diff options
| author | Arpi <arpi@thot.banki.hu> | 2001-08-03 18:33:03 +0000 |
|---|---|---|
| committer | Arpi <arpi@thot.banki.hu> | 2001-08-03 18:33:03 +0000 |
| commit | 4af7bcc1857e8abfa7ae9a8e3c54c93723219438 (patch) | |
| tree | 22217576fcce734d75eaac61ce399839ec168044 /libavcodec/i386 | |
| parent | 2d6d0c1d66accc7976325190f17dc93ebd9b665d (diff) | |
| download | ffmpeg-4af7bcc1857e8abfa7ae9a8e3c54c93723219438.tar.gz | |
MMX/MMXEXT iDCT support, using external functions currently defined in libmpeg2
Gives average 13-20% mpeg decoding speedup on x86 systems.
Originally committed as revision 30 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/i386')
| -rw-r--r-- | libavcodec/i386/dsputil_mmx.c | 28 |
1 files changed, 28 insertions, 0 deletions
diff --git a/libavcodec/i386/dsputil_mmx.c b/libavcodec/i386/dsputil_mmx.c index d4a07c3a77..d9028cf2b9 100644 --- a/libavcodec/i386/dsputil_mmx.c +++ b/libavcodec/i386/dsputil_mmx.c @@ -29,6 +29,16 @@ int pix_abs16x16_x2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h); int pix_abs16x16_y2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h); int pix_abs16x16_xy2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h); +#ifdef USE_MMX_IDCT +/* external functions, defined in libmpeg2 */ +void mmx_idct(DCTELEM *block); +void mmxext_idct(DCTELEM *block); +/* this should be in dsputil.h? -- A'rpi */ +extern UINT8 ff_alternate_horizontal_scan[64]; +extern UINT8 ff_alternate_vertical_scan[64]; +extern UINT8 zigzag_direct[64]; +#endif + /* pixel operations */ static const unsigned long long int mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001; static const unsigned long long int mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002; @@ -1039,5 +1049,23 @@ void dsputil_init_mmx(void) sub_pixels_tab[1] = sub_pixels_x2_3dnow; sub_pixels_tab[2] = sub_pixels_y2_3dnow; } + +#ifdef USE_MMX_IDCT + /* use MMX / MMXEXT iDCT code from libmpeg2 */ + //printf("LIBAVCODEC: Using MMX%s iDCT code\n",(mm_flags & MM_MMXEXT)?"EXT":""); + ff_idct = (mm_flags & MM_MMXEXT) ? mmxext_idct : mmx_idct; + /* the mmx/mmxext idct uses a reordered input, so we patch scan tables */ + { int i,j; + for (i = 0; i < 64; i++) { + j = zigzag_direct[i]; + zigzag_direct[i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2); + j = ff_alternate_horizontal_scan[i]; + ff_alternate_horizontal_scan[i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2); + j = ff_alternate_vertical_scan[i]; + ff_alternate_vertical_scan[i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2); + } + } +#endif + } } |