1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00

rv34: Intra 16x16 handling

Extract processing of intra 16x16 blocks from intra macroblock
processing.
Also implement a function performing inverse transform and block
reconstruction for DC-only blocks in 1 pass instead of 2.
This commit is contained in:
Christophe GISQUET 2012-01-03 00:22:11 +01:00 committed by Janne Grunau
parent 3eeb755763
commit d78062386e
5 changed files with 297 additions and 145 deletions

View File

@ -982,15 +982,6 @@ static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int
r->h.pred4x4[itype](dst, prev, stride); r->h.pred4x4[itype](dst, prev, stride);
} }
/** add_pixels_clamped for 4x4 block */
static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
{
int x, y;
for(y = 0; y < 4; y++)
for(x = 0; x < 4; x++)
dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
}
static inline int adjust_pred16(int itype, int up, int left) static inline int adjust_pred16(int itype, int up, int left)
{ {
if(!up && !left) if(!up && !left)
@ -1007,15 +998,20 @@ static inline int adjust_pred16(int itype, int up, int left)
return itype; return itype;
} }
static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16) static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
{ {
MpegEncContext *s = &r->s; LOCAL_ALIGNED_16(DCTELEM, block16, [64]);
DSPContext *dsp = &s->dsp; MpegEncContext *s = &r->s;
int i, j; DSPContext *dsp = &s->dsp;
uint8_t *Y, *U, *V; GetBitContext *gb = &s->gb;
int itype; int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
int avail[6*8] = {0}; q_ac = rv34_qscale_tab[s->qscale];
int idx; uint8_t *dst = s->dest[0];
DCTELEM *ptr = s->block[0];
int avail[6*8] = {0};
int i, j, itype, has_ac;
memset(block16, 0, 64 * sizeof(*block16));
// Set neighbour information. // Set neighbour information.
if(r->avail_cache[1]) if(r->avail_cache[1])
@ -1031,52 +1027,142 @@ static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int c
if(r->avail_cache[9]) if(r->avail_cache[9])
avail[24] = avail[32] = 1; avail[24] = avail[32] = 1;
Y = s->dest[0]; has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
U = s->dest[1]; if(has_ac)
V = s->dest[2]; r->rdsp.rv34_inv_transform_tab[1](block16);
if(!is16){ else
for(j = 0; j < 4; j++){ r->rdsp.rv34_inv_transform_dc_tab[1](block16);
idx = 9 + j*8;
for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
avail[idx] = 1;
if(cbp & 1)
rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
}
Y += s->linesize * 4 - 4*4;
intra_types += r->intra_types_stride;
}
intra_types -= r->intra_types_stride * 4;
fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
for(j = 0; j < 2; j++){
idx = 6 + j*4;
for(i = 0; i < 2; i++, cbp >>= 1, idx++){
rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
r->avail_cache[idx] = 1;
if(cbp & 0x01)
rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
if(cbp & 0x10)
rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
}
}
}else{
itype = ittrans16[intra_types[0]];
itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
r->h.pred16x16[itype](Y, s->linesize);
dsp->add_pixels_clamped(s->block[0], Y, s->linesize);
dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
Y += s->linesize * 8;
dsp->add_pixels_clamped(s->block[2], Y, s->linesize);
dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
itype = ittrans16[intra_types[0]]; itype = ittrans16[intra_types[0]];
if(itype == PLANE_PRED8x8) itype = DC_PRED8x8; itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]); r->h.pred16x16[itype](dst, s->linesize);
r->h.pred8x8[itype](U, s->uvlinesize);
dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize); dsp->clear_block(ptr);
r->h.pred8x8[itype](V, s->uvlinesize); for(j = 0; j < 4; j++){
dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize); for(i = 0; i < 4; i++, cbp >>= 1){
int dc = block16[i + j*8];
if(cbp & 1){
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(has_ac){
ptr[0] = dc;
r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
dsp->clear_block(ptr);
}else
r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
}
dst += 4*s->linesize;
}
itype = ittrans16[intra_types[0]];
if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
for(j = 1; j < 3; j++){
dst = s->dest[j];
r->h.pred8x8[itype](dst, s->uvlinesize);
for(i = 0; i < 4; i++, cbp >>= 1){
uint8_t *pdst;
if(!(cbp & 1)) continue;
pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->chroma_vlc, 1, q_dc, q_ac, q_ac);
if(has_ac){
r->rdsp.rv34_idct_add(pdst, s->uvlinesize, ptr);
dsp->clear_block(ptr);
}else{
r->rdsp.rv34_idct_dc_add(pdst, s->uvlinesize, ptr[0]);
ptr[0] = 0;
}
}
}
}
static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
{
MpegEncContext *s = &r->s;
DSPContext *dsp = &s->dsp;
GetBitContext *gb = &s->gb;
DCTELEM *ptr = s->block[0];
uint8_t *dst = s->dest[0];
int avail[6*8] = {0};
int i, j, k;
int idx, has_ac;
int q_ac, q_dc;
// Set neighbour information.
if(r->avail_cache[1])
avail[0] = 1;
if(r->avail_cache[2])
avail[1] = avail[2] = 1;
if(r->avail_cache[3])
avail[3] = avail[4] = 1;
if(r->avail_cache[4])
avail[5] = 1;
if(r->avail_cache[5])
avail[8] = avail[16] = 1;
if(r->avail_cache[9])
avail[24] = avail[32] = 1;
q_ac = rv34_qscale_tab[s->qscale];
for(j = 0; j < 4; j++){
idx = 9 + j*8;
for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
avail[idx] = 1;
if(!(cbp & 1)) continue;
has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
if(has_ac){
r->rdsp.rv34_idct_add(dst, s->linesize, ptr);
dsp->clear_block(ptr);
}else{
r->rdsp.rv34_idct_dc_add(dst, s->linesize, ptr[0]);
ptr[0] = 0;
}
}
dst += s->linesize * 4 - 4*4;
intra_types += r->intra_types_stride;
}
intra_types -= r->intra_types_stride * 4;
q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
for(k = 0; k < 2; k++){
dst = s->dest[1+k];
fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
for(j = 0; j < 2; j++){
int* acache = r->avail_cache + 6 + j*4;
for(i = 0; i < 2; i++, cbp >>= 1, acache++){
int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
acache[0] = 1;
if(!(cbp&1)) continue;
has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->chroma_vlc, 1, q_dc, q_ac, q_ac);
if(has_ac){
r->rdsp.rv34_idct_add(dst + 4*i, s->uvlinesize, ptr);
dsp->clear_block(ptr);
}
else {
r->rdsp.rv34_idct_dc_add(dst + 4*i, s->uvlinesize, ptr[0]);
ptr[0] = 0;
}
}
dst += 4*s->uvlinesize;
}
} }
} }
@ -1185,6 +1271,12 @@ static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
if(cbp == -1) if(cbp == -1)
return -1; return -1;
if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
else rv34_output_intra(r, intra_types, cbp);
return 0;
}
if(r->is16){ if(r->is16){
int luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 int luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16
? r->luma_dc_quant_p[s->qscale] ? r->luma_dc_quant_p[s->qscale]
@ -1243,10 +1335,7 @@ static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
else else
r->rdsp.rv34_inv_transform_dc_tab[0](ptr); r->rdsp.rv34_inv_transform_dc_tab[0](ptr);
} }
if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])) rv34_apply_differences(r, cbp2);
rv34_output_macroblock(r, intra_types, cbp2, r->is16);
else
rv34_apply_differences(r, cbp2);
return 0; return 0;
} }
@ -1254,12 +1343,7 @@ static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types) static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
{ {
MpegEncContext *s = &r->s; MpegEncContext *s = &r->s;
GetBitContext *gb = &s->gb; int cbp, dist;
int cbp, cbp2;
int q_dc, q_ac, has_ac;
int i, blknum, blkoff;
LOCAL_ALIGNED_16(DCTELEM, block16, [64]);
int dist;
int mb_pos = s->mb_x + s->mb_y * s->mb_stride; int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
// Calculate which neighbours are available. Maybe it's worth optimizing too. // Calculate which neighbours are available. Maybe it's worth optimizing too.
@ -1278,74 +1362,21 @@ static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1]; r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
s->qscale = r->si.quant; s->qscale = r->si.quant;
cbp = cbp2 = rv34_decode_intra_mb_header(r, intra_types); cbp = rv34_decode_intra_mb_header(r, intra_types);
r->cbp_luma [mb_pos] = cbp; r->cbp_luma [mb_pos] = cbp;
r->cbp_chroma[mb_pos] = cbp >> 16; r->cbp_chroma[mb_pos] = cbp >> 16;
r->deblock_coefs[mb_pos] = 0xFFFF; r->deblock_coefs[mb_pos] = 0xFFFF;
s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale; s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
if(cbp == -1) if(cbp == -1)
return -1; return -1;
if(r->is16){ if(r->is16){
int luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 rv34_output_i16x16(r, intra_types, cbp);
? r->luma_dc_quant_p[s->qscale] return 0;
: r->luma_dc_quant_i[s->qscale];
q_dc = rv34_qscale_tab[luma_dc_quant];
q_ac = rv34_qscale_tab[s->qscale];
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;
blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
blkoff = ((i & 1) << 2) + ((i & 4) << 3);
ptr = s->block[blknum] + blkoff;
if(cbp & 1)
has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
else
has_ac = 0;
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);
}
}else{
q_ac = rv34_qscale_tab[s->qscale];
for(i = 0; i < 16; i++, cbp >>= 1){
DCTELEM *ptr;
if(!(cbp & 1)) continue;
blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
blkoff = ((i & 1) << 2) + ((i & 4) << 3);
ptr = s->block[blknum] + blkoff;
has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
if(has_ac)
r->rdsp.rv34_inv_transform_tab[0](ptr);
else
r->rdsp.rv34_inv_transform_dc_tab[0](ptr);
}
} }
q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]]; rv34_output_intra(r, intra_types, cbp);
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);
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);
}
rv34_output_macroblock(r, intra_types, cbp2, r->is16);
return 0; return 0;
} }

View File

@ -32,7 +32,7 @@
* @{ * @{
*/ */
static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block) static av_always_inline void rv34_row_transform(int temp[16], const DCTELEM *block)
{ {
int i; int i;
@ -72,6 +72,32 @@ static void rv34_inv_transform_c(DCTELEM *block){
} }
} }
/**
* Real Video 3.0/4.0 inverse transform + sample reconstruction
* Code is almost the same as in SVQ3, only scaling is different.
*/
static void rv34_idct_add_c(uint8_t *dst, int stride, const DCTELEM *block){
int temp[16];
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
int i;
rv34_row_transform(temp, block);
for(i = 0; i < 4; i++){
const int z0 = 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
const int z1 = 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
const int z2 = 7* temp[4*1+i] - 17*temp[4*3+i];
const int z3 = 17* temp[4*1+i] + 7*temp[4*3+i];
dst[0] = cm[ dst[0] + ( (z0 + z3) >> 10 ) ];
dst[1] = cm[ dst[1] + ( (z1 + z2) >> 10 ) ];
dst[2] = cm[ dst[2] + ( (z1 - z2) >> 10 ) ];
dst[3] = cm[ dst[3] + ( (z0 - z3) >> 10 ) ];
dst += stride;
}
}
/** /**
* RealVideo 3.0/4.0 inverse transform for DC block * RealVideo 3.0/4.0 inverse transform for DC block
* *
@ -97,6 +123,22 @@ static void rv34_inv_transform_noround_c(DCTELEM *block){
} }
} }
static void rv34_idct_dc_add_c(uint8_t *dst, int stride, int dc)
{
const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
int i, j;
cm += (13*13*dc + 0x200) >> 10;
for (i = 0; i < 4; i++)
{
for (j = 0; j < 4; j++)
dst[j] = cm[ dst[j] ];
dst += stride;
}
}
static void rv34_inv_transform_dc_c(DCTELEM *block) static void rv34_inv_transform_dc_c(DCTELEM *block)
{ {
DCTELEM dc = (13 * 13 * block[0] + 0x200) >> 10; DCTELEM dc = (13 * 13 * block[0] + 0x200) >> 10;
@ -126,6 +168,9 @@ av_cold void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp) {
c->rv34_inv_transform_dc_tab[0] = rv34_inv_transform_dc_c; c->rv34_inv_transform_dc_tab[0] = rv34_inv_transform_dc_c;
c->rv34_inv_transform_dc_tab[1] = rv34_inv_transform_dc_noround_c; c->rv34_inv_transform_dc_tab[1] = rv34_inv_transform_dc_noround_c;
c->rv34_idct_add = rv34_idct_add_c;
c->rv34_idct_dc_add = rv34_idct_dc_add_c;
if (HAVE_NEON) if (HAVE_NEON)
ff_rv34dsp_init_neon(c, dsp); ff_rv34dsp_init_neon(c, dsp);
if (HAVE_MMX) if (HAVE_MMX)

View File

@ -36,6 +36,11 @@ typedef void (*rv40_weight_func)(uint8_t *dst/*align width (8 or 16)*/,
typedef void (*rv34_inv_transform_func)(DCTELEM *block); typedef void (*rv34_inv_transform_func)(DCTELEM *block);
typedef void (*rv34_idct_add_func)(uint8_t *dst, int stride,
const DCTELEM *block);
typedef void (*rv34_idct_dc_add_func)(uint8_t *dst, int stride,
int dc);
typedef void (*rv40_weak_loop_filter_func)(uint8_t *src, int stride, typedef void (*rv40_weak_loop_filter_func)(uint8_t *src, int stride,
int filter_p1, int filter_q1, int filter_p1, int filter_q1,
int alpha, int beta, int alpha, int beta,
@ -57,6 +62,8 @@ typedef struct RV34DSPContext {
rv40_weight_func rv40_weight_pixels_tab[2]; rv40_weight_func rv40_weight_pixels_tab[2];
rv34_inv_transform_func rv34_inv_transform_tab[2]; rv34_inv_transform_func rv34_inv_transform_tab[2];
void (*rv34_inv_transform_dc_tab[2])(DCTELEM *block); void (*rv34_inv_transform_dc_tab[2])(DCTELEM *block);
rv34_idct_add_func rv34_idct_add;
rv34_idct_dc_add_func rv34_idct_dc_add;
rv40_weak_loop_filter_func rv40_weak_loop_filter[2]; rv40_weak_loop_filter_func rv40_weak_loop_filter[2];
rv40_strong_loop_filter_func rv40_strong_loop_filter[2]; rv40_strong_loop_filter_func rv40_strong_loop_filter[2];
rv40_loop_filter_strength_func rv40_loop_filter_strength[2]; rv40_loop_filter_strength_func rv40_loop_filter_strength[2];

View File

@ -35,21 +35,84 @@ SECTION .text
sar %1, 10 sar %1, 10
%endmacro %endmacro
%macro rv34_idct_dequant4x4_dc 1 %macro rv34_idct 1
cglobal rv34_idct_dequant4x4_%1_mmx2, 1, 2, 0 cglobal rv34_idct_%1_mmx2, 1, 2, 0
movsx r1, word [r0] movsx r1, word [r0]
IDCT_DC r1 IDCT_DC r1
movd mm0, r1 movd m0, r1
pshufw mm0, mm0, 0 pshufw m0, m0, 0
movq [r0+ 0], mm0 movq [r0+ 0], m0
movq [r0+16], mm0 movq [r0+16], m0
movq [r0+32], mm0 movq [r0+32], m0
movq [r0+48], mm0 movq [r0+48], m0
REP_RET REP_RET
%endmacro %endmacro
INIT_MMX INIT_MMX
%define IDCT_DC IDCT_DC_ROUND %define IDCT_DC IDCT_DC_ROUND
rv34_idct_dequant4x4_dc dc rv34_idct dc
%define IDCT_DC IDCT_DC_NOROUND %define IDCT_DC IDCT_DC_NOROUND
rv34_idct_dequant4x4_dc dc_noround rv34_idct dc_noround
; ff_rv34_idct_dc_add_mmx(uint8_t *dst, int stride, int dc);
cglobal rv34_idct_dc_add_mmx, 3, 3
; calculate DC
IDCT_DC_ROUND r2
pxor m1, m1
movd m0, r2
psubw m1, m0
packuswb m0, m0
packuswb m1, m1
punpcklbw m0, m0
punpcklbw m1, m1
punpcklwd m0, m0
punpcklwd m1, m1
; add DC
lea r2, [r0+r1*2]
movh m2, [r0]
movh m3, [r0+r1]
movh m4, [r2]
movh m5, [r2+r1]
paddusb m2, m0
paddusb m3, m0
paddusb m4, m0
paddusb m5, m0
psubusb m2, m1
psubusb m3, m1
psubusb m4, m1
psubusb m5, m1
movh [r0], m2
movh [r0+r1], m3
movh [r2], m4
movh [r2+r1], m5
RET
; ff_rv34_idct_dc_add_sse4(uint8_t *dst, int stride, int dc);
INIT_XMM
cglobal rv34_idct_dc_add_sse4, 3, 3, 6
; load data
IDCT_DC_ROUND r2
pxor m1, m1
; calculate DC
movd m0, r2
lea r2, [r0+r1*2]
movd m2, [r0]
movd m3, [r0+r1]
pshuflw m0, m0, 0
movd m4, [r2]
movd m5, [r2+r1]
punpcklqdq m0, m0
punpckldq m2, m3
punpckldq m4, m5
punpcklbw m2, m1
punpcklbw m4, m1
paddw m2, m0
paddw m4, m0
packuswb m2, m4
movd [r0], m2
pextrd [r0+r1], m2, 1
pextrd [r2], m2, 2
pextrd [r2+r1], m2, 3
RET

View File

@ -24,17 +24,23 @@
#include "libavcodec/dsputil.h" #include "libavcodec/dsputil.h"
#include "libavcodec/rv34dsp.h" #include "libavcodec/rv34dsp.h"
void ff_rv34_idct_dequant4x4_dc_mmx2(DCTELEM *block); void ff_rv34_idct_dc_mmx2(DCTELEM *block);
void ff_rv34_idct_dequant4x4_dc_noround_mmx2(DCTELEM *block); void ff_rv34_idct_dc_noround_mmx2(DCTELEM *block);
void ff_rv34_idct_dc_add_mmx(uint8_t *dst, int stride, int dc);
void ff_rv34_idct_dc_add_sse4(uint8_t *dst, int stride, int dc);
av_cold void ff_rv34dsp_init_x86(RV34DSPContext* c, DSPContext *dsp) av_cold void ff_rv34dsp_init_x86(RV34DSPContext* c, DSPContext *dsp)
{ {
#if HAVE_YASM #if HAVE_YASM
int mm_flags = av_get_cpu_flags(); int mm_flags = av_get_cpu_flags();
if (mm_flags & AV_CPU_FLAG_MMX)
c->rv34_idct_dc_add = ff_rv34_idct_dc_add_mmx;
if (mm_flags & AV_CPU_FLAG_MMX2) { if (mm_flags & AV_CPU_FLAG_MMX2) {
c->rv34_inv_transform_dc_tab[0] = ff_rv34_idct_dequant4x4_dc_mmx2; c->rv34_inv_transform_dc_tab[0] = ff_rv34_idct_dc_mmx2;
c->rv34_inv_transform_dc_tab[1] = ff_rv34_idct_dequant4x4_dc_noround_mmx2; c->rv34_inv_transform_dc_tab[1] = ff_rv34_idct_dc_noround_mmx2;
} }
if (mm_flags & AV_CPU_FLAG_SSE4)
c->rv34_idct_dc_add = ff_rv34_idct_dc_add_sse4;
#endif #endif
} }