dsputil_mmx.c 90.5 KB
Newer Older
Fabrice Bellard's avatar
Fabrice Bellard committed
1 2
/*
 * MMX optimized DSP utils
3
 * Copyright (c) 2000, 2001 Fabrice Bellard
4
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
Fabrice Bellard's avatar
Fabrice Bellard committed
5
 *
6 7
 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
 *
8
 * This file is part of Libav.
9
 *
10
 * Libav is free software; you can redistribute it and/or
Fabrice Bellard's avatar
Fabrice Bellard committed
11 12
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
13
 * version 2.1 of the License, or (at your option) any later version.
Fabrice Bellard's avatar
Fabrice Bellard committed
14
 *
15
 * Libav is distributed in the hope that it will be useful,
Fabrice Bellard's avatar
Fabrice Bellard committed
16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
Fabrice Bellard's avatar
Fabrice Bellard committed
17 18
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
Fabrice Bellard's avatar
Fabrice Bellard committed
19
 *
Fabrice Bellard's avatar
Fabrice Bellard committed
20
 * You should have received a copy of the GNU Lesser General Public
21
 * License along with Libav; if not, write to the Free Software
22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Fabrice Bellard's avatar
Fabrice Bellard committed
23 24
 */

25
#include "libavutil/attributes.h"
26
#include "libavutil/cpu.h"
27
#include "libavutil/x86/asm.h"
28
#include "libavcodec/dsputil.h"
29
#include "libavcodec/h264dsp.h"
30 31
#include "libavcodec/mpegvideo.h"
#include "libavcodec/simple_idct.h"
32
#include "dsputil_mmx.h"
33
#include "idct_xvid.h"
Fabrice Bellard's avatar
Fabrice Bellard committed
34

35 36 37
//#undef NDEBUG
//#include <assert.h>

Fabrice Bellard's avatar
Fabrice Bellard committed
38
/* pixel operations */
39 40
DECLARE_ALIGNED(8,  const uint64_t, ff_bone) = 0x0101010101010101ULL;
DECLARE_ALIGNED(8,  const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
Fabrice Bellard's avatar
Fabrice Bellard committed
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
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_1)    = { 0x0001000100010001ULL, 0x0001000100010001ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_2)    = { 0x0002000200020002ULL, 0x0002000200020002ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_3)    = { 0x0003000300030003ULL, 0x0003000300030003ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_4)    = { 0x0004000400040004ULL, 0x0004000400040004ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_5)    = { 0x0005000500050005ULL, 0x0005000500050005ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_8)    = { 0x0008000800080008ULL, 0x0008000800080008ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_9)    = { 0x0009000900090009ULL, 0x0009000900090009ULL };
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_15)   =   0x000F000F000F000FULL;
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_16)   = { 0x0010001000100010ULL, 0x0010001000100010ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_17)   = { 0x0011001100110011ULL, 0x0011001100110011ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_18)   = { 0x0012001200120012ULL, 0x0012001200120012ULL };
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_20)   =   0x0014001400140014ULL;
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_27)   = { 0x001B001B001B001BULL, 0x001B001B001B001BULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_28)   = { 0x001C001C001C001CULL, 0x001C001C001C001CULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_32)   = { 0x0020002000200020ULL, 0x0020002000200020ULL };
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_42)   =   0x002A002A002A002AULL;
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_53)   =   0x0035003500350035ULL;
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_63)   = { 0x003F003F003F003FULL, 0x003F003F003F003FULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_64)   = { 0x0040004000400040ULL, 0x0040004000400040ULL };
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_96)   =   0x0060006000600060ULL;
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_128)  =   0x0080008000800080ULL;
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_255)  =   0x00ff00ff00ff00ffULL;
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_512)  = { 0x0200020002000200ULL, 0x0200020002000200ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_1019) = { 0x03FB03FB03FB03FBULL, 0x03FB03FB03FB03FBULL };

DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_0)    = { 0x0000000000000000ULL, 0x0000000000000000ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_1)    = { 0x0101010101010101ULL, 0x0101010101010101ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_3)    = { 0x0303030303030303ULL, 0x0303030303030303ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_4)    = { 0x0404040404040404ULL, 0x0404040404040404ULL };
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_7)    =   0x0707070707070707ULL;
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1F)   =   0x1F1F1F1F1F1F1F1FULL;
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3F)   =   0x3F3F3F3F3F3F3F3FULL;
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_80)   = { 0x8080808080808080ULL, 0x8080808080808080ULL };
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_81)   =   0x8181818181818181ULL;
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_A1)   = { 0xA1A1A1A1A1A1A1A1ULL, 0xA1A1A1A1A1A1A1A1ULL };
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_F8)   = { 0xF8F8F8F8F8F8F8F8ULL, 0xF8F8F8F8F8F8F8F8ULL };
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_FC)   =   0xFCFCFCFCFCFCFCFCULL;
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_FE)   = { 0xFEFEFEFEFEFEFEFEULL, 0xFEFEFEFEFEFEFEFEULL };
80 81 82

DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };
DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };
83

84

85
#if HAVE_YASM
86
void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
87
                              ptrdiff_t line_size, int h);
88
void ff_put_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
89
                             ptrdiff_t line_size, int h);
90 91 92 93 94 95 96 97
void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
                              int dstStride, int src1Stride, int h);
void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1,
                                     uint8_t *src2, int dstStride,
                                     int src1Stride, int h);
void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
                              int dstStride, int src1Stride, int h);
void ff_put_pixels16_x2_mmxext(uint8_t *block, const uint8_t *pixels,
98
                               ptrdiff_t line_size, int h);
99
void ff_put_pixels16_x2_3dnow(uint8_t *block, const uint8_t *pixels,
100
                              ptrdiff_t line_size, int h);
101 102 103 104 105 106 107
void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
                               int dstStride, int src1Stride, int h);
void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
                               int dstStride, int src1Stride, int h);
void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
                                      int dstStride, int src1Stride, int h);
void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
108
                                     ptrdiff_t line_size, int h);
109
void ff_put_no_rnd_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
110
                                    ptrdiff_t line_size, int h);
111 112
void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block,
                                           const uint8_t *pixels,
113
                                           ptrdiff_t line_size, int h);
114 115
void ff_put_no_rnd_pixels8_x2_exact_3dnow(uint8_t *block,
                                          const uint8_t *pixels,
116
                                          ptrdiff_t line_size, int h);
117
void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
118
                              ptrdiff_t line_size, int h);
119
void ff_put_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
120
                             ptrdiff_t line_size, int h);
121
void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
122
                                     ptrdiff_t line_size, int h);
123
void ff_put_no_rnd_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
124
                                    ptrdiff_t line_size, int h);
125 126
void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block,
                                           const uint8_t *pixels,
127
                                           ptrdiff_t line_size, int h);
128 129
void ff_put_no_rnd_pixels8_y2_exact_3dnow(uint8_t *block,
                                          const uint8_t *pixels,
130
                                          ptrdiff_t line_size, int h);
131
void ff_avg_pixels8_mmxext(uint8_t *block, const uint8_t *pixels,
132
                           ptrdiff_t line_size, int h);
133
void ff_avg_pixels8_3dnow(uint8_t *block, const uint8_t *pixels,
134
                          ptrdiff_t line_size, int h);
135
void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
136
                              ptrdiff_t line_size, int h);
137
void ff_avg_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
138
                             ptrdiff_t line_size, int h);
139
void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
140
                              ptrdiff_t line_size, int h);
141
void ff_avg_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
142
                             ptrdiff_t line_size, int h);
143
void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
144
                               ptrdiff_t line_size, int h);
145
void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
146
                              ptrdiff_t line_size, int h);
147

148
void ff_put_pixels8_mmxext(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h);
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
static void ff_put_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
                                   int line_size, int h)
{
    ff_put_pixels8_mmxext(block,     pixels,     line_size, h);
    ff_put_pixels8_mmxext(block + 8, pixels + 8, line_size, h);
}

void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                         int dstStride, int srcStride, int h);
void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                         int dstStride, int srcStride, int h);
void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                                 int dstStride, int srcStride,
                                                 int h);
void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                        int dstStride, int srcStride, int h);
void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                        int dstStride, int srcStride, int h);
void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                                int dstStride, int srcStride,
                                                int h);
void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                         int dstStride, int srcStride);
void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                         int dstStride, int srcStride);
void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                                 int dstStride, int srcStride);
void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                        int dstStride, int srcStride);
void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                        int dstStride, int srcStride);
void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
                                                int dstStride, int srcStride);
#define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmxext
#define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmxext
184
#endif /* HAVE_YASM */
185 186


187 188
#if HAVE_INLINE_ASM

189 190
#define JUMPALIGN()     __asm__ volatile (".p2align 3"::)
#define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%"#regd", %%"#regd ::)
191

192 193 194 195
#define MOVQ_BFE(regd)                                  \
    __asm__ volatile (                                  \
        "pcmpeqd %%"#regd", %%"#regd"   \n\t"           \
        "paddb   %%"#regd", %%"#regd"   \n\t" ::)
196

197
#ifndef PIC
198 199
#define MOVQ_BONE(regd) __asm__ volatile ("movq %0, %%"#regd" \n\t" :: "m"(ff_bone))
#define MOVQ_WTWO(regd) __asm__ volatile ("movq %0, %%"#regd" \n\t" :: "m"(ff_wtwo))
200 201 202
#else
// for shared library it's better to use this way for accessing constants
// pcmpeqd -> -1
203 204 205 206 207 208 209 210 211 212 213
#define MOVQ_BONE(regd)                                 \
    __asm__ volatile (                                  \
        "pcmpeqd  %%"#regd", %%"#regd"  \n\t"           \
        "psrlw          $15, %%"#regd"  \n\t"           \
        "packuswb %%"#regd", %%"#regd"  \n\t" ::)

#define MOVQ_WTWO(regd)                                 \
    __asm__ volatile (                                  \
        "pcmpeqd %%"#regd", %%"#regd"   \n\t"           \
        "psrlw         $15, %%"#regd"   \n\t"           \
        "psllw          $1, %%"#regd"   \n\t"::)
214

215 216
#endif

217
// using regr as temporary and for the output result
218
// first argument is unmodifed and second is trashed
219
// regfe is supposed to contain 0xfefefefefefefefe
220 221 222 223 224 225 226 227 228 229 230 231 232 233 234
#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe)                \
    "movq   "#rega", "#regr"            \n\t"                    \
    "pand   "#regb", "#regr"            \n\t"                    \
    "pxor   "#rega", "#regb"            \n\t"                    \
    "pand  "#regfe", "#regb"            \n\t"                    \
    "psrlq       $1, "#regb"            \n\t"                    \
    "paddb  "#regb", "#regr"            \n\t"

#define PAVGB_MMX(rega, regb, regr, regfe)                       \
    "movq   "#rega", "#regr"            \n\t"                    \
    "por    "#regb", "#regr"            \n\t"                    \
    "pxor   "#rega", "#regb"            \n\t"                    \
    "pand  "#regfe", "#regb"            \n\t"                    \
    "psrlq       $1, "#regb"            \n\t"                    \
    "psubb  "#regb", "#regr"            \n\t"
235

236
// mm6 is supposed to contain 0xfefefefefefefefe
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
#define PAVGBP_MMX_NO_RND(rega, regb, regr,  regc, regd, regp)   \
    "movq  "#rega", "#regr"             \n\t"                    \
    "movq  "#regc", "#regp"             \n\t"                    \
    "pand  "#regb", "#regr"             \n\t"                    \
    "pand  "#regd", "#regp"             \n\t"                    \
    "pxor  "#rega", "#regb"             \n\t"                    \
    "pxor  "#regc", "#regd"             \n\t"                    \
    "pand    %%mm6, "#regb"             \n\t"                    \
    "pand    %%mm6, "#regd"             \n\t"                    \
    "psrlq      $1, "#regb"             \n\t"                    \
    "psrlq      $1, "#regd"             \n\t"                    \
    "paddb "#regb", "#regr"             \n\t"                    \
    "paddb "#regd", "#regp"             \n\t"

#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp)           \
    "movq  "#rega", "#regr"             \n\t"                    \
    "movq  "#regc", "#regp"             \n\t"                    \
    "por   "#regb", "#regr"             \n\t"                    \
    "por   "#regd", "#regp"             \n\t"                    \
    "pxor  "#rega", "#regb"             \n\t"                    \
    "pxor  "#regc", "#regd"             \n\t"                    \
    "pand    %%mm6, "#regb"             \n\t"                    \
    "pand    %%mm6, "#regd"             \n\t"                    \
    "psrlq      $1, "#regd"             \n\t"                    \
    "psrlq      $1, "#regb"             \n\t"                    \
    "psubb "#regb", "#regr"             \n\t"                    \
    "psubb "#regd", "#regp"             \n\t"
264

265 266
/***********************************/
/* MMX no rounding */
267
#define NO_RND 1
268
#define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
269
#define SET_RND  MOVQ_WONE
270 271
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
#define PAVGB(a, b, c, e)               PAVGB_MMX_NO_RND(a, b, c, e)
272
#define OP_AVG(a, b, c, e)              PAVGB_MMX(a, b, c, e)
273

274
#include "dsputil_rnd_template.c"
275 276

#undef DEF
277
#undef SET_RND
278
#undef PAVGBP
279
#undef PAVGB
280
#undef NO_RND
281 282 283
/***********************************/
/* MMX rounding */

284
#define DEF(x, y) x ## _ ## y ## _mmx
285
#define SET_RND  MOVQ_WTWO
286 287
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX(a, b, c, d, e, f)
#define PAVGB(a, b, c, e)               PAVGB_MMX(a, b, c, e)
288

289
#include "dsputil_rnd_template.c"
290 291

#undef DEF
292
#undef SET_RND
293
#undef PAVGBP
294
#undef PAVGB
295
#undef OP_AVG
296

297 298 299 300 301 302
#endif /* HAVE_INLINE_ASM */


#if HAVE_YASM
#define ff_put_pixels8_mmx ff_put_pixels8_mmxext

Fabrice Bellard's avatar
Fabrice Bellard committed
303 304 305 306 307
/***********************************/
/* 3Dnow specific */

#define DEF(x) x ## _3dnow

308
#include "dsputil_avg_template.c"
Fabrice Bellard's avatar
Fabrice Bellard committed
309 310 311 312

#undef DEF

/***********************************/
313
/* MMXEXT specific */
Fabrice Bellard's avatar
Fabrice Bellard committed
314

315
#define DEF(x) x ## _mmxext
Fabrice Bellard's avatar
Fabrice Bellard committed
316

317
#include "dsputil_avg_template.c"
Fabrice Bellard's avatar
Fabrice Bellard committed
318 319 320

#undef DEF

321 322 323 324
#endif /* HAVE_YASM */


#if HAVE_INLINE_ASM
Loren Merritt's avatar
Loren Merritt committed
325 326
#define put_no_rnd_pixels16_mmx put_pixels16_mmx
#define put_no_rnd_pixels8_mmx put_pixels8_mmx
327 328 329 330 331
#define put_pixels16_mmxext put_pixels16_mmx
#define put_pixels8_mmxext put_pixels8_mmx
#define put_pixels4_mmxext put_pixels4_mmx
#define put_no_rnd_pixels16_mmxext put_no_rnd_pixels16_mmx
#define put_no_rnd_pixels8_mmxext put_no_rnd_pixels8_mmx
332

Fabrice Bellard's avatar
Fabrice Bellard committed
333 334 335
/***********************************/
/* standard MMX */

Diego Biurrun's avatar
Diego Biurrun committed
336
void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
337
                               int line_size)
Fabrice Bellard's avatar
Fabrice Bellard committed
338
{
Diego Biurrun's avatar
Diego Biurrun committed
339
    const int16_t *p;
340
    uint8_t *pix;
Fabrice Bellard's avatar
Fabrice Bellard committed
341 342

    /* read the pixels */
343
    p   = block;
Fabrice Bellard's avatar
Fabrice Bellard committed
344
    pix = pixels;
345
    /* unrolled loop */
346
    __asm__ volatile (
347 348 349 350 351 352 353 354
        "movq      (%3), %%mm0          \n\t"
        "movq     8(%3), %%mm1          \n\t"
        "movq    16(%3), %%mm2          \n\t"
        "movq    24(%3), %%mm3          \n\t"
        "movq    32(%3), %%mm4          \n\t"
        "movq    40(%3), %%mm5          \n\t"
        "movq    48(%3), %%mm6          \n\t"
        "movq    56(%3), %%mm7          \n\t"
355 356 357 358 359 360 361 362 363
        "packuswb %%mm1, %%mm0          \n\t"
        "packuswb %%mm3, %%mm2          \n\t"
        "packuswb %%mm5, %%mm4          \n\t"
        "packuswb %%mm7, %%mm6          \n\t"
        "movq     %%mm0, (%0)           \n\t"
        "movq     %%mm2, (%0, %1)       \n\t"
        "movq     %%mm4, (%0, %1, 2)    \n\t"
        "movq     %%mm6, (%0, %2)       \n\t"
        :: "r"(pix), "r"((x86_reg)line_size), "r"((x86_reg)line_size * 3),
364
           "r"(p)
365
        : "memory");
366 367
    pix += line_size * 4;
    p   += 32;
368 369 370 371

    // if here would be an exact copy of the code above
    // compiler would generate some very strange code
    // thus using "r"
372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390
    __asm__ volatile (
        "movq       (%3), %%mm0         \n\t"
        "movq      8(%3), %%mm1         \n\t"
        "movq     16(%3), %%mm2         \n\t"
        "movq     24(%3), %%mm3         \n\t"
        "movq     32(%3), %%mm4         \n\t"
        "movq     40(%3), %%mm5         \n\t"
        "movq     48(%3), %%mm6         \n\t"
        "movq     56(%3), %%mm7         \n\t"
        "packuswb  %%mm1, %%mm0         \n\t"
        "packuswb  %%mm3, %%mm2         \n\t"
        "packuswb  %%mm5, %%mm4         \n\t"
        "packuswb  %%mm7, %%mm6         \n\t"
        "movq      %%mm0, (%0)          \n\t"
        "movq      %%mm2, (%0, %1)      \n\t"
        "movq      %%mm4, (%0, %1, 2)   \n\t"
        "movq      %%mm6, (%0, %2)      \n\t"
        :: "r"(pix), "r"((x86_reg)line_size), "r"((x86_reg)line_size * 3), "r"(p)
        : "memory");
Fabrice Bellard's avatar
Fabrice Bellard committed
391 392
}

393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409
#define put_signed_pixels_clamped_mmx_half(off)             \
    "movq          "#off"(%2), %%mm1        \n\t"           \
    "movq     16 + "#off"(%2), %%mm2        \n\t"           \
    "movq     32 + "#off"(%2), %%mm3        \n\t"           \
    "movq     48 + "#off"(%2), %%mm4        \n\t"           \
    "packsswb  8 + "#off"(%2), %%mm1        \n\t"           \
    "packsswb 24 + "#off"(%2), %%mm2        \n\t"           \
    "packsswb 40 + "#off"(%2), %%mm3        \n\t"           \
    "packsswb 56 + "#off"(%2), %%mm4        \n\t"           \
    "paddb              %%mm0, %%mm1        \n\t"           \
    "paddb              %%mm0, %%mm2        \n\t"           \
    "paddb              %%mm0, %%mm3        \n\t"           \
    "paddb              %%mm0, %%mm4        \n\t"           \
    "movq               %%mm1, (%0)         \n\t"           \
    "movq               %%mm2, (%0, %3)     \n\t"           \
    "movq               %%mm3, (%0, %3, 2)  \n\t"           \
    "movq               %%mm4, (%0, %1)     \n\t"
410

Diego Biurrun's avatar
Diego Biurrun committed
411
void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
412
                                      int line_size)
413
{
414
    x86_reg line_skip = line_size;
415
    x86_reg line_skip3;
416 417

    __asm__ volatile (
418 419 420 421 422 423 424 425
        "movq "MANGLE(ff_pb_80)", %%mm0     \n\t"
        "lea         (%3, %3, 2), %1        \n\t"
        put_signed_pixels_clamped_mmx_half(0)
        "lea         (%0, %3, 4), %0        \n\t"
        put_signed_pixels_clamped_mmx_half(64)
        : "+&r"(pixels), "=&r"(line_skip3)
        : "r"(block), "r"(line_skip)
        : "memory");
426 427
}

Diego Biurrun's avatar
Diego Biurrun committed
428
void ff_add_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
429
                               int line_size)
Fabrice Bellard's avatar
Fabrice Bellard committed
430
{
Diego Biurrun's avatar
Diego Biurrun committed
431
    const int16_t *p;
432
    uint8_t *pix;
Fabrice Bellard's avatar
Fabrice Bellard committed
433 434 435
    int i;

    /* read the pixels */
436
    p   = block;
Fabrice Bellard's avatar
Fabrice Bellard committed
437
    pix = pixels;
438 439
    MOVQ_ZERO(mm7);
    i = 4;
440
    do {
441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464
        __asm__ volatile (
            "movq        (%2), %%mm0    \n\t"
            "movq       8(%2), %%mm1    \n\t"
            "movq      16(%2), %%mm2    \n\t"
            "movq      24(%2), %%mm3    \n\t"
            "movq          %0, %%mm4    \n\t"
            "movq          %1, %%mm6    \n\t"
            "movq       %%mm4, %%mm5    \n\t"
            "punpcklbw  %%mm7, %%mm4    \n\t"
            "punpckhbw  %%mm7, %%mm5    \n\t"
            "paddsw     %%mm4, %%mm0    \n\t"
            "paddsw     %%mm5, %%mm1    \n\t"
            "movq       %%mm6, %%mm5    \n\t"
            "punpcklbw  %%mm7, %%mm6    \n\t"
            "punpckhbw  %%mm7, %%mm5    \n\t"
            "paddsw     %%mm6, %%mm2    \n\t"
            "paddsw     %%mm5, %%mm3    \n\t"
            "packuswb   %%mm1, %%mm0    \n\t"
            "packuswb   %%mm3, %%mm2    \n\t"
            "movq       %%mm0, %0       \n\t"
            "movq       %%mm2, %1       \n\t"
            : "+m"(*pix), "+m"(*(pix + line_size))
            : "r"(p)
            : "memory");
465 466
        pix += line_size * 2;
        p   += 16;
467
    } while (--i);
Fabrice Bellard's avatar
Fabrice Bellard committed
468 469
}

470 471
static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
                            int line_size, int h)
Fabrice Bellard's avatar
Fabrice Bellard committed
472
{
473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493
    __asm__ volatile (
        "lea   (%3, %3), %%"REG_a"      \n\t"
        ".p2align     3                 \n\t"
        "1:                             \n\t"
        "movq  (%1    ), %%mm0          \n\t"
        "movq  (%1, %3), %%mm1          \n\t"
        "movq     %%mm0, (%2)           \n\t"
        "movq     %%mm1, (%2, %3)       \n\t"
        "add  %%"REG_a", %1             \n\t"
        "add  %%"REG_a", %2             \n\t"
        "movq  (%1    ), %%mm0          \n\t"
        "movq  (%1, %3), %%mm1          \n\t"
        "movq     %%mm0, (%2)           \n\t"
        "movq     %%mm1, (%2, %3)       \n\t"
        "add  %%"REG_a", %1             \n\t"
        "add  %%"REG_a", %2             \n\t"
        "subl        $4, %0             \n\t"
        "jnz         1b                 \n\t"
        : "+g"(h), "+r"(pixels),  "+r"(block)
        : "r"((x86_reg)line_size)
        : "%"REG_a, "memory"
494
        );
Fabrice Bellard's avatar
Fabrice Bellard committed
495 496
}

497 498
static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels,
                             int line_size, int h)
Michael Niedermayer's avatar
Michael Niedermayer committed
499
{
500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528
    __asm__ volatile (
        "lea   (%3, %3), %%"REG_a"      \n\t"
        ".p2align     3                 \n\t"
        "1:                             \n\t"
        "movq  (%1    ), %%mm0          \n\t"
        "movq 8(%1    ), %%mm4          \n\t"
        "movq  (%1, %3), %%mm1          \n\t"
        "movq 8(%1, %3), %%mm5          \n\t"
        "movq     %%mm0,  (%2)          \n\t"
        "movq     %%mm4, 8(%2)          \n\t"
        "movq     %%mm1,  (%2, %3)      \n\t"
        "movq     %%mm5, 8(%2, %3)      \n\t"
        "add  %%"REG_a", %1             \n\t"
        "add  %%"REG_a", %2             \n\t"
        "movq  (%1    ), %%mm0          \n\t"
        "movq 8(%1    ), %%mm4          \n\t"
        "movq  (%1, %3), %%mm1          \n\t"
        "movq 8(%1, %3), %%mm5          \n\t"
        "movq     %%mm0,  (%2)          \n\t"
        "movq     %%mm4, 8(%2)          \n\t"
        "movq     %%mm1,  (%2, %3)      \n\t"
        "movq     %%mm5, 8(%2, %3)      \n\t"
        "add  %%"REG_a", %1             \n\t"
        "add  %%"REG_a", %2             \n\t"
        "subl        $4, %0             \n\t"
        "jnz         1b                 \n\t"
        : "+g"(h), "+r"(pixels),  "+r"(block)
        : "r"((x86_reg)line_size)
        : "%"REG_a, "memory"
529
        );
Michael Niedermayer's avatar
Michael Niedermayer committed
530 531
}

532
#define CLEAR_BLOCKS(name, n)                           \
Diego Biurrun's avatar
Diego Biurrun committed
533
static void name(int16_t *blocks)                       \
534 535 536 537 538 539 540 541 542 543 544 545 546 547 548
{                                                       \
    __asm__ volatile (                                  \
        "pxor %%mm7, %%mm7              \n\t"           \
        "mov     %1,        %%"REG_a"   \n\t"           \
        "1:                             \n\t"           \
        "movq %%mm7,   (%0, %%"REG_a")  \n\t"           \
        "movq %%mm7,  8(%0, %%"REG_a")  \n\t"           \
        "movq %%mm7, 16(%0, %%"REG_a")  \n\t"           \
        "movq %%mm7, 24(%0, %%"REG_a")  \n\t"           \
        "add    $32, %%"REG_a"          \n\t"           \
        "js      1b                     \n\t"           \
        :: "r"(((uint8_t *)blocks) + 128 * n),          \
           "i"(-128 * n)                                \
        : "%"REG_a                                      \
        );                                              \
Loren Merritt's avatar
Loren Merritt committed
549 550 551 552
}
CLEAR_BLOCKS(clear_blocks_mmx, 6)
CLEAR_BLOCKS(clear_block_mmx, 1)

Diego Biurrun's avatar
Diego Biurrun committed
553
static void clear_block_sse(int16_t *block)
554
{
555 556 557 558 559 560 561 562 563 564
    __asm__ volatile (
        "xorps  %%xmm0, %%xmm0          \n"
        "movaps %%xmm0,    (%0)         \n"
        "movaps %%xmm0,  16(%0)         \n"
        "movaps %%xmm0,  32(%0)         \n"
        "movaps %%xmm0,  48(%0)         \n"
        "movaps %%xmm0,  64(%0)         \n"
        "movaps %%xmm0,  80(%0)         \n"
        "movaps %%xmm0,  96(%0)         \n"
        "movaps %%xmm0, 112(%0)         \n"
Loren Merritt's avatar
Loren Merritt committed
565 566 567
        :: "r"(block)
        : "memory"
    );
568 569
}

Diego Biurrun's avatar
Diego Biurrun committed
570
static void clear_blocks_sse(int16_t *blocks)
571
{
572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587
    __asm__ volatile (
        "xorps  %%xmm0, %%xmm0              \n"
        "mov        %1,         %%"REG_a"   \n"
        "1:                                 \n"
        "movaps %%xmm0,    (%0, %%"REG_a")  \n"
        "movaps %%xmm0,  16(%0, %%"REG_a")  \n"
        "movaps %%xmm0,  32(%0, %%"REG_a")  \n"
        "movaps %%xmm0,  48(%0, %%"REG_a")  \n"
        "movaps %%xmm0,  64(%0, %%"REG_a")  \n"
        "movaps %%xmm0,  80(%0, %%"REG_a")  \n"
        "movaps %%xmm0,  96(%0, %%"REG_a")  \n"
        "movaps %%xmm0, 112(%0, %%"REG_a")  \n"
        "add      $128,         %%"REG_a"   \n"
        "js         1b                      \n"
        :: "r"(((uint8_t *)blocks) + 128 * 6),
           "i"(-128 * 6)
588 589 590 591
        : "%"REG_a
    );
}

592 593 594
static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w)
{
    x86_reg i = 0;
595 596
    __asm__ volatile (
        "jmp          2f                \n\t"
597
        "1:                             \n\t"
598 599 600 601 602 603 604 605 606
        "movq   (%1, %0), %%mm0         \n\t"
        "movq   (%2, %0), %%mm1         \n\t"
        "paddb     %%mm0, %%mm1         \n\t"
        "movq      %%mm1, (%2, %0)      \n\t"
        "movq  8(%1, %0), %%mm0         \n\t"
        "movq  8(%2, %0), %%mm1         \n\t"
        "paddb     %%mm0, %%mm1         \n\t"
        "movq      %%mm1, 8(%2, %0)     \n\t"
        "add         $16, %0            \n\t"
607
        "2:                             \n\t"
608 609 610 611
        "cmp          %3, %0            \n\t"
        "js           1b                \n\t"
        : "+r"(i)
        : "r"(src), "r"(dst), "r"((x86_reg)w - 15)
Michael Niedermayer's avatar
huffyuv  
Michael Niedermayer committed
612
    );
613 614
    for ( ; i < w; i++)
        dst[i + 0] += src[i + 0];
Michael Niedermayer's avatar
huffyuv  
Michael Niedermayer committed
615 616
}

617
#if HAVE_7REGS
618 619 620 621
static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top,
                                            const uint8_t *diff, int w,
                                            int *left, int *left_top)
{
622 623
    x86_reg w2 = -w;
    x86_reg x;
624
    int l  = *left     & 0xff;
625 626
    int tl = *left_top & 0xff;
    int t;
627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648
    __asm__ volatile (
        "mov          %7, %3            \n"
        "1:                             \n"
        "movzbl (%3, %4), %2            \n"
        "mov          %2, %k3           \n"
        "sub         %b1, %b3           \n"
        "add         %b0, %b3           \n"
        "mov          %2, %1            \n"
        "cmp          %0, %2            \n"
        "cmovg        %0, %2            \n"
        "cmovg        %1, %0            \n"
        "cmp         %k3, %0            \n"
        "cmovg       %k3, %0            \n"
        "mov          %7, %3            \n"
        "cmp          %2, %0            \n"
        "cmovl        %2, %0            \n"
        "add    (%6, %4), %b0           \n"
        "mov         %b0, (%5, %4)      \n"
        "inc          %4                \n"
        "jl           1b                \n"
        : "+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
        : "r"(dst + w), "r"(diff + w), "rm"(top + w)
649
    );
650
    *left     = l;
651 652 653 654
    *left_top = tl;
}
#endif

655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682
static inline void transpose4x4(uint8_t *dst, uint8_t *src, x86_reg dst_stride, x86_reg src_stride){
    __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...
        "movd  (%1), %%mm0              \n\t"
        "add   %3, %1                   \n\t"
        "movd  (%1), %%mm1              \n\t"
        "movd  (%1,%3,1), %%mm2         \n\t"
        "movd  (%1,%3,2), %%mm3         \n\t"
        "punpcklbw %%mm1, %%mm0         \n\t"
        "punpcklbw %%mm3, %%mm2         \n\t"
        "movq %%mm0, %%mm1              \n\t"
        "punpcklwd %%mm2, %%mm0         \n\t"
        "punpckhwd %%mm2, %%mm1         \n\t"
        "movd  %%mm0, (%0)              \n\t"
        "add   %2, %0                   \n\t"
        "punpckhdq %%mm0, %%mm0         \n\t"
        "movd  %%mm0, (%0)              \n\t"
        "movd  %%mm1, (%0,%2,1)         \n\t"
        "punpckhdq %%mm1, %%mm1         \n\t"
        "movd  %%mm1, (%0,%2,2)         \n\t"

        :  "+&r" (dst),
           "+&r" (src)
        :  "r" (dst_stride),
           "r" (src_stride)
        :  "memory"
    );
}

683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752
#define H263_LOOP_FILTER                        \
    "pxor      %%mm7, %%mm7             \n\t"   \
    "movq         %0, %%mm0             \n\t"   \
    "movq         %0, %%mm1             \n\t"   \
    "movq         %3, %%mm2             \n\t"   \
    "movq         %3, %%mm3             \n\t"   \
    "punpcklbw %%mm7, %%mm0             \n\t"   \
    "punpckhbw %%mm7, %%mm1             \n\t"   \
    "punpcklbw %%mm7, %%mm2             \n\t"   \
    "punpckhbw %%mm7, %%mm3             \n\t"   \
    "psubw     %%mm2, %%mm0             \n\t"   \
    "psubw     %%mm3, %%mm1             \n\t"   \
    "movq         %1, %%mm2             \n\t"   \
    "movq         %1, %%mm3             \n\t"   \
    "movq         %2, %%mm4             \n\t"   \
    "movq         %2, %%mm5             \n\t"   \
    "punpcklbw %%mm7, %%mm2             \n\t"   \
    "punpckhbw %%mm7, %%mm3             \n\t"   \
    "punpcklbw %%mm7, %%mm4             \n\t"   \
    "punpckhbw %%mm7, %%mm5             \n\t"   \
    "psubw     %%mm2, %%mm4             \n\t"   \
    "psubw     %%mm3, %%mm5             \n\t"   \
    "psllw        $2, %%mm4             \n\t"   \
    "psllw        $2, %%mm5             \n\t"   \
    "paddw     %%mm0, %%mm4             \n\t"   \
    "paddw     %%mm1, %%mm5             \n\t"   \
    "pxor      %%mm6, %%mm6             \n\t"   \
    "pcmpgtw   %%mm4, %%mm6             \n\t"   \
    "pcmpgtw   %%mm5, %%mm7             \n\t"   \
    "pxor      %%mm6, %%mm4             \n\t"   \
    "pxor      %%mm7, %%mm5             \n\t"   \
    "psubw     %%mm6, %%mm4             \n\t"   \
    "psubw     %%mm7, %%mm5             \n\t"   \
    "psrlw        $3, %%mm4             \n\t"   \
    "psrlw        $3, %%mm5             \n\t"   \
    "packuswb  %%mm5, %%mm4             \n\t"   \
    "packsswb  %%mm7, %%mm6             \n\t"   \
    "pxor      %%mm7, %%mm7             \n\t"   \
    "movd         %4, %%mm2             \n\t"   \
    "punpcklbw %%mm2, %%mm2             \n\t"   \
    "punpcklbw %%mm2, %%mm2             \n\t"   \
    "punpcklbw %%mm2, %%mm2             \n\t"   \
    "psubusb   %%mm4, %%mm2             \n\t"   \
    "movq      %%mm2, %%mm3             \n\t"   \
    "psubusb   %%mm4, %%mm3             \n\t"   \
    "psubb     %%mm3, %%mm2             \n\t"   \
    "movq         %1, %%mm3             \n\t"   \
    "movq         %2, %%mm4             \n\t"   \
    "pxor      %%mm6, %%mm3             \n\t"   \
    "pxor      %%mm6, %%mm4             \n\t"   \
    "paddusb   %%mm2, %%mm3             \n\t"   \
    "psubusb   %%mm2, %%mm4             \n\t"   \
    "pxor      %%mm6, %%mm3             \n\t"   \
    "pxor      %%mm6, %%mm4             \n\t"   \
    "paddusb   %%mm2, %%mm2             \n\t"   \
    "packsswb  %%mm1, %%mm0             \n\t"   \
    "pcmpgtb   %%mm0, %%mm7             \n\t"   \
    "pxor      %%mm7, %%mm0             \n\t"   \
    "psubb     %%mm7, %%mm0             \n\t"   \
    "movq      %%mm0, %%mm1             \n\t"   \
    "psubusb   %%mm2, %%mm0             \n\t"   \
    "psubb     %%mm0, %%mm1             \n\t"   \
    "pand         %5, %%mm1             \n\t"   \
    "psrlw        $2, %%mm1             \n\t"   \
    "pxor      %%mm7, %%mm1             \n\t"   \
    "psubb     %%mm7, %%mm1             \n\t"   \
    "movq         %0, %%mm5             \n\t"   \
    "movq         %3, %%mm6             \n\t"   \
    "psubb     %%mm1, %%mm5             \n\t"   \
    "paddb     %%mm1, %%mm6             \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
753

754 755 756 757
static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale)
{
    if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
        const int strength = ff_h263_loop_filter_strength[qscale];
Michael Niedermayer's avatar
Michael Niedermayer committed
758

759 760 761 762 763 764 765 766 767 768 769 770 771
        __asm__ volatile (
            H263_LOOP_FILTER

            "movq %%mm3, %1             \n\t"
            "movq %%mm4, %2             \n\t"
            "movq %%mm5, %0             \n\t"
            "movq %%mm6, %3             \n\t"
            : "+m"(*(uint64_t*)(src - 2 * stride)),
              "+m"(*(uint64_t*)(src - 1 * stride)),
              "+m"(*(uint64_t*)(src + 0 * stride)),
              "+m"(*(uint64_t*)(src + 1 * stride))
            : "g"(2 * strength), "m"(ff_pb_FC)
            );
772
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
773 774
}

775 776 777 778 779 780
static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale)
{
    if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
        const int strength = ff_h263_loop_filter_strength[qscale];
        DECLARE_ALIGNED(8, uint64_t, temp)[4];
        uint8_t *btemp = (uint8_t*)temp;
781

782
        src -= 2;
Michael Niedermayer's avatar
Michael Niedermayer committed
783

784 785
        transpose4x4(btemp,     src,              8, stride);
        transpose4x4(btemp + 4, src + 4 * stride, 8, stride);
786 787 788 789 790 791 792 793 794
        __asm__ volatile (
            H263_LOOP_FILTER // 5 3 4 6

            : "+m"(temp[0]),
              "+m"(temp[1]),
              "+m"(temp[2]),
              "+m"(temp[3])
            : "g"(2 * strength), "m"(ff_pb_FC)
            );
Michael Niedermayer's avatar
Michael Niedermayer committed
795

796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825
        __asm__ volatile (
            "movq      %%mm5, %%mm1         \n\t"
            "movq      %%mm4, %%mm0         \n\t"
            "punpcklbw %%mm3, %%mm5         \n\t"
            "punpcklbw %%mm6, %%mm4         \n\t"
            "punpckhbw %%mm3, %%mm1         \n\t"
            "punpckhbw %%mm6, %%mm0         \n\t"
            "movq      %%mm5, %%mm3         \n\t"
            "movq      %%mm1, %%mm6         \n\t"
            "punpcklwd %%mm4, %%mm5         \n\t"
            "punpcklwd %%mm0, %%mm1         \n\t"
            "punpckhwd %%mm4, %%mm3         \n\t"
            "punpckhwd %%mm0, %%mm6         \n\t"
            "movd      %%mm5, (%0)          \n\t"
            "punpckhdq %%mm5, %%mm5         \n\t"
            "movd      %%mm5, (%0, %2)      \n\t"
            "movd      %%mm3, (%0, %2, 2)   \n\t"
            "punpckhdq %%mm3, %%mm3         \n\t"
            "movd      %%mm3, (%0, %3)      \n\t"
            "movd      %%mm1, (%1)          \n\t"
            "punpckhdq %%mm1, %%mm1         \n\t"
            "movd      %%mm1, (%1, %2)      \n\t"
            "movd      %%mm6, (%1, %2, 2)   \n\t"
            "punpckhdq %%mm6, %%mm6         \n\t"
            "movd      %%mm6, (%1, %3)      \n\t"
            :: "r"(src),
               "r"(src + 4 * stride),
               "r"((x86_reg)stride),
               "r"((x86_reg)(3 * stride))
            );
826
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
827 828
}

829 830 831 832
/* Draw the edges of width 'w' of an image of size width, height
 * this MMX version can only handle w == 8 || w == 16. */
static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
                           int w, int h, int sides)
833 834 835 836 837 838 839
{
    uint8_t *ptr, *last_line;
    int i;

    last_line = buf + (height - 1) * wrap;
    /* left and right */
    ptr = buf;
840
    if (w == 8) {
841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858
        __asm__ volatile (
            "1:                             \n\t"
            "movd            (%0), %%mm0    \n\t"
            "punpcklbw      %%mm0, %%mm0    \n\t"
            "punpcklwd      %%mm0, %%mm0    \n\t"
            "punpckldq      %%mm0, %%mm0    \n\t"
            "movq           %%mm0, -8(%0)   \n\t"
            "movq      -8(%0, %2), %%mm1    \n\t"
            "punpckhbw      %%mm1, %%mm1    \n\t"
            "punpckhwd      %%mm1, %%mm1    \n\t"
            "punpckhdq      %%mm1, %%mm1    \n\t"
            "movq           %%mm1, (%0, %2) \n\t"
            "add               %1, %0       \n\t"
            "cmp               %3, %0       \n\t"
            "jb                1b           \n\t"
            : "+r"(ptr)
            : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
            );
859
    } else {
860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879
        __asm__ volatile (
            "1:                                 \n\t"
            "movd            (%0), %%mm0        \n\t"
            "punpcklbw      %%mm0, %%mm0        \n\t"
            "punpcklwd      %%mm0, %%mm0        \n\t"
            "punpckldq      %%mm0, %%mm0        \n\t"
            "movq           %%mm0, -8(%0)       \n\t"
            "movq           %%mm0, -16(%0)      \n\t"
            "movq      -8(%0, %2), %%mm1        \n\t"
            "punpckhbw      %%mm1, %%mm1        \n\t"
            "punpckhwd      %%mm1, %%mm1        \n\t"
            "punpckhdq      %%mm1, %%mm1        \n\t"
            "movq           %%mm1,  (%0, %2)    \n\t"
            "movq           %%mm1, 8(%0, %2)    \n\t"
            "add               %1, %0           \n\t"
            "cmp               %3, %0           \n\t"
            "jb                1b               \n\t"
            : "+r"(ptr)
            : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
            );
880 881
    }

882
    /* top and bottom (and hopefully also the corners) */
883 884 885
    if (sides & EDGE_TOP) {
        for (i = 0; i < h; i += 4) {
            ptr = buf - (i + 1) * wrap - w;
886 887 888 889 890 891 892 893 894 895 896 897 898 899
            __asm__ volatile (
                "1:                             \n\t"
                "movq (%1, %0), %%mm0           \n\t"
                "movq    %%mm0, (%0)            \n\t"
                "movq    %%mm0, (%0, %2)        \n\t"
                "movq    %%mm0, (%0, %2, 2)     \n\t"
                "movq    %%mm0, (%0, %3)        \n\t"
                "add        $8, %0              \n\t"
                "cmp        %4, %0              \n\t"
                "jb         1b                  \n\t"
                : "+r"(ptr)
                : "r"((x86_reg)buf - (x86_reg)ptr - w), "r"((x86_reg) -wrap),
                  "r"((x86_reg) -wrap * 3), "r"(ptr + width + 2 * w)
                );
900 901 902
        }
    }

903 904 905
    if (sides & EDGE_BOTTOM) {
        for (i = 0; i < h; i += 4) {
            ptr = last_line + (i + 1) * wrap - w;
906 907 908 909 910 911 912 913 914 915 916 917 918 919 920
            __asm__ volatile (
                "1:                             \n\t"
                "movq (%1, %0), %%mm0           \n\t"
                "movq    %%mm0, (%0)            \n\t"
                "movq    %%mm0, (%0, %2)        \n\t"
                "movq    %%mm0, (%0, %2, 2)     \n\t"
                "movq    %%mm0, (%0, %3)        \n\t"
                "add        $8, %0              \n\t"
                "cmp        %4, %0              \n\t"
                "jb         1b                  \n\t"
                : "+r"(ptr)
                : "r"((x86_reg)last_line - (x86_reg)ptr - w),
                  "r"((x86_reg)wrap), "r"((x86_reg)wrap * 3),
                  "r"(ptr + width + 2 * w)
                );
921
        }
922 923
    }
}
924
#endif /* HAVE_INLINE_ASM */
925

926

927
#if HAVE_YASM
928 929 930 931
#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)                          \
static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
932
    ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);              \
933 934 935 936 937 938 939
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t temp[8];                                                   \
    uint8_t * const half = (uint8_t*)temp;                              \
940 941 942 943
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
                                                   stride, 8);          \
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
                                        stride, stride, 8);             \
944 945 946 947 948
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
949 950
    ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride,    \
                                                   stride, 8);          \
951 952 953 954 955 956 957
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t temp[8];                                                   \
    uint8_t * const half = (uint8_t*)temp;                              \
958 959 960 961
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
                                                   stride, 8);          \
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride,     \
                                        stride, 8);                     \
962 963 964 965 966 967 968
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t temp[8];                                                   \
    uint8_t * const half = (uint8_t*)temp;                              \
969 970 971 972
    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
                                                   8, stride);          \
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
                                        stride, stride, 8);             \
973 974 975 976 977
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
978 979
    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src,            \
                                                   stride, stride);     \
980 981 982 983 984 985 986
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t temp[8];                                                   \
    uint8_t * const half = (uint8_t*)temp;                              \
987 988 989 990
    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
                                                   8, stride);          \
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\
                                        stride, 8);                     \
991 992 993 994 995 996 997 998
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t half[8 + 9];                                               \
    uint8_t * const halfH  = ((uint8_t*)half) + 64;                     \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
999 1000 1001 1002 1003 1004 1005
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
                                                   stride, 9);          \
    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
                                        stride, 9);                     \
    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
                                        stride, 8, 8);                  \
1006 1007 1008 1009 1010 1011 1012 1013
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t half[8 + 9];                                               \
    uint8_t * const halfH  = ((uint8_t*)half) + 64;                     \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
1014 1015 1016 1017 1018 1019 1020
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
                                                   stride, 9);          \
    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
                                        stride, 9);                     \
    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
                                        stride, 8, 8);                  \
1021 1022 1023 1024 1025 1026 1027 1028
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t half[8 + 9];                                               \
    uint8_t * const halfH  = ((uint8_t*)half) + 64;                     \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
1029 1030 1031 1032 1033 1034 1035
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
                                                   stride, 9);          \
    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
                                        stride, 9);                     \
    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
                                        stride, 8, 8);                  \
1036 1037 1038 1039 1040 1041 1042 1043
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t half[8 + 9];                                               \
    uint8_t * const halfH  = ((uint8_t*)half) + 64;                     \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
1044 1045 1046 1047 1048 1049 1050
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
                                                   stride, 9);          \
    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
                                        stride, 9);                     \
    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
                                        stride, 8, 8);                  \
1051 1052 1053 1054 1055 1056 1057 1058
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t half[8 + 9];                                               \
    uint8_t * const halfH  = ((uint8_t*)half) + 64;                     \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
1059 1060 1061 1062 1063
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
                                                   stride, 9);          \
    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
                                        stride, 8, 8);                  \
1064 1065 1066 1067 1068 1069 1070 1071
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t half[8 + 9];                                               \
    uint8_t * const halfH  = ((uint8_t*)half) + 64;                     \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
1072 1073 1074 1075 1076
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
                                                   stride, 9);          \
    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
                                        stride, 8, 8);                  \
1077 1078 1079 1080 1081 1082 1083
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t half[8 + 9];                                               \
    uint8_t * const halfH = ((uint8_t*)half);                           \
1084 1085 1086 1087 1088 1089
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
                                                   stride, 9);          \
    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH,              \
                                        8, stride, 9);                  \
    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
                                                   stride, 8);          \
1090 1091 1092 1093 1094 1095 1096
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t half[8 + 9];                                               \
    uint8_t * const halfH = ((uint8_t*)half);                           \
1097 1098 1099 1100 1101 1102
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
                                                   stride, 9);          \
    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
                                        stride, 9);                     \
    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
                                                   stride, 8);          \
1103 1104 1105 1106 1107 1108 1109
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    uint64_t half[9];                                                   \
    uint8_t * const halfH = ((uint8_t*)half);                           \
1110 1111 1112 1113
    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
                                                   stride, 9);          \
    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
                                                   stride, 8);          \
1114 1115 1116 1117 1118
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src,  \
                                           int stride)                  \
{                                                                       \
1119
    ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);            \
1120 1121 1122 1123 1124 1125 1126
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t temp[32];                                                  \
    uint8_t * const half = (uint8_t*)temp;                              \