virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2026-06-19 19:03:00 +02:00
Code Issues Releases Activity
17,735 Commits 39 Branches 425 Tags
6166516d1f0df6cc8f96e7685e78ef75541b9698
Commit Graph
100 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Loren Merritt 6166516d1f re-enable mid_pred asm on x86_64. (broke in r16681) 
Originally committed as revision 17058 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2009-02-08 17:44:04 +00:00
Loren Merritt 5a446bc88e fix an overflow in vc1 intra overlap filter 
Originally committed as revision 16845 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2009-01-28 11:12:19 +00:00
Loren Merritt 1cca8d2423 flatten an array, since gcc fails at optimizing multidimensional arrays 
h264_idct8_add_c: 780 -> 735 cycles on conroe

Originally committed as revision 16307 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-12-25 01:20:37 +00:00
Loren Merritt 00794ed0d9 1.6x faster pred*_vertical_add_c 
Originally committed as revision 16202 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-12-18 01:05:06 +00:00
Loren Merritt 5fecfb7d58 clear_block mmx 
Originally committed as revision 16045 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-12-10 21:35:17 +00:00
Loren Merritt 5fac277602 fix progressive jpeg: 
support refinement passes
remove intermediate clipping
remove redundant idct

Originally committed as revision 16044 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-12-10 21:26:00 +00:00
Loren Merritt 7fadc32de4 document some dsp alignments 
Originally committed as revision 14935 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-24 04:03:02 +00:00
Loren Merritt 2272e04aa7 document some dsp alignments 
Originally committed as revision 14771 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-15 00:14:07 +00:00
Loren Merritt 7ca7d5fae0 file which should have been added in r14749 
Originally committed as revision 14751 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-14 05:00:25 +00:00
Loren Merritt 75ac287517 missing prototype 
Originally committed as revision 14750 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-14 04:41:02 +00:00
Loren Merritt ebceaa1cd5 gcc chokes on the 7 registers needed for float_to_int16_interleave6 (even inside HAVE_7REGS), so write it in yasm 
Originally committed as revision 14749 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-14 04:40:46 +00:00
Loren Merritt ee46753739 gcc chokes on xmm constraints, so pessimize int32_to_float_fmul_scalar_sse a little 
Originally committed as revision 14748 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-14 04:39:59 +00:00
Loren Merritt 675872382f special case 6 channel version of float_to_int16_interleave 
5% faster ac3

Originally committed as revision 14744 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-13 23:36:37 +00:00
Loren Merritt 911e21a306 simd int->float 
20% faster ac3 if downmixing, 15% if not

Originally committed as revision 14743 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-13 23:35:40 +00:00
Loren Merritt ac2e556456 simd downmix 
13% faster ac3 if downmixing

Originally committed as revision 14742 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-13 23:33:48 +00:00
Loren Merritt 45d9d61889 don't mark the delayed samples for upmixing if they haven't been downmixed 
Originally committed as revision 14741 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-13 23:30:53 +00:00
Loren Merritt 3b6516f7e7 oops, non-normalized downmix to mono contaminated add_bias for non-simd float-to-int. 
Originally committed as revision 14740 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-13 23:30:28 +00:00
Loren Merritt 557ac0c448 optimize ac3_downmix. 
1.3x faster 5.1->stereo, 1.9x faster 5.1->mono.

Originally committed as revision 14719 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 23:45:46 +00:00
Loren Merritt 81b060faf9 align the audio decoding buffer, since some codecs write to it with simd 
Originally committed as revision 14707 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 05:59:12 +00:00
Loren Merritt 72745cff20 use float_to_int16_interleave in ac3 
Originally committed as revision 14706 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 03:01:17 +00:00
Loren Merritt 916d5d6c32 use imdct_half in ac3 
Originally committed as revision 14705 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 01:30:24 +00:00
Loren Merritt 862b98d42c cosmetics in dsp init 
Originally committed as revision 14704 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 00:51:45 +00:00
Loren Merritt d46ac5bfde mdct wrapper function to match fft 
Originally committed as revision 14703 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 00:38:30 +00:00
Loren Merritt 0a570e826d remove mdct tmp buffer 
Originally committed as revision 14702 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 00:36:36 +00:00
Loren Merritt e8dd7b0c9a remove a float->double->float conversion. 
1.8x faster imdct_c with fpmath=sse, no change with x87

Originally committed as revision 14701 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 00:34:08 +00:00
Loren Merritt 46803f4f67 optimize imdct_half: 
remove tmp buffer.
skip fft reinterleave pass, leaving data in a format more convenient for simd.
merge post-rotate with post-reorder.

Originally committed as revision 14700 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 00:33:34 +00:00
Loren Merritt 49c0dd754c indent 
Originally committed as revision 14699 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 00:27:21 +00:00
Loren Merritt 5d0ddd1a9f split-radix FFT 
c is 1.9x faster than previous c (on various x86 cpus), sse is 1.6x faster than previous sse.

Originally committed as revision 14698 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-12 00:26:58 +00:00
Loren Merritt bafad220a7 import yasm macros from x264 
Originally committed as revision 14697 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-11 23:54:09 +00:00
Loren Merritt 304d586de6 yasm buildsystem 
Originally committed as revision 14696 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-08-11 23:53:31 +00:00
Loren Merritt 8a37920c9e r14205 broke vorbis float_to_int on ppc 
Originally committed as revision 14466 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-07-29 14:17:13 +00:00
Loren Merritt 5eb0f2a425 float_to_int16_interleave: change src to an array of pointers instead of assuming it's contiguous. 
this has no immediate effect, but will allow it to be used in more codecs.

Originally committed as revision 14252 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-07-16 00:50:12 +00:00
Loren Merritt 633d9def9d vorbis: reduce memory by reusing buffers 
Originally committed as revision 14251 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-07-15 23:04:28 +00:00
Loren Merritt 1a32536737 unroll another common case in vorbis_residue_decode 
Originally committed as revision 14250 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-07-15 23:01:15 +00:00
Loren Merritt 4342a7f30b 10l, float_to_int16_interleave_sse/3dnow wrote the wrong samples 
Originally committed as revision 14236 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-07-15 04:11:30 +00:00
Loren Merritt b9fa32082c exploit mdct symmetry 
2% faster vorbis on conroe, k8. 7% on celeron.

Originally committed as revision 14207 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-07-13 15:03:58 +00:00
Loren Merritt f27e1d645e simplify vorbis windowing 
Originally committed as revision 14205 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-07-13 14:56:01 +00:00
Loren Merritt 359bbdabf3 rename me=tesa 
Originally committed as revision 13661 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-06-05 02:20:15 +00:00
Loren MerrittandMike Melanson 96275520a3 Fix H.264 interframe decoding when compiling with icc. Patch by Loren 
Merritt:

"It seems that icc copies the constants from their global var onto the
stack, at which point they're not aligned, hence the crash.
[This change] really shouldn't mean anything different, but maybe it'll
confuse icc into not performing that 'optimization'."

Originally committed as revision 12772 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-04-08 23:49:34 +00:00
Loren Merritt ce53144bac h264 chroma mc ssse3 
width8: 180->92, width4: 78->63 cycles (core2)

Originally committed as revision 12661 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-04-01 04:51:28 +00:00
Loren Merritt be0995b166 more aspect ratios added in one of the ammendments 
Originally committed as revision 12282 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-03-01 01:58:25 +00:00
Loren Merritt 042e0adda9 decode mng color decorrelation 
Originally committed as revision 12228 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-26 10:21:33 +00:00
Loren Merritt 00829b9bc1 simplify 
Originally committed as revision 12224 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-26 07:35:56 +00:00
Loren Merritt 4a5b619df7 png filters 
Originally committed as revision 12195 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-25 10:34:23 +00:00
Loren Merritt bd58e7f730 remove a memcpy 
Originally committed as revision 12193 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-25 05:46:53 +00:00
Loren Merritt 917f55cc03 cleaner way of writing packed-byte constants. 
suggested by Uoti Urpala

Originally committed as revision 12177 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-22 16:53:26 +00:00
Loren Merritt 31304587da indent 
Originally committed as revision 12166 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-21 07:55:42 +00:00
Loren Merritt 469bd7b150 pseudo-simd add_bytes and diff_bytes 
2x faster than scalar in 32bit, 4x faster in 64bit (as opposed to 8x in mmx)

Originally committed as revision 12165 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-21 07:54:46 +00:00
Loren Merritt 4a9ca0a279 simd and unroll png_filter_row 
cycles per 1000 pixels on core2:
left: 9211->5170
top: 9283->2138
avg: 12215->7611
paeth: 64024->17360
overall rgb png decoding speed: +45%
overall greyscale png decoding speed: +6%

Originally committed as revision 12164 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-21 07:10:46 +00:00
Loren Merritt ec199cc94c asm argument that might be in memory needs a size 
Originally committed as revision 11890 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-10 01:45:42 +00:00
Loren Merritt e1813a2f1e 30% faster ff_vorbis_floor1_render_list, 3% faster overall 
Originally committed as revision 11883 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-09 13:53:15 +00:00
Loren Merritt 2c70770e33 use fewer registers in apply_welch_window_sse2 
Originally committed as revision 11882 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-09 05:29:47 +00:00
Loren Merritt 1d67b037f7 sse2 h264 motion compensation. not new code, just separate out the cases that didn't need ssse3. 
Originally committed as revision 11877 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-06 12:32:31 +00:00
Loren Merritt 20d565be6d put loop counter in a register if possible. makes some of the qpel functions 3% faster. 
Originally committed as revision 11876 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-06 04:44:21 +00:00
Loren Merritt 7080ec2937 fix aliasing warnings. simpler too. 
Originally committed as revision 11875 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-06 04:14:07 +00:00
Loren Merritt a2b7bc8e71 constant was excessively aligned 
Originally committed as revision 11874 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-06 03:51:53 +00:00
Loren Merritt ddf969705f ssse3 h264 motion compensation. 
25% faster tham mmx on core2, 35% if you discount fullpel, 4% overall decoding.

Originally committed as revision 11871 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-05 11:22:55 +00:00
Loren Merritt b64dfbb8d2 add qpel rounder once during hv rather than twice during hv and whatever it's averaged with 
Originally committed as revision 11870 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-05 03:58:13 +00:00
Loren Merritt fa9b873e08 clean up an ugliness introduced in r11826. this syntax will require fewer changes when adding future sse2 code. 
Originally committed as revision 11868 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-05 01:16:48 +00:00
Loren Merritt b2f775860b reduce code duplication 
Originally committed as revision 11863 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-04 16:20:46 +00:00
Loren Merritt b313e8159c avg_pixels4_mmx2 
Originally committed as revision 11829 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-03 17:04:33 +00:00
Loren Merritt 6c01d0069d use mmx2/3dnow avg functions in avg_qpel*_mc00 
Originally committed as revision 11828 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-03 16:19:37 +00:00
Loren Merritt ed5d7a531c ff_h264_idct8_add_sse2. 
compared to mmx, 217->126 cycles on core2, 262->220 on k8.

Originally committed as revision 11826 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-03 07:05:11 +00:00
Loren Merritt 51f0ac65df remove some movq in ff_h264_idct8_add_mmx. 225->217 cycles on core2. 
Originally committed as revision 11825 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-03 03:21:47 +00:00
Loren Merritt ffbc5e04ce "swap first two elements of L1 if L0 and L1 are identical" applies after splitting the lists into fields 
Originally committed as revision 11808 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-02 10:54:20 +00:00
Loren Merritt 02f7695b0b fix mbaff diagonal neighbor mv 
Originally committed as revision 11807 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-02 10:53:15 +00:00
Loren Merritt 50b3ab0fa5 mbaff spatial direct 
Originally committed as revision 11806 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2008-02-02 10:51:56 +00:00
Loren Merritt 90f0344165 remove a branch from an inner loop. 
1% faster flac encoding.

Originally committed as revision 10865 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-10-27 13:39:19 +00:00
Loren Merritt 286127c5dd div -> mul 
Originally committed as revision 10630 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-10-01 05:56:30 +00:00
Loren Merritt b8de342919 simplify lpc 
Originally committed as revision 10628 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-30 03:36:13 +00:00
Loren Merritt d1a5c4216c 20% faster lpc, 6% overall flac decoding 
Originally committed as revision 10627 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-30 03:01:56 +00:00
Loren Merritt 08965b22e2 replace FIR with finite differences. 
3x faster decode_subframe_fixed().
overall flac decoding: 10% faster if file was encoded with fixed predictors.

Originally committed as revision 10626 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-30 02:12:03 +00:00
Loren Merritt bfdd5bc1f1 unroll finite differences to avoid swapping registers. 
1.5x faster encode_residual_fixed().
overall flac encoding: 3% faster at compression_levels 0-2.

Originally committed as revision 10625 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-29 23:44:34 +00:00
Loren Merritt a309dce75f encode_residual_fixed(): replace FIR with finite differences. 
4x faster order 2, 3.5x order 3, 3x order 4.
overall flac encoding: 35% faster at compression_levels 0-2, no effect at higher levels.

Originally committed as revision 10624 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-29 23:21:57 +00:00
Loren Merritt 6810b93a81 sse2 version of compute_autocorr(). 
4x faster than c (somehow, even though doubles only allow 2x simd).
overal flac encoding: 15-50% faster on core2, 4-11% on k8, 3-13% on p4.

Originally committed as revision 10621 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-29 22:31:18 +00:00
Loren Merritt 19b9c7cc66 was computing one more autocorrelation coefficient that was actually used 
Originally committed as revision 10613 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-29 07:43:52 +00:00
Loren Merritt 14bd2a9f25 replace brute force find_optimal_param() with a closed-form solution. 
overall flac encoding: 4-15% faster.
output is not identical to the previous algorithm due to occasional rounding
errors, but the differece is less than .0005% bitrate.

Originally committed as revision 10612 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-29 05:41:27 +00:00
Loren Merritt f6215b1b06 gcc isn't smart enough to factor out duplicate stores 
Originally committed as revision 10611 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-29 01:54:25 +00:00
Loren Merritt 28e968c4a4 2.5x faster compute_autocorr() 
overall flac encoding: 15-50% faster on core2, 8-30% on k8, 2-20% on p4 (depending on compression_level)

Originally committed as revision 10606 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-28 06:06:18 +00:00
Loren Merritt 815d96a65d oops, revert unrelated change 
Originally committed as revision 10597 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-27 03:16:25 +00:00
Loren Merritt dc44d4ad64 unroll encode_residual_lpc(). speedup varies between 1.2x and 1.8x depending on lpc order. 
Originally committed as revision 10596 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-27 02:42:00 +00:00
Loren Merritt 0d2caa37c5 oops, potential overflow on really large blocks 
Originally committed as revision 10587 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-25 23:34:48 +00:00
Loren Merritt bd49d4fd49 optimize decode_subframe_lpc() 
50%/67%/43% faster on core2/k8/p4, making flac decoding overall 24%/25%/11% faster

Originally committed as revision 10586 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-25 23:31:13 +00:00
Loren Merritt f74471e043 optimize encode_residual_lpc() 
37%/45%/90% faster on core2/k8/p4, making flac encoding overall 15%/17%/40% faster at compression_level>=8 (less at low levels).

Originally committed as revision 10585 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-09-25 23:30:29 +00:00
Loren Merritt 2e1ad4a28e simd gmc wasn't impossible. 
Originally committed as revision 10191 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-08-23 04:56:29 +00:00
Loren Merritt 2bf4aa2e93 prevent huffyuv from generating codewords of length 32. (regression in r9069) 
Originally committed as revision 9921 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-08-05 00:29:02 +00:00
Loren MerrittandJustin Ruggles afb46fc092 AC-3 decoder, soc revision 148, Sep 28 18:38:27 2006 UTC by banan 
Enable sse imdct, patch by Loren Merrit.

Originally committed as revision 9672 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-07-14 16:05:39 +00:00
Loren Merritt 8be6dac8dc merge huffman tables for rgb too 
Originally committed as revision 9124 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-25 00:16:14 +00:00
Loren Merritt f45fcba303 cosmetics (reduce nesting) 
Originally committed as revision 9119 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-24 19:04:19 +00:00
Loren Merritt acc0e0c8b6 use sparse huffman tables. 1.5% faster huffyuv decoding. 
Originally committed as revision 9118 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-24 17:40:02 +00:00
Loren Merritt b613bacca9 add init_vlc_sparse(). faster than init_vlc() if there are lots of holes in the tables. 
Originally committed as revision 9117 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-24 17:38:56 +00:00
Loren Merritt 19716d6478 merge huffman tables so that we read 2 symbols at a time. 30% faster huffyuv decoding. 
Originally committed as revision 9104 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-23 09:02:07 +00:00
Loren Merritt 98ef8c324c change brute force search to min-heap. 3.6x faster generate_len_table, 8% faster ffvhuff encoding. 
Originally committed as revision 9069 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-19 02:32:59 +00:00
Loren Merritt bf1a5a46d6 cosmetics: macrofy duplicate code in huffyuv 
Originally committed as revision 9055 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-18 08:58:30 +00:00
Loren Merritt 390f9f3127 r3938 broke 2pass huffyuv (not that anyone uses it) 
Originally committed as revision 9054 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-18 08:56:46 +00:00
Loren Merritt ff506a906e sse2 & ssse3 versions of dct_quantize. 
core2: mmx2=154 sse2=73 ssse3=66 (cycles)
k8: mmx2=179 sse2=149
p4: mmx2=284 sse2=194

Originally committed as revision 9003 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-12 05:55:09 +00:00
Loren Merritt 8e38071003 configure check for whether we can compile ssse3 
Originally committed as revision 9002 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-12 03:20:41 +00:00
Loren Merritt 1edbfe1994 factor sum_abs_dctelem out of dct_sad, and simd it. 
sum_abs_dctelem_* alone:
core2: c=186 mmx2=39 sse2=21 ssse3=13 (cycles)
k8: c=163 mmx2=33 sse2=31
p4: c=370 mmx2=60 sse2=60
 dct_sad including sum_abs_dctelem_*:
core2: c=405 mmx2=258 sse2=240 ssse3=232
k8: c=624 mmx2=394 sse2=392
p4: c=849 mmx2=556 sse2=556

Originally committed as revision 9001 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-12 02:41:25 +00:00
Loren Merritt 561f940c03 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels. 
core2: before mmx2=193 cycles. after mmx2=174 sse2=122 ssse3=115 (cycles).
k8: before mmx2=205. after mmx2=184 sse2=180.
p4: before mmx2=342. after mmx2=314 sse2=309.

Originally committed as revision 9000 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-12 01:16:06 +00:00
Loren Merritt 5b67307a68 fix a crash with svq1 + sse2. 
EMU_EDGE broke memory alignment when the desired alignment is >32 bytes.

Originally committed as revision 8999 to svn://svn.ffmpeg.org/ffmpeg/trunk
 2007-05-12 00:31:48 +00:00