generic-library/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
69,730 Commits 24 Branches 241 Tags 173 MiB
3e2714992b
Commit Graph

1917 Commits

Author SHA1 Message Date
James Almer
1679d68dbf x86/hevc_mc: optimize AVX2 mc functions 
Before
40766 decicycles in ff_hevc_put_hevc_qpel_h64_8_avx2, 8192 runs, 0 skips

After
37975 decicycles in ff_hevc_put_hevc_qpel_h64_8_avx2, 8192 runs, 0 skips

Reviewed-by: Christophe Gisquet <christophe.gisquet@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-12 13:21:58 -03:00
James Almer
14b44c1614 x86/hevc_sao: make sao_edge_filter_{10,12} work on x86_32 
Reviewed-by: Michael Niedermayer <michaelni@gmx.at>
Reviewed-by: Christophe Gisquet <christophe.gisquet@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-12 13:21:30 -03:00
James Almer
06fe6dfe12 x86/hevc_sao: make sao_band_filter work on x86_32 
Reviewed-by: Christophe Gisquet <christophe.gisquet@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-09 20:41:21 -03:00
Christophe Gisquet
b61b9e4919 x86: hevc_mc: remove lea in EPEL_LOAD 
The second parameter to the macro is always an immediate address,
so no lea is needed.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-08 22:19:35 +01:00
Christophe Gisquet
4919b38421 x86: hevc_mc: fewer gpr autoloads for _v filters 
In that case, it's just to load my, but mx/r3src is not used.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-08 22:19:34 +01:00
James Almer
92d903afaa x86/vp9dsp: fix clobbering of xmm6 on IDCT sse2 functions 
Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-08 00:50:39 -03:00
Christophe Gisquet
626d6184ce x86: lavc/hevc_mc: fix comments 
The width parameter is now completely at the back, and actually
never used. This helps understanding the actual parameter list.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-07 20:52:03 +01:00
Christophe Gisquet
ed450d4acf x86: lavc: share more constant through defines 
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-07 17:48:14 +01:00
Christophe Gisquet
691b7f5e9e lavc/lossless_audiodsp: revert various commits 
Their intent was to make the DSP work with wmalossless pro.
The later was fixed to work with the DSP.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-07 15:15:19 +01:00
Christophe Gisquet
9dc45d1f42 x86: lavc: share more constants 
Reviewed-by: "Ronald S. Bultje" <rsbultje@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-06 23:35:02 +01:00
Mickaël Raulet
6ecc3fd612 x86/hevc_mc: use aligned loads 
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-06 21:38:00 +01:00
James Almer
383fddeec6 x86/lossless_audiodsp: fix compilation with --disable-yasm 
Reviewed-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-06 17:30:17 -03:00
James Almer
aea29a891f x86/hevc_sao: fix loading of RIP address 
pb_eo must be handled as a rip relative address for MSVC64, so an
intermediate register is needed. Should fix link failures.

Suggested by Hendrik Leppkes and Christophe Gisquet.

Tested-By: Hendrik Leppkes <h.leppkes@gmail.com>
Reviewed-by: Christophe Gisquet <christophe.gisquet@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-06 15:06:15 -03:00
Mickaël Raulet
bcb0925115 x86/hevc: use CLIPW macro when possible 
Conflicts:
	libavcodec/x86/hevc_mc.asm

Reviewed-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-06 17:38:47 +01:00
Christophe Gisquet
5eedd36df1 x86: hevc_mc: use epel_hv 16-wide function 
The epel_hv functions were still relying on only epel_hv 8-wide
being the maximum width instanciated.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-06 17:37:56 +01:00
Pierre Edouard Lepere
a0d1300f71 x86: hevc_mc: add AVX2 optimizations 
before
33304 decicycles in luma_bi_1, 523066 runs, 1222 skips
38138 decicycles in luma_bi_2, 523427 runs, 861 skips
13490 decicycles in luma_uni, 516138 runs, 8150 skips
after
20185 decicycles in luma_bi_1, 519970 runs, 4318 skips
24620 decicycles in luma_bi_2, 521024 runs, 3264 skips
10397 decicycles in luma_uni, 515715 runs, 8573 skips

Conflicts:
	libavcodec/x86/hevc_mc.asm
	libavcodec/x86/hevcdsp_init.c

Reviewed-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-06 17:20:47 +01:00
Michael Niedermayer
a6c2c8fe3f Revert "avcodec/x86/lossless_audiodsp: Make scalarproduct_and_madd_int16 prototypes more similar" 
This reverts commit 3b4ffba3af.

Unbreaks the SSSE3 code on mingw32

Conflicts:

	libavcodec/x86/lossless_audiodsp.asm

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-06 02:31:45 +01:00
Michael Niedermayer
f1214763af avcodec/x86/lossless_audiodsp: Move order&8 fallback into C code 
This is simpler and more robust, and fixes mismatching XMM save restore
mismatches

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-06 02:18:54 +01:00
Michael Niedermayer
3b4ffba3af avcodec/x86/lossless_audiodsp: Make scalarproduct_and_madd_int16 prototypes more similar 
This is needed as the mmx code is used as fallback from the ssse3 code

Suggested-by: jamrial
Tested-by: wm4
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2015-02-06 00:20:59 +01:00
James Almer
15574c505b x86/hevcdsp: add ff_hevc_sao_edge_filter_{10,12}_{sse2,avx2} 
Original x86 intrinsics code by Pierre-Edouard Lepere.
Yasm port, refactoring and optimizations by James Almer.

Benchmarks of BQTerrace_1920x1080_60_qp22.bin with an Intel Core i5-4200U

Width 32
342694 decicycles in sao_edge_filter_10, 16384 runs, 0 skips
29476 decicycles in ff_hevc_sao_edge_filter_32_10_ssse3, 16384 runs, 0 skips
13996 decicycles in ff_hevc_sao_edge_filter_32_10_avx2, 16381 runs, 3 skips

Width 64
581163 decicycles in sao_edge_filter_10, 8192 runs, 0 skips
59774 decicycles in ff_hevc_sao_edge_filter_64_10_ssse3, 8192 runs, 0 skips
28383 decicycles in ff_hevc_sao_edge_filter_64_10_avx2, 8191 runs, 1 skips

Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-05 15:02:33 -03:00
James Almer
042c1159fc x86/hevcdsp: add ff_hevc_sao_edge_filter_8_{ssse3,avx2} 
Original x86 intrinsics code and initial yasm port by Pierre-Edouard Lepere.
Refactoring and optimizations by James Almer.

Benchmarks of BQTerrace_1920x1080_60_qp22.bin with an Intel Core i5-4200U

Width 32
158583 decicycles in edge, sao_edge_filter_8 runs, 0 skips
5205 decicycles in ff_hevc_sao_edge_filter_32_8_ssse3, 32767 runs, 1 skips
2942 decicycles in ff_hevc_sao_edge_filter_32_8_avx2, 32767 runs, 1 skips

Width 64
705639 decicycles in sao_edge_filter_8, 262144 runs, 0 skips
19224 decicycles in ff_hevc_sao_edge_filter_64_8_ssse3, 262111 runs, 33 skips
10433 decicycles in ff_hevc_sao_edge_filter_64_8_avx2, 262115 runs, 29 skips

Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-05 15:02:27 -03:00
James Almer
aa945dc112 x86/hevcdsp: add missing vzeroupper in ff_hevc_sao_band_filter_48_*_avx2 
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-02 00:01:35 -03:00
James Almer
71e2cb4706 x86/hevcdsp: add missing guards to ff_hevc_sao_band_filter_avx2 
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-01 21:45:52 -03:00
Christophe Gisquet
bff7feb328 x86: hevc/sao: aligned source buffers 
Usefull for at least band filter, for which:
- Band filter call only:
           32      64
Before:  16556    54015
After:   16497    52355
- Whole case:
           32      64
Before:  37031   103008
After:   32045    93952
 2015-02-01 20:22:54 -03:00
James Almer
fa3eccb4f9 x86/hevc: add ff_hevc_sao_band_filter_{8,10,12}_{sse2,avx,avx2} 
Original x86 intrinsics code and initial 8bit yasm port by Pierre-Edouard Lepere.
10/12bit yasm ports, refactoring and optimizations by James Almer

Benchmarks of BQTerrace_1920x1080_60_qp22.bin with an Intel Core i5-4200U

width 32
40338 decicycles in sao_band_filter_0_8, 2048 runs, 0 skips
8056 decicycles in ff_hevc_sao_band_filter_8_32_sse2, 2048 runs, 0 skips
7458 decicycles in ff_hevc_sao_band_filter_8_32_avx, 2048 runs, 0 skips
4504 decicycles in ff_hevc_sao_band_filter_8_32_avx2, 2048 runs, 0 skips

width 64
136046 decicycles in sao_band_filter_0_8, 16384 runs, 0 skips
28576 decicycles in ff_hevc_sao_band_filter_8_32_sse2, 16384 runs, 0 skips
26707 decicycles in ff_hevc_sao_band_filter_8_32_avx, 16384 runs, 0 skips
14387 decicycles in ff_hevc_sao_band_filter_8_32_avx2, 16384 runs, 0 skips

Reviewed-by: Christophe Gisquet <christophe.gisquet@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-02-01 20:22:35 -03:00
Christophe Gisquet
7aeafacfd0 x86/sbrdsp: Use different mem moves 
Before
2843 decicycles in ff_sbr_autocorrelate_sse3, 262086 runs, 58 skips

After
2693 decicycles in ff_sbr_autocorrelate_sse3, 262117 runs, 27 skips

Signed-off-by: James Almer <jamrial@gmail.com>
 2015-01-25 18:20:43 -03:00
James Almer
449b21bfab x86/sbrdsp: add ff_sbr_autocorrelate_{sse,sse3} 
2 to 2.5 times faster.

Signed-off-by: James Almer <jamrial@gmail.com>
 2015-01-25 18:20:39 -03:00
James Almer
08810a8895 x86/flacdsp: remove unneeded ifdeffery 
x86inc can translate r*m into a register or stack on its own

Reviewed-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: James Almer <jamrial@gmail.com>
 2015-01-05 16:29:28 -03:00
James Almer
37b35feb64 x86/swr: add SSE2/AVX pack_8ch functions 
Reviewed-by: Michael Niedermayer <michaelni@gmx.at>
Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
 2014-12-30 23:05:27 -03:00
Ronald S. Bultje
3aefca68ca vp9/x86: add myself to copyright holders for loopfilter assembly. 2014-12-27 16:55:16 -05:00
Ronald S. Bultje
afd8c464b7 vp9/x86: make filter_16_h work on 32-bit. 2014-12-27 16:55:16 -05:00
Ronald S. Bultje
b26bc3520f vp9/x86: make filter_48/84/88_h work on 32-bit. 2014-12-27 16:55:15 -05:00
Ronald S. Bultje
8a1cff1c35 vp9/x86: make filter_44_h work on 32-bit. 2014-12-27 16:55:15 -05:00
Ronald S. Bultje
047088b8c6 vp9/x86: make filter_16_v work on 32-bit. 2014-12-27 16:55:14 -05:00
Ronald S. Bultje
0cc9c23ea1 vp9/x86: make filter_48/84_v work on 32-bit. 2014-12-27 16:55:14 -05:00
Ronald S. Bultje
6433a9133f vp9/x86: make filter_88_v work on 32-bit. 2014-12-27 16:55:14 -05:00
Ronald S. Bultje
75f8e52089 vp9/x86: make filter_44_v work on 32-bit. 2014-12-27 16:55:13 -05:00
Ronald S. Bultje
7f80c3344c vp8/x86: save one register in SIGN_ADD/SUB. 2014-12-27 16:55:13 -05:00
Ronald S. Bultje
8ea2194ebb vp9/x86: store unpacked intermediates for filter6/14 on stack. 
filter16 goes from 508 to 482 (h) or 346 to 314 (v) cycles; filter88
goes from 240 to 238 (h) or 174 to 165 (v) cycles, measured on TOS.
 2014-12-27 16:55:13 -05:00
Ronald S. Bultje
e42409479f vp8/x86: move variable assigned inside macro branch. 
The value is not used outside the branch.
 2014-12-27 16:55:12 -05:00
Ronald S. Bultje
418c202c63 vp9/x86: simplify ABSSUM_CMP by inverting the comparison meaning. 2014-12-27 16:55:12 -05:00
Ronald S. Bultje
d1c55654e1 vp8/x86: remove unused register from ABSSUB_CMP macro. 2014-12-27 16:55:12 -05:00
Ronald S. Bultje
e59bd08986 vp9/x86: slightly simplify 44/48/84/88 h stores. 2014-12-27 16:55:11 -05:00
Ronald S. Bultje
8132629bd5 vp9/x86: make cglobal statement more conservative in register allocation. 2014-12-27 16:55:11 -05:00
Ronald S. Bultje
c013ca58c5 vp9/x86: save one register in loopfilter surface coverage. 2014-12-27 16:55:11 -05:00
James Almer
32c836cb11 x86/vp9: remove duplicate function prototypes 
Fixes "redundant redeclaration" warnings.

Signed-off-by: James Almer <jamrial@gmail.com>
 2014-12-23 00:56:51 -03:00
James Almer
7696e429c7 x86/vp3dsp: port put_vp_no_rnd_pixels8_l2_mmx to yasm 
Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2014-12-20 13:25:43 +01:00
James Almer
a4d62f7775 x86/constants: fix alignment of pw_255 
Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2014-12-19 20:21:34 +01:00
Ronald S. Bultje
bdc1e3e3b2 vp9/x86: intra prediction sse2/32bit support. 
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2014-12-19 14:07:19 +01:00
Ronald S. Bultje
b6e1711223 vp9/x86: invert hu_ipred left array ordering. 
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
 2014-12-19 14:07:18 +01:00