www.pudn.com > MPEG4Codec.zip > deblock_horiz_lpf9.c
/**************************************************************************
* *
* 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
*
**/
/***
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 */
const static uint64_t mm64_0008 = 0x0008000800080008;
const static uint64_t mm64_0101 = 0x0101010101010101;
static uint64_t mm64_temp;
const static uint64_t mm64_coefs[18] = {
0x0001000200040006, /* p1 left */ 0x0000000000000001, /* v1 right */
0x0001000200020004, /* v1 left */ 0x0000000000010001, /* v2 right */
0x0002000200040002, /* v2 left */ 0x0000000100010002, /* v3 right */
0x0002000400020002, /* v3 left */ 0x0001000100020002, /* v4 right */
0x0004000200020001, /* v4 left */ 0x0001000200020004, /* v5 right */
0x0002000200010001, /* v5 left */ 0x0002000200040002, /* v6 right */
0x0002000100010000, /* v6 left */ 0x0002000400020002, /* v7 right */
0x0001000100000000, /* v7 left */ 0x0004000200020001, /* v8 right */
0x0001000000000000, /* v8 left */ 0x0006000400020001 /* p2 right */
};
static uint32_t mm32_p1p2;
static uint8_t *pmm1;
/* The 9-tap low pass filter used in "DC" regions */
/* I'm not sure that I like this implementation any more...! */
INLINE void deblock_horiz_lpf9(uint8_t *v, int stride, int QP) {
int y, p1, p2;
#ifdef PP_SELF_CHECK
uint8_t selfcheck[9];
int psum;
uint8_t *vv;
int i;
#endif
for (y=0; y<4; y++) {
p1 = (ABS(v[0+y*stride]-v[1+y*stride]) < QP ) ? v[0+y*stride] : v[1+y*stride];
p2 = (ABS(v[8+y*stride]-v[9+y*stride]) < QP ) ? v[9+y*stride] : v[8+y*stride];
mm32_p1p2 = 0x0101 * ((p2 << 16) + p1);
#ifdef PP_SELF_CHECK
/* generate a self-check version of the filter result in selfcheck[9] */
/* low pass filtering (LPF9: 1 1 2 2 4 2 2 1 1) */
vv = &(v[y*stride]);
psum = p1 + p1 + p1 + vv[1] + vv[2] + vv[3] + vv[4] + 4;
selfcheck[1] = (((psum + vv[1]) << 1) - (vv[4] - vv[5])) >> 4;
psum += vv[5] - p1;
selfcheck[2] = (((psum + vv[2]) << 1) - (vv[5] - vv[6])) >> 4;
psum += vv[6] - p1;
selfcheck[3] = (((psum + vv[3]) << 1) - (vv[6] - vv[7])) >> 4;
psum += vv[7] - p1;
selfcheck[4] = (((psum + vv[4]) << 1) + p1 - vv[1] - (vv[7] - vv[8])) >> 4;
psum += vv[8] - vv[1];
selfcheck[5] = (((psum + vv[5]) << 1) + (vv[1] - vv[2]) - vv[8] + p2) >> 4;
psum += p2 - vv[2];
selfcheck[6] = (((psum + vv[6]) << 1) + (vv[2] - vv[3])) >> 4;
psum += p2 - vv[3];
selfcheck[7] = (((psum + vv[7]) << 1) + (vv[3] - vv[4])) >> 4;
psum += p2 - vv[4];
selfcheck[8] = (((psum + vv[8]) << 1) + (vv[4] - vv[5])) >> 4;
#endif
pmm1 = (&(v[y*stride-3])); /* this is 64-aligned */
/* mm7 = 0, mm6 is left hand accumulator, mm5 is right hand acc */
__asm {
push eax
push ebx
mov eax, pmm1
lea ebx, mm64_coefs
#ifdef PREFETCH_ENABLE
prefetcht0 32[ebx]
#endif
movd mm0, mm32_p1p2 /* mm0 = ________p2p2p1p1 0w1 2 3 4 5 6 7 */
punpcklbw mm0, mm0 /* mm0 = p2p2p2p2p1p1p1p1 0m1 2 3 4 5 6 7 */
movq mm2, qword ptr [eax] /* mm2 = v4v3v2v1xxxxxxxx 0 1 2w3 4 5 6 7 */
pxor mm7, mm7 /* mm7 = 0000000000000000 0 1 2 3 4 5 6 7w */
movq mm6, mm64_0008 /* mm6 = 0008000800080008 0 1 2 3 4 5 6w7 */
punpckhbw mm2, mm2 /* mm2 = v4__v3__v2__v1__ 0 1 2m3 4 5 6 7 */
movq mm64_temp, mm0 /*temp = p2p2p2p2p1p1p1p1 0r1 2 3 4 5 6 7 */
punpcklbw mm0, mm7 /* mm0 = __p1__p1__p1__p1 0m1 2 3 4 5 6 7 */
movq mm5, mm6 /* mm5 = 0008000800080008 0 1 2 3 4 5w6r7 */
pmullw mm0, [ebx] /* mm0 *= mm64_coefs[0] 0m1 2 3 4 5 6 7 */
movq mm1, mm2 /* mm1 = v4v4v3v3v2v2v1v1 0 1w2r3 4 5 6 7 */
punpcklbw mm2, mm2 /* mm2 = v2v2v2v2v1v1v1v1 0 1 2m3 4 5 6 7 */
punpckhbw mm1, mm1 /* mm1 = v4v4v4v4v3v3v3v3 0 1m2 3 4 5 6 7 */
#ifdef PREFETCH_ENABLE
prefetcht0 32[ebx]
#endif
movq mm3, mm2 /* mm3 = v2v2v2v2v1v1v1v1 0 1 2r3w4 5 6 7 */
punpcklbw mm2, mm7 /* mm2 = __v1__v1__v1__v1 0 1 2m3 4 5 6 7 */
punpckhbw mm3, mm7 /* mm3 = __v2__v2__v2__v2 0 1 2 3m4 5 6 7 */
paddw mm6, mm0 /* mm6 += mm0 0r1 2 3 4 5 6m7 */
movq mm0, mm2 /* mm0 = __v1__v1__v1__v1 0w1 2r3 4 5 6 7 */
pmullw mm0, 8[ebx] /* mm2 *= mm64_coefs[1] 0m1 2 3 4 5 6 7 */
movq mm4, mm3 /* mm4 = __v2__v2__v2__v2 0 1 2 3r4w5 6 7 */
pmullw mm2, 16[ebx] /* mm2 *= mm64_coefs[2] 0 1 2m3 4 5 6 7 */
pmullw mm3, 32[ebx] /* mm3 *= mm64_coefs[4] 0 1 2 3m4 5 6 7 */
pmullw mm4, 24[ebx] /* mm3 *= mm64_coefs[3] 0 1 2 3 4m5 6 7 */
paddw mm5, mm0 /* mm5 += mm0 0r1 2 3 4 5m6 7 */
paddw mm6, mm2 /* mm6 += mm2 0 1 2r3 4 5 6m7 */
movq mm2, mm1 /* mm2 = v4v4v4v4v3v3v3v3 0 1 2 3 4 5 6 7 */
punpckhbw mm2, mm7 /* mm2 = __v4__v4__v4__v4 0 1 2m3 4 5 6 7r */
paddw mm5, mm4 /* mm5 += mm4 0 1 2 3 4r5m6 7 */
punpcklbw mm1, mm7 /* mm1 = __v3__v3__v3__v3 0 1m2 3 4 5 6 7r */
paddw mm6, mm3 /* mm6 += mm3 0 1 2 3r4 5 6m7 */
#ifdef PREFETCH_ENABLE
prefetcht0 64[ebx]
#endif
movq mm0, mm1 /* mm0 = __v3__v3__v3__v3 0w1 2 3 4 5 6 7 */
pmullw mm1, 48[ebx] /* mm1 *= mm64_coefs[6] 0 1m2 3 4 5 6 7 */
pmullw mm0, 40[ebx] /* mm0 *= mm64_coefs[5] 0m1 2 3 4 5 6 7 */
movq mm4, mm2 /* mm4 = __v4__v4__v4__v4 0 1 2r3 4w5 6 7 */
pmullw mm2, 64[ebx] /* mm2 *= mm64_coefs[8] 0 1 2 3 4 5 6 7 */
paddw mm6, mm1 /* mm6 += mm1 0 1 2 3 4 5 6 7 */
pmullw mm4, 56[ebx] /* mm4 *= mm64_coefs[7] 0 1 2 3 4m5 6 7 */
pxor mm3, mm3 /* mm3 = 0000000000000000 0 1 2 3w4 5 6 7 */
movq mm1, 8[eax] /* mm1 = xxxxxxxxv8v7v6v5 0 1w2 3 4 5 6 7 */
paddw mm5, mm0 /* mm5 += mm0 0r1 2 3 4 5 6 7 */
punpcklbw mm1, mm1 /* mm1 = v8v8v7v7v6v6v5v5 0 1m2 3m4 5 6 7 */
paddw mm6, mm2 /* mm6 += mm2 0 1 2r3 4 5 6 7 */
#ifdef PREFETCH_ENABLE
prefetcht0 96[ebx]
#endif
movq mm2, mm1 /* mm2 = v8v8v7v7v6v6v5v5 0 1r2w3 4 5 6 7 */
paddw mm5, mm4 /* mm5 += mm4 0 1 2 3 4r5 6 7 */
punpcklbw mm2, mm2 /* mm2 = v6v6v6v6v5v5v5v5 0 1 2m3 4 5 6 7 */
punpckhbw mm1, mm1 /* mm1 = v8v8v8v8v7v7v7v7 0 1m2 3 4 5 6 7 */
movq mm3, mm2 /* mm3 = v6v6v6v6v5v5v5v5 0 1 2r3w4 5 6 7 */
punpcklbw mm2, mm7 /* mm2 = __v5__v5__v5__v5 0 1 2m3 4 5 6 7r */
punpckhbw mm3, mm7 /* mm3 = __v6__v6__v6__v6 0 1 2 3m4 5 6 7r */
movq mm0, mm2 /* mm0 = __v5__v5__v5__v5 0w1 2b3 4 5 6 7 */
pmullw mm0, 72[ebx] /* mm0 *= mm64_coefs[9] 0m1 2 3 4 5 6 7 */
movq mm4, mm3 /* mm4 = __v6__v6__v6__v6 0 1 2 3 4w5 6 7 */
pmullw mm2, 80[ebx] /* mm2 *= mm64_coefs[10] 0 1 2m3 4 5 6 7 */
pmullw mm3, 96[ebx] /* mm3 *= mm64_coefs[12] 0 1 2 3m4 5 6 7 */
pmullw mm4, 88[ebx] /* mm4 *= mm64_coefs[11] 0 1 2 3 4m5 6 7 */
paddw mm5, mm0 /* mm5 += mm0 0r1 2 3 4 5 6 7 */
paddw mm6, mm2 /* mm6 += mm2 0 1 2r3 4 5 6 7 */
movq mm2, mm1 /* mm2 = v8v8v8v8v7v7v7v7 0 1r2w3 4 5 6 7 */
paddw mm6, mm3 /* mm6 += mm3 0 1 2 3r4 5 6 7 */
punpcklbw mm1, mm7 /* mm1 = __v7__v7__v7__v7 0 1m2 3 4 5 6 7r */
paddw mm5, mm4 /* mm5 += mm4 0 1 2 3 4r5 6 7 */
punpckhbw mm2, mm7 /* mm2 = __v8__v8__v8__v8 0 1 2m3 4 5 6 7 */
#ifdef PREFETCH_ENABLE
prefetcht0 128[ebx]
#endif
movq mm3, mm64_temp /* mm0 = p2p2p2p2p1p1p1p1 0 1 2 3w4 5 6 7 */
movq mm0, mm1 /* mm0 = __v7__v7__v7__v7 0w1r2 3 4 5 6 7 */
pmullw mm0, 104[ebx] /* mm0 *= mm64_coefs[13] 0m1b2 3 4 5 6 7 */
movq mm4, mm2 /* mm4 = __v8__v8__v8__v8 0 1 2r3 4w5 6 7 */
pmullw mm1, 112[ebx] /* mm1 *= mm64_coefs[14] 0 1w2 3 4 5 6 7 */
punpckhbw mm3, mm7 /* mm0 = __p2__p2__p2__p2 0 1 2 3 4 5 6 7 */
pmullw mm2, 128[ebx] /* mm2 *= mm64_coefs[16] 0 1b2m3 4 5 6 7 */
pmullw mm4, 120[ebx] /* mm4 *= mm64_coefs[15] 0 1b2 3 4m5 6 7 */
paddw mm5, mm0 /* mm5 += mm0 0r1 2 3 4 5m6 7 */
pmullw mm3, 136[ebx] /* mm0 *= mm64_coefs[17] 0 1 2 3m4 5 6 7 */
paddw mm6, mm1 /* mm6 += mm1 0 1w2 3 4 5 6m7 */
paddw mm6, mm2 /* mm6 += mm2 0 1 2r3 4 5 6m7 */
paddw mm5, mm4 /* mm5 += mm4 0 1 2 3 4r5m6 7 */
psrlw mm6, 4 /* mm6 /= 16 0 1 2 3 4 5 6m7 */
paddw mm5, mm3 /* mm6 += mm0 0 1 2 3r4 5m6 7 */
psrlw mm5, 4 /* mm5 /= 16 0 1 2 3 4 5m6 7 */
packuswb mm6, mm5 /* pack result into mm6 0 1 2 3 4 5r6m7 */
movq 4[eax], mm6 /* v[] = mm6 0 1 2 3 4 5 6r7 */
pop ebx
pop eax
};
#ifdef PP_SELF_CHECK
for (i=1; i<=8; i++) {
if (selfcheck[i] != v[i+y*stride]) {
printf("ERROR: MMX version of horiz lpf9 is incorrect at %d\n", i);
}
}
#endif
}
}