The second stride is always the internal buffer one, MAX_PB_SIZE (times 2 to
get the value in bytes).
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
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>
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>
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>
In some cases, 2 or 3 calls are performed to functions for unusual
widths. Instead, perform 2 calls for different widths to split the
workload.
The 8+16 and 4+8 widths for respectively 8 and more than 8 bits can't
be processed that way without modifications: some calls use unaligned
buffers, and having branches to handle this was resulting in no
micro-benchmark benefit.
For block_w == 12 (around 1% of the pixels of the sequence):
Before:
12758 decicycles in epel_uni, 4093 runs, 3 skips
19389 decicycles in qpel_uni, 8187 runs, 5 skips
22699 decicycles in epel_bi, 32743 runs, 25 skips
34736 decicycles in qpel_bi, 32733 runs, 35 skips
After:
11929 decicycles in epel_uni, 4096 runs, 0 skips
18131 decicycles in qpel_uni, 8184 runs, 8 skips
20065 decicycles in epel_bi, 32750 runs, 18 skips
31458 decicycles in qpel_bi, 32753 runs, 15 skips
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
* Reduced xmm register count to 7 (As such they are now enabled for x86_32).
* Removed four movdqa (affects the sse2 version only).
* pxor is now used to clear m0 only once.
~5% faster.
Reviewed-by: Christophe Gisquet <christophe.gisquet@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
Only 8-bit and 10-bit idct_dc() functions are included (adding others should be trivial).
Benchmarks on an Intel Core i5-4200U:
idct8x8_dc
SSE2 MMXEXT C
cycles 22 26 57
idct16x16_dc
AVX2 SSE2 C
cycles 27 32 249
idct32x32_dc
AVX2 SSE2 C
cycles 62 126 1375
Signed-off-by: James Almer <jamrial@gmail.com>
Reviewed-by: Mickaël Raulet <mraulet@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Additional contributions by James Almer <jamrial@gmail.com>,
Carl Eugen Hoyos <cehoyos@ag.or.at>, Fiona Glaser <fiona@x264.com> and
Anton Khirnov <anton@khirnov.net>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
