Mercurial > hg > js-dsp-test

comparison fft/fftw/fftw-3.3.4/rdft/scalar/r2cf/hc2cfdft2_4.c @ 19:26056e866c29

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add FFTW to comparison table
author Chris Cannam
date Tue, 06 Oct 2015 13:08:39 +0100
parents
children
comparison
equal deleted inserted replaced
18:8db794ca3e0b 19:26056e866c29
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 Tue Mar 4 13:49:31 EST 2014 */
23
24 #include "codelet-rdft.h"
25
26 #ifdef HAVE_FMA
27
28 /* Generated by: ../../../genfft/gen_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 4 -dit -name hc2cfdft2_4 -include hc2cf.h */
29
30 /*
31 * This function contains 32 FP additions, 24 FP multiplications,
32 * (or, 24 additions, 16 multiplications, 8 fused multiply/add),
33 * 33 stack variables, 1 constants, and 16 memory accesses
34 */
35 #include "hc2cf.h"
36
37 static void hc2cfdft2_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
38 {
39 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
40 {
41 INT m;
42 for (m = mb, W = W + ((mb - 1) * 4); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 4, MAKE_VOLATILE_STRIDE(16, rs)) {
43 E T1, T5, T2, T4;
44 T1 = W[0];
45 T5 = W[3];
46 T2 = W[2];
47 T4 = W[1];
48 {
49 E Tc, T6, Tp, Tj, Tw, Tt, T9, TE, To, TC, Ta, Tr, Tf, Tl, Tm;
50 {
51 E Th, Tb, T3, Ti;
52 Th = Ip[0];
53 Tb = T1 * T5;
54 T3 = T1 * T2;
55 Ti = Im[0];
56 Tl = Rm[0];
57 Tc = FNMS(T4, T2, Tb);
58 T6 = FMA(T4, T5, T3);
59 Tp = Th + Ti;
60 Tj = Th - Ti;
61 Tm = Rp[0];
62 }
63 {
64 E T7, T8, Td, Tn, Te;
65 T7 = Ip[WS(rs, 1)];
66 T8 = Im[WS(rs, 1)];
67 Td = Rp[WS(rs, 1)];
68 Tw = Tm + Tl;
69 Tn = Tl - Tm;
70 Tt = T7 + T8;
71 T9 = T7 - T8;
72 Te = Rm[WS(rs, 1)];
73 TE = T4 * Tn;
74 To = T1 * Tn;
75 TC = T2 * Tt;
76 Ta = T6 * T9;
77 Tr = Td - Te;
78 Tf = Td + Te;
79 }
80 {
81 E Tq, Tk, TB, Ty, Tu, TI, TG, TF;
82 Tq = FNMS(T4, Tp, To);
83 TF = FMA(T1, Tp, TE);
84 {
85 E Tg, Tx, TD, Ts;
86 Tg = FNMS(Tc, Tf, Ta);
87 Tx = T6 * Tf;
88 TD = FNMS(T5, Tr, TC);
89 Ts = T2 * Tr;
90 Tk = Tg + Tj;
91 TB = Tj - Tg;
92 Ty = FMA(Tc, T9, Tx);
93 Tu = FMA(T5, Tt, Ts);
94 TI = TD + TF;
95 TG = TD - TF;
96 }
97 {
98 E Tz, TH, Tv, TA;
99 Tz = Tw - Ty;
100 TH = Tw + Ty;
101 Tv = Tq - Tu;
102 TA = Tu + Tq;
103 Rp[0] = KP500000000 * (TH + TI);
104 Rm[WS(rs, 1)] = KP500000000 * (TH - TI);
105 Rm[0] = KP500000000 * (Tz - TA);
106 Im[WS(rs, 1)] = KP500000000 * (Tv - Tk);
107 Ip[0] = KP500000000 * (Tk + Tv);
108 Im[0] = KP500000000 * (TG - TB);
109 Rp[WS(rs, 1)] = KP500000000 * (Tz + TA);
110 Ip[WS(rs, 1)] = KP500000000 * (TB + TG);
111 }
112 }
113 }
114 }
115 }
116 }
117
118 static const tw_instr twinstr[] = {
119 {TW_CEXP, 1, 1},
120 {TW_CEXP, 1, 3},
121 {TW_NEXT, 1, 0}
122 };
123
124 static const hc2c_desc desc = { 4, "hc2cfdft2_4", twinstr, &GENUS, {24, 16, 8, 0} };
125
126 void X(codelet_hc2cfdft2_4) (planner *p) {
127 X(khc2c_register) (p, hc2cfdft2_4, &desc, HC2C_VIA_DFT);
128 }
129 #else /* HAVE_FMA */
130
131 /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 4 -dit -name hc2cfdft2_4 -include hc2cf.h */
132
133 /*
134 * This function contains 32 FP additions, 24 FP multiplications,
135 * (or, 24 additions, 16 multiplications, 8 fused multiply/add),
136 * 24 stack variables, 1 constants, and 16 memory accesses
137 */
138 #include "hc2cf.h"
139
140 static void hc2cfdft2_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
141 {
142 DK(KP500000000, +0.500000000000000000000000000000000000000000000);
143 {
144 INT m;
145 for (m = mb, W = W + ((mb - 1) * 4); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 4, MAKE_VOLATILE_STRIDE(16, rs)) {
146 E T1, T3, T2, T4, T5, T9;
147 T1 = W[0];
148 T3 = W[1];
149 T2 = W[2];
150 T4 = W[3];
151 T5 = FMA(T1, T2, T3 * T4);
152 T9 = FNMS(T3, T2, T1 * T4);
153 {
154 E Tg, Tr, Tm, Tx, Td, Tw, Tp, Ts;
155 {
156 E Te, Tf, Tl, Ti, Tj, Tk;
157 Te = Ip[0];
158 Tf = Im[0];
159 Tl = Te + Tf;
160 Ti = Rm[0];
161 Tj = Rp[0];
162 Tk = Ti - Tj;
163 Tg = Te - Tf;
164 Tr = Tj + Ti;
165 Tm = FNMS(T3, Tl, T1 * Tk);
166 Tx = FMA(T3, Tk, T1 * Tl);
167 }
168 {
169 E T8, To, Tc, Tn;
170 {
171 E T6, T7, Ta, Tb;
172 T6 = Ip[WS(rs, 1)];
173 T7 = Im[WS(rs, 1)];
174 T8 = T6 - T7;
175 To = T6 + T7;
176 Ta = Rp[WS(rs, 1)];
177 Tb = Rm[WS(rs, 1)];
178 Tc = Ta + Tb;
179 Tn = Ta - Tb;
180 }
181 Td = FNMS(T9, Tc, T5 * T8);
182 Tw = FNMS(T4, Tn, T2 * To);
183 Tp = FMA(T2, Tn, T4 * To);
184 Ts = FMA(T5, Tc, T9 * T8);
185 }
186 {
187 E Th, Tq, Tz, TA;
188 Th = Td + Tg;
189 Tq = Tm - Tp;
190 Ip[0] = KP500000000 * (Th + Tq);
191 Im[WS(rs, 1)] = KP500000000 * (Tq - Th);
192 Tz = Tr + Ts;
193 TA = Tw + Tx;
194 Rm[WS(rs, 1)] = KP500000000 * (Tz - TA);
195 Rp[0] = KP500000000 * (Tz + TA);
196 }
197 {
198 E Tt, Tu, Tv, Ty;
199 Tt = Tr - Ts;
200 Tu = Tp + Tm;
201 Rm[0] = KP500000000 * (Tt - Tu);
202 Rp[WS(rs, 1)] = KP500000000 * (Tt + Tu);
203 Tv = Tg - Td;
204 Ty = Tw - Tx;
205 Ip[WS(rs, 1)] = KP500000000 * (Tv + Ty);
206 Im[0] = KP500000000 * (Ty - Tv);
207 }
208 }
209 }
210 }
211 }
212
213 static const tw_instr twinstr[] = {
214 {TW_CEXP, 1, 1},
215 {TW_CEXP, 1, 3},
216 {TW_NEXT, 1, 0}
217 };
218
219 static const hc2c_desc desc = { 4, "hc2cfdft2_4", twinstr, &GENUS, {24, 16, 8, 0} };
220
221 void X(codelet_hc2cfdft2_4) (planner *p) {
222 X(khc2c_register) (p, hc2cfdft2_4, &desc, HC2C_VIA_DFT);
223 }
224 #endif /* HAVE_FMA */