dsputil_mmx.c 152 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 8
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
9 10
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
Fabrice Bellard's avatar
Fabrice Bellard committed
12
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
Fabrice Bellard's avatar
Fabrice Bellard committed
14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 16
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
Fabrice Bellard's avatar
Fabrice Bellard committed
17
 *
18
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with FFmpeg; if not, write to the Free Software
20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Fabrice Bellard's avatar
Fabrice Bellard committed
21 22 23 24
 *
 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
 */

25
#include "libavutil/cpu.h"
26
#include "libavutil/x86/asm.h"
27
#include "libavcodec/dsputil.h"
28
#include "libavcodec/h264dsp.h"
29 30
#include "libavcodec/mpegvideo.h"
#include "libavcodec/simple_idct.h"
31
#include "dsputil_mmx.h"
32
#include "idct_xvid.h"
33
#include "diracdsp_mmx.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
DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
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
    { 0x8000000080000000ULL, 0x8000000080000000ULL };

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 };
83 84 85

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 };
86

87 88
#if HAVE_INLINE_ASM

89 90
#define JUMPALIGN()     __asm__ volatile (".p2align 3"::)
#define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%"#regd", %%"#regd ::)
91

92 93 94 95
#define MOVQ_BFE(regd)                                  \
    __asm__ volatile (                                  \
        "pcmpeqd %%"#regd", %%"#regd"   \n\t"           \
        "paddb   %%"#regd", %%"#regd"   \n\t" ::)
96

97
#ifndef PIC
98 99
#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))
100 101 102
#else
// for shared library it's better to use this way for accessing constants
// pcmpeqd -> -1
103 104 105 106 107 108 109 110 111 112 113
#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"::)
114

115 116
#endif

117
// using regr as temporary and for the output result
118
// first argument is unmodifed and second is trashed
119
// regfe is supposed to contain 0xfefefefefefefefe
120 121 122 123 124 125 126 127 128 129 130 131 132 133 134
#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"
135

136
// mm6 is supposed to contain 0xfefefefefefefefe
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
#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"
164

165 166
/***********************************/
/* MMX no rounding */
167
#define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
168
#define SET_RND  MOVQ_WONE
169 170
#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)
171
#define OP_AVG(a, b, c, e)              PAVGB_MMX(a, b, c, e)
172

173
#include "dsputil_rnd_template.c"
174 175

#undef DEF
176
#undef SET_RND
177
#undef PAVGBP
178
#undef PAVGB
179 180 181
/***********************************/
/* MMX rounding */

182
#define DEF(x, y) x ## _ ## y ## _mmx
183
#define SET_RND  MOVQ_WTWO
184 185
#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)
186

187
#include "dsputil_rnd_template.c"
188 189

#undef DEF
190
#undef SET_RND
191
#undef PAVGBP
192
#undef PAVGB
193
#undef OP_AVG
194

Fabrice Bellard's avatar
Fabrice Bellard committed
195 196 197 198 199
/***********************************/
/* 3Dnow specific */

#define DEF(x) x ## _3dnow
#define PAVGB "pavgusb"
200
#define OP_AVG PAVGB
Fabrice Bellard's avatar
Fabrice Bellard committed
201

202
#include "dsputil_avg_template.c"
Fabrice Bellard's avatar
Fabrice Bellard committed
203 204 205

#undef DEF
#undef PAVGB
206
#undef OP_AVG
Fabrice Bellard's avatar
Fabrice Bellard committed
207 208 209 210

/***********************************/
/* MMX2 specific */

211
#define DEF(x) x ## _mmx2
Fabrice Bellard's avatar
Fabrice Bellard committed
212 213 214

/* Introduced only in MMX2 set */
#define PAVGB "pavgb"
215
#define OP_AVG PAVGB
Fabrice Bellard's avatar
Fabrice Bellard committed
216

217
#include "dsputil_avg_template.c"
Fabrice Bellard's avatar
Fabrice Bellard committed
218 219 220

#undef DEF
#undef PAVGB
221
#undef OP_AVG
Fabrice Bellard's avatar
Fabrice Bellard committed
222

Loren Merritt's avatar
Loren Merritt committed
223 224
#define put_no_rnd_pixels16_mmx put_pixels16_mmx
#define put_no_rnd_pixels8_mmx put_pixels8_mmx
225 226 227 228 229 230 231 232 233 234 235
#define put_pixels16_mmx2 put_pixels16_mmx
#define put_pixels8_mmx2 put_pixels8_mmx
#define put_pixels4_mmx2 put_pixels4_mmx
#define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx
#define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx
#define put_pixels16_3dnow put_pixels16_mmx
#define put_pixels8_3dnow put_pixels8_mmx
#define put_pixels4_3dnow put_pixels4_mmx
#define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx
#define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx

Fabrice Bellard's avatar
Fabrice Bellard committed
236 237 238
/***********************************/
/* standard MMX */

239 240
void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels,
                               int line_size)
Fabrice Bellard's avatar
Fabrice Bellard committed
241 242
{
    const DCTELEM *p;
243
    uint8_t *pix;
Fabrice Bellard's avatar
Fabrice Bellard committed
244 245

    /* read the pixels */
246
    p   = block;
Fabrice Bellard's avatar
Fabrice Bellard committed
247
    pix = pixels;
248
    /* unrolled loop */
249
    __asm__ volatile (
250 251 252 253 254 255 256 257
        "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"
258 259 260 261 262 263 264 265 266
        "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),
267
           "r"(p)
268
        : "memory");
269 270
    pix += line_size * 4;
    p   += 32;
271 272 273 274

    // if here would be an exact copy of the code above
    // compiler would generate some very strange code
    // thus using "r"
275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293
    __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
294 295
}

296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312
#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"
313

314 315
void ff_put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels,
                                      int line_size)
316
{
317
    x86_reg line_skip = line_size;
318
    x86_reg line_skip3;
319 320

    __asm__ volatile (
321 322 323 324 325 326 327 328
        "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");
329 330
}

331 332
void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels,
                               int line_size)
Fabrice Bellard's avatar
Fabrice Bellard committed
333 334
{
    const DCTELEM *p;
335
    uint8_t *pix;
Fabrice Bellard's avatar
Fabrice Bellard committed
336 337 338
    int i;

    /* read the pixels */
339
    p   = block;
Fabrice Bellard's avatar
Fabrice Bellard committed
340
    pix = pixels;
341 342
    MOVQ_ZERO(mm7);
    i = 4;
343
    do {
344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
        __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");
368 369
        pix += line_size * 2;
        p   += 16;
370
    } while (--i);
Fabrice Bellard's avatar
Fabrice Bellard committed
371 372
}

373 374
static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels,
                            int line_size, int h)
375
{
376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396
    __asm__ volatile (
        "lea   (%3, %3), %%"REG_a"      \n\t"
        ".p2align     3                 \n\t"
        "1:                             \n\t"
        "movd  (%1    ), %%mm0          \n\t"
        "movd  (%1, %3), %%mm1          \n\t"
        "movd     %%mm0, (%2)           \n\t"
        "movd     %%mm1, (%2, %3)       \n\t"
        "add  %%"REG_a", %1             \n\t"
        "add  %%"REG_a", %2             \n\t"
        "movd  (%1    ), %%mm0          \n\t"
        "movd  (%1, %3), %%mm1          \n\t"
        "movd     %%mm0, (%2)           \n\t"
        "movd     %%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"
397
        );
398 399
}

400 401
static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
                            int line_size, int h)
Fabrice Bellard's avatar
Fabrice Bellard committed
402
{
403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423
    __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"
424
        );
Fabrice Bellard's avatar
Fabrice Bellard committed
425 426
}

427 428
static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels,
                             int line_size, int h)
Michael Niedermayer's avatar
Michael Niedermayer committed
429
{
430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458
    __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"
459
        );
Michael Niedermayer's avatar
Michael Niedermayer committed
460 461
}

462 463
static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
                              int line_size, int h)
464
{
465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481
    __asm__ volatile (
        "1:                              \n\t"
        "movdqu (%1       ), %%xmm0      \n\t"
        "movdqu (%1, %3   ), %%xmm1      \n\t"
        "movdqu (%1, %3, 2), %%xmm2      \n\t"
        "movdqu (%1, %4   ), %%xmm3      \n\t"
        "lea    (%1, %3, 4), %1          \n\t"
        "movdqa      %%xmm0, (%2)        \n\t"
        "movdqa      %%xmm1, (%2, %3)    \n\t"
        "movdqa      %%xmm2, (%2, %3, 2) \n\t"
        "movdqa      %%xmm3, (%2, %4)    \n\t"
        "subl            $4, %0          \n\t"
        "lea    (%2, %3, 4), %2          \n\t"
        "jnz             1b              \n\t"
        : "+g"(h), "+r"(pixels),  "+r"(block)
        : "r"((x86_reg)line_size), "r"((x86_reg)3L * line_size)
        : "memory"
482 483 484
        );
}

485 486
static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
                              int line_size, int h)
487
{
488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508
    __asm__ volatile (
        "1:                                 \n\t"
        "movdqu (%1       ), %%xmm0         \n\t"
        "movdqu (%1, %3   ), %%xmm1         \n\t"
        "movdqu (%1, %3, 2), %%xmm2         \n\t"
        "movdqu (%1, %4   ), %%xmm3         \n\t"
        "lea    (%1, %3, 4), %1             \n\t"
        "pavgb  (%2       ), %%xmm0         \n\t"
        "pavgb  (%2, %3   ), %%xmm1         \n\t"
        "pavgb  (%2, %3, 2), %%xmm2         \n\t"
        "pavgb     (%2, %4), %%xmm3         \n\t"
        "movdqa      %%xmm0, (%2)           \n\t"
        "movdqa      %%xmm1, (%2, %3)       \n\t"
        "movdqa      %%xmm2, (%2, %3, 2)    \n\t"
        "movdqa      %%xmm3, (%2, %4)       \n\t"
        "subl            $4, %0             \n\t"
        "lea    (%2, %3, 4), %2             \n\t"
        "jnz             1b                 \n\t"
        : "+g"(h), "+r"(pixels),  "+r"(block)
        : "r"((x86_reg)line_size), "r"((x86_reg)3L * line_size)
        : "memory"
509 510 511
        );
}

512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528
#define CLEAR_BLOCKS(name, n)                           \
static void name(DCTELEM *blocks)                       \
{                                                       \
    __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
529 530 531 532 533
}
CLEAR_BLOCKS(clear_blocks_mmx, 6)
CLEAR_BLOCKS(clear_block_mmx, 1)

static void clear_block_sse(DCTELEM *block)
534
{
535 536 537 538 539 540 541 542 543 544
    __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
545 546 547
        :: "r"(block)
        : "memory"
    );
548 549
}

550
static void clear_blocks_sse(DCTELEM *blocks)
551
{
552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567
    __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)
568 569 570 571
        : "%"REG_a
    );
}

572 573 574
static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w)
{
    x86_reg i = 0;
575 576
    __asm__ volatile (
        "jmp          2f                \n\t"
577
        "1:                             \n\t"
578 579 580 581 582 583 584 585 586
        "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"
587
        "2:                             \n\t"
588 589 590 591
        "cmp          %3, %0            \n\t"
        "js           1b                \n\t"
        : "+r"(i)
        : "r"(src), "r"(dst), "r"((x86_reg)w - 15)
Michael Niedermayer's avatar
Michael Niedermayer committed
592
    );
593 594
    for ( ; i < w; i++)
        dst[i + 0] += src[i + 0];
Michael Niedermayer's avatar
Michael Niedermayer committed
595 596
}

597
#if HAVE_7REGS
598 599 600 601
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)
{
602 603
    x86_reg w2 = -w;
    x86_reg x;
604
    int l  = *left     & 0xff;
605 606
    int tl = *left_top & 0xff;
    int t;
607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628
    __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)
629
    );
630
    *left     = l;
631 632 633 634
    *left_top = tl;
}
#endif

635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662
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"
    );
}

663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732
#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"
733

734 735 736 737
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];
738

739 740 741 742 743 744 745 746 747 748 749 750 751
        __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)
            );
752
    }
753 754
}

755 756 757 758 759 760
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;
761

762
        src -= 2;
763

764 765
        transpose4x4(btemp,     src,              8, stride);
        transpose4x4(btemp + 4, src + 4 * stride, 8, stride);
766 767 768 769 770 771 772 773 774
        __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)
            );
775

776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805
        __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))
            );
806
    }
807 808
}

809 810 811 812
/* 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)
813 814 815 816 817 818 819
{
    uint8_t *ptr, *last_line;
    int i;

    last_line = buf + (height - 1) * wrap;
    /* left and right */
    ptr = buf;
820
    if (w == 8) {
821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838
        __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)
            );
839
    } else if(w==16){
840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856
        __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"
857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875
            : "+r"(ptr)
            : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
            );
    } else {
        av_assert1(w == 4);
        __asm__ volatile (
            "1:                             \n\t"
            "movd            (%0), %%mm0    \n\t"
            "punpcklbw      %%mm0, %%mm0    \n\t"
            "punpcklwd      %%mm0, %%mm0    \n\t"
            "movd           %%mm0, -4(%0)   \n\t"
            "movd      -4(%0, %2), %%mm1    \n\t"
            "punpcklbw      %%mm1, %%mm1    \n\t"
            "punpckhwd      %%mm1, %%mm1    \n\t"
            "punpckhdq      %%mm1, %%mm1    \n\t"
            "movd           %%mm1, (%0, %2) \n\t"
            "add               %1, %0       \n\t"
            "cmp               %3, %0       \n\t"
            "jb                1b           \n\t"
876 877 878
            : "+r"(ptr)
            : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
            );
879 880
    }

881
    /* top and bottom (and hopefully also the corners) */
882 883 884
    if (sides & EDGE_TOP) {
        for (i = 0; i < h; i += 4) {
            ptr = buf - (i + 1) * wrap - w;
885 886 887 888 889 890 891 892 893 894 895 896 897 898
            __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)
                );
899
        }
900
    }
901

902 903 904
    if (sides & EDGE_BOTTOM) {
        for (i = 0; i < h; i += 4) {
            ptr = last_line + (i + 1) * wrap - w;
905 906 907 908 909 910 911 912 913 914 915 916 917 918 919
            __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)
                );
920
        }
921 922 923
    }
}

924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
#define QPEL_V_LOW(m3, m4, m5, m6, pw_20, pw_3, rnd,                      \
                   in0, in1, in2, in7, out, OP)                           \
    "paddw               "#m4", "#m3"   \n\t" /* x1 */                    \
    "movq   "MANGLE(ff_pw_20)", %%mm4   \n\t" /* 20 */                    \
    "pmullw              "#m3", %%mm4   \n\t" /* 20x1 */                  \
    "movq               "#in7", "#m3"   \n\t" /* d */                     \
    "movq               "#in0", %%mm5   \n\t" /* D */                     \
    "paddw               "#m3", %%mm5   \n\t" /* x4 */                    \
    "psubw               %%mm5, %%mm4   \n\t" /* 20x1 - x4 */             \
    "movq               "#in1", %%mm5   \n\t" /* C */                     \
    "movq               "#in2", %%mm6   \n\t" /* B */                     \
    "paddw               "#m6", %%mm5   \n\t" /* x3 */                    \
    "paddw               "#m5", %%mm6   \n\t" /* x2 */                    \
    "paddw               %%mm6, %%mm6   \n\t" /* 2x2 */                   \
    "psubw               %%mm6, %%mm5   \n\t" /* -2x2 + x3 */             \
    "pmullw  "MANGLE(ff_pw_3)", %%mm5   \n\t" /* -6x2 + 3x3 */            \
    "paddw              "#rnd", %%mm4   \n\t" /* x2 */                    \
    "paddw               %%mm4, %%mm5   \n\t" /* 20x1 - 6x2 + 3x3 - x4 */ \
    "psraw                  $5, %%mm5   \n\t"                             \
    "packuswb            %%mm5, %%mm5   \n\t"                             \
    OP(%%mm5, out, %%mm7, d)

#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)                \
static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst,           \
                                                  uint8_t *src,           \
                                                  int dstStride,          \
                                                  int srcStride,          \
                                                  int h)                  \
{                                                                         \
    uint64_t temp;                                                        \
                                                                          \
    __asm__ volatile (                                                    \
        "pxor      %%mm7, %%mm7             \n\t"                         \
        "1:                                 \n\t"                         \
        "movq       (%0), %%mm0             \n\t" /* ABCDEFGH */          \
        "movq      %%mm0, %%mm1             \n\t" /* ABCDEFGH */          \
        "movq      %%mm0, %%mm2             \n\t" /* ABCDEFGH */          \
        "punpcklbw %%mm7, %%mm0             \n\t" /* 0A0B0C0D */          \
        "punpckhbw %%mm7, %%mm1             \n\t" /* 0E0F0G0H */          \
        "pshufw    $0x90, %%mm0, %%mm5      \n\t" /* 0A0A0B0C */          \
        "pshufw    $0x41, %%mm0, %%mm6      \n\t" /* 0B0A0A0B */          \
        "movq      %%mm2, %%mm3             \n\t" /* ABCDEFGH */          \
        "movq      %%mm2, %%mm4             \n\t" /* ABCDEFGH */          \
        "psllq        $8, %%mm2             \n\t" /* 0ABCDEFG */          \
        "psllq       $16, %%mm3             \n\t" /* 00ABCDEF */          \
        "psllq       $24, %%mm4             \n\t" /* 000ABCDE */          \
        "punpckhbw %%mm7, %%mm2             \n\t" /* 0D0E0F0G */          \
        "punpckhbw %%mm7, %%mm3             \n\t" /* 0C0D0E0F */          \
        "punpckhbw %%mm7, %%mm4             \n\t" /* 0B0C0D0E */          \
        "paddw     %%mm3, %%mm5             \n\t" /* b */                 \
        "paddw     %%mm2, %%mm6             \n\t" /* c */                 \
        "paddw     %%mm5, %%mm5             \n\t" /* 2b */                \
        "psubw     %%mm5, %%mm6             \n\t" /* c - 2b */            \
        "pshufw    $0x06, %%mm0, %%mm5      \n\t" /* 0C0B0A0A */          \
        "pmullw "MANGLE(ff_pw_3)", %%mm6    \n\t" /* 3c - 6b */           \
        "paddw     %%mm4, %%mm0             \n\t" /* a */                 \
        "paddw     %%mm1, %%mm5             \n\t" /* d */                 \
        "pmullw "MANGLE(ff_pw_20)", %%mm0   \n\t" /* 20a */               \
        "psubw     %%mm5, %%mm0             \n\t" /* 20a - d */           \
        "paddw        %6, %%mm6             \n\t"                         \
        "paddw     %%mm6, %%mm0             \n\t" /* 20a - 6b + 3c - d */ \
        "psraw        $5, %%mm0             \n\t"                         \
        "movq      %%mm0, %5                \n\t"                         \
        /* mm1 = EFGH, mm2 = DEFG, mm3 = CDEF, mm4 = BCDE, mm7 = 0 */     \
                                                                          \
        "movq      5(%0), %%mm0             \n\t" /* FGHIJKLM */          \
        "movq      %%mm0, %%mm5             \n\t" /* FGHIJKLM */          \
        "movq      %%mm0, %%mm6             \n\t" /* FGHIJKLM */          \
        "psrlq        $8, %%mm0             \n\t" /* GHIJKLM0 */          \
        "psrlq       $16, %%mm5             \n\t" /* HIJKLM00 */          \
        "punpcklbw %%mm7, %%mm0             \n\t" /* 0G0H0I0J */          \
        "punpcklbw %%mm7, %%mm5             \n\t" /* 0H0I0J0K */          \
        "paddw     %%mm0, %%mm2             \n\t" /* b */                 \
        "paddw     %%mm5, %%mm3             \n\t" /* c */                 \
        "paddw     %%mm2, %%mm2             \n\t" /* 2b */                \
        "psubw     %%mm2, %%mm3             \n\t" /* c - 2b */            \
        "movq      %%mm6, %%mm2             \n\t" /* FGHIJKLM */          \
        "psrlq       $24, %%mm6             \n\t" /* IJKLM000 */          \
        "punpcklbw %%mm7, %%mm2             \n\t" /* 0F0G0H0I */          \
        "punpcklbw %%mm7, %%mm6             \n\t" /* 0I0J0K0L */          \
        "pmullw "MANGLE(ff_pw_3)", %%mm3    \n\t" /* 3c - 6b */           \
        "paddw     %%mm2, %%mm1             \n\t" /* a */                 \
        "paddw     %%mm6, %%mm4             \n\t" /* d */                 \
        "pmullw "MANGLE(ff_pw_20)", %%mm1   \n\t" /* 20a */               \
        "psubw     %%mm4, %%mm3             \n\t" /* - 6b +3c - d */      \
        "paddw        %6, %%mm1             \n\t"                         \
        "paddw     %%mm1, %%mm3             \n\t" /* 20a - 6b +3c - d */  \
        "psraw        $5, %%mm3             \n\t"                         \
        "movq         %5, %%mm1             \n\t"                         \
        "packuswb  %%mm3, %%mm1             \n\t"                         \
        OP_MMX2(%%mm1, (%1), %%mm4, q)                                    \
        /* mm0 = GHIJ, mm2 = FGHI, mm5 = HIJK, mm6 = IJKL, mm7 = 0 */     \
                                                                          \
        "movq      9(%0), %%mm1             \n\t" /* JKLMNOPQ */          \
        "movq      %%mm1, %%mm4             \n\t" /* JKLMNOPQ */          \
        "movq      %%mm1, %%mm3             \n\t" /* JKLMNOPQ */          \
        "psrlq        $8, %%mm1             \n\t" /* KLMNOPQ0 */          \
        "psrlq       $16, %%mm4             \n\t" /* LMNOPQ00 */          \
        "punpcklbw %%mm7, %%mm1             \n\t" /* 0K0L0M0N */          \
        "punpcklbw %%mm7, %%mm4             \n\t" /* 0L0M0N0O */          \
        "paddw     %%mm1, %%mm5             \n\t" /* b */                 \
        "paddw     %%mm4, %%mm0             \n\t" /* c */                 \
        "paddw     %%mm5, %%mm5             \n\t" /* 2b */                \
        "psubw     %%mm5, %%mm0             \n\t" /* c - 2b */            \
        "movq      %%mm3, %%mm5             \n\t" /* JKLMNOPQ */          \
        "psrlq       $24, %%mm3             \n\t" /* MNOPQ000 */          \
        "pmullw "MANGLE(ff_pw_3)", %%mm0    \n\t" /* 3c - 6b */           \
        "punpcklbw %%mm7, %%mm3             \n\t" /* 0M0N0O0P */          \
        "paddw     %%mm3, %%mm2             \n\t" /* d */                 \
        "psubw     %%mm2, %%mm0             \n\t" /* -6b + 3c - d */      \
        "movq      %%mm5, %%mm2             \n\t" /* JKLMNOPQ */          \
        "punpcklbw %%mm7, %%mm2             \n\t" /* 0J0K0L0M */          \
        "punpckhbw %%mm7, %%mm5             \n\t" /* 0N0O0P0Q */          \
        "paddw     %%mm2, %%mm6             \n\t" /* a */                 \
        "pmullw "MANGLE(ff_pw_20)", %%mm6   \n\t" /* 20a */               \
        "paddw        %6, %%mm0             \n\t"                         \
        "paddw     %%mm6, %%mm0             \n\t" /* 20a - 6b + 3c - d */ \
        "psraw        $5, %%mm0             \n\t"                         \
        /* mm1 = KLMN, mm2 = JKLM, mm3 = MNOP, */                         \
        /* mm4 = LMNO, mm5 = NOPQ mm7 = 0 */                              \
                                                                          \
        "paddw    %%mm5, %%mm3              \n\t" /* a */                 \
        "pshufw   $0xF9, %%mm5, %%mm6       \n\t" /* 0O0P0Q0Q */          \
        "paddw    %%mm4, %%mm6              \n\t" /* b */                 \
        "pshufw   $0xBE, %%mm5, %%mm4       \n\t" /* 0P0Q0Q0P */          \
        "pshufw   $0x6F, %%mm5, %%mm5       \n\t" /* 0Q0Q0P0O */          \
        "paddw    %%mm1, %%mm4              \n\t" /* c */                 \
        "paddw    %%mm2, %%mm5              \n\t" /* d */                 \
        "paddw    %%mm6, %%mm6              \n\t" /* 2b */                \
        "psubw    %%mm6, %%mm4              \n\t" /* c - 2b */            \
        "pmullw "MANGLE(ff_pw_20)", %%mm3   \n\t" /* 20a */               \
        "pmullw  "MANGLE(ff_pw_3)", %%mm4   \n\t" /* 3c - 6b */           \
        "psubw    %%mm5, %%mm3              \n\t" /* -6b + 3c - d */      \
        "paddw       %6, %%mm4              \n\t"                         \
        "paddw    %%mm3, %%mm4              \n\t" /* 20a - 6b + 3c - d */ \
        "psraw       $5, %%mm4              \n\t"                         \
        "packuswb %%mm4, %%mm0              \n\t"                         \
        OP_MMX2(%%mm0, 8(%1), %%mm4, q)                                   \
                                                                          \
        "add         %3, %0                 \n\t"                         \
        "add         %4, %1                 \n\t"                         \
        "decl        %2                     \n\t"                         \
        "jnz         1b                     \n\t"                         \
        : "+a"(src), "+c"(dst), "+D"(h)                                   \
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride),               \
          /* "m"(ff_pw_20), "m"(ff_pw_3), */ "m"(temp), "m"(ROUNDER)      \
        : "memory"                                                        \
        );                                                                \
}                                                                         \
                                                                          \
static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst,          \
                                                   uint8_t *src,          \
                                                   int dstStride,         \
                                                   int srcStride,         \
                                                   int h)                 \
{                                                                         \
    int i;                                                                \
    int16_t temp[16];                                                     \
    /* quick HACK, XXX FIXME MUST be optimized */                         \
    for (i = 0; i < h; i++) {                                             \
        temp[ 0] = (src[ 0] + src[ 1]) * 20 - (src[ 0] + src[ 2]) * 6 +   \
                   (src[ 1] + src[ 3]) *  3 - (src[ 2] + src[ 4]);        \
        temp[ 1] = (src[ 1] + src[ 2]) * 20 - (src[ 0] + src[ 3]) * 6 +   \
                   (src[ 0] + src[ 4]) *  3 - (src[ 1] + src[ 5]);        \
        temp[ 2] = (src[ 2] + src[ 3]) * 20 - (src[ 1] + src[ 4]) * 6 +   \
                   (src[ 0] + src[ 5]) *  3 - (src[ 0] + src[ 6]);        \
        temp[ 3] = (src[ 3] + src[ 4]) * 20 - (src[ 2] + src[ 5]) * 6 +   \
                   (src[ 1] + src[ 6]) *  3 - (src[ 0] + src[ 7]);        \
        temp[ 4] = (src[ 4] + src[ 5]) * 20 - (src[ 3] + src[ 6]) * 6 +   \
                   (src[ 2] + src[ 7]) *  3 - (src[ 1] + src[ 8]);        \
        temp[ 5] = (src[ 5] + src[ 6]) * 20 - (src[ 4] + src[ 7]) * 6 +   \
                   (src[ 3] + src[ 8]) *  3 - (src[ 2] + src[ 9]);        \
        temp[ 6] = (src[ 6] + src[ 7]) * 20 - (src[ 5] + src[ 8]) * 6 +   \
                   (src[ 4] + src[ 9]) *  3 - (src[ 3] + src[10]);        \
        temp[ 7] = (src[ 7] + src[ 8]) * 20 - (src[ 6] + src[ 9]) * 6 +   \
                   (src[ 5] + src[10]) *  3 - (src[ 4] + src[11]);        \
        temp[ 8] = (src[ 8] + src[ 9]) * 20 - (src[ 7] + src[10]) * 6 +   \
                   (src[ 6] + src[11]) *  3 - (src[ 5] + src[12]);        \
        temp[ 9] = (src[ 9] + src[10]) * 20 - (src[ 8] + src[11]) * 6 +   \
                   (src[ 7] + src[12]) *  3 - (src[ 6] + src[13]);        \
        temp[10] = (src[10] + src[11]) * 20 - (src[ 9] + src[12]) * 6 +   \
                   (src[ 8] + src[13]) *  3 - (src[ 7] + src[14]);        \
        temp[11] = (src[11] + src[12]) * 20 - (src[10] + src[13]) * 6 +   \
                   (src[ 9] + src[14]) *  3 - (src[ 8] + src[15]);        \
        temp[12] = (src[12] + src[13]) * 20 - (src[11] + src[14]) * 6 +   \
                   (src[10] + src[15]) *  3 - (src[ 9] + src[16]);        \
        temp[13] = (src[13] + src[14]) * 20 - (src[12] + src[15]) * 6 +   \
                   (src[11] + src[16]) *  3 - (src[10] + src[16]);        \
        temp[14] = (src[14] + src[15]) * 20 - (src[13] + src[16]) * 6 +   \
                   (src[12] + src[16]) *  3 - (src[11] + src[15]);        \
        temp[15] = (src[15] + src[16]) * 20 - (src[14] + src[16]) * 6 +   \
                   (src[13] + src[15]) *  3 - (src[12] + src[14]);        \
        __asm__ volatile (                                                \
            "movq      (%0), %%mm0          \n\t"                         \
            "movq     8(%0), %%mm1          \n\t"                         \
            "paddw       %2, %%mm0          \n\t"                         \
            "paddw       %2, %%mm1          \n\t"                         \
            "psraw       $5, %%mm0          \n\t"                         \
            "psraw       $5, %%mm1          \n\t"                         \
            "packuswb %%mm1, %%mm0          \n\t"                         \
            OP_3DNOW(%%mm0, (%1), %%mm1, q)                               \
            "movq    16(%0), %%mm0          \n\t"                         \
            "movq    24(%0), %%mm1          \n\t"                         \
            "paddw       %2, %%mm0          \n\t"                         \
            "paddw       %2, %%mm1          \n\t"                         \
            "psraw       $5, %%mm0          \n\t"                         \
            "psraw       $5, %%mm1          \n\t"                         \
            "packuswb %%mm1, %%mm0          \n\t"                         \
            OP_3DNOW(%%mm0, 8(%1), %%mm1, q)                              \
            :: "r"(temp), "r"(dst), "m"(ROUNDER)                          \
            : "memory"                                                    \
            );                                                            \
        dst += dstStride;                                                 \
        src += srcStride;                                                 \
    }                                                                     \
}                                                                         \
                                                                          \
static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst,            \
                                                 uint8_t *src,            \
                                                 int dstStride,           \
                                                 int srcStride,           \
                                                 int h)                   \
{                                                                         \
    __asm__ volatile (                                                    \
        "pxor      %%mm7, %%mm7             \n\t"                         \
        "1:                                 \n\t"                         \
        "movq       (%0), %%mm0             \n\t" /* ABCDEFGH */          \
        "movq      %%mm0, %%mm1             \n\t" /* ABCDEFGH */          \
        "movq      %%mm0, %%mm2             \n\t" /* ABCDEFGH */          \
        "punpcklbw %%mm7, %%mm0             \n\t" /* 0A0B0C0D */          \
        "punpckhbw %%mm7, %%mm1             \n\t" /* 0E0F0G0H */          \
        "pshufw    $0x90, %%mm0, %%mm5      \n\t" /* 0A0A0B0C */          \
        "pshufw    $0x41, %%mm0, %%mm6      \n\t" /* 0B0A0A0B */          \
        "movq      %%mm2, %%mm3             \n\t" /* ABCDEFGH */          \
        "movq      %%mm2, %%mm4             \n\t" /* ABCDEFGH */          \
        "psllq        $8, %%mm2             \n\t" /* 0ABCDEFG */          \
        "psllq       $16, %%mm3             \n\t" /* 00ABCDEF */          \
        "psllq       $24, %%mm4             \n\t" /* 000ABCDE */          \
        "punpckhbw %%mm7, %%mm2             \n\t" /* 0D0E0F0G */          \
        "punpckhbw %%mm7, %%mm3             \n\t" /* 0C0D0E0F */          \
        "punpckhbw %%mm7, %%mm4             \n\t" /* 0B0C0D0E */          \
        "paddw     %%mm3, %%mm5             \n\t" /* b */                 \
        "paddw     %%mm2, %%mm6             \n\t" /* c */                 \
        "paddw     %%mm5, %%mm5             \n\t" /* 2b */                \
        "psubw     %%mm5, %%mm6             \n\t" /* c - 2b */            \
        "pshufw    $0x06, %%mm0, %%mm5      \n\t" /* 0C0B0A0A */          \
        "pmullw "MANGLE(ff_pw_3)", %%mm6    \n\t" /* 3c - 6b */           \
        "paddw     %%mm4, %%mm0             \n\t" /* a */                 \
        "paddw     %%mm1, %%mm5             \n\t" /* d */                 \
        "pmullw "MANGLE(ff_pw_20)", %%mm0   \n\t" /* 20a */               \
        "psubw     %%mm5, %%mm0             \n\t" /* 20a - d */           \
        "paddw        %5, %%mm6             \n\t"                         \
        "paddw     %%mm6, %%mm0             \n\t" /* 20a - 6b + 3c - d */ \
        "psraw        $5, %%mm0             \n\t"                         \
        /* mm1 = EFGH, mm2 = DEFG, mm3 = CDEF, mm4 = BCDE, mm7 = 0 */     \
                                                                          \
        "movd      5(%0), %%mm5             \n\t" /* FGHI */              \
        "punpcklbw %%mm7, %%mm5             \n\t" /* 0F0G0H0I */          \
        "pshufw    $0xF9, %%mm5, %%mm6      \n\t" /* 0G0H0I0I */          \
        "paddw     %%mm5, %%mm1             \n\t" /* a */                 \
        "paddw     %%mm6, %%mm2             \n\t" /* b */                 \
        "pshufw    $0xBE, %%mm5, %%mm6      \n\t" /* 0H0I0I0H */          \
        "pshufw    $0x6F, %%mm5, %%mm5      \n\t" /* 0I0I0H0G */          \
        "paddw     %%mm6, %%mm3             \n\t" /* c */                 \
        "paddw     %%mm5, %%mm4             \n\t" /* d */                 \
        "paddw     %%mm2, %%mm2             \n\t" /* 2b */                \
        "psubw     %%mm2, %%mm3             \n\t" /* c - 2b */            \
        "pmullw "MANGLE(ff_pw_20)", %%mm1   \n\t" /* 20a */               \
        "pmullw  "MANGLE(ff_pw_3)", %%mm3   \n\t" /* 3c - 6b */           \
        "psubw     %%mm4, %%mm3             \n\t" /* -6b + 3c - d */      \
        "paddw        %5, %%mm1             \n\t"                         \
        "paddw     %%mm1, %%mm3             \n\t" /* 20a - 6b + 3c - d */ \
        "psraw        $5, %%mm3             \n\t"                         \
        "packuswb  %%mm3, %%mm0             \n\t"                         \
        OP_MMX2(%%mm0, (%1), %%mm4, q)                                    \
                                                                          \
        "add          %3, %0                \n\t"                         \
        "add          %4, %1                \n\t"                         \
        "decl         %2                    \n\t"                         \
        "jnz          1b                    \n\t"                         \
        : "+a"(src), "+c"(dst), "+d"(h)                                   \
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride),               \
          /* "m"(ff_pw_20), "m"(ff_pw_3), */ "m"(ROUNDER)                 \
        : "memory"                                                        \
        );                                                                \
}                                                                         \
                                                                          \
static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst,           \
                                                  uint8_t *src,           \
                                                  int dstStride,          \
                                                  int srcStride,          \
                                                  int h)                  \
{                                                                         \
    int i;                                                                \
    int16_t temp[8];                                                      \
    /* quick HACK, XXX FIXME MUST be optimized */                         \
    for (i = 0; i < h; i++) {                                             \
        temp[0] = (src[0] + src[1]) * 20 - (src[0] + src[2]) * 6 +        \
                  (src[1] + src[3]) *  3 - (src[2] + src[4]);             \
        temp[1] = (src[1] + src[2]) * 20 - (src[0] + src[3]) * 6 +        \
                  (src[0] + src[4]) *  3 - (src[1] + src[5]);             \
        temp[2] = (src[2] + src[3]) * 20 - (src[1] + src[4]) * 6 +        \
                  (src[0] + src[5]) *  3 - (src[0] + src[6]);             \
        temp[3] = (src[3] + src[4]) * 20 - (src[2] + src[5]) * 6 +        \
                  (src[1] + src[6]) *  3 - (src[0] + src[7]);             \
        temp[4] = (src[4] + src[5]) * 20 - (src[3] + src[6]) * 6 +        \
                  (src[2] + src[7]) *  3 - (src[1] + src[8]);             \
        temp[5] = (src[5] + src[6]) * 20 - (src[4] + src[7]) * 6 +        \
                  (src[3] + src[8]) *  3 - (src[2] + src[8]);             \
        temp[6] = (src[6] + src[7]) * 20 - (src[5] + src[8]) * 6 +        \
                  (src[4] + src[8]) *  3 - (src[3] + src[7]);             \
        temp[7] = (src[7] + src[8]) * 20 - (src[6] + src[8]) * 6 +        \
                  (src[5] + src[7]) *  3 - (src[4] + src[6]);             \
        __asm__ volatile (                                                \
            "movq      (%0), %%mm0      \n\t"                             \
            "movq     8(%0), %%mm1      \n\t"                             \
            "paddw       %2, %%mm0      \n\t"                             \
            "paddw       %2, %%mm1      \n\t"                             \
            "psraw       $5, %%mm0      \n\t"                             \
            "psraw       $5, %%mm1      \n\t"                             \
            "packuswb %%mm1, %%mm0      \n\t"                             \
            OP_3DNOW(%%mm0, (%1), %%mm1, q)                               \
            :: "r"(temp), "r"(dst), "m"(ROUNDER)                          \
            : "memory"                                                    \
            );                                                            \
        dst += dstStride;                                                 \
        src += srcStride;                                                 \
    }                                                                     \
1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263
}

#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)                          \
static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst,      \
                                                     uint8_t *src,      \
                                                     int dstStride,     \
                                                     int srcStride)     \
{                                                                       \
    uint64_t temp[17 * 4];                                              \
    uint64_t *temp_ptr = temp;                                          \
    int count = 17;                                                     \
                                                                        \
1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287
    /* FIXME unroll */                                                  \
    __asm__ volatile (                                                  \
        "pxor      %%mm7, %%mm7             \n\t"                       \
        "1:                                 \n\t"                       \
        "movq       (%0), %%mm0             \n\t"                       \
        "movq       (%0), %%mm1             \n\t"                       \
        "movq      8(%0), %%mm2             \n\t"                       \
        "movq      8(%0), %%mm3             \n\t"                       \
        "punpcklbw %%mm7, %%mm0             \n\t"                       \
        "punpckhbw %%mm7, %%mm1             \n\t"                       \
        "punpcklbw %%mm7, %%mm2             \n\t"                       \
        "punpckhbw %%mm7, %%mm3             \n\t"                       \
        "movq      %%mm0, (%1)              \n\t"                       \
        "movq      %%mm1, 17 * 8(%1)        \n\t"                       \
        "movq      %%mm2, 2 * 17 * 8(%1)    \n\t"                       \
        "movq      %%mm3, 3 * 17 * 8(%1)    \n\t"                       \
        "add          $8, %1                \n\t"                       \
        "add          %3, %0                \n\t"                       \
        "decl         %2                    \n\t"                       \
        "jnz          1b                    \n\t"                       \
        : "+r"(src), "+r"(temp_ptr), "+r"(count)                        \
        : "r"((x86_reg)srcStride)                                       \
        : "memory"                                                      \
        );                                                              \
1288 1289 1290 1291
                                                                        \
    temp_ptr = temp;                                                    \
    count    = 4;                                                       \
                                                                        \
1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336
    /* FIXME reorder for speed */                                       \
    __asm__ volatile (                                                  \
        /* "pxor  %%mm7, %%mm7            \n\t" */                      \
        "1:                             \n\t"                           \
        "movq    (%0), %%mm0            \n\t"                           \
        "movq   8(%0), %%mm1            \n\t"                           \
        "movq  16(%0), %%mm2            \n\t"                           \
        "movq  24(%0), %%mm3            \n\t"                           \
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),   8(%0),    (%0),  32(%0), (%1),     OP) \
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),    (%0),    (%0),  40(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),    (%0),   8(%0),  48(%0), (%1),     OP) \
                                                                        \
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),   8(%0),  16(%0),  56(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0),  16(%0),  24(%0),  64(%0), (%1),     OP) \
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0),  24(%0),  32(%0),  72(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0),  32(%0),  40(%0),  80(%0), (%1),     OP) \
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0),  40(%0),  48(%0),  88(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0),  48(%0),  56(%0),  96(%0), (%1),     OP) \
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0),  56(%0),  64(%0), 104(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0),  64(%0),  72(%0), 112(%0), (%1),     OP) \
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0),  72(%0),  80(%0), 120(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0),  80(%0),  88(%0), 128(%0), (%1),     OP) \
                                                                        \
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0),  88(%0),  96(%0), 128(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0),  96(%0), 104(%0), 120(%0), (%1),     OP) \
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0), 104(%0), 112(%0), 112(%0), (%1, %3), OP) \
                                                                        \
        "add     $136, %0               \n\t"                           \
        "add       %6, %1               \n\t"                           \
        "decl      %2                   \n\t"                           \
        "jnz       1b                   \n\t"                           \
                                                                        \
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)                        \
        : "r"((x86_reg)dstStride), "r"(2 * (x86_reg)dstStride),         \
          /* "m"(ff_pw_20), "m"(ff_pw_3), */ "m"(ROUNDER),              \
          "g"(4 - 14 * (x86_reg)dstStride)                              \
        : "memory"                                                      \
        );                                                              \
1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347
}                                                                       \
                                                                        \
static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst,       \
                                                    uint8_t *src,       \
                                                    int dstStride,      \
                                                    int srcStride)      \
{                                                                       \
    uint64_t temp[9 * 2];                                               \
    uint64_t *temp_ptr = temp;                                          \
    int count = 9;                                                      \
                                                                        \
1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365
    /* FIXME unroll */                                                  \
    __asm__ volatile (                                                  \
        "pxor      %%mm7, %%mm7         \n\t"                           \
        "1:                             \n\t"                           \
        "movq       (%0), %%mm0         \n\t"                           \
        "movq       (%0), %%mm1         \n\t"                           \
        "punpcklbw %%mm7, %%mm0         \n\t"                           \
        "punpckhbw %%mm7, %%mm1         \n\t"                           \
        "movq      %%mm0, (%1)          \n\t"                           \
        "movq      %%mm1, 9*8(%1)       \n\t"                           \
        "add          $8, %1            \n\t"                           \
        "add          %3, %0            \n\t"                           \
        "decl         %2                \n\t"                           \
        "jnz          1b                \n\t"                           \
        : "+r"(src), "+r"(temp_ptr), "+r"(count)                        \
        : "r"((x86_reg)srcStride)                                       \
        : "memory"                                                      \
        );                                                              \
1366 1367 1368 1369
                                                                        \
    temp_ptr = temp;                                                    \
    count    = 2;                                                       \
                                                                        \
1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402
    /* FIXME reorder for speed */                                       \
    __asm__ volatile (                                                  \
        /* "pxor  %%mm7, %%mm7            \n\t" */                      \
        "1:                             \n\t"                           \
        "movq    (%0), %%mm0            \n\t"                           \
        "movq   8(%0), %%mm1            \n\t"                           \
        "movq  16(%0), %%mm2            \n\t"                           \
        "movq  24(%0), %%mm3            \n\t"                           \
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)     \
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)     \
                                                                        \
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)     \
                                                                        \
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP) \
        "add       %4, %1               \n\t"                           \
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)     \
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP) \
                                                                        \
        "add      $72, %0               \n\t"                           \
        "add       %6, %1               \n\t"                           \
        "decl      %2                   \n\t"                           \
        "jnz       1b                   \n\t"                           \
                                                                        \
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)                        \
        : "r"((x86_reg)dstStride), "r"(2 * (x86_reg)dstStride),         \
          /* "m"(ff_pw_20), "m"(ff_pw_3), */ "m"(ROUNDER),              \
          "g"(4 - 6 * (x86_reg)dstStride)                               \
        : "memory"                                                      \
        );                                                              \
1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);                     \
}                                                                       \
                                                                        \
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;                              \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,           \
                                                stride, 8);             \
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);    \
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride,           \
                                            stride, 8);                 \
}                                                                       \
                                                                        \
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;                              \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,           \
                                                stride, 8);             \
    OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride,            \
                                 stride, 8);                            \
}                                                                       \
                                                                        \
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;                              \
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);  \
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);    \
}                                                                       \
                                                                        \
static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                         int stride)                    \
{                                                                       \
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);  \
}                                                                       \
                                                                        \
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;                              \
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);  \
    OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,       \
                                 stride, 8);                            \
}                                                                       \
                                                                        \
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);                          \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,          \
                                                stride, 9);             \
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);  \
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);   \
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);     \
}                                                                       \
                                                                        \
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);                          \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,          \
                                                stride, 9);             \
    put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,          \
                                     stride, 9);                        \
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);   \
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);     \
}                                                                       \
                                                                        \
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);                          \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,          \
                                                stride, 9);             \
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);  \
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);   \
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, stride, 8, 8); \
}                                                                       \
                                                                        \
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);                          \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,          \
                                                stride, 9);             \
    put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,          \
                                     stride, 9);                        \
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);   \
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, stride, 8, 8); \
}                                                                       \
                                                                        \
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);                          \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,          \
                                                stride, 9);             \
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);   \
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);     \
}                                                                       \
                                                                        \
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);                          \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,          \
                                                stride, 9);             \
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);   \
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, stride, 8, 8); \
}                                                                       \
                                                                        \
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);                           \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,          \
                                                stride, 9);             \
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);  \
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);     \
}                                                                       \
                                                                        \
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);                           \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,          \
                                                stride, 9);             \
    put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,          \
                                     stride, 9);                        \
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);     \
}                                                                       \
                                                                        \
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);                           \
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,          \
                                                stride, 9);             \
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);     \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src,  \
                                           int stride)                  \
{                                                                       \
    OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);                   \
}                                                                       \
                                                                        \
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;                              \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,         \
                                                 stride, 16);           \
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);  \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src,                  \
                                             stride, stride, 16);       \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t temp[32];                                                  \
    uint8_t * const half = (uint8_t*)temp;                              \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,         \
                                                 stride, 16);           \
    OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half,                   \
                                  stride, stride, 16);                  \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t temp[32];                                                  \
    uint8_t * const half = (uint8_t*)temp;                              \
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,         \
                                                 stride);               \
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);  \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride); \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t temp[32];                                                  \
    uint8_t * const half = (uint8_t*)temp;                              \
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,         \
                                                 stride);               \
    OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half,                \
                                  stride, stride, 16);                  \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t half[16 * 2 + 17 * 2];                                     \
    uint8_t * const halfH  = ((uint8_t*)half) + 256;                    \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,        \
                                                 stride, 17);           \
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,            \
                                      stride, 17);                      \
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,         \
                                                 16, 16);               \
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);  \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t half[16 * 2 + 17 * 2];                                     \
    uint8_t * const halfH  = ((uint8_t*)half) + 256;                    \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,        \
                                                 stride, 17);           \
    put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,        \
                                      stride, 17);                      \
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,         \
                                                 16, 16);               \
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);  \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t half[16 * 2 + 17 * 2];                                     \
    uint8_t * const halfH  = ((uint8_t*)half) + 256;                    \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,        \
                                                 stride, 17);           \
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,            \
                                      stride, 17);                      \
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,         \
                                                 16, 16);               \
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, stride,      \
                                  16, 16);                              \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t half[16 * 2 + 17 * 2];                                     \
    uint8_t * const halfH  = ((uint8_t*)half) + 256;                    \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,        \
                                                 stride, 17);           \
    put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,        \
                                      stride, 17);                      \
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,         \
                                                 16, 16);               \
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, stride,      \
                                  16, 16);                              \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t half[16 * 2 + 17 * 2];                                     \
    uint8_t * const halfH  = ((uint8_t*)half) + 256;                    \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,        \
                                                 stride, 17);           \
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,         \
                                                 16, 16);               \
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);  \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t half[16 * 2 + 17 * 2];                                     \
    uint8_t * const halfH  = ((uint8_t*)half) + 256;                    \
    uint8_t * const halfHV = ((uint8_t*)half);                          \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,        \
                                                 stride, 17);           \
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,         \
                                                 16, 16);               \
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, stride,      \
                                  16, 16);                              \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t half[17 * 2];                                              \
    uint8_t * const halfH = ((uint8_t*)half);                           \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,        \
                                                 stride, 17);           \
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,            \
                                      stride, 17);                      \
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);   \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t half[17 * 2];                                              \
    uint8_t * const halfH = ((uint8_t*)half);                           \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,        \
                                                 stride, 17);           \
    put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,        \
                                      stride, 17);                      \
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);   \
}                                                                       \
                                                                        \
static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src,   \
                                          int stride)                   \
{                                                                       \
    uint64_t half[17 * 2];                                              \
    uint8_t * const halfH = ((uint8_t*)half);                           \
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,        \
                                                 stride, 17);           \
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);   \
Michael Niedermayer's avatar
Michael Niedermayer committed
1757 1758
}

1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780
#define PUT_OP(a, b, temp, size)                \
    "mov"#size"        "#a", "#b"       \n\t"

#define AVG_3DNOW_OP(a, b, temp, size)          \
    "mov"#size"        "#b", "#temp"    \n\t"   \
    "pavgusb        "#temp", "#a"       \n\t"   \
    "mov"#size"        "#a", "#b"       \n\t"

#define AVG_MMX2_OP(a, b, temp, size)           \
    "mov"#size"        "#b", "#temp"    \n\t"   \
    "pavgb          "#temp", "#a"       \n\t"   \
    "mov"#size"        "#a", "#b"       \n\t"

QPEL_BASE(put_,        ff_pw_16, _,        PUT_OP,       PUT_OP)
QPEL_BASE(avg_,        ff_pw_16, _,        AVG_MMX2_OP,  AVG_3DNOW_OP)
QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP,       PUT_OP)
QPEL_OP(put_,          ff_pw_16, _,        PUT_OP,       3dnow)
QPEL_OP(avg_,          ff_pw_16, _,        AVG_3DNOW_OP, 3dnow)
QPEL_OP(put_no_rnd_,   ff_pw_15, _no_rnd_, PUT_OP,       3dnow)
QPEL_OP(put_,          ff_pw_16, _,        PUT_OP,       mmx2)
QPEL_OP(avg_,          ff_pw_16, _,        AVG_MMX2_OP,  mmx2)
QPEL_OP(put_no_rnd_,   ff_pw_15, _no_rnd_, PUT_OP,       mmx2)
Michael Niedermayer's avatar
Michael Niedermayer committed
1781

1782 1783 1784
/***********************************/
/* bilinear qpel: not compliant to any spec, only for -lavdopts fast */

1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832
#define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)                              \
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, \
                                                                 uint8_t *src, \
                                                                 int stride)   \
{                                                                              \
    OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);                  \
}

#define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)                        \
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, \
                                                                 uint8_t *src, \
                                                                 int stride)   \
{                                                                              \
    OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src + S0, stride, SIZE,    \
                                               S1, S2);                        \
}

#define QPEL_2TAP(OPNAME, SIZE, MMX)                                        \
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)                            \
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)                            \
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)                               \
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =    \
    OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;                                \
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =    \
    OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;                           \
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =    \
    OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;                           \
static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst,      \
                                                         uint8_t *src,      \
                                                         int stride)        \
{                                                                           \
    OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src + 1, stride, SIZE);    \
}                                                                           \
static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst,      \
                                                         uint8_t *src,      \
                                                         int stride)        \
{                                                                           \
    OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src + stride,              \
                                            stride, SIZE);                  \
}                                                                           \
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0,           1,       0)                \
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1,          -1,       0)                \
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0,           stride,  0)                \
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride,     -stride,  0)                \
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0,           stride,  1)                \
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1,           stride, -1)                \
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride,     -stride,  1)                \
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride + 1, -stride, -1)                \
1833 1834 1835 1836 1837 1838 1839 1840 1841 1842

QPEL_2TAP(put_, 16, mmx2)
QPEL_2TAP(avg_, 16, mmx2)
QPEL_2TAP(put_,  8, mmx2)
QPEL_2TAP(avg_,  8, mmx2)
QPEL_2TAP(put_, 16, 3dnow)
QPEL_2TAP(avg_, 16, 3dnow)
QPEL_2TAP(put_,  8, 3dnow)
QPEL_2TAP(avg_,  8, 3dnow)

1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858
void ff_put_rv40_qpel8_mc33_mmx(uint8_t *dst, uint8_t *src, int stride)
{
  put_pixels8_xy2_mmx(dst, src, stride, 8);
}
void ff_put_rv40_qpel16_mc33_mmx(uint8_t *dst, uint8_t *src, int stride)
{
  put_pixels16_xy2_mmx(dst, src, stride, 16);
}
void ff_avg_rv40_qpel8_mc33_mmx(uint8_t *dst, uint8_t *src, int stride)
{
  avg_pixels8_xy2_mmx(dst, src, stride, 8);
}
void ff_avg_rv40_qpel16_mc33_mmx(uint8_t *dst, uint8_t *src, int stride)
{
  avg_pixels16_xy2_mmx(dst, src, stride, 16);
}
1859

1860 1861
#endif /* HAVE_INLINE_ASM */

1862
#if HAVE_YASM
1863 1864 1865 1866 1867
typedef void emu_edge_core_func(uint8_t *buf, const uint8_t *src,
                                x86_reg linesize, x86_reg start_y,
                                x86_reg end_y, x86_reg block_h,
                                x86_reg start_x, x86_reg end_x,
                                x86_reg block_w);
1868 1869 1870
extern emu_edge_core_func ff_emu_edge_core_mmx;
extern emu_edge_core_func ff_emu_edge_core_sse;

1871 1872 1873 1874 1875 1876
static av_always_inline void emulated_edge_mc(uint8_t *buf, const uint8_t *src,
                                              int linesize,
                                              int block_w, int block_h,
                                              int src_x, int src_y,
                                              int w, int h,
                                              emu_edge_core_func *core_fn)
1877
{
1878 1879 1880 1881 1882 1883 1884 1885
    int start_y, start_x, end_y, end_x, src_y_add = 0;

    if (src_y >= h) {
        src_y_add = h - 1 - src_y;
        src_y     = h - 1;
    } else if (src_y <= -block_h) {
        src_y_add = 1 - block_h - src_y;
        src_y     = 1 - block_h;
1886
    }
1887 1888 1889 1890 1891 1892
    if (src_x >= w) {
        src   += w - 1 - src_x;
        src_x  = w - 1;
    } else if (src_x <= -block_w) {
        src   += 1 - block_w - src_x;
        src_x  = 1 - block_w;
1893 1894
    }

1895 1896 1897 1898
    start_y = FFMAX(0, -src_y);
    start_x = FFMAX(0, -src_x);
    end_y   = FFMIN(block_h, h-src_y);
    end_x   = FFMIN(block_w, w-src_x);
1899 1900
    av_assert2(start_x < end_x && block_w > 0);
    av_assert2(start_y < end_y && block_h > 0);
1901 1902

    // fill in the to-be-copied part plus all above/below
1903
    src += (src_y_add + start_y) * linesize + start_x;
1904
    buf += start_x;
1905 1906
    core_fn(buf, src, linesize, start_y, end_y,
            block_h, start_x, end_x, block_w);
1907 1908 1909
}

#if ARCH_X86_32
1910 1911 1912 1913
static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src,
                                             int linesize,
                                             int block_w, int block_h,
                                             int src_x, int src_y, int w, int h)
1914 1915 1916 1917 1918
{
    emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
                     w, h, &ff_emu_edge_core_mmx);
}
#endif
1919 1920 1921 1922 1923

static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src,
                                             int linesize,
                                             int block_w, int block_h,
                                             int src_x, int src_y, int w, int h)
1924 1925 1926 1927 1928 1929
{
    emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
                     w, h, &ff_emu_edge_core_sse);
}
#endif /* HAVE_YASM */

1930 1931
#if HAVE_INLINE_ASM

1932 1933 1934 1935 1936 1937 1938 1939 1940
typedef void emulated_edge_mc_func(uint8_t *dst, const uint8_t *src,
                                   int linesize, int block_w, int block_h,
                                   int src_x, int src_y, int w, int h);

static av_always_inline void gmc(uint8_t *dst, uint8_t *src,
                                 int stride, int h, int ox, int oy,
                                 int dxx, int dxy, int dyx, int dyy,
                                 int shift, int r, int width, int height,
                                 emulated_edge_mc_func *emu_edge_fn)
1941
{
1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954
    const int w    = 8;
    const int ix   = ox  >> (16 + shift);
    const int iy   = oy  >> (16 + shift);
    const int oxs  = ox  >> 4;
    const int oys  = oy  >> 4;
    const int dxxs = dxx >> 4;
    const int dxys = dxy >> 4;
    const int dyxs = dyx >> 4;
    const int dyys = dyy >> 4;
    const uint16_t r4[4]   = { r, r, r, r };
    const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys };
    const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys };
    const uint64_t shift2 = 2 * shift;
1955 1956 1957
#define MAX_STRIDE 4096U
#define MAX_H 8U
    uint8_t edge_buf[(MAX_H + 1) * MAX_STRIDE];
1958 1959
    int x, y;

1960 1961 1962 1963
    const int dxw = (dxx - (1 << (16 + shift))) * (w - 1);
    const int dyh = (dyy - (1 << (16 + shift))) * (h - 1);
    const int dxh = dxy * (h - 1);
    const int dyw = dyx * (w - 1);
1964 1965 1966
    int need_emu =  (unsigned)ix >= width  - w ||
                    (unsigned)iy >= height - h;

1967 1968 1969
    if ( // non-constant fullpel offset (3% of blocks)
        ((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) |
         (oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift)
1970
        // uses more than 16 bits of subpel mv (only at huge resolution)
1971
        || (dxx | dxy | dyx | dyy) & 15
1972
        || (need_emu && (h > MAX_H || stride > MAX_STRIDE))) {
1973
        // FIXME could still use mmx for some of the rows
1974 1975
        ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy,
                 shift, r, width, height);
1976 1977 1978
        return;
    }

1979
    src += ix + iy * stride;
1980
    if (need_emu) {
1981 1982 1983
        emu_edge_fn(edge_buf, src, stride, w + 1, h + 1, ix, iy, width, height);
        src = edge_buf;
    }
Loren Merritt's avatar
Loren Merritt committed
1984

1985 1986 1987 1988 1989
    __asm__ volatile (
        "movd         %0, %%mm6         \n\t"
        "pxor      %%mm7, %%mm7         \n\t"
        "punpcklwd %%mm6, %%mm6         \n\t"
        "punpcklwd %%mm6, %%mm6         \n\t"
Loren Merritt's avatar
Loren Merritt committed
1990 1991
        :: "r"(1<<shift)
    );
1992

1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
    for (x = 0; x < w; x += 4) {
        uint16_t dx4[4] = { oxs - dxys + dxxs * (x + 0),
                            oxs - dxys + dxxs * (x + 1),
                            oxs - dxys + dxxs * (x + 2),
                            oxs - dxys + dxxs * (x + 3) };
        uint16_t dy4[4] = { oys - dyys + dyxs * (x + 0),
                            oys - dyys + dyxs * (x + 1),
                            oys - dyys + dyxs * (x + 2),
                            oys - dyys + dyxs * (x + 3) };

        for (y = 0; y < h; y++) {
2004 2005 2006 2007 2008 2009 2010 2011 2012
            __asm__ volatile (
                "movq      %0, %%mm4    \n\t"
                "movq      %1, %%mm5    \n\t"
                "paddw     %2, %%mm4    \n\t"
                "paddw     %3, %%mm5    \n\t"
                "movq   %%mm4, %0       \n\t"
                "movq   %%mm5, %1       \n\t"
                "psrlw    $12, %%mm4    \n\t"
                "psrlw    $12, %%mm5    \n\t"
2013 2014 2015 2016
                : "+m"(*dx4), "+m"(*dy4)
                : "m"(*dxy4), "m"(*dyy4)
            );

2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030
            __asm__ volatile (
                "movq      %%mm6, %%mm2 \n\t"
                "movq      %%mm6, %%mm1 \n\t"
                "psubw     %%mm4, %%mm2 \n\t"
                "psubw     %%mm5, %%mm1 \n\t"
                "movq      %%mm2, %%mm0 \n\t"
                "movq      %%mm4, %%mm3 \n\t"
                "pmullw    %%mm1, %%mm0 \n\t" // (s - dx) * (s - dy)
                "pmullw    %%mm5, %%mm3 \n\t" // dx * dy
                "pmullw    %%mm5, %%mm2 \n\t" // (s - dx) * dy
                "pmullw    %%mm4, %%mm1 \n\t" // dx * (s - dy)

                "movd         %4, %%mm5 \n\t"
                "movd         %3, %%mm4 \n\t"
2031 2032
                "punpcklbw %%mm7, %%mm5 \n\t"
                "punpcklbw %%mm7, %%mm4 \n\t"
2033 2034
                "pmullw    %%mm5, %%mm3 \n\t" // src[1, 1] * dx * dy
                "pmullw    %%mm4, %%mm2 \n\t" // src[0, 1] * (s - dx) * dy
2035

2036 2037
                "movd         %2, %%mm5 \n\t"
                "movd         %1, %%mm4 \n\t"
2038 2039
                "punpcklbw %%mm7, %%mm5 \n\t"
                "punpcklbw %%mm7, %%mm4 \n\t"
2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051
                "pmullw    %%mm5, %%mm1 \n\t" // src[1, 0] * dx * (s - dy)
                "pmullw    %%mm4, %%mm0 \n\t" // src[0, 0] * (s - dx) * (s - dy)
                "paddw        %5, %%mm1 \n\t"
                "paddw     %%mm3, %%mm2 \n\t"
                "paddw     %%mm1, %%mm0 \n\t"
                "paddw     %%mm2, %%mm0 \n\t"

                "psrlw        %6, %%mm0 \n\t"
                "packuswb  %%mm0, %%mm0 \n\t"
                "movd      %%mm0, %0    \n\t"

                : "=m"(dst[x + y * stride])
2052
                : "m"(src[0]), "m"(src[1]),
2053
                  "m"(src[stride]), "m"(src[stride + 1]),
2054 2055 2056 2057
                  "m"(*r4), "m"(shift2)
            );
            src += stride;
        }
2058
        src += 4 - h * stride;
2059 2060 2061
    }
}

2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091
#if HAVE_YASM
#if ARCH_X86_32
static void gmc_mmx(uint8_t *dst, uint8_t *src,
                    int stride, int h, int ox, int oy,
                    int dxx, int dxy, int dyx, int dyy,
                    int shift, int r, int width, int height)
{
    gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
        width, height, &emulated_edge_mc_mmx);
}
#endif
static void gmc_sse(uint8_t *dst, uint8_t *src,
                    int stride, int h, int ox, int oy,
                    int dxx, int dxy, int dyx, int dyy,
                    int shift, int r, int width, int height)
{
    gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
        width, height, &emulated_edge_mc_sse);
}
#else
static void gmc_mmx(uint8_t *dst, uint8_t *src,
                    int stride, int h, int ox, int oy,
                    int dxx, int dxy, int dyx, int dyy,
                    int shift, int r, int width, int height)
{
    gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
        width, height, &ff_emulated_edge_mc_8);
}
#endif

2092 2093 2094 2095 2096
#define PREFETCH(name, op)                      \
static void name(void *mem, int stride, int h)  \
{                                               \
    const uint8_t *p = mem;                     \
    do {                                        \
2097
        __asm__ volatile (#op" %0" :: "m"(*p)); \
2098 2099
        p += stride;                            \
    } while (--h);                              \
2100
}
2101

2102 2103 2104 2105
PREFETCH(prefetch_mmx2,  prefetcht0)
PREFETCH(prefetch_3dnow, prefetch)
#undef PREFETCH

2106 2107
#endif /* HAVE_INLINE_ASM */

2108
#include "h264_qpel.c"
2109

2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141
void ff_put_h264_chroma_mc8_mmx_rnd  (uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);
void ff_avg_h264_chroma_mc8_mmx2_rnd (uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);
void ff_avg_h264_chroma_mc8_3dnow_rnd(uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);

void ff_put_h264_chroma_mc4_mmx      (uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);
void ff_avg_h264_chroma_mc4_mmx2     (uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);
void ff_avg_h264_chroma_mc4_3dnow    (uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);

void ff_put_h264_chroma_mc2_mmx2     (uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);
void ff_avg_h264_chroma_mc2_mmx2     (uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);

void ff_put_h264_chroma_mc8_ssse3_rnd(uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);
void ff_put_h264_chroma_mc4_ssse3    (uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);

void ff_avg_h264_chroma_mc8_ssse3_rnd(uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);
void ff_avg_h264_chroma_mc4_ssse3    (uint8_t *dst, uint8_t *src,
                                      int stride, int h, int x, int y);

#define CHROMA_MC(OP, NUM, DEPTH, OPT)                                  \
void ff_ ## OP ## _h264_chroma_mc ## NUM ## _ ## DEPTH ## _ ## OPT      \
                                      (uint8_t *dst, uint8_t *src,      \
2142 2143
                                       int stride, int h, int x, int y);

2144 2145 2146 2147
CHROMA_MC(put, 2, 10, mmx2)
CHROMA_MC(avg, 2, 10, mmx2)
CHROMA_MC(put, 4, 10, mmx2)
CHROMA_MC(avg, 4, 10, mmx2)
2148 2149 2150 2151
CHROMA_MC(put, 8, 10, sse2)
CHROMA_MC(avg, 8, 10, sse2)
CHROMA_MC(put, 8, 10, avx)
CHROMA_MC(avg, 8, 10, avx)
2152

2153 2154
#if HAVE_INLINE_ASM

2155 2156 2157
/* CAVS-specific */
void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride)
{
2158 2159
    put_pixels8_mmx(dst, src, stride, 8);
}
2160 2161 2162

void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride)
{
2163 2164
    avg_pixels8_mmx(dst, src, stride, 8);
}
2165 2166 2167

void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride)
{
2168 2169
    put_pixels16_mmx(dst, src, stride, 16);
}
2170 2171 2172

void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride)
{
2173 2174 2175
    avg_pixels16_mmx(dst, src, stride, 16);
}

2176 2177 2178 2179
/* VC-1-specific */
void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src,
                               int stride, int rnd)
{
2180 2181
    put_pixels8_mmx(dst, src, stride, 8);
}
2182 2183 2184 2185

void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src,
                                int stride, int rnd)
{
2186 2187
    avg_pixels8_mmx2(dst, src, stride, 8);
}
2188

David Conrad's avatar
David Conrad committed
2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226
/* only used in VP3/5/6 */
static void put_vp_no_rnd_pixels8_l2_mmx(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h)
{
//    START_TIMER
    MOVQ_BFE(mm6);
    __asm__ volatile(
        "1:                             \n\t"
        "movq   (%1), %%mm0             \n\t"
        "movq   (%2), %%mm1             \n\t"
        "movq   (%1,%4), %%mm2          \n\t"
        "movq   (%2,%4), %%mm3          \n\t"
        PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
        "movq   %%mm4, (%3)             \n\t"
        "movq   %%mm5, (%3,%4)          \n\t"

        "movq   (%1,%4,2), %%mm0        \n\t"
        "movq   (%2,%4,2), %%mm1        \n\t"
        "movq   (%1,%5), %%mm2          \n\t"
        "movq   (%2,%5), %%mm3          \n\t"
        "lea    (%1,%4,4), %1           \n\t"
        "lea    (%2,%4,4), %2           \n\t"
        PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
        "movq   %%mm4, (%3,%4,2)        \n\t"
        "movq   %%mm5, (%3,%5)          \n\t"
        "lea    (%3,%4,4), %3           \n\t"
        "subl   $4, %0                  \n\t"
        "jnz    1b                      \n\t"
        :"+r"(h), "+r"(a), "+r"(b), "+r"(dst)
        :"r"((x86_reg)stride), "r"((x86_reg)3L*stride)
        :"memory");
//    STOP_TIMER("put_vp_no_rnd_pixels8_l2_mmx")
}
static void put_vp_no_rnd_pixels16_l2_mmx(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h)
{
    put_vp_no_rnd_pixels8_l2_mmx(dst, a, b, stride, h);
    put_vp_no_rnd_pixels8_l2_mmx(dst+8, a+8, b+8, stride, h);
}

2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266
#if CONFIG_DIRAC_DECODER
#define DIRAC_PIXOP(OPNAME, EXT)\
void ff_ ## OPNAME ## _dirac_pixels8_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
{\
    OPNAME ## _pixels8_ ## EXT(dst, src[0], stride, h);\
}\
void ff_ ## OPNAME ## _dirac_pixels16_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
{\
    OPNAME ## _pixels16_ ## EXT(dst, src[0], stride, h);\
}\
void ff_ ## OPNAME ## _dirac_pixels32_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
{\
    OPNAME ## _pixels16_ ## EXT(dst   , src[0]   , stride, h);\
    OPNAME ## _pixels16_ ## EXT(dst+16, src[0]+16, stride, h);\
}

DIRAC_PIXOP(put, mmx)
DIRAC_PIXOP(avg, mmx)
DIRAC_PIXOP(avg, mmx2)

void ff_put_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
{
    put_pixels16_sse2(dst, src[0], stride, h);
}
void ff_avg_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
{
    avg_pixels16_sse2(dst, src[0], stride, h);
}
void ff_put_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
{
    put_pixels16_sse2(dst   , src[0]   , stride, h);
    put_pixels16_sse2(dst+16, src[0]+16, stride, h);
}
void ff_avg_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
{
    avg_pixels16_sse2(dst   , src[0]   , stride, h);
    avg_pixels16_sse2(dst+16, src[0]+16, stride, h);
}
#endif

2267 2268
/* XXX: Those functions should be suppressed ASAP when all IDCTs are
 * converted. */
2269
#if CONFIG_GPL
2270 2271
static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size,
                                    DCTELEM *block)
2272
{
2273
    ff_mmx_idct(block);
2274
    ff_put_pixels_clamped_mmx(block, dest, line_size);
2275
}
2276 2277 2278

static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size,
                                    DCTELEM *block)
2279
{
2280
    ff_mmx_idct(block);
2281
    ff_add_pixels_clamped_mmx(block, dest, line_size);
2282
}
2283 2284 2285

static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size,
                                     DCTELEM *block)
2286
{
2287
    ff_mmxext_idct(block);
2288
    ff_put_pixels_clamped_mmx(block, dest, line_size);
2289
}
2290 2291 2292

static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size,
                                     DCTELEM *block)
2293
{
2294
    ff_mmxext_idct(block);
2295
    ff_add_pixels_clamped_mmx(block, dest, line_size);
2296
}
2297
#endif
2298

2299
static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)
2300 2301
{
    int i;
2302
    __asm__ volatile ("pxor %%mm7, %%mm7":);
2303
    for (i = 0; i < blocksize; i += 2) {
2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321
        __asm__ volatile (
            "movq       %0, %%mm0   \n\t"
            "movq       %1, %%mm1   \n\t"
            "movq    %%mm0, %%mm2   \n\t"
            "movq    %%mm1, %%mm3   \n\t"
            "pfcmpge %%mm7, %%mm2   \n\t" // m <= 0.0
            "pfcmpge %%mm7, %%mm3   \n\t" // a <= 0.0
            "pslld     $31, %%mm2   \n\t" // keep only the sign bit
            "pxor    %%mm2, %%mm1   \n\t"
            "movq    %%mm3, %%mm4   \n\t"
            "pand    %%mm1, %%mm3   \n\t"
            "pandn   %%mm1, %%mm4   \n\t"
            "pfadd   %%mm0, %%mm3   \n\t" // a = m + ((a < 0) & (a ^ sign(m)))
            "pfsub   %%mm4, %%mm0   \n\t" // m = m + ((a > 0) & (a ^ sign(m)))
            "movq    %%mm3, %1      \n\t"
            "movq    %%mm0, %0      \n\t"
            : "+m"(mag[i]), "+m"(ang[i])
            :: "memory"
2322 2323
        );
    }
2324
    __asm__ volatile ("femms");
2325
}
2326

2327
static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
2328 2329
{
    int i;
2330

2331 2332 2333
    __asm__ volatile (
        "movaps  %0, %%xmm5 \n\t"
        :: "m"(ff_pdw_80000000[0])
2334
    );
2335
    for (i = 0; i < blocksize; i += 4) {
2336 2337 2338
        __asm__ volatile (
            "movaps      %0, %%xmm0 \n\t"
            "movaps      %1, %%xmm1 \n\t"
2339 2340
            "xorps   %%xmm2, %%xmm2 \n\t"
            "xorps   %%xmm3, %%xmm3 \n\t"
2341 2342
            "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0
            "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0
2343 2344
            "andps   %%xmm5, %%xmm2 \n\t" // keep only the sign bit
            "xorps   %%xmm2, %%xmm1 \n\t"
2345
            "movaps  %%xmm3, %%xmm4 \n\t"
2346 2347
            "andps   %%xmm1, %%xmm3 \n\t"
            "andnps  %%xmm1, %%xmm4 \n\t"
2348 2349
            "addps   %%xmm0, %%xmm3 \n\t" // a = m + ((a < 0) & (a ^ sign(m)))
            "subps   %%xmm4, %%xmm0 \n\t" // m = m + ((a > 0) & (a ^ sign(m)))
2350 2351
            "movaps  %%xmm3, %1     \n\t"
            "movaps  %%xmm0, %0     \n\t"
2352 2353
            : "+m"(mag[i]), "+m"(ang[i])
            :: "memory"
2354 2355 2356 2357
        );
    }
}

2358
#if HAVE_6REGS
2359 2360 2361
static void vector_fmul_window_3dnowext(float *dst, const float *src0,
                                        const float *src1, const float *win,
                                        int len)
2362 2363 2364
{
    x86_reg i = -len * 4;
    x86_reg j =  len * 4 - 8;
2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387
    __asm__ volatile (
        "1:                             \n"
        "pswapd (%5, %1), %%mm1         \n"
        "movq   (%5, %0), %%mm0         \n"
        "pswapd (%4, %1), %%mm5         \n"
        "movq   (%3, %0), %%mm4         \n"
        "movq      %%mm0, %%mm2         \n"
        "movq      %%mm1, %%mm3         \n"
        "pfmul     %%mm4, %%mm2         \n" // src0[len + i] * win[len + i]
        "pfmul     %%mm5, %%mm3         \n" // src1[j]       * win[len + j]
        "pfmul     %%mm4, %%mm1         \n" // src0[len + i] * win[len + j]
        "pfmul     %%mm5, %%mm0         \n" // src1[j]       * win[len + i]
        "pfadd     %%mm3, %%mm2         \n"
        "pfsub     %%mm0, %%mm1         \n"
        "pswapd    %%mm2, %%mm2         \n"
        "movq      %%mm1, (%2, %0)      \n"
        "movq      %%mm2, (%2, %1)      \n"
        "sub          $8, %1            \n"
        "add          $8, %0            \n"
        "jl           1b                \n"
        "femms                          \n"
        : "+r"(i), "+r"(j)
        : "r"(dst + len), "r"(src0 + len), "r"(src1), "r"(win + len)
Justin Ruggles's avatar
Justin Ruggles committed
2388
    );
Loren Merritt's avatar
Loren Merritt committed
2389 2390
}

2391 2392 2393 2394 2395
static void vector_fmul_window_sse(float *dst, const float *src0,
                                   const float *src1, const float *win, int len)
{
    x86_reg i = -len * 4;
    x86_reg j =  len * 4 - 16;
2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419
    __asm__ volatile (
        "1:                             \n"
        "movaps      (%5, %1), %%xmm1   \n"
        "movaps      (%5, %0), %%xmm0   \n"
        "movaps      (%4, %1), %%xmm5   \n"
        "movaps      (%3, %0), %%xmm4   \n"
        "shufps $0x1b, %%xmm1, %%xmm1   \n"
        "shufps $0x1b, %%xmm5, %%xmm5   \n"
        "movaps        %%xmm0, %%xmm2   \n"
        "movaps        %%xmm1, %%xmm3   \n"
        "mulps         %%xmm4, %%xmm2   \n" // src0[len + i] * win[len + i]
        "mulps         %%xmm5, %%xmm3   \n" // src1[j]       * win[len + j]
        "mulps         %%xmm4, %%xmm1   \n" // src0[len + i] * win[len + j]
        "mulps         %%xmm5, %%xmm0   \n" // src1[j]       * win[len + i]
        "addps         %%xmm3, %%xmm2   \n"
        "subps         %%xmm0, %%xmm1   \n"
        "shufps $0x1b, %%xmm2, %%xmm2   \n"
        "movaps        %%xmm1, (%2, %0) \n"
        "movaps        %%xmm2, (%2, %1) \n"
        "sub              $16, %1       \n"
        "add              $16, %0       \n"
        "jl                1b           \n"
        : "+r"(i), "+r"(j)
        : "r"(dst + len), "r"(src0 + len), "r"(src1), "r"(win + len)
Justin Ruggles's avatar
Justin Ruggles committed
2420
    );
2421
}
2422
#endif /* HAVE_6REGS */
2423

2424 2425
static void vector_clipf_sse(float *dst, const float *src,
                             float min, float max, int len)
2426
{
2427
    x86_reg i = (len - 16) * 4;
2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454
    __asm__ volatile (
        "movss          %3, %%xmm4      \n\t"
        "movss          %4, %%xmm5      \n\t"
        "shufps $0, %%xmm4, %%xmm4      \n\t"
        "shufps $0, %%xmm5, %%xmm5      \n\t"
        "1:                             \n\t"
        "movaps   (%2, %0), %%xmm0      \n\t" // 3/1 on intel
        "movaps 16(%2, %0), %%xmm1      \n\t"
        "movaps 32(%2, %0), %%xmm2      \n\t"
        "movaps 48(%2, %0), %%xmm3      \n\t"
        "maxps      %%xmm4, %%xmm0      \n\t"
        "maxps      %%xmm4, %%xmm1      \n\t"
        "maxps      %%xmm4, %%xmm2      \n\t"
        "maxps      %%xmm4, %%xmm3      \n\t"
        "minps      %%xmm5, %%xmm0      \n\t"
        "minps      %%xmm5, %%xmm1      \n\t"
        "minps      %%xmm5, %%xmm2      \n\t"
        "minps      %%xmm5, %%xmm3      \n\t"
        "movaps     %%xmm0,   (%1, %0)  \n\t"
        "movaps     %%xmm1, 16(%1, %0)  \n\t"
        "movaps     %%xmm2, 32(%1, %0)  \n\t"
        "movaps     %%xmm3, 48(%1, %0)  \n\t"
        "sub           $64, %0          \n\t"
        "jge            1b              \n\t"
        : "+&r"(i)
        : "r"(dst), "r"(src), "m"(min), "m"(max)
        : "memory"
2455 2456 2457
    );
}

2458 2459
#endif /* HAVE_INLINE_ASM */

2460
int32_t ff_scalarproduct_int16_mmx2(const int16_t *v1, const int16_t *v2,
2461
                                    int order);
2462
int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2,
2463
                                    int order);
2464 2465 2466 2467 2468 2469 2470 2471 2472
int32_t ff_scalarproduct_and_madd_int16_mmx2(int16_t *v1, const int16_t *v2,
                                             const int16_t *v3,
                                             int order, int mul);
int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2,
                                             const int16_t *v3,
                                             int order, int mul);
int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2,
                                              const int16_t *v3,
                                              int order, int mul);
2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486

void ff_apply_window_int16_mmxext    (int16_t *output, const int16_t *input,
                                      const int16_t *window, unsigned int len);
void ff_apply_window_int16_mmxext_ba (int16_t *output, const int16_t *input,
                                      const int16_t *window, unsigned int len);
void ff_apply_window_int16_sse2      (int16_t *output, const int16_t *input,
                                      const int16_t *window, unsigned int len);
void ff_apply_window_int16_sse2_ba   (int16_t *output, const int16_t *input,
                                      const int16_t *window, unsigned int len);
void ff_apply_window_int16_ssse3     (int16_t *output, const int16_t *input,
                                      const int16_t *window, unsigned int len);
void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
                                      const int16_t *window, unsigned int len);

2487 2488 2489
void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w);
void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w);

2490 2491 2492 2493 2494 2495 2496
void ff_add_hfyu_median_prediction_mmx2(uint8_t *dst, const uint8_t *top,
                                        const uint8_t *diff, int w,
                                        int *left, int *left_top);
int  ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src,
                                       int w, int left);
int  ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src,
                                      int w, int left);
2497

2498 2499
float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);

2500 2501 2502 2503 2504 2505 2506 2507 2508 2509
void ff_vector_fmul_reverse_sse(float *dst, const float *src0,
                                const float *src1, int len);
void ff_vector_fmul_reverse_avx(float *dst, const float *src0,
                                const float *src1, int len);

void ff_vector_fmul_add_sse(float *dst, const float *src0, const float *src1,
                            const float *src2, int len);
void ff_vector_fmul_add_avx(float *dst, const float *src0, const float *src1,
                            const float *src2, int len);

2510 2511 2512 2513 2514 2515 2516 2517
void ff_vector_clip_int32_mmx     (int32_t *dst, const int32_t *src,
                                   int32_t min, int32_t max, unsigned int len);
void ff_vector_clip_int32_sse2    (int32_t *dst, const int32_t *src,
                                   int32_t min, int32_t max, unsigned int len);
void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src,
                                   int32_t min, int32_t max, unsigned int len);
void ff_vector_clip_int32_sse4    (int32_t *dst, const int32_t *src,
                                   int32_t min, int32_t max, unsigned int len);
2518

2519 2520 2521 2522 2523
extern void ff_butterflies_float_interleave_sse(float *dst, const float *src0,
                                                const float *src1, int len);
extern void ff_butterflies_float_interleave_avx(float *dst, const float *src0,
                                                const float *src1, int len);

2524
#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
2525
    do {                                                                     \
2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540
    c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
    c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
2541 2542
    c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
    } while (0)
2543

2544 2545 2546 2547 2548 2549
#define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU)                                     \
    do {                                                                        \
        c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _     ## CPU; \
        c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_  ## CPU; \
        c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_  ## CPU; \
        c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \
2550
    } while (0)
2551

2552 2553 2554 2555 2556 2557
#define H264_QPEL_FUNCS(x, y, CPU)                                                            \
    do {                                                                                      \
        c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \
        c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc  ## x ## y ## _ ## CPU; \
        c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \
        c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc  ## x ## y ## _ ## CPU; \
2558
    } while (0)
2559

2560 2561 2562 2563 2564 2565
#define H264_QPEL_FUNCS_10(x, y, CPU)                                                               \
    do {                                                                                            \
        c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
        c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc  ## x ## y ## _10_ ## CPU; \
        c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
        c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc  ## x ## y ## _10_ ## CPU; \
2566
    } while (0)
2567 2568 2569 2570 2571

static void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx, int mm_flags)
{
    const int high_bit_depth = avctx->bits_per_raw_sample > 8;

2572
#if HAVE_INLINE_ASM
2573 2574 2575 2576 2577 2578 2579 2580 2581
    c->put_pixels_clamped        = ff_put_pixels_clamped_mmx;
    c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
    c->add_pixels_clamped        = ff_add_pixels_clamped_mmx;

    if (!high_bit_depth) {
        c->clear_block  = clear_block_mmx;
        c->clear_blocks = clear_blocks_mmx;
        c->draw_edges   = draw_edges_mmx;

2582
        SET_HPEL_FUNCS(put,        0, 16, mmx);
2583
        SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx);
2584
        SET_HPEL_FUNCS(avg,        0, 16, mmx);
2585
        SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx);
2586 2587 2588 2589
        SET_HPEL_FUNCS(put,        1,  8, mmx);
        SET_HPEL_FUNCS(put_no_rnd, 1,  8, mmx);
        SET_HPEL_FUNCS(avg,        1,  8, mmx);
        SET_HPEL_FUNCS(avg_no_rnd, 1,  8, mmx);
2590 2591
    }

2592
#if ARCH_X86_32 || !HAVE_YASM
2593
    c->gmc = gmc_mmx;
2594
#endif
2595 2596 2597

    c->add_bytes = add_bytes_mmx;

2598 2599 2600
    c->put_no_rnd_pixels_l2[0]= put_vp_no_rnd_pixels16_l2_mmx;
    c->put_no_rnd_pixels_l2[1]= put_vp_no_rnd_pixels8_l2_mmx;

2601 2602 2603 2604
    if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
        c->h263_v_loop_filter = h263_v_loop_filter_mmx;
        c->h263_h_loop_filter = h263_h_loop_filter_mmx;
    }
2605
#endif /* HAVE_INLINE_ASM */
2606 2607

#if HAVE_YASM
2608 2609 2610 2611 2612
#if ARCH_X86_32
    if (!high_bit_depth)
        c->emulated_edge_mc = emulated_edge_mc_mmx;
#endif

2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628
    if (!high_bit_depth && CONFIG_H264CHROMA) {
        c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_mmx_rnd;
        c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_mmx;
    }

    c->vector_clip_int32 = ff_vector_clip_int32_mmx;
#endif

}

static void dsputil_init_mmx2(DSPContext *c, AVCodecContext *avctx,
                              int mm_flags)
{
    const int bit_depth      = avctx->bits_per_raw_sample;
    const int high_bit_depth = bit_depth > 8;

2629
#if HAVE_INLINE_ASM
2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659
    c->prefetch = prefetch_mmx2;

    if (!high_bit_depth) {
        c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
        c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;

        c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
        c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
        c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;

        c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
        c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;

        c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
        c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
        c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
    }

    if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
        if (!high_bit_depth) {
            c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
            c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
            c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
            c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;

            c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
            c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
        }
    }

2660 2661
    if (CONFIG_VP3_DECODER && (avctx->codec_id == AV_CODEC_ID_VP3 ||
                               avctx->codec_id == AV_CODEC_ID_THEORA)) {
2662 2663 2664
        c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_mmx2;
        c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_mmx2;
    }
2665
#endif /* HAVE_INLINE_ASM */
2666

2667
    if (CONFIG_H264QPEL) {
2668
#if HAVE_INLINE_ASM
2669 2670 2671 2672 2673 2674
        SET_QPEL_FUNCS(put_qpel,        0, 16, mmx2, );
        SET_QPEL_FUNCS(put_qpel,        1,  8, mmx2, );
        SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2, );
        SET_QPEL_FUNCS(put_no_rnd_qpel, 1,  8, mmx2, );
        SET_QPEL_FUNCS(avg_qpel,        0, 16, mmx2, );
        SET_QPEL_FUNCS(avg_qpel,        1,  8, mmx2, );
2675
#endif /* HAVE_INLINE_ASM */
2676

2677
        if (!high_bit_depth) {
2678
#if HAVE_INLINE_ASM
2679 2680 2681 2682 2683 2684
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2, );
            SET_QPEL_FUNCS(put_h264_qpel, 1,  8, mmx2, );
            SET_QPEL_FUNCS(put_h264_qpel, 2,  4, mmx2, );
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2, );
            SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, mmx2, );
            SET_QPEL_FUNCS(avg_h264_qpel, 2,  4, mmx2, );
2685
#endif /* HAVE_INLINE_ASM */
2686
        } else if (bit_depth == 10) {
2687 2688
#if HAVE_YASM
#if !ARCH_X86_64
2689 2690 2691 2692
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);
            SET_QPEL_FUNCS(put_h264_qpel, 1,  8, 10_mmxext, ff_);
            SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, 10_mmxext, ff_);
2693
#endif
2694 2695
            SET_QPEL_FUNCS(put_h264_qpel, 2, 4,  10_mmxext, ff_);
            SET_QPEL_FUNCS(avg_h264_qpel, 2, 4,  10_mmxext, ff_);
2696
#endif /* HAVE_YASM */
2697
        }
2698

2699
#if HAVE_INLINE_ASM
2700 2701 2702 2703
        SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2, );
        SET_QPEL_FUNCS(put_2tap_qpel, 1,  8, mmx2, );
        SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2, );
        SET_QPEL_FUNCS(avg_2tap_qpel, 1,  8, mmx2, );
2704
#endif /* HAVE_INLINE_ASM */
2705
    }
2706 2707 2708 2709 2710 2711 2712 2713 2714

#if HAVE_YASM
    if (!high_bit_depth && CONFIG_H264CHROMA) {
        c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_mmx2_rnd;
        c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_mmx2;
        c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_mmx2;
        c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_mmx2;
    }
    if (bit_depth == 10 && CONFIG_H264CHROMA) {
2715 2716 2717 2718
        c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_10_mmx2;
        c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_10_mmx2;
        c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_10_mmx2;
        c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_10_mmx2;
2719 2720
    }

2721 2722 2723
    /* slower than cmov version on AMD */
    if (!(mm_flags & AV_CPU_FLAG_3DNOW))
        c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
2724 2725 2726 2727 2728 2729 2730 2731 2732

    c->scalarproduct_int16          = ff_scalarproduct_int16_mmx2;
    c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmx2;

    if (avctx->flags & CODEC_FLAG_BITEXACT) {
        c->apply_window_int16 = ff_apply_window_int16_mmxext_ba;
    } else {
        c->apply_window_int16 = ff_apply_window_int16_mmxext;
    }
2733
#endif /* HAVE_YASM */
2734 2735 2736 2737
}

static void dsputil_init_3dnow(DSPContext *c, AVCodecContext *avctx,
                               int mm_flags)
Fabrice Bellard's avatar
Fabrice Bellard committed
2738
{
2739
    const int high_bit_depth = avctx->bits_per_raw_sample > 8;
2740

2741
#if HAVE_INLINE_ASM
2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769
    c->prefetch = prefetch_3dnow;

    if (!high_bit_depth) {
        c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
        c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;

        c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
        c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
        c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;

        c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
        c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;

        c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
        c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
        c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;

        if (!(avctx->flags & CODEC_FLAG_BITEXACT)){
            c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
            c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
            c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
            c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;

            c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
            c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
        }
    }

2770 2771
    if (CONFIG_VP3_DECODER && (avctx->codec_id == AV_CODEC_ID_VP3 ||
                               avctx->codec_id == AV_CODEC_ID_THEORA)) {
2772 2773 2774 2775
        c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_3dnow;
        c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_3dnow;
    }

2776
    if (CONFIG_H264QPEL) {
2777 2778 2779 2780 2781 2782
        SET_QPEL_FUNCS(put_qpel,        0, 16, 3dnow, );
        SET_QPEL_FUNCS(put_qpel,        1,  8, 3dnow, );
        SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow, );
        SET_QPEL_FUNCS(put_no_rnd_qpel, 1,  8, 3dnow, );
        SET_QPEL_FUNCS(avg_qpel,        0, 16, 3dnow, );
        SET_QPEL_FUNCS(avg_qpel,        1,  8, 3dnow, );
2783

2784 2785 2786 2787 2788 2789 2790 2791
        if (!high_bit_depth) {
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow, );
            SET_QPEL_FUNCS(put_h264_qpel, 1,  8, 3dnow, );
            SET_QPEL_FUNCS(put_h264_qpel, 2,  4, 3dnow, );
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow, );
            SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, 3dnow, );
            SET_QPEL_FUNCS(avg_h264_qpel, 2,  4, 3dnow, );
        }
2792

2793 2794 2795 2796
        SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow, );
        SET_QPEL_FUNCS(put_2tap_qpel, 1,  8, 3dnow, );
        SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow, );
        SET_QPEL_FUNCS(avg_2tap_qpel, 1,  8, 3dnow, );
2797
    }
2798 2799

    c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
2800 2801 2802 2803 2804 2805 2806
#endif /* HAVE_INLINE_ASM */

#if HAVE_YASM
    if (!high_bit_depth && CONFIG_H264CHROMA) {
        c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_3dnow_rnd;
        c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_3dnow;
    }
2807
#endif /* HAVE_YASM */
2808 2809
}

2810 2811
static void dsputil_init_3dnowext(DSPContext *c, AVCodecContext *avctx,
                                  int mm_flags)
2812
{
2813
#if HAVE_AMD3DNOWEXT_INLINE && HAVE_6REGS
2814
    c->vector_fmul_window  = vector_fmul_window_3dnowext;
2815 2816 2817 2818
#endif
}

static void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx, int mm_flags)
Fabrice Bellard's avatar
Fabrice Bellard committed
2819
{
2820
    const int high_bit_depth = avctx->bits_per_raw_sample > 8;
2821

2822
#if HAVE_INLINE_ASM
2823
    if (!high_bit_depth) {
2824
        if (!(CONFIG_MPEG_XVMC_DECODER && avctx->xvmc_acceleration > 1)) {
2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837
            /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
            c->clear_block  = clear_block_sse;
            c->clear_blocks = clear_blocks_sse;
        }
    }

    c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;

#if HAVE_6REGS
    c->vector_fmul_window = vector_fmul_window_sse;
#endif

    c->vector_clipf = vector_clipf_sse;
2838
#endif /* HAVE_INLINE_ASM */
2839 2840

#if HAVE_YASM
2841 2842 2843
    c->vector_fmul_reverse = ff_vector_fmul_reverse_sse;
    c->vector_fmul_add     = ff_vector_fmul_add_sse;

2844 2845 2846 2847 2848
    c->scalarproduct_float          = ff_scalarproduct_float_sse;
    c->butterflies_float_interleave = ff_butterflies_float_interleave_sse;

    if (!high_bit_depth)
        c->emulated_edge_mc = emulated_edge_mc_sse;
2849 2850 2851
#if HAVE_INLINE_ASM
    c->gmc = gmc_sse;
#endif
2852
#endif /* HAVE_YASM */
2853 2854 2855 2856 2857 2858
}

static void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
                              int mm_flags)
{
    const int bit_depth      = avctx->bits_per_raw_sample;
2859 2860

#if HAVE_INLINE_ASM
2861 2862
    const int high_bit_depth = bit_depth > 8;

2863
    if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
2864 2865 2866 2867 2868
        // these functions are slower than mmx on AMD, but faster on Intel
        if (!high_bit_depth) {
            c->put_pixels_tab[0][0]        = put_pixels16_sse2;
            c->put_no_rnd_pixels_tab[0][0] = put_pixels16_sse2;
            c->avg_pixels_tab[0][0]        = avg_pixels16_sse2;
2869 2870
            if (CONFIG_H264QPEL)
                H264_QPEL_FUNCS(0, 0, sse2);
2871 2872 2873
        }
    }

2874
    if (!high_bit_depth && CONFIG_H264QPEL) {
2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887
        H264_QPEL_FUNCS(0, 1, sse2);
        H264_QPEL_FUNCS(0, 2, sse2);
        H264_QPEL_FUNCS(0, 3, sse2);
        H264_QPEL_FUNCS(1, 1, sse2);
        H264_QPEL_FUNCS(1, 2, sse2);
        H264_QPEL_FUNCS(1, 3, sse2);
        H264_QPEL_FUNCS(2, 1, sse2);
        H264_QPEL_FUNCS(2, 2, sse2);
        H264_QPEL_FUNCS(2, 3, sse2);
        H264_QPEL_FUNCS(3, 1, sse2);
        H264_QPEL_FUNCS(3, 2, sse2);
        H264_QPEL_FUNCS(3, 3, sse2);
    }
2888
#endif /* HAVE_INLINE_ASM */
2889 2890 2891

#if HAVE_YASM
    if (bit_depth == 10) {
2892
        if (CONFIG_H264QPEL) {
2893 2894 2895 2896 2897 2898 2899
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
            SET_QPEL_FUNCS(put_h264_qpel, 1,  8, 10_sse2, ff_);
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
            SET_QPEL_FUNCS(avg_h264_qpel, 1,  8, 10_sse2, ff_);
            H264_QPEL_FUNCS_10(1, 0, sse2_cache64);
            H264_QPEL_FUNCS_10(2, 0, sse2_cache64);
            H264_QPEL_FUNCS_10(3, 0, sse2_cache64);
2900
        }
2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915
        if (CONFIG_H264CHROMA) {
            c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_10_sse2;
            c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_10_sse2;
        }
    }

    c->scalarproduct_int16          = ff_scalarproduct_int16_sse2;
    c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
    if (mm_flags & AV_CPU_FLAG_ATOM) {
        c->vector_clip_int32 = ff_vector_clip_int32_int_sse2;
    } else {
        c->vector_clip_int32 = ff_vector_clip_int32_sse2;
    }
    if (avctx->flags & CODEC_FLAG_BITEXACT) {
        c->apply_window_int16 = ff_apply_window_int16_sse2_ba;
2916
    } else if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
2917 2918 2919
        c->apply_window_int16 = ff_apply_window_int16_sse2;
    }
    c->bswap_buf = ff_bswap32_buf_sse2;
2920
#endif /* HAVE_YASM */
2921 2922 2923 2924 2925 2926 2927 2928
}

static void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx,
                               int mm_flags)
{
    const int high_bit_depth = avctx->bits_per_raw_sample > 8;
    const int bit_depth      = avctx->bits_per_raw_sample;

2929
#if HAVE_SSSE3_INLINE
2930
    if (!high_bit_depth && CONFIG_H264QPEL) {
2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943
        H264_QPEL_FUNCS(1, 0, ssse3);
        H264_QPEL_FUNCS(1, 1, ssse3);
        H264_QPEL_FUNCS(1, 2, ssse3);
        H264_QPEL_FUNCS(1, 3, ssse3);
        H264_QPEL_FUNCS(2, 0, ssse3);
        H264_QPEL_FUNCS(2, 1, ssse3);
        H264_QPEL_FUNCS(2, 2, ssse3);
        H264_QPEL_FUNCS(2, 3, ssse3);
        H264_QPEL_FUNCS(3, 0, ssse3);
        H264_QPEL_FUNCS(3, 1, ssse3);
        H264_QPEL_FUNCS(3, 2, ssse3);
        H264_QPEL_FUNCS(3, 3, ssse3);
    }
2944 2945 2946
#endif /* HAVE_SSSE3_INLINE */

#if HAVE_SSSE3_EXTERNAL
2947
    if (bit_depth == 10 && CONFIG_H264QPEL) {
2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961
        H264_QPEL_FUNCS_10(1, 0, ssse3_cache64);
        H264_QPEL_FUNCS_10(2, 0, ssse3_cache64);
        H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
    }
    if (!high_bit_depth && CONFIG_H264CHROMA) {
        c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_ssse3_rnd;
        c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_ssse3_rnd;
        c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_ssse3;
        c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_ssse3;
    }
    c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
    if (mm_flags & AV_CPU_FLAG_SSE4) // not really sse4, just slow on Conroe
        c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;

2962
    if (mm_flags & AV_CPU_FLAG_ATOM)
2963
        c->apply_window_int16 = ff_apply_window_int16_ssse3_atom;
2964
    else
2965
        c->apply_window_int16 = ff_apply_window_int16_ssse3;
2966
    if (!(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW))) // cachesplit
2967 2968
        c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
    c->bswap_buf = ff_bswap32_buf_ssse3;
2969
#endif /* HAVE_SSSE3_EXTERNAL */
2970 2971 2972 2973 2974
}

static void dsputil_init_sse4(DSPContext *c, AVCodecContext *avctx,
                              int mm_flags)
{
2975
#if HAVE_SSE4_EXTERNAL
2976
    c->vector_clip_int32 = ff_vector_clip_int32_sse4;
2977
#endif /* HAVE_SSE4_EXTERNAL */
2978 2979 2980 2981
}

static void dsputil_init_avx(DSPContext *c, AVCodecContext *avctx, int mm_flags)
{
2982
#if HAVE_AVX_EXTERNAL
2983
    const int bit_depth = avctx->bits_per_raw_sample;
Zdenek Kabelac's avatar
Zdenek Kabelac committed
2984

2985 2986 2987
    if (bit_depth == 10) {
        // AVX implies !cache64.
        // TODO: Port cache(32|64) detection from x264.
2988
        if (CONFIG_H264QPEL) {
2989 2990 2991
            H264_QPEL_FUNCS_10(1, 0, sse2);
            H264_QPEL_FUNCS_10(2, 0, sse2);
            H264_QPEL_FUNCS_10(3, 0, sse2);
2992
        }
2993 2994 2995 2996 2997 2998 2999

        if (CONFIG_H264CHROMA) {
            c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_10_avx;
            c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_10_avx;
        }
    }
    c->butterflies_float_interleave = ff_butterflies_float_interleave_avx;
3000 3001
    c->vector_fmul_reverse = ff_vector_fmul_reverse_avx;
    c->vector_fmul_add = ff_vector_fmul_add_avx;
3002
#endif /* HAVE_AVX_EXTERNAL */
3003 3004
}

3005
void ff_dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx)
3006 3007 3008
{
    int mm_flags = av_get_cpu_flags();

3009 3010 3011 3012 3013
#if HAVE_7REGS && HAVE_INLINE_ASM
    if (mm_flags & AV_CPU_FLAG_CMOV)
        c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;
#endif

3014
    if (mm_flags & AV_CPU_FLAG_MMX) {
3015
#if HAVE_INLINE_ASM
3016
        const int idct_algo = avctx->idct_algo;
3017

Michael Niedermayer's avatar
Michael Niedermayer committed
3018
        if (avctx->lowres == 0 && avctx->bits_per_raw_sample <= 8) {
3019 3020 3021 3022 3023
            if (idct_algo == FF_IDCT_AUTO || idct_algo == FF_IDCT_SIMPLEMMX) {
                c->idct_put              = ff_simple_idct_put_mmx;
                c->idct_add              = ff_simple_idct_add_mmx;
                c->idct                  = ff_simple_idct_mmx;
                c->idct_permutation_type = FF_SIMPLE_IDCT_PERM;
3024
#if CONFIG_GPL
3025 3026 3027 3028 3029 3030 3031 3032 3033
            } else if (idct_algo == FF_IDCT_LIBMPEG2MMX) {
                if (mm_flags & AV_CPU_FLAG_MMX2) {
                    c->idct_put = ff_libmpeg2mmx2_idct_put;
                    c->idct_add = ff_libmpeg2mmx2_idct_add;
                    c->idct     = ff_mmxext_idct;
                } else {
                    c->idct_put = ff_libmpeg2mmx_idct_put;
                    c->idct_add = ff_libmpeg2mmx_idct_add;
                    c->idct     = ff_mmx_idct;
3034
                }
3035
                c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM;
3036
#endif
3037 3038 3039 3040 3041 3042
            } else if (idct_algo == FF_IDCT_XVIDMMX) {
                if (mm_flags & AV_CPU_FLAG_SSE2) {
                    c->idct_put              = ff_idct_xvid_sse2_put;
                    c->idct_add              = ff_idct_xvid_sse2_add;
                    c->idct                  = ff_idct_xvid_sse2;
                    c->idct_permutation_type = FF_SSE2_IDCT_PERM;
3043
                } else if (mm_flags & AV_CPU_FLAG_MMXEXT) {
3044 3045 3046 3047 3048 3049 3050
                    c->idct_put              = ff_idct_xvid_mmx2_put;
                    c->idct_add              = ff_idct_xvid_mmx2_add;
                    c->idct                  = ff_idct_xvid_mmx2;
                } else {
                    c->idct_put              = ff_idct_xvid_mmx_put;
                    c->idct_add              = ff_idct_xvid_mmx_add;
                    c->idct                  = ff_idct_xvid_mmx;
3051
                }
3052
            }
3053
        }
3054
#endif /* HAVE_INLINE_ASM */
3055

3056 3057
        dsputil_init_mmx(c, avctx, mm_flags);
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
3058

3059
    if (mm_flags & AV_CPU_FLAG_MMXEXT)
3060
        dsputil_init_mmx2(c, avctx, mm_flags);
3061

3062
    if (mm_flags & AV_CPU_FLAG_3DNOW)
3063
        dsputil_init_3dnow(c, avctx, mm_flags);
Loren Merritt's avatar
Loren Merritt committed
3064

3065
    if (mm_flags & AV_CPU_FLAG_3DNOWEXT)
3066
        dsputil_init_3dnowext(c, avctx, mm_flags);
3067

3068
    if (mm_flags & AV_CPU_FLAG_SSE)
3069
        dsputil_init_sse(c, avctx, mm_flags);
3070

3071 3072
    if (mm_flags & AV_CPU_FLAG_SSE2)
        dsputil_init_sse2(c, avctx, mm_flags);
3073

3074 3075
    if (mm_flags & AV_CPU_FLAG_SSSE3)
        dsputil_init_ssse3(c, avctx, mm_flags);
3076

3077
    if (mm_flags & AV_CPU_FLAG_SSE4)
3078
        dsputil_init_sse4(c, avctx, mm_flags);
3079

3080 3081
    if (mm_flags & AV_CPU_FLAG_AVX)
        dsputil_init_avx(c, avctx, mm_flags);
3082

3083
    if (CONFIG_ENCODERS)
3084
        ff_dsputilenc_init_mmx(c, avctx);
Fabrice Bellard's avatar
Fabrice Bellard committed
3085
}