📄 deblock_vert_default_filter.c
字号:
/**************************************************************************
* *
* This code has been developed by John Funnell. This software is an *
* implementation of a part of one or more MPEG-4 Video tools as *
* specified in ISO/IEC 14496-2 standard. Those intending to use this *
* software module in hardware or software products are advised that its *
* use may infringe existing patents or copyrights, and any such use *
* would be at such party's own risk. The original developer of this *
* software module and his/her company, and subsequent editors and their *
* companies (including Project Mayo), will have no liability for use of *
* this software or modifications or derivatives thereof. *
* *
* Project Mayo gives users of the Codec a license to this software *
* module or modifications thereof for use in hardware or software *
* products claiming conformance to the MPEG-4 Video Standard as *
* described in the Open DivX license. *
* *
* The complete Open DivX license can be found at *
* http://www.projectmayo.com/opendivx/license.php *
* *
**************************************************************************/
/**
* Copyright (C) 2001 - Project Mayo
*
* John Funnell
*
* DivX Advanced Research Center <darc@projectmayo.com>
*
**/
/***
References:
* ISO/IEC 14496-2
* MoMuSys-FDIS-V1.0-990812
* Intel Architecture Software Developer's Manual
Volume 2: Instruction Set Reference
***/
#include "postprocess_mmx.h"
/* John Funnell, December 2000 */
/* Vertical deblocking filter for use in non-flat picture regions */
INLINE void deblock_vert_default_filter(uint8_t *v, int stride, int QP) {
uint64_t *pmm1;
const uint64_t mm_0020 = 0x0020002000200020;
uint64_t mm_8_x_QP;
int i;
#ifdef PP_SELF_CHECK
/* define semi-constants to enable us to move up and down the picture easily... */
int l1 = 1 * stride;
int l2 = 2 * stride;
int l3 = 3 * stride;
int l4 = 4 * stride;
int l5 = 5 * stride;
int l6 = 6 * stride;
int l7 = 7 * stride;
int l8 = 8 * stride;
int x, y, a3_0_SC, a3_1_SC, a3_2_SC, d_SC, q_SC;
uint8_t selfcheck[8][2];
#endif
#ifdef PP_SELF_CHECK
/* compute selfcheck matrix for later comparison */
for (x=0; x<8; x++) {
a3_0_SC = 2*v[l3+x] - 5*v[l4+x] + 5*v[l5+x] - 2*v[l6+x];
a3_1_SC = 2*v[l1+x] - 5*v[l2+x] + 5*v[l3+x] - 2*v[l4+x];
a3_2_SC = 2*v[l5+x] - 5*v[l6+x] + 5*v[l7+x] - 2*v[l8+x];
q_SC = (v[l4+x] - v[l5+x]) / 2;
if (ABS(a3_0_SC) < 8*QP) {
d_SC = ABS(a3_0_SC) - MIN(ABS(a3_1_SC), ABS(a3_2_SC));
if (d_SC < 0) d_SC=0;
d_SC = (5*d_SC + 32) >> 6;
d_SC *= SIGN(-a3_0_SC);
//printf("d_SC[%d] preclip=%d\n", x, d_SC);
/* clip d in the range 0 ... q */
if (q_SC > 0) {
d_SC = d_SC<0 ? 0 : d_SC;
d_SC = d_SC>q_SC ? q_SC : d_SC;
} else {
d_SC = d_SC>0 ? 0 : d_SC;
d_SC = d_SC<q_SC ? q_SC : d_SC;
}
} else {
d_SC = 0;
}
selfcheck[x][0] = v[l4+x] - d_SC;
selfcheck[x][1] = v[l5+x] + d_SC;
}
#endif
((uint32_t *)&mm_8_x_QP)[0] =
((uint32_t *)&mm_8_x_QP)[1] = 0x00080008 * QP;
/* working in 4-pixel wide columns, left to right */
for (i=0; i<2; i++) { /*i=0 in left, i=1 in right */
/* v should be 64-bit aligned here */
pmm1 = (uint64_t *)(&(v[4*i]));
/* pmm1 will be 32-bit aligned but this doesn't matter as we'll use movd not movq */
__asm {
push ecx
mov ecx, pmm1
pxor mm7, mm7 /* mm7 = 0000000000000000 0 1 2 3 4 5 6 7w */
add ecx, stride /* %0 += stride 0 1 2 3 4 5 6 7 */
movd mm0, [ecx] /* mm0 = v1v1v1v1v1v1v1v1 0w1 2 3 4 5 6 7 */
punpcklbw mm0, mm7 /* mm0 = __v1__v1__v1__v1 L 0m1 2 3 4 5 6 7r */
add ecx, stride /* ecx += stride 0 1 2 3 4 5 6 7 */
movd mm1, [ecx] /* mm1 = v2v2v2v2v2v2v2v2 0 1w2 3 4 5 6 7 */
add ecx, stride /* ecx += stride 0 1 2 3 4 5 6 7 */
punpcklbw mm1, mm7 /* mm1 = __v2__v2__v2__v2 L 0 1m2 3 4 5 6 7r */
movd mm2, [ecx] /* mm2 = v3v3v3v3v3v3v3v3 0 1 2w3 4 5 6 7 */
add ecx, stride /* ecx += stride 0 1 2 3 4 5 6 7 */
punpcklbw mm2, mm7 /* mm2 = __v3__v3__v3__v3 L 0 1 2m3 4 5 6 7r */
movd mm3, [ecx] /* mm3 = v4v4v4v4v4v4v4v4 0 1 2 3w4 5 6 7 */
punpcklbw mm3, mm7 /* mm3 = __v4__v4__v4__v4 L 0 1 2 3m4 5 6 7r */
psubw mm1, mm2 /* mm1 = v2 - v3 L 0 1m2r3 4 5 6 7 */
movq mm4, mm1 /* mm4 = v2 - v3 L 0 1r2 3 4w5 6 7 */
psllw mm1, 2 /* mm1 = 4 * (v2 - v3) L 0 1m2 3 4 5 6 7 */
paddw mm1, mm4 /* mm1 = 5 * (v2 - v3) L 0 1m2 3 4r5 6 7 */
psubw mm0, mm3 /* mm0 = v1 - v4 L 0m1 2 3r4 5 6 7 */
psllw mm0, 1 /* mm0 = 2 * (v1 - v4) L 0m1 2 3 4 5 6 7 */
psubw mm0, mm1 /* mm0 = a3_1 L 0m1r2 3 4 5 6 7 */
pxor mm1, mm1 /* mm1 = 0000000000000000 0 1w2 3 4 5 6 7 */
pcmpgtw mm1, mm0 /* is 0 > a3_1 ? L 0r1m2 3 4 5 6 7 */
add ecx, stride /* ecx += stride 0 1 2 3 4 5 6 7 */
pxor mm0, mm1 /* mm0 = ABS(a3_1) step 1 L 0m1r2 3 4 5 6 7 */
psubw mm0, mm1 /* mm0 = ABS(a3_1) step 2 L 0m1r2 3 4 5 6 7 */
movd mm1, [ecx] /* mm1 = v5v5v5v5v5v5v5v5 0 1w2 3 4 5 6 7 */
punpcklbw mm1, mm7 /* mm1 = __v5__v5__v5__v5 L 0 1m2 3 4 5 6 7r */
add ecx, stride /* ecx += stride 0 1 2 3 4 5 6 7 */
psubw mm3, mm1 /* mm3 = v4 - v5 L 0 1r2 3m4 5 6 7 */
movd mm4, [ecx] /* mm4 = v6v6v6v6v6v6v6v6 0 1 2 3 4w5 6 7 */
punpcklbw mm4, mm7 /* mm4 = __v6__v6__v6__v6 L 0 1 2 3 4m5 6 7r */
add ecx, stride /* ecx += stride 0 1 2 3 4 5 6 7 */
movd mm5, [ecx] /* mm5 = v7v7v7v7v7v7v7v7 0 1 2 3 4 5w6 7 */
psubw mm2, mm4 /* mm2 = v3 - v6 L 0 1 2m3 4r5 6 7 */
punpcklbw mm5, mm7 /* mm5 = __v7__v7__v7__v7 L 0 1 2 3 4 5m6 7r */
add ecx, stride /* ecx += stride 0 1 2 3 4 5 6 7 */
psubw mm5, mm4 /* mm5 = v7 - v6 L 0 1 2 3 4r5m6 7 */
movq mm4, mm5 /* mm4 = v7 - v6 L 0 1 2 3 4w5r6 7 */
psllw mm4, 2 /* mm4 = 4 * (v7 - v6) L 0 1 2 3 4 5m6 7 */
paddw mm5, mm4 /* mm5 = 5 * (v7 - v6) L 0 1 2 3 4r5m6 7 */
movd mm4, [ecx] /* mm4 = v8v8v8v8v8v8v8v8 0 1 2 3 4w5 6 7 */
punpcklbw mm4, mm7 /* mm4 = __v8__v8__v8__v8 L 0 1 2 3 4m5 6 7r */
psubw mm1, mm4 /* mm1 = v5 - v8 L 0 1m2 3 4r5 6 7 */
pxor mm4, mm4 /* mm4 = 0000000000000000 0 1 2 3 4w5 6 7 */
psllw mm1, 1 /* mm1 = 2 * (v5 - v8) L 0 1m2 3 4 5 6 7 */
paddw mm1, mm5 /* mm1 = a3_2 L 0 1m2 3 4 5r6 7 */
pcmpgtw mm4, mm1 /* is 0 > a3_2 ? L 0 1r2 3 4m5 6 7 */
pxor mm1, mm4 /* mm1 = ABS(a3_2) step 1 L 0 1m2 3 4r5 6 7 */
psubw mm1, mm4 /* mm1 = ABS(a3_2) step 2 L 0 1m2 3 4r5 6 7 */
/* at this point, mm0 = ABS(a3_1), mm1 = ABS(a3_2), mm2 = v3 - v6, mm3 = v4 - v5 */
movq mm4, mm1 /* mm4 = ABS(a3_2) L 0 1r2 3 4w5 6 7 */
pcmpgtw mm1, mm0 /* is ABS(a3_2) > ABS(a3_1) 0r1m2 3 4 5 6 7 */
pand mm0, mm1 /* min() step 1 L 0m1r2 3 4 5 6 7 */
pandn mm1, mm4 /* min() step 2 L 0 1m2 3 4r5 6 7 */
por mm0, mm1 /* min() step 3 L 0m1r2 3 4 5 6 7 */
/* at this point, mm0 = MIN( ABS(a3_1), ABS(a3_2), mm2 = v3 - v6, mm3 = v4 - v5 */
movq mm1, mm3 /* mm1 = v4 - v5 L 0 1w2 3r4 5 6 7 */
psllw mm3, 2 /* mm3 = 4 * (v4 - v5) L 0 1 2 3m4 5 6 7 */
paddw mm3, mm1 /* mm3 = 5 * (v4 - v5) L 0 1r2 3m4 5 6 7 */
psllw mm2, 1 /* mm2 = 2 * (v3 - v6) L 0 1 2m3 4 5 6 7 */
psubw mm2, mm3 /* mm2 = a3_0 L 0 1 2m3r4 5 6 7 */
/* at this point, mm0 = MIN( ABS(a3_1), ABS(a3_2), mm1 = v4 - v5, mm2 = a3_0 */
movq mm4, mm2 /* mm4 = a3_0 L 0 1 2r3 4w5 6 7 */
pxor mm3, mm3 /* mm3 = 0000000000000000 0 1 2 3w4 5 6 7 */
pcmpgtw mm3, mm2 /* is 0 > a3_0 ? L 0 1 2r3m4 5 6 7 */
movq mm2, mm_8_x_QP /* mm4 = 8*QP 0 1 2w3 4 5 6 7 */
pxor mm4, mm3 /* mm4 = ABS(a3_0) step 1 L 0 1 2 3r4m5 6 7 */
psubw mm4, mm3 /* mm4 = ABS(a3_0) step 2 L 0 1 2 3r4m5 6 7 */
/* compare a3_0 against 8*QP */
pcmpgtw mm2, mm4 /* is 8*QP > ABS(d) ? L 0 1 2m3 4r5 6 7 */
pand mm2, mm4 /* if no, d = 0 L 0 1 2m3 4r5 6 7 */
movq mm4, mm2 /* mm2 = a3_0 L 0 1 2r3 4w5 6 7 */
/* at this point, mm0 = MIN( ABS(a3_1), ABS(a3_2), mm1 = v4 - v5, mm2 = a3_0 , mm3 = SGN(a3_0), mm4 = ABS(a3_0) */
psubusw mm4, mm0 /* mm0 = (A3_0 - a3_0) L 0r1 2 3 4m5 6 7 */
movq mm0, mm4 /* mm0=ABS(d) L 0w1 2 3 4r5 6 7 */
psllw mm0, 2 /* mm0 = 4 * (A3_0-a3_0) L 0m1 2 3 4 5 6 7 */
paddw mm0, mm4 /* mm0 = 5 * (A3_0-a3_0) L 0m1 2 3 4r5 6 7 */
paddw mm0, mm_0020 /* mm0 += 32 L 0m1 2 3 4 5 6 7 */
psraw mm0, 6 /* mm0 >>= 6 L 0m1 2 3 4 5 6 7 */
/* at this point, mm0 = ABS(d), mm1 = v4 - v5, mm3 = SGN(a3_0) */
pxor mm2, mm2 /* mm2 = 0000000000000000 0 1 2w3 4 5 6 7 */
pcmpgtw mm2, mm1 /* is 0 > (v4 - v5) ? L 0 1r2m3 4 5 6 7 */
pxor mm1, mm2 /* mm1 = ABS(mm1) step 1 L 0 1m2r3 4 5 6 7 */
psubw mm1, mm2 /* mm1 = ABS(mm1) step 2 L 0 1m2r3 4 5 6 7 */
psraw mm1, 1 /* mm1 >>= 2 L 0 1m2 3 4 5 6 7 */
/* OK at this point, mm0 = ABS(d), mm1 = ABS(q), mm2 = SGN(q), mm3 = SGN(-d) */
movq mm4, mm2 /* mm4 = SGN(q) L 0 1 2r3 4w5 6 7 */
pxor mm4, mm3 /* mm4 = SGN(q) ^ SGN(-d) L 0 1 2 3r4m5 6 7 */
movq mm5, mm0 /* mm5 = ABS(d) L 0r1 2 3 4 5w6 7 */
pcmpgtw mm5, mm1 /* is ABS(d) > ABS(q) ? L 0 1r2 3 4 5m6 7 */
pand mm1, mm5 /* min() step 1 L 0m1 2 3 4 5r6 7 */
pandn mm5, mm0 /* min() step 2 L 0 1r2 3 4 5m6 7 */
por mm1, mm5 /* min() step 3 L 0m1 2 3 4 5r6 7 */
pand mm1, mm4 /* if signs differ, set 0 L 0m1 2 3 4r5 6 7 */
pxor mm1, mm2 /* Apply sign step 1 L 0m1 2r3 4 5 6 7 */
psubw mm1, mm2 /* Apply sign step 2 L 0m1 2r3 4 5 6 7 */
/* at this point we have d in mm1 */
pop ecx
};
if (i==0) {
__asm {
movq mm6, mm1
};
}
}
/* add d to rows l4 and l5 in memory... */
pmm1 = (uint64_t *)(&(v[4*stride]));
__asm {
push ecx
mov ecx, pmm1
packsswb mm6, mm1
movq mm0, [ecx]
psubb mm0, mm6
movq [ecx], mm0
add ecx, stride /* %0 += stride 0 1 2 3 4 5 6 7 */
paddb mm6, [ecx]
movq [ecx], mm6
pop ecx
};
#ifdef PP_SELF_CHECK
/* do selfcheck */
for (x=0; x<8; x++) {
for (y=0; y<2; y++) {
if (selfcheck[x][y] != v[l4+x+y*stride]) {
printf("ERROR: problem with vertical default filter in col %d, row %d\n", x, y);
printf("%d should be %d\n", v[l4+x+y*stride], selfcheck[x][y]);
}
}
}
#endif
}
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -