Newer
Older
"movq %%mm1, %%mm4 \n\t"\
"movq %%mm2, %%mm5 \n\t"\
"psubusb %%mm1, %%mm3 \n\t"\
"psubusb %%mm2, %%mm4 \n\t"\
"psubusb %%mm0, %%mm5 \n\t"\
"pcmpeqb %%mm7, %%mm3 \n\t"\
"pcmpeqb %%mm7, %%mm4 \n\t"\
"pcmpeqb %%mm7, %%mm5 \n\t"\
"movq %%mm3, %%mm6 \n\t"\
"pxor %%mm4, %%mm3 \n\t"\
"pxor %%mm5, %%mm4 \n\t"\
"pxor %%mm6, %%mm5 \n\t"\
"por %%mm3, %%mm1 \n\t"\
"por %%mm4, %%mm2 \n\t"\
"por %%mm5, %%mm0 \n\t"\
"pand %%mm2, %%mm0 \n\t"\
"pand %%mm1, %%mm0 \n\t"\
"movq %%mm0, " #b " \n\t"
Aurelien Jacobs
committed
#define MEDIAN(a,b,c) REAL_MEDIAN(a,b,c)
MEDIAN((%0) , (%%REGa) , (%%REGa, %1))
Aurelien Jacobs
committed
MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
MEDIAN((%0, %1, 4) , (%%REGd) , (%%REGd, %1))
Aurelien Jacobs
committed
MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
: : "r" (src), "r" ((long)stride)
: "%"REG_a, "%"REG_d
);
#endif //HAVE_MMX2
#else //HAVE_MMX
int x, y;
src+= 4*stride;
// FIXME - there should be a way to do a few columns in parallel like w/mmx
for(x=0; x<8; x++)
{
uint8_t *colsrc = src;
for (y=0; y<4; y++)
{
int a, b, c, d, e, f;
a = colsrc[0 ];
b = colsrc[stride ];
c = colsrc[stride*2];
d = (a-b)>>31;
e = (b-c)>>31;
f = (c-a)>>31;
colsrc[stride ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
colsrc += stride*2;
}
src++;
}
#endif //HAVE_MMX
Michael Niedermayer
committed
}
Michael Niedermayer
committed
#ifdef HAVE_MMX
/**
* transposes and shift the given 8x8 Block into dst1 and dst2
*/
static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
{
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
asm(
"lea (%0, %1), %%"REG_a" \n\t"
// 0 1 2 3 4 5 6 7 8 9
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
"movq (%0), %%mm0 \n\t" // 12345678
"movq (%%"REG_a"), %%mm1 \n\t" // abcdefgh
"movq %%mm0, %%mm2 \n\t" // 12345678
"punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
"punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
"movq (%%"REG_a", %1), %%mm1 \n\t"
"movq (%%"REG_a", %1, 2), %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"punpcklbw %%mm3, %%mm1 \n\t"
"punpckhbw %%mm3, %%mm4 \n\t"
"movq %%mm0, %%mm3 \n\t"
"punpcklwd %%mm1, %%mm0 \n\t"
"punpckhwd %%mm1, %%mm3 \n\t"
"movq %%mm2, %%mm1 \n\t"
"punpcklwd %%mm4, %%mm2 \n\t"
"punpckhwd %%mm4, %%mm1 \n\t"
"movd %%mm0, 128(%2) \n\t"
"psrlq $32, %%mm0 \n\t"
"movd %%mm0, 144(%2) \n\t"
"movd %%mm3, 160(%2) \n\t"
"psrlq $32, %%mm3 \n\t"
"movd %%mm3, 176(%2) \n\t"
"movd %%mm3, 48(%3) \n\t"
"movd %%mm2, 192(%2) \n\t"
"movd %%mm2, 64(%3) \n\t"
"psrlq $32, %%mm2 \n\t"
"movd %%mm2, 80(%3) \n\t"
"movd %%mm1, 96(%3) \n\t"
"psrlq $32, %%mm1 \n\t"
"movd %%mm1, 112(%3) \n\t"
"lea (%%"REG_a", %1, 4), %%"REG_a" \n\t"
"movq (%0, %1, 4), %%mm0 \n\t" // 12345678
"movq (%%"REG_a"), %%mm1 \n\t" // abcdefgh
"movq %%mm0, %%mm2 \n\t" // 12345678
"punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
"punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
"movq (%%"REG_a", %1), %%mm1 \n\t"
"movq (%%"REG_a", %1, 2), %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"punpcklbw %%mm3, %%mm1 \n\t"
"punpckhbw %%mm3, %%mm4 \n\t"
"movq %%mm0, %%mm3 \n\t"
"punpcklwd %%mm1, %%mm0 \n\t"
"punpckhwd %%mm1, %%mm3 \n\t"
"movq %%mm2, %%mm1 \n\t"
"punpcklwd %%mm4, %%mm2 \n\t"
"punpckhwd %%mm4, %%mm1 \n\t"
"movd %%mm0, 132(%2) \n\t"
"psrlq $32, %%mm0 \n\t"
"movd %%mm0, 148(%2) \n\t"
"movd %%mm3, 164(%2) \n\t"
"psrlq $32, %%mm3 \n\t"
"movd %%mm3, 180(%2) \n\t"
"movd %%mm3, 52(%3) \n\t"
"movd %%mm2, 196(%2) \n\t"
"movd %%mm2, 68(%3) \n\t"
"psrlq $32, %%mm2 \n\t"
"movd %%mm2, 84(%3) \n\t"
"movd %%mm1, 100(%3) \n\t"
"psrlq $32, %%mm1 \n\t"
"movd %%mm1, 116(%3) \n\t"
:: "r" (src), "r" ((long)srcStride), "r" (dst1), "r" (dst2)
: "%"REG_a
);
}
/**
* transposes the given 8x8 block
*/
static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
{
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
asm(
"lea (%0, %1), %%"REG_a" \n\t"
"lea (%%"REG_a",%1,4), %%"REG_d" \n\t"
// 0 1 2 3 4 5 6 7 8 9
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
"movq (%2), %%mm0 \n\t" // 12345678
"movq 16(%2), %%mm1 \n\t" // abcdefgh
"movq %%mm0, %%mm2 \n\t" // 12345678
"punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
"punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
"movq 32(%2), %%mm1 \n\t"
"movq 48(%2), %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"punpcklbw %%mm3, %%mm1 \n\t"
"punpckhbw %%mm3, %%mm4 \n\t"
"movq %%mm0, %%mm3 \n\t"
"punpcklwd %%mm1, %%mm0 \n\t"
"punpckhwd %%mm1, %%mm3 \n\t"
"movq %%mm2, %%mm1 \n\t"
"punpcklwd %%mm4, %%mm2 \n\t"
"punpckhwd %%mm4, %%mm1 \n\t"
"movd %%mm0, (%0) \n\t"
"psrlq $32, %%mm0 \n\t"
"movd %%mm0, (%%"REG_a") \n\t"
"movd %%mm3, (%%"REG_a", %1) \n\t"
"psrlq $32, %%mm3 \n\t"
"movd %%mm3, (%%"REG_a", %1, 2) \n\t"
"movd %%mm2, (%0, %1, 4) \n\t"
"psrlq $32, %%mm2 \n\t"
"movd %%mm2, (%%"REG_d") \n\t"
"movd %%mm1, (%%"REG_d", %1) \n\t"
"psrlq $32, %%mm1 \n\t"
"movd %%mm1, (%%"REG_d", %1, 2) \n\t"
"movq 64(%2), %%mm0 \n\t" // 12345678
"movq 80(%2), %%mm1 \n\t" // abcdefgh
"movq %%mm0, %%mm2 \n\t" // 12345678
"punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
"punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
"movq 96(%2), %%mm1 \n\t"
"movq 112(%2), %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"punpcklbw %%mm3, %%mm1 \n\t"
"punpckhbw %%mm3, %%mm4 \n\t"
"movq %%mm0, %%mm3 \n\t"
"punpcklwd %%mm1, %%mm0 \n\t"
"punpckhwd %%mm1, %%mm3 \n\t"
"movq %%mm2, %%mm1 \n\t"
"punpcklwd %%mm4, %%mm2 \n\t"
"punpckhwd %%mm4, %%mm1 \n\t"
"movd %%mm0, 4(%0) \n\t"
"psrlq $32, %%mm0 \n\t"
"movd %%mm0, 4(%%"REG_a") \n\t"
"movd %%mm3, 4(%%"REG_a", %1) \n\t"
"psrlq $32, %%mm3 \n\t"
"movd %%mm3, 4(%%"REG_a", %1, 2) \n\t"
"movd %%mm2, 4(%0, %1, 4) \n\t"
"psrlq $32, %%mm2 \n\t"
"movd %%mm2, 4(%%"REG_d") \n\t"
"movd %%mm1, 4(%%"REG_d", %1) \n\t"
"psrlq $32, %%mm1 \n\t"
"movd %%mm1, 4(%%"REG_d", %1, 2) \n\t"
:: "r" (dst), "r" ((long)dstStride), "r" (src)
: "%"REG_a, "%"REG_d
);
}
#endif //HAVE_MMX
Aurelien Jacobs
committed
//static long test=0;
static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
// to save a register (FIXME do this outside of the loops)
tempBluredPast[127]= maxNoise[0];
tempBluredPast[128]= maxNoise[1];
tempBluredPast[129]= maxNoise[2];
Michael Niedermayer
committed
#define FAST_L2_DIFF
//#define L1_DIFF //u should change the thresholds too if u try that one
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
asm volatile(
"lea (%2, %2, 2), %%"REG_a" \n\t" // 3*stride
"lea (%2, %2, 4), %%"REG_d" \n\t" // 5*stride
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
// 0 1 2 3 4 5 6 7 8 9
// %x %x+%2 %x+2%2 %x+eax %x+4%2 %x+edx %x+2eax %x+ecx %x+8%2
Michael Niedermayer
committed
//FIXME reorder?
#ifdef L1_DIFF //needs mmx2
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
"movq (%0), %%mm0 \n\t" // L0
"psadbw (%1), %%mm0 \n\t" // |L0-R0|
"movq (%0, %2), %%mm1 \n\t" // L1
"psadbw (%1, %2), %%mm1 \n\t" // |L1-R1|
"movq (%0, %2, 2), %%mm2 \n\t" // L2
"psadbw (%1, %2, 2), %%mm2 \n\t" // |L2-R2|
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
"psadbw (%1, %%"REG_a"), %%mm3 \n\t" // |L3-R3|
"movq (%0, %2, 4), %%mm4 \n\t" // L4
"paddw %%mm1, %%mm0 \n\t"
"psadbw (%1, %2, 4), %%mm4 \n\t" // |L4-R4|
"movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
"paddw %%mm2, %%mm0 \n\t"
"psadbw (%1, %%"REG_d"), %%mm5 \n\t" // |L5-R5|
"movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
"paddw %%mm3, %%mm0 \n\t"
"psadbw (%1, %%"REG_a", 2), %%mm6 \n\t" // |L6-R6|
"movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
"paddw %%mm4, %%mm0 \n\t"
"psadbw (%1, %%"REG_c"), %%mm7 \n\t" // |L7-R7|
"paddw %%mm5, %%mm6 \n\t"
"paddw %%mm7, %%mm6 \n\t"
"paddw %%mm6, %%mm0 \n\t"
#else //L1_DIFF
Aurelien Jacobs
committed
#if defined (FAST_L2_DIFF)
"pcmpeqb %%mm7, %%mm7 \n\t"
"movq "MANGLE(b80)", %%mm6 \n\t"
"pxor %%mm0, %%mm0 \n\t"
Aurelien Jacobs
committed
#define REAL_L2_DIFF_CORE(a, b)\
"movq " #a ", %%mm5 \n\t"\
"movq " #b ", %%mm2 \n\t"\
"pxor %%mm7, %%mm2 \n\t"\
PAVGB(%%mm2, %%mm5)\
"paddb %%mm6, %%mm5 \n\t"\
"movq %%mm5, %%mm2 \n\t"\
"psllw $8, %%mm5 \n\t"\
"pmaddwd %%mm5, %%mm5 \n\t"\
"pmaddwd %%mm2, %%mm2 \n\t"\
"paddd %%mm2, %%mm5 \n\t"\
"psrld $14, %%mm5 \n\t"\
"paddd %%mm5, %%mm0 \n\t"
Michael Niedermayer
committed
#else //defined (FAST_L2_DIFF)
"pxor %%mm7, %%mm7 \n\t"
"pxor %%mm0, %%mm0 \n\t"
Aurelien Jacobs
committed
#define REAL_L2_DIFF_CORE(a, b)\
"movq " #a ", %%mm5 \n\t"\
"movq " #b ", %%mm2 \n\t"\
"movq %%mm5, %%mm1 \n\t"\
"movq %%mm2, %%mm3 \n\t"\
"punpcklbw %%mm7, %%mm5 \n\t"\
"punpckhbw %%mm7, %%mm1 \n\t"\
"punpcklbw %%mm7, %%mm2 \n\t"\
"punpckhbw %%mm7, %%mm3 \n\t"\
"psubw %%mm2, %%mm5 \n\t"\
"psubw %%mm3, %%mm1 \n\t"\
"pmaddwd %%mm5, %%mm5 \n\t"\
"pmaddwd %%mm1, %%mm1 \n\t"\
"paddd %%mm1, %%mm5 \n\t"\
"paddd %%mm5, %%mm0 \n\t"
Michael Niedermayer
committed
#endif //defined (FAST_L2_DIFF)
Aurelien Jacobs
committed
#define L2_DIFF_CORE(a, b) REAL_L2_DIFF_CORE(a, b)
L2_DIFF_CORE((%0) , (%1))
L2_DIFF_CORE((%0, %2) , (%1, %2))
L2_DIFF_CORE((%0, %2, 2) , (%1, %2, 2))
L2_DIFF_CORE((%0, %%REGa) , (%1, %%REGa))
L2_DIFF_CORE((%0, %2, 4) , (%1, %2, 4))
L2_DIFF_CORE((%0, %%REGd) , (%1, %%REGd))
Aurelien Jacobs
committed
L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))
L2_DIFF_CORE((%0, %%REGc) , (%1, %%REGc))
Michael Niedermayer
committed
#endif //L1_DIFF
Michael Niedermayer
committed
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
"movq %%mm0, %%mm4 \n\t"
"psrlq $32, %%mm0 \n\t"
"paddd %%mm0, %%mm4 \n\t"
"movd %%mm4, %%ecx \n\t"
"shll $2, %%ecx \n\t"
"mov %3, %%"REG_d" \n\t"
"addl -4(%%"REG_d"), %%ecx \n\t"
"addl 4(%%"REG_d"), %%ecx \n\t"
"addl -1024(%%"REG_d"), %%ecx \n\t"
"addl $4, %%ecx \n\t"
"addl 1024(%%"REG_d"), %%ecx \n\t"
"shrl $3, %%ecx \n\t"
"movl %%ecx, (%%"REG_d") \n\t"
// "mov %3, %%"REG_c" \n\t"
// "mov %%"REG_c", test \n\t"
// "jmp 4f \n\t"
"cmpl 512(%%"REG_d"), %%ecx \n\t"
" jb 2f \n\t"
"cmpl 516(%%"REG_d"), %%ecx \n\t"
" jb 1f \n\t"
"lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
"movq (%0), %%mm0 \n\t" // L0
"movq (%0, %2), %%mm1 \n\t" // L1
"movq (%0, %2, 2), %%mm2 \n\t" // L2
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
"movq (%0, %2, 4), %%mm4 \n\t" // L4
"movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
"movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
"movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
"movq %%mm0, (%1) \n\t" // L0
"movq %%mm1, (%1, %2) \n\t" // L1
"movq %%mm2, (%1, %2, 2) \n\t" // L2
"movq %%mm3, (%1, %%"REG_a") \n\t" // L3
"movq %%mm4, (%1, %2, 4) \n\t" // L4
"movq %%mm5, (%1, %%"REG_d") \n\t" // L5
"movq %%mm6, (%1, %%"REG_a", 2) \n\t" // L6
"movq %%mm7, (%1, %%"REG_c") \n\t" // L7
"jmp 4f \n\t"
"1: \n\t"
"lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
"movq (%0), %%mm0 \n\t" // L0
PAVGB((%1), %%mm0) // L0
"movq (%0, %2), %%mm1 \n\t" // L1
PAVGB((%1, %2), %%mm1) // L1
"movq (%0, %2, 2), %%mm2 \n\t" // L2
PAVGB((%1, %2, 2), %%mm2) // L2
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
PAVGB((%1, %%REGa), %%mm3) // L3
"movq (%0, %2, 4), %%mm4 \n\t" // L4
PAVGB((%1, %2, 4), %%mm4) // L4
"movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
PAVGB((%1, %%REGd), %%mm5) // L5
"movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
PAVGB((%1, %%REGa, 2), %%mm6) // L6
"movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
PAVGB((%1, %%REGc), %%mm7) // L7
"movq %%mm0, (%1) \n\t" // R0
"movq %%mm1, (%1, %2) \n\t" // R1
"movq %%mm2, (%1, %2, 2) \n\t" // R2
"movq %%mm3, (%1, %%"REG_a") \n\t" // R3
"movq %%mm4, (%1, %2, 4) \n\t" // R4
"movq %%mm5, (%1, %%"REG_d") \n\t" // R5
"movq %%mm6, (%1, %%"REG_a", 2) \n\t" // R6
"movq %%mm7, (%1, %%"REG_c") \n\t" // R7
"movq %%mm0, (%0) \n\t" // L0
"movq %%mm1, (%0, %2) \n\t" // L1
"movq %%mm2, (%0, %2, 2) \n\t" // L2
"movq %%mm3, (%0, %%"REG_a") \n\t" // L3
"movq %%mm4, (%0, %2, 4) \n\t" // L4
"movq %%mm5, (%0, %%"REG_d") \n\t" // L5
"movq %%mm6, (%0, %%"REG_a", 2) \n\t" // L6
"movq %%mm7, (%0, %%"REG_c") \n\t" // L7
"jmp 4f \n\t"
"2: \n\t"
"cmpl 508(%%"REG_d"), %%ecx \n\t"
" jb 3f \n\t"
"lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
"movq (%0), %%mm0 \n\t" // L0
"movq (%0, %2), %%mm1 \n\t" // L1
"movq (%0, %2, 2), %%mm2 \n\t" // L2
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
"movq (%1), %%mm4 \n\t" // R0
"movq (%1, %2), %%mm5 \n\t" // R1
"movq (%1, %2, 2), %%mm6 \n\t" // R2
"movq (%1, %%"REG_a"), %%mm7 \n\t" // R3
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
"movq %%mm0, (%1) \n\t" // R0
"movq %%mm1, (%1, %2) \n\t" // R1
"movq %%mm2, (%1, %2, 2) \n\t" // R2
"movq %%mm3, (%1, %%"REG_a") \n\t" // R3
"movq %%mm0, (%0) \n\t" // L0
"movq %%mm1, (%0, %2) \n\t" // L1
"movq %%mm2, (%0, %2, 2) \n\t" // L2
"movq %%mm3, (%0, %%"REG_a") \n\t" // L3
"movq (%0, %2, 4), %%mm0 \n\t" // L4
"movq (%0, %%"REG_d"), %%mm1 \n\t" // L5
"movq (%0, %%"REG_a", 2), %%mm2 \n\t" // L6
"movq (%0, %%"REG_c"), %%mm3 \n\t" // L7
"movq (%1, %2, 4), %%mm4 \n\t" // R4
"movq (%1, %%"REG_d"), %%mm5 \n\t" // R5
"movq (%1, %%"REG_a", 2), %%mm6 \n\t" // R6
"movq (%1, %%"REG_c"), %%mm7 \n\t" // R7
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
"movq %%mm0, (%1, %2, 4) \n\t" // R4
"movq %%mm1, (%1, %%"REG_d") \n\t" // R5
"movq %%mm2, (%1, %%"REG_a", 2) \n\t" // R6
"movq %%mm3, (%1, %%"REG_c") \n\t" // R7
"movq %%mm0, (%0, %2, 4) \n\t" // L4
"movq %%mm1, (%0, %%"REG_d") \n\t" // L5
"movq %%mm2, (%0, %%"REG_a", 2) \n\t" // L6
"movq %%mm3, (%0, %%"REG_c") \n\t" // L7
"jmp 4f \n\t"
"3: \n\t"
"lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
"movq (%0), %%mm0 \n\t" // L0
"movq (%0, %2), %%mm1 \n\t" // L1
"movq (%0, %2, 2), %%mm2 \n\t" // L2
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
"movq (%1), %%mm4 \n\t" // R0
"movq (%1, %2), %%mm5 \n\t" // R1
"movq (%1, %2, 2), %%mm6 \n\t" // R2
"movq (%1, %%"REG_a"), %%mm7 \n\t" // R3
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
"movq %%mm0, (%1) \n\t" // R0
"movq %%mm1, (%1, %2) \n\t" // R1
"movq %%mm2, (%1, %2, 2) \n\t" // R2
"movq %%mm3, (%1, %%"REG_a") \n\t" // R3
"movq %%mm0, (%0) \n\t" // L0
"movq %%mm1, (%0, %2) \n\t" // L1
"movq %%mm2, (%0, %2, 2) \n\t" // L2
"movq %%mm3, (%0, %%"REG_a") \n\t" // L3
"movq (%0, %2, 4), %%mm0 \n\t" // L4
"movq (%0, %%"REG_d"), %%mm1 \n\t" // L5
"movq (%0, %%"REG_a", 2), %%mm2 \n\t" // L6
"movq (%0, %%"REG_c"), %%mm3 \n\t" // L7
"movq (%1, %2, 4), %%mm4 \n\t" // R4
"movq (%1, %%"REG_d"), %%mm5 \n\t" // R5
"movq (%1, %%"REG_a", 2), %%mm6 \n\t" // R6
"movq (%1, %%"REG_c"), %%mm7 \n\t" // R7
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
PAVGB(%%mm4, %%mm0)
PAVGB(%%mm5, %%mm1)
PAVGB(%%mm6, %%mm2)
PAVGB(%%mm7, %%mm3)
"movq %%mm0, (%1, %2, 4) \n\t" // R4
"movq %%mm1, (%1, %%"REG_d") \n\t" // R5
"movq %%mm2, (%1, %%"REG_a", 2) \n\t" // R6
"movq %%mm3, (%1, %%"REG_c") \n\t" // R7
"movq %%mm0, (%0, %2, 4) \n\t" // L4
"movq %%mm1, (%0, %%"REG_d") \n\t" // L5
"movq %%mm2, (%0, %%"REG_a", 2) \n\t" // L6
"movq %%mm3, (%0, %%"REG_c") \n\t" // L7
"4: \n\t"
:: "r" (src), "r" (tempBlured), "r"((long)stride), "m" (tempBluredPast)
: "%"REG_a, "%"REG_d, "%"REG_c, "memory"
);
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
int y;
int d=0;
// int sysd=0;
int i;
for(y=0; y<8; y++)
{
int x;
for(x=0; x<8; x++)
{
int ref= tempBlured[ x + y*stride ];
int cur= src[ x + y*stride ];
int d1=ref - cur;
// if(x==0 || x==7) d1+= d1>>1;
// if(y==0 || y==7) d1+= d1>>1;
d+= d1*d1;
// sysd+= d1;
}
}
i=d;
d= (
4*d
+(*(tempBluredPast-256))
+(*(tempBluredPast-1))+ (*(tempBluredPast+1))
+(*(tempBluredPast+256))
+4)>>3;
*tempBluredPast=i;
// ((*tempBluredPast)*3 + d + 2)>>2;
/*
Switch between
1 0 0 0 0 0 0 (0)
64 32 16 8 4 2 1 (1)
64 48 36 27 20 15 11 (33) (approx)
64 56 49 43 37 33 29 (200) (approx)
*/
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
if(d > maxNoise[1])
{
if(d < maxNoise[2])
{
for(y=0; y<8; y++)
{
int x;
for(x=0; x<8; x++)
{
int ref= tempBlured[ x + y*stride ];
int cur= src[ x + y*stride ];
tempBlured[ x + y*stride ]=
src[ x + y*stride ]=
(ref + cur + 1)>>1;
}
}
}
else
{
for(y=0; y<8; y++)
{
int x;
for(x=0; x<8; x++)
{
tempBlured[ x + y*stride ]= src[ x + y*stride ];
}
}
}
}
else
{
if(d < maxNoise[0])
{
for(y=0; y<8; y++)
{
int x;
for(x=0; x<8; x++)
{
int ref= tempBlured[ x + y*stride ];
int cur= src[ x + y*stride ];
tempBlured[ x + y*stride ]=
src[ x + y*stride ]=
(ref*7 + cur + 4)>>3;
}
}
}
else
{
for(y=0; y<8; y++)
{
int x;
for(x=0; x<8; x++)
{
int ref= tempBlured[ x + y*stride ];
int cur= src[ x + y*stride ];
tempBlured[ x + y*stride ]=
src[ x + y*stride ]=
(ref*3 + cur + 2)>>2;
}
}
}
}
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
#endif //HAVE_ALTIVEC
#ifdef HAVE_MMX
/**
* accurate deblock filter
*/
static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
int64_t dc_mask, eq_mask, both_masks;
int64_t sums[10*8*2];
src+= step*3; // src points to begin of the 8x8 Block
//START_TIMER
asm volatile(
"movq %0, %%mm7 \n\t"
"movq %1, %%mm6 \n\t"
: : "m" (c->mmxDcOffset[c->nonBQP]), "m" (c->mmxDcThreshold[c->nonBQP])
);
asm volatile(
"lea (%2, %3), %%"REG_a" \n\t"
// 0 1 2 3 4 5 6 7 8 9
// %1 eax eax+%2 eax+2%2 %1+4%2 ecx ecx+%2 ecx+2%2 %1+8%2 ecx+4%2
"movq (%2), %%mm0 \n\t"
"movq (%%"REG_a"), %%mm1 \n\t"
"movq %%mm1, %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
"paddb %%mm7, %%mm0 \n\t"
"pcmpgtb %%mm6, %%mm0 \n\t"
"movq (%%"REG_a",%3), %%mm2 \n\t"
PMAXUB(%%mm2, %%mm4)
PMINUB(%%mm2, %%mm3, %%mm5)
"psubb %%mm2, %%mm1 \n\t"
"paddb %%mm7, %%mm1 \n\t"
"pcmpgtb %%mm6, %%mm1 \n\t"
"paddb %%mm1, %%mm0 \n\t"
"movq (%%"REG_a", %3, 2), %%mm1 \n\t"
PMAXUB(%%mm1, %%mm4)
PMINUB(%%mm1, %%mm3, %%mm5)
"psubb %%mm1, %%mm2 \n\t"
"paddb %%mm7, %%mm2 \n\t"
"pcmpgtb %%mm6, %%mm2 \n\t"
"paddb %%mm2, %%mm0 \n\t"
"lea (%%"REG_a", %3, 4), %%"REG_a" \n\t"
"movq (%2, %3, 4), %%mm2 \n\t"
PMAXUB(%%mm2, %%mm4)
PMINUB(%%mm2, %%mm3, %%mm5)
"psubb %%mm2, %%mm1 \n\t"
"paddb %%mm7, %%mm1 \n\t"
"pcmpgtb %%mm6, %%mm1 \n\t"
"paddb %%mm1, %%mm0 \n\t"
"movq (%%"REG_a"), %%mm1 \n\t"
PMAXUB(%%mm1, %%mm4)
PMINUB(%%mm1, %%mm3, %%mm5)
"psubb %%mm1, %%mm2 \n\t"
"paddb %%mm7, %%mm2 \n\t"
"pcmpgtb %%mm6, %%mm2 \n\t"
"paddb %%mm2, %%mm0 \n\t"
"movq (%%"REG_a", %3), %%mm2 \n\t"
PMAXUB(%%mm2, %%mm4)
PMINUB(%%mm2, %%mm3, %%mm5)
"psubb %%mm2, %%mm1 \n\t"
"paddb %%mm7, %%mm1 \n\t"
"pcmpgtb %%mm6, %%mm1 \n\t"
"paddb %%mm1, %%mm0 \n\t"
"movq (%%"REG_a", %3, 2), %%mm1 \n\t"
PMAXUB(%%mm1, %%mm4)
PMINUB(%%mm1, %%mm3, %%mm5)
"psubb %%mm1, %%mm2 \n\t"
"paddb %%mm7, %%mm2 \n\t"
"pcmpgtb %%mm6, %%mm2 \n\t"
"paddb %%mm2, %%mm0 \n\t"
"movq (%2, %3, 8), %%mm2 \n\t"
PMAXUB(%%mm2, %%mm4)
PMINUB(%%mm2, %%mm3, %%mm5)
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
"psubb %%mm2, %%mm1 \n\t"
"paddb %%mm7, %%mm1 \n\t"
"pcmpgtb %%mm6, %%mm1 \n\t"
"paddb %%mm1, %%mm0 \n\t"
"movq (%%"REG_a", %3, 4), %%mm1 \n\t"
"psubb %%mm1, %%mm2 \n\t"
"paddb %%mm7, %%mm2 \n\t"
"pcmpgtb %%mm6, %%mm2 \n\t"
"paddb %%mm2, %%mm0 \n\t"
"psubusb %%mm3, %%mm4 \n\t"
"pxor %%mm6, %%mm6 \n\t"
"movq %4, %%mm7 \n\t" // QP,..., QP
"paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
"psubusb %%mm4, %%mm7 \n\t" // Diff >=2QP -> 0
"pcmpeqb %%mm6, %%mm7 \n\t" // Diff < 2QP -> 0
"pcmpeqb %%mm6, %%mm7 \n\t" // Diff < 2QP -> 0
"movq %%mm7, %1 \n\t"
"movq %5, %%mm7 \n\t"
"punpcklbw %%mm7, %%mm7 \n\t"
"punpcklbw %%mm7, %%mm7 \n\t"
"punpcklbw %%mm7, %%mm7 \n\t"
"psubb %%mm0, %%mm6 \n\t"
"pcmpgtb %%mm7, %%mm6 \n\t"
"movq %%mm6, %0 \n\t"
: "=m" (eq_mask), "=m" (dc_mask)
: "r" (src), "r" ((long)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
: "%"REG_a
);
both_masks = dc_mask & eq_mask;
if(both_masks){
long offset= -8*step;
int64_t *temp_sums= sums;
asm volatile(
"movq %2, %%mm0 \n\t" // QP,..., QP
"pxor %%mm4, %%mm4 \n\t"
"movq (%0), %%mm6 \n\t"
"movq (%0, %1), %%mm5 \n\t"
"movq %%mm5, %%mm1 \n\t"
"movq %%mm6, %%mm2 \n\t"
"psubusb %%mm6, %%mm5 \n\t"
"psubusb %%mm1, %%mm2 \n\t"
"por %%mm5, %%mm2 \n\t" // ABS Diff of lines
"psubusb %%mm2, %%mm0 \n\t" // diff >= QP -> 0
"pcmpeqb %%mm4, %%mm0 \n\t" // diff >= QP -> FF
"pxor %%mm6, %%mm1 \n\t"
"pand %%mm0, %%mm1 \n\t"
"pxor %%mm1, %%mm6 \n\t"
// 0:QP 6:First
"movq (%0, %1, 8), %%mm5 \n\t"
"add %1, %0 \n\t" // %0 points to line 1 not 0
"movq (%0, %1, 8), %%mm7 \n\t"
"movq %%mm5, %%mm1 \n\t"
"movq %%mm7, %%mm2 \n\t"
"psubusb %%mm7, %%mm5 \n\t"
"psubusb %%mm1, %%mm2 \n\t"
"por %%mm5, %%mm2 \n\t" // ABS Diff of lines
"movq %2, %%mm0 \n\t" // QP,..., QP
"psubusb %%mm2, %%mm0 \n\t" // diff >= QP -> 0
"pcmpeqb %%mm4, %%mm0 \n\t" // diff >= QP -> FF
"pxor %%mm7, %%mm1 \n\t"
"pand %%mm0, %%mm1 \n\t"
"pxor %%mm1, %%mm7 \n\t"
"movq %%mm6, %%mm5 \n\t"
"punpckhbw %%mm4, %%mm6 \n\t"
"punpcklbw %%mm4, %%mm5 \n\t"
// 4:0 5/6:First 7:Last
"movq %%mm5, %%mm0 \n\t"
"movq %%mm6, %%mm1 \n\t"
"psllw $2, %%mm0 \n\t"
"psllw $2, %%mm1 \n\t"
"paddw "MANGLE(w04)", %%mm0 \n\t"
"paddw "MANGLE(w04)", %%mm1 \n\t"
"movq (%0), %%mm2 \n\t"\
"movq (%0), %%mm3 \n\t"\
"add %1, %0 \n\t"\
"punpcklbw %%mm4, %%mm2 \n\t"\
"punpckhbw %%mm4, %%mm3 \n\t"\
"paddw %%mm2, %%mm0 \n\t"\
"paddw %%mm3, %%mm1 \n\t"
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
"movq (%0), %%mm2 \n\t"\
"movq (%0), %%mm3 \n\t"\
"add %1, %0 \n\t"\
"punpcklbw %%mm4, %%mm2 \n\t"\
"punpckhbw %%mm4, %%mm3 \n\t"\
"psubw %%mm2, %%mm0 \n\t"\
"psubw %%mm3, %%mm1 \n\t"
NEXT //0
NEXT //1
NEXT //2
"movq %%mm0, (%3) \n\t"
"movq %%mm1, 8(%3) \n\t"
NEXT //3
"psubw %%mm5, %%mm0 \n\t"
"psubw %%mm6, %%mm1 \n\t"
"movq %%mm0, 16(%3) \n\t"
"movq %%mm1, 24(%3) \n\t"
NEXT //4
"psubw %%mm5, %%mm0 \n\t"
"psubw %%mm6, %%mm1 \n\t"
"movq %%mm0, 32(%3) \n\t"
"movq %%mm1, 40(%3) \n\t"
NEXT //5
"psubw %%mm5, %%mm0 \n\t"
"psubw %%mm6, %%mm1 \n\t"
"movq %%mm0, 48(%3) \n\t"
"movq %%mm1, 56(%3) \n\t"
NEXT //6
"psubw %%mm5, %%mm0 \n\t"
"psubw %%mm6, %%mm1 \n\t"
"movq %%mm0, 64(%3) \n\t"
"movq %%mm1, 72(%3) \n\t"
"movq %%mm7, %%mm6 \n\t"
"punpckhbw %%mm4, %%mm7 \n\t"
"punpcklbw %%mm4, %%mm6 \n\t"
NEXT //7
"mov %4, %0 \n\t"
"add %1, %0 \n\t"
PREV //0
"movq %%mm0, 80(%3) \n\t"
"movq %%mm1, 88(%3) \n\t"
PREV //1
"paddw %%mm6, %%mm0 \n\t"
"paddw %%mm7, %%mm1 \n\t"
"movq %%mm0, 96(%3) \n\t"
"movq %%mm1, 104(%3) \n\t"
PREV //2
"paddw %%mm6, %%mm0 \n\t"
"paddw %%mm7, %%mm1 \n\t"
"movq %%mm0, 112(%3) \n\t"
"movq %%mm1, 120(%3) \n\t"
PREV //3
"paddw %%mm6, %%mm0 \n\t"
"paddw %%mm7, %%mm1 \n\t"
"movq %%mm0, 128(%3) \n\t"
"movq %%mm1, 136(%3) \n\t"
PREV //4
"paddw %%mm6, %%mm0 \n\t"
"paddw %%mm7, %%mm1 \n\t"
"movq %%mm0, 144(%3) \n\t"
"movq %%mm1, 152(%3) \n\t"
"mov %4, %0 \n\t" //FIXME
: "+&r"(src)
: "r" ((long)step), "m" (c->pQPb), "r"(sums), "g"(src)
);
src+= step; // src points to begin of the 8x8 Block
asm volatile(
"movq %4, %%mm6 \n\t"
"pcmpeqb %%mm5, %%mm5 \n\t"
"pxor %%mm6, %%mm5 \n\t"
"pxor %%mm7, %%mm7 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm1 \n\t"
"paddw 32(%1), %%mm0 \n\t"
"paddw 40(%1), %%mm1 \n\t"
"movq (%0, %3), %%mm2 \n\t"
"movq %%mm2, %%mm3 \n\t"
"movq %%mm2, %%mm4 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t"
"punpckhbw %%mm7, %%mm3 \n\t"
"paddw %%mm2, %%mm0 \n\t"
"paddw %%mm3, %%mm1 \n\t"
"paddw %%mm2, %%mm0 \n\t"
"paddw %%mm3, %%mm1 \n\t"
"psrlw $4, %%mm0 \n\t"
"psrlw $4, %%mm1 \n\t"
"packuswb %%mm1, %%mm0 \n\t"
"pand %%mm6, %%mm0 \n\t"
"pand %%mm5, %%mm4 \n\t"
"por %%mm4, %%mm0 \n\t"
"movq %%mm0, (%0, %3) \n\t"
"add $16, %1 \n\t"
"add %2, %0 \n\t"
" js 1b \n\t"
: "+r"(offset), "+r"(temp_sums)
: "r" ((long)step), "r"(src - offset), "m"(both_masks)
);
}else
src+= step; // src points to begin of the 8x8 Block
if(eq_mask != -1LL){
uint8_t *temp_src= src;
asm volatile(
"pxor %%mm7, %%mm7 \n\t"
"lea -40(%%"REG_SP"), %%"REG_c" \n\t" // make space for 4 8-byte vars
"and "ALIGN_MASK", %%"REG_c" \n\t" // align
// 0 1 2 3 4 5 6 7 8 9
// %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %1+8%1 ecx+4%1
"movq (%0), %%mm0 \n\t"
"movq %%mm0, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
"punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
"movq (%0, %1), %%mm2 \n\t"
"lea (%0, %1, 2), %%"REG_a" \n\t"
"movq %%mm2, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
"punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
"movq (%%"REG_a"), %%mm4 \n\t"
"movq %%mm4, %%mm5 \n\t"
"punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
"punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
"paddw %%mm0, %%mm0 \n\t" // 2L0
"paddw %%mm1, %%mm1 \n\t" // 2H0
"psubw %%mm4, %%mm2 \n\t" // L1 - L2
"psubw %%mm5, %%mm3 \n\t" // H1 - H2
"psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
"psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
"psllw $2, %%mm2 \n\t" // 4L1 - 4L2
"psllw $2, %%mm3 \n\t" // 4H1 - 4H2
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
"movq (%%"REG_a", %1), %%mm2 \n\t"
"movq %%mm2, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t" // L3
"punpckhbw %%mm7, %%mm3 \n\t" // H3
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
"movq %%mm0, (%%"REG_c") \n\t" // 2L0 - 5L1 + 5L2 - 2L3
"movq %%mm1, 8(%%"REG_c") \n\t" // 2H0 - 5H1 + 5H2 - 2H3
"movq (%%"REG_a", %1, 2), %%mm0 \n\t"
"movq %%mm0, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t" // L4
"punpckhbw %%mm7, %%mm1 \n\t" // H4
"psubw %%mm0, %%mm2 \n\t" // L3 - L4
"psubw %%mm1, %%mm3 \n\t" // H3 - H4
"movq %%mm2, 16(%%"REG_c") \n\t" // L3 - L4
"movq %%mm3, 24(%%"REG_c") \n\t" // H3 - H4
"paddw %%mm4, %%mm4 \n\t" // 2L2
"paddw %%mm5, %%mm5 \n\t" // 2H2
"psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
"psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
"lea (%%"REG_a", %1), %0 \n\t"
"psllw $2, %%mm2 \n\t" // 4L3 - 4L4
"psllw $2, %%mm3 \n\t" // 4H3 - 4H4
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
//50 opcodes so far
"movq (%0, %1, 2), %%mm2 \n\t"
"movq %%mm2, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t" // L5
"punpckhbw %%mm7, %%mm3 \n\t" // H5
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5