

#if defined(__x86_64__) || defined(__i386__)
445 446
static inline void bfly2_tw1(__m128i *x0, __m128i *x1, __m128i *y0, __m128i *y1)__attribute__((always_inline));

447 448
static inline void bfly2_tw1(__m128i *x0, __m128i *x1, __m128i *y0, __m128i *y1)
{
449 450 451 452 453 454

  *y0  = _mm_adds_epi16(*x0,*x1);
  *y1  = _mm_subs_epi16(*x0,*x1);

}

455 456 457 458 459 460 461 462 463 464 465 466 467 468
#elif defined(__arm__)

static inline void bfly2_tw1(int16x8_t *x0, int16x8_t *x1, int16x8_t *y0, int16x8_t *y1)__attribute__((always_inline));

static inline void bfly2_tw1(int16x8_t *x0, int16x8_t *x1, int16x8_t *y0, int16x8_t *y1)
{

  *y0  = vqaddq_s16(*x0,*x1);
  *y1  = vqsubq_s16(*x0,*x1);

}
#endif

#if defined(__x86_64__) || defined(__i386__)
469 470
static inline void bfly2_16(__m128i *x0, __m128i *x1, __m128i *y0, __m128i *y1, __m128i *tw, __m128i *twb)__attribute__((always_inline));

471 472
static inline void bfly2_16(__m128i *x0, __m128i *x1, __m128i *y0, __m128i *y1, __m128i *tw, __m128i *twb)
{
473 474 475 476

  register __m128i x1t;

  x1t = packed_cmult2(*(x1),*(tw),*(twb));
477

478 479 480 481 482 483
  *y0  = _mm_adds_epi16(*x0,x1t);
  *y1  = _mm_subs_epi16(*x0,x1t);

}


484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502
#elif defined(__arm__)

static inline void bfly2_16(int16x8_t *x0, int16x8_t *x1, int16x8_t *y0, int16x8_t *y1, int16x8_t *tw, int16x8_t *twb)__attribute__((always_inline));

static inline void bfly2_16(int16x8_t *x0, int16x8_t *x1, int16x8_t *y0, int16x8_t *y1, int16x8_t *tw, int16x8_t *twb)
{

  register int16x8_t x1t;

  x1t = packed_cmult2(*(x1),*(tw),*(twb));

  *y0  = vqaddq_s16(*x0,x1t);
  *y1  = vqsubq_s16(*x0,x1t);

}

#endif

#if defined(__x86_64__) || defined(__i386__)
503 504
static inline void ibfly2(__m128i *x0, __m128i *x1,__m128i *y0, __m128i *y1,__m128i *tw)__attribute__((always_inline));

505 506
static inline void ibfly2(__m128i *x0, __m128i *x1,__m128i *y0, __m128i *y1,__m128i *tw)
{
507 508 509 510 511 512 513 514 515 516 517 518

  __m128i x0r_2,x0i_2,x1r_2,x1i_2,dy0r,dy1r,dy0i,dy1i;
  __m128i bfly2_tmp1,bfly2_tmp2;

  cmultc(*(x0),*(W0),&x0r_2,&x0i_2);
  cmultc(*(x1),*(tw),&x1r_2,&x1i_2);

  dy0r = _mm_srai_epi32(_mm_add_epi32(x0r_2,x1r_2),15);
  dy1r = _mm_srai_epi32(_mm_sub_epi32(x0r_2,x1r_2),15);
  dy0i = _mm_srai_epi32(_mm_add_epi32(x0i_2,x1i_2),15);
  //  printf("y0i %d\n",((int16_t *)y0i)[0]);
  dy1i = _mm_srai_epi32(_mm_sub_epi32(x0i_2,x1i_2),15);
519

520 521 522 523 524 525 526 527
  bfly2_tmp1 = _mm_unpacklo_epi32(dy0r,dy0i);
  bfly2_tmp2 = _mm_unpackhi_epi32(dy0r,dy0i);
  *y0 = _mm_packs_epi32(bfly2_tmp1,bfly2_tmp2);

  bfly2_tmp1 = _mm_unpacklo_epi32(dy1r,dy1i);
  bfly2_tmp2 = _mm_unpackhi_epi32(dy1r,dy1i);
  *y1 = _mm_packs_epi32(bfly2_tmp1,bfly2_tmp2);
}
528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548
#elif defined(__arm__)
static inline void ibfly2(int16x8_t *x0, int16x8_t *x1,int16x8_t *y0, int16x8_t *y1,int16x8_t *tw)
{

  int32x4_t x0r_2,x0i_2,x1r_2,x1i_2,dy0r,dy1r,dy0i,dy1i;

  cmultc(*(x0),*(W0),&x0r_2,&x0i_2);
  cmultc(*(x1),*(tw),&x1r_2,&x1i_2);

  dy0r = vqaddq_s32(x0r_2,x1r_2);
  dy1r = vqsubq_s32(x0r_2,x1r_2);
  dy0i = vqaddq_s32(x0i_2,x1i_2);
  dy1i = vqsubq_s32(x0i_2,x1i_2);

  *y0 = cpack(dy0r,dy0i);
  *y1 = cpack(dy1r,dy1i);

}

#endif

549 550 551 552



// This is the radix-3 butterfly (fft)
553 554 555

#if defined(__x86_64__) || defined(__i386__)

556
static inline void bfly3(__m128i *x0,__m128i *x1,__m128i *x2,
557 558
                         __m128i *y0,__m128i *y1,__m128i *y2,
                         __m128i *tw1,__m128i *tw2) __attribute__((always_inline));
559 560

static inline void bfly3(__m128i *x0,__m128i *x1,__m128i *x2,
561 562 563
                         __m128i *y0,__m128i *y1,__m128i *y2,
                         __m128i *tw1,__m128i *tw2)
{
564 565 566

  __m128i tmpre,tmpim,x1_2,x2_2;

567 568 569 570 571 572
  packed_cmult(*(x1),*(tw1),&x1_2);
  packed_cmult(*(x2),*(tw2),&x2_2);
  *(y0)  = _mm_adds_epi16(*(x0),_mm_adds_epi16(x1_2,x2_2));
  cmult(x1_2,*(W13),&tmpre,&tmpim);
  cmac(x2_2,*(W23),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
573
  *(y1) = _mm_adds_epi16(*(x0),*(y1));
574 575 576
  cmult(x1_2,*(W23),&tmpre,&tmpim);
  cmac(x2_2,*(W13),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
577 578 579
  *(y2) = _mm_adds_epi16(*(x0),*(y2));
}

580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608
#elif defined(__arm__)
static inline void bfly3(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,
                         int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,
                         int16x8_t *tw1,int16x8_t *tw2) __attribute__((always_inline));

static inline void bfly3(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,
                         int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,
                         int16x8_t *tw1,int16x8_t *tw2)
{

  int32x4_t tmpre,tmpim;
  int16x8_t x1_2,x2_2;

  packed_cmult(*(x1),*(tw1),&x1_2);
  packed_cmult(*(x2),*(tw2),&x2_2);
  *(y0)  = vqaddq_s16(*(x0),vqaddq_s16(x1_2,x2_2));
  cmult(x1_2,*(W13),&tmpre,&tmpim);
  cmac(x2_2,*(W23),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
  *(y1) = vqaddq_s16(*(x0),*(y1));
  cmult(x1_2,*(W23),&tmpre,&tmpim);
  cmac(x2_2,*(W13),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
  *(y2) = vqaddq_s16(*(x0),*(y2));
}

#endif

#if defined(__x86_64__) || defined(__i386__)
609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632
static inline void ibfly3(__m128i *x0,__m128i *x1,__m128i *x2,
			  __m128i *y0,__m128i *y1,__m128i *y2,
			  __m128i *tw1,__m128i *tw2) __attribute__((always_inline));

static inline void ibfly3(__m128i *x0,__m128i *x1,__m128i *x2,
			  __m128i *y0,__m128i *y1,__m128i *y2,
			  __m128i *tw1,__m128i *tw2)
{

  __m128i tmpre,tmpim,x1_2,x2_2;

  packed_cmultc(*(x1),*(tw1),&x1_2);
  packed_cmultc(*(x2),*(tw2),&x2_2);
  *(y0)  = _mm_adds_epi16(*(x0),_mm_adds_epi16(x1_2,x2_2));
  cmultc(x1_2,*(W13),&tmpre,&tmpim);
  cmacc(x2_2,*(W23),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
  *(y1) = _mm_adds_epi16(*(x0),*(y1));
  cmultc(x1_2,*(W23),&tmpre,&tmpim);
  cmacc(x2_2,*(W13),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
  *(y2) = _mm_adds_epi16(*(x0),*(y2));
}

633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660
#elif defined(__arm__)
static inline void ibfly3(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,
			  int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,
			  int16x8_t *tw1,int16x8_t *tw2) __attribute__((always_inline));

static inline void ibfly3(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,
			  int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,
			  int16x8_t *tw1,int16x8_t *tw2)
{

  int32x4_t tmpre,tmpim;
  int16x8_t x1_2,x2_2;

  packed_cmultc(*(x1),*(tw1),&x1_2);
  packed_cmultc(*(x2),*(tw2),&x2_2);
  *(y0)  = vqaddq_s16(*(x0),vqaddq_s16(x1_2,x2_2));
  cmultc(x1_2,*(W13),&tmpre,&tmpim);
  cmacc(x2_2,*(W23),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
  *(y1) = vqaddq_s16(*(x0),*(y1));
  cmultc(x1_2,*(W23),&tmpre,&tmpim);
  cmacc(x2_2,*(W13),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
  *(y2) = vqaddq_s16(*(x0),*(y2));
}
#endif

#if defined(__x86_64__) || defined(__i386__)
661
static inline void bfly3_tw1(__m128i *x0,__m128i *x1,__m128i *x2,
662
                             __m128i *y0,__m128i *y1,__m128i *y2) __attribute__((always_inline));
663 664

static inline void bfly3_tw1(__m128i *x0,__m128i *x1,__m128i *x2,
665 666
                             __m128i *y0,__m128i *y1,__m128i *y2)
{
667 668 669

  __m128i tmpre,tmpim;

670 671 672 673
  *(y0) = _mm_adds_epi16(*(x0),_mm_adds_epi16(*(x1),*(x2)));
  cmult(*(x1),*(W13),&tmpre,&tmpim);
  cmac(*(x2),*(W23),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
674
  *(y1) = _mm_adds_epi16(*(x0),*(y1));
675 676 677
  cmult(*(x1),*(W23),&tmpre,&tmpim);
  cmac(*(x2),*(W13),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
678 679
  *(y2) = _mm_adds_epi16(*(x0),*(y2));
}
680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700
#elif defined(__arm__)
static inline void bfly3_tw1(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,
                             int16x8_t *y0,int16x8_t *y1,int16x8_t *y2) __attribute__((always_inline));

static inline void bfly3_tw1(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,
                             int16x8_t *y0,int16x8_t *y1,int16x8_t *y2)
{

  int32x4_t tmpre,tmpim;

  *(y0) = vqaddq_s16(*(x0),vqaddq_s16(*(x1),*(x2)));
  cmult(*(x1),*(W13),&tmpre,&tmpim);
  cmac(*(x2),*(W23),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
  *(y1) = vqaddq_s16(*(x0),*(y1));
  cmult(*(x1),*(W23),&tmpre,&tmpim);
  cmac(*(x2),*(W13),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
  *(y2) = vqaddq_s16(*(x0),*(y2));

}
701

702
#endif
703

704
#if defined(__x86_64__) || defined(__i386__)
705
static inline void bfly4(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
706 707
                         __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3,
                         __m128i *tw1,__m128i *tw2,__m128i *tw3)__attribute__((always_inline));
708 709

static inline void bfly4(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
710 711 712
                         __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3,
                         __m128i *tw1,__m128i *tw2,__m128i *tw3)
{
713 714 715

  __m128i x1r_2,x1i_2,x2r_2,x2i_2,x3r_2,x3i_2,dy0r,dy0i,dy1r,dy1i,dy2r,dy2i,dy3r,dy3i;

716
  //  cmult(*(x0),*(W0),&x0r_2,&x0i_2);
717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745
  cmult(*(x1),*(tw1),&x1r_2,&x1i_2);
  cmult(*(x2),*(tw2),&x2r_2,&x2i_2);
  cmult(*(x3),*(tw3),&x3r_2,&x3i_2);
  //  dy0r = _mm_add_epi32(x0r_2,_mm_add_epi32(x1r_2,_mm_add_epi32(x2r_2,x3r_2)));
  //  dy0i = _mm_add_epi32(x0i_2,_mm_add_epi32(x1i_2,_mm_add_epi32(x2i_2,x3i_2)));
  //  *(y0)  = cpack(dy0r,dy0i);
  dy0r = _mm_add_epi32(x1r_2,_mm_add_epi32(x2r_2,x3r_2));
  dy0i = _mm_add_epi32(x1i_2,_mm_add_epi32(x2i_2,x3i_2));
  *(y0)  = _mm_add_epi16(*(x0),cpack(dy0r,dy0i));
  //  dy1r = _mm_add_epi32(x0r_2,_mm_sub_epi32(x1i_2,_mm_add_epi32(x2r_2,x3i_2)));
  //  dy1i = _mm_sub_epi32(x0i_2,_mm_add_epi32(x1r_2,_mm_sub_epi32(x2i_2,x3r_2)));
  //  *(y1)  = cpack(dy1r,dy1i);
  dy1r = _mm_sub_epi32(x1i_2,_mm_add_epi32(x2r_2,x3i_2));
  dy1i = _mm_sub_epi32(_mm_sub_epi32(x3r_2,x2i_2),x1r_2);
  *(y1)  = _mm_add_epi16(*(x0),cpack(dy1r,dy1i));
  //  dy2r = _mm_sub_epi32(x0r_2,_mm_sub_epi32(x1r_2,_mm_sub_epi32(x2r_2,x3r_2)));
  //  dy2i = _mm_sub_epi32(x0i_2,_mm_sub_epi32(x1i_2,_mm_sub_epi32(x2i_2,x3i_2)));
  //  *(y2)  = cpack(dy2r,dy2i);
  dy2r = _mm_sub_epi32(_mm_sub_epi32(x2r_2,x3r_2),x1r_2);
  dy2i = _mm_sub_epi32(_mm_sub_epi32(x2i_2,x3i_2),x1i_2);
  *(y2)  = _mm_add_epi16(*(x0),cpack(dy2r,dy2i));
  //  dy3r = _mm_sub_epi32(x0r_2,_mm_add_epi32(x1i_2,_mm_sub_epi32(x2r_2,x3i_2)));
  //  dy3i = _mm_add_epi32(x0i_2,_mm_sub_epi32(x1r_2,_mm_add_epi32(x2i_2,x3r_2)));
  //  *(y3) = cpack(dy3r,dy3i);
  dy3r = _mm_sub_epi32(_mm_sub_epi32(x3i_2,x2r_2),x1i_2);
  dy3i = _mm_sub_epi32(x1r_2,_mm_add_epi32(x2i_2,x3r_2));
  *(y3) = _mm_add_epi16(*(x0),cpack(dy3r,dy3i));
}

746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790
#elif defined(__arm__)
static inline void bfly4(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
                         int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3,
                         int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3)__attribute__((always_inline));

static inline void bfly4(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
                         int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3,
                         int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3)
{

  int32x4_t x1r_2,x1i_2,x2r_2,x2i_2,x3r_2,x3i_2,dy0r,dy0i,dy1r,dy1i,dy2r,dy2i,dy3r,dy3i;

  //  cmult(*(x0),*(W0),&x0r_2,&x0i_2);
  cmult(*(x1),*(tw1),&x1r_2,&x1i_2);
  cmult(*(x2),*(tw2),&x2r_2,&x2i_2);
  cmult(*(x3),*(tw3),&x3r_2,&x3i_2);
  //  dy0r = _mm_add_epi32(x0r_2,_mm_add_epi32(x1r_2,_mm_add_epi32(x2r_2,x3r_2)));
  //  dy0i = _mm_add_epi32(x0i_2,_mm_add_epi32(x1i_2,_mm_add_epi32(x2i_2,x3i_2)));
  //  *(y0)  = cpack(dy0r,dy0i);
  dy0r = vqaddq_s32(x1r_2,vqaddq_s32(x2r_2,x3r_2));
  dy0i = vqaddq_s32(x1i_2,vqaddq_s32(x2i_2,x3i_2));
  *(y0)  = vqaddq_s16(*(x0),cpack(dy0r,dy0i));
  //  dy1r = _mm_add_epi32(x0r_2,_mm_sub_epi32(x1i_2,_mm_add_epi32(x2r_2,x3i_2)));
  //  dy1i = _mm_sub_epi32(x0i_2,_mm_add_epi32(x1r_2,_mm_sub_epi32(x2i_2,x3r_2)));
  //  *(y1)  = cpack(dy1r,dy1i);
  dy1r = vqsubq_s32(x1i_2,vqaddq_s32(x2r_2,x3i_2));
  dy1i = vqsubq_s32(vqsubq_s32(x3r_2,x2i_2),x1r_2);
  *(y1)  = vqaddq_s16(*(x0),cpack(dy1r,dy1i));
  //  dy2r = _mm_sub_epi32(x0r_2,_mm_sub_epi32(x1r_2,_mm_sub_epi32(x2r_2,x3r_2)));
  //  dy2i = _mm_sub_epi32(x0i_2,_mm_sub_epi32(x1i_2,_mm_sub_epi32(x2i_2,x3i_2)));
  //  *(y2)  = cpack(dy2r,dy2i);
  dy2r = vqsubq_s32(vqsubq_s32(x2r_2,x3r_2),x1r_2);
  dy2i = vqsubq_s32(vqsubq_s32(x2i_2,x3i_2),x1i_2);
  *(y2)  = vqaddq_s16(*(x0),cpack(dy2r,dy2i));
  //  dy3r = _mm_sub_epi32(x0r_2,_mm_add_epi32(x1i_2,_mm_sub_epi32(x2r_2,x3i_2)));
  //  dy3i = _mm_add_epi32(x0i_2,_mm_sub_epi32(x1r_2,_mm_add_epi32(x2i_2,x3r_2)));
  //  *(y3) = cpack(dy3r,dy3i);
  dy3r = vqsubq_s32(vqsubq_s32(x3i_2,x2r_2),x1i_2);
  dy3i = vqsubq_s32(x1r_2,vqaddq_s32(x2i_2,x3r_2));
  *(y3) = vqaddq_s16(*(x0),cpack(dy3r,dy3i));
}

#endif

#if defined(__x86_64__) || defined(__i386__)
791
static inline void ibfly4(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
792 793 794
                          __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3,
                          __m128i *tw1,__m128i *tw2,__m128i *tw3)__attribute__((always_inline));

795
static inline void ibfly4(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
796 797 798
                          __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3,
                          __m128i *tw1,__m128i *tw2,__m128i *tw3)
{
799 800 801

  __m128i x1r_2,x1i_2,x2r_2,x2i_2,x3r_2,x3i_2,dy0r,dy0i,dy1r,dy1i,dy2r,dy2i,dy3r,dy3i;

802

803 804 805
  cmultc(*(x1),*(tw1),&x1r_2,&x1i_2);
  cmultc(*(x2),*(tw2),&x2r_2,&x2i_2);
  cmultc(*(x3),*(tw3),&x3r_2,&x3i_2);
806

807 808 809 810 811 812 813 814 815 816 817 818 819 820
  dy0r = _mm_add_epi32(x1r_2,_mm_add_epi32(x2r_2,x3r_2));
  dy0i = _mm_add_epi32(x1i_2,_mm_add_epi32(x2i_2,x3i_2));
  *(y0)  = _mm_add_epi16(*(x0),cpack(dy0r,dy0i));
  dy3r = _mm_sub_epi32(x1i_2,_mm_add_epi32(x2r_2,x3i_2));
  dy3i = _mm_sub_epi32(_mm_sub_epi32(x3r_2,x2i_2),x1r_2);
  *(y3)  = _mm_add_epi16(*(x0),cpack(dy3r,dy3i));
  dy2r = _mm_sub_epi32(_mm_sub_epi32(x2r_2,x3r_2),x1r_2);
  dy2i = _mm_sub_epi32(_mm_sub_epi32(x2i_2,x3i_2),x1i_2);
  *(y2)  = _mm_add_epi16(*(x0),cpack(dy2r,dy2i));
  dy1r = _mm_sub_epi32(_mm_sub_epi32(x3i_2,x2r_2),x1i_2);
  dy1i = _mm_sub_epi32(x1r_2,_mm_add_epi32(x2i_2,x3r_2));
  *(y1) = _mm_add_epi16(*(x0),cpack(dy1r,dy1i));
}

821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855
#elif defined(__arm__)

static inline void ibfly4(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
                          int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3,
                          int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3)__attribute__((always_inline));

static inline void ibfly4(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
                          int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3,
                          int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3)
{

  int32x4_t x1r_2,x1i_2,x2r_2,x2i_2,x3r_2,x3i_2,dy0r,dy0i,dy1r,dy1i,dy2r,dy2i,dy3r,dy3i;


  cmultc(*(x1),*(tw1),&x1r_2,&x1i_2);
  cmultc(*(x2),*(tw2),&x2r_2,&x2i_2);
  cmultc(*(x3),*(tw3),&x3r_2,&x3i_2);

  dy0r  = vqaddq_s32(x1r_2,vqaddq_s32(x2r_2,x3r_2));
  dy0i  = vqaddq_s32(x1i_2,vqaddq_s32(x2i_2,x3i_2));
  *(y0) = vqaddq_s16(*(x0),cpack(dy0r,dy0i));
  dy3r  = vqsubq_s32(x1i_2,vqaddq_s32(x2r_2,x3i_2));
  dy3i  = vqsubq_s32(vqsubq_s32(x3r_2,x2i_2),x1r_2);
  *(y3) = vqaddq_s16(*(x0),cpack(dy3r,dy3i));
  dy2r  = vqsubq_s32(vqsubq_s32(x2r_2,x3r_2),x1r_2);
  dy2i  = vqsubq_s32(vqsubq_s32(x2i_2,x3i_2),x1i_2);
  *(y2) = vqaddq_s16(*(x0),cpack(dy2r,dy2i));
  dy1r  = vqsubq_s32(vqsubq_s32(x3i_2,x2r_2),x1i_2);
  dy1i  = vqsubq_s32(x1r_2,vqaddq_s32(x2i_2,x3r_2));
  *(y1) = vqaddq_s16(*(x0),cpack(dy1r,dy1i));
}

#endif

#if defined(__x86_64__) || defined(__i386__)
856 857

static inline void bfly4_tw1(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
858
                             __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3)__attribute__((always_inline));
859 860

static inline void bfly4_tw1(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
861 862
                             __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3)
{
863 864 865 866 867 868 869 870 871 872 873 874 875 876 877
  register __m128i x1_flip,x3_flip,x02t,x13t;
  register __m128i complex_shuffle = _mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2);

  x02t    = _mm_adds_epi16(*(x0),*(x2));
  x13t    = _mm_adds_epi16(*(x1),*(x3));
  *(y0)   = _mm_adds_epi16(x02t,x13t);
  *(y2)   = _mm_subs_epi16(x02t,x13t);
  x1_flip = _mm_sign_epi16(*(x1),*(__m128i*)conjugatedft);
  x1_flip = _mm_shuffle_epi8(x1_flip,complex_shuffle);
  x3_flip = _mm_sign_epi16(*(x3),*(__m128i*)conjugatedft);
  x3_flip = _mm_shuffle_epi8(x3_flip,complex_shuffle);
  x02t    = _mm_subs_epi16(*(x0),*(x2));
  x13t    = _mm_subs_epi16(x1_flip,x3_flip);
  *(y1)   = _mm_adds_epi16(x02t,x13t);  // x0 + x1f - x2 - x3f
  *(y3)   = _mm_subs_epi16(x02t,x13t);  // x0 - x1f - x2 + x3f
878

879
  /*
880
  *(y0) = _mm_adds_epi16(*(x0),_mm_adds_epi16(*(x1),_mm_adds_epi16(*(x2),*(x3))));
881
  x1_flip = _mm_sign_epi16(*(x1),*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
882
  x1_flip = _mm_shuffle_epi8(x1_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
883
  x3_flip = _mm_sign_epi16(*(x3),*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
884
  x3_flip = _mm_shuffle_epi8(x3_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
885 886 887
  *(y1)   = _mm_adds_epi16(*(x0),_mm_subs_epi16(x1_flip,_mm_adds_epi16(*(x2),x3_flip)));
  *(y2)   = _mm_subs_epi16(*(x0),_mm_subs_epi16(*(x1),_mm_subs_epi16(*(x2),*(x3))));
  *(y3)   = _mm_subs_epi16(*(x0),_mm_adds_epi16(x1_flip,_mm_subs_epi16(*(x2),x3_flip)));
888
  */
889 890
}

891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913
#elif defined(__arm__)

static inline void bfly4_tw1(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
                             int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3)__attribute__((always_inline));

static inline void bfly4_tw1(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
                             int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3)
{

  register int16x8_t x1_flip,x3_flip;

  *(y0) = vqaddq_s16(*(x0),vqaddq_s16(*(x1),vqaddq_s16(*(x2),*(x3))));
  x1_flip = vrev32q_s16(vmulq_s16(*(x1),*(int16x8_t*)conjugatedft));
  x3_flip = vrev32q_s16(vmulq_s16(*(x3),*(int16x8_t*)conjugatedft));
  *(y1)   = vqaddq_s16(*(x0),vqsubq_s16(x1_flip,vqaddq_s16(*(x2),x3_flip)));
  *(y2)   = vqsubq_s16(*(x0),vqsubq_s16(*(x1),vqsubq_s16(*(x2),*(x3))));
  *(y3)   = vqsubq_s16(*(x0),vqaddq_s16(x1_flip,vqsubq_s16(*(x2),x3_flip)));
}

#endif

#if defined(__x86_64__) || defined(__i386__)

914
static inline void ibfly4_tw1(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
915
                              __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3)__attribute__((always_inline));
916 917

static inline void ibfly4_tw1(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
918 919 920
                              __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3)
{

921 922
  register __m128i x1_flip,x3_flip;

923
  *(y0) = _mm_adds_epi16(*(x0),_mm_adds_epi16(*(x1),_mm_adds_epi16(*(x2),*(x3))));
924 925

  x1_flip = _mm_sign_epi16(*(x1),*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
926 927 928
  //  x1_flip = _mm_shufflelo_epi16(x1_flip,_MM_SHUFFLE(2,3,0,1));
  //  x1_flip = _mm_shufflehi_epi16(x1_flip,_MM_SHUFFLE(2,3,0,1));
  x1_flip = _mm_shuffle_epi8(x1_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
929
  x3_flip = _mm_sign_epi16(*(x3),*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
930 931 932
  //  x3_flip = _mm_shufflelo_epi16(x3_flip,_MM_SHUFFLE(2,3,0,1));
  //  x3_flip = _mm_shufflehi_epi16(x3_flip,_MM_SHUFFLE(2,3,0,1));
  x3_flip = _mm_shuffle_epi8(x3_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
933 934 935 936 937
  *(y1)   = _mm_subs_epi16(*(x0),_mm_adds_epi16(x1_flip,_mm_subs_epi16(*(x2),x3_flip)));
  *(y2)   = _mm_subs_epi16(*(x0),_mm_subs_epi16(*(x1),_mm_subs_epi16(*(x2),*(x3))));
  *(y3)   = _mm_adds_epi16(*(x0),_mm_subs_epi16(x1_flip,_mm_adds_epi16(*(x2),x3_flip)));
}

938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959

#elif defined(__arm__)
static inline void ibfly4_tw1(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
			      int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3)__attribute__((always_inline));

static inline void ibfly4_tw1(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
			      int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3)
{

  register int16x8_t x1_flip,x3_flip;

  *(y0) = vqaddq_s16(*(x0),vqaddq_s16(*(x1),vqaddq_s16(*(x2),*(x3))));
  x1_flip = vrev32q_s16(vmulq_s16(*(x1),*(int16x8_t*)conjugatedft));
  x3_flip = vrev32q_s16(vmulq_s16(*(x3),*(int16x8_t*)conjugatedft));
  *(y1)   = vqsubq_s16(*(x0),vqaddq_s16(x1_flip,vqsubq_s16(*(x2),x3_flip)));
  *(y2)   = vqsubq_s16(*(x0),vqsubq_s16(*(x1),vqsubq_s16(*(x2),*(x3))));
  *(y3)   = vqaddq_s16(*(x0),vqsubq_s16(x1_flip,vqaddq_s16(*(x2),x3_flip)));
}

#endif

#if defined(__x86_64__) || defined(__i386__)
960
static inline void bfly4_16(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
961 962 963
                            __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3,
                            __m128i *tw1,__m128i *tw2,__m128i *tw3,
                            __m128i *tw1b,__m128i *tw2b,__m128i *tw3b)__attribute__((always_inline));
964 965

static inline void bfly4_16(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
966 967 968 969
                            __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3,
                            __m128i *tw1,__m128i *tw2,__m128i *tw3,
                            __m128i *tw1b,__m128i *tw2b,__m128i *tw3b)
{
970 971 972 973 974 975 976 977 978 979 980 981 982

  register __m128i x1t,x2t,x3t,x02t,x13t;
  register __m128i x1_flip,x3_flip;

  x1t = packed_cmult2(*(x1),*(tw1),*(tw1b));
  x2t = packed_cmult2(*(x2),*(tw2),*(tw2b));
  x3t = packed_cmult2(*(x3),*(tw3),*(tw3b));


  //  bfly4_tw1(x0,&x1t,&x2t,&x3t,y0,y1,y2,y3);
  x02t  = _mm_adds_epi16(*(x0),x2t);
  x13t  = _mm_adds_epi16(x1t,x3t);
  /*
983
  *(y0) = _mm_adds_epi16(*(x0),_mm_adds_epi16(x1t,_mm_adds_epi16(x2t,x3t)));
984 985 986 987 988 989
  *(y2)   = _mm_subs_epi16(*(x0),_mm_subs_epi16(x1t,_mm_subs_epi16(x2t,x3t)));
  */
  *(y0)   = _mm_adds_epi16(x02t,x13t);
  *(y2)   = _mm_subs_epi16(x02t,x13t);

  x1_flip = _mm_sign_epi16(x1t,*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
990 991 992
  //  x1_flip = _mm_shufflelo_epi16(x1_flip,_MM_SHUFFLE(2,3,0,1));
  //  x1_flip = _mm_shufflehi_epi16(x1_flip,_MM_SHUFFLE(2,3,0,1));
  x1_flip = _mm_shuffle_epi8(x1_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
993
  x3_flip = _mm_sign_epi16(x3t,*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
994 995 996
  //  x3_flip = _mm_shufflelo_epi16(x3_flip,_MM_SHUFFLE(2,3,0,1));
  //  x3_flip = _mm_shufflehi_epi16(x3_flip,_MM_SHUFFLE(2,3,0,1));
  x3_flip = _mm_shuffle_epi8(x3_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
997 998 999 1000 1001 1002 1003 1004 1005 1006 1007
  x02t  = _mm_subs_epi16(*(x0),x2t);
  x13t  = _mm_subs_epi16(x1_flip,x3_flip);
  /*
  *(y1)   = _mm_adds_epi16(*(x0),_mm_subs_epi16(x1_flip,_mm_adds_epi16(x2t,x3_flip)));  // x0 + x1f - x2 - x3f
  *(y3)   = _mm_subs_epi16(*(x0),_mm_adds_epi16(x1_flip,_mm_subs_epi16(x2t,x3_flip)));  // x0 - x1f - x2 + x3f
  */
  *(y1)   = _mm_adds_epi16(x02t,x13t);  // x0 + x1f - x2 - x3f
  *(y3)   = _mm_subs_epi16(x02t,x13t);  // x0 - x1f - x2 + x3f

}

1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043
#elif defined(__arm__)

static inline void bfly4_16(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
                            int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3,
                            int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3,
                            int16x8_t *tw1b,int16x8_t *tw2b,int16x8_t *tw3b)__attribute__((always_inline));

static inline void bfly4_16(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
                            int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3,
                            int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3,
                            int16x8_t *tw1b,int16x8_t *tw2b,int16x8_t *tw3b)
{

  register int16x8_t x1t,x2t,x3t,x02t,x13t;
  register int16x8_t x1_flip,x3_flip;

  x1t = packed_cmult2(*(x1),*(tw1),*(tw1b));
  x2t = packed_cmult2(*(x2),*(tw2),*(tw2b));
  x3t = packed_cmult2(*(x3),*(tw3),*(tw3b));



  x02t  = vqaddq_s16(*(x0),x2t);
  x13t  = vqaddq_s16(x1t,x3t);
  *(y0)   = vqaddq_s16(x02t,x13t);
  *(y2)   = vqsubq_s16(x02t,x13t);
  x1_flip = vrev32q_s16(vmulq_s16(x1t,*(int16x8_t*)conjugatedft));
  x3_flip = vrev32q_s16(vmulq_s16(x3t,*(int16x8_t*)conjugatedft));
  x02t  = vqsubq_s16(*(x0),x2t);
  x13t  = vqsubq_s16(x1_flip,x3_flip);
  *(y1)   = vqaddq_s16(x02t,x13t);  // x0 + x1f - x2 - x3f
  *(y3)   = vqsubq_s16(x02t,x13t);  // x0 - x1f - x2 + x3f
}
#endif

#if defined(__x86_64__) || defined(__i386__)
1044
static inline void ibfly4_16(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
1045 1046 1047
                             __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3,
                             __m128i *tw1,__m128i *tw2,__m128i *tw3,
                             __m128i *tw1b,__m128i *tw2b,__m128i *tw3b)__attribute__((always_inline));
1048 1049

static inline void ibfly4_16(__m128i *x0,__m128i *x1,__m128i *x2,__m128i *x3,
1050 1051 1052 1053 1054
                             __m128i *y0,__m128i *y1,__m128i *y2,__m128i *y3,
                             __m128i *tw1,__m128i *tw2,__m128i *tw3,
                             __m128i *tw1b,__m128i *tw2b,__m128i *tw3b)
{

1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066
  register __m128i x1t,x2t,x3t,x02t,x13t;
  register __m128i x1_flip,x3_flip;

  x1t = packed_cmult2(*(x1),*(tw1),*(tw1b));
  x2t = packed_cmult2(*(x2),*(tw2),*(tw2b));
  x3t = packed_cmult2(*(x3),*(tw3),*(tw3b));


  //  bfly4_tw1(x0,&x1t,&x2t,&x3t,y0,y1,y2,y3);
  x02t  = _mm_adds_epi16(*(x0),x2t);
  x13t  = _mm_adds_epi16(x1t,x3t);
  /*
1067
  *(y0) = _mm_adds_epi16(*(x0),_mm_adds_epi16(x1t,_mm_adds_epi16(x2t,x3t)));
1068 1069 1070 1071 1072 1073
  *(y2)   = _mm_subs_epi16(*(x0),_mm_subs_epi16(x1t,_mm_subs_epi16(x2t,x3t)));
  */
  *(y0)   = _mm_adds_epi16(x02t,x13t);
  *(y2)   = _mm_subs_epi16(x02t,x13t);

  x1_flip = _mm_sign_epi16(x1t,*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
1074 1075 1076
  //  x1_flip = _mm_shufflelo_epi16(x1_flip,_MM_SHUFFLE(2,3,0,1));
  //  x1_flip = _mm_shufflehi_epi16(x1_flip,_MM_SHUFFLE(2,3,0,1));
  x1_flip = _mm_shuffle_epi8(x1_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
1077
  x3_flip = _mm_sign_epi16(x3t,*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
1078 1079 1080
  //  x3_flip = _mm_shufflelo_epi16(x3_flip,_MM_SHUFFLE(2,3,0,1));
  //  x3_flip = _mm_shufflehi_epi16(x3_flip,_MM_SHUFFLE(2,3,0,1));
  x3_flip = _mm_shuffle_epi8(x3_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091
  x02t  = _mm_subs_epi16(*(x0),x2t);
  x13t  = _mm_subs_epi16(x1_flip,x3_flip);
  /*
  *(y1)   = _mm_adds_epi16(*(x0),_mm_subs_epi16(x1_flip,_mm_adds_epi16(x2t,x3_flip)));  // x0 + x1f - x2 - x3f
  *(y3)   = _mm_subs_epi16(*(x0),_mm_adds_epi16(x1_flip,_mm_subs_epi16(x2t,x3_flip)));  // x0 - x1f - x2 + x3f
  */
  *(y3)   = _mm_adds_epi16(x02t,x13t);  // x0 + x1f - x2 - x3f
  *(y1)   = _mm_subs_epi16(x02t,x13t);  // x0 - x1f - x2 + x3f

}

1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125
#elif defined(__arm__)
static inline void ibfly4_16(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
			     int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3,
			     int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3,
			     int16x8_t *tw1b,int16x8_t *tw2b,int16x8_t *tw3b)__attribute__((always_inline));

static inline void ibfly4_16(int16x8_t *x0,int16x8_t *x1,int16x8_t *x2,int16x8_t *x3,
			     int16x8_t *y0,int16x8_t *y1,int16x8_t *y2,int16x8_t *y3,
			     int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3,
			     int16x8_t *tw1b,int16x8_t *tw2b,int16x8_t *tw3b)
{

  register int16x8_t x1t,x2t,x3t,x02t,x13t;
  register int16x8_t x1_flip,x3_flip;

  x1t = packed_cmult2(*(x1),*(tw1),*(tw1b));
  x2t = packed_cmult2(*(x2),*(tw2),*(tw2b));
  x3t = packed_cmult2(*(x3),*(tw3),*(tw3b));

  x02t    = vqaddq_s16(*(x0),x2t);
  x13t    = vqaddq_s16(x1t,x3t);
  *(y0)   = vqaddq_s16(x02t,x13t);
  *(y2)   = vqsubq_s16(x02t,x13t);
  x1_flip = vrev32q_s16(vmulq_s16(x1t,*(int16x8_t*)conjugatedft));
  x3_flip = vrev32q_s16(vmulq_s16(x3t,*(int16x8_t*)conjugatedft));
  x02t    = vqsubq_s16(*(x0),x2t);
  x13t    = vqsubq_s16(x1_flip,x3_flip);
  *(y3)   = vqaddq_s16(x02t,x13t);  // x0 - x1f - x2 + x3f
  *(y1)   = vqsubq_s16(x02t,x13t);  // x0 + x1f - x2 - x3f
}

#endif

#if defined(__x86_64__) || defined(__i386__)
1126
static inline void bfly5(__m128i *x0, __m128i *x1, __m128i *x2, __m128i *x3,__m128i *x4,
1127 1128
                         __m128i *y0, __m128i *y1, __m128i *y2, __m128i *y3,__m128i *y4,
                         __m128i *tw1,__m128i *tw2,__m128i *tw3,__m128i *tw4)__attribute__((always_inline));
1129 1130

static inline void bfly5(__m128i *x0, __m128i *x1, __m128i *x2, __m128i *x3,__m128i *x4,
1131 1132 1133
                         __m128i *y0, __m128i *y1, __m128i *y2, __m128i *y3,__m128i *y4,
                         __m128i *tw1,__m128i *tw2,__m128i *tw3,__m128i *tw4)
{
1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144



  __m128i x1_2,x2_2,x3_2,x4_2,tmpre,tmpim;

  packed_cmult(*(x1),*(tw1),&x1_2);
  packed_cmult(*(x2),*(tw2),&x2_2);
  packed_cmult(*(x3),*(tw3),&x3_2);
  packed_cmult(*(x4),*(tw4),&x4_2);

  *(y0)  = _mm_adds_epi16(*(x0),_mm_adds_epi16(x1_2,_mm_adds_epi16(x2_2,_mm_adds_epi16(x3_2,x4_2))));
1145 1146 1147 1148 1149
  cmult(x1_2,*(W15),&tmpre,&tmpim);
  cmac(x2_2,*(W25),&tmpre,&tmpim);
  cmac(x3_2,*(W35),&tmpre,&tmpim);
  cmac(x4_2,*(W45),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
1150 1151
  *(y1) = _mm_adds_epi16(*(x0),*(y1));

1152 1153 1154 1155 1156
  cmult(x1_2,*(W25),&tmpre,&tmpim);
  cmac(x2_2,*(W45),&tmpre,&tmpim);
  cmac(x3_2,*(W15),&tmpre,&tmpim);
  cmac(x4_2,*(W35),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
1157 1158
  *(y2) = _mm_adds_epi16(*(x0),*(y2));

1159 1160 1161 1162 1163
  cmult(x1_2,*(W35),&tmpre,&tmpim);
  cmac(x2_2,*(W15),&tmpre,&tmpim);
  cmac(x3_2,*(W45),&tmpre,&tmpim);
  cmac(x4_2,*(W25),&tmpre,&tmpim);
  *(y3) = cpack(tmpre,tmpim);
1164 1165
  *(y3) = _mm_adds_epi16(*(x0),*(y3));

1166 1167 1168 1169 1170
  cmult(x1_2,*(W45),&tmpre,&tmpim);
  cmac(x2_2,*(W35),&tmpre,&tmpim);
  cmac(x3_2,*(W25),&tmpre,&tmpim);
  cmac(x4_2,*(W15),&tmpre,&tmpim);
  *(y4) = cpack(tmpre,tmpim);
1171 1172 1173
  *(y4) = _mm_adds_epi16(*(x0),*(y4));


1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225
}

#elif defined(__arm__)
static inline void bfly5(int16x8_t *x0, int16x8_t *x1, int16x8_t *x2, int16x8_t *x3,int16x8_t *x4,
                         int16x8_t *y0, int16x8_t *y1, int16x8_t *y2, int16x8_t *y3,int16x8_t *y4,
                         int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3,int16x8_t *tw4)__attribute__((always_inline));

static inline void bfly5(int16x8_t *x0, int16x8_t *x1, int16x8_t *x2, int16x8_t *x3,int16x8_t *x4,
                         int16x8_t *y0, int16x8_t *y1, int16x8_t *y2, int16x8_t *y3,int16x8_t *y4,
                         int16x8_t *tw1,int16x8_t *tw2,int16x8_t *tw3,int16x8_t *tw4)
{



  int16x8_t x1_2,x2_2,x3_2,x4_2;
  int32x4_t tmpre,tmpim;

  packed_cmult(*(x1),*(tw1),&x1_2);
  packed_cmult(*(x2),*(tw2),&x2_2);
  packed_cmult(*(x3),*(tw3),&x3_2);
  packed_cmult(*(x4),*(tw4),&x4_2);

  *(y0)  = vqaddq_s16(*(x0),vqaddq_s16(x1_2,vqaddq_s16(x2_2,vqaddq_s16(x3_2,x4_2))));
  cmult(x1_2,*(W15),&tmpre,&tmpim);
  cmac(x2_2,*(W25),&tmpre,&tmpim);
  cmac(x3_2,*(W35),&tmpre,&tmpim);
  cmac(x4_2,*(W45),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
  *(y1) = vqaddq_s16(*(x0),*(y1));

  cmult(x1_2,*(W25),&tmpre,&tmpim);
  cmac(x2_2,*(W45),&tmpre,&tmpim);
  cmac(x3_2,*(W15),&tmpre,&tmpim);
  cmac(x4_2,*(W35),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
  *(y2) = vqaddq_s16(*(x0),*(y2));

  cmult(x1_2,*(W35),&tmpre,&tmpim);
  cmac(x2_2,*(W15),&tmpre,&tmpim);
  cmac(x3_2,*(W45),&tmpre,&tmpim);
  cmac(x4_2,*(W25),&tmpre,&tmpim);
  *(y3) = cpack(tmpre,tmpim);
  *(y3) = vqaddq_s16(*(x0),*(y3));

  cmult(x1_2,*(W45),&tmpre,&tmpim);
  cmac(x2_2,*(W35),&tmpre,&tmpim);
  cmac(x3_2,*(W25),&tmpre,&tmpim);
  cmac(x4_2,*(W15),&tmpre,&tmpim);
  *(y4) = cpack(tmpre,tmpim);
  *(y4) = vqaddq_s16(*(x0),*(y4));


1226
}
1227 1228


1229
#endif
1230

1231
#if defined(__x86_64__) || defined(__i386__)
1232
static inline void bfly5_tw1(__m128i *x0, __m128i *x1, __m128i *x2, __m128i *x3,__m128i *x4,
1233
                             __m128i *y0, __m128i *y1, __m128i *y2, __m128i *y3,__m128i *y4) __attribute__((always_inline));
1234 1235

static inline void bfly5_tw1(__m128i *x0, __m128i *x1, __m128i *x2, __m128i *x3,__m128i *x4,
1236 1237
                             __m128i *y0, __m128i *y1, __m128i *y2, __m128i *y3,__m128i *y4)
{
1238 1239 1240

  __m128i tmpre,tmpim;

1241 1242 1243 1244 1245 1246
  *(y0) = _mm_adds_epi16(*(x0),_mm_adds_epi16(*(x1),_mm_adds_epi16(*(x2),_mm_adds_epi16(*(x3),*(x4)))));
  cmult(*(x1),*(W15),&tmpre,&tmpim);
  cmac(*(x2),*(W25),&tmpre,&tmpim);
  cmac(*(x3),*(W35),&tmpre,&tmpim);
  cmac(*(x4),*(W45),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
1247
  *(y1) = _mm_adds_epi16(*(x0),*(y1));
1248 1249 1250 1251 1252
  cmult(*(x1),*(W25),&tmpre,&tmpim);
  cmac(*(x2),*(W45),&tmpre,&tmpim);
  cmac(*(x3),*(W15),&tmpre,&tmpim);
  cmac(*(x4),*(W35),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
1253
  *(y2) = _mm_adds_epi16(*(x0),*(y2));
1254 1255 1256 1257 1258
  cmult(*(x1),*(W35),&tmpre,&tmpim);
  cmac(*(x2),*(W15),&tmpre,&tmpim);
  cmac(*(x3),*(W45),&tmpre,&tmpim);
  cmac(*(x4),*(W25),&tmpre,&tmpim);
  *(y3) = cpack(tmpre,tmpim);
1259
  *(y3) = _mm_adds_epi16(*(x0),*(y3));
1260 1261 1262 1263 1264
  cmult(*(x1),*(W45),&tmpre,&tmpim);
  cmac(*(x2),*(W35),&tmpre,&tmpim);
  cmac(*(x3),*(W25),&tmpre,&tmpim);
  cmac(*(x4),*(W15),&tmpre,&tmpim);
  *(y4) = cpack(tmpre,tmpim);
1265 1266 1267
  *(y4) = _mm_adds_epi16(*(x0),*(y4));
}

1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305
#elif defined(__arm__)
static inline void bfly5_tw1(int16x8_t *x0, int16x8_t *x1, int16x8_t *x2, int16x8_t *x3,int16x8_t *x4,
                             int16x8_t *y0, int16x8_t *y1, int16x8_t *y2, int16x8_t *y3,int16x8_t *y4) __attribute__((always_inline));

static inline void bfly5_tw1(int16x8_t *x0, int16x8_t *x1, int16x8_t *x2, int16x8_t *x3,int16x8_t *x4,
                             int16x8_t *y0, int16x8_t *y1, int16x8_t *y2, int16x8_t *y3,int16x8_t *y4)
{

  int32x4_t tmpre,tmpim;

  *(y0) = vqaddq_s16(*(x0),vqaddq_s16(*(x1),vqaddq_s16(*(x2),vqaddq_s16(*(x3),*(x4)))));
  cmult(*(x1),*(W15),&tmpre,&tmpim);
  cmac(*(x2),*(W25),&tmpre,&tmpim);
  cmac(*(x3),*(W35),&tmpre,&tmpim);
  cmac(*(x4),*(W45),&tmpre,&tmpim);
  *(y1) = cpack(tmpre,tmpim);
  *(y1) = vqaddq_s16(*(x0),*(y1));
  cmult(*(x1),*(W25),&tmpre,&tmpim);
  cmac(*(x2),*(W45),&tmpre,&tmpim);
  cmac(*(x3),*(W15),&tmpre,&tmpim);
  cmac(*(x4),*(W35),&tmpre,&tmpim);
  *(y2) = cpack(tmpre,tmpim);
  *(y2) = vqaddq_s16(*(x0),*(y2));
  cmult(*(x1),*(W35),&tmpre,&tmpim);
  cmac(*(x2),*(W15),&tmpre,&tmpim);
  cmac(*(x3),*(W45),&tmpre,&tmpim);
  cmac(*(x4),*(W25),&tmpre,&tmpim);
  *(y3) = cpack(tmpre,tmpim);
  *(y3) = vqaddq_s16(*(x0),*(y3));
  cmult(*(x1),*(W45),&tmpre,&tmpim);
  cmac(*(x2),*(W35),&tmpre,&tmpim);
  cmac(*(x3),*(W25),&tmpre,&tmpim);
  cmac(*(x4),*(W15),&tmpre,&tmpim);
  *(y4) = cpack(tmpre,tmpim);
  *(y4) = vqaddq_s16(*(x0),*(y4));
}

#endif
1306 1307 1308
// performs 4x4 transpose of input x (complex interleaved) using 128bit SIMD intrinsics
// i.e. x = [x0r x0i x1r x1i ... x15r x15i], y = [x0r x0i x4r x4i x8r x8i x12r x12i x1r x1i x5r x5i x9r x9i x13r x13i x2r x2i ... x15r x15i]

1309
#if defined(__x86_64__) || defined(__i386__)
1310
static inline void transpose16(__m128i *x,__m128i *y) __attribute__((always_inline));
1311 1312
static inline void transpose16(__m128i *x,__m128i *y)
{
1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324
  register __m128i ytmp0,ytmp1,ytmp2,ytmp3;

  ytmp0 = _mm_unpacklo_epi32(x[0],x[1]);
  ytmp1 = _mm_unpackhi_epi32(x[0],x[1]);
  ytmp2 = _mm_unpacklo_epi32(x[2],x[3]);
  ytmp3 = _mm_unpackhi_epi32(x[2],x[3]);
  y[0]    = _mm_unpacklo_epi64(ytmp0,ytmp2);
  y[1]    = _mm_unpackhi_epi64(ytmp0,ytmp2);
  y[2]    = _mm_unpacklo_epi64(ytmp1,ytmp3);
  y[3]    = _mm_unpackhi_epi64(ytmp1,ytmp3);
}

1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340
#elif defined(__arm__)
static inline void transpose16(int16x8_t *x,int16x8_t *y) __attribute__((always_inline));
static inline void transpose16(int16x8_t *x,int16x8_t *y)
{
  register uint32x4x2_t ytmp0,ytmp1;

  ytmp0 = vtrnq_u32((uint32x4_t)(x[0]),(uint32x4_t)(x[1]));
  ytmp1 = vtrnq_u32((uint32x4_t)(x[2]),(uint32x4_t)(x[3]));

  y[0]  = vcombine_s16(vget_low_s16((int16x8_t)ytmp0.val[0]),vget_low_s16((int16x8_t)ytmp1.val[0]));
  y[1]  = vcombine_s16(vget_high_s16((int16x8_t)ytmp0.val[0]),vget_high_s16((int16x8_t)ytmp1.val[0]));
  y[2]  = vcombine_s16(vget_low_s16((int16x8_t)ytmp0.val[1]),vget_low_s16((int16x8_t)ytmp1.val[1]));
  y[3]  = vcombine_s16(vget_high_s16((int16x8_t)ytmp0.val[1]),vget_high_s16((int16x8_t)ytmp1.val[1]));
}

# endif
1341
// same as above but output is offset by off
1342
#if defined(__x86_64__) || defined(__i386__)
1343 1344
static inline void transpose16_ooff(__m128i *x,__m128i *y,int off) __attribute__((always_inline));

1345 1346
static inline void transpose16_ooff(__m128i *x,__m128i *y,int off)
{
1347 1348 1349 1350 1351 1352 1353
  register __m128i ytmp0,ytmp1,ytmp2,ytmp3;
  __m128i *y2=y;

  ytmp0 = _mm_unpacklo_epi32(x[0],x[1]);
  ytmp1 = _mm_unpackhi_epi32(x[0],x[1]);
  ytmp2 = _mm_unpacklo_epi32(x[2],x[3]);
  ytmp3 = _mm_unpackhi_epi32(x[2],x[3]);
1354 1355 1356 1357 1358 1359
  *y2     = _mm_unpacklo_epi64(ytmp0,ytmp2);
  y2+=off;
  *y2     = _mm_unpackhi_epi64(ytmp0,ytmp2);
  y2+=off;
  *y2     = _mm_unpacklo_epi64(ytmp1,ytmp3);
  y2+=off;
1360 1361 1362
  *y2     = _mm_unpackhi_epi64(ytmp1,ytmp3);
}

1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385
#elif defined(__arm__)
static inline void transpose16_ooff(int16x8_t *x,int16x8_t *y,int off) __attribute__((always_inline));

static inline void transpose16_ooff(int16x8_t *x,int16x8_t *y,int off)
{
  int16x8_t *y2=y;
  register uint32x4x2_t ytmp0,ytmp1;

  ytmp0 = vtrnq_u32((uint32x4_t)(x[0]),(uint32x4_t)(x[1]));
  ytmp1 = vtrnq_u32((uint32x4_t)(x[2]),(uint32x4_t)(x[3]));

  *y2   = (int16x8_t)vcombine_s16(vget_low_s16((int16x8_t)ytmp0.val[0]),vget_low_s16((int16x8_t)ytmp1.val[0])); y2+=off;
  *y2   = (int16x8_t)vcombine_s16(vget_low_s16((int16x8_t)ytmp0.val[1]),vget_low_s16((int16x8_t)ytmp1.val[1])); y2+=off;
  *y2   = (int16x8_t)vcombine_s16(vget_high_s16((int16x8_t)ytmp0.val[0]),vget_high_s16((int16x8_t)ytmp1.val[0])); y2+=off;
  *y2   = (int16x8_t)vcombine_s16(vget_high_s16((int16x8_t)ytmp0.val[1]),vget_high_s16((int16x8_t)ytmp1.val[1]));


}

#endif

#if defined(__x86_64__) || defined(__i386__)

1386
static inline void transpose4_ooff(__m64 *x,__m64 *y,int off)__attribute__((always_inline));
1387 1388
static inline void transpose4_ooff(__m64 *x,__m64 *y,int off)
{
1389 1390 1391
  y[0]   = _mm_unpacklo_pi32(x[0],x[1]);
  y[off] = _mm_unpackhi_pi32(x[0],x[1]);
}
1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
#elif (__arm__)

static inline void transpose4_ooff(int16x4_t *x,int16x4_t *y,int off)__attribute__((always_inline));
static inline void transpose4_ooff(int16x4_t *x,int16x4_t *y,int off)
{
  uint32x2x2_t ytmp = vtrn_u32((uint32x2_t)x[0],(uint32x2_t)x[1]);

  y[0]   = (int16x4_t)ytmp.val[0];
  y[off] = (int16x4_t)ytmp.val[1];
}

#endif
1404 1405 1406 1407

// 16-point optimized DFT kernel

int16_t tw16[24] __attribute__((aligned(16))) = { 32767,0,30272,-12540,23169 ,-23170,12539 ,-30273,
1408 1409 1410
                                                  32767,0,23169,-23170,0     ,-32767,-23170,-23170,
                                                  32767,0,12539,-30273,-23170,-23170,-30273,12539
                                                };
1411 1412

int16_t tw16a[24] __attribute__((aligned(16))) = {32767,0,30272,12540,23169 ,23170,12539 ,30273,
1413 1414 1415
                                                  32767,0,23169,23170,0     ,32767,-23170,23170,
                                                  32767,0,12539,30273,-23170,23170,-30273,-12539
                                                 };
1416 1417

int16_t tw16b[24] __attribute__((aligned(16))) = { 0,32767,-12540,30272,-23170,23169 ,-30273,12539,
1418 1419 1420
                                                   0,32767,-23170,23169,-32767,0     ,-23170,-23170,
                                                   0,32767,-30273,12539,-23170,-23170,12539 ,-30273
                                                 };
1421 1422

int16_t tw16c[24] __attribute__((aligned(16))) = { 0,32767,12540,30272,23170,23169 ,30273 ,12539,
1423 1424 1425
                                                   0,32767,23170,23169,32767,0     ,23170 ,-23170,
                                                   0,32767,30273,12539,23170,-23170,-12539,-30273
                                                 };
1426

1427 1428


1429 1430
static inline void dft16(int16_t *x,int16_t *y) __attribute__((always_inline));

1431 1432
static inline void dft16(int16_t *x,int16_t *y)
{
1433

1434 1435
#if defined(__x86_64__) || defined(__i386__)

knopp's avatar
 
knopp committed
1436
  __m128i *tw16a_128=(__m128i *)tw16a,*tw16b_128=(__m128i *)tw16b,*x128=(__m128i *)x,*y128=(__m128i *)y;
1437

1438 1439 1440


#ifndef AVX2
1441 1442
  /*  This is the original version before unrolling

1443
  bfly4_tw1(x128,x128+1,x128+2,x128+3,
1444
      y128,y128+1,y128+2,y128+3);
1445 1446 1447 1448

  transpose16(y128,ytmp);

  bfly4_16(ytmp,ytmp+1,ytmp+2,ytmp+3,
1449 1450
     y128,y128+1,y128+2,y128+3,
     tw16_128,tw16_128+1,tw16_128+2);
1451 1452 1453 1454
  */

  register __m128i x1_flip,x3_flip,x02t,x13t;
  register __m128i ytmp0,ytmp1,ytmp2,ytmp3,xtmp0,xtmp1,xtmp2,xtmp3;
1455
  register __m128i complex_shuffle = _mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2);
1456 1457 1458 1459 1460 1461 1462 1463

  // First stage : 4 Radix-4 butterflies without input twiddles

  x02t    = _mm_adds_epi16(x128[0],x128[2]);
  x13t    = _mm_adds_epi16(x128[1],x128[3]);
  xtmp0   = _mm_adds_epi16(x02t,x13t);
  xtmp2   = _mm_subs_epi16(x02t,x13t);
  x1_flip = _mm_sign_epi16(x128[1],*(__m128i*)conjugatedft);
1464
  x1_flip = _mm_shuffle_epi8(x1_flip,complex_shuffle);
1465
  x3_flip = _mm_sign_epi16(x128[3],*(__m128i*)conjugatedft);
1466
  x3_flip = _mm_shuffle_epi8(x3_flip,complex_shuffle);
1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490
  x02t    = _mm_subs_epi16(x128[0],x128[2]);
  x13t    = _mm_subs_epi16(x1_flip,x3_flip);
  xtmp1   = _mm_adds_epi16(x02t,x13t);  // x0 + x1f - x2 - x3f
  xtmp3   = _mm_subs_epi16(x02t,x13t);  // x0 - x1f - x2 + x3f

  ytmp0   = _mm_unpacklo_epi32(xtmp0,xtmp1);
  ytmp1   = _mm_unpackhi_epi32(xtmp0,xtmp1);
  ytmp2   = _mm_unpacklo_epi32(xtmp2,xtmp3);
  ytmp3   = _mm_unpackhi_epi32(xtmp2,xtmp3);
  xtmp0   = _mm_unpacklo_epi64(ytmp0,ytmp2);
  xtmp1   = _mm_unpackhi_epi64(ytmp0,ytmp2);
  xtmp2   = _mm_unpacklo_epi64(ytmp1,ytmp3);
  xtmp3   = _mm_unpackhi_epi64(ytmp1,ytmp3);

  // Second stage : 4 Radix-4 butterflies with input twiddles
  xtmp1 = packed_cmult2(xtmp1,tw16a_128[0],tw16b_128[0]);
  xtmp2 = packed_cmult2(xtmp2,tw16a_128[1],tw16b_128[1]);
  xtmp3 = packed_cmult2(xtmp3,tw16a_128[2],tw16b_128[2]);

  x02t    = _mm_adds_epi16(xtmp0,xtmp2);
  x13t    = _mm_adds_epi16(xtmp1,xtmp3);
  y128[0] = _mm_adds_epi16(x02t,x13t);
  y128[2] = _mm_subs_epi16(x02t,x13t);
  x1_flip = _mm_sign_epi16(xtmp1,*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
1491
  x1_flip = _mm_shuffle_epi8(x1_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
1492
  x3_flip = _mm_sign_epi16(xtmp3,*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
1493
  x3_flip = _mm_shuffle_epi8(x3_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
1494 1495 1496 1497
  x02t    = _mm_subs_epi16(xtmp0,xtmp2);
  x13t    = _mm_subs_epi16(x1_flip,x3_flip);
  y128[1] = _mm_adds_epi16(x02t,x13t);  // x0 + x1f - x2 - x3f
  y128[3] = _mm_subs_epi16(x02t,x13t);  // x0 - x1f - x2 + x3f
1498

1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523
#else

  //  x02t    = _mm_adds_epi16(x128[0],x128[2]);
  //  x13t    = _mm_adds_epi16(x128[1],x128[3]);

  xt      = _mm256_adds_epi16(x256[0],x256[1]);

  xtmp0   = _mm_adds_epi16(x02t,x13t);
  xtmp2   = _mm_subs_epi16(x02t,x13t);

  x13_flip
  x1_flip = _mm_sign_epi16(x128[1],*(__m128i*)conjugatedft);
  x1_flip = _mm_shuffle_epi8(x1_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
  x3_flip = _mm_sign_epi16(x128[3],*(__m128i*)conjugatedft);
  x3_flip = _mm_shuffle_epi8(x3_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));

  //  x02t    = _mm_subs_epi16(x128[0],x128[2]);
  //  x13t    = _mm_subs_epi16(x1_flip,x3_flip);
  xt      = _mm256_subs_epi16(x256flip0,x256flip1);

  xtmp1   = _mm_adds_epi16(x02t,x13t);  // x0 + x1f - x2 - x3f
  xtmp3   = _mm_subs_epi16(x02t,x13t);  // x0 - x1f - x2 + x3f


#endif
1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538

#elif defined(__arm__)

  int16x8_t *tw16a_128=(int16x8_t *)tw16a,*tw16b_128=(int16x8_t *)tw16b,*x128=(int16x8_t *)x,*y128=(int16x8_t *)y;

  /*  This is the original version before unrolling

  bfly4_tw1(x128,x128+1,x128+2,x128+3,
      y128,y128+1,y128+2,y128+3);

  transpose16(y128,ytmp);

  bfly4_16(ytmp,ytmp+1,ytmp+2,ytmp+3,
     y128,y128+1,y128+2,y128+3,
     tw16_128,tw16_128+1,tw16_128+2);
1539
  */
1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592

  register int16x8_t x1_flip,x3_flip,x02t,x13t;
  register int16x8_t xtmp0,xtmp1,xtmp2,xtmp3;
  register uint32x4x2_t ytmp0,ytmp1;
  register int16x8_t ytmp0b,ytmp1b,ytmp2b,ytmp3b;

  // First stage : 4 Radix-4 butterflies without input twiddles
  
  x02t    = vqaddq_s16(x128[0],x128[2]);
  x13t    = vqaddq_s16(x128[1],x128[3]);
  xtmp0   = vqaddq_s16(x02t,x13t);
  xtmp2   = vqsubq_s16(x02t,x13t);
  x1_flip = vrev32q_s16(vmulq_s16(x128[1],*(int16x8_t*)conjugatedft));
  x3_flip = vrev32q_s16(vmulq_s16(x128[3],*(int16x8_t*)conjugatedft));
  x02t    = vqsubq_s16(x128[0],x128[2]);
  x13t    = vqsubq_s16(x1_flip,x3_flip);
  xtmp1   = vqaddq_s16(x02t,x13t);  // x0 + x1f - x2 - x3f
  xtmp3   = vqsubq_s16(x02t,x13t);  // x0 - x1f - x2 + x3f

  ytmp0  = vtrnq_u32((uint32x4_t)(xtmp0),(uint32x4_t)(xtmp1));
// y0[0] = [x00 x10 x02 x12], y0[1] = [x01 x11 x03 x13]
  ytmp1  = vtrnq_u32((uint32x4_t)(xtmp2),(uint32x4_t)(xtmp3));
// y1[0] = [x20 x30 x22 x32], y1[1] = [x21 x31 x23 x33]


  ytmp0b = vcombine_s16(vget_low_s16((int16x8_t)ytmp0.val[0]),vget_low_s16((int16x8_t)ytmp1.val[0]));
// y0 = [x00 x10 x20 x30] 
  ytmp1b = vcombine_s16(vget_low_s16((int16x8_t)ytmp0.val[1]),vget_low_s16((int16x8_t)ytmp1.val[1]));
// t1 = [x01 x11 x21 x31] 
  ytmp2b = vcombine_s16(vget_high_s16((int16x8_t)ytmp0.val[0]),vget_high_s16((int16x8_t)ytmp1.val[0]));
// t2 = [x02 x12 x22 x32]
  ytmp3b = vcombine_s16(vget_high_s16((int16x8_t)ytmp0.val[1]),vget_high_s16((int16x8_t)ytmp1.val[1]));
// t3 = [x03 x13 x23 x33]


  // Second stage : 4 Radix-4 butterflies with input twiddles
  xtmp1 = packed_cmult2(ytmp1b,tw16a_128[0],tw16b_128[0]);
  xtmp2 = packed_cmult2(ytmp2b,tw16a_128[1],tw16b_128[1]);
  xtmp3 = packed_cmult2(ytmp3b,tw16a_128[2],tw16b_128[2]);

  x02t    = vqaddq_s16(ytmp0b,xtmp2);
  x13t    = vqaddq_s16(xtmp1,xtmp3);
  y128[0] = vqaddq_s16(x02t,x13t);
  y128[2] = vqsubq_s16(x02t,x13t);
  x1_flip = vrev32q_s16(vmulq_s16(xtmp1,*(int16x8_t*)conjugatedft));
  x3_flip = vrev32q_s16(vmulq_s16(xtmp3,*(int16x8_t*)conjugatedft));
  x02t    = vqsubq_s16(ytmp0b,xtmp2);
  x13t    = vqsubq_s16(x1_flip,x3_flip);
  y128[1] = vqaddq_s16(x02t,x13t);  // x0 + x1f - x2 - x3f
  y128[3] = vqsubq_s16(x02t,x13t);  // x0 - x1f - x2 + x3f


#endif
1593 1594 1595 1596
}

static inline void idft16(int16_t *x,int16_t *y) __attribute__((always_inline));

1597 1598
static inline void idft16(int16_t *x,int16_t *y)
{
1599

1600
#if defined(__x86_64__) || defined(__i386__)
knopp's avatar
 
knopp committed
1601
  __m128i *tw16a_128=(__m128i *)tw16,*tw16b_128=(__m128i *)tw16c,*x128=(__m128i *)x,*y128=(__m128i *)y;
1602

1603 1604
  /*
  bfly4_tw1(x128,x128+1,x128+2,x128+3,
1605
      y128,y128+1,y128+2,y128+3);
1606 1607 1608 1609

  transpose16(y128,ytmp);

  bfly4_16(ytmp,ytmp+1,ytmp+2,ytmp+3,
1610 1611
     y128,y128+1,y128+2,y128+3,
     tw16_128,tw16_128+1,tw16_128+2);
1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623
  */

  register __m128i x1_flip,x3_flip,x02t,x13t;
  register __m128i ytmp0,ytmp1,ytmp2,ytmp3,xtmp0,xtmp1,xtmp2,xtmp3;

  // First stage : 4 Radix-4 butterflies without input twiddles

  x02t    = _mm_adds_epi16(x128[0],x128[2]);
  x13t    = _mm_adds_epi16(x128[1],x128[3]);
  xtmp0   = _mm_adds_epi16(x02t,x13t);
  xtmp2   = _mm_subs_epi16(x02t,x13t);
  x1_flip = _mm_sign_epi16(x128[1],*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
1624
  x1_flip = _mm_shuffle_epi8(x1_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
1625
  x3_flip = _mm_sign_epi16(x128[3],*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
1626
  x3_flip = _mm_shuffle_epi8(x3_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650
  x02t    = _mm_subs_epi16(x128[0],x128[2]);
  x13t    = _mm_subs_epi16(x1_flip,x3_flip);
  xtmp3   = _mm_adds_epi16(x02t,x13t);  // x0 + x1f - x2 - x3f
  xtmp1   = _mm_subs_epi16(x02t,x13t);  // x0 - x1f - x2 + x3f

  ytmp0   = _mm_unpacklo_epi32(xtmp0,xtmp1);
  ytmp1   = _mm_unpackhi_epi32(xtmp0,xtmp1);
  ytmp2   = _mm_unpacklo_epi32(xtmp2,xtmp3);
  ytmp3   = _mm_unpackhi_epi32(xtmp2,xtmp3);
  xtmp0   = _mm_unpacklo_epi64(ytmp0,ytmp2);
  xtmp1   = _mm_unpackhi_epi64(ytmp0,ytmp2);
  xtmp2   = _mm_unpacklo_epi64(ytmp1,ytmp3);
  xtmp3   = _mm_unpackhi_epi64(ytmp1,ytmp3);

  // Second stage : 4 Radix-4 butterflies with input twiddles
  xtmp1 = packed_cmult2(xtmp1,tw16a_128[0],tw16b_128[0]);
  xtmp2 = packed_cmult2(xtmp2,tw16a_128[1],tw16b_128[1]);
  xtmp3 = packed_cmult2(xtmp3,tw16a_128[2],tw16b_128[2]);

  x02t    = _mm_adds_epi16(xtmp0,xtmp2);
  x13t    = _mm_adds_epi16(xtmp1,xtmp3);
  y128[0] = _mm_adds_epi16(x02t,x13t);
  y128[2] = _mm_subs_epi16(x02t,x13t);
  x1_flip = _mm_sign_epi16(xtmp1,*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
1651
  x1_flip = _mm_shuffle_epi8(x1_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
1652
  x3_flip = _mm_sign_epi16(xtmp3,*(__m128i*)conjugatedft);
knopp's avatar
 
knopp committed
1653
  x3_flip = _mm_shuffle_epi8(x3_flip,_mm_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2));
1654 1655 1656 1657 1658
  x02t    = _mm_subs_epi16(xtmp0,xtmp2);
  x13t    = _mm_subs_epi16(x1_flip,x3_flip);
  y128[3] = _mm_adds_epi16(x02t,x13t);  // x0 + x1f - x2 - x3f
  y128[1] = _mm_subs_epi16(x02t,x13t);  // x0 - x1f - x2 + x3f

1659 1660
#elif defined(__arm__)
  int16x8_t *tw16a_128=(int16x8_t *)tw16,*tw16b_128=(int16x8_t *)tw16c,*x128=(int16x8_t *)x,*y128=(int16x8_t *)y;
1661

1662
  /*  This is the original version before unrolling
1663

1664
  bfly4_tw1(x128,x128+1,x128+2,x128+3,
1665
      y128,y128+1,y128+2,y128+3);
1666 1667 1668

  transpose16(y128,ytmp);

1669 1670 1671 1672
  bfly4_16(ytmp,ytmp+1,ytmp+2,ytmp+3,
     y128,y128+1,y128+2,y128+3,
     tw16_128,tw16_128+1,tw16_128+2);
  */
1673

1674 1675 1676 1677
  register int16x8_t x1_flip,x3_flip,x02t,x13t;
  register int16x8_t xtmp0,xtmp1,xtmp2,xtmp3;
  register uint32x4x2_t ytmp0,ytmp1;
  register int16x8_t ytmp0b,ytmp1b,ytmp2b,ytmp3b;
1678

1679
  // First stage : 4 Radix-4 butterflies without input twiddles
1680

1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705
  x02t    = vqaddq_s16(x128[0],x128[2]);
  x13t    = vqaddq_s16(x128[1],x128[3]);
  xtmp0   = vqaddq_s16(x02t,x13t);
  xtmp2   = vqsubq_s16(x02t,x13t);
  x1_flip = vrev32q_s16(vmulq_s16(x128[1],*(int16x8_t*)conjugatedft));
  x3_flip = vrev32q_s16(vmulq_s16(x128[3],*(int16x8_t*)conjugatedft));
  x02t    = vqsubq_s16(x128[0],x128[2]);
  x13t    = vqsubq_s16(x1_flip,x3_flip);
  xtmp3   = vqaddq_s16(x02t,x13t);  // x0 + x1f - x2 - x3f
  xtmp1   = vqsubq_s16(x02t,x13t);  // x0 - x1f - x2 + x3f

  ytmp0  = vtrnq_u32((uint32x4_t)(xtmp0),(uint32x4_t)(xtmp1));
// y0[0] = [x00 x10 x02 x12], y0[1] = [x01 x11 x03 x13]
  ytmp1  = vtrnq_u32((uint32x4_t)(xtmp2),(uint32x4_t)(xtmp3));
// y1[0] = [x20 x30 x22 x32], y1[1] = [x21 x31 x23 x33]


  ytmp0b = vcombine_s16(vget_low_s16((int16x8_t)ytmp0.val[0]),vget_low_s16((int16x8_t)ytmp1.val[0]));
// y0 = [x00 x10 x20 x30] 
  ytmp1b = vcombine_s16(vget_low_s16((int16x8_t)ytmp0.val[1]),vget_low_s16((int16x8_t)ytmp1.val[1]));
// t1 = [x01 x11 x21 x31] 
  ytmp2b = vcombine_s16(vget_high_s16((int16x8_t)ytmp0.val[0]),vget_high_s16((int16x8_t)ytmp1.val[0]));
// t2 = [x02 x12 x22 x32]
  ytmp3b = vcombine_s16(vget_high_s16((int16x8_t)ytmp0.val[1]),vget_high_s16((int16x8_t)ytmp1.val[1]));
// t3 = [x03 x13 x23 x33]
1706

1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727
  // Second stage : 4 Radix-4 butterflies with input twiddles
  xtmp1 = packed_cmult2(ytmp1b,tw16a_128[0],tw16b_128[0]);
  xtmp2 = packed_cmult2(ytmp2b,tw16a_128[1],tw16b_128[1]);
  xtmp3 = packed_cmult2(ytmp3b,tw16a_128[2],tw16b_128[2]);

  x02t    = vqaddq_s16(ytmp0b,xtmp2);
  x13t    = vqaddq_s16(xtmp1,xtmp3);
  y128[0] = vqaddq_s16(x02t,x13t);
  y128[2] = vqsubq_s16(x02t,x13t);
  x1_flip = vrev32q_s16(vmulq_s16(xtmp1,*(int16x8_t*)conjugatedft));
  x3_flip = vrev32q_s16(vmulq_s16(xtmp3,*(int16x8_t*)conjugatedft));
  x02t    = vqsubq_s16(ytmp0b,xtmp2);
  x13t    = vqsubq_s16(x1_flip,x3_flip);
  y128[3] = vqaddq_s16(x02t,x13t);  // x0 + x1f - x2 - x3f
  y128[1] = vqsubq_s16(x02t,x13t);  // x0 - x1f - x2 + x3f

#endif
}


// 64-point optimized DFT
1728 1729

int16_t tw64[96] __attribute__((aligned(16))) = { 32767,0,32609,-3212,32137,-6393,31356,-9512,30272,-12540,28897,-15447,27244,-18205,25329,-20788,23169,-23170,20787,-25330,18204,-27245,15446,-28898,12539,-30273,9511,-31357,6392,-32138,3211,-32610,
1730 1731 1732
                                                  32767,0,32137,-6393,30272,-12540,27244,-18205,23169,-23170,18204,-27245,12539,-30273,6392,-32138,0,-32767,-6393,-32138,-12540,-30273,-18205,-27245,-23170,-23170,-27245,-18205,-30273,-12540,-32138,-6393,
                                                  32767,0,31356,-9512,27244,-18205,20787,-25330,12539,-30273,3211,-32610,-6393,-32138,-15447,-28898,-23170,-23170,-28898,-15447,-32138,-6393,-32610,3211,-30273,12539,-25330,20787,-18205,27244,-9512,31356
                                                };
1733 1734

int16_t tw64a[96] __attribute__((aligned(16))) = { 32767,0,32609,3212,32137,6393,31356,9512,30272,12540,28897,15447,27244,18205,25329,20788,23169,23170,20787,25330,18204,27245,15446,28898,12539,30273,9511,31357,6392,32138,3211,32610,
1735 1736 1737
                                                   32767,0,32137,6393,30272,12540,27244,18205,23169,23170,18204,27245,12539,30273,6392,32138,0,32767,-6393,32138,-12540,30273,-18205,27245,-23170,23170,-27245,18205,-30273,12540,-32138,6393,
                                                   32767,0,31356,9512,27244,18205,20787,25330,12539,30273,3211,32610,-6393,32138,-15447,28898,-23170,23170,-28898,15447,-32138,6393,-32610,-3211,-30273,-12539,-25330,-20787,-18205,-27244,-9512,-31356
                                                 };
1738 1739

int16_t tw64b[96] __attribute__((aligned(16))) = { 0,32767,-3212,32609,-6393,32137,-9512,31356,-12540,30272,-15447,28897,-18205,27244,-20788,25329,-23170,23169,-25330,20787,-27245,18204,-28898,15446,-30273,12539,-31357,9511,-32138,6392,-32610,3211,
1740 1741 1742
                                                   0,32767,-6393,32137,-12540,30272,-18205,27244,-23170,23169,-27245,18204,-30273,12539,-32138,6392,-32767,0,-32138,-6393,-30273,-12540,-27245,-18205,-23170,-23170,-18205,-27245,-12540,-30273,-6393,-32138,
                                                   0,32767,-9512,31356,-18205,27244,-25330,20787,-30273,12539,-32610,3211,-32138,-6393,-28898,-15447,-23170,-23170,-15447,-28898,-6393,-32138,3211,-32610,12539,-30273,20787,-25330,27244,-18205,31356,-9512
                                                 };
1743 1744

int16_t tw64c[96] __attribute__((aligned(16))) = { 0,32767,3212,32609,6393,32137,9512,31356,12540,30272,15447,28897,18205,27244,20788,25329,23170,23169,25330,20787,27245,18204,28898,15446,30273,12539,31357,9511,32138,6392,32610,3211,
1745 1746 1747
                                                   0,32767,6393,32137,12540,30272,18205,27244,23170,23169,27245,18204,30273,12539,32138,6392,32767,0,32138,-6393,30273,-12540,27245,-18205,23170,-23170,18205,-27245,12540,-30273,6393,-32138,
                                                   0,32767,9512,31356,18205,27244,25330,20787,30273,12539,32610,3211,32138,-6393,28898,-15447,23170,-23170,15447,-28898,6393,-32138,-3211,-32610,-12539,-30273,-20787,-25330,-27244,-18205,-31356,-9512
                                                 };
1748 1749


1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764
#if defined(__x86_64__) || defined(__i386__)
#define simd_q15_t __m128i
#define simdshort_q15_t __m64
#define shiftright_int16(a,shift) _mm_srai_epi16(a,shift)
#define set1_int16(a) _mm_set1_epi16(a);
#define mulhi_int16(a,b) _mm_slli_epi16(_mm_mulhi_epi16(a,b),1);
#elif defined(__arm__)
#define simd_q15_t int16x8_t
#define simdshort_q15_t int16x4_t
#define shiftright_int16(a,shift) vshrq_n_s16(a,shift)
#define set1_int16(a) vdupq_n_s16(a)
#define mulhi_int16(a,b) vqdmulhq_s16(a,b);
#define _mm_empty() 
#define _m_empty()
#endif
1765

1766 1767
void dft64(int16_t *x,int16_t *y,int scale)
{
1768

1769
  simd_q15_t xtmp[16],ytmp[16],*tw64a_128=(simd_q15_t *)tw64a,*tw64b_128=(simd_q15_t *)tw64b,*x128=(simd_q15_t *)x,*y128=(simd_q15_t *)y;
1770

1771

1772 1773 1774 1775 1776 1777 1778 1779
#ifdef D64STATS
  time_stats_t ts_t,ts_d,ts_b;

  reset_meas(&ts_t);
  reset_meas(&ts_d);
  reset_meas(&ts_b);
  start_meas(&ts_t);
#endif
1780