1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_cfft_q15.c
4 * Description: Combined Radix Decimation in Q15 Frequency CFFT processing function
5 *
6 * $Date: 23 April 2021
7 * $Revision: V1.9.0
8 *
9 * Target Processor: Cortex-M and Cortex-A cores
10 * -------------------------------------------------------------------- */
11 /*
12 * Copyright (C) 2010-2021 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 "dsp/transform_functions.h"
30
31 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
32
33 #include "arm_vec_fft.h"
34
35
_arm_radix4_butterfly_q15_mve(const arm_cfft_instance_q15 * S,q15_t * pSrc,uint32_t fftLen)36 static void _arm_radix4_butterfly_q15_mve(
37 const arm_cfft_instance_q15 * S,
38 q15_t *pSrc,
39 uint32_t fftLen)
40 {
41 q15x8_t vecTmp0, vecTmp1;
42 q15x8_t vecSum0, vecDiff0, vecSum1, vecDiff1;
43 q15x8_t vecA, vecB, vecC, vecD;
44 uint32_t blkCnt;
45 uint32_t n1, n2;
46 uint32_t stage = 0;
47 int32_t iter = 1;
48 static const int32_t strides[4] = {
49 (0 - 16) * (int32_t)sizeof(q15_t *), (4 - 16) * (int32_t)sizeof(q15_t *),
50 (8 - 16) * (int32_t)sizeof(q15_t *), (12 - 16) * (int32_t)sizeof(q15_t *)
51 };
52
53 /*
54 * Process first stages
55 * Each stage in middle stages provides two down scaling of the input
56 */
57 n2 = fftLen;
58 n1 = n2;
59 n2 >>= 2u;
60
61 for (int k = fftLen / 4u; k > 1; k >>= 2u)
62 {
63 q15_t const *p_rearranged_twiddle_tab_stride2 =
64 &S->rearranged_twiddle_stride2[
65 S->rearranged_twiddle_tab_stride2_arr[stage]];
66 q15_t const *p_rearranged_twiddle_tab_stride3 = &S->rearranged_twiddle_stride3[
67 S->rearranged_twiddle_tab_stride3_arr[stage]];
68 q15_t const *p_rearranged_twiddle_tab_stride1 =
69 &S->rearranged_twiddle_stride1[
70 S->rearranged_twiddle_tab_stride1_arr[stage]];
71
72 q15_t * pBase = pSrc;
73 for (int i = 0; i < iter; i++)
74 {
75 q15_t *inA = pBase;
76 q15_t *inB = inA + n2 * CMPLX_DIM;
77 q15_t *inC = inB + n2 * CMPLX_DIM;
78 q15_t *inD = inC + n2 * CMPLX_DIM;
79 q15_t const *pW1 = p_rearranged_twiddle_tab_stride1;
80 q15_t const *pW2 = p_rearranged_twiddle_tab_stride2;
81 q15_t const *pW3 = p_rearranged_twiddle_tab_stride3;
82 q15x8_t vecW;
83
84 blkCnt = n2 / 4;
85 /*
86 * load 4 x q15 complex pair
87 */
88 vecA = vldrhq_s16(inA);
89 vecC = vldrhq_s16(inC);
90 while (blkCnt > 0U)
91 {
92 vecB = vldrhq_s16(inB);
93 vecD = vldrhq_s16(inD);
94
95 vecSum0 = vhaddq(vecA, vecC);
96 vecDiff0 = vhsubq(vecA, vecC);
97
98 vecSum1 = vhaddq(vecB, vecD);
99 vecDiff1 = vhsubq(vecB, vecD);
100 /*
101 * [ 1 1 1 1 ] * [ A B C D ]' .* 1
102 */
103 vecTmp0 = vhaddq(vecSum0, vecSum1);
104 vst1q(inA, vecTmp0);
105 inA += 8;
106 /*
107 * [ 1 -1 1 -1 ] * [ A B C D ]'
108 */
109 vecTmp0 = vhsubq(vecSum0, vecSum1);
110 /*
111 * [ 1 -1 1 -1 ] * [ A B C D ]'.* W2
112 */
113 vecW = vld1q(pW2);
114 pW2 += 8;
115 vecTmp1 = MVE_CMPLX_MULT_FX_AxB(vecW, vecTmp0, q15x8_t);
116
117 vst1q(inB, vecTmp1);
118 inB += 8;
119 /*
120 * [ 1 -i -1 +i ] * [ A B C D ]'
121 */
122 vecTmp0 = MVE_CMPLX_SUB_FX_A_ixB(vecDiff0, vecDiff1);
123 /*
124 * [ 1 -i -1 +i ] * [ A B C D ]'.* W1
125 */
126 vecW = vld1q(pW1);
127 pW1 += 8;
128 vecTmp1 = MVE_CMPLX_MULT_FX_AxB(vecW, vecTmp0, q15x8_t);
129 vst1q(inC, vecTmp1);
130 inC += 8;
131
132 /*
133 * [ 1 +i -1 -i ] * [ A B C D ]'
134 */
135 vecTmp0 = MVE_CMPLX_ADD_FX_A_ixB(vecDiff0, vecDiff1);
136 /*
137 * [ 1 +i -1 -i ] * [ A B C D ]'.* W3
138 */
139 vecW = vld1q(pW3);
140 pW3 += 8;
141 vecTmp1 = MVE_CMPLX_MULT_FX_AxB(vecW, vecTmp0, q15x8_t);
142 vst1q(inD, vecTmp1);
143 inD += 8;
144
145 vecA = vldrhq_s16(inA);
146 vecC = vldrhq_s16(inC);
147
148 blkCnt--;
149 }
150 pBase += CMPLX_DIM * n1;
151 }
152 n1 = n2;
153 n2 >>= 2u;
154 iter = iter << 2;
155 stage++;
156 }
157
158 /*
159 * start of Last stage process
160 */
161 uint32x4_t vecScGathAddr = vld1q_u32 ((uint32_t*)strides);
162 vecScGathAddr = vecScGathAddr + (uint32_t) pSrc;
163
164 /*
165 * load scheduling
166 */
167 vecA = (q15x8_t) vldrwq_gather_base_wb_s32(&vecScGathAddr, 64);
168 vecC = (q15x8_t) vldrwq_gather_base_s32(vecScGathAddr, 8);
169
170 blkCnt = (fftLen >> 4);
171 while (blkCnt > 0U)
172 {
173 vecSum0 = vhaddq(vecA, vecC);
174 vecDiff0 = vhsubq(vecA, vecC);
175
176 vecB = (q15x8_t) vldrwq_gather_base_s32(vecScGathAddr, 4);
177 vecD = (q15x8_t) vldrwq_gather_base_s32(vecScGathAddr, 12);
178
179 vecSum1 = vhaddq(vecB, vecD);
180 vecDiff1 = vhsubq(vecB, vecD);
181 /*
182 * pre-load for next iteration
183 */
184 vecA = (q15x8_t) vldrwq_gather_base_wb_s32(&vecScGathAddr, 64);
185 vecC = (q15x8_t) vldrwq_gather_base_s32(vecScGathAddr, 8);
186
187 vecTmp0 = vhaddq(vecSum0, vecSum1);
188 vstrwq_scatter_base_s32(vecScGathAddr, -64, (int32x4_t) vecTmp0);
189
190 vecTmp0 = vhsubq(vecSum0, vecSum1);
191 vstrwq_scatter_base_s32(vecScGathAddr, -64 + 4, (int32x4_t) vecTmp0);
192
193 vecTmp0 = MVE_CMPLX_SUB_FX_A_ixB(vecDiff0, vecDiff1);
194 vstrwq_scatter_base_s32(vecScGathAddr, -64 + 8, (int32x4_t) vecTmp0);
195
196 vecTmp0 = MVE_CMPLX_ADD_FX_A_ixB(vecDiff0, vecDiff1);
197 vstrwq_scatter_base_s32(vecScGathAddr, -64 + 12, (int32x4_t) vecTmp0);
198
199 blkCnt--;
200 }
201
202 }
203
arm_cfft_radix4by2_q15_mve(const arm_cfft_instance_q15 * S,q15_t * pSrc,uint32_t fftLen)204 static void arm_cfft_radix4by2_q15_mve(const arm_cfft_instance_q15 *S, q15_t *pSrc, uint32_t fftLen)
205 {
206 uint32_t n2;
207 q15_t *pIn0;
208 q15_t *pIn1;
209 const q15_t *pCoef = S->pTwiddle;
210 uint32_t blkCnt;
211 q15x8_t vecIn0, vecIn1, vecSum, vecDiff;
212 q15x8_t vecCmplxTmp, vecTw;
213 q15_t const *pCoefVec;
214
215 n2 = fftLen >> 1;
216
217 pIn0 = pSrc;
218 pIn1 = pSrc + fftLen;
219 pCoefVec = pCoef;
220
221 blkCnt = n2 / 4;
222
223 while (blkCnt > 0U)
224 {
225 vecIn0 = *(q15x8_t *) pIn0;
226 vecIn1 = *(q15x8_t *) pIn1;
227
228 vecIn0 = vecIn0 >> 1;
229 vecIn1 = vecIn1 >> 1;
230 vecSum = vhaddq(vecIn0, vecIn1);
231 vst1q(pIn0, vecSum);
232 pIn0 += 8;
233
234 vecTw = vld1q(pCoefVec);
235 pCoefVec += 8;
236
237 vecDiff = vhsubq(vecIn0, vecIn1);
238 vecCmplxTmp = MVE_CMPLX_MULT_FX_AxConjB(vecDiff, vecTw, q15x8_t);
239 vst1q(pIn1, vecCmplxTmp);
240 pIn1 += 8;
241
242 blkCnt--;
243 }
244
245 _arm_radix4_butterfly_q15_mve(S, pSrc, n2);
246
247 _arm_radix4_butterfly_q15_mve(S, pSrc + fftLen, n2);
248
249
250 pIn0 = pSrc;
251 blkCnt = (fftLen << 1) >> 3;
252 while (blkCnt > 0U)
253 {
254 vecIn0 = *(q15x8_t *) pIn0;
255 vecIn0 = vecIn0 << 1;
256 vst1q(pIn0, vecIn0);
257 pIn0 += 8;
258 blkCnt--;
259 }
260 /*
261 * tail
262 * (will be merged thru tail predication)
263 */
264 blkCnt = (fftLen << 1) & 7;
265 if (blkCnt > 0U)
266 {
267 mve_pred16_t p0 = vctp16q(blkCnt);
268
269 vecIn0 = *(q15x8_t *) pIn0;
270 vecIn0 = vecIn0 << 1;
271 vstrhq_p(pIn0, vecIn0, p0);
272 }
273 }
274
_arm_radix4_butterfly_inverse_q15_mve(const arm_cfft_instance_q15 * S,q15_t * pSrc,uint32_t fftLen)275 static void _arm_radix4_butterfly_inverse_q15_mve(const arm_cfft_instance_q15 *S,q15_t *pSrc, uint32_t fftLen)
276 {
277 q15x8_t vecTmp0, vecTmp1;
278 q15x8_t vecSum0, vecDiff0, vecSum1, vecDiff1;
279 q15x8_t vecA, vecB, vecC, vecD;
280 uint32_t blkCnt;
281 uint32_t n1, n2;
282 uint32_t stage = 0;
283 int32_t iter = 1;
284 static const int32_t strides[4] = {
285 (0 - 16) * (int32_t)sizeof(q15_t *), (4 - 16) * (int32_t)sizeof(q15_t *),
286 (8 - 16) * (int32_t)sizeof(q15_t *), (12 - 16) * (int32_t)sizeof(q15_t *)
287 };
288
289
290 /*
291 * Process first stages
292 * Each stage in middle stages provides two down scaling of the input
293 */
294 n2 = fftLen;
295 n1 = n2;
296 n2 >>= 2u;
297
298 for (int k = fftLen / 4u; k > 1; k >>= 2u)
299 {
300 q15_t const *p_rearranged_twiddle_tab_stride2 =
301 &S->rearranged_twiddle_stride2[
302 S->rearranged_twiddle_tab_stride2_arr[stage]];
303 q15_t const *p_rearranged_twiddle_tab_stride3 = &S->rearranged_twiddle_stride3[
304 S->rearranged_twiddle_tab_stride3_arr[stage]];
305 q15_t const *p_rearranged_twiddle_tab_stride1 =
306 &S->rearranged_twiddle_stride1[
307 S->rearranged_twiddle_tab_stride1_arr[stage]];
308
309 q15_t * pBase = pSrc;
310 for (int i = 0; i < iter; i++)
311 {
312 q15_t *inA = pBase;
313 q15_t *inB = inA + n2 * CMPLX_DIM;
314 q15_t *inC = inB + n2 * CMPLX_DIM;
315 q15_t *inD = inC + n2 * CMPLX_DIM;
316 q15_t const *pW1 = p_rearranged_twiddle_tab_stride1;
317 q15_t const *pW2 = p_rearranged_twiddle_tab_stride2;
318 q15_t const *pW3 = p_rearranged_twiddle_tab_stride3;
319 q15x8_t vecW;
320
321
322 blkCnt = n2 / 4;
323 /*
324 * load 4 x q15 complex pair
325 */
326 vecA = vldrhq_s16(inA);
327 vecC = vldrhq_s16(inC);
328 while (blkCnt > 0U)
329 {
330 vecB = vldrhq_s16(inB);
331 vecD = vldrhq_s16(inD);
332
333 vecSum0 = vhaddq(vecA, vecC);
334 vecDiff0 = vhsubq(vecA, vecC);
335
336 vecSum1 = vhaddq(vecB, vecD);
337 vecDiff1 = vhsubq(vecB, vecD);
338 /*
339 * [ 1 1 1 1 ] * [ A B C D ]' .* 1
340 */
341 vecTmp0 = vhaddq(vecSum0, vecSum1);
342 vst1q(inA, vecTmp0);
343 inA += 8;
344 /*
345 * [ 1 -1 1 -1 ] * [ A B C D ]'
346 */
347 vecTmp0 = vhsubq(vecSum0, vecSum1);
348 /*
349 * [ 1 -1 1 -1 ] * [ A B C D ]'.* W2
350 */
351 vecW = vld1q(pW2);
352 pW2 += 8;
353 vecTmp1 = MVE_CMPLX_MULT_FX_AxConjB(vecTmp0, vecW, q15x8_t);
354
355 vst1q(inB, vecTmp1);
356 inB += 8;
357 /*
358 * [ 1 -i -1 +i ] * [ A B C D ]'
359 */
360 vecTmp0 = MVE_CMPLX_ADD_FX_A_ixB(vecDiff0, vecDiff1);
361 /*
362 * [ 1 -i -1 +i ] * [ A B C D ]'.* W1
363 */
364 vecW = vld1q(pW1);
365 pW1 += 8;
366 vecTmp1 = MVE_CMPLX_MULT_FX_AxConjB(vecTmp0, vecW, q15x8_t);
367 vst1q(inC, vecTmp1);
368 inC += 8;
369 /*
370 * [ 1 +i -1 -i ] * [ A B C D ]'
371 */
372 vecTmp0 = MVE_CMPLX_SUB_FX_A_ixB(vecDiff0, vecDiff1);
373 /*
374 * [ 1 +i -1 -i ] * [ A B C D ]'.* W3
375 */
376 vecW = vld1q(pW3);
377 pW3 += 8;
378 vecTmp1 = MVE_CMPLX_MULT_FX_AxConjB(vecTmp0, vecW, q15x8_t);
379 vst1q(inD, vecTmp1);
380 inD += 8;
381
382 vecA = vldrhq_s16(inA);
383 vecC = vldrhq_s16(inC);
384
385 blkCnt--;
386 }
387 pBase += CMPLX_DIM * n1;
388 }
389 n1 = n2;
390 n2 >>= 2u;
391 iter = iter << 2;
392 stage++;
393 }
394
395 /*
396 * start of Last stage process
397 */
398 uint32x4_t vecScGathAddr = vld1q_u32((uint32_t*)strides);
399 vecScGathAddr = vecScGathAddr + (uint32_t) pSrc;
400
401 /*
402 * load scheduling
403 */
404 vecA = (q15x8_t) vldrwq_gather_base_wb_s32(&vecScGathAddr, 64);
405 vecC = (q15x8_t) vldrwq_gather_base_s32(vecScGathAddr, 8);
406
407 blkCnt = (fftLen >> 4);
408 while (blkCnt > 0U)
409 {
410 vecSum0 = vhaddq(vecA, vecC);
411 vecDiff0 = vhsubq(vecA, vecC);
412
413 vecB = (q15x8_t) vldrwq_gather_base_s32(vecScGathAddr, 4);
414 vecD = (q15x8_t) vldrwq_gather_base_s32(vecScGathAddr, 12);
415
416 vecSum1 = vhaddq(vecB, vecD);
417 vecDiff1 = vhsubq(vecB, vecD);
418 /*
419 * pre-load for next iteration
420 */
421 vecA = (q15x8_t) vldrwq_gather_base_wb_s32(&vecScGathAddr, 64);
422 vecC = (q15x8_t) vldrwq_gather_base_s32(vecScGathAddr, 8);
423
424 vecTmp0 = vhaddq(vecSum0, vecSum1);
425 vstrwq_scatter_base_s32(vecScGathAddr, -64, (int32x4_t) vecTmp0);
426
427 vecTmp0 = vhsubq(vecSum0, vecSum1);
428 vstrwq_scatter_base_s32(vecScGathAddr, -64 + 4, (int32x4_t) vecTmp0);
429
430 vecTmp0 = MVE_CMPLX_ADD_FX_A_ixB(vecDiff0, vecDiff1);
431 vstrwq_scatter_base_s32(vecScGathAddr, -64 + 8, (int32x4_t) vecTmp0);
432
433 vecTmp0 = MVE_CMPLX_SUB_FX_A_ixB(vecDiff0, vecDiff1);
434 vstrwq_scatter_base_s32(vecScGathAddr, -64 + 12, (int32x4_t) vecTmp0);
435
436 blkCnt--;
437 }
438 }
439
arm_cfft_radix4by2_inverse_q15_mve(const arm_cfft_instance_q15 * S,q15_t * pSrc,uint32_t fftLen)440 static void arm_cfft_radix4by2_inverse_q15_mve(const arm_cfft_instance_q15 *S, q15_t *pSrc, uint32_t fftLen)
441 {
442 uint32_t n2;
443 q15_t *pIn0;
444 q15_t *pIn1;
445 const q15_t *pCoef = S->pTwiddle;
446
447 uint32_t blkCnt;
448 q15x8_t vecIn0, vecIn1, vecSum, vecDiff;
449 q15x8_t vecCmplxTmp, vecTw;
450 q15_t const *pCoefVec;
451
452 n2 = fftLen >> 1;
453
454 pIn0 = pSrc;
455 pIn1 = pSrc + fftLen;
456 pCoefVec = pCoef;
457
458 blkCnt = n2 / 4;
459
460 while (blkCnt > 0U)
461 {
462 vecIn0 = *(q15x8_t *) pIn0;
463 vecIn1 = *(q15x8_t *) pIn1;
464
465 vecIn0 = vecIn0 >> 1;
466 vecIn1 = vecIn1 >> 1;
467 vecSum = vhaddq(vecIn0, vecIn1);
468 vst1q(pIn0, vecSum);
469 pIn0 += 8;
470
471 vecTw = vld1q(pCoefVec);
472 pCoefVec += 8;
473
474 vecDiff = vhsubq(vecIn0, vecIn1);
475 vecCmplxTmp = vqrdmlsdhq(vuninitializedq_s16() , vecDiff, vecTw);
476 vecCmplxTmp = vqrdmladhxq(vecCmplxTmp, vecDiff, vecTw);
477 vst1q(pIn1, vecCmplxTmp);
478 pIn1 += 8;
479
480 blkCnt--;
481 }
482
483
484 _arm_radix4_butterfly_inverse_q15_mve(S, pSrc, n2);
485
486 _arm_radix4_butterfly_inverse_q15_mve(S, pSrc + fftLen, n2);
487
488 pIn0 = pSrc;
489 blkCnt = (fftLen << 1) >> 3;
490 while (blkCnt > 0U)
491 {
492 vecIn0 = *(q15x8_t *) pIn0;
493 vecIn0 = vecIn0 << 1;
494 vst1q(pIn0, vecIn0);
495 pIn0 += 8;
496 blkCnt--;
497 }
498 /*
499 * tail
500 * (will be merged thru tail predication)
501 */
502 blkCnt = (fftLen << 1) & 7;
503 while (blkCnt > 0U)
504 {
505 mve_pred16_t p0 = vctp16q(blkCnt);
506
507 vecIn0 = *(q15x8_t *) pIn0;
508 vecIn0 = vecIn0 << 1;
509 vstrhq_p(pIn0, vecIn0, p0);
510 }
511 }
512
513 /**
514 @addtogroup ComplexFFTQ15
515 @{
516 */
517
518 /**
519 @brief Processing function for Q15 complex FFT.
520 @param[in] S points to an instance of Q15 CFFT structure
521 @param[in,out] p1 points to the complex data buffer of size <code>2*fftLen</code>. Processing occurs in-place
522 @param[in] ifftFlag flag that selects transform direction
523 - value = 0: forward transform
524 - value = 1: inverse transform
525 @param[in] bitReverseFlag flag that enables / disables bit reversal of output
526 - value = 0: disables bit reversal of output
527 - value = 1: enables bit reversal of output
528 */
arm_cfft_q15(const arm_cfft_instance_q15 * S,q15_t * pSrc,uint8_t ifftFlag,uint8_t bitReverseFlag)529 ARM_DSP_ATTRIBUTE void arm_cfft_q15(
530 const arm_cfft_instance_q15 * S,
531 q15_t * pSrc,
532 uint8_t ifftFlag,
533 uint8_t bitReverseFlag)
534 {
535 uint32_t fftLen = S->fftLen;
536
537 if (ifftFlag == 1U) {
538
539 switch (fftLen) {
540 case 16:
541 case 64:
542 case 256:
543 case 1024:
544 case 4096:
545 _arm_radix4_butterfly_inverse_q15_mve(S, pSrc, fftLen);
546 break;
547
548 case 32:
549 case 128:
550 case 512:
551 case 2048:
552 arm_cfft_radix4by2_inverse_q15_mve(S, pSrc, fftLen);
553 break;
554 }
555 } else {
556 switch (fftLen) {
557 case 16:
558 case 64:
559 case 256:
560 case 1024:
561 case 4096:
562 _arm_radix4_butterfly_q15_mve(S, pSrc, fftLen);
563 break;
564
565 case 32:
566 case 128:
567 case 512:
568 case 2048:
569 arm_cfft_radix4by2_q15_mve(S, pSrc, fftLen);
570 break;
571 }
572 }
573
574
575 if (bitReverseFlag)
576 {
577
578 arm_bitreversal_16_inpl_mve((uint16_t*)pSrc, S->bitRevLength, S->pBitRevTable);
579
580 }
581 }
582
583 #else
584
585 extern void arm_radix4_butterfly_q15(
586 q15_t * pSrc,
587 uint32_t fftLen,
588 const q15_t * pCoef,
589 uint32_t twidCoefModifier);
590
591 extern void arm_radix4_butterfly_inverse_q15(
592 q15_t * pSrc,
593 uint32_t fftLen,
594 const q15_t * pCoef,
595 uint32_t twidCoefModifier);
596
597 extern void arm_bitreversal_16(
598 uint16_t * pSrc,
599 const uint16_t bitRevLen,
600 const uint16_t * pBitRevTable);
601
602 ARM_DSP_ATTRIBUTE void arm_cfft_radix4by2_q15(
603 q15_t * pSrc,
604 uint32_t fftLen,
605 const q15_t * pCoef);
606
607 ARM_DSP_ATTRIBUTE void arm_cfft_radix4by2_inverse_q15(
608 q15_t * pSrc,
609 uint32_t fftLen,
610 const q15_t * pCoef);
611
612
613
614 /**
615 @addtogroup ComplexFFTQ15
616 @{
617 */
618
619 /**
620 @brief Processing function for Q15 complex FFT.
621 @param[in] S points to an instance of Q15 CFFT structure
622 @param[in,out] p1 points to the complex data buffer of size <code>2*fftLen</code>. Processing occurs in-place
623 @param[in] ifftFlag flag that selects transform direction
624 - value = 0: forward transform
625 - value = 1: inverse transform
626 @param[in] bitReverseFlag flag that enables / disables bit reversal of output
627 - value = 0: disables bit reversal of output
628 - value = 1: enables bit reversal of output
629 */
630
arm_cfft_q15(const arm_cfft_instance_q15 * S,q15_t * p1,uint8_t ifftFlag,uint8_t bitReverseFlag)631 ARM_DSP_ATTRIBUTE void arm_cfft_q15(
632 const arm_cfft_instance_q15 * S,
633 q15_t * p1,
634 uint8_t ifftFlag,
635 uint8_t bitReverseFlag)
636 {
637 uint32_t L = S->fftLen;
638
639 if (ifftFlag == 1U)
640 {
641 switch (L)
642 {
643 case 16:
644 case 64:
645 case 256:
646 case 1024:
647 case 4096:
648 arm_radix4_butterfly_inverse_q15 ( p1, L, (q15_t*)S->pTwiddle, 1 );
649 break;
650
651 case 32:
652 case 128:
653 case 512:
654 case 2048:
655 arm_cfft_radix4by2_inverse_q15 ( p1, L, S->pTwiddle );
656 break;
657 }
658 }
659 else
660 {
661 switch (L)
662 {
663 case 16:
664 case 64:
665 case 256:
666 case 1024:
667 case 4096:
668 arm_radix4_butterfly_q15 ( p1, L, (q15_t*)S->pTwiddle, 1 );
669 break;
670
671 case 32:
672 case 128:
673 case 512:
674 case 2048:
675 arm_cfft_radix4by2_q15 ( p1, L, S->pTwiddle );
676 break;
677 }
678 }
679
680 if ( bitReverseFlag )
681 arm_bitreversal_16 ((uint16_t*) p1, S->bitRevLength, S->pBitRevTable);
682 }
683
684 /**
685 @} end of ComplexFFTQ15 group
686 */
687
arm_cfft_radix4by2_q15(q15_t * pSrc,uint32_t fftLen,const q15_t * pCoef)688 ARM_DSP_ATTRIBUTE void arm_cfft_radix4by2_q15(
689 q15_t * pSrc,
690 uint32_t fftLen,
691 const q15_t * pCoef)
692 {
693 uint32_t i;
694 uint32_t n2;
695 q15_t p0, p1, p2, p3;
696 #if defined (ARM_MATH_DSP)
697 q31_t T, S, R;
698 q31_t coeff, out1, out2;
699 const q15_t *pC = pCoef;
700 q15_t *pSi = pSrc;
701 q15_t *pSl = pSrc + fftLen;
702 #else
703 uint32_t l;
704 q15_t xt, yt, cosVal, sinVal;
705 #endif
706
707 n2 = fftLen >> 1U;
708
709 #if defined (ARM_MATH_DSP)
710
711 for (i = n2; i > 0; i--)
712 {
713 coeff = read_q15x2_ia (&pC);
714
715 T = read_q15x2 (pSi);
716 T = __SHADD16(T, 0); /* this is just a SIMD arithmetic shift right by 1 */
717
718 S = read_q15x2 (pSl);
719 S = __SHADD16(S, 0); /* this is just a SIMD arithmetic shift right by 1 */
720
721 R = __QSUB16(T, S);
722
723 write_q15x2_ia (&pSi, __SHADD16(T, S));
724
725 #ifndef ARM_MATH_BIG_ENDIAN
726 out1 = __SMUAD(coeff, R) >> 16U;
727 out2 = __SMUSDX(coeff, R);
728 #else
729 out1 = __SMUSDX(R, coeff) >> 16U;
730 out2 = __SMUAD(coeff, R);
731 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
732
733 write_q15x2_ia (&pSl, (q31_t)__PKHBT( out1, out2, 0 ) );
734 }
735
736 #else /* #if defined (ARM_MATH_DSP) */
737
738 for (i = 0; i < n2; i++)
739 {
740 cosVal = pCoef[2 * i];
741 sinVal = pCoef[2 * i + 1];
742
743 l = i + n2;
744
745 xt = (pSrc[2 * i] >> 1U) - (pSrc[2 * l] >> 1U);
746 pSrc[2 * i] = ((pSrc[2 * i] >> 1U) + (pSrc[2 * l] >> 1U)) >> 1U;
747
748 yt = (pSrc[2 * i + 1] >> 1U) - (pSrc[2 * l + 1] >> 1U);
749 pSrc[2 * i + 1] = ((pSrc[2 * l + 1] >> 1U) + (pSrc[2 * i + 1] >> 1U)) >> 1U;
750
751 pSrc[2 * l] = (((int16_t) (((q31_t) xt * cosVal) >> 16U)) +
752 ((int16_t) (((q31_t) yt * sinVal) >> 16U)) );
753
754 pSrc[2 * l + 1] = (((int16_t) (((q31_t) yt * cosVal) >> 16U)) -
755 ((int16_t) (((q31_t) xt * sinVal) >> 16U)) );
756 }
757
758 #endif /* #if defined (ARM_MATH_DSP) */
759
760 /* first col */
761 arm_radix4_butterfly_q15( pSrc, n2, (q15_t*)pCoef, 2U);
762
763 /* second col */
764 arm_radix4_butterfly_q15( pSrc + fftLen, n2, (q15_t*)pCoef, 2U);
765
766 n2 = fftLen >> 1U;
767 for (i = 0; i < n2; i++)
768 {
769 p0 = pSrc[4 * i + 0];
770 p1 = pSrc[4 * i + 1];
771 p2 = pSrc[4 * i + 2];
772 p3 = pSrc[4 * i + 3];
773
774 p0 <<= 1U;
775 p1 <<= 1U;
776 p2 <<= 1U;
777 p3 <<= 1U;
778
779 pSrc[4 * i + 0] = p0;
780 pSrc[4 * i + 1] = p1;
781 pSrc[4 * i + 2] = p2;
782 pSrc[4 * i + 3] = p3;
783 }
784
785 }
786
arm_cfft_radix4by2_inverse_q15(q15_t * pSrc,uint32_t fftLen,const q15_t * pCoef)787 ARM_DSP_ATTRIBUTE void arm_cfft_radix4by2_inverse_q15(
788 q15_t * pSrc,
789 uint32_t fftLen,
790 const q15_t * pCoef)
791 {
792 uint32_t i;
793 uint32_t n2;
794 q15_t p0, p1, p2, p3;
795 #if defined (ARM_MATH_DSP)
796 q31_t T, S, R;
797 q31_t coeff, out1, out2;
798 const q15_t *pC = pCoef;
799 q15_t *pSi = pSrc;
800 q15_t *pSl = pSrc + fftLen;
801 #else
802 uint32_t l;
803 q15_t xt, yt, cosVal, sinVal;
804 #endif
805
806 n2 = fftLen >> 1U;
807
808 #if defined (ARM_MATH_DSP)
809
810 for (i = n2; i > 0; i--)
811 {
812 coeff = read_q15x2_ia (&pC);
813
814 T = read_q15x2 (pSi);
815 T = __SHADD16(T, 0); /* this is just a SIMD arithmetic shift right by 1 */
816
817 S = read_q15x2 (pSl);
818 S = __SHADD16(S, 0); /* this is just a SIMD arithmetic shift right by 1 */
819
820 R = __QSUB16(T, S);
821
822 write_q15x2_ia (&pSi, __SHADD16(T, S));
823
824 #ifndef ARM_MATH_BIG_ENDIAN
825 out1 = __SMUSD(coeff, R) >> 16U;
826 out2 = __SMUADX(coeff, R);
827 #else
828 out1 = __SMUADX(R, coeff) >> 16U;
829 out2 = __SMUSD(__QSUB(0, coeff), R);
830 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
831
832 write_q15x2_ia (&pSl, (q31_t)__PKHBT( out1, out2, 0 ));
833 }
834
835 #else /* #if defined (ARM_MATH_DSP) */
836
837 for (i = 0; i < n2; i++)
838 {
839 cosVal = pCoef[2 * i];
840 sinVal = pCoef[2 * i + 1];
841
842 l = i + n2;
843
844 xt = (pSrc[2 * i] >> 1U) - (pSrc[2 * l] >> 1U);
845 pSrc[2 * i] = ((pSrc[2 * i] >> 1U) + (pSrc[2 * l] >> 1U)) >> 1U;
846
847 yt = (pSrc[2 * i + 1] >> 1U) - (pSrc[2 * l + 1] >> 1U);
848 pSrc[2 * i + 1] = ((pSrc[2 * l + 1] >> 1U) + (pSrc[2 * i + 1] >> 1U)) >> 1U;
849
850 pSrc[2 * l] = (((int16_t) (((q31_t) xt * cosVal) >> 16U)) -
851 ((int16_t) (((q31_t) yt * sinVal) >> 16U)) );
852
853 pSrc[2 * l + 1] = (((int16_t) (((q31_t) yt * cosVal) >> 16U)) +
854 ((int16_t) (((q31_t) xt * sinVal) >> 16U)) );
855 }
856
857 #endif /* #if defined (ARM_MATH_DSP) */
858
859 /* first col */
860 arm_radix4_butterfly_inverse_q15( pSrc, n2, (q15_t*)pCoef, 2U);
861
862 /* second col */
863 arm_radix4_butterfly_inverse_q15( pSrc + fftLen, n2, (q15_t*)pCoef, 2U);
864
865 n2 = fftLen >> 1U;
866 for (i = 0; i < n2; i++)
867 {
868 p0 = pSrc[4 * i + 0];
869 p1 = pSrc[4 * i + 1];
870 p2 = pSrc[4 * i + 2];
871 p3 = pSrc[4 * i + 3];
872
873 p0 <<= 1U;
874 p1 <<= 1U;
875 p2 <<= 1U;
876 p3 <<= 1U;
877
878 pSrc[4 * i + 0] = p0;
879 pSrc[4 * i + 1] = p1;
880 pSrc[4 * i + 2] = p2;
881 pSrc[4 * i + 3] = p3;
882 }
883 }
884
885 #endif /* defined(ARM_MATH_MVEI) */
886
