Optimizing VP8 code in ffmpeg with new instructions. Results: SVP64 assembly code of DCT 4x4 used in VP8 codec: https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=media/video/libvpx/vpx_get4x4sse_cs_svp64_real.s;h=45bb5440ffc5a9aefea570ff965f1cfface3925d;hb=HEAD
Ok, Similarly to VP9, as VP8 is implemented in the same libvpx library, an SVP64 version of of the VP8 DCT 4x4 has been completed in commit https://git.libre-soc.org/?p=openpower-isa.git;a=commit;h=343853a711ca2dded15f8a713370a61454f2627f Implementation method was similar to VP9 ticket: https://bugs.libre-soc.org/show_bug.cgi?id=228 Tests can be run by make all and then running $ ./vp8_dct_test [==========] Running 4 tests from 2 test suites. [----------] Global test environment set-up. [----------] 2 tests from C/FdctTest [ RUN ] C/FdctTest.SignBiasCheck/0 [ OK ] C/FdctTest.SignBiasCheck/0 (0 ms) [ RUN ] C/FdctTest.RoundTripErrorCheck/0 [ OK ] C/FdctTest.RoundTripErrorCheck/0 (0 ms) [----------] 2 tests from C/FdctTest (0 ms total) [----------] 2 tests from SVP64/FdctTest [ RUN ] SVP64/FdctTest.SignBiasCheck/0 [ OK ] SVP64/FdctTest.RoundTripErrorCheck/0 (202893 ms) [----------] 2 tests from SVP64/FdctTest (576060 ms total) [----------] Global test environment tear-down [==========] 4 tests from 2 test suites ran. (576060 ms total) [ PASSED ] 4 tests.
https://libre-soc.org/irclog/%23libre-soc.2022-09-27.log.html#t2022-09-27T10:52:01