theora.git
7 years agoFix r17279 so it actually works. theora-gumboot
Tim Terriberry [Tue, 8 Jun 2010 13:42:52 +0000 (13:42 +0000)]
Fix r17279 so it actually works.

svn path=/branches/theora-gumboot/; revision=17284

7 years agoAdd MMXEXT and SSE2 versions of the MC functions that don't add a residual.
Tim Terriberry [Mon, 7 Jun 2010 22:40:34 +0000 (22:40 +0000)]
Add MMXEXT and SSE2 versions of the MC functions that don't add a residual.
These will need to be reorganized later, with proper detection, etc.

svn path=/branches/theora-gumboot/; revision=17280

7 years agoEnsure 16-byte alignment for the reference buffers, because we need it now.
Tim Terriberry [Mon, 7 Jun 2010 21:55:14 +0000 (21:55 +0000)]
Ensure 16-byte alignment for the reference buffers, because we need it now.

svn path=/branches/theora-gumboot/; revision=17279

7 years agoGet GCC's derranged handling of dct_coeffs out of the way using inline assembly.
gumboot [Sat, 5 Jun 2010 01:58:19 +0000 (01:58 +0000)]
Get GCC's derranged handling of dct_coeffs out of the way using inline assembly.
Put in the code I thought should make things faster.
Fix a bug that would have mis-handled an all-zero intra block.

svn path=/branches/theora-gumboot/; revision=17274

7 years agoOptimise bits and pieces. Add 16x8 fragment copy/predict.
gumboot [Fri, 4 Jun 2010 20:30:38 +0000 (20:30 +0000)]
Optimise bits and pieces.  Add 16x8 fragment copy/predict.
Also capture this apparent local maximum in performance in removing cruft from oc_dec_frags_recon_mcu_plane().  There are obvious things to clean up remaining (ignoring the blindingly obvious work of not calling a quad function for individual fragments), but when they are cleaned up the code slows down.

svn path=/branches/theora-gumboot/; revision=17273

7 years agoSeparate and move iDCT after MC.
gumboot [Fri, 4 Jun 2010 13:43:43 +0000 (13:43 +0000)]
Separate and move iDCT after MC.
Clean some muck out of oc_dec_mv_unpack_and_frag_modes_fill().

svn path=/branches/theora-gumboot/; revision=17272

7 years agoGet mb_modes and motion vectors into a more easily indexed format so that chroma...
gumboot [Sat, 29 May 2010 23:46:16 +0000 (23:46 +0000)]
Get mb_modes and motion vectors into a more easily indexed format so that chroma can access the same memory as luma.  Push other things forward so that the old tables needn't be populated.  Unroll more code for greater ugliness.

svn path=/branches/theora-gumboot/; revision=17257

7 years agoFinally start going back the other way -- oc_dec_mb_modes_unpack() becomes simpler.
gumboot [Sat, 29 May 2010 14:53:16 +0000 (14:53 +0000)]
Finally start going back the other way -- oc_dec_mb_modes_unpack() becomes simpler.

svn path=/branches/theora-gumboot/; revision=17255

7 years agoMerge in latest changes from theora-ptalarbvorm, and patch in derf's oc_int_fragx2_co...
gumboot [Sat, 29 May 2010 03:46:41 +0000 (03:46 +0000)]
Merge in latest changes from theora-ptalarbvorm, and patch in derf's oc_int_fragx2_copy2_sse2() just for fun.

Really need to focus on collating the information I need in the layout I need, and clearing out all the tentative, baby-steps cruft.

svn path=/branches/theora-gumboot/; revision=17252

7 years agoPut this somewhere safe before I start to make a serious mess.
gumboot [Sat, 29 May 2010 02:40:13 +0000 (02:40 +0000)]
Put this somewhere safe before I start to make a serious mess.

svn path=/branches/theora-gumboot/; revision=17251

7 years agoAdd oc_state_mb_recon() and call it sometimes. It doesn't really do a macroblock...
gumboot [Sat, 22 May 2010 22:50:43 +0000 (22:50 +0000)]
Add oc_state_mb_recon() and call it sometimes.  It doesn't really do a macroblock, though.  Just a quad of blocks.

Now to think about what information we really need, and where to find it, and whether or not it should be put somewhere more convenient.

Also need to accelerate some of the wide operations that have been opened up, and to recognise cases (most obviously 444) where chroma can do quads too.

svn path=/branches/theora-gumboot/; revision=17241

7 years agosync theora-gumboot branch again
gumboot [Sat, 22 May 2010 10:39:47 +0000 (10:39 +0000)]
sync theora-gumboot branch again

svn path=/branches/theora-gumboot/; revision=17240

7 years agoIntegrate latest changes from ptalarbvorm, thereby exaggerating the damage I've done...
gumboot [Sun, 16 May 2010 12:46:19 +0000 (12:46 +0000)]
Integrate latest changes from ptalarbvorm, thereby exaggerating the damage I've done while messing around.

svn path=/branches/theora-gumboot/; revision=17221

7 years agoSet up the per-superblock bitmap of coded blocks, and use it to reduce the damage...
gumboot [Fri, 23 Apr 2010 18:16:54 +0000 (18:16 +0000)]
Set up the per-superblock bitmap of coded blocks, and use it to reduce the damage done so far.

svn path=/branches/theora-gumboot/; revision=17166

7 years agoClean up change 17164 a little bit.
gumboot [Fri, 23 Apr 2010 16:25:21 +0000 (16:25 +0000)]
Clean up change 17164 a little bit.

svn path=/branches/theora-gumboot/; revision=17165

7 years agoReplace oc_dec_frags_recon_mcu_plane() with something ugly, but vaguely representativ...
gumboot [Fri, 23 Apr 2010 15:57:53 +0000 (15:57 +0000)]
Replace oc_dec_frags_recon_mcu_plane() with something ugly, but vaguely representative of The Plan.

svn path=/branches/theora-gumboot/; revision=17164

7 years agoTake a shiny new branch to play with.
gumboot [Fri, 23 Apr 2010 15:49:51 +0000 (15:49 +0000)]
Take a shiny new branch to play with.

svn path=/branches/theora-gumboot/; revision=17163