virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-11-23 21:54:53 +02:00
Code Issues Releases Activity

rv34: DC-only inverse transform

Browse Source 
When decoding coefficients, detect whether the block is DC-only, and take
advantage of this knowledge to perform DC-only inverse transform.

This is achieved by:
- first, changing the 108x4 element modulo_three_table into a 108 element
  table (kind of base4), and accessing each value using mask and shifts.
- then, checking low bits for 0 (as they represent the presence of higher
  frequency coefficients)

Also provide x86 SIMD code for the DC-only inverse transform.

Signed-off-by: Kostya Shishkov <kostya.shishkov@gmail.com>
This commit is contained in:
Christophe GISQUET
2012-01-01 18:33:22 +01:00
committed by Kostya Shishkov
parent b2ce3b998b
commit 3faa303a47
7 changed files with 200 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

95
libavcodec/rv34.c
View File

@@ -204,7 +204,7 @@ static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
}
for(i = 0; i < 4; i++){
t = modulo_three_table[code][i];
t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
if(t == 1)
cbp |= cbp_masks[get_bits1(gb)] << i;
if(t == 2)
@@ -238,41 +238,42 @@ static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *
*/
static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
{
int coeffs[4];
int flags = modulo_three_table[code];
coeffs[0] = modulo_three_table[code][0];
coeffs[1] = modulo_three_table[code][1];
coeffs[2] = modulo_three_table[code][2];
coeffs[3] = modulo_three_table[code][3];
decode_coeff(dst , coeffs[0], 3, gb, vlc, q);
decode_coeff( dst+0, (flags >> 6) , 3, gb, vlc, q);
if(is_block2){
decode_coeff(dst+8, coeffs[1], 2, gb, vlc, q);
decode_coeff(dst+1, coeffs[2], 2, gb, vlc, q);
decode_coeff(dst+8, (flags >> 4) & 3, 2, gb, vlc, q);
decode_coeff(dst+1, (flags >> 2) & 3, 2, gb, vlc, q);
}else{
decode_coeff(dst+1, coeffs[1], 2, gb, vlc, q);
decode_coeff(dst+8, coeffs[2], 2, gb, vlc, q);
decode_coeff(dst+1, (flags >> 4) & 3, 2, gb, vlc, q);
decode_coeff(dst+8, (flags >> 2) & 3, 2, gb, vlc, q);
}
decode_coeff(dst+9, coeffs[3], 2, gb, vlc, q);
decode_coeff( dst+9, (flags >> 0) & 3, 2, gb, vlc, q);
}
/**
* Decode a single coefficient.
*/
static inline void decode_subblock1(DCTELEM *dst, int code, GetBitContext *gb, VLC *vlc, int q)
{
int coeff = modulo_three_table[code] >> 6;
decode_coeff(dst, coeff, 3, gb, vlc, q);
}
static inline void decode_subblock3(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc,
int q_dc, int q_ac1, int q_ac2)
{
int coeffs[4];
int flags = modulo_three_table[code];
coeffs[0] = modulo_three_table[code][0];
coeffs[1] = modulo_three_table[code][1];
coeffs[2] = modulo_three_table[code][2];
coeffs[3] = modulo_three_table[code][3];
decode_coeff(dst , coeffs[0], 3, gb, vlc, q_dc);
decode_coeff( dst+0, (flags >> 6) , 3, gb, vlc, q_dc);
if(is_block2){
decode_coeff(dst+8, coeffs[1], 2, gb, vlc, q_ac1);
decode_coeff(dst+1, coeffs[2], 2, gb, vlc, q_ac1);
decode_coeff(dst+8, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
decode_coeff(dst+1, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
}else{
decode_coeff(dst+1, coeffs[1], 2, gb, vlc, q_ac1);
decode_coeff(dst+8, coeffs[2], 2, gb, vlc, q_ac1);
decode_coeff(dst+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
decode_coeff(dst+8, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
}
decode_coeff(dst+9, coeffs[3], 2, gb, vlc, q_ac2);
decode_coeff( dst+9, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
}
/**
@@ -286,16 +287,24 @@ static inline void decode_subblock3(DCTELEM *dst, int code, const int is_block2,
* o--o
*/
static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
static inline int rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
{
int code, pattern;
int code, pattern, has_ac = 1;
code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
pattern = code & 0x7;
code >>= 3;
decode_subblock3(dst, code, 0, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
if (modulo_three_table[code] & 0x3F) {
decode_subblock3(dst, code, 0, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
} else {
decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
if (!pattern)
return 0;
has_ac = 0;
}
if(pattern & 4){
code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
@@ -309,7 +318,7 @@ static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *r
code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
}
return has_ac || pattern;
}
/**
@@ -1118,7 +1127,7 @@ static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
MpegEncContext *s = &r->s;
GetBitContext *gb = &s->gb;
int cbp, cbp2;
int q_dc, q_ac;
int q_dc, q_ac, has_ac;
int i, blknum, blkoff;
LOCAL_ALIGNED_16(DCTELEM, block16, [64]);
int luma_dc_quant;
@@ -1157,33 +1166,45 @@ static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
if(r->is16){
q_dc = rv34_qscale_tab[luma_dc_quant];
q_ac = rv34_qscale_tab[s->qscale];
memset(block16, 0, 64 * sizeof(*block16));
rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
r->rdsp.rv34_inv_transform_tab[1](block16);
s->dsp.clear_block(block16);
if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
r->rdsp.rv34_inv_transform_tab[1](block16);
else
r->rdsp.rv34_inv_transform_dc_tab[1](block16);
}
q_ac = rv34_qscale_tab[s->qscale];
for(i = 0; i < 16; i++, cbp >>= 1){
DCTELEM *ptr;
if(!r->is16 && !(cbp & 1)) continue;
blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
blkoff = ((i & 1) << 2) + ((i & 4) << 3);
ptr = s->block[blknum] + blkoff;
if(cbp & 1)
rv34_decode_block(s->block[blknum] + blkoff, gb,
r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
else
has_ac = 0;
if(r->is16) //FIXME: optimize
s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
r->rdsp.rv34_inv_transform_tab[0](s->block[blknum] + blkoff);
ptr[0] = block16[(i & 3) | ((i & 0xC) << 1)];
if(has_ac)
r->rdsp.rv34_inv_transform_tab[0](ptr);
else
r->rdsp.rv34_inv_transform_dc_tab[0](ptr);
}
if(r->block_type == RV34_MB_P_MIX16x16)
r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
for(; i < 24; i++, cbp >>= 1){
DCTELEM *ptr;
if(!(cbp & 1)) continue;
blknum = ((i & 4) >> 2) + 4;
blkoff = ((i & 1) << 2) + ((i & 2) << 4);
rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1, q_dc, q_ac, q_ac);
r->rdsp.rv34_inv_transform_tab[0](s->block[blknum] + blkoff);
ptr = s->block[blknum] + blkoff;
if (rv34_decode_block(ptr, gb, r->cur_vlcs, r->chroma_vlc, 1, q_dc, q_ac, q_ac))
r->rdsp.rv34_inv_transform_tab[0](ptr);
else
r->rdsp.rv34_inv_transform_dc_tab[0](ptr);
}
if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos]))
rv34_output_macroblock(r, intra_types, cbp2, r->is16);