1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
|
/*
* Copyright (c) 2003, 2007-14 Matteo Frigo
* Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
*
* Generic128d added by Romain Dolbeau, and turned into simd-generic128.h
* with single & double precision by Erik Lindahl.
* Romain Dolbeau hereby places his modifications in the public domain.
* Erik Lindahl hereby places his modifications in the public domain.
*
* 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
*
*/
#if defined(FFTW_LDOUBLE) || defined(FFTW_QUAD)
# error "Generic simd128 only works in single or double precision"
#endif
#define SIMD_SUFFIX _generic_simd128 /* for renaming */
#ifdef FFTW_SINGLE
# define DS(d,s) s /* single-precision option */
# define VDUPL(x) (V){x[0],x[0],x[2],x[2]}
# define VDUPH(x) (V){x[1],x[1],x[3],x[3]}
# define DVK(var, val) V var = {val,val,val,val}
#else
# define DS(d,s) d /* double-precision option */
# define VDUPL(x) (V){x[0],x[0]}
# define VDUPH(x) (V){x[1],x[1]}
# define DVK(var, val) V var = {val, val}
#endif
#define VL DS(1,2) /* SIMD vector length, in term of complex numbers */
#define SIMD_VSTRIDE_OKA(x) DS(1,((x) == 2))
#define SIMD_STRIDE_OKPAIR SIMD_STRIDE_OK
typedef DS(double,float) V __attribute__ ((vector_size(16)));
#define VADD(a,b) ((a)+(b))
#define VSUB(a,b) ((a)-(b))
#define VMUL(a,b) ((a)*(b))
#define LDK(x) x
static inline V LDA(const R *x, INT ivs, const R *aligned_like)
{
(void)aligned_like; /* UNUSED */
(void)ivs; /* UNUSED */
return *(const V *)x;
}
static inline void STA(R *x, V v, INT ovs, const R *aligned_like)
{
(void)aligned_like; /* UNUSED */
(void)ovs; /* UNUSED */
*(V *)x = v;
}
static inline V LD(const R *x, INT ivs, const R *aligned_like)
{
(void)aligned_like; /* UNUSED */
V res;
res[0] = x[0];
res[1] = x[1];
#ifdef FFTW_SINGLE
res[2] = x[ivs];
res[3] = x[ivs+1];
#endif
return res;
}
#ifdef FFTW_SINGLE
/* ST has to be separate due to the storage hack requiring reverse order */
static inline void ST(R *x, V v, INT ovs, const R *aligned_like)
{
(void)aligned_like; /* UNUSED */
(void)ovs; /* UNUSED */
*(x + ovs ) = v[2];
*(x + ovs + 1) = v[3];
*(x ) = v[0];
*(x + 1) = v[1];
}
#else
/* FFTW_DOUBLE */
# define ST STA
#endif
#ifdef FFTW_SINGLE
#define STM2 ST
#define STN2(x, v0, v1, ovs) /* nop */
static inline void STN4(R *x, V v0, V v1, V v2, V v3, INT ovs)
{
*(x ) = v0[0];
*(x + 1) = v1[0];
*(x + 2) = v2[0];
*(x + 3) = v3[0];
*(x + ovs ) = v0[1];
*(x + ovs + 1) = v1[1];
*(x + ovs + 2) = v2[1];
*(x + ovs + 3) = v3[1];
*(x + 2 * ovs ) = v0[2];
*(x + 2 * ovs + 1) = v1[2];
*(x + 2 * ovs + 2) = v2[2];
*(x + 2 * ovs + 3) = v3[2];
*(x + 3 * ovs ) = v0[3];
*(x + 3 * ovs + 1) = v1[3];
*(x + 3 * ovs + 2) = v2[3];
*(x + 3 * ovs + 3) = v3[3];
}
#define STM4(x, v, ovs, aligned_like) /* no-op */
#else
/* FFTW_DOUBLE */
#define STM2 STA
#define STN2(x, v0, v1, ovs) /* nop */
static inline void STM4(R *x, V v, INT ovs, const R *aligned_like)
{
(void)aligned_like; /* UNUSED */
*(x) = v[0];
*(x+ovs) = v[1];
}
# define STN4(x, v0, v1, v2, v3, ovs) /* nothing */
#endif
static inline V FLIP_RI(V x)
{
#ifdef FFTW_SINGLE
return (V){x[1],x[0],x[3],x[2]};
#else
return (V){x[1],x[0]};
#endif
}
static inline V VCONJ(V x)
{
#ifdef FFTW_SINGLE
return (V){x[0],-x[1],x[2],-x[3]};
#else
return (V){x[0],-x[1]};
#endif
}
static inline V VBYI(V x)
{
x = VCONJ(x);
x = FLIP_RI(x);
return x;
}
/* FMA support */
#define VFMA(a, b, c) VADD(c, VMUL(a, b))
#define VFNMS(a, b, c) VSUB(c, VMUL(a, b))
#define VFMS(a, b, c) VSUB(VMUL(a, b), c)
#define VFMAI(b, c) VADD(c, VBYI(b))
#define VFNMSI(b, c) VSUB(c, VBYI(b))
#define VFMACONJ(b,c) VADD(VCONJ(b),c)
#define VFMSCONJ(b,c) VSUB(VCONJ(b),c)
#define VFNMSCONJ(b,c) VSUB(c, VCONJ(b))
static inline V VZMUL(V tx, V sr)
{
V tr = VDUPL(tx);
V ti = VDUPH(tx);
tr = VMUL(sr, tr);
sr = VBYI(sr);
return VFMA(ti, sr, tr);
}
static inline V VZMULJ(V tx, V sr)
{
V tr = VDUPL(tx);
V ti = VDUPH(tx);
tr = VMUL(sr, tr);
sr = VBYI(sr);
return VFNMS(ti, sr, tr);
}
static inline V VZMULI(V tx, V sr)
{
V tr = VDUPL(tx);
V ti = VDUPH(tx);
ti = VMUL(ti, sr);
sr = VBYI(sr);
return VFMS(tr, sr, ti);
}
static inline V VZMULIJ(V tx, V sr)
{
V tr = VDUPL(tx);
V ti = VDUPH(tx);
ti = VMUL(ti, sr);
sr = VBYI(sr);
return VFMA(tr, sr, ti);
}
/* twiddle storage #1: compact, slower */
#ifdef FFTW_SINGLE
# define VTW1(v,x) \
{TW_COS, v, x}, {TW_COS, v+1, x}, {TW_SIN, v, x}, {TW_SIN, v+1, x}
static inline V BYTW1(const R *t, V sr)
{
return VZMUL(LDA(t, 2, t), sr);
}
static inline V BYTWJ1(const R *t, V sr)
{
return VZMULJ(LDA(t, 2, t), sr);
}
#else /* !FFTW_SINGLE */
# define VTW1(v,x) {TW_CEXP, v, x}
static inline V BYTW1(const R *t, V sr)
{
V tx = LD(t, 1, t);
return VZMUL(tx, sr);
}
static inline V BYTWJ1(const R *t, V sr)
{
V tx = LD(t, 1, t);
return VZMULJ(tx, sr);
}
#endif
#define TWVL1 (VL)
/* twiddle storage #2: twice the space, faster (when in cache) */
#ifdef FFTW_SINGLE
# define VTW2(v,x) \
{TW_COS, v, x}, {TW_COS, v, x}, {TW_COS, v+1, x}, {TW_COS, v+1, x}, \
{TW_SIN, v, -x}, {TW_SIN, v, x}, {TW_SIN, v+1, -x}, {TW_SIN, v+1, x}
#else /* !FFTW_SINGLE */
# define VTW2(v,x) \
{TW_COS, v, x}, {TW_COS, v, x}, {TW_SIN, v, -x}, {TW_SIN, v, x}
#endif
#define TWVL2 (2 * VL)
static inline V BYTW2(const R *t, V sr)
{
const V *twp = (const V *)t;
V si = FLIP_RI(sr);
V tr = twp[0], ti = twp[1];
return VFMA(tr, sr, VMUL(ti, si));
}
static inline V BYTWJ2(const R *t, V sr)
{
const V *twp = (const V *)t;
V si = FLIP_RI(sr);
V tr = twp[0], ti = twp[1];
return VFNMS(ti, si, VMUL(tr, sr));
}
/* twiddle storage #3 */
#ifdef FFTW_SINGLE
# define VTW3(v,x) {TW_CEXP, v, x}, {TW_CEXP, v+1, x}
# define TWVL3 (VL)
#else
# define VTW3(v,x) VTW1(v,x)
# define TWVL3 TWVL1
#endif
/* twiddle storage for split arrays */
#ifdef FFTW_SINGLE
# define VTWS(v,x) \
{TW_COS, v, x}, {TW_COS, v+1, x}, {TW_COS, v+2, x}, {TW_COS, v+3, x}, \
{TW_SIN, v, x}, {TW_SIN, v+1, x}, {TW_SIN, v+2, x}, {TW_SIN, v+3, x}
#else
# define VTWS(v,x) \
{TW_COS, v, x}, {TW_COS, v+1, x}, {TW_SIN, v, x}, {TW_SIN, v+1, x}
#endif
#define TWVLS (2 * VL)
#define VLEAVE() /* nothing */
#include "simd-common.h"
|