mirror of
https://github.com/tildearrow/furnace.git
synced 2024-11-05 12:25:05 +00:00
489 lines
13 KiB
C
489 lines
13 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 10:46:13 EDT 2021 */
|
|
|
|
#include "rdft/codelet-rdft.h"
|
|
|
|
#if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
|
|
|
|
/* Generated by: ../../../genfft/gen_hc2hc.native -fma -compact -variables 4 -pipeline-latency 4 -n 10 -dit -name hf_10 -include rdft/scalar/hf.h */
|
|
|
|
/*
|
|
* This function contains 102 FP additions, 72 FP multiplications,
|
|
* (or, 48 additions, 18 multiplications, 54 fused multiply/add),
|
|
* 47 stack variables, 4 constants, and 40 memory accesses
|
|
*/
|
|
#include "rdft/scalar/hf.h"
|
|
|
|
static void hf_10(R *cr, R *ci, 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, cr = cr + ms, ci = ci - ms, W = W + 18, MAKE_VOLATILE_STRIDE(20, rs)) {
|
|
E T8, T23, T12, T1U, TM, TZ, T10, T1F, T1G, T25, T16, T17, T18, T1s, T1x;
|
|
E T1P, Tl, Ty, Tz, T1I, T1J, T24, T13, T14, T15, T1h, T1m, T1O;
|
|
{
|
|
E T1, T1R, T3, T6, T4, T1S, T2, T7, T1T, T5;
|
|
T1 = cr[0];
|
|
T1R = ci[0];
|
|
T3 = cr[WS(rs, 5)];
|
|
T6 = ci[WS(rs, 5)];
|
|
T2 = W[8];
|
|
T4 = T2 * T3;
|
|
T1S = T2 * T6;
|
|
T5 = W[9];
|
|
T7 = FMA(T5, T6, T4);
|
|
T1T = FNMS(T5, T3, T1S);
|
|
T8 = T1 - T7;
|
|
T23 = T1T + T1R;
|
|
T12 = T1 + T7;
|
|
T1U = T1R - T1T;
|
|
}
|
|
{
|
|
E TF, T1w, TY, T1p, TL, T1u, TS, T1r;
|
|
{
|
|
E TB, TE, TC, T1v, TA, TD;
|
|
TB = cr[WS(rs, 4)];
|
|
TE = ci[WS(rs, 4)];
|
|
TA = W[6];
|
|
TC = TA * TB;
|
|
T1v = TA * TE;
|
|
TD = W[7];
|
|
TF = FMA(TD, TE, TC);
|
|
T1w = FNMS(TD, TB, T1v);
|
|
}
|
|
{
|
|
E TU, TX, TV, T1o, TT, TW;
|
|
TU = cr[WS(rs, 1)];
|
|
TX = ci[WS(rs, 1)];
|
|
TT = W[0];
|
|
TV = TT * TU;
|
|
T1o = TT * TX;
|
|
TW = W[1];
|
|
TY = FMA(TW, TX, TV);
|
|
T1p = FNMS(TW, TU, T1o);
|
|
}
|
|
{
|
|
E TH, TK, TI, T1t, TG, TJ;
|
|
TH = cr[WS(rs, 9)];
|
|
TK = ci[WS(rs, 9)];
|
|
TG = W[16];
|
|
TI = TG * TH;
|
|
T1t = TG * TK;
|
|
TJ = W[17];
|
|
TL = FMA(TJ, TK, TI);
|
|
T1u = FNMS(TJ, TH, T1t);
|
|
}
|
|
{
|
|
E TO, TR, TP, T1q, TN, TQ;
|
|
TO = cr[WS(rs, 6)];
|
|
TR = ci[WS(rs, 6)];
|
|
TN = W[10];
|
|
TP = TN * TO;
|
|
T1q = TN * TR;
|
|
TQ = W[11];
|
|
TS = FMA(TQ, TR, TP);
|
|
T1r = FNMS(TQ, TO, T1q);
|
|
}
|
|
TM = TF - TL;
|
|
TZ = TS - TY;
|
|
T10 = TM + TZ;
|
|
T1F = T1w + T1u;
|
|
T1G = T1r + T1p;
|
|
T25 = T1F + T1G;
|
|
T16 = TF + TL;
|
|
T17 = TS + TY;
|
|
T18 = T16 + T17;
|
|
T1s = T1p - T1r;
|
|
T1x = T1u - T1w;
|
|
T1P = T1x + T1s;
|
|
}
|
|
{
|
|
E Te, T1l, Tx, T1e, Tk, T1j, Tr, T1g;
|
|
{
|
|
E Ta, Td, Tb, T1k, T9, Tc;
|
|
Ta = cr[WS(rs, 2)];
|
|
Td = ci[WS(rs, 2)];
|
|
T9 = W[2];
|
|
Tb = T9 * Ta;
|
|
T1k = T9 * Td;
|
|
Tc = W[3];
|
|
Te = FMA(Tc, Td, Tb);
|
|
T1l = FNMS(Tc, Ta, T1k);
|
|
}
|
|
{
|
|
E Tt, Tw, Tu, T1d, Ts, Tv;
|
|
Tt = cr[WS(rs, 3)];
|
|
Tw = ci[WS(rs, 3)];
|
|
Ts = W[4];
|
|
Tu = Ts * Tt;
|
|
T1d = Ts * Tw;
|
|
Tv = W[5];
|
|
Tx = FMA(Tv, Tw, Tu);
|
|
T1e = FNMS(Tv, Tt, T1d);
|
|
}
|
|
{
|
|
E Tg, Tj, Th, T1i, Tf, Ti;
|
|
Tg = cr[WS(rs, 7)];
|
|
Tj = ci[WS(rs, 7)];
|
|
Tf = W[12];
|
|
Th = Tf * Tg;
|
|
T1i = Tf * Tj;
|
|
Ti = W[13];
|
|
Tk = FMA(Ti, Tj, Th);
|
|
T1j = FNMS(Ti, Tg, T1i);
|
|
}
|
|
{
|
|
E Tn, Tq, To, T1f, Tm, Tp;
|
|
Tn = cr[WS(rs, 8)];
|
|
Tq = ci[WS(rs, 8)];
|
|
Tm = W[14];
|
|
To = Tm * Tn;
|
|
T1f = Tm * Tq;
|
|
Tp = W[15];
|
|
Tr = FMA(Tp, Tq, To);
|
|
T1g = FNMS(Tp, Tn, T1f);
|
|
}
|
|
Tl = Te - Tk;
|
|
Ty = Tr - Tx;
|
|
Tz = Tl + Ty;
|
|
T1I = T1l + T1j;
|
|
T1J = T1g + T1e;
|
|
T24 = T1I + T1J;
|
|
T13 = Te + Tk;
|
|
T14 = Tr + Tx;
|
|
T15 = T13 + T14;
|
|
T1h = T1e - T1g;
|
|
T1m = T1j - T1l;
|
|
T1O = T1m + T1h;
|
|
}
|
|
{
|
|
E T1b, T11, T1a, T1z, T1B, T1n, T1y, T1A, T1c;
|
|
T1b = Tz - T10;
|
|
T11 = Tz + T10;
|
|
T1a = FNMS(KP250000000, T11, T8);
|
|
T1n = T1h - T1m;
|
|
T1y = T1s - T1x;
|
|
T1z = FMA(KP618033988, T1y, T1n);
|
|
T1B = FNMS(KP618033988, T1n, T1y);
|
|
ci[WS(rs, 4)] = T8 + T11;
|
|
T1A = FNMS(KP559016994, T1b, T1a);
|
|
ci[WS(rs, 2)] = FNMS(KP951056516, T1B, T1A);
|
|
cr[WS(rs, 3)] = FMA(KP951056516, T1B, T1A);
|
|
T1c = FMA(KP559016994, T1b, T1a);
|
|
ci[0] = FNMS(KP951056516, T1z, T1c);
|
|
cr[WS(rs, 1)] = FMA(KP951056516, T1z, T1c);
|
|
}
|
|
{
|
|
E T1D, T19, T1C, T1L, T1N, T1H, T1K, T1M, T1E;
|
|
T1D = T15 - T18;
|
|
T19 = T15 + T18;
|
|
T1C = FNMS(KP250000000, T19, T12);
|
|
T1H = T1F - T1G;
|
|
T1K = T1I - T1J;
|
|
T1L = FNMS(KP618033988, T1K, T1H);
|
|
T1N = FMA(KP618033988, T1H, T1K);
|
|
cr[0] = T12 + T19;
|
|
T1M = FMA(KP559016994, T1D, T1C);
|
|
cr[WS(rs, 4)] = FNMS(KP951056516, T1N, T1M);
|
|
ci[WS(rs, 3)] = FMA(KP951056516, T1N, T1M);
|
|
T1E = FNMS(KP559016994, T1D, T1C);
|
|
cr[WS(rs, 2)] = FNMS(KP951056516, T1L, T1E);
|
|
ci[WS(rs, 1)] = FMA(KP951056516, T1L, T1E);
|
|
}
|
|
{
|
|
E T1W, T1Q, T1V, T20, T22, T1Y, T1Z, T21, T1X;
|
|
T1W = T1P - T1O;
|
|
T1Q = T1O + T1P;
|
|
T1V = FMA(KP250000000, T1Q, T1U);
|
|
T1Y = TZ - TM;
|
|
T1Z = Ty - Tl;
|
|
T20 = FNMS(KP618033988, T1Z, T1Y);
|
|
T22 = FMA(KP618033988, T1Y, T1Z);
|
|
cr[WS(rs, 5)] = T1Q - T1U;
|
|
T21 = FMA(KP559016994, T1W, T1V);
|
|
cr[WS(rs, 9)] = FMS(KP951056516, T22, T21);
|
|
ci[WS(rs, 8)] = FMA(KP951056516, T22, T21);
|
|
T1X = FNMS(KP559016994, T1W, T1V);
|
|
cr[WS(rs, 7)] = FMS(KP951056516, T20, T1X);
|
|
ci[WS(rs, 6)] = FMA(KP951056516, T20, T1X);
|
|
}
|
|
{
|
|
E T28, T26, T27, T2c, T2e, T2a, T2b, T2d, T29;
|
|
T28 = T24 - T25;
|
|
T26 = T24 + T25;
|
|
T27 = FNMS(KP250000000, T26, T23);
|
|
T2a = T13 - T14;
|
|
T2b = T16 - T17;
|
|
T2c = FMA(KP618033988, T2b, T2a);
|
|
T2e = FNMS(KP618033988, T2a, T2b);
|
|
ci[WS(rs, 9)] = T26 + T23;
|
|
T2d = FNMS(KP559016994, T28, T27);
|
|
cr[WS(rs, 8)] = FMS(KP951056516, T2e, T2d);
|
|
ci[WS(rs, 7)] = FMA(KP951056516, T2e, T2d);
|
|
T29 = FMA(KP559016994, T28, T27);
|
|
cr[WS(rs, 6)] = FMS(KP951056516, T2c, T29);
|
|
ci[WS(rs, 5)] = FMA(KP951056516, T2c, T29);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
static const tw_instr twinstr[] = {
|
|
{ TW_FULL, 1, 10 },
|
|
{ TW_NEXT, 1, 0 }
|
|
};
|
|
|
|
static const hc2hc_desc desc = { 10, "hf_10", twinstr, &GENUS, { 48, 18, 54, 0 } };
|
|
|
|
void X(codelet_hf_10) (planner *p) {
|
|
X(khc2hc_register) (p, hf_10, &desc);
|
|
}
|
|
#else
|
|
|
|
/* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 10 -dit -name hf_10 -include rdft/scalar/hf.h */
|
|
|
|
/*
|
|
* This function contains 102 FP additions, 60 FP multiplications,
|
|
* (or, 72 additions, 30 multiplications, 30 fused multiply/add),
|
|
* 45 stack variables, 4 constants, and 40 memory accesses
|
|
*/
|
|
#include "rdft/scalar/hf.h"
|
|
|
|
static void hf_10(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms)
|
|
{
|
|
DK(KP587785252, +0.587785252292473129168705954639072768597652438);
|
|
DK(KP951056516, +0.951056516295153572116439333379382143405698634);
|
|
DK(KP250000000, +0.250000000000000000000000000000000000000000000);
|
|
DK(KP559016994, +0.559016994374947424102293417182819058860154590);
|
|
{
|
|
INT m;
|
|
for (m = mb, W = W + ((mb - 1) * 18); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 18, MAKE_VOLATILE_STRIDE(20, rs)) {
|
|
E T7, T1R, TT, T1C, TF, TQ, TR, T1o, T1p, T1P, TX, TY, TZ, T1d, T1g;
|
|
E T1x, Ti, Tt, Tu, T1r, T1s, T1O, TU, TV, TW, T16, T19, T1y;
|
|
{
|
|
E T1, T1A, T6, T1B;
|
|
T1 = cr[0];
|
|
T1A = ci[0];
|
|
{
|
|
E T3, T5, T2, T4;
|
|
T3 = cr[WS(rs, 5)];
|
|
T5 = ci[WS(rs, 5)];
|
|
T2 = W[8];
|
|
T4 = W[9];
|
|
T6 = FMA(T2, T3, T4 * T5);
|
|
T1B = FNMS(T4, T3, T2 * T5);
|
|
}
|
|
T7 = T1 - T6;
|
|
T1R = T1B + T1A;
|
|
TT = T1 + T6;
|
|
T1C = T1A - T1B;
|
|
}
|
|
{
|
|
E Tz, T1b, TP, T1e, TE, T1c, TK, T1f;
|
|
{
|
|
E Tw, Ty, Tv, Tx;
|
|
Tw = cr[WS(rs, 4)];
|
|
Ty = ci[WS(rs, 4)];
|
|
Tv = W[6];
|
|
Tx = W[7];
|
|
Tz = FMA(Tv, Tw, Tx * Ty);
|
|
T1b = FNMS(Tx, Tw, Tv * Ty);
|
|
}
|
|
{
|
|
E TM, TO, TL, TN;
|
|
TM = cr[WS(rs, 1)];
|
|
TO = ci[WS(rs, 1)];
|
|
TL = W[0];
|
|
TN = W[1];
|
|
TP = FMA(TL, TM, TN * TO);
|
|
T1e = FNMS(TN, TM, TL * TO);
|
|
}
|
|
{
|
|
E TB, TD, TA, TC;
|
|
TB = cr[WS(rs, 9)];
|
|
TD = ci[WS(rs, 9)];
|
|
TA = W[16];
|
|
TC = W[17];
|
|
TE = FMA(TA, TB, TC * TD);
|
|
T1c = FNMS(TC, TB, TA * TD);
|
|
}
|
|
{
|
|
E TH, TJ, TG, TI;
|
|
TH = cr[WS(rs, 6)];
|
|
TJ = ci[WS(rs, 6)];
|
|
TG = W[10];
|
|
TI = W[11];
|
|
TK = FMA(TG, TH, TI * TJ);
|
|
T1f = FNMS(TI, TH, TG * TJ);
|
|
}
|
|
TF = Tz - TE;
|
|
TQ = TK - TP;
|
|
TR = TF + TQ;
|
|
T1o = T1b + T1c;
|
|
T1p = T1f + T1e;
|
|
T1P = T1o + T1p;
|
|
TX = Tz + TE;
|
|
TY = TK + TP;
|
|
TZ = TX + TY;
|
|
T1d = T1b - T1c;
|
|
T1g = T1e - T1f;
|
|
T1x = T1g - T1d;
|
|
}
|
|
{
|
|
E Tc, T14, Ts, T18, Th, T15, Tn, T17;
|
|
{
|
|
E T9, Tb, T8, Ta;
|
|
T9 = cr[WS(rs, 2)];
|
|
Tb = ci[WS(rs, 2)];
|
|
T8 = W[2];
|
|
Ta = W[3];
|
|
Tc = FMA(T8, T9, Ta * Tb);
|
|
T14 = FNMS(Ta, T9, T8 * Tb);
|
|
}
|
|
{
|
|
E Tp, Tr, To, Tq;
|
|
Tp = cr[WS(rs, 3)];
|
|
Tr = ci[WS(rs, 3)];
|
|
To = W[4];
|
|
Tq = W[5];
|
|
Ts = FMA(To, Tp, Tq * Tr);
|
|
T18 = FNMS(Tq, Tp, To * Tr);
|
|
}
|
|
{
|
|
E Te, Tg, Td, Tf;
|
|
Te = cr[WS(rs, 7)];
|
|
Tg = ci[WS(rs, 7)];
|
|
Td = W[12];
|
|
Tf = W[13];
|
|
Th = FMA(Td, Te, Tf * Tg);
|
|
T15 = FNMS(Tf, Te, Td * Tg);
|
|
}
|
|
{
|
|
E Tk, Tm, Tj, Tl;
|
|
Tk = cr[WS(rs, 8)];
|
|
Tm = ci[WS(rs, 8)];
|
|
Tj = W[14];
|
|
Tl = W[15];
|
|
Tn = FMA(Tj, Tk, Tl * Tm);
|
|
T17 = FNMS(Tl, Tk, Tj * Tm);
|
|
}
|
|
Ti = Tc - Th;
|
|
Tt = Tn - Ts;
|
|
Tu = Ti + Tt;
|
|
T1r = T14 + T15;
|
|
T1s = T17 + T18;
|
|
T1O = T1r + T1s;
|
|
TU = Tc + Th;
|
|
TV = Tn + Ts;
|
|
TW = TU + TV;
|
|
T16 = T14 - T15;
|
|
T19 = T17 - T18;
|
|
T1y = T16 + T19;
|
|
}
|
|
{
|
|
E T11, TS, T12, T1i, T1k, T1a, T1h, T1j, T13;
|
|
T11 = KP559016994 * (Tu - TR);
|
|
TS = Tu + TR;
|
|
T12 = FNMS(KP250000000, TS, T7);
|
|
T1a = T16 - T19;
|
|
T1h = T1d + T1g;
|
|
T1i = FMA(KP951056516, T1a, KP587785252 * T1h);
|
|
T1k = FNMS(KP587785252, T1a, KP951056516 * T1h);
|
|
ci[WS(rs, 4)] = T7 + TS;
|
|
T1j = T12 - T11;
|
|
ci[WS(rs, 2)] = T1j - T1k;
|
|
cr[WS(rs, 3)] = T1j + T1k;
|
|
T13 = T11 + T12;
|
|
ci[0] = T13 - T1i;
|
|
cr[WS(rs, 1)] = T13 + T1i;
|
|
}
|
|
{
|
|
E T1m, T10, T1l, T1u, T1w, T1q, T1t, T1v, T1n;
|
|
T1m = KP559016994 * (TW - TZ);
|
|
T10 = TW + TZ;
|
|
T1l = FNMS(KP250000000, T10, TT);
|
|
T1q = T1o - T1p;
|
|
T1t = T1r - T1s;
|
|
T1u = FNMS(KP587785252, T1t, KP951056516 * T1q);
|
|
T1w = FMA(KP951056516, T1t, KP587785252 * T1q);
|
|
cr[0] = TT + T10;
|
|
T1v = T1m + T1l;
|
|
cr[WS(rs, 4)] = T1v - T1w;
|
|
ci[WS(rs, 3)] = T1v + T1w;
|
|
T1n = T1l - T1m;
|
|
cr[WS(rs, 2)] = T1n - T1u;
|
|
ci[WS(rs, 1)] = T1n + T1u;
|
|
}
|
|
{
|
|
E T1H, T1z, T1G, T1F, T1J, T1D, T1E, T1K, T1I;
|
|
T1H = KP559016994 * (T1y + T1x);
|
|
T1z = T1x - T1y;
|
|
T1G = FMA(KP250000000, T1z, T1C);
|
|
T1D = Ti - Tt;
|
|
T1E = TQ - TF;
|
|
T1F = FMA(KP587785252, T1D, KP951056516 * T1E);
|
|
T1J = FNMS(KP951056516, T1D, KP587785252 * T1E);
|
|
cr[WS(rs, 5)] = T1z - T1C;
|
|
T1K = T1H + T1G;
|
|
cr[WS(rs, 9)] = T1J - T1K;
|
|
ci[WS(rs, 8)] = T1J + T1K;
|
|
T1I = T1G - T1H;
|
|
cr[WS(rs, 7)] = T1F - T1I;
|
|
ci[WS(rs, 6)] = T1F + T1I;
|
|
}
|
|
{
|
|
E T1Q, T1S, T1T, T1N, T1V, T1L, T1M, T1W, T1U;
|
|
T1Q = KP559016994 * (T1O - T1P);
|
|
T1S = T1O + T1P;
|
|
T1T = FNMS(KP250000000, T1S, T1R);
|
|
T1L = TU - TV;
|
|
T1M = TX - TY;
|
|
T1N = FMA(KP951056516, T1L, KP587785252 * T1M);
|
|
T1V = FNMS(KP587785252, T1L, KP951056516 * T1M);
|
|
ci[WS(rs, 9)] = T1S + T1R;
|
|
T1W = T1T - T1Q;
|
|
cr[WS(rs, 8)] = T1V - T1W;
|
|
ci[WS(rs, 7)] = T1V + T1W;
|
|
T1U = T1Q + T1T;
|
|
cr[WS(rs, 6)] = T1N - T1U;
|
|
ci[WS(rs, 5)] = T1N + T1U;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
static const tw_instr twinstr[] = {
|
|
{ TW_FULL, 1, 10 },
|
|
{ TW_NEXT, 1, 0 }
|
|
};
|
|
|
|
static const hc2hc_desc desc = { 10, "hf_10", twinstr, &GENUS, { 72, 30, 30, 0 } };
|
|
|
|
void X(codelet_hf_10) (planner *p) {
|
|
X(khc2hc_register) (p, hf_10, &desc);
|
|
}
|
|
#endif
|