Current Patches, Where to get them, How they affect speed/output

[Dark Shikari]

@DS, i realy confused how to turn on psytrellis? what switch?Its just a modification to trellis that activates when psy-rd is on. To activate it, just have psy-rd and trellis both on.

[shon3i]

trellis 1 or 2?

[Yoshiyuki Blade]

trellis 1 or 2?

I'm pretty sure either mode will work.

[gav1577]

You know, you can just use --b-adapt 1 for the regular B-frame decision ;). In fact, I think its default.

Thanks i forgot about the --b-adapt 1 :)

[gav1577]

One question in have though is with the new patch is there any point in lowering the
chroma QP offset as i usually do now that it scales automatically ?

[martino]

trellis 1 or 2?
psy-rd is off with 1.

[Sharktooth]

D_S said the modified version of trellis (psy-trellis) can be both 1 or 2 when psy-rdo is enabled.
while the old/normal trellis version will be:
0 - disabled
1 - enabled on the final encode of macroblock if subme < 6, automatically disabled if subme >=6, always enabled on the final encode of macroblock if psy-rd is disabled (--psy-rd 0).
2 - always enabled on all all RD calls, in the final encode of macroblock and in any case.
psy-rd is enabled by defualt and always on unless you turn it off with --psy-rd 0

[Dark Shikari]

D_S said the modified version of trellis (psy-trellis) can be both 1 or 2 when psy-rdo is enabled.
while the old/normal trellis version will be:
0 - disabled
1 - enabled on I frames if subme < 6, automatically disabled if subme >=6. always enabled on I frames if psy-rd is disabled.
2 - always enabled on all kind of frames and in any case.
psy-rd is enabled by defualt and always on unless you turn it off with --psy-rd 0I'm noticing this a lot, but please don't respond unless you actually know the correct answer. It has nothing to do with frametype. Trellis 1 acts on the final encode of a macroblock--always. Trellis 2 acts on every RD call, in addition to the final encode.psy-rd is off with 1.No its not...One question in have though is with the new patch is there any point in lowering the
chroma QP offset as i usually do now that it scales automatically ?Probably not.

[Sharktooth]

argh... sorry, ill correct it (done), i confused it with another thing from another encoder...
today i have a bit of confusion in my head... i hope it is only a temporary effect of the meds.

[skystrife]

x264.930.modified.exe (http://www.mediafire.com/?ijadanmuaal) - Alternate Download (http://skystrife.com/x264/x264.930.modified.exe)
libx264-61.930.modified.dll (http://www.mediafire.com/?idjiacazwt4) - Alternate Download (http://skystrife.com/x264/libx264-61.930.modified.dll)
(Source) (http://www.mediafire.com/?87fkhdhj8ix) - Alternate Download (http://skystrife.com/x264/x264.930.modified.source.7z)

Patches used:

x264_psy_rdo.0.5+psy_trellis_01_r929.diff <-- This patch is experimental. See the PsyRDO thread for more details.
x264_new_bframe_decision_03.diff <-- This patch is highly experimental.
x264_hrd_pulldown.09_interlace.diff
x264.progress.indication.01.diff

All patches are included with the source.

gcc 3.4.5 fprofiled build.

[martino]

I see that nobody has posted yet a build without psy-trellis, so here (http://www.mediafire.com/?d4rjwwvqvrk) is one; r930 modified, for those that like them less experimental...

Patches:
x264.progress.indication.01.diff
x264.psyrdo.0.5.r929.diff (modified by kemuri)

gcc 3.4.5 fprofiled.

[DeeGee]

Seems like all the builds after (skystrife's) 920 keep crashing on pass 2. I guess i need to do some testing to isolate what causes it.

[edit] After doing all kinds of testing I couldn't get it to crash anymore... I guess I did something Stupid™ the first time around.

[gigah72]

Seems like all the builds after (skystrife's) 920 keep crashing on pass 2. I guess i need to do some testing to isolate what causes it.

is this happening on all your sources or just some, what kind of source do you have?
what x264 settings do you use?

[kemuri-_9]

I've been trying sky's builds for benchmarking against my own and they work for me on the settings i use for both passes.

didn't most of the people who were having crashes say they were using DGIndex/DGDecode?

[Blue_MiSfit]

I'm running Skystrife's patched 930 build tonight on a CRF20 SD encode, with most settings maxed out... We'll see if it turns out okay! 12fps and it's saturating my Q6600 (no filtering).

~MiSfit

[gav1577]

I see that nobody has posted yet a build without psy-trellis, so here (http://www.mediafire.com/?d4rjwwvqvrk) is one; r930 modified, for those that like them less experimental...

Patches:
x264.progress.indication.01.diff
x264.psyrdo.0.5.r929.diff (modified by kemuri)

gcc 3.4.5 fprofiled.

Thanks for the build ;)

[Ranguvar]

*wishes for an r934 build with DS's VAQ2, Psy RDO, Psy-Trellis, progress indication, nal-hrd, and the new b-frame decision patch* =]

Zanku!!

[skystrife]

x264.934.modified.exe (http://www.mediafire.com/?evnbmckbedf) - Alternate Download (http://skystrife.com/x264/x264.934.modified.exe)
libx264-61.934.modified.dll (http://www.mediafire.com/?iaabgzqapvk) - Alternate Download (http://skystrife.com/x264/libx264-61.934.modified.dll)
(Source) (http://www.mediafire.com/?dhu40zlwtiy) - Alternate Download (http://skystrife.com/x264/x264.934.modified.source.7z)

Patches used:

x264_psy_rdo.0.5+psy_trellis_01_r929.diff <-- This patch is experimental. See the PsyRDO thread for more details.
x264_new_bframe_decision_03.diff <-- This patch is highly experimental.
x264_hrd_pulldown.09_interlace.diff
x264.progress.indication.01.diff

All patches are included with the source.

gcc 3.4.5 fprofiled build.

[ajp_anton]

Using the build from the post above, I'm getting some strange artifacts when using multithreading. Is this kinda "known" or should I post pics?

[Dark Shikari]

Using the build from the post above, I'm getting some strange artifacts when using multithreading. Is this kinda "known" or should I post pics?You should upload a stream.

[ajp_anton]

I see differences between identical encodes, and not just small ones, and only in small parts of the picture. They are done on different computers (still a bug but could narrow it down).
Splitting them up into >1MB parts with mkvmerge creates 6, 12 and 16 parts for the 3 "identical" encodes.
Will do some more testing before I upload anything. I've noticed something strange with the official build too.

The streams should only be completely identical if multithreading is off, right?

[Dark Shikari]

The streams should only be completely identical if multithreading is off, right?Yes, two encodes with the same build and same settings and same source and same statsfile with no threads should be exactly the same.

[ajp_anton]

Did some test encodes over the night:

- Two of my computers produce identical streams while the 3rd is a bit different. Source is lagarith+avisource. Saved the avisynth stream into avi with VDub (direct stream copy) on all computers and they were all identical. Any ideas?
E4300 and X2 4400+ are identical, 3rd one is E4300 (just to show that it's not about the CPU).

- You'd think the artifacts (just some insane blocking (like 16x16 pixels of a single color)) comes from the somewhat low bitrate, but they just aren't there in singlethreaded encodes. But I was able to recreate them with... *drumroll*... --pre-scenecut. Making the first (and only the first) pass singlethreaded seems to help.
edit: The test clip I used has lots of <1s scenes, --min-keyint = 1s
edit: Oh, and this only occurs with --pre-scenecut and only with the patched build.

[Dark Shikari]

Did some test encodes over the night:

- Two of my computers produce identical streams while the 3rd is a bit different. Source is lagarith+avisource. Saved the avisynth stream into avi with VDub (direct stream copy) on all computers and they were all identical. Any ideas?
E4300 and X2 4400+ are identical, 3rd one is E4300 (just to show that it's not about the CPU).

- You'd think the artifacts (just some insane blocking (like 16x16 pixels of a single color)) comes from the somewhat low bitrate, but they just aren't there in singlethreaded encodes. But I was able to recreate them with... *drumroll*... --pre-scenecut. Making the first (and only the first) pass singlethreaded seems to help.
edit: The test clip I used has lots of <1s scenes, --min-keyint = 1s
edit: Oh, and this only occurs with --pre-scenecut and only with the patched build.Post the streams; are you using VBV? If so, its possible that the artifacting is due to VBV; x264 tries rather hard in 1pass mode to obey VBV strictly, which can result in significant quality drops. Don't use 1pass VBV unless you cannot use 2pass for some reason.

[ajp_anton]

Meh, VBV was it. Feels a bit stupid after all these tests and never even thinking of VBV. Well it's still a little strange when bitrate is 6000 and both VBV at 15000 (2-pass). Wonder why they even were 15000...
Anyway... I'm still a bit surprised how much --pre-scenecut changes things, and I still have no idea why this one computer's lagarith-avisynth output is different when using x264 but identical to the others' with VDub. Oh well, it's not that I can see the differences anyway...

[skystrife]

x264.935.modified.exe (http://www.mediafire.com/?rdmzpzu2jh0) - Alternate Download (http://skystrife.com/x264/x264.935.modified.exe)
libx264-61.935.modified.dll (http://www.mediafire.com/?db7hy1d2n7v) - Alternate Download (http://skystrife.com/x264/libx264-61.935.modified.dll)
(Source) (http://www.mediafire.com/?ojwl3gxj0e5) - Alternate Download (http://skystrife.com/x264/x264.935.modified.source.7z)

Patches used:

x264_psy_rdo.0.5+psytrellis_02.diff <-- This patch is experimental. See the PsyRDO thread for more details. Now allows controlling the strength of psyrdo and psytrellis individually.
x264_new_bframe_decision_04.diff (http://skystrife.com/x264/x264_new_bframe_decision_04.diff) <-- This patch is highly experimental. Only change from 03 is to remove the change of X264_BUILD in x264.h because the psytrellis patch already changes it.
x264_hrd_pulldown.09_interlace.diff
x264.progress.indication.01.diff

All patches are included with the source.

gcc 3.4.5 fprofiled build.

[Sharktooth]

request: r935 -> CLI build for megui
patches: psy-rdo
hrd-pulldown
progress indicarion - optional.
usual fprofiled build with gcc 3.4.x

[techouse]

Here you go Shark:

x264_x86_r935_techouse (http://techouse.project357.com/builds/x264_x86_r935_techouse.7z)

Source: x264 r935 GIT (git://git.videolan.org/x264.git)

Applied patches (current versions):

x264_progress.diff

x264_psy_rdo.0.5+psy_trellis_01_r929.diff

x264_hrd_pulldown.09_interlace.diff


Please check http://forum.doom9.org/showthread.php?t=130364 and http://git.videolan.org/gitweb.cgi?p=x264.git;a=shortlog for more info

Compiled by techouse on August 15th 2008, 16:00 CEST with GCC-4.3.1 on Windows Vista Ultimate SP-1 32-bit.

Commandline used: ./configure --extra-cflags="-march=core2 -pipe" && make fprofiled

Platform: X86
System: MINGW
asm: yes
avis input: yes
mp4 output: yes
pthread: yes
gtk: no
debug: no
gprof: no
PIC: no
shared: no
visualize: no

[Sharktooth]

-march=core2 ... no gprof.... :( sorry cant use it for megui autoupdate.
thanks anyway.

[gav1577]

Hi i have a request please for the latest build without psy trellis
just psy rdo Thanks :)

[Dark Shikari]

Hi i have a request please for the latest build without psy trellis
just psy rdo Thanks :)Latest psy-RD build lets you disable psy trellis... :p

[gav1577]

Latest psy-RD build lets you disable psy trellis... :p

Thanks for the info dark shikari :). so just to clarify to use old trellis and disable the new would this be correct --psy-rd 1:0 ? i am still still quite a noob at all this Thanks

[Dark Shikari]

Thanks for the info dark shikari :). so just to clarify to use old trellis and disable the new would this be correct --psy-rd 1:0 ? i am still still quite a noob at all this ThanksCorrect

[Sharktooth]

i think i will include the psy-trellis enabled build in megui. so i have to disable psy-trellis in presets or is almost safe to use it at default settings?

[Dark Shikari]

i think i will include the psy-trellis enabled build in megui. so i have to disable psy-trellis in presets or is almost safe to use it at default settings?Well hopefully I'm going to commit this soon so the bickering about it can stop. :p

[bob0r]

-march=core2 ... no gprof.... :( sorry cant use it for megui autoupdate.
thanks anyway.

gprof: no
All my builds have that too (its not fprofiled if you think that), do you?

[martino]

Hi i have a request please for the latest build without psy trellis
just psy rdo Thanks :)
If you still want one despite what DS said...

x264 r935 modified (http://users.shinsen-subs.org/martino/x264/x264.935.7z)

x264.progress.01
x264.psyrdo.05.r929.diff

gcc 3.4.5 fprofiled.

[skystrife]

x264.936.modified.exe (http://www.mediafire.com/?uc5xlvdtblh) - Alternate Download (http://skystrife.com/x264/x264.936.modified.exe)
libx264-61.936.modified.dll (http://www.mediafire.com/?a0qa6wbaqm0) - Alternate Download (http://skystrife.com/x264/libx264-61.936.modified.dll)
(Source) (http://www.mediafire.com/?mbplvdr0day) - Alternate Download (http://skystrife.com/x264/x264.936.modified.source.7z)

Patches used:

x264_psy_rdo.0.5+psytrellis_02.diff <-- This patch is experimental. See the PsyRDO thread for more details. Now allows controlling the strength of psyrdo and psytrellis individually.
x264_new_bframe_decision_04.diff (http://skystrife.com/x264/x264_new_bframe_decision_04.diff) <-- This patch is highly experimental. Only change from 03 is to remove the change of X264_BUILD in x264.h because the psytrellis patch already changes it.
x264_hrd_pulldown.09_interlace.diff
x264.progress.indication.01.diff

All patches are included with the source.

gcc 3.4.5 fprofiled build.

[Sharktooth]

gprof: no
All my builds have that too (its not fprofiled if you think that), do you?
profiling = very, very slight speed advantage but for free.

[bob0r]

profiling = very, very slight speed advantage but for free.

Unrelated?

[Sharktooth]

well, not properly. lets say fprofiling could be usefull for those running x264 on old PCs or using insane settings.
however, my request was for a fprofiled generuic build.
march=core2 isnt generic and that build was not fprofiled.

[kemuri-_9]

I've also noticed that profiled builds are usually smaller than non:

non profiled builds run about 1.18-1.2 MB,
profiled builds run about 1 MB
this has been my experience with them at least.

btw, gprofiling and fprofling are different profiling methods too, so be sure not to confuse them:
GNU gprof vs GCC flags: -fprofile-generate/-fprofile-use
x264 configure can accept gprof and compile with gprofile enabled code, but default is to use fprofiling through 'make fprofiled'

[Sharktooth]

"g" and "f" keys are really close on the keyoard... and my left hand and fingers are actually not so accurate... i usually edit my posts several times to correct the typos but i just cant catch some of them
i meant fprofiled btw.

[skystrife]

x264.937.modified.exe (http://www.mediafire.com/?rhacap8ja1e) - Alternate Download (http://skystrife.com/x264/x264.937.modified.exe)
libx264-61.937.modified.dll (http://www.mediafire.com/?yazcfgy3iv0) - Alternate Download (http://skystrife.com/x264/libx264-61.937.modified.dll)
(Source) (http://www.mediafire.com/?nenhokfdomb) - Alternate Download (http://skystrife.com/x264/x264.937.modified.source.7z)

Patches used:

x264_psy_rdo_0.6.diff <-- Just a name change, no different from the patch used in 936 modified (this means it includes psytrellis).
x264_new_bframe_decision_04.diff <-- This patch is highly experimental, only enabled with --b-adapt 2.
x264_hrd_pulldown.09_interlace.diff
x264.progress.indication.01.diff

All patches are included with the source.

gcc 3.4.5 fprofiled build.

[qyqgpower]

@skystrife
Could you include revision number in x264.exe , or I can't identify which build is used to encode the specified file. Maybe I should put revision number in the names of the encoded files? :)
your build:
x264 --version
x264 0.61.x

Jarod's build:
x264 --version
x264 0.60.928M 4816918

[bob0r]

x264.937.modified.01.exe (http://files.x264.nl/x264.937.modified.01.exe)
libx264-61.937.modified.01.dll (http://files.x264.nl/libx264-61.937.modified.01.dll)

x264-psyrd-0.6.diff (on by default, adjust with: --psy-rd)
x264.new.bframes.decision.04.diff (highly experimental, enabled with: --b-adapt 2)
x264_hrd_pulldown.09_interlace.diff
x264.progress.indication.01.diff

[Sharktooth]

bobor, is it fprofiled?

[bob0r]

bobor, is it fprofiled?

You even have to ask?
When i put an url on http://x264.nl yes it's fprofiled.

[kemuri-_9]

@skystrife
Could you include revision number in x264.exe , or I can't identify which build is used to encode the specified file. Maybe I should put revision number in the names of the encoded files? :)
your build:
x264 --version
x264 0.61.x

Jarod's build:
x264 --version
x264 0.60.928M 4816918

if the build doesn't include the revision number which it's automatically configured to do on compilation, then it doesn't have access to some CLI utils:
GIT CLI : git-rev-list, git-status
binutils: awk (usually via gawk or mawk), grep, head, join, sed, sort, wc
be sure these binaries are on the PATH environment variable or it won't be able to include the revision number within the build.

[skystrife]

Thanks for the heads up qyqgpower, and kemuri for the explanation, this further builds from me will have this issue fixed. =)

And I feel really dumb for not catching that earlier.