Mercurial > hg > sv-dependency-builds

comparison src/fftw-3.3.8/rdft/scalar/r2cb/hc2cb2_8.c @ 167:bd3cc4d1df30

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add FFTW 3.3.8 source, and a Linux build
author Chris Cannam <cannam@all-day-breakfast.com>
date Tue, 19 Nov 2019 14:52:55 +0000
parents
children
comparison
equal deleted inserted replaced
166:cbd6d7e562c7 167:bd3cc4d1df30
1 /*
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 */
20
21 /* This file was automatically generated --- DO NOT EDIT */
22 /* Generated on Thu May 24 08:07:54 EDT 2018 */
23
24 #include "rdft/codelet-rdft.h"
25
26 #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
27
28 /* Generated by: ../../../genfft/gen_hc2c.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 8 -dif -name hc2cb2_8 -include rdft/scalar/hc2cb.h */
29
30 /*
31 * This function contains 74 FP additions, 50 FP multiplications,
32 * (or, 44 additions, 20 multiplications, 30 fused multiply/add),
33 * 47 stack variables, 1 constants, and 32 memory accesses
34 */
35 #include "rdft/scalar/hc2cb.h"
36
37 static void hc2cb2_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
38 {
39 DK(KP707106781, +0.707106781186547524400844362104849039284835938);
40 {
41 INT m;
42 for (m = mb, W = W + ((mb - 1) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(32, rs)) {
43 E Tf, Tg, Tl, Tp, Ti, Tj, Tk, T1b, T1u, T1e, T1o, To, Tq, TK;
44 {
45 E Th, T1n, T1t, Tn, Tm, TJ;
46 Tf = W[0];
47 Tg = W[2];
48 Th = Tf * Tg;
49 Tl = W[4];
50 T1n = Tf * Tl;
51 Tp = W[5];
52 T1t = Tf * Tp;
53 Ti = W[1];
54 Tj = W[3];
55 Tn = Tf * Tj;
56 Tk = FMA(Ti, Tj, Th);
57 T1b = FNMS(Ti, Tj, Th);
58 T1u = FNMS(Ti, Tl, T1t);
59 T1e = FMA(Ti, Tg, Tn);
60 T1o = FMA(Ti, Tp, T1n);
61 Tm = Tk * Tl;
62 TJ = Tk * Tp;
63 To = FNMS(Ti, Tg, Tn);
64 Tq = FMA(To, Tp, Tm);
65 TK = FNMS(To, Tl, TJ);
66 }
67 {
68 E T7, T1p, T1v, Tv, TP, T13, T1h, TZ, Te, T1k, T1w, T1q, TQ, TR, T10;
69 E TG, T14;
70 {
71 E T3, Tr, TO, T1f, T6, TL, Tu, T1g;
72 {
73 E T1, T2, TM, TN;
74 T1 = Rp[0];
75 T2 = Rm[WS(rs, 3)];
76 T3 = T1 + T2;
77 Tr = T1 - T2;
78 TM = Ip[0];
79 TN = Im[WS(rs, 3)];
80 TO = TM + TN;
81 T1f = TM - TN;
82 }
83 {
84 E T4, T5, Ts, Tt;
85 T4 = Rp[WS(rs, 2)];
86 T5 = Rm[WS(rs, 1)];
87 T6 = T4 + T5;
88 TL = T4 - T5;
89 Ts = Ip[WS(rs, 2)];
90 Tt = Im[WS(rs, 1)];
91 Tu = Ts + Tt;
92 T1g = Ts - Tt;
93 }
94 T7 = T3 + T6;
95 T1p = T3 - T6;
96 T1v = T1f - T1g;
97 Tv = Tr - Tu;
98 TP = TL + TO;
99 T13 = TO - TL;
100 T1h = T1f + T1g;
101 TZ = Tr + Tu;
102 }
103 {
104 E Ta, Tw, Tz, T1i, Td, TB, TE, T1j, TA, TF;
105 {
106 E T8, T9, Tx, Ty;
107 T8 = Rp[WS(rs, 1)];
108 T9 = Rm[WS(rs, 2)];
109 Ta = T8 + T9;
110 Tw = T8 - T9;
111 Tx = Ip[WS(rs, 1)];
112 Ty = Im[WS(rs, 2)];
113 Tz = Tx + Ty;
114 T1i = Tx - Ty;
115 }
116 {
117 E Tb, Tc, TC, TD;
118 Tb = Rm[0];
119 Tc = Rp[WS(rs, 3)];
120 Td = Tb + Tc;
121 TB = Tb - Tc;
122 TC = Ip[WS(rs, 3)];
123 TD = Im[0];
124 TE = TC + TD;
125 T1j = TC - TD;
126 }
127 Te = Ta + Td;
128 T1k = T1i + T1j;
129 T1w = Ta - Td;
130 T1q = T1j - T1i;
131 TQ = Tw + Tz;
132 TR = TB + TE;
133 T10 = TQ + TR;
134 TA = Tw - Tz;
135 TF = TB - TE;
136 TG = TA + TF;
137 T14 = TA - TF;
138 }
139 Rp[0] = T7 + Te;
140 Rm[0] = T1h + T1k;
141 {
142 E T11, T12, T15, T16;
143 T11 = FNMS(KP707106781, T10, TZ);
144 T12 = Tg * T11;
145 T15 = FMA(KP707106781, T14, T13);
146 T16 = Tg * T15;
147 Ip[WS(rs, 1)] = FNMS(Tj, T15, T12);
148 Im[WS(rs, 1)] = FMA(Tj, T11, T16);
149 }
150 {
151 E T1z, T1A, T1B, T1C;
152 T1z = T1p + T1q;
153 T1A = Tk * T1z;
154 T1B = T1w + T1v;
155 T1C = Tk * T1B;
156 Rp[WS(rs, 1)] = FNMS(To, T1B, T1A);
157 Rm[WS(rs, 1)] = FMA(To, T1z, T1C);
158 }
159 {
160 E T17, T18, T19, T1a;
161 T17 = FMA(KP707106781, T10, TZ);
162 T18 = Tl * T17;
163 T19 = FNMS(KP707106781, T14, T13);
164 T1a = Tl * T19;
165 Ip[WS(rs, 3)] = FNMS(Tp, T19, T18);
166 Im[WS(rs, 3)] = FMA(Tp, T17, T1a);
167 }
168 {
169 E T1l, T1d, T1m, T1c;
170 T1l = T1h - T1k;
171 T1c = T7 - Te;
172 T1d = T1b * T1c;
173 T1m = T1e * T1c;
174 Rp[WS(rs, 2)] = FNMS(T1e, T1l, T1d);
175 Rm[WS(rs, 2)] = FMA(T1b, T1l, T1m);
176 }
177 {
178 E T1r, T1s, T1x, T1y;
179 T1r = T1p - T1q;
180 T1s = T1o * T1r;
181 T1x = T1v - T1w;
182 T1y = T1o * T1x;
183 Rp[WS(rs, 3)] = FNMS(T1u, T1x, T1s);
184 Rm[WS(rs, 3)] = FMA(T1u, T1r, T1y);
185 }
186 {
187 E TT, TX, TW, TY, TI, TU, TS, TV, TH;
188 TS = TQ - TR;
189 TT = FNMS(KP707106781, TS, TP);
190 TX = FMA(KP707106781, TS, TP);
191 TV = FMA(KP707106781, TG, Tv);
192 TW = Tf * TV;
193 TY = Ti * TV;
194 TH = FNMS(KP707106781, TG, Tv);
195 TI = Tq * TH;
196 TU = TK * TH;
197 Ip[WS(rs, 2)] = FNMS(TK, TT, TI);
198 Im[WS(rs, 2)] = FMA(Tq, TT, TU);
199 Ip[0] = FNMS(Ti, TX, TW);
200 Im[0] = FMA(Tf, TX, TY);
201 }
202 }
203 }
204 }
205 }
206
207 static const tw_instr twinstr[] = {
208 {TW_CEXP, 1, 1},
209 {TW_CEXP, 1, 3},
210 {TW_CEXP, 1, 7},
211 {TW_NEXT, 1, 0}
212 };
213
214 static const hc2c_desc desc = { 8, "hc2cb2_8", twinstr, &GENUS, {44, 20, 30, 0} };
215
216 void X(codelet_hc2cb2_8) (planner *p) {
217 X(khc2c_register) (p, hc2cb2_8, &desc, HC2C_VIA_RDFT);
218 }
219 #else
220
221 /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 8 -dif -name hc2cb2_8 -include rdft/scalar/hc2cb.h */
222
223 /*
224 * This function contains 74 FP additions, 44 FP multiplications,
225 * (or, 56 additions, 26 multiplications, 18 fused multiply/add),
226 * 46 stack variables, 1 constants, and 32 memory accesses
227 */
228 #include "rdft/scalar/hc2cb.h"
229
230 static void hc2cb2_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
231 {
232 DK(KP707106781, +0.707106781186547524400844362104849039284835938);
233 {
234 INT m;
235 for (m = mb, W = W + ((mb - 1) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(32, rs)) {
236 E Tf, Ti, Tg, Tj, Tl, Tp, TP, TR, TF, TG, TH, T15, TL, TT;
237 {
238 E Th, To, Tk, Tn;
239 Tf = W[0];
240 Ti = W[1];
241 Tg = W[2];
242 Tj = W[3];
243 Th = Tf * Tg;
244 To = Ti * Tg;
245 Tk = Ti * Tj;
246 Tn = Tf * Tj;
247 Tl = Th - Tk;
248 Tp = Tn + To;
249 TP = Th + Tk;
250 TR = Tn - To;
251 TF = W[4];
252 TG = W[5];
253 TH = FMA(Tf, TF, Ti * TG);
254 T15 = FNMS(TR, TF, TP * TG);
255 TL = FNMS(Ti, TF, Tf * TG);
256 TT = FMA(TP, TF, TR * TG);
257 }
258 {
259 E T7, T1f, T1i, Tw, TI, TW, T18, TM, Te, T19, T1a, TD, TJ, TZ, T12;
260 E TN, Tm, TE;
261 {
262 E T3, TU, Ts, T17, T6, T16, Tv, TV;
263 {
264 E T1, T2, Tq, Tr;
265 T1 = Rp[0];
266 T2 = Rm[WS(rs, 3)];
267 T3 = T1 + T2;
268 TU = T1 - T2;
269 Tq = Ip[0];
270 Tr = Im[WS(rs, 3)];
271 Ts = Tq - Tr;
272 T17 = Tq + Tr;
273 }
274 {
275 E T4, T5, Tt, Tu;
276 T4 = Rp[WS(rs, 2)];
277 T5 = Rm[WS(rs, 1)];
278 T6 = T4 + T5;
279 T16 = T4 - T5;
280 Tt = Ip[WS(rs, 2)];
281 Tu = Im[WS(rs, 1)];
282 Tv = Tt - Tu;
283 TV = Tt + Tu;
284 }
285 T7 = T3 + T6;
286 T1f = TU + TV;
287 T1i = T17 - T16;
288 Tw = Ts + Tv;
289 TI = T3 - T6;
290 TW = TU - TV;
291 T18 = T16 + T17;
292 TM = Ts - Tv;
293 }
294 {
295 E Ta, TX, Tz, TY, Td, T10, TC, T11;
296 {
297 E T8, T9, Tx, Ty;
298 T8 = Rp[WS(rs, 1)];
299 T9 = Rm[WS(rs, 2)];
300 Ta = T8 + T9;
301 TX = T8 - T9;
302 Tx = Ip[WS(rs, 1)];
303 Ty = Im[WS(rs, 2)];
304 Tz = Tx - Ty;
305 TY = Tx + Ty;
306 }
307 {
308 E Tb, Tc, TA, TB;
309 Tb = Rm[0];
310 Tc = Rp[WS(rs, 3)];
311 Td = Tb + Tc;
312 T10 = Tb - Tc;
313 TA = Ip[WS(rs, 3)];
314 TB = Im[0];
315 TC = TA - TB;
316 T11 = TA + TB;
317 }
318 Te = Ta + Td;
319 T19 = TX + TY;
320 T1a = T10 + T11;
321 TD = Tz + TC;
322 TJ = TC - Tz;
323 TZ = TX - TY;
324 T12 = T10 - T11;
325 TN = Ta - Td;
326 }
327 Rp[0] = T7 + Te;
328 Rm[0] = Tw + TD;
329 Tm = T7 - Te;
330 TE = Tw - TD;
331 Rp[WS(rs, 2)] = FNMS(Tp, TE, Tl * Tm);
332 Rm[WS(rs, 2)] = FMA(Tp, Tm, Tl * TE);
333 {
334 E TQ, TS, TK, TO;
335 TQ = TI + TJ;
336 TS = TN + TM;
337 Rp[WS(rs, 1)] = FNMS(TR, TS, TP * TQ);
338 Rm[WS(rs, 1)] = FMA(TP, TS, TR * TQ);
339 TK = TI - TJ;
340 TO = TM - TN;
341 Rp[WS(rs, 3)] = FNMS(TL, TO, TH * TK);
342 Rm[WS(rs, 3)] = FMA(TH, TO, TL * TK);
343 }
344 {
345 E T1h, T1l, T1k, T1m, T1g, T1j;
346 T1g = KP707106781 * (T19 + T1a);
347 T1h = T1f - T1g;
348 T1l = T1f + T1g;
349 T1j = KP707106781 * (TZ - T12);
350 T1k = T1i + T1j;
351 T1m = T1i - T1j;
352 Ip[WS(rs, 1)] = FNMS(Tj, T1k, Tg * T1h);
353 Im[WS(rs, 1)] = FMA(Tg, T1k, Tj * T1h);
354 Ip[WS(rs, 3)] = FNMS(TG, T1m, TF * T1l);
355 Im[WS(rs, 3)] = FMA(TF, T1m, TG * T1l);
356 }
357 {
358 E T14, T1d, T1c, T1e, T13, T1b;
359 T13 = KP707106781 * (TZ + T12);
360 T14 = TW - T13;
361 T1d = TW + T13;
362 T1b = KP707106781 * (T19 - T1a);
363 T1c = T18 - T1b;
364 T1e = T18 + T1b;
365 Ip[WS(rs, 2)] = FNMS(T15, T1c, TT * T14);
366 Im[WS(rs, 2)] = FMA(T15, T14, TT * T1c);
367 Ip[0] = FNMS(Ti, T1e, Tf * T1d);
368 Im[0] = FMA(Ti, T1d, Tf * T1e);
369 }
370 }
371 }
372 }
373 }
374
375 static const tw_instr twinstr[] = {
376 {TW_CEXP, 1, 1},
377 {TW_CEXP, 1, 3},
378 {TW_CEXP, 1, 7},
379 {TW_NEXT, 1, 0}
380 };
381
382 static const hc2c_desc desc = { 8, "hc2cb2_8", twinstr, &GENUS, {56, 26, 18, 0} };
383
384 void X(codelet_hc2cb2_8) (planner *p) {
385 X(khc2c_register) (p, hc2cb2_8, &desc, HC2C_VIA_RDFT);
386 }
387 #endif