dsputil_mmx.c 147 KB
Newer Older
Fabrice Bellard's avatar
Fabrice Bellard committed
1 2
/*
 * MMX optimized DSP utils
3
 * Copyright (c) 2000, 2001 Fabrice Bellard
4
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
Fabrice Bellard's avatar
Fabrice Bellard committed
5
 *
6 7
 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
 *
8
 * This file is part of Libav.
9
 *
10
 * Libav is free software; you can redistribute it and/or
Fabrice Bellard's avatar
Fabrice Bellard committed
11 12
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
13
 * version 2.1 of the License, or (at your option) any later version.
Fabrice Bellard's avatar
Fabrice Bellard committed
14
 *
15
 * Libav is distributed in the hope that it will be useful,
Fabrice Bellard's avatar
Fabrice Bellard committed
16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
Fabrice Bellard's avatar
Fabrice Bellard committed
17 18
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
Fabrice Bellard's avatar
Fabrice Bellard committed
19
 *
Fabrice Bellard's avatar
Fabrice Bellard committed
20
 * You should have received a copy of the GNU Lesser General Public
21
 * License along with Libav; if not, write to the Free Software
22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Fabrice Bellard's avatar
Fabrice Bellard committed
23 24
 */

25
#include "libavutil/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"
Fabrice Bellard's avatar
Fabrice Bellard committed
33

34 35 36
//#undef NDEBUG
//#include <assert.h>

Fabrice Bellard's avatar
Fabrice Bellard committed
37
/* pixel operations */
38 39
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
40

41
DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81
    { 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 };
82 83 84

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 };
85

86 87
#if HAVE_INLINE_ASM

88 89
#define JUMPALIGN()     __asm__ volatile (".p2align 3"::)
#define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%"#regd", %%"#regd ::)
90

91 92 93 94
#define MOVQ_BFE(regd)                                  \
    __asm__ volatile (                                  \
        "pcmpeqd %%"#regd", %%"#regd"   \n\t"           \
        "paddb   %%"#regd", %%"#regd"   \n\t" ::)
95

96
#ifndef PIC
97 98
#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))
99 100 101
#else
// for shared library it's better to use this way for accessing constants
// pcmpeqd -> -1
102 103 104 105 106 107 108 109 110 111 112
#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"::)
113

114 115
#endif

116
// using regr as temporary and for the output result
117
// first argument is unmodifed and second is trashed
118
// regfe is supposed to contain 0xfefefefefefefefe
119 120 121 122 123 124 125 126 127 128 129 130 131 132 133
#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"
134

135
// mm6 is supposed to contain 0xfefefefefefefefe
136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
#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"
163

164 165
/***********************************/
/* MMX no rounding */
166
#define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
167
#define SET_RND  MOVQ_WONE
168 169
#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)
170
#define OP_AVG(a, b, c, e)              PAVGB_MMX(a, b, c, e)
171

172
#include "dsputil_rnd_template.c"
173 174

#undef DEF
175
#undef SET_RND
176
#undef PAVGBP
177
#undef PAVGB
178 179 180
/***********************************/
/* MMX rounding */

181
#define DEF(x, y) x ## _ ## y ## _mmx
182
#define SET_RND  MOVQ_WTWO
183 184
#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)
185

186
#include "dsputil_rnd_template.c"
187 188

#undef DEF
189
#undef SET_RND
190
#undef PAVGBP
191
#undef PAVGB
192
#undef OP_AVG
193

Fabrice Bellard's avatar
Fabrice Bellard committed
194 195 196 197 198
/***********************************/
/* 3Dnow specific */

#define DEF(x) x ## _3dnow
#define PAVGB "pavgusb"
199
#define OP_AVG PAVGB
Fabrice Bellard's avatar
Fabrice Bellard committed
200

201
#include "dsputil_avg_template.c"
Fabrice Bellard's avatar
Fabrice Bellard committed
202 203 204

#undef DEF
#undef PAVGB
205
#undef OP_AVG
Fabrice Bellard's avatar
Fabrice Bellard committed
206 207 208 209

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

210
#define DEF(x) x ## _mmx2
Fabrice Bellard's avatar
Fabrice Bellard committed
211 212 213

/* Introduced only in MMX2 set */
#define PAVGB "pavgb"
214
#define OP_AVG PAVGB
Fabrice Bellard's avatar
Fabrice Bellard committed
215

216
#include "dsputil_avg_template.c"
Fabrice Bellard's avatar
Fabrice Bellard committed
217 218 219

#undef DEF
#undef PAVGB
220
#undef OP_AVG
Fabrice Bellard's avatar
Fabrice Bellard committed
221

Loren Merritt's avatar
Loren Merritt committed
222 223
#define put_no_rnd_pixels16_mmx put_pixels16_mmx
#define put_no_rnd_pixels8_mmx put_pixels8_mmx
224 225 226 227 228 229 230 231 232 233 234
#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
235 236 237
/***********************************/
/* standard MMX */

238 239
void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels,
                               int line_size)
Fabrice Bellard's avatar
Fabrice Bellard committed
240 241
{
    const DCTELEM *p;
242
    uint8_t *pix;
Fabrice Bellard's avatar
Fabrice Bellard committed
243 244

    /* read the pixels */
245
    p   = block;
Fabrice Bellard's avatar
Fabrice Bellard committed
246
    pix = pixels;
247
    /* unrolled loop */
248
    __asm__ volatile (
249 250 251 252 253 254 255 256
        "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"
257 258 259 260 261 262 263 264 265
        "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),
266
           "r"(p)
267
        : "memory");
268 269
    pix += line_size * 4;
    p   += 32;
270 271 272 273

    // if here would be an exact copy of the code above
    // compiler would generate some very strange code
    // thus using "r"
274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292
    __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
293 294
}

295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311
#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"
312

313 314
void ff_put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels,
                                      int line_size)
315
{
316
    x86_reg line_skip = line_size;
317
    x86_reg line_skip3;
318 319

    __asm__ volatile (
320 321 322 323 324 325 326 327
        "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");
328 329
}

330 331
void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels,
                               int line_size)
Fabrice Bellard's avatar
Fabrice Bellard committed
332 333
{
    const DCTELEM *p;
334
    uint8_t *pix;
Fabrice Bellard's avatar
Fabrice Bellard committed
335 336 337
    int i;

    /* read the pixels */
338
    p   = block;
Fabrice Bellard's avatar
Fabrice Bellard committed
339
    pix = pixels;
340 341
    MOVQ_ZERO(mm7);
    i = 4;
342
    do {
343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366
        __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");
367 368
        pix += line_size * 2;
        p   += 16;
369
    } while (--i);
Fabrice Bellard's avatar
Fabrice Bellard committed
370 371
}

372 373
static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels,
                            int line_size, int h)
374
{
375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
    __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"
396
        );
397 398
}

399 400
static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
                            int line_size, int h)
Fabrice Bellard's avatar
Fabrice Bellard committed
401
{
402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422
    __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"
423
        );
Fabrice Bellard's avatar
Fabrice Bellard committed
424 425
}

426 427
static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels,
                             int line_size, int h)
Michael Niedermayer's avatar
Michael Niedermayer committed
428
{
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
    __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"
458
        );
Michael Niedermayer's avatar
Michael Niedermayer committed
459 460
}

461 462
static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
                              int line_size, int h)
463
{
464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480
    __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"
481 482 483
        );
}

484 485
static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
                              int line_size, int h)
486
{
487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507
    __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"
508 509 510
        );
}

511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527
#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
528 529 530 531 532
}
CLEAR_BLOCKS(clear_blocks_mmx, 6)
CLEAR_BLOCKS(clear_block_mmx, 1)

static void clear_block_sse(DCTELEM *block)
533
{
534 535 536 537 538 539 540 541 542 543
    __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
544 545 546
        :: "r"(block)
        : "memory"
    );
547 548
}

549
static void clear_blocks_sse(DCTELEM *blocks)
550
{
551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566
    __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)
567 568 569 570
        : "%"REG_a
    );
}

571 572 573
static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w)
{
    x86_reg i = 0;
574 575
    __asm__ volatile (
        "jmp          2f                \n\t"
576
        "1:                             \n\t"
577 578 579 580 581 582 583 584 585
        "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"
586
        "2:                             \n\t"
587 588 589 590
        "cmp          %3, %0            \n\t"
        "js           1b                \n\t"
        : "+r"(i)
        : "r"(src), "r"(dst), "r"((x86_reg)w - 15)
Michael Niedermayer's avatar
huffyuv  
Michael Niedermayer committed
591
    );
592 593
    for ( ; i < w; i++)
        dst[i + 0] += src[i + 0];
Michael Niedermayer's avatar
huffyuv  
Michael Niedermayer committed
594 595
}

596
#if HAVE_7REGS
597 598 599 600
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)
{
601 602
    x86_reg w2 = -w;
    x86_reg x;
603
    int l  = *left     & 0xff;
604 605
    int tl = *left_top & 0xff;
    int t;
606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627
    __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)
628
    );
629
    *left     = l;
630 631 632 633
    *left_top = tl;
}
#endif

634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661
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"
    );
}

662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731
#define H263_LOOP_FILTER                        \
    "pxor      %%mm7, %%mm7             \n\t"   \
    "movq         %0, %%mm0             \n\t"   \
    "movq         %0, %%mm1             \n\t"   \
    "movq         %3, %%mm2             \n\t"   \
    "movq         %3, %%mm3             \n\t"   \
    "punpcklbw %%mm7, %%mm0             \n\t"   \
    "punpckhbw %%mm7, %%mm1             \n\t"   \
    "punpcklbw %%mm7, %%mm2             \n\t"   \
    "punpckhbw %%mm7, %%mm3             \n\t"   \
    "psubw     %%mm2, %%mm0             \n\t"   \
    "psubw     %%mm3, %%mm1             \n\t"   \
    "movq         %1, %%mm2             \n\t"   \
    "movq         %1, %%mm3             \n\t"   \
    "movq         %2, %%mm4             \n\t"   \
    "movq         %2, %%mm5             \n\t"   \
    "punpcklbw %%mm7, %%mm2             \n\t"   \
    "punpckhbw %%mm7, %%mm3             \n\t"   \
    "punpcklbw %%mm7, %%mm4             \n\t"   \
    "punpckhbw %%mm7, %%mm5             \n\t"   \
    "psubw     %%mm2, %%mm4             \n\t"   \
    "psubw     %%mm3, %%mm5             \n\t"   \
    "psllw        $2, %%mm4             \n\t"   \
    "psllw        $2, %%mm5             \n\t"   \
    "paddw     %%mm0, %%mm4             \n\t"   \
    "paddw     %%mm1, %%mm5             \n\t"   \
    "pxor      %%mm6, %%mm6             \n\t"   \
    "pcmpgtw   %%mm4, %%mm6             \n\t"   \
    "pcmpgtw   %%mm5, %%mm7             \n\t"   \
    "pxor      %%mm6, %%mm4             \n\t"   \
    "pxor      %%mm7, %%mm5             \n\t"   \
    "psubw     %%mm6, %%mm4             \n\t"   \
    "psubw     %%mm7, %%mm5             \n\t"   \
    "psrlw        $3, %%mm4             \n\t"   \
    "psrlw        $3, %%mm5             \n\t"   \
    "packuswb  %%mm5, %%mm4             \n\t"   \
    "packsswb  %%mm7, %%mm6             \n\t"   \
    "pxor      %%mm7, %%mm7             \n\t"   \
    "movd         %4, %%mm2             \n\t"   \
    "punpcklbw %%mm2, %%mm2             \n\t"   \
    "punpcklbw %%mm2, %%mm2             \n\t"   \
    "punpcklbw %%mm2, %%mm2             \n\t"   \
    "psubusb   %%mm4, %%mm2             \n\t"   \
    "movq      %%mm2, %%mm3             \n\t"   \
    "psubusb   %%mm4, %%mm3             \n\t"   \
    "psubb     %%mm3, %%mm2             \n\t"   \
    "movq         %1, %%mm3             \n\t"   \
    "movq         %2, %%mm4             \n\t"   \
    "pxor      %%mm6, %%mm3             \n\t"   \
    "pxor      %%mm6, %%mm4             \n\t"   \
    "paddusb   %%mm2, %%mm3             \n\t"   \
    "psubusb   %%mm2, %%mm4             \n\t"   \
    "pxor      %%mm6, %%mm3             \n\t"   \
    "pxor      %%mm6, %%mm4             \n\t"   \
    "paddusb   %%mm2, %%mm2             \n\t"   \
    "packsswb  %%mm1, %%mm0             \n\t"   \
    "pcmpgtb   %%mm0, %%mm7             \n\t"   \
    "pxor      %%mm7, %%mm0             \n\t"   \
    "psubb     %%mm7, %%mm0             \n\t"   \
    "movq      %%mm0, %%mm1             \n\t"   \
    "psubusb   %%mm2, %%mm0             \n\t"   \
    "psubb     %%mm0, %%mm1             \n\t"   \
    "pand         %5, %%mm1             \n\t"   \
    "psrlw        $2, %%mm1             \n\t"   \
    "pxor      %%mm7, %%mm1             \n\t"   \
    "psubb     %%mm7, %%mm1             \n\t"   \
    "movq         %0, %%mm5             \n\t"   \
    "movq         %3, %%mm6             \n\t"   \
    "psubb     %%mm1, %%mm5             \n\t"   \
    "paddb     %%mm1, %%mm6             \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
732

733 734 735 736
static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale)
{
    if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
        const int strength = ff_h263_loop_filter_strength[qscale];
Michael Niedermayer's avatar
Michael Niedermayer committed
737

738 739 740 741 742 743 744 745 746 747 748 749 750
        __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)
            );
751
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
752 753
}

754 755 756 757 758 759
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;
760

761
        src -= 2;
Michael Niedermayer's avatar
Michael Niedermayer committed
762

763 764
        transpose4x4(btemp,     src,              8, stride);
        transpose4x4(btemp + 4, src + 4 * stride, 8, stride);
765 766 767 768 769 770 771 772 773
        __asm__ volatile (
            H263_LOOP_FILTER // 5 3 4 6

            : "+m"(temp[0]),
              "+m"(temp[1]),
              "+m"(temp[2]),
              "+m"(temp[3])
            : "g"(2 * strength), "m"(ff_pb_FC)
            );
Michael Niedermayer's avatar
Michael Niedermayer committed
774

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
        __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))
            );
805
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
806 807
}

808 809 810 811
/* 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)
812 813 814 815 816 817 818
{
    uint8_t *ptr, *last_line;
    int i;

    last_line = buf + (height - 1) * wrap;
    /* left and right */
    ptr = buf;
819
    if (w == 8) {
820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837
        __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)
            );
838
    } else {
839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858
        __asm__ volatile (
            "1:                                 \n\t"
            "movd            (%0), %%mm0        \n\t"
            "punpcklbw      %%mm0, %%mm0        \n\t"
            "punpcklwd      %%mm0, %%mm0        \n\t"
            "punpckldq      %%mm0, %%mm0        \n\t"
            "movq           %%mm0, -8(%0)       \n\t"
            "movq           %%mm0, -16(%0)      \n\t"
            "movq      -8(%0, %2), %%mm1        \n\t"
            "punpckhbw      %%mm1, %%mm1        \n\t"
            "punpckhwd      %%mm1, %%mm1        \n\t"
            "punpckhdq      %%mm1, %%mm1        \n\t"
            "movq           %%mm1,  (%0, %2)    \n\t"
            "movq           %%mm1, 8(%0, %2)    \n\t"
            "add               %1, %0           \n\t"
            "cmp               %3, %0           \n\t"
            "jb                1b               \n\t"
            : "+r"(ptr)
            : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
            );
859 860
    }

861
    /* top and bottom (and hopefully also the corners) */
862 863 864
    if (sides & EDGE_TOP) {
        for (i = 0; i < h; i += 4) {
            ptr = buf - (i + 1) * wrap - w;
865 866 867 868 869 870 871 872 873 874 875 876 877 878
            __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)
                );
879 880 881
        }
    }

882 883 884
    if (sides & EDGE_BOTTOM) {
        for (i = 0; i < h; i += 4) {
            ptr = last_line + (i + 1) * wrap - w;
885 886 887 888 889 890 891 892 893 894 895 896 897 898 899
            __asm__ volatile (
                "1:                             \n\t"
                "movq (%1, %0), %%mm0           \n\t"
                "movq    %%mm0, (%0)            \n\t"
                "movq    %%mm0, (%0, %2)        \n\t"
                "movq    %%mm0, (%0, %2, 2)     \n\t"
                "movq    %%mm0, (%0, %3)        \n\t"
                "add        $8, %0              \n\t"
                "cmp        %4, %0              \n\t"
                "jb         1b                  \n\t"
                : "+r"(ptr)
                : "r"((x86_reg)last_line - (x86_reg)ptr - w),
                  "r"((x86_reg)wrap), "r"((x86_reg)wrap * 3),
                  "r"(ptr + width + 2 * w)
                );
900
        }
901 902 903
    }
}

904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 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
#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;                                                 \
    }                                                                     \
1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243
}

#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;                                                     \
                                                                        \
1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267
    /* 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"                                                      \
        );                                                              \
1268 1269 1270 1271
                                                                        \
    temp_ptr = temp;                                                    \
    count    = 4;                                                       \
                                                                        \
1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 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
    /* 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"                                                      \
        );                                                              \
1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327
}                                                                       \
                                                                        \
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;                                                      \
                                                                        \
1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345
    /* 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"                                                      \
        );                                                              \
1346 1347 1348 1349
                                                                        \
    temp_ptr = temp;                                                    \
    count    = 2;                                                       \
                                                                        \
1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382
    /* 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"                                                      \
        );                                                              \
1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413