Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 56 | mjames | 1 | /* ---------------------------------------------------------------------- |
| 2 | * Project: CMSIS DSP Library |
||
| 3 | * Title: arm_cfft_radix4_f32.c |
||
| 4 | * Description: Radix-4 Decimation in Frequency CFFT & CIFFT Floating point processing function |
||
| 5 | * |
||
| 6 | * $Date: 27. January 2017 |
||
| 7 | * $Revision: V.1.5.1 |
||
| 8 | * |
||
| 9 | * Target Processor: Cortex-M cores |
||
| 10 | * -------------------------------------------------------------------- */ |
||
| 11 | /* |
||
| 12 | * Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved. |
||
| 13 | * |
||
| 14 | * SPDX-License-Identifier: Apache-2.0 |
||
| 15 | * |
||
| 16 | * Licensed under the Apache License, Version 2.0 (the License); you may |
||
| 17 | * not use this file except in compliance with the License. |
||
| 18 | * You may obtain a copy of the License at |
||
| 19 | * |
||
| 20 | * www.apache.org/licenses/LICENSE-2.0 |
||
| 21 | * |
||
| 22 | * Unless required by applicable law or agreed to in writing, software |
||
| 23 | * distributed under the License is distributed on an AS IS BASIS, WITHOUT |
||
| 24 | * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
||
| 25 | * See the License for the specific language governing permissions and |
||
| 26 | * limitations under the License. |
||
| 27 | */ |
||
| 28 | |||
| 29 | #include "arm_math.h" |
||
| 30 | |||
| 31 | extern void arm_bitreversal_f32( |
||
| 32 | float32_t * pSrc, |
||
| 33 | uint16_t fftSize, |
||
| 34 | uint16_t bitRevFactor, |
||
| 35 | uint16_t * pBitRevTab); |
||
| 36 | |||
| 37 | void arm_radix4_butterfly_f32( |
||
| 38 | float32_t * pSrc, |
||
| 39 | uint16_t fftLen, |
||
| 40 | float32_t * pCoef, |
||
| 41 | uint16_t twidCoefModifier); |
||
| 42 | |||
| 43 | void arm_radix4_butterfly_inverse_f32( |
||
| 44 | float32_t * pSrc, |
||
| 45 | uint16_t fftLen, |
||
| 46 | float32_t * pCoef, |
||
| 47 | uint16_t twidCoefModifier, |
||
| 48 | float32_t onebyfftLen); |
||
| 49 | |||
| 50 | |||
| 51 | /** |
||
| 52 | * @ingroup groupTransforms |
||
| 53 | */ |
||
| 54 | |||
| 55 | /** |
||
| 56 | * @addtogroup ComplexFFT |
||
| 57 | * @{ |
||
| 58 | */ |
||
| 59 | |||
| 60 | /** |
||
| 61 | * @details |
||
| 62 | * @brief Processing function for the floating-point Radix-4 CFFT/CIFFT. |
||
| 63 | * @deprecated Do not use this function. It has been superseded by \ref arm_cfft_f32 and will be removed |
||
| 64 | * in the future. |
||
| 65 | * @param[in] *S points to an instance of the floating-point Radix-4 CFFT/CIFFT structure. |
||
| 66 | * @param[in, out] *pSrc points to the complex data buffer of size <code>2*fftLen</code>. Processing occurs in-place. |
||
| 67 | * @return none. |
||
| 68 | */ |
||
| 69 | |||
| 70 | void arm_cfft_radix4_f32( |
||
| 71 | const arm_cfft_radix4_instance_f32 * S, |
||
| 72 | float32_t * pSrc) |
||
| 73 | { |
||
| 74 | if (S->ifftFlag == 1U) |
||
| 75 | { |
||
| 76 | /* Complex IFFT radix-4 */ |
||
| 77 | arm_radix4_butterfly_inverse_f32(pSrc, S->fftLen, S->pTwiddle, S->twidCoefModifier, S->onebyfftLen); |
||
| 78 | } |
||
| 79 | else |
||
| 80 | { |
||
| 81 | /* Complex FFT radix-4 */ |
||
| 82 | arm_radix4_butterfly_f32(pSrc, S->fftLen, S->pTwiddle, S->twidCoefModifier); |
||
| 83 | } |
||
| 84 | |||
| 85 | if (S->bitReverseFlag == 1U) |
||
| 86 | { |
||
| 87 | /* Bit Reversal */ |
||
| 88 | arm_bitreversal_f32(pSrc, S->fftLen, S->bitRevFactor, S->pBitRevTable); |
||
| 89 | } |
||
| 90 | |||
| 91 | } |
||
| 92 | |||
| 93 | /** |
||
| 94 | * @} end of ComplexFFT group |
||
| 95 | */ |
||
| 96 | |||
| 97 | /* ---------------------------------------------------------------------- |
||
| 98 | * Internal helper function used by the FFTs |
||
| 99 | * ---------------------------------------------------------------------- */ |
||
| 100 | |||
| 101 | /* |
||
| 102 | * @brief Core function for the floating-point CFFT butterfly process. |
||
| 103 | * @param[in, out] *pSrc points to the in-place buffer of floating-point data type. |
||
| 104 | * @param[in] fftLen length of the FFT. |
||
| 105 | * @param[in] *pCoef points to the twiddle coefficient buffer. |
||
| 106 | * @param[in] twidCoefModifier twiddle coefficient modifier that supports different size FFTs with the same twiddle factor table. |
||
| 107 | * @return none. |
||
| 108 | */ |
||
| 109 | |||
| 110 | void arm_radix4_butterfly_f32( |
||
| 111 | float32_t * pSrc, |
||
| 112 | uint16_t fftLen, |
||
| 113 | float32_t * pCoef, |
||
| 114 | uint16_t twidCoefModifier) |
||
| 115 | { |
||
| 116 | |||
| 117 | float32_t co1, co2, co3, si1, si2, si3; |
||
| 118 | uint32_t ia1, ia2, ia3; |
||
| 119 | uint32_t i0, i1, i2, i3; |
||
| 120 | uint32_t n1, n2, j, k; |
||
| 121 | |||
| 122 | #if defined (ARM_MATH_DSP) |
||
| 123 | |||
| 124 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
||
| 125 | |||
| 126 | float32_t xaIn, yaIn, xbIn, ybIn, xcIn, ycIn, xdIn, ydIn; |
||
| 127 | float32_t Xaplusc, Xbplusd, Yaplusc, Ybplusd, Xaminusc, Xbminusd, Yaminusc, |
||
| 128 | Ybminusd; |
||
| 129 | float32_t Xb12C_out, Yb12C_out, Xc12C_out, Yc12C_out, Xd12C_out, Yd12C_out; |
||
| 130 | float32_t Xb12_out, Yb12_out, Xc12_out, Yc12_out, Xd12_out, Yd12_out; |
||
| 131 | float32_t *ptr1; |
||
| 132 | float32_t p0,p1,p2,p3,p4,p5; |
||
| 133 | float32_t a0,a1,a2,a3,a4,a5,a6,a7; |
||
| 134 | |||
| 135 | /* Initializations for the first stage */ |
||
| 136 | n2 = fftLen; |
||
| 137 | n1 = n2; |
||
| 138 | |||
| 139 | /* n2 = fftLen/4 */ |
||
| 140 | n2 >>= 2U; |
||
| 141 | i0 = 0U; |
||
| 142 | ia1 = 0U; |
||
| 143 | |||
| 144 | j = n2; |
||
| 145 | |||
| 146 | /* Calculation of first stage */ |
||
| 147 | do |
||
| 148 | { |
||
| 149 | /* index calculation for the input as, */ |
||
| 150 | /* pSrc[i0 + 0], pSrc[i0 + fftLen/4], pSrc[i0 + fftLen/2], pSrc[i0 + 3fftLen/4] */ |
||
| 151 | i1 = i0 + n2; |
||
| 152 | i2 = i1 + n2; |
||
| 153 | i3 = i2 + n2; |
||
| 154 | |||
| 155 | xaIn = pSrc[(2U * i0)]; |
||
| 156 | yaIn = pSrc[(2U * i0) + 1U]; |
||
| 157 | |||
| 158 | xbIn = pSrc[(2U * i1)]; |
||
| 159 | ybIn = pSrc[(2U * i1) + 1U]; |
||
| 160 | |||
| 161 | xcIn = pSrc[(2U * i2)]; |
||
| 162 | ycIn = pSrc[(2U * i2) + 1U]; |
||
| 163 | |||
| 164 | xdIn = pSrc[(2U * i3)]; |
||
| 165 | ydIn = pSrc[(2U * i3) + 1U]; |
||
| 166 | |||
| 167 | /* xa + xc */ |
||
| 168 | Xaplusc = xaIn + xcIn; |
||
| 169 | /* xb + xd */ |
||
| 170 | Xbplusd = xbIn + xdIn; |
||
| 171 | /* ya + yc */ |
||
| 172 | Yaplusc = yaIn + ycIn; |
||
| 173 | /* yb + yd */ |
||
| 174 | Ybplusd = ybIn + ydIn; |
||
| 175 | |||
| 176 | /* index calculation for the coefficients */ |
||
| 177 | ia2 = ia1 + ia1; |
||
| 178 | co2 = pCoef[ia2 * 2U]; |
||
| 179 | si2 = pCoef[(ia2 * 2U) + 1U]; |
||
| 180 | |||
| 181 | /* xa - xc */ |
||
| 182 | Xaminusc = xaIn - xcIn; |
||
| 183 | /* xb - xd */ |
||
| 184 | Xbminusd = xbIn - xdIn; |
||
| 185 | /* ya - yc */ |
||
| 186 | Yaminusc = yaIn - ycIn; |
||
| 187 | /* yb - yd */ |
||
| 188 | Ybminusd = ybIn - ydIn; |
||
| 189 | |||
| 190 | /* xa' = xa + xb + xc + xd */ |
||
| 191 | pSrc[(2U * i0)] = Xaplusc + Xbplusd; |
||
| 192 | /* ya' = ya + yb + yc + yd */ |
||
| 193 | pSrc[(2U * i0) + 1U] = Yaplusc + Ybplusd; |
||
| 194 | |||
| 195 | /* (xa - xc) + (yb - yd) */ |
||
| 196 | Xb12C_out = (Xaminusc + Ybminusd); |
||
| 197 | /* (ya - yc) + (xb - xd) */ |
||
| 198 | Yb12C_out = (Yaminusc - Xbminusd); |
||
| 199 | /* (xa + xc) - (xb + xd) */ |
||
| 200 | Xc12C_out = (Xaplusc - Xbplusd); |
||
| 201 | /* (ya + yc) - (yb + yd) */ |
||
| 202 | Yc12C_out = (Yaplusc - Ybplusd); |
||
| 203 | /* (xa - xc) - (yb - yd) */ |
||
| 204 | Xd12C_out = (Xaminusc - Ybminusd); |
||
| 205 | /* (ya - yc) + (xb - xd) */ |
||
| 206 | Yd12C_out = (Xbminusd + Yaminusc); |
||
| 207 | |||
| 208 | co1 = pCoef[ia1 * 2U]; |
||
| 209 | si1 = pCoef[(ia1 * 2U) + 1U]; |
||
| 210 | |||
| 211 | /* index calculation for the coefficients */ |
||
| 212 | ia3 = ia2 + ia1; |
||
| 213 | co3 = pCoef[ia3 * 2U]; |
||
| 214 | si3 = pCoef[(ia3 * 2U) + 1U]; |
||
| 215 | |||
| 216 | Xb12_out = Xb12C_out * co1; |
||
| 217 | Yb12_out = Yb12C_out * co1; |
||
| 218 | Xc12_out = Xc12C_out * co2; |
||
| 219 | Yc12_out = Yc12C_out * co2; |
||
| 220 | Xd12_out = Xd12C_out * co3; |
||
| 221 | Yd12_out = Yd12C_out * co3; |
||
| 222 | |||
| 223 | /* xb' = (xa+yb-xc-yd)co1 - (ya-xb-yc+xd)(si1) */ |
||
| 224 | //Xb12_out -= Yb12C_out * si1; |
||
| 225 | p0 = Yb12C_out * si1; |
||
| 226 | /* yb' = (ya-xb-yc+xd)co1 + (xa+yb-xc-yd)(si1) */ |
||
| 227 | //Yb12_out += Xb12C_out * si1; |
||
| 228 | p1 = Xb12C_out * si1; |
||
| 229 | /* xc' = (xa-xb+xc-xd)co2 - (ya-yb+yc-yd)(si2) */ |
||
| 230 | //Xc12_out -= Yc12C_out * si2; |
||
| 231 | p2 = Yc12C_out * si2; |
||
| 232 | /* yc' = (ya-yb+yc-yd)co2 + (xa-xb+xc-xd)(si2) */ |
||
| 233 | //Yc12_out += Xc12C_out * si2; |
||
| 234 | p3 = Xc12C_out * si2; |
||
| 235 | /* xd' = (xa-yb-xc+yd)co3 - (ya+xb-yc-xd)(si3) */ |
||
| 236 | //Xd12_out -= Yd12C_out * si3; |
||
| 237 | p4 = Yd12C_out * si3; |
||
| 238 | /* yd' = (ya+xb-yc-xd)co3 + (xa-yb-xc+yd)(si3) */ |
||
| 239 | //Yd12_out += Xd12C_out * si3; |
||
| 240 | p5 = Xd12C_out * si3; |
||
| 241 | |||
| 242 | Xb12_out += p0; |
||
| 243 | Yb12_out -= p1; |
||
| 244 | Xc12_out += p2; |
||
| 245 | Yc12_out -= p3; |
||
| 246 | Xd12_out += p4; |
||
| 247 | Yd12_out -= p5; |
||
| 248 | |||
| 249 | /* xc' = (xa-xb+xc-xd)co2 + (ya-yb+yc-yd)(si2) */ |
||
| 250 | pSrc[2U * i1] = Xc12_out; |
||
| 251 | |||
| 252 | /* yc' = (ya-yb+yc-yd)co2 - (xa-xb+xc-xd)(si2) */ |
||
| 253 | pSrc[(2U * i1) + 1U] = Yc12_out; |
||
| 254 | |||
| 255 | /* xb' = (xa+yb-xc-yd)co1 + (ya-xb-yc+xd)(si1) */ |
||
| 256 | pSrc[2U * i2] = Xb12_out; |
||
| 257 | |||
| 258 | /* yb' = (ya-xb-yc+xd)co1 - (xa+yb-xc-yd)(si1) */ |
||
| 259 | pSrc[(2U * i2) + 1U] = Yb12_out; |
||
| 260 | |||
| 261 | /* xd' = (xa-yb-xc+yd)co3 + (ya+xb-yc-xd)(si3) */ |
||
| 262 | pSrc[2U * i3] = Xd12_out; |
||
| 263 | |||
| 264 | /* yd' = (ya+xb-yc-xd)co3 - (xa-yb-xc+yd)(si3) */ |
||
| 265 | pSrc[(2U * i3) + 1U] = Yd12_out; |
||
| 266 | |||
| 267 | /* Twiddle coefficients index modifier */ |
||
| 268 | ia1 += twidCoefModifier; |
||
| 269 | |||
| 270 | /* Updating input index */ |
||
| 271 | i0++; |
||
| 272 | |||
| 273 | } |
||
| 274 | while (--j); |
||
| 275 | |||
| 276 | twidCoefModifier <<= 2U; |
||
| 277 | |||
| 278 | /* Calculation of second stage to excluding last stage */ |
||
| 279 | for (k = fftLen >> 2U; k > 4U; k >>= 2U) |
||
| 280 | { |
||
| 281 | /* Initializations for the first stage */ |
||
| 282 | n1 = n2; |
||
| 283 | n2 >>= 2U; |
||
| 284 | ia1 = 0U; |
||
| 285 | |||
| 286 | /* Calculation of first stage */ |
||
| 287 | j = 0; |
||
| 288 | do |
||
| 289 | { |
||
| 290 | /* index calculation for the coefficients */ |
||
| 291 | ia2 = ia1 + ia1; |
||
| 292 | ia3 = ia2 + ia1; |
||
| 293 | co1 = pCoef[ia1 * 2U]; |
||
| 294 | si1 = pCoef[(ia1 * 2U) + 1U]; |
||
| 295 | co2 = pCoef[ia2 * 2U]; |
||
| 296 | si2 = pCoef[(ia2 * 2U) + 1U]; |
||
| 297 | co3 = pCoef[ia3 * 2U]; |
||
| 298 | si3 = pCoef[(ia3 * 2U) + 1U]; |
||
| 299 | |||
| 300 | /* Twiddle coefficients index modifier */ |
||
| 301 | ia1 += twidCoefModifier; |
||
| 302 | |||
| 303 | i0 = j; |
||
| 304 | do |
||
| 305 | { |
||
| 306 | /* index calculation for the input as, */ |
||
| 307 | /* pSrc[i0 + 0], pSrc[i0 + fftLen/4], pSrc[i0 + fftLen/2], pSrc[i0 + 3fftLen/4] */ |
||
| 308 | i1 = i0 + n2; |
||
| 309 | i2 = i1 + n2; |
||
| 310 | i3 = i2 + n2; |
||
| 311 | |||
| 312 | xaIn = pSrc[(2U * i0)]; |
||
| 313 | yaIn = pSrc[(2U * i0) + 1U]; |
||
| 314 | |||
| 315 | xbIn = pSrc[(2U * i1)]; |
||
| 316 | ybIn = pSrc[(2U * i1) + 1U]; |
||
| 317 | |||
| 318 | xcIn = pSrc[(2U * i2)]; |
||
| 319 | ycIn = pSrc[(2U * i2) + 1U]; |
||
| 320 | |||
| 321 | xdIn = pSrc[(2U * i3)]; |
||
| 322 | ydIn = pSrc[(2U * i3) + 1U]; |
||
| 323 | |||
| 324 | /* xa - xc */ |
||
| 325 | Xaminusc = xaIn - xcIn; |
||
| 326 | /* (xb - xd) */ |
||
| 327 | Xbminusd = xbIn - xdIn; |
||
| 328 | /* ya - yc */ |
||
| 329 | Yaminusc = yaIn - ycIn; |
||
| 330 | /* (yb - yd) */ |
||
| 331 | Ybminusd = ybIn - ydIn; |
||
| 332 | |||
| 333 | /* xa + xc */ |
||
| 334 | Xaplusc = xaIn + xcIn; |
||
| 335 | /* xb + xd */ |
||
| 336 | Xbplusd = xbIn + xdIn; |
||
| 337 | /* ya + yc */ |
||
| 338 | Yaplusc = yaIn + ycIn; |
||
| 339 | /* yb + yd */ |
||
| 340 | Ybplusd = ybIn + ydIn; |
||
| 341 | |||
| 342 | /* (xa - xc) + (yb - yd) */ |
||
| 343 | Xb12C_out = (Xaminusc + Ybminusd); |
||
| 344 | /* (ya - yc) - (xb - xd) */ |
||
| 345 | Yb12C_out = (Yaminusc - Xbminusd); |
||
| 346 | /* xa + xc -(xb + xd) */ |
||
| 347 | Xc12C_out = (Xaplusc - Xbplusd); |
||
| 348 | /* (ya + yc) - (yb + yd) */ |
||
| 349 | Yc12C_out = (Yaplusc - Ybplusd); |
||
| 350 | /* (xa - xc) - (yb - yd) */ |
||
| 351 | Xd12C_out = (Xaminusc - Ybminusd); |
||
| 352 | /* (ya - yc) + (xb - xd) */ |
||
| 353 | Yd12C_out = (Xbminusd + Yaminusc); |
||
| 354 | |||
| 355 | pSrc[(2U * i0)] = Xaplusc + Xbplusd; |
||
| 356 | pSrc[(2U * i0) + 1U] = Yaplusc + Ybplusd; |
||
| 357 | |||
| 358 | Xb12_out = Xb12C_out * co1; |
||
| 359 | Yb12_out = Yb12C_out * co1; |
||
| 360 | Xc12_out = Xc12C_out * co2; |
||
| 361 | Yc12_out = Yc12C_out * co2; |
||
| 362 | Xd12_out = Xd12C_out * co3; |
||
| 363 | Yd12_out = Yd12C_out * co3; |
||
| 364 | |||
| 365 | /* xb' = (xa+yb-xc-yd)co1 - (ya-xb-yc+xd)(si1) */ |
||
| 366 | //Xb12_out -= Yb12C_out * si1; |
||
| 367 | p0 = Yb12C_out * si1; |
||
| 368 | /* yb' = (ya-xb-yc+xd)co1 + (xa+yb-xc-yd)(si1) */ |
||
| 369 | //Yb12_out += Xb12C_out * si1; |
||
| 370 | p1 = Xb12C_out * si1; |
||
| 371 | /* xc' = (xa-xb+xc-xd)co2 - (ya-yb+yc-yd)(si2) */ |
||
| 372 | //Xc12_out -= Yc12C_out * si2; |
||
| 373 | p2 = Yc12C_out * si2; |
||
| 374 | /* yc' = (ya-yb+yc-yd)co2 + (xa-xb+xc-xd)(si2) */ |
||
| 375 | //Yc12_out += Xc12C_out * si2; |
||
| 376 | p3 = Xc12C_out * si2; |
||
| 377 | /* xd' = (xa-yb-xc+yd)co3 - (ya+xb-yc-xd)(si3) */ |
||
| 378 | //Xd12_out -= Yd12C_out * si3; |
||
| 379 | p4 = Yd12C_out * si3; |
||
| 380 | /* yd' = (ya+xb-yc-xd)co3 + (xa-yb-xc+yd)(si3) */ |
||
| 381 | //Yd12_out += Xd12C_out * si3; |
||
| 382 | p5 = Xd12C_out * si3; |
||
| 383 | |||
| 384 | Xb12_out += p0; |
||
| 385 | Yb12_out -= p1; |
||
| 386 | Xc12_out += p2; |
||
| 387 | Yc12_out -= p3; |
||
| 388 | Xd12_out += p4; |
||
| 389 | Yd12_out -= p5; |
||
| 390 | |||
| 391 | /* xc' = (xa-xb+xc-xd)co2 + (ya-yb+yc-yd)(si2) */ |
||
| 392 | pSrc[2U * i1] = Xc12_out; |
||
| 393 | |||
| 394 | /* yc' = (ya-yb+yc-yd)co2 - (xa-xb+xc-xd)(si2) */ |
||
| 395 | pSrc[(2U * i1) + 1U] = Yc12_out; |
||
| 396 | |||
| 397 | /* xb' = (xa+yb-xc-yd)co1 + (ya-xb-yc+xd)(si1) */ |
||
| 398 | pSrc[2U * i2] = Xb12_out; |
||
| 399 | |||
| 400 | /* yb' = (ya-xb-yc+xd)co1 - (xa+yb-xc-yd)(si1) */ |
||
| 401 | pSrc[(2U * i2) + 1U] = Yb12_out; |
||
| 402 | |||
| 403 | /* xd' = (xa-yb-xc+yd)co3 + (ya+xb-yc-xd)(si3) */ |
||
| 404 | pSrc[2U * i3] = Xd12_out; |
||
| 405 | |||
| 406 | /* yd' = (ya+xb-yc-xd)co3 - (xa-yb-xc+yd)(si3) */ |
||
| 407 | pSrc[(2U * i3) + 1U] = Yd12_out; |
||
| 408 | |||
| 409 | i0 += n1; |
||
| 410 | } while (i0 < fftLen); |
||
| 411 | j++; |
||
| 412 | } while (j <= (n2 - 1U)); |
||
| 413 | twidCoefModifier <<= 2U; |
||
| 414 | } |
||
| 415 | |||
| 416 | j = fftLen >> 2; |
||
| 417 | ptr1 = &pSrc[0]; |
||
| 418 | |||
| 419 | /* Calculations of last stage */ |
||
| 420 | do |
||
| 421 | { |
||
| 422 | xaIn = ptr1[0]; |
||
| 423 | yaIn = ptr1[1]; |
||
| 424 | xbIn = ptr1[2]; |
||
| 425 | ybIn = ptr1[3]; |
||
| 426 | xcIn = ptr1[4]; |
||
| 427 | ycIn = ptr1[5]; |
||
| 428 | xdIn = ptr1[6]; |
||
| 429 | ydIn = ptr1[7]; |
||
| 430 | |||
| 431 | /* xa + xc */ |
||
| 432 | Xaplusc = xaIn + xcIn; |
||
| 433 | |||
| 434 | /* xa - xc */ |
||
| 435 | Xaminusc = xaIn - xcIn; |
||
| 436 | |||
| 437 | /* ya + yc */ |
||
| 438 | Yaplusc = yaIn + ycIn; |
||
| 439 | |||
| 440 | /* ya - yc */ |
||
| 441 | Yaminusc = yaIn - ycIn; |
||
| 442 | |||
| 443 | /* xb + xd */ |
||
| 444 | Xbplusd = xbIn + xdIn; |
||
| 445 | |||
| 446 | /* yb + yd */ |
||
| 447 | Ybplusd = ybIn + ydIn; |
||
| 448 | |||
| 449 | /* (xb-xd) */ |
||
| 450 | Xbminusd = xbIn - xdIn; |
||
| 451 | |||
| 452 | /* (yb-yd) */ |
||
| 453 | Ybminusd = ybIn - ydIn; |
||
| 454 | |||
| 455 | /* xa' = xa + xb + xc + xd */ |
||
| 456 | a0 = (Xaplusc + Xbplusd); |
||
| 457 | /* ya' = ya + yb + yc + yd */ |
||
| 458 | a1 = (Yaplusc + Ybplusd); |
||
| 459 | /* xc' = (xa-xb+xc-xd) */ |
||
| 460 | a2 = (Xaplusc - Xbplusd); |
||
| 461 | /* yc' = (ya-yb+yc-yd) */ |
||
| 462 | a3 = (Yaplusc - Ybplusd); |
||
| 463 | /* xb' = (xa+yb-xc-yd) */ |
||
| 464 | a4 = (Xaminusc + Ybminusd); |
||
| 465 | /* yb' = (ya-xb-yc+xd) */ |
||
| 466 | a5 = (Yaminusc - Xbminusd); |
||
| 467 | /* xd' = (xa-yb-xc+yd)) */ |
||
| 468 | a6 = (Xaminusc - Ybminusd); |
||
| 469 | /* yd' = (ya+xb-yc-xd) */ |
||
| 470 | a7 = (Xbminusd + Yaminusc); |
||
| 471 | |||
| 472 | ptr1[0] = a0; |
||
| 473 | ptr1[1] = a1; |
||
| 474 | ptr1[2] = a2; |
||
| 475 | ptr1[3] = a3; |
||
| 476 | ptr1[4] = a4; |
||
| 477 | ptr1[5] = a5; |
||
| 478 | ptr1[6] = a6; |
||
| 479 | ptr1[7] = a7; |
||
| 480 | |||
| 481 | /* increment pointer by 8 */ |
||
| 482 | ptr1 += 8U; |
||
| 483 | } while (--j); |
||
| 484 | |||
| 485 | #else |
||
| 486 | |||
| 487 | float32_t t1, t2, r1, r2, s1, s2; |
||
| 488 | |||
| 489 | /* Run the below code for Cortex-M0 */ |
||
| 490 | |||
| 491 | /* Initializations for the fft calculation */ |
||
| 492 | n2 = fftLen; |
||
| 493 | n1 = n2; |
||
| 494 | for (k = fftLen; k > 1U; k >>= 2U) |
||
| 495 | { |
||
| 496 | /* Initializations for the fft calculation */ |
||
| 497 | n1 = n2; |
||
| 498 | n2 >>= 2U; |
||
| 499 | ia1 = 0U; |
||
| 500 | |||
| 501 | /* FFT Calculation */ |
||
| 502 | j = 0; |
||
| 503 | do |
||
| 504 | { |
||
| 505 | /* index calculation for the coefficients */ |
||
| 506 | ia2 = ia1 + ia1; |
||
| 507 | ia3 = ia2 + ia1; |
||
| 508 | co1 = pCoef[ia1 * 2U]; |
||
| 509 | si1 = pCoef[(ia1 * 2U) + 1U]; |
||
| 510 | co2 = pCoef[ia2 * 2U]; |
||
| 511 | si2 = pCoef[(ia2 * 2U) + 1U]; |
||
| 512 | co3 = pCoef[ia3 * 2U]; |
||
| 513 | si3 = pCoef[(ia3 * 2U) + 1U]; |
||
| 514 | |||
| 515 | /* Twiddle coefficients index modifier */ |
||
| 516 | ia1 = ia1 + twidCoefModifier; |
||
| 517 | |||
| 518 | i0 = j; |
||
| 519 | do |
||
| 520 | { |
||
| 521 | /* index calculation for the input as, */ |
||
| 522 | /* pSrc[i0 + 0], pSrc[i0 + fftLen/4], pSrc[i0 + fftLen/2], pSrc[i0 + 3fftLen/4] */ |
||
| 523 | i1 = i0 + n2; |
||
| 524 | i2 = i1 + n2; |
||
| 525 | i3 = i2 + n2; |
||
| 526 | |||
| 527 | /* xa + xc */ |
||
| 528 | r1 = pSrc[(2U * i0)] + pSrc[(2U * i2)]; |
||
| 529 | |||
| 530 | /* xa - xc */ |
||
| 531 | r2 = pSrc[(2U * i0)] - pSrc[(2U * i2)]; |
||
| 532 | |||
| 533 | /* ya + yc */ |
||
| 534 | s1 = pSrc[(2U * i0) + 1U] + pSrc[(2U * i2) + 1U]; |
||
| 535 | |||
| 536 | /* ya - yc */ |
||
| 537 | s2 = pSrc[(2U * i0) + 1U] - pSrc[(2U * i2) + 1U]; |
||
| 538 | |||
| 539 | /* xb + xd */ |
||
| 540 | t1 = pSrc[2U * i1] + pSrc[2U * i3]; |
||
| 541 | |||
| 542 | /* xa' = xa + xb + xc + xd */ |
||
| 543 | pSrc[2U * i0] = r1 + t1; |
||
| 544 | |||
| 545 | /* xa + xc -(xb + xd) */ |
||
| 546 | r1 = r1 - t1; |
||
| 547 | |||
| 548 | /* yb + yd */ |
||
| 549 | t2 = pSrc[(2U * i1) + 1U] + pSrc[(2U * i3) + 1U]; |
||
| 550 | |||
| 551 | /* ya' = ya + yb + yc + yd */ |
||
| 552 | pSrc[(2U * i0) + 1U] = s1 + t2; |
||
| 553 | |||
| 554 | /* (ya + yc) - (yb + yd) */ |
||
| 555 | s1 = s1 - t2; |
||
| 556 | |||
| 557 | /* (yb - yd) */ |
||
| 558 | t1 = pSrc[(2U * i1) + 1U] - pSrc[(2U * i3) + 1U]; |
||
| 559 | |||
| 560 | /* (xb - xd) */ |
||
| 561 | t2 = pSrc[2U * i1] - pSrc[2U * i3]; |
||
| 562 | |||
| 563 | /* xc' = (xa-xb+xc-xd)co2 + (ya-yb+yc-yd)(si2) */ |
||
| 564 | pSrc[2U * i1] = (r1 * co2) + (s1 * si2); |
||
| 565 | |||
| 566 | /* yc' = (ya-yb+yc-yd)co2 - (xa-xb+xc-xd)(si2) */ |
||
| 567 | pSrc[(2U * i1) + 1U] = (s1 * co2) - (r1 * si2); |
||
| 568 | |||
| 569 | /* (xa - xc) + (yb - yd) */ |
||
| 570 | r1 = r2 + t1; |
||
| 571 | |||
| 572 | /* (xa - xc) - (yb - yd) */ |
||
| 573 | r2 = r2 - t1; |
||
| 574 | |||
| 575 | /* (ya - yc) - (xb - xd) */ |
||
| 576 | s1 = s2 - t2; |
||
| 577 | |||
| 578 | /* (ya - yc) + (xb - xd) */ |
||
| 579 | s2 = s2 + t2; |
||
| 580 | |||
| 581 | /* xb' = (xa+yb-xc-yd)co1 + (ya-xb-yc+xd)(si1) */ |
||
| 582 | pSrc[2U * i2] = (r1 * co1) + (s1 * si1); |
||
| 583 | |||
| 584 | /* yb' = (ya-xb-yc+xd)co1 - (xa+yb-xc-yd)(si1) */ |
||
| 585 | pSrc[(2U * i2) + 1U] = (s1 * co1) - (r1 * si1); |
||
| 586 | |||
| 587 | /* xd' = (xa-yb-xc+yd)co3 + (ya+xb-yc-xd)(si3) */ |
||
| 588 | pSrc[2U * i3] = (r2 * co3) + (s2 * si3); |
||
| 589 | |||
| 590 | /* yd' = (ya+xb-yc-xd)co3 - (xa-yb-xc+yd)(si3) */ |
||
| 591 | pSrc[(2U * i3) + 1U] = (s2 * co3) - (r2 * si3); |
||
| 592 | |||
| 593 | i0 += n1; |
||
| 594 | } while ( i0 < fftLen); |
||
| 595 | j++; |
||
| 596 | } while (j <= (n2 - 1U)); |
||
| 597 | twidCoefModifier <<= 2U; |
||
| 598 | } |
||
| 599 | |||
| 600 | #endif /* #if defined (ARM_MATH_DSP) */ |
||
| 601 | |||
| 602 | } |
||
| 603 | |||
| 604 | /* |
||
| 605 | * @brief Core function for the floating-point CIFFT butterfly process. |
||
| 606 | * @param[in, out] *pSrc points to the in-place buffer of floating-point data type. |
||
| 607 | * @param[in] fftLen length of the FFT. |
||
| 608 | * @param[in] *pCoef points to twiddle coefficient buffer. |
||
| 609 | * @param[in] twidCoefModifier twiddle coefficient modifier that supports different size FFTs with the same twiddle factor table. |
||
| 610 | * @param[in] onebyfftLen value of 1/fftLen. |
||
| 611 | * @return none. |
||
| 612 | */ |
||
| 613 | |||
| 614 | void arm_radix4_butterfly_inverse_f32( |
||
| 615 | float32_t * pSrc, |
||
| 616 | uint16_t fftLen, |
||
| 617 | float32_t * pCoef, |
||
| 618 | uint16_t twidCoefModifier, |
||
| 619 | float32_t onebyfftLen) |
||
| 620 | { |
||
| 621 | float32_t co1, co2, co3, si1, si2, si3; |
||
| 622 | uint32_t ia1, ia2, ia3; |
||
| 623 | uint32_t i0, i1, i2, i3; |
||
| 624 | uint32_t n1, n2, j, k; |
||
| 625 | |||
| 626 | #if defined (ARM_MATH_DSP) |
||
| 627 | |||
| 628 | float32_t xaIn, yaIn, xbIn, ybIn, xcIn, ycIn, xdIn, ydIn; |
||
| 629 | float32_t Xaplusc, Xbplusd, Yaplusc, Ybplusd, Xaminusc, Xbminusd, Yaminusc, |
||
| 630 | Ybminusd; |
||
| 631 | float32_t Xb12C_out, Yb12C_out, Xc12C_out, Yc12C_out, Xd12C_out, Yd12C_out; |
||
| 632 | float32_t Xb12_out, Yb12_out, Xc12_out, Yc12_out, Xd12_out, Yd12_out; |
||
| 633 | float32_t *ptr1; |
||
| 634 | float32_t p0,p1,p2,p3,p4,p5,p6,p7; |
||
| 635 | float32_t a0,a1,a2,a3,a4,a5,a6,a7; |
||
| 636 | |||
| 637 | |||
| 638 | /* Initializations for the first stage */ |
||
| 639 | n2 = fftLen; |
||
| 640 | n1 = n2; |
||
| 641 | |||
| 642 | /* n2 = fftLen/4 */ |
||
| 643 | n2 >>= 2U; |
||
| 644 | i0 = 0U; |
||
| 645 | ia1 = 0U; |
||
| 646 | |||
| 647 | j = n2; |
||
| 648 | |||
| 649 | /* Calculation of first stage */ |
||
| 650 | do |
||
| 651 | { |
||
| 652 | /* index calculation for the input as, */ |
||
| 653 | /* pSrc[i0 + 0], pSrc[i0 + fftLen/4], pSrc[i0 + fftLen/2], pSrc[i0 + 3fftLen/4] */ |
||
| 654 | i1 = i0 + n2; |
||
| 655 | i2 = i1 + n2; |
||
| 656 | i3 = i2 + n2; |
||
| 657 | |||
| 658 | /* Butterfly implementation */ |
||
| 659 | xaIn = pSrc[(2U * i0)]; |
||
| 660 | yaIn = pSrc[(2U * i0) + 1U]; |
||
| 661 | |||
| 662 | xcIn = pSrc[(2U * i2)]; |
||
| 663 | ycIn = pSrc[(2U * i2) + 1U]; |
||
| 664 | |||
| 665 | xbIn = pSrc[(2U * i1)]; |
||
| 666 | ybIn = pSrc[(2U * i1) + 1U]; |
||
| 667 | |||
| 668 | xdIn = pSrc[(2U * i3)]; |
||
| 669 | ydIn = pSrc[(2U * i3) + 1U]; |
||
| 670 | |||
| 671 | /* xa + xc */ |
||
| 672 | Xaplusc = xaIn + xcIn; |
||
| 673 | /* xb + xd */ |
||
| 674 | Xbplusd = xbIn + xdIn; |
||
| 675 | /* ya + yc */ |
||
| 676 | Yaplusc = yaIn + ycIn; |
||
| 677 | /* yb + yd */ |
||
| 678 | Ybplusd = ybIn + ydIn; |
||
| 679 | |||
| 680 | /* index calculation for the coefficients */ |
||
| 681 | ia2 = ia1 + ia1; |
||
| 682 | co2 = pCoef[ia2 * 2U]; |
||
| 683 | si2 = pCoef[(ia2 * 2U) + 1U]; |
||
| 684 | |||
| 685 | /* xa - xc */ |
||
| 686 | Xaminusc = xaIn - xcIn; |
||
| 687 | /* xb - xd */ |
||
| 688 | Xbminusd = xbIn - xdIn; |
||
| 689 | /* ya - yc */ |
||
| 690 | Yaminusc = yaIn - ycIn; |
||
| 691 | /* yb - yd */ |
||
| 692 | Ybminusd = ybIn - ydIn; |
||
| 693 | |||
| 694 | /* xa' = xa + xb + xc + xd */ |
||
| 695 | pSrc[(2U * i0)] = Xaplusc + Xbplusd; |
||
| 696 | |||
| 697 | /* ya' = ya + yb + yc + yd */ |
||
| 698 | pSrc[(2U * i0) + 1U] = Yaplusc + Ybplusd; |
||
| 699 | |||
| 700 | /* (xa - xc) - (yb - yd) */ |
||
| 701 | Xb12C_out = (Xaminusc - Ybminusd); |
||
| 702 | /* (ya - yc) + (xb - xd) */ |
||
| 703 | Yb12C_out = (Yaminusc + Xbminusd); |
||
| 704 | /* (xa + xc) - (xb + xd) */ |
||
| 705 | Xc12C_out = (Xaplusc - Xbplusd); |
||
| 706 | /* (ya + yc) - (yb + yd) */ |
||
| 707 | Yc12C_out = (Yaplusc - Ybplusd); |
||
| 708 | /* (xa - xc) + (yb - yd) */ |
||
| 709 | Xd12C_out = (Xaminusc + Ybminusd); |
||
| 710 | /* (ya - yc) - (xb - xd) */ |
||
| 711 | Yd12C_out = (Yaminusc - Xbminusd); |
||
| 712 | |||
| 713 | co1 = pCoef[ia1 * 2U]; |
||
| 714 | si1 = pCoef[(ia1 * 2U) + 1U]; |
||
| 715 | |||
| 716 | /* index calculation for the coefficients */ |
||
| 717 | ia3 = ia2 + ia1; |
||
| 718 | co3 = pCoef[ia3 * 2U]; |
||
| 719 | si3 = pCoef[(ia3 * 2U) + 1U]; |
||
| 720 | |||
| 721 | Xb12_out = Xb12C_out * co1; |
||
| 722 | Yb12_out = Yb12C_out * co1; |
||
| 723 | Xc12_out = Xc12C_out * co2; |
||
| 724 | Yc12_out = Yc12C_out * co2; |
||
| 725 | Xd12_out = Xd12C_out * co3; |
||
| 726 | Yd12_out = Yd12C_out * co3; |
||
| 727 | |||
| 728 | /* xb' = (xa+yb-xc-yd)co1 - (ya-xb-yc+xd)(si1) */ |
||
| 729 | //Xb12_out -= Yb12C_out * si1; |
||
| 730 | p0 = Yb12C_out * si1; |
||
| 731 | /* yb' = (ya-xb-yc+xd)co1 + (xa+yb-xc-yd)(si1) */ |
||
| 732 | //Yb12_out += Xb12C_out * si1; |
||
| 733 | p1 = Xb12C_out * si1; |
||
| 734 | /* xc' = (xa-xb+xc-xd)co2 - (ya-yb+yc-yd)(si2) */ |
||
| 735 | //Xc12_out -= Yc12C_out * si2; |
||
| 736 | p2 = Yc12C_out * si2; |
||
| 737 | /* yc' = (ya-yb+yc-yd)co2 + (xa-xb+xc-xd)(si2) */ |
||
| 738 | //Yc12_out += Xc12C_out * si2; |
||
| 739 | p3 = Xc12C_out * si2; |
||
| 740 | /* xd' = (xa-yb-xc+yd)co3 - (ya+xb-yc-xd)(si3) */ |
||
| 741 | //Xd12_out -= Yd12C_out * si3; |
||
| 742 | p4 = Yd12C_out * si3; |
||
| 743 | /* yd' = (ya+xb-yc-xd)co3 + (xa-yb-xc+yd)(si3) */ |
||
| 744 | //Yd12_out += Xd12C_out * si3; |
||
| 745 | p5 = Xd12C_out * si3; |
||
| 746 | |||
| 747 | Xb12_out -= p0; |
||
| 748 | Yb12_out += p1; |
||
| 749 | Xc12_out -= p2; |
||
| 750 | Yc12_out += p3; |
||
| 751 | Xd12_out -= p4; |
||
| 752 | Yd12_out += p5; |
||
| 753 | |||
| 754 | /* xc' = (xa-xb+xc-xd)co2 - (ya-yb+yc-yd)(si2) */ |
||
| 755 | pSrc[2U * i1] = Xc12_out; |
||
| 756 | |||
| 757 | /* yc' = (ya-yb+yc-yd)co2 + (xa-xb+xc-xd)(si2) */ |
||
| 758 | pSrc[(2U * i1) + 1U] = Yc12_out; |
||
| 759 | |||
| 760 | /* xb' = (xa+yb-xc-yd)co1 - (ya-xb-yc+xd)(si1) */ |
||
| 761 | pSrc[2U * i2] = Xb12_out; |
||
| 762 | |||
| 763 | /* yb' = (ya-xb-yc+xd)co1 + (xa+yb-xc-yd)(si1) */ |
||
| 764 | pSrc[(2U * i2) + 1U] = Yb12_out; |
||
| 765 | |||
| 766 | /* xd' = (xa-yb-xc+yd)co3 - (ya+xb-yc-xd)(si3) */ |
||
| 767 | pSrc[2U * i3] = Xd12_out; |
||
| 768 | |||
| 769 | /* yd' = (ya+xb-yc-xd)co3 + (xa-yb-xc+yd)(si3) */ |
||
| 770 | pSrc[(2U * i3) + 1U] = Yd12_out; |
||
| 771 | |||
| 772 | /* Twiddle coefficients index modifier */ |
||
| 773 | ia1 = ia1 + twidCoefModifier; |
||
| 774 | |||
| 775 | /* Updating input index */ |
||
| 776 | i0 = i0 + 1U; |
||
| 777 | |||
| 778 | } while (--j); |
||
| 779 | |||
| 780 | twidCoefModifier <<= 2U; |
||
| 781 | |||
| 782 | /* Calculation of second stage to excluding last stage */ |
||
| 783 | for (k = fftLen >> 2U; k > 4U; k >>= 2U) |
||
| 784 | { |
||
| 785 | /* Initializations for the first stage */ |
||
| 786 | n1 = n2; |
||
| 787 | n2 >>= 2U; |
||
| 788 | ia1 = 0U; |
||
| 789 | |||
| 790 | /* Calculation of first stage */ |
||
| 791 | j = 0; |
||
| 792 | do |
||
| 793 | { |
||
| 794 | /* index calculation for the coefficients */ |
||
| 795 | ia2 = ia1 + ia1; |
||
| 796 | ia3 = ia2 + ia1; |
||
| 797 | co1 = pCoef[ia1 * 2U]; |
||
| 798 | si1 = pCoef[(ia1 * 2U) + 1U]; |
||
| 799 | co2 = pCoef[ia2 * 2U]; |
||
| 800 | si2 = pCoef[(ia2 * 2U) + 1U]; |
||
| 801 | co3 = pCoef[ia3 * 2U]; |
||
| 802 | si3 = pCoef[(ia3 * 2U) + 1U]; |
||
| 803 | |||
| 804 | /* Twiddle coefficients index modifier */ |
||
| 805 | ia1 = ia1 + twidCoefModifier; |
||
| 806 | |||
| 807 | i0 = j; |
||
| 808 | do |
||
| 809 | { |
||
| 810 | /* index calculation for the input as, */ |
||
| 811 | /* pSrc[i0 + 0], pSrc[i0 + fftLen/4], pSrc[i0 + fftLen/2], pSrc[i0 + 3fftLen/4] */ |
||
| 812 | i1 = i0 + n2; |
||
| 813 | i2 = i1 + n2; |
||
| 814 | i3 = i2 + n2; |
||
| 815 | |||
| 816 | xaIn = pSrc[(2U * i0)]; |
||
| 817 | yaIn = pSrc[(2U * i0) + 1U]; |
||
| 818 | |||
| 819 | xbIn = pSrc[(2U * i1)]; |
||
| 820 | ybIn = pSrc[(2U * i1) + 1U]; |
||
| 821 | |||
| 822 | xcIn = pSrc[(2U * i2)]; |
||
| 823 | ycIn = pSrc[(2U * i2) + 1U]; |
||
| 824 | |||
| 825 | xdIn = pSrc[(2U * i3)]; |
||
| 826 | ydIn = pSrc[(2U * i3) + 1U]; |
||
| 827 | |||
| 828 | /* xa - xc */ |
||
| 829 | Xaminusc = xaIn - xcIn; |
||
| 830 | /* (xb - xd) */ |
||
| 831 | Xbminusd = xbIn - xdIn; |
||
| 832 | /* ya - yc */ |
||
| 833 | Yaminusc = yaIn - ycIn; |
||
| 834 | /* (yb - yd) */ |
||
| 835 | Ybminusd = ybIn - ydIn; |
||
| 836 | |||
| 837 | /* xa + xc */ |
||
| 838 | Xaplusc = xaIn + xcIn; |
||
| 839 | /* xb + xd */ |
||
| 840 | Xbplusd = xbIn + xdIn; |
||
| 841 | /* ya + yc */ |
||
| 842 | Yaplusc = yaIn + ycIn; |
||
| 843 | /* yb + yd */ |
||
| 844 | Ybplusd = ybIn + ydIn; |
||
| 845 | |||
| 846 | /* (xa - xc) - (yb - yd) */ |
||
| 847 | Xb12C_out = (Xaminusc - Ybminusd); |
||
| 848 | /* (ya - yc) + (xb - xd) */ |
||
| 849 | Yb12C_out = (Yaminusc + Xbminusd); |
||
| 850 | /* xa + xc -(xb + xd) */ |
||
| 851 | Xc12C_out = (Xaplusc - Xbplusd); |
||
| 852 | /* (ya + yc) - (yb + yd) */ |
||
| 853 | Yc12C_out = (Yaplusc - Ybplusd); |
||
| 854 | /* (xa - xc) + (yb - yd) */ |
||
| 855 | Xd12C_out = (Xaminusc + Ybminusd); |
||
| 856 | /* (ya - yc) - (xb - xd) */ |
||
| 857 | Yd12C_out = (Yaminusc - Xbminusd); |
||
| 858 | |||
| 859 | pSrc[(2U * i0)] = Xaplusc + Xbplusd; |
||
| 860 | pSrc[(2U * i0) + 1U] = Yaplusc + Ybplusd; |
||
| 861 | |||
| 862 | Xb12_out = Xb12C_out * co1; |
||
| 863 | Yb12_out = Yb12C_out * co1; |
||
| 864 | Xc12_out = Xc12C_out * co2; |
||
| 865 | Yc12_out = Yc12C_out * co2; |
||
| 866 | Xd12_out = Xd12C_out * co3; |
||
| 867 | Yd12_out = Yd12C_out * co3; |
||
| 868 | |||
| 869 | /* xb' = (xa+yb-xc-yd)co1 - (ya-xb-yc+xd)(si1) */ |
||
| 870 | //Xb12_out -= Yb12C_out * si1; |
||
| 871 | p0 = Yb12C_out * si1; |
||
| 872 | /* yb' = (ya-xb-yc+xd)co1 + (xa+yb-xc-yd)(si1) */ |
||
| 873 | //Yb12_out += Xb12C_out * si1; |
||
| 874 | p1 = Xb12C_out * si1; |
||
| 875 | /* xc' = (xa-xb+xc-xd)co2 - (ya-yb+yc-yd)(si2) */ |
||
| 876 | //Xc12_out -= Yc12C_out * si2; |
||
| 877 | p2 = Yc12C_out * si2; |
||
| 878 | /* yc' = (ya-yb+yc-yd)co2 + (xa-xb+xc-xd)(si2) */ |
||
| 879 | //Yc12_out += Xc12C_out * si2; |
||
| 880 | p3 = Xc12C_out * si2; |
||
| 881 | /* xd' = (xa-yb-xc+yd)co3 - (ya+xb-yc-xd)(si3) */ |
||
| 882 | //Xd12_out -= Yd12C_out * si3; |
||
| 883 | p4 = Yd12C_out * si3; |
||
| 884 | /* yd' = (ya+xb-yc-xd)co3 + (xa-yb-xc+yd)(si3) */ |
||
| 885 | //Yd12_out += Xd12C_out * si3; |
||
| 886 | p5 = Xd12C_out * si3; |
||
| 887 | |||
| 888 | Xb12_out -= p0; |
||
| 889 | Yb12_out += p1; |
||
| 890 | Xc12_out -= p2; |
||
| 891 | Yc12_out += p3; |
||
| 892 | Xd12_out -= p4; |
||
| 893 | Yd12_out += p5; |
||
| 894 | |||
| 895 | /* xc' = (xa-xb+xc-xd)co2 - (ya-yb+yc-yd)(si2) */ |
||
| 896 | pSrc[2U * i1] = Xc12_out; |
||
| 897 | |||
| 898 | /* yc' = (ya-yb+yc-yd)co2 + (xa-xb+xc-xd)(si2) */ |
||
| 899 | pSrc[(2U * i1) + 1U] = Yc12_out; |
||
| 900 | |||
| 901 | /* xb' = (xa+yb-xc-yd)co1 - (ya-xb-yc+xd)(si1) */ |
||
| 902 | pSrc[2U * i2] = Xb12_out; |
||
| 903 | |||
| 904 | /* yb' = (ya-xb-yc+xd)co1 + (xa+yb-xc-yd)(si1) */ |
||
| 905 | pSrc[(2U * i2) + 1U] = Yb12_out; |
||
| 906 | |||
| 907 | /* xd' = (xa-yb-xc+yd)co3 - (ya+xb-yc-xd)(si3) */ |
||
| 908 | pSrc[2U * i3] = Xd12_out; |
||
| 909 | |||
| 910 | /* yd' = (ya+xb-yc-xd)co3 + (xa-yb-xc+yd)(si3) */ |
||
| 911 | pSrc[(2U * i3) + 1U] = Yd12_out; |
||
| 912 | |||
| 913 | i0 += n1; |
||
| 914 | } while (i0 < fftLen); |
||
| 915 | j++; |
||
| 916 | } while (j <= (n2 - 1U)); |
||
| 917 | twidCoefModifier <<= 2U; |
||
| 918 | } |
||
| 919 | /* Initializations of last stage */ |
||
| 920 | |||
| 921 | j = fftLen >> 2; |
||
| 922 | ptr1 = &pSrc[0]; |
||
| 923 | |||
| 924 | /* Calculations of last stage */ |
||
| 925 | do |
||
| 926 | { |
||
| 927 | xaIn = ptr1[0]; |
||
| 928 | yaIn = ptr1[1]; |
||
| 929 | xbIn = ptr1[2]; |
||
| 930 | ybIn = ptr1[3]; |
||
| 931 | xcIn = ptr1[4]; |
||
| 932 | ycIn = ptr1[5]; |
||
| 933 | xdIn = ptr1[6]; |
||
| 934 | ydIn = ptr1[7]; |
||
| 935 | |||
| 936 | /* Butterfly implementation */ |
||
| 937 | /* xa + xc */ |
||
| 938 | Xaplusc = xaIn + xcIn; |
||
| 939 | |||
| 940 | /* xa - xc */ |
||
| 941 | Xaminusc = xaIn - xcIn; |
||
| 942 | |||
| 943 | /* ya + yc */ |
||
| 944 | Yaplusc = yaIn + ycIn; |
||
| 945 | |||
| 946 | /* ya - yc */ |
||
| 947 | Yaminusc = yaIn - ycIn; |
||
| 948 | |||
| 949 | /* xb + xd */ |
||
| 950 | Xbplusd = xbIn + xdIn; |
||
| 951 | |||
| 952 | /* yb + yd */ |
||
| 953 | Ybplusd = ybIn + ydIn; |
||
| 954 | |||
| 955 | /* (xb-xd) */ |
||
| 956 | Xbminusd = xbIn - xdIn; |
||
| 957 | |||
| 958 | /* (yb-yd) */ |
||
| 959 | Ybminusd = ybIn - ydIn; |
||
| 960 | |||
| 961 | /* xa' = (xa+xb+xc+xd) * onebyfftLen */ |
||
| 962 | a0 = (Xaplusc + Xbplusd); |
||
| 963 | /* ya' = (ya+yb+yc+yd) * onebyfftLen */ |
||
| 964 | a1 = (Yaplusc + Ybplusd); |
||
| 965 | /* xc' = (xa-xb+xc-xd) * onebyfftLen */ |
||
| 966 | a2 = (Xaplusc - Xbplusd); |
||
| 967 | /* yc' = (ya-yb+yc-yd) * onebyfftLen */ |
||
| 968 | a3 = (Yaplusc - Ybplusd); |
||
| 969 | /* xb' = (xa-yb-xc+yd) * onebyfftLen */ |
||
| 970 | a4 = (Xaminusc - Ybminusd); |
||
| 971 | /* yb' = (ya+xb-yc-xd) * onebyfftLen */ |
||
| 972 | a5 = (Yaminusc + Xbminusd); |
||
| 973 | /* xd' = (xa-yb-xc+yd) * onebyfftLen */ |
||
| 974 | a6 = (Xaminusc + Ybminusd); |
||
| 975 | /* yd' = (ya-xb-yc+xd) * onebyfftLen */ |
||
| 976 | a7 = (Yaminusc - Xbminusd); |
||
| 977 | |||
| 978 | p0 = a0 * onebyfftLen; |
||
| 979 | p1 = a1 * onebyfftLen; |
||
| 980 | p2 = a2 * onebyfftLen; |
||
| 981 | p3 = a3 * onebyfftLen; |
||
| 982 | p4 = a4 * onebyfftLen; |
||
| 983 | p5 = a5 * onebyfftLen; |
||
| 984 | p6 = a6 * onebyfftLen; |
||
| 985 | p7 = a7 * onebyfftLen; |
||
| 986 | |||
| 987 | /* xa' = (xa+xb+xc+xd) * onebyfftLen */ |
||
| 988 | ptr1[0] = p0; |
||
| 989 | /* ya' = (ya+yb+yc+yd) * onebyfftLen */ |
||
| 990 | ptr1[1] = p1; |
||
| 991 | /* xc' = (xa-xb+xc-xd) * onebyfftLen */ |
||
| 992 | ptr1[2] = p2; |
||
| 993 | /* yc' = (ya-yb+yc-yd) * onebyfftLen */ |
||
| 994 | ptr1[3] = p3; |
||
| 995 | /* xb' = (xa-yb-xc+yd) * onebyfftLen */ |
||
| 996 | ptr1[4] = p4; |
||
| 997 | /* yb' = (ya+xb-yc-xd) * onebyfftLen */ |
||
| 998 | ptr1[5] = p5; |
||
| 999 | /* xd' = (xa-yb-xc+yd) * onebyfftLen */ |
||
| 1000 | ptr1[6] = p6; |
||
| 1001 | /* yd' = (ya-xb-yc+xd) * onebyfftLen */ |
||
| 1002 | ptr1[7] = p7; |
||
| 1003 | |||
| 1004 | /* increment source pointer by 8 for next calculations */ |
||
| 1005 | ptr1 = ptr1 + 8U; |
||
| 1006 | |||
| 1007 | } while (--j); |
||
| 1008 | |||
| 1009 | #else |
||
| 1010 | |||
| 1011 | float32_t t1, t2, r1, r2, s1, s2; |
||
| 1012 | |||
| 1013 | /* Run the below code for Cortex-M0 */ |
||
| 1014 | |||
| 1015 | /* Initializations for the first stage */ |
||
| 1016 | n2 = fftLen; |
||
| 1017 | n1 = n2; |
||
| 1018 | |||
| 1019 | /* Calculation of first stage */ |
||
| 1020 | for (k = fftLen; k > 4U; k >>= 2U) |
||
| 1021 | { |
||
| 1022 | /* Initializations for the first stage */ |
||
| 1023 | n1 = n2; |
||
| 1024 | n2 >>= 2U; |
||
| 1025 | ia1 = 0U; |
||
| 1026 | |||
| 1027 | /* Calculation of first stage */ |
||
| 1028 | j = 0; |
||
| 1029 | do |
||
| 1030 | { |
||
| 1031 | /* index calculation for the coefficients */ |
||
| 1032 | ia2 = ia1 + ia1; |
||
| 1033 | ia3 = ia2 + ia1; |
||
| 1034 | co1 = pCoef[ia1 * 2U]; |
||
| 1035 | si1 = pCoef[(ia1 * 2U) + 1U]; |
||
| 1036 | co2 = pCoef[ia2 * 2U]; |
||
| 1037 | si2 = pCoef[(ia2 * 2U) + 1U]; |
||
| 1038 | co3 = pCoef[ia3 * 2U]; |
||
| 1039 | si3 = pCoef[(ia3 * 2U) + 1U]; |
||
| 1040 | |||
| 1041 | /* Twiddle coefficients index modifier */ |
||
| 1042 | ia1 = ia1 + twidCoefModifier; |
||
| 1043 | |||
| 1044 | i0 = j; |
||
| 1045 | do |
||
| 1046 | { |
||
| 1047 | /* index calculation for the input as, */ |
||
| 1048 | /* pSrc[i0 + 0], pSrc[i0 + fftLen/4], pSrc[i0 + fftLen/2], pSrc[i0 + 3fftLen/4] */ |
||
| 1049 | i1 = i0 + n2; |
||
| 1050 | i2 = i1 + n2; |
||
| 1051 | i3 = i2 + n2; |
||
| 1052 | |||
| 1053 | /* xa + xc */ |
||
| 1054 | r1 = pSrc[(2U * i0)] + pSrc[(2U * i2)]; |
||
| 1055 | |||
| 1056 | /* xa - xc */ |
||
| 1057 | r2 = pSrc[(2U * i0)] - pSrc[(2U * i2)]; |
||
| 1058 | |||
| 1059 | /* ya + yc */ |
||
| 1060 | s1 = pSrc[(2U * i0) + 1U] + pSrc[(2U * i2) + 1U]; |
||
| 1061 | |||
| 1062 | /* ya - yc */ |
||
| 1063 | s2 = pSrc[(2U * i0) + 1U] - pSrc[(2U * i2) + 1U]; |
||
| 1064 | |||
| 1065 | /* xb + xd */ |
||
| 1066 | t1 = pSrc[2U * i1] + pSrc[2U * i3]; |
||
| 1067 | |||
| 1068 | /* xa' = xa + xb + xc + xd */ |
||
| 1069 | pSrc[2U * i0] = r1 + t1; |
||
| 1070 | |||
| 1071 | /* xa + xc -(xb + xd) */ |
||
| 1072 | r1 = r1 - t1; |
||
| 1073 | |||
| 1074 | /* yb + yd */ |
||
| 1075 | t2 = pSrc[(2U * i1) + 1U] + pSrc[(2U * i3) + 1U]; |
||
| 1076 | |||
| 1077 | /* ya' = ya + yb + yc + yd */ |
||
| 1078 | pSrc[(2U * i0) + 1U] = s1 + t2; |
||
| 1079 | |||
| 1080 | /* (ya + yc) - (yb + yd) */ |
||
| 1081 | s1 = s1 - t2; |
||
| 1082 | |||
| 1083 | /* (yb - yd) */ |
||
| 1084 | t1 = pSrc[(2U * i1) + 1U] - pSrc[(2U * i3) + 1U]; |
||
| 1085 | |||
| 1086 | /* (xb - xd) */ |
||
| 1087 | t2 = pSrc[2U * i1] - pSrc[2U * i3]; |
||
| 1088 | |||
| 1089 | /* xc' = (xa-xb+xc-xd)co2 - (ya-yb+yc-yd)(si2) */ |
||
| 1090 | pSrc[2U * i1] = (r1 * co2) - (s1 * si2); |
||
| 1091 | |||
| 1092 | /* yc' = (ya-yb+yc-yd)co2 + (xa-xb+xc-xd)(si2) */ |
||
| 1093 | pSrc[(2U * i1) + 1U] = (s1 * co2) + (r1 * si2); |
||
| 1094 | |||
| 1095 | /* (xa - xc) - (yb - yd) */ |
||
| 1096 | r1 = r2 - t1; |
||
| 1097 | |||
| 1098 | /* (xa - xc) + (yb - yd) */ |
||
| 1099 | r2 = r2 + t1; |
||
| 1100 | |||
| 1101 | /* (ya - yc) + (xb - xd) */ |
||
| 1102 | s1 = s2 + t2; |
||
| 1103 | |||
| 1104 | /* (ya - yc) - (xb - xd) */ |
||
| 1105 | s2 = s2 - t2; |
||
| 1106 | |||
| 1107 | /* xb' = (xa+yb-xc-yd)co1 - (ya-xb-yc+xd)(si1) */ |
||
| 1108 | pSrc[2U * i2] = (r1 * co1) - (s1 * si1); |
||
| 1109 | |||
| 1110 | /* yb' = (ya-xb-yc+xd)co1 + (xa+yb-xc-yd)(si1) */ |
||
| 1111 | pSrc[(2U * i2) + 1U] = (s1 * co1) + (r1 * si1); |
||
| 1112 | |||
| 1113 | /* xd' = (xa-yb-xc+yd)co3 - (ya+xb-yc-xd)(si3) */ |
||
| 1114 | pSrc[2U * i3] = (r2 * co3) - (s2 * si3); |
||
| 1115 | |||
| 1116 | /* yd' = (ya+xb-yc-xd)co3 + (xa-yb-xc+yd)(si3) */ |
||
| 1117 | pSrc[(2U * i3) + 1U] = (s2 * co3) + (r2 * si3); |
||
| 1118 | |||
| 1119 | i0 += n1; |
||
| 1120 | } while ( i0 < fftLen); |
||
| 1121 | j++; |
||
| 1122 | } while (j <= (n2 - 1U)); |
||
| 1123 | twidCoefModifier <<= 2U; |
||
| 1124 | } |
||
| 1125 | /* Initializations of last stage */ |
||
| 1126 | n1 = n2; |
||
| 1127 | n2 >>= 2U; |
||
| 1128 | |||
| 1129 | /* Calculations of last stage */ |
||
| 1130 | for (i0 = 0U; i0 <= (fftLen - n1); i0 += n1) |
||
| 1131 | { |
||
| 1132 | /* index calculation for the input as, */ |
||
| 1133 | /* pSrc[i0 + 0], pSrc[i0 + fftLen/4], pSrc[i0 + fftLen/2], pSrc[i0 + 3fftLen/4] */ |
||
| 1134 | i1 = i0 + n2; |
||
| 1135 | i2 = i1 + n2; |
||
| 1136 | i3 = i2 + n2; |
||
| 1137 | |||
| 1138 | /* Butterfly implementation */ |
||
| 1139 | /* xa + xc */ |
||
| 1140 | r1 = pSrc[2U * i0] + pSrc[2U * i2]; |
||
| 1141 | |||
| 1142 | /* xa - xc */ |
||
| 1143 | r2 = pSrc[2U * i0] - pSrc[2U * i2]; |
||
| 1144 | |||
| 1145 | /* ya + yc */ |
||
| 1146 | s1 = pSrc[(2U * i0) + 1U] + pSrc[(2U * i2) + 1U]; |
||
| 1147 | |||
| 1148 | /* ya - yc */ |
||
| 1149 | s2 = pSrc[(2U * i0) + 1U] - pSrc[(2U * i2) + 1U]; |
||
| 1150 | |||
| 1151 | /* xc + xd */ |
||
| 1152 | t1 = pSrc[2U * i1] + pSrc[2U * i3]; |
||
| 1153 | |||
| 1154 | /* xa' = xa + xb + xc + xd */ |
||
| 1155 | pSrc[2U * i0] = (r1 + t1) * onebyfftLen; |
||
| 1156 | |||
| 1157 | /* (xa + xb) - (xc + xd) */ |
||
| 1158 | r1 = r1 - t1; |
||
| 1159 | |||
| 1160 | /* yb + yd */ |
||
| 1161 | t2 = pSrc[(2U * i1) + 1U] + pSrc[(2U * i3) + 1U]; |
||
| 1162 | |||
| 1163 | /* ya' = ya + yb + yc + yd */ |
||
| 1164 | pSrc[(2U * i0) + 1U] = (s1 + t2) * onebyfftLen; |
||
| 1165 | |||
| 1166 | /* (ya + yc) - (yb + yd) */ |
||
| 1167 | s1 = s1 - t2; |
||
| 1168 | |||
| 1169 | /* (yb-yd) */ |
||
| 1170 | t1 = pSrc[(2U * i1) + 1U] - pSrc[(2U * i3) + 1U]; |
||
| 1171 | |||
| 1172 | /* (xb-xd) */ |
||
| 1173 | t2 = pSrc[2U * i1] - pSrc[2U * i3]; |
||
| 1174 | |||
| 1175 | /* xc' = (xa-xb+xc-xd)co2 - (ya-yb+yc-yd)(si2) */ |
||
| 1176 | pSrc[2U * i1] = r1 * onebyfftLen; |
||
| 1177 | |||
| 1178 | /* yc' = (ya-yb+yc-yd)co2 + (xa-xb+xc-xd)(si2) */ |
||
| 1179 | pSrc[(2U * i1) + 1U] = s1 * onebyfftLen; |
||
| 1180 | |||
| 1181 | /* (xa - xc) - (yb-yd) */ |
||
| 1182 | r1 = r2 - t1; |
||
| 1183 | |||
| 1184 | /* (xa - xc) + (yb-yd) */ |
||
| 1185 | r2 = r2 + t1; |
||
| 1186 | |||
| 1187 | /* (ya - yc) + (xb-xd) */ |
||
| 1188 | s1 = s2 + t2; |
||
| 1189 | |||
| 1190 | /* (ya - yc) - (xb-xd) */ |
||
| 1191 | s2 = s2 - t2; |
||
| 1192 | |||
| 1193 | /* xb' = (xa+yb-xc-yd)co1 - (ya-xb-yc+xd)(si1) */ |
||
| 1194 | pSrc[2U * i2] = r1 * onebyfftLen; |
||
| 1195 | |||
| 1196 | /* yb' = (ya-xb-yc+xd)co1 + (xa+yb-xc-yd)(si1) */ |
||
| 1197 | pSrc[(2U * i2) + 1U] = s1 * onebyfftLen; |
||
| 1198 | |||
| 1199 | /* xd' = (xa-yb-xc+yd)co3 - (ya+xb-yc-xd)(si3) */ |
||
| 1200 | pSrc[2U * i3] = r2 * onebyfftLen; |
||
| 1201 | |||
| 1202 | /* yd' = (ya+xb-yc-xd)co3 + (xa-yb-xc+yd)(si3) */ |
||
| 1203 | pSrc[(2U * i3) + 1U] = s2 * onebyfftLen; |
||
| 1204 | } |
||
| 1205 | |||
| 1206 | #endif /* #if defined (ARM_MATH_DSP) */ |
||
| 1207 | } |
||
| 1208 | |||
| 1209 |