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 Dec 10 07:06:27 EST 2020 */
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_hc2hc.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 8 -dif -name hb_8 -include rdft/scalar/hb.h */
29
30 /*
31 * This function contains 66 FP additions, 36 FP multiplications,
32 * (or, 44 additions, 14 multiplications, 22 fused multiply/add),
33 * 33 stack variables, 1 constants, and 32 memory accesses
34 */
35 #include "rdft/scalar/hb.h"
36
hb_8(R * cr,R * ci,const R * W,stride rs,INT mb,INT me,INT ms)37 static void hb_8(R *cr, R *ci, 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) * 14); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 14, MAKE_VOLATILE_STRIDE(16, rs)) {
43 E T7, T1i, T1n, Tk, TD, TV, T1b, TQ, Te, T1e, T1o, T1j, TE, TF, TR;
44 E Tv, TW;
45 {
46 E T3, Tg, TC, T19, T6, Tz, Tj, T1a;
47 {
48 E T1, T2, TA, TB;
49 T1 = cr[0];
50 T2 = ci[WS(rs, 3)];
51 T3 = T1 + T2;
52 Tg = T1 - T2;
53 TA = ci[WS(rs, 7)];
54 TB = cr[WS(rs, 4)];
55 TC = TA + TB;
56 T19 = TA - TB;
57 }
58 {
59 E T4, T5, Th, Ti;
60 T4 = cr[WS(rs, 2)];
61 T5 = ci[WS(rs, 1)];
62 T6 = T4 + T5;
63 Tz = T4 - T5;
64 Th = ci[WS(rs, 5)];
65 Ti = cr[WS(rs, 6)];
66 Tj = Th + Ti;
67 T1a = Th - Ti;
68 }
69 T7 = T3 + T6;
70 T1i = T3 - T6;
71 T1n = T19 - T1a;
72 Tk = Tg - Tj;
73 TD = Tz + TC;
74 TV = TC - Tz;
75 T1b = T19 + T1a;
76 TQ = Tg + Tj;
77 }
78 {
79 E Ta, Tl, Tt, T1d, Td, Tq, To, T1c, Tp, Tu;
80 {
81 E T8, T9, Tr, Ts;
82 T8 = cr[WS(rs, 1)];
83 T9 = ci[WS(rs, 2)];
84 Ta = T8 + T9;
85 Tl = T8 - T9;
86 Tr = ci[WS(rs, 4)];
87 Ts = cr[WS(rs, 7)];
88 Tt = Tr + Ts;
89 T1d = Tr - Ts;
90 }
91 {
92 E Tb, Tc, Tm, Tn;
93 Tb = ci[0];
94 Tc = cr[WS(rs, 3)];
95 Td = Tb + Tc;
96 Tq = Tb - Tc;
97 Tm = ci[WS(rs, 6)];
98 Tn = cr[WS(rs, 5)];
99 To = Tm + Tn;
100 T1c = Tm - Tn;
101 }
102 Te = Ta + Td;
103 T1e = T1c + T1d;
104 T1o = Ta - Td;
105 T1j = T1d - T1c;
106 TE = Tl + To;
107 TF = Tq + Tt;
108 TR = TE + TF;
109 Tp = Tl - To;
110 Tu = Tq - Tt;
111 Tv = Tp + Tu;
112 TW = Tp - Tu;
113 }
114 cr[0] = T7 + Te;
115 ci[0] = T1b + T1e;
116 {
117 E TS, TX, TT, TY, TP, TU;
118 TS = FNMS(KP707106781, TR, TQ);
119 TX = FMA(KP707106781, TW, TV);
120 TP = W[4];
121 TT = TP * TS;
122 TY = TP * TX;
123 TU = W[5];
124 cr[WS(rs, 3)] = FNMS(TU, TX, TT);
125 ci[WS(rs, 3)] = FMA(TU, TS, TY);
126 }
127 {
128 E T1s, T1v, T1t, T1w, T1r, T1u;
129 T1s = T1i + T1j;
130 T1v = T1o + T1n;
131 T1r = W[2];
132 T1t = T1r * T1s;
133 T1w = T1r * T1v;
134 T1u = W[3];
135 cr[WS(rs, 2)] = FNMS(T1u, T1v, T1t);
136 ci[WS(rs, 2)] = FMA(T1u, T1s, T1w);
137 }
138 {
139 E T10, T13, T11, T14, TZ, T12;
140 T10 = FMA(KP707106781, TR, TQ);
141 T13 = FNMS(KP707106781, TW, TV);
142 TZ = W[12];
143 T11 = TZ * T10;
144 T14 = TZ * T13;
145 T12 = W[13];
146 cr[WS(rs, 7)] = FNMS(T12, T13, T11);
147 ci[WS(rs, 7)] = FMA(T12, T10, T14);
148 }
149 {
150 E T1f, T15, T17, T18, T1g, T16;
151 T1f = T1b - T1e;
152 T16 = T7 - Te;
153 T15 = W[6];
154 T17 = T15 * T16;
155 T18 = W[7];
156 T1g = T18 * T16;
157 cr[WS(rs, 4)] = FNMS(T18, T1f, T17);
158 ci[WS(rs, 4)] = FMA(T15, T1f, T1g);
159 }
160 {
161 E T1k, T1p, T1l, T1q, T1h, T1m;
162 T1k = T1i - T1j;
163 T1p = T1n - T1o;
164 T1h = W[10];
165 T1l = T1h * T1k;
166 T1q = T1h * T1p;
167 T1m = W[11];
168 cr[WS(rs, 6)] = FNMS(T1m, T1p, T1l);
169 ci[WS(rs, 6)] = FMA(T1m, T1k, T1q);
170 }
171 {
172 E TH, TN, TJ, TL, TM, TO, Tf, Tx, Ty, TI, TG, TK, Tw;
173 TG = TE - TF;
174 TH = FNMS(KP707106781, TG, TD);
175 TN = FMA(KP707106781, TG, TD);
176 TK = FMA(KP707106781, Tv, Tk);
177 TJ = W[0];
178 TL = TJ * TK;
179 TM = W[1];
180 TO = TM * TK;
181 Tw = FNMS(KP707106781, Tv, Tk);
182 Tf = W[8];
183 Tx = Tf * Tw;
184 Ty = W[9];
185 TI = Ty * Tw;
186 cr[WS(rs, 5)] = FNMS(Ty, TH, Tx);
187 ci[WS(rs, 5)] = FMA(Tf, TH, TI);
188 cr[WS(rs, 1)] = FNMS(TM, TN, TL);
189 ci[WS(rs, 1)] = FMA(TJ, TN, TO);
190 }
191 }
192 }
193 }
194
195 static const tw_instr twinstr[] = {
196 { TW_FULL, 1, 8 },
197 { TW_NEXT, 1, 0 }
198 };
199
200 static const hc2hc_desc desc = { 8, "hb_8", twinstr, &GENUS, { 44, 14, 22, 0 } };
201
X(codelet_hb_8)202 void X(codelet_hb_8) (planner *p) {
203 X(khc2hc_register) (p, hb_8, &desc);
204 }
205 #else
206
207 /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 8 -dif -name hb_8 -include rdft/scalar/hb.h */
208
209 /*
210 * This function contains 66 FP additions, 32 FP multiplications,
211 * (or, 52 additions, 18 multiplications, 14 fused multiply/add),
212 * 30 stack variables, 1 constants, and 32 memory accesses
213 */
214 #include "rdft/scalar/hb.h"
215
hb_8(R * cr,R * ci,const R * W,stride rs,INT mb,INT me,INT ms)216 static void hb_8(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms)
217 {
218 DK(KP707106781, +0.707106781186547524400844362104849039284835938);
219 {
220 INT m;
221 for (m = mb, W = W + ((mb - 1) * 14); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 14, MAKE_VOLATILE_STRIDE(16, rs)) {
222 E T7, T18, T1c, To, Ty, TM, TY, TC, Te, TZ, T10, Tv, Tz, TP, TS;
223 E TD;
224 {
225 E T3, TK, Tn, TL, T6, TW, Tk, TX;
226 {
227 E T1, T2, Tl, Tm;
228 T1 = cr[0];
229 T2 = ci[WS(rs, 3)];
230 T3 = T1 + T2;
231 TK = T1 - T2;
232 Tl = ci[WS(rs, 5)];
233 Tm = cr[WS(rs, 6)];
234 Tn = Tl - Tm;
235 TL = Tl + Tm;
236 }
237 {
238 E T4, T5, Ti, Tj;
239 T4 = cr[WS(rs, 2)];
240 T5 = ci[WS(rs, 1)];
241 T6 = T4 + T5;
242 TW = T4 - T5;
243 Ti = ci[WS(rs, 7)];
244 Tj = cr[WS(rs, 4)];
245 Tk = Ti - Tj;
246 TX = Ti + Tj;
247 }
248 T7 = T3 + T6;
249 T18 = TK + TL;
250 T1c = TX - TW;
251 To = Tk + Tn;
252 Ty = T3 - T6;
253 TM = TK - TL;
254 TY = TW + TX;
255 TC = Tk - Tn;
256 }
257 {
258 E Ta, TN, Tu, TR, Td, TQ, Tr, TO;
259 {
260 E T8, T9, Ts, Tt;
261 T8 = cr[WS(rs, 1)];
262 T9 = ci[WS(rs, 2)];
263 Ta = T8 + T9;
264 TN = T8 - T9;
265 Ts = ci[WS(rs, 4)];
266 Tt = cr[WS(rs, 7)];
267 Tu = Ts - Tt;
268 TR = Ts + Tt;
269 }
270 {
271 E Tb, Tc, Tp, Tq;
272 Tb = ci[0];
273 Tc = cr[WS(rs, 3)];
274 Td = Tb + Tc;
275 TQ = Tb - Tc;
276 Tp = ci[WS(rs, 6)];
277 Tq = cr[WS(rs, 5)];
278 Tr = Tp - Tq;
279 TO = Tp + Tq;
280 }
281 Te = Ta + Td;
282 TZ = TN + TO;
283 T10 = TQ + TR;
284 Tv = Tr + Tu;
285 Tz = Tu - Tr;
286 TP = TN - TO;
287 TS = TQ - TR;
288 TD = Ta - Td;
289 }
290 cr[0] = T7 + Te;
291 ci[0] = To + Tv;
292 {
293 E Tg, Tw, Tf, Th;
294 Tg = T7 - Te;
295 Tw = To - Tv;
296 Tf = W[6];
297 Th = W[7];
298 cr[WS(rs, 4)] = FNMS(Th, Tw, Tf * Tg);
299 ci[WS(rs, 4)] = FMA(Th, Tg, Tf * Tw);
300 }
301 {
302 E TG, TI, TF, TH;
303 TG = Ty + Tz;
304 TI = TD + TC;
305 TF = W[2];
306 TH = W[3];
307 cr[WS(rs, 2)] = FNMS(TH, TI, TF * TG);
308 ci[WS(rs, 2)] = FMA(TF, TI, TH * TG);
309 }
310 {
311 E TA, TE, Tx, TB;
312 TA = Ty - Tz;
313 TE = TC - TD;
314 Tx = W[10];
315 TB = W[11];
316 cr[WS(rs, 6)] = FNMS(TB, TE, Tx * TA);
317 ci[WS(rs, 6)] = FMA(Tx, TE, TB * TA);
318 }
319 {
320 E T1a, T1g, T1e, T1i, T19, T1d;
321 T19 = KP707106781 * (TZ + T10);
322 T1a = T18 - T19;
323 T1g = T18 + T19;
324 T1d = KP707106781 * (TP - TS);
325 T1e = T1c + T1d;
326 T1i = T1c - T1d;
327 {
328 E T17, T1b, T1f, T1h;
329 T17 = W[4];
330 T1b = W[5];
331 cr[WS(rs, 3)] = FNMS(T1b, T1e, T17 * T1a);
332 ci[WS(rs, 3)] = FMA(T17, T1e, T1b * T1a);
333 T1f = W[12];
334 T1h = W[13];
335 cr[WS(rs, 7)] = FNMS(T1h, T1i, T1f * T1g);
336 ci[WS(rs, 7)] = FMA(T1f, T1i, T1h * T1g);
337 }
338 }
339 {
340 E TU, T14, T12, T16, TT, T11;
341 TT = KP707106781 * (TP + TS);
342 TU = TM - TT;
343 T14 = TM + TT;
344 T11 = KP707106781 * (TZ - T10);
345 T12 = TY - T11;
346 T16 = TY + T11;
347 {
348 E TJ, TV, T13, T15;
349 TJ = W[8];
350 TV = W[9];
351 cr[WS(rs, 5)] = FNMS(TV, T12, TJ * TU);
352 ci[WS(rs, 5)] = FMA(TV, TU, TJ * T12);
353 T13 = W[0];
354 T15 = W[1];
355 cr[WS(rs, 1)] = FNMS(T15, T16, T13 * T14);
356 ci[WS(rs, 1)] = FMA(T15, T14, T13 * T16);
357 }
358 }
359 }
360 }
361 }
362
363 static const tw_instr twinstr[] = {
364 { TW_FULL, 1, 8 },
365 { TW_NEXT, 1, 0 }
366 };
367
368 static const hc2hc_desc desc = { 8, "hb_8", twinstr, &GENUS, { 52, 18, 14, 0 } };
369
X(codelet_hb_8)370 void X(codelet_hb_8) (planner *p) {
371 X(khc2hc_register) (p, hb_8, &desc);
372 }
373 #endif
374