Op 02-02-16 om 16:54 schreef lvqcl:> BTW, 64-bit flac can benefit from 64-bit words in > bitreader/bitwriter routines. Currently it requires > --enable-64-bit-words switch in ./configure parameters. It > would be interesting to test its effect, especially on ARM.Okay, I might test that as well. I've checked performance with the brand new laptop my employer gave me, which is why it runs Windows. There is no change between current git and 1.3.1, so I did a check with AVX disabled to check whether AVX is still faster (I suppose some SSE routines have been sped up?) and it indeed is. Results are here: http://www.audiograaf.nl/misc_stuff/Comparison on 64-bit Windows (2014 Intel, AVX2).pdf
Op 02-02-16 om 18:22 schreef Martijn van Beurden:> Op 02-02-16 om 16:54 schreef lvqcl: >> BTW, 64-bit flac can benefit from 64-bit words in >> bitreader/bitwriter routines. Currently it requires >> --enable-64-bit-words switch in ./configure parameters. It >> would be interesting to test its effect, especially on ARM. > > Okay, I might test that as well.So, here are some performance checks for the 64-bit word change. http://www.audiograaf.nl/misc_stuff/Comparison 64 bit words 32-bit ARM (Raspberry Pi).pdf http://www.audiograaf.nl/misc_stuff/Comparison 64-bit word on 64-bit Linux (2008 Intel).pdf http://www.audiograaf.nl/misc_stuff/Comparison 64-bit word on 64-bit Windows (2014 Intel).pdf As you can see, both 64-bit systems seem to benefit, but the Raspberry Pi suffers slightly. Sadly I do not have any 64-bit ARM systems to run these tests on. I noticed that the improvements seem to increase the faster the coding proces is, something it has in common with the calculation of the MD5 sum. So, I decided to check whether the MD5 sum is the 'source' of this improvement. See the next file http://www.audiograaf.nl/misc_stuff/Comparison 64-bit word on 64-bit Windows (2014 Intel) - No MD5 sum.pdf It turns out it is not in the MD5 sum calculation, otherwise the difference should have vanished (or shrunk) on disabling the MD5 sum calculation.
Hi Martijn, I don?t know if it?s just my Firefox or whether something breaks during the parsing of the email but none of your links seem to work for me. I feel it might have something to do with the spaces in the file names of the PDFs, which doesn?t translate well to URLs. Thanks, Lennard> On 3 Feb 2016, at 18:04, Martijn van Beurden <mvanb1 at gmail.com> wrote: > > Op 02-02-16 om 18:22 schreef Martijn van Beurden: >> Op 02-02-16 om 16:54 schreef lvqcl: >>> BTW, 64-bit flac can benefit from 64-bit words in >>> bitreader/bitwriter routines. Currently it requires >>> --enable-64-bit-words switch in ./configure parameters. It >>> would be interesting to test its effect, especially on ARM. >> >> Okay, I might test that as well. > > So, here are some performance checks for the 64-bit word change. > > http://www.audiograaf.nl/misc_stuff/Comparison 64 bit words > 32-bit ARM (Raspberry Pi).pdf > http://www.audiograaf.nl/misc_stuff/Comparison 64-bit word on > 64-bit Linux (2008 Intel).pdf > http://www.audiograaf.nl/misc_stuff/Comparison 64-bit word on > 64-bit Windows (2014 Intel).pdf > > As you can see, both 64-bit systems seem to benefit, but the > Raspberry Pi suffers slightly. Sadly I do not have any 64-bit > ARM systems to run these tests on. > > I noticed that the improvements seem to increase the faster the > coding proces is, something it has in common with the > calculation of the MD5 sum. So, I decided to check whether the > MD5 sum is the 'source' of this improvement. See the next file > > http://www.audiograaf.nl/misc_stuff/Comparison 64-bit word on > 64-bit Windows (2014 Intel) - No MD5 sum.pdf > > It turns out it is not in the MD5 sum calculation, otherwise the > difference should have vanished (or shrunk) on disabling the MD5 > sum calculation. > _______________________________________________ > flac-dev mailing list > flac-dev at xiph.org > http://lists.xiph.org/mailman/listinfo/flac-dev