WebSVN - shark - Blame - Rev 2 - /shark/branches/pj/ports/fftw/rfftw/fhb

Rev	Author	Line No.	Line
2	pj	1	/*
		2	* Copyright (c) 1997-1999 Massachusetts Institute of Technology
		3	*
		4	* This program is free software; you can redistribute it and/or modify
		5	* it under the terms of the GNU General Public License as published by
		6	* the Free Software Foundation; either version 2 of the License, or
		7	* (at your option) any later version.
		8	*
		9	* This program is distributed in the hope that it will be useful,
		10	* but WITHOUT ANY WARRANTY; without even the implied warranty of
		11	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
		12	* GNU General Public License for more details.
		13	*
		14	* You should have received a copy of the GNU General Public License
		15	* along with this program; if not, write to the Free Software
		16	* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
		17	*
		18	*/
		19
		20	/* This file was automatically generated --- DO NOT EDIT */
		21	/* Generated on Tue May 18 13:56:02 EDT 1999 */
		22
		23	#include <ports/fftw-int.h>
		24	#include <ports/fftw.h>
		25
		26	/* Generated by: ./genfft -magic-alignment-check -magic-twiddle-load-all -magic-variables 4 -magic-loopi -hc2hc-backward 7 */
		27
		28	/*
		29	* This function contains 120 FP additions, 98 FP multiplications,
		30	* (or, 106 additions, 84 multiplications, 14 fused multiply/add),
		31	* 32 stack variables, and 56 memory accesses
		32	*/
		33	static const fftw_real K222520933 = FFTW_KONST(+0.222520933956314404288902564496794759466355569);
		34	static const fftw_real K900968867 = FFTW_KONST(+0.900968867902419126236102319507445051165919162);
		35	static const fftw_real K623489801 = FFTW_KONST(+0.623489801858733530525004884004239810632274731);
		36	static const fftw_real K781831482 = FFTW_KONST(+0.781831482468029808708444526674057750232334519);
		37	static const fftw_real K974927912 = FFTW_KONST(+0.974927912181823607018131682993931217232785801);
		38	static const fftw_real K433883739 = FFTW_KONST(+0.433883739117558120475768332848358754609990728);
		39	static const fftw_real K2_000000000 = FFTW_KONST(+2.000000000000000000000000000000000000000000000);
		40	static const fftw_real K1_801937735 = FFTW_KONST(+1.801937735804838252472204639014890102331838324);
		41	static const fftw_real K445041867 = FFTW_KONST(+0.445041867912628808577805128993589518932711138);
		42	static const fftw_real K1_246979603 = FFTW_KONST(+1.246979603717467061050009768008479621264549462);
		43	static const fftw_real K867767478 = FFTW_KONST(+0.867767478235116240951536665696717509219981456);
		44	static const fftw_real K1_949855824 = FFTW_KONST(+1.949855824363647214036263365987862434465571601);
		45	static const fftw_real K1_563662964 = FFTW_KONST(+1.563662964936059617416889053348115500464669037);
		46
		47	/*
		48	* Generator Id's :
		49	* $Id: fhb_7.c,v 1.1.1.1 2002-03-29 14:12:58 pj Exp $
		50	* $Id: fhb_7.c,v 1.1.1.1 2002-03-29 14:12:58 pj Exp $
		51	* $Id: fhb_7.c,v 1.1.1.1 2002-03-29 14:12:58 pj Exp $
		52	*/
		53
		54	void fftw_hc2hc_backward_7(fftw_real A, const fftw_complex W, int iostride, int m, int dist)
		55	{
		56	int i;
		57	fftw_real *X;
		58	fftw_real *Y;
		59	X = A;
		60	Y = A + (7 * iostride);
		61	{
		62	fftw_real tmp84;
		63	fftw_real tmp88;
		64	fftw_real tmp86;
		65	fftw_real tmp76;
		66	fftw_real tmp79;
		67	fftw_real tmp77;
		68	fftw_real tmp78;
		69	fftw_real tmp80;
		70	fftw_real tmp87;
		71	fftw_real tmp85;
		72	fftw_real tmp81;
		73	fftw_real tmp83;
		74	fftw_real tmp82;
		75	ASSERT_ALIGNED_DOUBLE();
		76	tmp81 = Y[-2 * iostride];
		77	tmp83 = Y[-iostride];
		78	tmp82 = Y[-3 * iostride];
		79	tmp84 = (K1_563662964 * tmp81) - (K1_949855824 * tmp82) - (K867767478 * tmp83);
		80	tmp88 = (K867767478 * tmp81) + (K1_563662964 * tmp82) - (K1_949855824 * tmp83);
		81	tmp86 = (K1_563662964 * tmp83) + (K1_949855824 * tmp81) + (K867767478 * tmp82);
		82	tmp76 = X[0];
		83	tmp79 = X[3 * iostride];
		84	tmp77 = X[iostride];
		85	tmp78 = X[2 * iostride];
		86	tmp80 = tmp76 + (K1_246979603 * tmp78) - (K445041867 * tmp79) - (K1_801937735 * tmp77);
		87	tmp87 = tmp76 + (K1_246979603 * tmp79) - (K1_801937735 * tmp78) - (K445041867 * tmp77);
		88	tmp85 = tmp76 + (K1_246979603 * tmp77) - (K1_801937735 * tmp79) - (K445041867 * tmp78);
		89	X[4 * iostride] = tmp80 - tmp84;
		90	X[3 * iostride] = tmp80 + tmp84;
		91	X[0] = tmp76 + (K2_000000000 * (tmp77 + tmp78 + tmp79));
		92	X[2 * iostride] = tmp87 + tmp88;
		93	X[5 * iostride] = tmp87 - tmp88;
		94	X[iostride] = tmp85 - tmp86;
		95	X[6 * iostride] = tmp85 + tmp86;
		96	}
		97	X = X + dist;
		98	Y = Y - dist;
		99	for (i = 2; i < m; i = i + 2, X = X + dist, Y = Y - dist, W = W + 6) {
		100	fftw_real tmp14;
		101	fftw_real tmp23;
		102	fftw_real tmp17;
		103	fftw_real tmp20;
		104	fftw_real tmp39;
		105	fftw_real tmp53;
		106	fftw_real tmp66;
		107	fftw_real tmp69;
		108	fftw_real tmp57;
		109	fftw_real tmp42;
		110	fftw_real tmp24;
		111	fftw_real tmp33;
		112	fftw_real tmp27;
		113	fftw_real tmp30;
		114	fftw_real tmp46;
		115	fftw_real tmp58;
		116	fftw_real tmp70;
		117	fftw_real tmp65;
		118	fftw_real tmp54;
		119	fftw_real tmp35;
		120	ASSERT_ALIGNED_DOUBLE();
		121	{
		122	fftw_real tmp37;
		123	fftw_real tmp36;
		124	fftw_real tmp38;
		125	fftw_real tmp21;
		126	fftw_real tmp22;
		127	ASSERT_ALIGNED_DOUBLE();
		128	tmp14 = X[0];
		129	tmp21 = X[3 * iostride];
		130	tmp22 = Y[-4 * iostride];
		131	tmp23 = tmp21 + tmp22;
		132	tmp37 = tmp21 - tmp22;
		133	{
		134	fftw_real tmp15;
		135	fftw_real tmp16;
		136	fftw_real tmp18;
		137	fftw_real tmp19;
		138	ASSERT_ALIGNED_DOUBLE();
		139	tmp15 = X[iostride];
		140	tmp16 = Y[-6 * iostride];
		141	tmp17 = tmp15 + tmp16;
		142	tmp36 = tmp15 - tmp16;
		143	tmp18 = X[2 * iostride];
		144	tmp19 = Y[-5 * iostride];
		145	tmp20 = tmp18 + tmp19;
		146	tmp38 = tmp18 - tmp19;
		147	}
		148	tmp39 = (K433883739 * tmp36) + (K974927912 * tmp37) - (K781831482 * tmp38);
		149	tmp53 = (K781831482 * tmp36) + (K974927912 * tmp38) + (K433883739 * tmp37);
		150	tmp66 = (K974927912 * tmp36) - (K781831482 * tmp37) - (K433883739 * tmp38);
		151	tmp69 = tmp14 + (K623489801 * tmp23) - (K900968867 * tmp20) - (K222520933 * tmp17);
		152	tmp57 = tmp14 + (K623489801 * tmp17) - (K900968867 * tmp23) - (K222520933 * tmp20);
		153	tmp42 = tmp14 + (K623489801 * tmp20) - (K222520933 * tmp23) - (K900968867 * tmp17);
		154	}
		155	{
		156	fftw_real tmp44;
		157	fftw_real tmp45;
		158	fftw_real tmp43;
		159	fftw_real tmp31;
		160	fftw_real tmp32;
		161	ASSERT_ALIGNED_DOUBLE();
		162	tmp24 = Y[0];
		163	tmp31 = Y[-3 * iostride];
		164	tmp32 = X[4 * iostride];
		165	tmp33 = tmp31 - tmp32;
		166	tmp44 = tmp31 + tmp32;
		167	{
		168	fftw_real tmp25;
		169	fftw_real tmp26;
		170	fftw_real tmp28;
		171	fftw_real tmp29;
		172	ASSERT_ALIGNED_DOUBLE();
		173	tmp25 = Y[-iostride];
		174	tmp26 = X[6 * iostride];
		175	tmp27 = tmp25 - tmp26;
		176	tmp45 = tmp25 + tmp26;
		177	tmp28 = Y[-2 * iostride];
		178	tmp29 = X[5 * iostride];
		179	tmp30 = tmp28 - tmp29;
		180	tmp43 = tmp28 + tmp29;
		181	}
		182	tmp46 = (K781831482 * tmp43) - (K974927912 * tmp44) - (K433883739 * tmp45);
		183	tmp58 = (K781831482 * tmp45) + (K974927912 * tmp43) + (K433883739 * tmp44);
		184	tmp70 = (K433883739 * tmp43) + (K781831482 * tmp44) - (K974927912 * tmp45);
		185	tmp65 = tmp24 + (K623489801 * tmp33) - (K900968867 * tmp30) - (K222520933 * tmp27);
		186	tmp54 = tmp24 + (K623489801 * tmp27) - (K900968867 * tmp33) - (K222520933 * tmp30);
		187	tmp35 = tmp24 + (K623489801 * tmp30) - (K222520933 * tmp33) - (K900968867 * tmp27);
		188	}
		189	X[0] = tmp14 + tmp17 + tmp20 + tmp23;
		190	{
		191	fftw_real tmp61;
		192	fftw_real tmp63;
		193	fftw_real tmp60;
		194	fftw_real tmp62;
		195	ASSERT_ALIGNED_DOUBLE();
		196	tmp61 = tmp54 - tmp53;
		197	tmp63 = tmp57 + tmp58;
		198	tmp60 = c_re(W[5]);
		199	tmp62 = c_im(W[5]);
		200	Y[0] = (tmp60 * tmp61) - (tmp62 * tmp63);
		201	X[6 * iostride] = (tmp62 * tmp61) + (tmp60 * tmp63);
		202	}
		203	{
		204	fftw_real tmp73;
		205	fftw_real tmp75;
		206	fftw_real tmp72;
		207	fftw_real tmp74;
		208	ASSERT_ALIGNED_DOUBLE();
		209	tmp73 = tmp66 + tmp65;
		210	tmp75 = tmp69 + tmp70;
		211	tmp72 = c_re(W[1]);
		212	tmp74 = c_im(W[1]);
		213	Y[-4 * iostride] = (tmp72 * tmp73) - (tmp74 * tmp75);
		214	X[2 * iostride] = (tmp74 * tmp73) + (tmp72 * tmp75);
		215	}
		216	{
		217	fftw_real tmp67;
		218	fftw_real tmp71;
		219	fftw_real tmp64;
		220	fftw_real tmp68;
		221	ASSERT_ALIGNED_DOUBLE();
		222	tmp67 = tmp65 - tmp66;
		223	tmp71 = tmp69 - tmp70;
		224	tmp64 = c_re(W[4]);
		225	tmp68 = c_im(W[4]);
		226	Y[-iostride] = (tmp64 * tmp67) - (tmp68 * tmp71);
		227	X[5 * iostride] = (tmp68 * tmp67) + (tmp64 * tmp71);
		228	}
		229	Y[-6 * iostride] = tmp24 + tmp27 + tmp30 + tmp33;
		230	{
		231	fftw_real tmp40;
		232	fftw_real tmp47;
		233	fftw_real tmp34;
		234	fftw_real tmp41;
		235	ASSERT_ALIGNED_DOUBLE();
		236	tmp40 = tmp35 - tmp39;
		237	tmp47 = tmp42 - tmp46;
		238	tmp34 = c_re(W[3]);
		239	tmp41 = c_im(W[3]);
		240	Y[-2 * iostride] = (tmp34 * tmp40) - (tmp41 * tmp47);
		241	X[4 * iostride] = (tmp41 * tmp40) + (tmp34 * tmp47);
		242	}
		243	{
		244	fftw_real tmp49;
		245	fftw_real tmp51;
		246	fftw_real tmp48;
		247	fftw_real tmp50;
		248	ASSERT_ALIGNED_DOUBLE();
		249	tmp49 = tmp39 + tmp35;
		250	tmp51 = tmp42 + tmp46;
		251	tmp48 = c_re(W[2]);
		252	tmp50 = c_im(W[2]);
		253	Y[-3 * iostride] = (tmp48 * tmp49) - (tmp50 * tmp51);
		254	X[3 * iostride] = (tmp50 * tmp49) + (tmp48 * tmp51);
		255	}
		256	{
		257	fftw_real tmp55;
		258	fftw_real tmp59;
		259	fftw_real tmp52;
		260	fftw_real tmp56;
		261	ASSERT_ALIGNED_DOUBLE();
		262	tmp55 = tmp53 + tmp54;
		263	tmp59 = tmp57 - tmp58;
		264	tmp52 = c_re(W[0]);
		265	tmp56 = c_im(W[0]);
		266	Y[-5 * iostride] = (tmp52 * tmp55) - (tmp56 * tmp59);
		267	X[iostride] = (tmp56 * tmp55) + (tmp52 * tmp59);
		268	}
		269	}
		270	if (i == m) {
		271	fftw_real tmp9;
		272	fftw_real tmp13;
		273	fftw_real tmp11;
		274	fftw_real tmp1;
		275	fftw_real tmp4;
		276	fftw_real tmp2;
		277	fftw_real tmp3;
		278	fftw_real tmp5;
		279	fftw_real tmp12;
		280	fftw_real tmp10;
		281	fftw_real tmp6;
		282	fftw_real tmp8;
		283	fftw_real tmp7;
		284	ASSERT_ALIGNED_DOUBLE();
		285	tmp6 = Y[-2 * iostride];
		286	tmp8 = Y[0];
		287	tmp7 = Y[-iostride];
		288	tmp9 = (K1_563662964 * tmp6) + (K1_949855824 * tmp7) + (K867767478 * tmp8);
		289	tmp13 = (K1_563662964 * tmp7) - (K1_949855824 * tmp8) - (K867767478 * tmp6);
		290	tmp11 = (K1_949855824 * tmp6) - (K1_563662964 * tmp8) - (K867767478 * tmp7);
		291	tmp1 = X[3 * iostride];
		292	tmp4 = X[0];
		293	tmp2 = X[2 * iostride];
		294	tmp3 = X[iostride];
		295	tmp5 = (K445041867 * tmp3) + (K1_801937735 * tmp4) - (K1_246979603 * tmp2) - tmp1;
		296	tmp12 = (K1_801937735 * tmp2) + (K445041867 * tmp4) - (K1_246979603 * tmp3) - tmp1;
		297	tmp10 = tmp1 + (K1_246979603 * tmp4) - (K1_801937735 * tmp3) - (K445041867 * tmp2);
		298	X[iostride] = tmp5 - tmp9;
		299	X[6 * iostride] = -(tmp5 + tmp9);
		300	X[0] = tmp1 + (K2_000000000 * (tmp2 + tmp3 + tmp4));
		301	X[4 * iostride] = tmp13 - tmp12;
		302	X[3 * iostride] = tmp12 + tmp13;
		303	X[5 * iostride] = tmp11 - tmp10;
		304	X[2 * iostride] = tmp10 + tmp11;
		305	}
		306	}
		307
		308	static const int twiddle_order[] =
		309	{1, 2, 3, 4, 5, 6};
		310	fftw_codelet_desc fftw_hc2hc_backward_7_desc =
		311	{
		312	"fftw_hc2hc_backward_7",
		313	(void (*)()) fftw_hc2hc_backward_7,
		314	7,
		315	FFTW_BACKWARD,
		316	FFTW_HC2HC,
		317	168,
		318	6,
		319	twiddle_order,
		320	};

Subversion Repositories shark

(root)/shark/branches/pj/ports/fftw/rfftw/fhb_7.c - Rev 2