--- motion_comp.c 2002/10/05 21:42:04 1.11.2.4 +++ motion_comp.c 2002/11/02 15:52:31 1.11.2.9 @@ -1,3 +1,4 @@ +// 30.10.2002 corrected qpel chroma rounding // 04.10.2002 added qpel support to MBMotionCompensation // 01.05.2002 updated MBMotionCompensationBVOP // 14.04.2002 bframe compensation @@ -11,6 +12,57 @@ #define ABS(X) (((X)>0)?(X):-(X)) #define SIGN(X) (((X)>0)?1:-1) +static __inline void +compensate16x16_interpolate(int16_t * const dct_codes, + uint8_t * const cur, + const uint8_t * const ref, + const uint8_t * const refh, + const uint8_t * const refv, + const uint8_t * const refhv, + const uint32_t x, + const uint32_t y, + const int32_t dx, + const int32_t dy, + const uint32_t stride, + const uint32_t quarterpel, + const uint32_t rounding) +{ + if(quarterpel) { + interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, + (uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); + + transfer_8to16sub(dct_codes, cur + y*stride + x, + refv + y*stride + x, stride); + transfer_8to16sub(dct_codes+64, cur + y*stride + x + 8, + refv + y*stride + x + 8, stride); + transfer_8to16sub(dct_codes+128, cur + y*stride + x + 8*stride, + refv + y*stride + x + 8*stride, stride); + transfer_8to16sub(dct_codes+192, cur + y*stride + x + 8*stride + 8, + refv + y*stride + x + 8*stride+8, stride); + + } + else + { + const uint8_t * reference; + + switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) + { + case 0: reference = ref + ((y + dy / 2) * stride + x + dx / 2); break; + case 1: reference = refv + ((y + (dy-1) / 2) * stride + x + dx / 2); break; + case 2: reference = refh + ((y + dy / 2) * stride + x + (dx-1) / 2); break; + default: // case 3: + reference = refhv + ((y + (dy-1) / 2) * stride + x + (dx-1) / 2); break; + } + transfer_8to16sub(dct_codes, cur + y * stride + x, + reference, stride); + transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, + reference + 8, stride); + transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, + reference + 8*stride, stride); + transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, + reference + 8*stride + 8, stride); + } +} static __inline void compensate8x8_interpolate(int16_t * const dct_codes, @@ -67,59 +119,47 @@ const uint32_t quarterpel, const uint32_t rounding) { - if (mb->mode == MODE_NOT_CODED) { - transfer8x8_copy(cur->y + 16 * (i + j * edged_width), - ref->y + 16 * (i + j * edged_width), - edged_width); - transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8, - ref->y + 16 * (i + j * edged_width) + 8, - edged_width); - transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * edged_width, - ref->y + 16 * (i + j * edged_width) + 8 * edged_width, - edged_width); - transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), - ref->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), - edged_width); + if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { - transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), - ref->u + 8 * (i + j * edged_width/2), - edged_width / 2); - transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), - ref->v + 8 * (i + j * edged_width/2), - edged_width / 2); - return; - } + int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); + int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); - if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { - int32_t dx = mb->mvs[0].x; - int32_t dy = mb->mvs[0].y; - - if(quarterpel) { - dx = mb->qmvs[0].x; - dy = mb->qmvs[0].y; + if ( (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ + transfer8x8_copy(cur->y + 16 * (i + j * edged_width), + ref->y + 16 * (i + j * edged_width), + edged_width); + transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8, + ref->y + 16 * (i + j * edged_width) + 8, + edged_width); + transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * edged_width, + ref->y + 16 * (i + j * edged_width) + 8 * edged_width, + edged_width); + transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), + ref->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), + edged_width); + + transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), + ref->u + 8 * (i + j * edged_width/2), + edged_width / 2); + transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), + ref->v + 8 * (i + j * edged_width/2), + edged_width / 2); + return; } + /* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ - compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, - refv->y, refhv->y, 16 * i, 16 * j, dx, dy, - edged_width, quarterpel, rounding); - compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, - refv->y, refhv->y, 16 * i + 8, 16 * j, dx, dy, - edged_width, quarterpel, rounding); - compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, - refv->y, refhv->y, 16 * i, 16 * j + 8, dx, dy, - edged_width, quarterpel, rounding); - compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, - refv->y, refhv->y, 16 * i + 8, 16 * j + 8, dx, - dy, edged_width, quarterpel, rounding); + compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, + refv->y, refhv->y, 16 * i, 16 * j, dx, + dy, edged_width, quarterpel, rounding); if (quarterpel) { - dx = (dx >> 1) | (dx & 1); - dy = (dy >> 1) | (dy & 1); + dx /= 2; + dy /= 2; } - dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; - dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; + dx = (dx >> 1) + roundtab_79[dx & 0x3]; + dy = (dy >> 1) + roundtab_79[dy & 0x3]; /* uv-block-based compensation */ transfer_8to16sub(&dct_codes[4 * 64], @@ -156,21 +196,19 @@ refv->y, refhv->y, 16 * i + 8, 16 * j + 8, mvs[3].x, mvs[3].y, edged_width, quarterpel, rounding); - sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; - if(quarterpel) - sum /= 2; - - dx = (sum ? SIGN(sum) * - (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); + sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); + else + sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; - sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; + dx = (sum >> 3) + roundtab_76[sum & 0xf]; if(quarterpel) - sum /= 2; + sum = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); + else + sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; - dy = (sum ? SIGN(sum) * - (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); + dy = (sum >> 3) + roundtab_76[sum & 0xf]; /* uv-block-based compensation */ transfer_8to16sub(&dct_codes[4 * 64], @@ -183,7 +221,6 @@ cur->v + 8 * j * edged_width / 2 + 8 * i, interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, dx, dy, edged_width / 2, rounding), - edged_width / 2); } }