static void MCFUNC(hl_motion)(const H264Context *h, H264SliceContext *sl,
uint8_t *dest_y,
uint8_t *dest_cb, uint8_t *dest_cr,
qpel_mc_func(*qpix_put)[16],
const h264_chroma_mc_func(*chroma_put),
qpel_mc_func(*qpix_avg)[16],
const h264_chroma_mc_func(*chroma_avg),
const h264_weight_func *weight_op,
const h264_biweight_func *weight_avg)
{
const int mb_xy = sl->mb_xy;
const int mb_type = h->cur_pic.mb_type[mb_xy];
av_assert2(IS_INTER(mb_type));
if (HAVE_THREADS && (h->avctx->active_thread_type & FF_THREAD_FRAME))
await_references(h, sl);
prefetch_motion(h, sl, 0, PIXEL_SHIFT, CHROMA_IDC);
if (IS_16X16(mb_type)) {
mc_part(h, sl, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
weight_op, weight_avg,
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
} else if (IS_16X8(mb_type)) {
mc_part(h, sl, 0, 0, 8, 8 << PIXEL_SHIFT, dest_y, dest_cb, dest_cr, 0, 0,
qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
weight_op, weight_avg,
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
mc_part(h, sl, 8, 0, 8, 8 << PIXEL_SHIFT, dest_y, dest_cb, dest_cr, 0, 4,
qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
weight_op, weight_avg,
IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
} else if (IS_8X16(mb_type)) {
mc_part(h, sl, 0, 0, 16, 8 * sl->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
&weight_op[1], &weight_avg[1],
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
mc_part(h, sl, 4, 0, 16, 8 * sl->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
&weight_op[1], &weight_avg[1],
IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
} else {
int i;
av_assert2(IS_8X8(mb_type));
for (i = 0; i < 4; i++) {
const int sub_mb_type = sl->sub_mb_type[i];
const int n = 4 * i;
int x_offset = (i & 1) << 2;
int y_offset = (i & 2) << 1;
if (IS_SUB_8X8(sub_mb_type)) {
mc_part(h, sl, n, 1, 8, 0, dest_y, dest_cb, dest_cr,
x_offset, y_offset,
qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
&weight_op[1], &weight_avg[1],
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
} else if (IS_SUB_8X4(sub_mb_type)) {
mc_part(h, sl, n, 0, 4, 4 << PIXEL_SHIFT, dest_y, dest_cb, dest_cr,
x_offset, y_offset,
qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
&weight_op[1], &weight_avg[1],
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
mc_part(h, sl, n + 2, 0, 4, 4 << PIXEL_SHIFT,
dest_y, dest_cb, dest_cr, x_offset, y_offset + 2,
qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
&weight_op[1], &weight_avg[1],
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
} else if (IS_SUB_4X8(sub_mb_type)) {
mc_part(h, sl, n, 0, 8, 4 * sl->mb_linesize,
dest_y, dest_cb, dest_cr, x_offset, y_offset,
qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
&weight_op[2], &weight_avg[2],
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
mc_part(h, sl, n + 1, 0, 8, 4 * sl->mb_linesize,
dest_y, dest_cb, dest_cr, x_offset + 2, y_offset,
qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
&weight_op[2], &weight_avg[2],
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
} else {
int j;
av_assert2(IS_SUB_4X4(sub_mb_type));
for (j = 0; j < 4; j++) {
int sub_x_offset = x_offset + 2 * (j & 1);
int sub_y_offset = y_offset + (j & 2);
mc_part(h, sl, n + j, 1, 4, 0,
dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
&weight_op[2], &weight_avg[2],
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
}
}
}
}
if (USES_LIST(mb_type, 1))
prefetch_motion(h, sl, 1, PIXEL_SHIFT, CHROMA_IDC);
}
void render_mbs()
{
H264Context *h = g_h;
GPUH264Context * const g = &h->gpu;
MpegEncContext * const s = &h->s;
H264mb* blockStore = g->block_buffer;
int i, l;
int lists = (h->slice_type==FF_B_TYPE)?2:1;
int dpb_pos = s->current_picture.gpu_dpb;
printf("Attempting to motion compensate %d blocks\n", (g->end-g->start+1));
glNewList(dispList, GL_COMPILE);
for(l=0; l < lists; l++)
{
glBegin(GL_QUADS);
for(i= g->start; i <= g->end; i++)
{
const int mb_x = blockStore[i].mb_x;
const int mb_y = blockStore[i].mb_y;
const int mb_xy = mb_x + mb_y*s->mb_stride;
const int mb_type = s->current_picture.mb_type[mb_xy];
int mv_x, mv_y, j;
//RUDD TODO ignoring Intra blocks for now
if(IS_INTER(mb_type))
{
if(IS_16X16(mb_type) && IS_DIR(mb_type, 0, l))
{
mv_x = blockStore[i].mv_cache[0][ scan8[0]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[0]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 16, 16, 0, 0,
h->ref_list[l][h->ref_cache[l][ scan8[0] ]].gpu_dpb, dpb_pos);
}
else if(IS_16X8(mb_type))
{
if(IS_DIR(mb_type, 0, l))
{
mv_x = blockStore[i].mv_cache[l][ scan8[0]][0];
mv_y = blockStore[i].mv_cache[l][ scan8[0]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 16, 8, 0, 0,
h->ref_list[l][h->ref_cache[l][ scan8[0] ]].gpu_dpb, dpb_pos);
}
if(IS_DIR(mb_type, 1, l))
{
mv_x = blockStore[i].mv_cache[0][ scan8[8]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[8]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 16, 8, 0, -8,
h->ref_list[l][h->ref_cache[l][ scan8[8] ]].gpu_dpb, dpb_pos);
}
}
else if(IS_8X16(mb_type))
{
if(IS_DIR(mb_type, 0, l))
{
mv_x = blockStore[i].mv_cache[0][ scan8[0]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[0]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 8, 16, 0, 0,
h->ref_list[l][h->ref_cache[l][ scan8[0] ]].gpu_dpb, dpb_pos);
}
if(IS_DIR(mb_type, 1, l))
{
mv_x = blockStore[i].mv_cache[0][ scan8[4]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[4]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 8, 16, 8, 0,
h->ref_list[l][h->ref_cache[l][ scan8[4] ]].gpu_dpb, dpb_pos);
}
}
else
{
assert(IS_8X8(mb_type));
int j;
for(j=0;j<4;j++)
{
const int sub_mb_type= h->sub_mb_type[j];
const int n= 4*j;
int x_offset= (j&1);
int y_offset= (j&2)>>1;
if(!IS_DIR(sub_mb_type, 0, l))
continue;
if(IS_SUB_8X8(sub_mb_type))
{
mv_x = blockStore[i].mv_cache[0][ scan8[n]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[n]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 8, 8, 8*x_offset,-8*y_offset,
h->ref_list[l][h->ref_cache[l][ scan8[n] ]].gpu_dpb, dpb_pos);
}
else if(IS_SUB_8X4(sub_mb_type))
{
mv_x = blockStore[i].mv_cache[0][ scan8[n]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[n]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 8, 4, 8*x_offset,-8*y_offset,
h->ref_list[l][h->ref_cache[l][ scan8[n] ]].gpu_dpb, dpb_pos);
mv_x = blockStore[i].mv_cache[0][ scan8[n+2]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[n+2]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 8, 4, 8*x_offset,-8*y_offset-4,
h->ref_list[l][h->ref_cache[l][ scan8[n+2] ]].gpu_dpb, dpb_pos);
}
else if(IS_SUB_4X8(sub_mb_type))
{
mv_x = blockStore[i].mv_cache[0][ scan8[n]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[n]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 4, 8, 8*x_offset,-8*y_offset,
h->ref_list[l][h->ref_cache[l][ scan8[n] ]].gpu_dpb, dpb_pos);
mv_x = blockStore[i].mv_cache[0][ scan8[n+1]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[n+1]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 4, 8, 8*x_offset+4,-8*y_offset,
h->ref_list[l][h->ref_cache[l][ scan8[n+1] ]].gpu_dpb, dpb_pos);
}
else
{
mv_x = blockStore[i].mv_cache[0][ scan8[n]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[n]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 4, 4, 8*x_offset,-8*y_offset,
h->ref_list[l][h->ref_cache[l][ scan8[n] ]].gpu_dpb, dpb_pos);
mv_x = blockStore[i].mv_cache[0][ scan8[n+1]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[n+1]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 4, 4, 8*x_offset+4,-8*y_offset,
h->ref_list[l][h->ref_cache[l][ scan8[n+1] ]].gpu_dpb, dpb_pos);
mv_x = blockStore[i].mv_cache[0][ scan8[n+2]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[n+2]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 4, 4, 8*x_offset,-8*y_offset-4,
h->ref_list[l][h->ref_cache[l][ scan8[n+2] ]].gpu_dpb, dpb_pos);
mv_x = blockStore[i].mv_cache[0][ scan8[n+3]][0];
mv_y = blockStore[i].mv_cache[0][ scan8[n+3]][1];
render_one_block(mb_x, mb_y, mv_x, mv_y, 4, 4, 8*x_offset+4,-8*y_offset-4,
h->ref_list[l][h->ref_cache[l][ scan8[n+3] ]].gpu_dpb, dpb_pos);
}
}
}
}
}
}
glEnd();
glEndList();
}
void motion_execute(H264_Slice_GlbARGs *SLICE_T, H264_MB_Ctrl_DecARGs *dmb, uint8_t *recon_buf,
uint8_t *motion_dha)
{
uint8_t *motion_douty, *motion_doutc;
uint8_t *motion_dsa = motion_dha + 0x108;
const int mb_type= dmb->mb_type;
motion_douty = recon_buf;
motion_doutc = recon_buf + PREVIOUS_OFFSET_U;
SET_REG1_DSTA(TCSM1_PADDR((int)motion_douty));
SET_REG1_DSA(TCSM1_PADDR((int)motion_dsa));
SET_REG2_DSTA(TCSM1_PADDR((int)motion_doutc));
SET_REG2_DSA(TCSM1_PADDR((int)motion_dsa));
volatile int *tdd = (int *)motion_dha;
int tkn = 0;
motion_dsa[0] = 0x0;
tdd++;
if(IS_16X16(mb_type)){
motion_task(SLICE_T, dmb, 0, 0, tdd, &tkn,
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
3/*blkh*/, 3/*blkw*/, 0/*boy*/, 0/*box*/);
}else if(IS_16X8(mb_type)){
motion_task(SLICE_T, dmb, 0, 0, tdd, &tkn,
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
2/*blkh*/, 3/*blkw*/, 0/*boy*/, 0/*box*/);
motion_task(SLICE_T, dmb, 1, 2, tdd, &tkn,
IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
2/*blkh*/, 3/*blkw*/, 2/*boy*/, 0/*box*/);
}else if(IS_8X16(mb_type)){
motion_task(SLICE_T, dmb, 0, 0, tdd, &tkn,
IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
3/*blkh*/, 2/*blkw*/, 0/*boy*/, 0/*box*/);
motion_task(SLICE_T, dmb, 1, 1, tdd, &tkn,
IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
3/*blkh*/, 2/*blkw*/, 0/*boy*/, 2/*box*/);
}else{
int i;
int mv_n=0;
for(i=0; i<4; i++){
const int sub_mb_type= dmb->sub_mb_type[i];
if(IS_SUB_8X8(sub_mb_type)){
motion_task(SLICE_T, dmb, mv_n, i, tdd, &tkn,
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
2/*blkh*/, 2/*blkw*/, (i & 0x2)/*boy*/, (i & 0x1)*2/*box*/);
mv_n++;
}else if(IS_SUB_8X4(sub_mb_type)){
motion_task(SLICE_T, dmb, mv_n, i, tdd, &tkn,
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
1/*blkh*/, 2/*blkw*/, (i & 0x2)/*boy*/, (i & 0x1)*2/*box*/);
mv_n++;
motion_task(SLICE_T, dmb, mv_n, i, tdd, &tkn,
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
1/*blkh*/, 2/*blkw*/, (i & 0x2)+1/*boy*/, (i & 0x1)*2/*box*/);
mv_n++;
}else if(IS_SUB_4X8(sub_mb_type)){
motion_task(SLICE_T, dmb, mv_n, i, tdd, &tkn,
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
2/*blkh*/, 1/*blkw*/, (i & 0x2)/*boy*/, (i & 0x1)*2/*box*/);
mv_n++;
motion_task(SLICE_T, dmb, mv_n, i, tdd, &tkn,
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
2/*blkh*/, 1/*blkw*/, (i & 0x2)/*boy*/, (i & 0x1)*2+1/*box*/);
mv_n++;
}else{
int j;
for(j=0; j<4; j++){
motion_task(SLICE_T, dmb, mv_n, i, tdd, &tkn,
IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
1/*blkh*/, 1/*blkw*/,
(i & 0x2) + (j & 0x2)/2/*boy*/, (i & 0x1)*2 + (j & 0x1)/*box*/);
mv_n++;
} //j
} //BLK4X4
} //i
} //BLK8X8
tdd[2*tkn-1] |= 0x1<<TDD_DOE_SFT;
tdd[-1] = TDD_HEAD(1,/*vld*/
1,/*lk*/
0,/*sync*/
1,/*ch1pel*/
2,/*ch2pel*/
TDD_POS_SPEC,/*posmd*/
TDD_MV_AUTO,/*mvmd*/
1,/*ch2en*/
tkn,/*tkn*/
dmb->mb_y,/*mby*/
dmb->mb_x/*mbx*/);
tdd[2*tkn] = TDD_HEAD(1,/*vld*/
0,/*lk*/
1,/*sync*/
1,/*ch1pel*/
2,/*ch2pel*/
TDD_POS_SPEC,/*posmd*/
TDD_MV_AUTO,/*mvmd*/
1,/*ch2en*/
0,/*tkn*/
0xFF,/*mby*/
0xFF/*mbx*/);
}
