dsputil_mmx.c 119 KB
Newer Older
Fabrice Bellard's avatar
Fabrice Bellard committed
1 2
/*
 * MMX optimized DSP utils
Fabrice Bellard's avatar
Fabrice Bellard committed
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
Fabrice Bellard's avatar
Fabrice Bellard committed
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
Fabrice Bellard's avatar
Fabrice Bellard committed
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
 *
Fabrice Bellard's avatar
Fabrice Bellard committed
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 26 27 28 29
#include "libavutil/x86_cpu.h"
#include "libavcodec/dsputil.h"
#include "libavcodec/h263.h"
#include "libavcodec/mpegvideo.h"
#include "libavcodec/simple_idct.h"
30
#include "dsputil_mmx.h"
31
#include "mmx.h"
32 33
#include "vp3dsp_mmx.h"
#include "vp3dsp_sse2.h"
34
#include "idct_xvid.h"
Fabrice Bellard's avatar
Fabrice Bellard committed
35

36 37 38
//#undef NDEBUG
//#include <assert.h>

Måns Rullgård's avatar
Måns Rullgård committed
39
int mm_flags; /* multimedia extension flags */
Michael Niedermayer's avatar
Michael Niedermayer committed
40

Fabrice Bellard's avatar
Fabrice Bellard committed
41
/* pixel operations */
42 43
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
44

45
DECLARE_ALIGNED_16(const uint64_t, ff_pdw_80000000[2]) =
46 47
{0x8000000080000000ULL, 0x8000000080000000ULL};

48 49
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_3  ) = 0x0003000300030003ULL;
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_4  ) = 0x0004000400040004ULL;
50
DECLARE_ALIGNED_16(const xmm_t,    ff_pw_5  ) = {0x0005000500050005ULL, 0x0005000500050005ULL};
51
DECLARE_ALIGNED_16(const xmm_t,    ff_pw_8  ) = {0x0008000800080008ULL, 0x0008000800080008ULL};
52
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL;
53
DECLARE_ALIGNED_16(const xmm_t,    ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL};
54
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL;
Loren Merritt's avatar
Loren Merritt committed
55
DECLARE_ALIGNED_16(const xmm_t,    ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL};
56
DECLARE_ALIGNED_16(const xmm_t,    ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL};
57 58 59
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL;
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_64 ) = 0x0040004000400040ULL;
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL;
60
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
Loren Merritt's avatar
Loren Merritt committed
61
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
Michael Niedermayer's avatar
Michael Niedermayer committed
62

63 64 65
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_1  ) = 0x0101010101010101ULL;
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3  ) = 0x0303030303030303ULL;
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_7  ) = 0x0707070707070707ULL;
David Conrad's avatar
David Conrad committed
66
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL;
67
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
David Conrad's avatar
David Conrad committed
68
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL;
69 70
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_A1 ) = 0xA1A1A1A1A1A1A1A1ULL;
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;
Michael Niedermayer's avatar
Michael Niedermayer committed
71

72 73
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 };
74

75 76
#define JUMPALIGN() __asm__ volatile (ASMALIGN(3)::)
#define MOVQ_ZERO(regd)  __asm__ volatile ("pxor %%" #regd ", %%" #regd ::)
77

78
#define MOVQ_BFE(regd) \
79
    __asm__ volatile ( \
80 81 82
    "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
    "paddb %%" #regd ", %%" #regd " \n\t" ::)

83
#ifndef PIC
84 85
#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))
86 87 88
#else
// for shared library it's better to use this way for accessing constants
// pcmpeqd -> -1
89
#define MOVQ_BONE(regd) \
90
    __asm__ volatile ( \
91 92 93
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
    "psrlw $15, %%" #regd " \n\t" \
    "packuswb %%" #regd ", %%" #regd " \n\t" ::)
94 95

#define MOVQ_WTWO(regd) \
96
    __asm__ volatile ( \
97 98 99
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
    "psrlw $15, %%" #regd " \n\t" \
    "psllw $1, %%" #regd " \n\t"::)
100

101 102
#endif

103
// using regr as temporary and for the output result
104
// first argument is unmodifed and second is trashed
105 106
// regfe is supposed to contain 0xfefefefefefefefe
#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
107 108 109 110 111 112
    "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"
113

114
#define PAVGB_MMX(rega, regb, regr, regfe) \
115 116 117 118 119 120
    "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"
121

122
// mm6 is supposed to contain 0xfefefefefefefefe
123
#define PAVGBP_MMX_NO_RND(rega, regb, regr,  regc, regd, regp) \
124 125 126 127 128 129 130 131 132 133 134 135
    "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"
136 137

#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
138 139 140 141 142 143 144 145 146 147 148 149
    "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"
150

151 152 153
/***********************************/
/* MMX no rounding */
#define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
154
#define SET_RND  MOVQ_WONE
155 156
#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)
157

158 159 160
#include "dsputil_mmx_rnd.h"

#undef DEF
161
#undef SET_RND
162
#undef PAVGBP
163
#undef PAVGB
164 165 166 167
/***********************************/
/* MMX rounding */

#define DEF(x, y) x ## _ ## y ##_mmx
168
#define SET_RND  MOVQ_WTWO
169 170
#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)
171

172 173 174
#include "dsputil_mmx_rnd.h"

#undef DEF
175
#undef SET_RND
176
#undef PAVGBP
177
#undef PAVGB
178

Fabrice Bellard's avatar
Fabrice Bellard committed
179 180 181 182 183 184 185 186 187 188 189 190 191 192
/***********************************/
/* 3Dnow specific */

#define DEF(x) x ## _3dnow
#define PAVGB "pavgusb"

#include "dsputil_mmx_avg.h"

#undef DEF
#undef PAVGB

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

193
#define DEF(x) x ## _mmx2
Fabrice Bellard's avatar
Fabrice Bellard committed
194 195 196 197 198 199 200 201 202

/* Introduced only in MMX2 set */
#define PAVGB "pavgb"

#include "dsputil_mmx_avg.h"

#undef DEF
#undef PAVGB

Loren Merritt's avatar
Loren Merritt committed
203 204
#define put_no_rnd_pixels16_mmx put_pixels16_mmx
#define put_no_rnd_pixels8_mmx put_pixels8_mmx
205 206 207 208 209 210 211 212 213 214 215
#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
216 217 218
/***********************************/
/* standard MMX */

219
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
Fabrice Bellard's avatar
Fabrice Bellard committed
220 221
{
    const DCTELEM *p;
222
    uint8_t *pix;
Fabrice Bellard's avatar
Fabrice Bellard committed
223 224 225 226

    /* read the pixels */
    p = block;
    pix = pixels;
227
    /* unrolled loop */
228
        __asm__ volatile(
229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244
                "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"
245
                ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p)
246
                :"memory");
Fabrice Bellard's avatar
Fabrice Bellard committed
247 248
        pix += line_size*4;
        p += 32;
249 250 251 252

    // if here would be an exact copy of the code above
    // compiler would generate some very strange code
    // thus using "r"
253
    __asm__ volatile(
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269
            "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"
270
            ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p)
271
            :"memory");
Fabrice Bellard's avatar
Fabrice Bellard committed
272 273
}

274
static DECLARE_ALIGNED_8(const unsigned char, vector128[8]) =
275 276
  { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };

277 278 279 280 281 282 283 284 285 286 287 288 289 290 291
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
{
    int i;

    movq_m2r(*vector128, mm1);
    for (i = 0; i < 8; i++) {
        movq_m2r(*(block), mm0);
        packsswb_m2r(*(block + 4), mm0);
        block += 8;
        paddb_r2r(mm1, mm0);
        movq_r2m(mm0, *pixels);
        pixels += line_size;
    }
}

292
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
Fabrice Bellard's avatar
Fabrice Bellard committed
293 294
{
    const DCTELEM *p;
295
    uint8_t *pix;
Fabrice Bellard's avatar
Fabrice Bellard committed
296 297 298 299 300
    int i;

    /* read the pixels */
    p = block;
    pix = pixels;
301 302
    MOVQ_ZERO(mm7);
    i = 4;
303
    do {
304
        __asm__ volatile(
305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327
                "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");
Fabrice Bellard's avatar
Fabrice Bellard committed
328 329
        pix += line_size*2;
        p += 16;
330
    } while (--i);
Fabrice Bellard's avatar
Fabrice Bellard committed
331 332
}

333 334
static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
335
    __asm__ volatile(
336
         "lea (%3, %3), %%"REG_a"       \n\t"
337
         ASMALIGN(3)
338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
         "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)
354
         : "r"((x86_reg)line_size)
355 356
         : "%"REG_a, "memory"
        );
357 358
}

359
static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
Fabrice Bellard's avatar
Fabrice Bellard committed
360
{
361
    __asm__ volatile(
362
         "lea (%3, %3), %%"REG_a"       \n\t"
363
         ASMALIGN(3)
364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379
         "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)
380
         : "r"((x86_reg)line_size)
381 382
         : "%"REG_a, "memory"
        );
Fabrice Bellard's avatar
Fabrice Bellard committed
383 384
}

385
static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
Michael Niedermayer's avatar
Michael Niedermayer committed
386
{
387
    __asm__ volatile(
388
         "lea (%3, %3), %%"REG_a"       \n\t"
389
         ASMALIGN(3)
390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413
         "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)
414
         : "r"((x86_reg)line_size)
415 416
         : "%"REG_a, "memory"
        );
Michael Niedermayer's avatar
Michael Niedermayer committed
417 418
}

419 420
static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
421
    __asm__ volatile(
422 423 424 425 426 427 428 429 430 431 432 433 434 435
         "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"
         "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 (%1,%3,4), %1             \n\t"
         "lea (%2,%3,4), %2             \n\t"
         "jnz 1b                        \n\t"
         : "+g"(h), "+r" (pixels),  "+r" (block)
436
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
437 438 439 440 441 442
         : "memory"
        );
}

static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
443
    __asm__ volatile(
444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461
         "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"
         "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 (%1,%3,4), %1             \n\t"
         "lea (%2,%3,4), %2             \n\t"
         "jnz 1b                        \n\t"
         : "+g"(h), "+r" (pixels),  "+r" (block)
462
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
463 464 465 466
         : "memory"
        );
}

467 468
static void clear_blocks_mmx(DCTELEM *blocks)
{
469
    __asm__ volatile(
470 471 472 473 474 475 476 477 478
                "pxor %%mm7, %%mm7              \n\t"
                "mov $-128*6, %%"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"
479 480
                : : "r" (((uint8_t *)blocks)+128*6)
                : "%"REG_a
481 482 483
        );
}

Michael Niedermayer's avatar
huffyuv  
Michael Niedermayer committed
484
static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
485
    x86_reg i=0;
486
    __asm__ volatile(
487
        "jmp 2f                         \n\t"
488 489 490 491 492 493 494 495 496 497
        "1:                             \n\t"
        "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"
498
        "2:                             \n\t"
499
        "cmp %3, %0                     \n\t"
500
        " js 1b                         \n\t"
Michael Niedermayer's avatar
huffyuv  
Michael Niedermayer committed
501
        : "+r" (i)
502
        : "r"(src), "r"(dst), "r"((x86_reg)w-15)
Michael Niedermayer's avatar
huffyuv  
Michael Niedermayer committed
503 504 505 506 507
    );
    for(; i<w; i++)
        dst[i+0] += src[i+0];
}

Loren Merritt's avatar
Loren Merritt committed
508
static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
509
    x86_reg i=0;
510
    __asm__ volatile(
511
        "jmp 2f                         \n\t"
Loren Merritt's avatar
Loren Merritt committed
512 513 514 515 516 517 518 519
        "1:                             \n\t"
        "movq   (%2, %0), %%mm0         \n\t"
        "movq  8(%2, %0), %%mm1         \n\t"
        "paddb  (%3, %0), %%mm0         \n\t"
        "paddb 8(%3, %0), %%mm1         \n\t"
        "movq %%mm0,  (%1, %0)          \n\t"
        "movq %%mm1, 8(%1, %0)          \n\t"
        "add $16, %0                    \n\t"
520
        "2:                             \n\t"
Loren Merritt's avatar
Loren Merritt committed
521
        "cmp %4, %0                     \n\t"
522
        " js 1b                         \n\t"
Loren Merritt's avatar
Loren Merritt committed
523
        : "+r" (i)
524
        : "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg)w-15)
Loren Merritt's avatar
Loren Merritt committed
525 526 527 528 529
    );
    for(; i<w; i++)
        dst[i] = src1[i] + src2[i];
}

Michael Niedermayer's avatar
Michael Niedermayer committed
530
#define H263_LOOP_FILTER \
531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599
        "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
600

Michael Niedermayer's avatar
Michael Niedermayer committed
601
static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
602
    if(ENABLE_ANY_H263) {
Michael Niedermayer's avatar
Michael Niedermayer committed
603 604
    const int strength= ff_h263_loop_filter_strength[qscale];

605
    __asm__ volatile(
606

Michael Niedermayer's avatar
Michael Niedermayer committed
607
        H263_LOOP_FILTER
608

609 610 611 612
        "movq %%mm3, %1                 \n\t"
        "movq %%mm4, %2                 \n\t"
        "movq %%mm5, %0                 \n\t"
        "movq %%mm6, %3                 \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
613 614 615 616 617 618
        : "+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)
    );
619
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
620 621
}

Michael Niedermayer's avatar
Michael Niedermayer committed
622
static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){
623
    __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...
624 625 626 627 628 629 630 631 632 633 634 635 636 637 638
        "movd  %4, %%mm0                \n\t"
        "movd  %5, %%mm1                \n\t"
        "movd  %6, %%mm2                \n\t"
        "movd  %7, %%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"
        "punpckhdq %%mm0, %%mm0         \n\t"
        "movd  %%mm0, %1                \n\t"
        "movd  %%mm1, %2                \n\t"
        "punpckhdq %%mm1, %%mm1         \n\t"
        "movd  %%mm1, %3                \n\t"
639

Michael Niedermayer's avatar
Michael Niedermayer committed
640 641 642 643 644 645 646 647 648 649 650 651
        : "=m" (*(uint32_t*)(dst + 0*dst_stride)),
          "=m" (*(uint32_t*)(dst + 1*dst_stride)),
          "=m" (*(uint32_t*)(dst + 2*dst_stride)),
          "=m" (*(uint32_t*)(dst + 3*dst_stride))
        :  "m" (*(uint32_t*)(src + 0*src_stride)),
           "m" (*(uint32_t*)(src + 1*src_stride)),
           "m" (*(uint32_t*)(src + 2*src_stride)),
           "m" (*(uint32_t*)(src + 3*src_stride))
    );
}

static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
652
    if(ENABLE_ANY_H263) {
Michael Niedermayer's avatar
Michael Niedermayer committed
653
    const int strength= ff_h263_loop_filter_strength[qscale];
Reimar Döffinger's avatar
Reimar Döffinger committed
654
    DECLARE_ALIGNED(8, uint64_t, temp[4]);
Michael Niedermayer's avatar
Michael Niedermayer committed
655
    uint8_t *btemp= (uint8_t*)temp;
656

Michael Niedermayer's avatar
Michael Niedermayer committed
657 658 659 660
    src -= 2;

    transpose4x4(btemp  , src           , 8, stride);
    transpose4x4(btemp+4, src + 4*stride, 8, stride);
661
    __asm__ volatile(
Michael Niedermayer's avatar
Michael Niedermayer committed
662
        H263_LOOP_FILTER // 5 3 4 6
663

Michael Niedermayer's avatar
Michael Niedermayer committed
664 665 666 667 668 669 670
        : "+m" (temp[0]),
          "+m" (temp[1]),
          "+m" (temp[2]),
          "+m" (temp[3])
        : "g" (2*strength), "m"(ff_pb_FC)
    );

671
    __asm__ volatile(
672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695
        "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"
696 697
        :: "r" (src),
           "r" (src + 4*stride),
698 699
           "r" ((x86_reg)   stride ),
           "r" ((x86_reg)(3*stride))
Michael Niedermayer's avatar
Michael Niedermayer committed
700
    );
701
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
702 703
}

704 705 706 707 708 709 710 711 712 713 714 715
/* 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)
{
    uint8_t *ptr, *last_line;
    int i;

    last_line = buf + (height - 1) * wrap;
    /* left and right */
    ptr = buf;
    if(w==8)
    {
716
        __asm__ volatile(
717 718 719 720 721 722 723 724 725 726 727 728 729 730 731
                "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)
732
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
733 734 735 736
        );
    }
    else
    {
737
        __asm__ volatile(
738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754
                "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)
755
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
756 757 758 759 760 761
        );
    }

    for(i=0;i<w;i+=4) {
        /* top and bottom (and hopefully also the corners) */
        ptr= buf - (i + 1) * wrap - w;
762
        __asm__ volatile(
763 764 765 766 767 768 769 770 771 772
                "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)
773
                : "r" ((x86_reg)buf - (x86_reg)ptr - w), "r" ((x86_reg)-wrap), "r" ((x86_reg)-wrap*3), "r" (ptr+width+2*w)
774 775
        );
        ptr= last_line + (i + 1) * wrap - w;
776
        __asm__ volatile(
777 778 779 780 781 782 783 784 785 786
                "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)
787
                : "r" ((x86_reg)last_line - (x86_reg)ptr - w), "r" ((x86_reg)wrap), "r" ((x86_reg)wrap*3), "r" (ptr+width+2*w)
788 789 790 791
        );
    }
}

Loren Merritt's avatar
Loren Merritt committed
792
#define PAETH(cpu, abs3)\
793
static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\
Loren Merritt's avatar
Loren Merritt committed
794
{\
795 796
    x86_reg i = -bpp;\
    x86_reg end = w-3;\
797
    __asm__ volatile(\
Loren Merritt's avatar
Loren Merritt committed
798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838
        "pxor      %%mm7, %%mm7 \n"\
        "movd    (%1,%0), %%mm0 \n"\
        "movd    (%2,%0), %%mm1 \n"\
        "punpcklbw %%mm7, %%mm0 \n"\
        "punpcklbw %%mm7, %%mm1 \n"\
        "add       %4, %0 \n"\
        "1: \n"\
        "movq      %%mm1, %%mm2 \n"\
        "movd    (%2,%0), %%mm1 \n"\
        "movq      %%mm2, %%mm3 \n"\
        "punpcklbw %%mm7, %%mm1 \n"\
        "movq      %%mm2, %%mm4 \n"\
        "psubw     %%mm1, %%mm3 \n"\
        "psubw     %%mm0, %%mm4 \n"\
        "movq      %%mm3, %%mm5 \n"\
        "paddw     %%mm4, %%mm5 \n"\
        abs3\
        "movq      %%mm4, %%mm6 \n"\
        "pminsw    %%mm5, %%mm6 \n"\
        "pcmpgtw   %%mm6, %%mm3 \n"\
        "pcmpgtw   %%mm5, %%mm4 \n"\
        "movq      %%mm4, %%mm6 \n"\
        "pand      %%mm3, %%mm4 \n"\
        "pandn     %%mm3, %%mm6 \n"\
        "pandn     %%mm0, %%mm3 \n"\
        "movd    (%3,%0), %%mm0 \n"\
        "pand      %%mm1, %%mm6 \n"\
        "pand      %%mm4, %%mm2 \n"\
        "punpcklbw %%mm7, %%mm0 \n"\
        "movq      %6,    %%mm5 \n"\
        "paddw     %%mm6, %%mm0 \n"\
        "paddw     %%mm2, %%mm3 \n"\
        "paddw     %%mm3, %%mm0 \n"\
        "pand      %%mm5, %%mm0 \n"\
        "movq      %%mm0, %%mm3 \n"\
        "packuswb  %%mm3, %%mm3 \n"\
        "movd      %%mm3, (%1,%0) \n"\
        "add       %4, %0 \n"\
        "cmp       %5, %0 \n"\
        "jle 1b \n"\
        :"+r"(i)\
839
        :"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\
Loren Merritt's avatar
Loren Merritt committed
840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865
         "m"(ff_pw_255)\
        :"memory"\
    );\
}

#define ABS3_MMX2\
        "psubw     %%mm5, %%mm7 \n"\
        "pmaxsw    %%mm7, %%mm5 \n"\
        "pxor      %%mm6, %%mm6 \n"\
        "pxor      %%mm7, %%mm7 \n"\
        "psubw     %%mm3, %%mm6 \n"\
        "psubw     %%mm4, %%mm7 \n"\
        "pmaxsw    %%mm6, %%mm3 \n"\
        "pmaxsw    %%mm7, %%mm4 \n"\
        "pxor      %%mm7, %%mm7 \n"

#define ABS3_SSSE3\
        "pabsw     %%mm3, %%mm3 \n"\
        "pabsw     %%mm4, %%mm4 \n"\
        "pabsw     %%mm5, %%mm5 \n"

PAETH(mmx2, ABS3_MMX2)
#ifdef HAVE_SSSE3
PAETH(ssse3, ABS3_SSSE3)
#endif

Michael Niedermayer's avatar
Michael Niedermayer committed
866
#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884
        "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"\
Michael Niedermayer's avatar
Michael Niedermayer committed
885 886
        OP(%%mm5, out, %%mm7, d)

887
#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
888
static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
Michael Niedermayer's avatar
Michael Niedermayer committed
889 890
    uint64_t temp;\
\
891
    __asm__ volatile(\
892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922
        "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"\
Michael Niedermayer's avatar
Michael Niedermayer committed
923 924
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
        \
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
        "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"\
950
        OP_MMX2(%%mm1, (%1),%%mm4, q)\
Michael Niedermayer's avatar
Michael Niedermayer committed
951 952
        /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
        \
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
        "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"\
Michael Niedermayer's avatar
Michael Niedermayer committed
978 979
        /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
        \
980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995
        "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"\
996
        OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
Michael Niedermayer's avatar
Michael Niedermayer committed
997
        \
998 999 1000 1001
        "add %3, %0                       \n\t"\
        "add %4, %1                       \n\t"\
        "decl %2                          \n\t"\
        " jnz 1b                          \n\t"\
1002
        : "+a"(src), "+c"(dst), "+D"(h)\
1003
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
Michael Niedermayer's avatar
Michael Niedermayer committed
1004
        : "memory"\
Michael Niedermayer's avatar
Michael Niedermayer committed
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
    );\
}\
\
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]);\
1030
        __asm__ volatile(\
1031 1032 1033 1034 1035 1036 1037
            "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"\
1038
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1039 1040 1041 1042 1043 1044 1045
            "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"\
1046
            OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
Michael Niedermayer's avatar
Michael Niedermayer committed
1047
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
Michael Niedermayer's avatar
Michael Niedermayer committed
1048
            : "memory"\
Michael Niedermayer's avatar
Michael Niedermayer committed
1049 1050 1051 1052 1053 1054
        );\
        dst+=dstStride;\
        src+=srcStride;\
    }\
}\
\
1055
static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1056
    __asm__ volatile(\
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
        "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 */\
Michael Niedermayer's avatar
Michael Niedermayer committed
1084
        "paddw %5, %%mm6                  \n\t"\
1085 1086
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
        "psraw $5, %%mm0                  \n\t"\
Michael Niedermayer's avatar
Michael Niedermayer committed
1087 1088
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
        \
1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
        "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 */\
Michael Niedermayer's avatar
Michael Niedermayer committed
1103
        "paddw %5, %%mm1                  \n\t"\
1104 1105 1106
        "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b + 3c - d */\
        "psraw $5, %%mm3                  \n\t"\
        "packuswb %%mm3, %%mm0            \n\t"\
1107
        OP_MMX2(%%mm0, (%1), %%mm4, q)\
Michael Niedermayer's avatar
Michael Niedermayer committed
1108
        \
1109 1110 1111 1112
        "add %3, %0                       \n\t"\
        "add %4, %1                       \n\t"\
        "decl %2                          \n\t"\
        " jnz 1b                          \n\t"\
1113
        : "+a"(src), "+c"(dst), "+d"(h)\
1114
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER)\
Michael Niedermayer's avatar
Michael Niedermayer committed
1115
        : "memory"\
Michael Niedermayer's avatar
Michael Niedermayer committed
1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132
    );\
}\
\
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]);\
1133
        __asm__ volatile(\