generic-library/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Altivec unscaled YV12 -> packed YUV patch by (Romain Dolbeau <dolbeau at irisa dot fr>)

Browse Source 
Originally committed as revision 12769 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc
This commit is contained in:
Romain Dolbeau 2004-07-06 13:52:56 +00:00 committed by Michael Niedermayer
parent 61b3b639e7
commit b71cf33c9b
2 changed files with 159 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
postproc/swscale.c
View File

@ -1909,6 +1909,18 @@ SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int
}
}
#ifdef HAVE_ALTIVEC
if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
((srcFormat == IMGFMT_YV12 &&
(dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
// unscaled YV12 -> packed YUV, we want speed
if (dstFormat == IMGFMT_YUY2)
c->swScale= yv12toyuy2_unscaled_altivec;
else
c->swScale= yv12touyvy_unscaled_altivec;
}
#endif
/* simple copy */
if( srcFormat == dstFormat
|| (isPlanarYUV(srcFormat) && isGray(dstFormat))

147
postproc/swscale_altivec_template.c
View File

@ -383,3 +383,150 @@ static inline void hScale_altivec_real(int16_t *dst, int dstW, uint8_t *src, int
}
}
}
static inline int yv12toyuy2_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
// yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
uint8_t *ysrc = src[0];
uint8_t *usrc = src[1];
uint8_t *vsrc = src[2];
const int width = c->srcW;
const int height = srcSliceH;
const int lumStride = srcStride[0];
const int chromStride = srcStride[1];
const int dstStride = dstStride_a[0];
const vector unsigned char yperm = vec_lvsl(0, ysrc);
const int vertLumPerChroma = 2;
register unsigned int y;
/* this code assume:
1) dst is 16 bytes-aligned
2) dstStride is a multiple of 16
3) width is a multiple of 16
4) lum&chrom stride are multiple of 8
*/
for(y=0; y<height; y++)
{
int i;
for (i = 0; i < width - 31; i+= 32) {
const unsigned int j = i >> 1;
vector unsigned char v_yA = vec_ld(i, ysrc);
vector unsigned char v_yB = vec_ld(i + 16, ysrc);
vector unsigned char v_yC = vec_ld(i + 32, ysrc);
vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
vector unsigned char v_uA = vec_ld(j, usrc);
vector unsigned char v_uB = vec_ld(j + 16, usrc);
vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
vector unsigned char v_vA = vec_ld(j, vsrc);
vector unsigned char v_vB = vec_ld(j + 16, vsrc);
vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b);
vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b);
vec_st(v_yuy2_0, (i << 1), dst);
vec_st(v_yuy2_1, (i << 1) + 16, dst);
vec_st(v_yuy2_2, (i << 1) + 32, dst);
vec_st(v_yuy2_3, (i << 1) + 48, dst);
}
if (i < width) {
const unsigned int j = i >> 1;
vector unsigned char v_y1 = vec_ld(i, ysrc);
vector unsigned char v_u = vec_ld(j, usrc);
vector unsigned char v_v = vec_ld(j, vsrc);
vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
vec_st(v_yuy2_0, (i << 1), dst);
vec_st(v_yuy2_1, (i << 1) + 16, dst);
}
if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
{
usrc += chromStride;
vsrc += chromStride;
}
ysrc += lumStride;
dst += dstStride;
}
return srcSliceH;
}
static inline int yv12touyvy_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
// yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
uint8_t *ysrc = src[0];
uint8_t *usrc = src[1];
uint8_t *vsrc = src[2];
const int width = c->srcW;
const int height = srcSliceH;
const int lumStride = srcStride[0];
const int chromStride = srcStride[1];
const int dstStride = dstStride_a[0];
const int vertLumPerChroma = 2;
const vector unsigned char yperm = vec_lvsl(0, ysrc);
register unsigned int y;
/* this code assume:
1) dst is 16 bytes-aligned
2) dstStride is a multiple of 16
3) width is a multiple of 16
4) lum&chrom stride are multiple of 8
*/
for(y=0; y<height; y++)
{
int i;
for (i = 0; i < width - 31; i+= 32) {
const unsigned int j = i >> 1;
vector unsigned char v_yA = vec_ld(i, ysrc);
vector unsigned char v_yB = vec_ld(i + 16, ysrc);
vector unsigned char v_yC = vec_ld(i + 32, ysrc);
vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
vector unsigned char v_uA = vec_ld(j, usrc);
vector unsigned char v_uB = vec_ld(j + 16, usrc);
vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
vector unsigned char v_vA = vec_ld(j, vsrc);
vector unsigned char v_vB = vec_ld(j + 16, vsrc);
vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2);
vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2);
vec_st(v_uyvy_0, (i << 1), dst);
vec_st(v_uyvy_1, (i << 1) + 16, dst);
vec_st(v_uyvy_2, (i << 1) + 32, dst);
vec_st(v_uyvy_3, (i << 1) + 48, dst);
}
if (i < width) {
const unsigned int j = i >> 1;
vector unsigned char v_y1 = vec_ld(i, ysrc);
vector unsigned char v_u = vec_ld(j, usrc);
vector unsigned char v_v = vec_ld(j, vsrc);
vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
vec_st(v_uyvy_0, (i << 1), dst);
vec_st(v_uyvy_1, (i << 1) + 16, dst);
}
if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
{
usrc += chromStride;
vsrc += chromStride;
}
ysrc += lumStride;
dst += dstStride;
}
return srcSliceH;
}