mirror of
https://github.com/tildearrow/furnace.git
synced 2024-11-05 20:35:06 +00:00
54e93db207
not reliable yet
545 lines
14 KiB
C
545 lines
14 KiB
C
/*
|
|
* Copyright (c) 2003, 2007-14 Matteo Frigo
|
|
* Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
|
|
*
|
|
* This program is free software; you can redistribute it and/or modify
|
|
* it under the terms of the GNU General Public License as published by
|
|
* the Free Software Foundation; either version 2 of the License, or
|
|
* (at your option) any later version.
|
|
*
|
|
* This program is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
* GNU General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License
|
|
* along with this program; if not, write to the Free Software
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
*
|
|
*/
|
|
|
|
/* This file was automatically generated --- DO NOT EDIT */
|
|
/* Generated on Tue Sep 14 10:47:12 EDT 2021 */
|
|
|
|
#include "rdft/codelet-rdft.h"
|
|
|
|
#if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
|
|
|
|
/* Generated by: ../../../genfft/gen_hc2cdft.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 10 -dif -name hc2cbdft_10 -include rdft/scalar/hc2cb.h */
|
|
|
|
/*
|
|
* This function contains 122 FP additions, 72 FP multiplications,
|
|
* (or, 68 additions, 18 multiplications, 54 fused multiply/add),
|
|
* 91 stack variables, 4 constants, and 40 memory accesses
|
|
*/
|
|
#include "rdft/scalar/hc2cb.h"
|
|
|
|
static void hc2cbdft_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
|
|
{
|
|
DK(KP951056516, +0.951056516295153572116439333379382143405698634);
|
|
DK(KP559016994, +0.559016994374947424102293417182819058860154590);
|
|
DK(KP618033988, +0.618033988749894848204586834365638117720309180);
|
|
DK(KP250000000, +0.250000000000000000000000000000000000000000000);
|
|
{
|
|
INT m;
|
|
for (m = mb, W = W + ((mb - 1) * 18); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 18, MAKE_VOLATILE_STRIDE(40, rs)) {
|
|
E T3, Tl, Tu, T14, Ti, T13, Ts, Tt, T1p, T23, TZ, T1z, TQ, T1g, TV;
|
|
E T1l, TT, TU, T1j, T1k, T1c, T1Y, TK, T1u;
|
|
{
|
|
E Td, Tp, Tg, Tq, Th, Tr, T6, Tm, T9, Tn, Ta, To, T1, T2;
|
|
T1 = Rp[0];
|
|
T2 = Rm[WS(rs, 4)];
|
|
T3 = T1 + T2;
|
|
Tl = T1 - T2;
|
|
{
|
|
E Tb, Tc, Te, Tf;
|
|
Tb = Rp[WS(rs, 4)];
|
|
Tc = Rm[0];
|
|
Td = Tb + Tc;
|
|
Tp = Tb - Tc;
|
|
Te = Rm[WS(rs, 3)];
|
|
Tf = Rp[WS(rs, 1)];
|
|
Tg = Te + Tf;
|
|
Tq = Te - Tf;
|
|
}
|
|
Th = Td + Tg;
|
|
Tr = Tp + Tq;
|
|
{
|
|
E T4, T5, T7, T8;
|
|
T4 = Rp[WS(rs, 2)];
|
|
T5 = Rm[WS(rs, 2)];
|
|
T6 = T4 + T5;
|
|
Tm = T4 - T5;
|
|
T7 = Rm[WS(rs, 1)];
|
|
T8 = Rp[WS(rs, 3)];
|
|
T9 = T7 + T8;
|
|
Tn = T7 - T8;
|
|
}
|
|
Ta = T6 + T9;
|
|
To = Tm + Tn;
|
|
Tu = To - Tr;
|
|
T14 = Ta - Th;
|
|
Ti = Ta + Th;
|
|
T13 = FNMS(KP250000000, Ti, T3);
|
|
Ts = To + Tr;
|
|
Tt = FNMS(KP250000000, Ts, Tl);
|
|
{
|
|
E T1n, T1o, TX, TY;
|
|
T1n = Td - Tg;
|
|
T1o = T6 - T9;
|
|
T1p = FNMS(KP618033988, T1o, T1n);
|
|
T23 = FMA(KP618033988, T1n, T1o);
|
|
TX = Tm - Tn;
|
|
TY = Tp - Tq;
|
|
TZ = FMA(KP618033988, TY, TX);
|
|
T1z = FNMS(KP618033988, TX, TY);
|
|
}
|
|
}
|
|
{
|
|
E TF, T16, TI, T17, TS, T1i, Ty, T19, TB, T1a, TR, T1h, TO, TP;
|
|
TO = Ip[0];
|
|
TP = Im[WS(rs, 4)];
|
|
TQ = TO + TP;
|
|
T1g = TO - TP;
|
|
{
|
|
E TD, TE, TG, TH;
|
|
TD = Ip[WS(rs, 4)];
|
|
TE = Im[0];
|
|
TF = TD + TE;
|
|
T16 = TD - TE;
|
|
TG = Im[WS(rs, 3)];
|
|
TH = Ip[WS(rs, 1)];
|
|
TI = TG + TH;
|
|
T17 = TH - TG;
|
|
}
|
|
TS = TF - TI;
|
|
T1i = T16 + T17;
|
|
{
|
|
E Tw, Tx, Tz, TA;
|
|
Tw = Ip[WS(rs, 2)];
|
|
Tx = Im[WS(rs, 2)];
|
|
Ty = Tw + Tx;
|
|
T19 = Tw - Tx;
|
|
Tz = Im[WS(rs, 1)];
|
|
TA = Ip[WS(rs, 3)];
|
|
TB = Tz + TA;
|
|
T1a = TA - Tz;
|
|
}
|
|
TR = Ty - TB;
|
|
T1h = T19 + T1a;
|
|
TV = TR - TS;
|
|
T1l = T1h - T1i;
|
|
TT = TR + TS;
|
|
TU = FNMS(KP250000000, TT, TQ);
|
|
T1j = T1h + T1i;
|
|
T1k = FNMS(KP250000000, T1j, T1g);
|
|
{
|
|
E T18, T1b, TC, TJ;
|
|
T18 = T16 - T17;
|
|
T1b = T19 - T1a;
|
|
T1c = FNMS(KP618033988, T1b, T18);
|
|
T1Y = FMA(KP618033988, T18, T1b);
|
|
TC = Ty + TB;
|
|
TJ = TF + TI;
|
|
TK = FMA(KP618033988, TJ, TC);
|
|
T1u = FNMS(KP618033988, TC, TJ);
|
|
}
|
|
}
|
|
{
|
|
E Tj, T2y, T2a, T1A, T2q, T10, T1Q, T24, T2k, T1q, T1K, T26, T28, T29, T2c;
|
|
E Tk, TM, TN, T2w, T1M, T1O, T1P, T1S, T1s, T1w, T1x, T1C, T2m, T2o, T2p;
|
|
E T2s, T12, T1e, T1f, T1E, T1G, T1I, T1J, T1U, T1W, T20, T21, T2e, T2g, T2i;
|
|
E T2j, T2u, T1y, TW, T22, T2l, T2r;
|
|
Tj = T3 + Ti;
|
|
T2y = T1g + T1j;
|
|
T2a = TQ + TT;
|
|
T1y = FNMS(KP559016994, TV, TU);
|
|
T1A = FMA(KP951056516, T1z, T1y);
|
|
T2q = FNMS(KP951056516, T1z, T1y);
|
|
TW = FMA(KP559016994, TV, TU);
|
|
T10 = FMA(KP951056516, TZ, TW);
|
|
T1Q = FNMS(KP951056516, TZ, TW);
|
|
T22 = FMA(KP559016994, T1l, T1k);
|
|
T24 = FNMS(KP951056516, T23, T22);
|
|
T2k = FMA(KP951056516, T23, T22);
|
|
{
|
|
E T1m, T1v, T2n, T1t;
|
|
T1m = FNMS(KP559016994, T1l, T1k);
|
|
T1q = FNMS(KP951056516, T1p, T1m);
|
|
T1K = FMA(KP951056516, T1p, T1m);
|
|
{
|
|
E T27, TL, T1N, Tv;
|
|
T27 = Tl + Ts;
|
|
T26 = W[9];
|
|
T28 = T26 * T27;
|
|
T29 = W[8];
|
|
T2c = T29 * T27;
|
|
Tv = FMA(KP559016994, Tu, Tt);
|
|
TL = FNMS(KP951056516, TK, Tv);
|
|
T1N = FMA(KP951056516, TK, Tv);
|
|
Tk = W[1];
|
|
TM = Tk * TL;
|
|
TN = W[0];
|
|
T2w = TN * TL;
|
|
T1M = W[17];
|
|
T1O = T1M * T1N;
|
|
T1P = W[16];
|
|
T1S = T1P * T1N;
|
|
}
|
|
T1t = FNMS(KP559016994, Tu, Tt);
|
|
T1v = FNMS(KP951056516, T1u, T1t);
|
|
T2n = FMA(KP951056516, T1u, T1t);
|
|
T1s = W[5];
|
|
T1w = T1s * T1v;
|
|
T1x = W[4];
|
|
T1C = T1x * T1v;
|
|
T2m = W[13];
|
|
T2o = T2m * T2n;
|
|
T2p = W[12];
|
|
T2s = T2p * T2n;
|
|
{
|
|
E T1d, T1H, T15, T1Z, T2h, T1X;
|
|
T15 = FNMS(KP559016994, T14, T13);
|
|
T1d = FMA(KP951056516, T1c, T15);
|
|
T1H = FNMS(KP951056516, T1c, T15);
|
|
T12 = W[2];
|
|
T1e = T12 * T1d;
|
|
T1f = W[3];
|
|
T1E = T1f * T1d;
|
|
T1G = W[14];
|
|
T1I = T1G * T1H;
|
|
T1J = W[15];
|
|
T1U = T1J * T1H;
|
|
T1X = FMA(KP559016994, T14, T13);
|
|
T1Z = FMA(KP951056516, T1Y, T1X);
|
|
T2h = FNMS(KP951056516, T1Y, T1X);
|
|
T1W = W[6];
|
|
T20 = T1W * T1Z;
|
|
T21 = W[7];
|
|
T2e = T21 * T1Z;
|
|
T2g = W[10];
|
|
T2i = T2g * T2h;
|
|
T2j = W[11];
|
|
T2u = T2j * T2h;
|
|
}
|
|
}
|
|
{
|
|
E T11, T2x, T1r, T1B;
|
|
T11 = FMA(TN, T10, TM);
|
|
Rp[0] = Tj - T11;
|
|
Rm[0] = Tj + T11;
|
|
T2x = FNMS(Tk, T10, T2w);
|
|
Im[0] = T2x - T2y;
|
|
Ip[0] = T2x + T2y;
|
|
T1r = FNMS(T1f, T1q, T1e);
|
|
T1B = FMA(T1x, T1A, T1w);
|
|
Rp[WS(rs, 1)] = T1r - T1B;
|
|
Rm[WS(rs, 1)] = T1B + T1r;
|
|
{
|
|
E T1D, T1F, T1L, T1R;
|
|
T1D = FNMS(T1s, T1A, T1C);
|
|
T1F = FMA(T12, T1q, T1E);
|
|
Im[WS(rs, 1)] = T1D - T1F;
|
|
Ip[WS(rs, 1)] = T1D + T1F;
|
|
T1L = FNMS(T1J, T1K, T1I);
|
|
T1R = FMA(T1P, T1Q, T1O);
|
|
Rp[WS(rs, 4)] = T1L - T1R;
|
|
Rm[WS(rs, 4)] = T1R + T1L;
|
|
}
|
|
}
|
|
{
|
|
E T1T, T1V, T2t, T2v;
|
|
T1T = FNMS(T1M, T1Q, T1S);
|
|
T1V = FMA(T1G, T1K, T1U);
|
|
Im[WS(rs, 4)] = T1T - T1V;
|
|
Ip[WS(rs, 4)] = T1T + T1V;
|
|
T2t = FNMS(T2m, T2q, T2s);
|
|
T2v = FMA(T2g, T2k, T2u);
|
|
Im[WS(rs, 3)] = T2t - T2v;
|
|
Ip[WS(rs, 3)] = T2t + T2v;
|
|
}
|
|
T2l = FNMS(T2j, T2k, T2i);
|
|
T2r = FMA(T2p, T2q, T2o);
|
|
Rp[WS(rs, 3)] = T2l - T2r;
|
|
Rm[WS(rs, 3)] = T2r + T2l;
|
|
{
|
|
E T25, T2b, T2d, T2f;
|
|
T25 = FNMS(T21, T24, T20);
|
|
T2b = FMA(T29, T2a, T28);
|
|
Rp[WS(rs, 2)] = T25 - T2b;
|
|
Rm[WS(rs, 2)] = T2b + T25;
|
|
T2d = FNMS(T26, T2a, T2c);
|
|
T2f = FMA(T1W, T24, T2e);
|
|
Im[WS(rs, 2)] = T2d - T2f;
|
|
Ip[WS(rs, 2)] = T2d + T2f;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
static const tw_instr twinstr[] = {
|
|
{ TW_FULL, 1, 10 },
|
|
{ TW_NEXT, 1, 0 }
|
|
};
|
|
|
|
static const hc2c_desc desc = { 10, "hc2cbdft_10", twinstr, &GENUS, { 68, 18, 54, 0 } };
|
|
|
|
void X(codelet_hc2cbdft_10) (planner *p) {
|
|
X(khc2c_register) (p, hc2cbdft_10, &desc, HC2C_VIA_DFT);
|
|
}
|
|
#else
|
|
|
|
/* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 10 -dif -name hc2cbdft_10 -include rdft/scalar/hc2cb.h */
|
|
|
|
/*
|
|
* This function contains 122 FP additions, 60 FP multiplications,
|
|
* (or, 92 additions, 30 multiplications, 30 fused multiply/add),
|
|
* 61 stack variables, 4 constants, and 40 memory accesses
|
|
*/
|
|
#include "rdft/scalar/hc2cb.h"
|
|
|
|
static void hc2cbdft_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
|
|
{
|
|
DK(KP951056516, +0.951056516295153572116439333379382143405698634);
|
|
DK(KP587785252, +0.587785252292473129168705954639072768597652438);
|
|
DK(KP250000000, +0.250000000000000000000000000000000000000000000);
|
|
DK(KP559016994, +0.559016994374947424102293417182819058860154590);
|
|
{
|
|
INT m;
|
|
for (m = mb, W = W + ((mb - 1) * 18); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 18, MAKE_VOLATILE_STRIDE(40, rs)) {
|
|
E T3, TS, TR, T13, Ti, T12, TT, TU, T1g, T1T, Tr, T1s, TJ, T1h, TG;
|
|
E T1m, TK, TL, T1k, T1l, T1b, T1P, TY, T1w;
|
|
{
|
|
E Td, To, Tg, Tp, Th, TQ, T6, Tl, T9, Tm, Ta, TP, T1, T2;
|
|
T1 = Rp[0];
|
|
T2 = Rm[WS(rs, 4)];
|
|
T3 = T1 + T2;
|
|
TS = T1 - T2;
|
|
{
|
|
E Tb, Tc, Te, Tf;
|
|
Tb = Rp[WS(rs, 4)];
|
|
Tc = Rm[0];
|
|
Td = Tb + Tc;
|
|
To = Tb - Tc;
|
|
Te = Rm[WS(rs, 3)];
|
|
Tf = Rp[WS(rs, 1)];
|
|
Tg = Te + Tf;
|
|
Tp = Te - Tf;
|
|
}
|
|
Th = Td + Tg;
|
|
TQ = To + Tp;
|
|
{
|
|
E T4, T5, T7, T8;
|
|
T4 = Rp[WS(rs, 2)];
|
|
T5 = Rm[WS(rs, 2)];
|
|
T6 = T4 + T5;
|
|
Tl = T4 - T5;
|
|
T7 = Rm[WS(rs, 1)];
|
|
T8 = Rp[WS(rs, 3)];
|
|
T9 = T7 + T8;
|
|
Tm = T7 - T8;
|
|
}
|
|
Ta = T6 + T9;
|
|
TP = Tl + Tm;
|
|
TR = KP559016994 * (TP - TQ);
|
|
T13 = KP559016994 * (Ta - Th);
|
|
Ti = Ta + Th;
|
|
T12 = FNMS(KP250000000, Ti, T3);
|
|
TT = TP + TQ;
|
|
TU = FNMS(KP250000000, TT, TS);
|
|
{
|
|
E T1e, T1f, Tn, Tq;
|
|
T1e = T6 - T9;
|
|
T1f = Td - Tg;
|
|
T1g = FNMS(KP951056516, T1f, KP587785252 * T1e);
|
|
T1T = FMA(KP951056516, T1e, KP587785252 * T1f);
|
|
Tn = Tl - Tm;
|
|
Tq = To - Tp;
|
|
Tr = FMA(KP951056516, Tn, KP587785252 * Tq);
|
|
T1s = FNMS(KP951056516, Tq, KP587785252 * Tn);
|
|
}
|
|
}
|
|
{
|
|
E TB, T18, TE, T19, TF, T1j, Tu, T15, Tx, T16, Ty, T1i, TH, TI;
|
|
TH = Ip[0];
|
|
TI = Im[WS(rs, 4)];
|
|
TJ = TH + TI;
|
|
T1h = TH - TI;
|
|
{
|
|
E Tz, TA, TC, TD;
|
|
Tz = Ip[WS(rs, 4)];
|
|
TA = Im[0];
|
|
TB = Tz + TA;
|
|
T18 = Tz - TA;
|
|
TC = Im[WS(rs, 3)];
|
|
TD = Ip[WS(rs, 1)];
|
|
TE = TC + TD;
|
|
T19 = TD - TC;
|
|
}
|
|
TF = TB - TE;
|
|
T1j = T18 + T19;
|
|
{
|
|
E Ts, Tt, Tv, Tw;
|
|
Ts = Ip[WS(rs, 2)];
|
|
Tt = Im[WS(rs, 2)];
|
|
Tu = Ts + Tt;
|
|
T15 = Ts - Tt;
|
|
Tv = Im[WS(rs, 1)];
|
|
Tw = Ip[WS(rs, 3)];
|
|
Tx = Tv + Tw;
|
|
T16 = Tw - Tv;
|
|
}
|
|
Ty = Tu - Tx;
|
|
T1i = T15 + T16;
|
|
TG = KP559016994 * (Ty - TF);
|
|
T1m = KP559016994 * (T1i - T1j);
|
|
TK = Ty + TF;
|
|
TL = FNMS(KP250000000, TK, TJ);
|
|
T1k = T1i + T1j;
|
|
T1l = FNMS(KP250000000, T1k, T1h);
|
|
{
|
|
E T17, T1a, TW, TX;
|
|
T17 = T15 - T16;
|
|
T1a = T18 - T19;
|
|
T1b = FNMS(KP951056516, T1a, KP587785252 * T17);
|
|
T1P = FMA(KP951056516, T17, KP587785252 * T1a);
|
|
TW = Tu + Tx;
|
|
TX = TB + TE;
|
|
TY = FMA(KP951056516, TW, KP587785252 * TX);
|
|
T1w = FNMS(KP951056516, TX, KP587785252 * TW);
|
|
}
|
|
}
|
|
{
|
|
E Tj, T2g, TN, T1H, T1U, T26, TZ, T1J, T1Q, T24, T1c, T1C, T1t, T29, T1o;
|
|
E T1E, T1x, T2b, T20, T21, TM, T1S, TV;
|
|
Tj = T3 + Ti;
|
|
T2g = T1h + T1k;
|
|
TM = TG + TL;
|
|
TN = Tr + TM;
|
|
T1H = TM - Tr;
|
|
T1S = T1m + T1l;
|
|
T1U = T1S - T1T;
|
|
T26 = T1T + T1S;
|
|
TV = TR + TU;
|
|
TZ = TV - TY;
|
|
T1J = TV + TY;
|
|
{
|
|
E T1O, T14, T1r, T1n, T1v;
|
|
T1O = T13 + T12;
|
|
T1Q = T1O + T1P;
|
|
T24 = T1O - T1P;
|
|
T14 = T12 - T13;
|
|
T1c = T14 - T1b;
|
|
T1C = T14 + T1b;
|
|
T1r = TL - TG;
|
|
T1t = T1r - T1s;
|
|
T29 = T1s + T1r;
|
|
T1n = T1l - T1m;
|
|
T1o = T1g + T1n;
|
|
T1E = T1n - T1g;
|
|
T1v = TU - TR;
|
|
T1x = T1v + T1w;
|
|
T2b = T1v - T1w;
|
|
{
|
|
E T1X, T1Z, T1W, T1Y;
|
|
T1X = TS + TT;
|
|
T1Z = TJ + TK;
|
|
T1W = W[9];
|
|
T1Y = W[8];
|
|
T20 = FMA(T1W, T1X, T1Y * T1Z);
|
|
T21 = FNMS(T1W, T1Z, T1Y * T1X);
|
|
}
|
|
}
|
|
{
|
|
E T10, T2f, Tk, TO;
|
|
Tk = W[0];
|
|
TO = W[1];
|
|
T10 = FMA(Tk, TN, TO * TZ);
|
|
T2f = FNMS(TO, TN, Tk * TZ);
|
|
Rp[0] = Tj - T10;
|
|
Ip[0] = T2f + T2g;
|
|
Rm[0] = Tj + T10;
|
|
Im[0] = T2f - T2g;
|
|
}
|
|
{
|
|
E T1V, T22, T1N, T1R;
|
|
T1N = W[6];
|
|
T1R = W[7];
|
|
T1V = FNMS(T1R, T1U, T1N * T1Q);
|
|
T22 = FMA(T1R, T1Q, T1N * T1U);
|
|
Rp[WS(rs, 2)] = T1V - T20;
|
|
Ip[WS(rs, 2)] = T21 + T22;
|
|
Rm[WS(rs, 2)] = T20 + T1V;
|
|
Im[WS(rs, 2)] = T21 - T22;
|
|
}
|
|
{
|
|
E T1p, T1A, T1y, T1z;
|
|
{
|
|
E T11, T1d, T1q, T1u;
|
|
T11 = W[2];
|
|
T1d = W[3];
|
|
T1p = FNMS(T1d, T1o, T11 * T1c);
|
|
T1A = FMA(T1d, T1c, T11 * T1o);
|
|
T1q = W[4];
|
|
T1u = W[5];
|
|
T1y = FMA(T1q, T1t, T1u * T1x);
|
|
T1z = FNMS(T1u, T1t, T1q * T1x);
|
|
}
|
|
Rp[WS(rs, 1)] = T1p - T1y;
|
|
Ip[WS(rs, 1)] = T1z + T1A;
|
|
Rm[WS(rs, 1)] = T1y + T1p;
|
|
Im[WS(rs, 1)] = T1z - T1A;
|
|
}
|
|
{
|
|
E T1F, T1M, T1K, T1L;
|
|
{
|
|
E T1B, T1D, T1G, T1I;
|
|
T1B = W[14];
|
|
T1D = W[15];
|
|
T1F = FNMS(T1D, T1E, T1B * T1C);
|
|
T1M = FMA(T1D, T1C, T1B * T1E);
|
|
T1G = W[16];
|
|
T1I = W[17];
|
|
T1K = FMA(T1G, T1H, T1I * T1J);
|
|
T1L = FNMS(T1I, T1H, T1G * T1J);
|
|
}
|
|
Rp[WS(rs, 4)] = T1F - T1K;
|
|
Ip[WS(rs, 4)] = T1L + T1M;
|
|
Rm[WS(rs, 4)] = T1K + T1F;
|
|
Im[WS(rs, 4)] = T1L - T1M;
|
|
}
|
|
{
|
|
E T27, T2e, T2c, T2d;
|
|
{
|
|
E T23, T25, T28, T2a;
|
|
T23 = W[10];
|
|
T25 = W[11];
|
|
T27 = FNMS(T25, T26, T23 * T24);
|
|
T2e = FMA(T25, T24, T23 * T26);
|
|
T28 = W[12];
|
|
T2a = W[13];
|
|
T2c = FMA(T28, T29, T2a * T2b);
|
|
T2d = FNMS(T2a, T29, T28 * T2b);
|
|
}
|
|
Rp[WS(rs, 3)] = T27 - T2c;
|
|
Ip[WS(rs, 3)] = T2d + T2e;
|
|
Rm[WS(rs, 3)] = T2c + T27;
|
|
Im[WS(rs, 3)] = T2d - T2e;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
static const tw_instr twinstr[] = {
|
|
{ TW_FULL, 1, 10 },
|
|
{ TW_NEXT, 1, 0 }
|
|
};
|
|
|
|
static const hc2c_desc desc = { 10, "hc2cbdft_10", twinstr, &GENUS, { 92, 30, 30, 0 } };
|
|
|
|
void X(codelet_hc2cbdft_10) (planner *p) {
|
|
X(khc2c_register) (p, hc2cbdft_10, &desc, HC2C_VIA_DFT);
|
|
}
|
|
#endif
|