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

[puffpio]

is there much of a difference w/ -march=pentium2 vs core2?

and for the 64 bit version, shouldn't the minimum -march be k8 or nocona?

[LoRd_MuldeR]

is there much of a difference w/ -march=pentium2 vs core2?

Nope. Most (all?) performance-critical functions in x264 are written as hand-optimized assembler code.

x264 will detect your CPU capabilities at runtime and it will choose the optimized assembler functions accordingly.

Compiler optimizations matter for the pure C code only...

[komisar]

LoRd_MuldeR, sure, but if gcc-libs compiled with -march -- we get ~1-5 more fps...

puffpio, little latter i compile x264 for generic (pentium2), k8, core2. Minimum for x86_64 -- "-mtune=generic" (without march)

[Dark Shikari]

LoRd_MuldeR, sure, but if gcc-libs compiled with -march -- we get ~1-5 more fps...Yes, because of cmov.

[akupenguin]

1-5 more fps
1-5 out of what? There's a big difference between gaining 5fps if it was 1 before, vs gaining 5 from 1000.
No one should ever post any speed results in the form of a diff without a baseline. The only valid forms are ratios.

[vmrsss]

hi everybody.

In theme of speed differences, would there be much using gcc-4-2.1 rather than gcc-4.3.3 for x86_64? (For -march=core2 rather than nothing I assume the answer is above, that is a few fps due do cmov)

In general, how would I go about timing one build of x264 against another?

Thx.

[vmrsss]

PS. What does make fprofiled do for you?

[J_Darnley]

1 - On nix: use time or modify x264 to print the time
On Windows: modify x264 to print the time

[imk]

r1127M ICC Builds:

SSE2 x32 (http://imk.cx/pc/x264/x264.r1127M.SSE2.x32.imk.exe) (w/MP4 Output (http://imk.cx/pc/x264/x264.r1127M.SSE2.x32.mp4.imk.exe))
SSSE3 x32 (http://imk.cx/pc/x264/x264.r1127M.SSSE3.x32.imk.exe) (w/MP4 Output (http://imk.cx/pc/x264/x264.r1127M.SSSE3.x32.mp4.imk.exe))

SSE2 x64 (http://imk.cx/pc/x264/x264.r1127M.SSE2.x64.imk.exe) (w/MP4 Output (http://imk.cx/pc/x264/x264.r1127M.SSE2.x64.mp4.imk.exe))
SSSE3 x64 (http://imk.cx/pc/x264/x264.r1127M.SSSE3.x64.imk.exe) (w/MP4 Output (http://imk.cx/pc/x264/x264.r1127M.SSSE3.x64.mp4.imk.exe))


Older builds, build scripts, information, etc. can all be found here (http://imk.cx/pc/x264/).

[LoRd_MuldeR]

PS. What does make fprofiled do for you?

It will analyze the code in execution and then apply additional optimizations.

Since "make fprofiled" will encode the specified video clip several time (to cover all x264 options/codepaths) it will take much longer to build...

http://en.wikipedia.org/wiki/Profiler_(computer_science)

[vmrsss]

Thx. What criteria are useful when picking the video clips to feed make fprofiled?

[LoRd_MuldeR]

Thx. What criteria are useful when picking the video clips to feed make fprofiled?

I think you should pick something that equals the kind of video that you are going to encode. A few hundred frames should be enough.

Choosing a "synthetic" clip, like solid black or random noise, wouldn't give optimal profiling results, I guess...

[skystrife]

x264 r1128 x64 (unpatched) (http://www.mediafire.com/?wnjg5utuzin) - Alternate Download (http://skystrife.com/x264/revision1128/x264.exe)
gcc 4.3.4 fprofiled build.

-------------------------

x264.1128M.x86.exe (http://www.mediafire.com/?mytyyrkwchg) - Alternate Download (http://skystrife.com/x264/x264.1128M.x86.exe) / x264.1128M.x64.exe (http://www.mediafire.com/?mmzmmmmnmxt) - Alternate Download (http://skystrife.com/x264/x264.1128M.x64.exe)
gcc 3.4.5 fprofiled build with -march=pentium2. / gcc 4.3.4 fprofiled build.

Patches used:

x264_hrd_pulldown.10_interlace.diff
x264_win_zone_parse_fix_05.diff

[techouse]

x264_x86_r1129_techouse (http://techouse.digitalpulse.us/builds/x264_x86_r1129_techouse.7z) | INFO (http://techouse.digitalpulse.us/nfo/x264_x86_r1129_techouse.txt)
GCC 4.3.3, fprofiled, -march=core2

x264_x64_r1129_techouse (http://techouse.digitalpulse.us/builds/x264_x64_r1129_techouse.7z) | INFO (http://techouse.digitalpulse.us/nfo/x264_x64_r1129_techouse.txt)
GCC 4.3.4 20090220 (prerelease) (x64.generic.Komisar), fprofiled, -march=core2

Patches used:

x264_hrd_pulldown.10_interlace.diff
x264_win_zone_parse_fix_05.diff

[Kurtnoise]

Patches used:

x264_hrd_pulldown.10_interlace.diff
x264_win_zone_parse_fix_05.diff
can I have a link for those, please ?

[LoRd_MuldeR]

can I have a link for those, please ?

http://forum.doom9.org/showpost.php?p=1257673&postcount=1726
http://forum.doom9.org/showpost.php?p=1230476&postcount=1535

:search:

[Sharktooth]

Request: updated skystrife's builds.

[Trahald]

Here is the HRD patch release 11. I rewrote most of the calculations . Because most/all the values calculations are changed, i'd consider 11 an alpha until its been around a bit.

[squid_80]

Still looks like it's got a typo to me (TBT,BT,BTB,BT).

[Trahald]

Still looks like it's got a typo to me (TBT,BT,BTB,BT).

HAH. now fixed.

[bob0r]

Request: updated skystrife's builds.

Hmm, still not active.
Maybe we can use techouse's builds? Or someone else can make them?

Also i would need x264 64bit (gpac/pthreads also 64bit) + fprofiled on the same 64bit system.
If someone can do this, all you need to do is host them on http. Example: http://yourdomain.com/x264/revisionXXXX/
Files inside revisionXXXX should be x264.exe and x264.md5
( To create the md5 for x264.nl: md5sum x264.exe | awk '{print $1}' >x264.md5 )

Then my mirror script checks for the x264.md5 file every hour and mirror it then)

[LoRd_MuldeR]

Why you worry so much? You have a r1128 build up and the differences between r1128 and r1129 are very minor...

[techouse]

Hmm, still not active.
Maybe we can use techouse's builds? Or someone else can make them?

Also i would need x264 64bit (gpac/pthreads also 64bit) + fprofiled on the same 64bit system.
If someone can do this, all you need to do is host them on http. Example: http://yourdomain.com/x264/revisionXXXX/
Files inside revisionXXXX should be x264.exe and x264.md5
( To create the md5 for x264.nl: md5sum x264.exe | awk '{print $1}' >x264.md5 )

Then my mirror script checks for the x264.md5 file every hour and mirror it then)
Sure, you can use mine :) I build&profile them on Vista x64, but i'm not sure how you want me to host the builds on my servers.

Cheers ;)

[techouse]

x264_x86_r1130_techouse (http://techouse.project357.com/builds/x264_x86_r1130_techouse.7z) | INFO (http://techouse.project357.com/nfo/x264_x86_r1130_techouse.txt)
GCC 4.3.3, fprofiled, -march=core2

x264_x64_r1130_techouse (http://techouse.project357.com/builds/x264_x64_r1130_techouse.7z) | INFO (http://techouse.project357.com/nfo/x264_x64_r1130_techouse.txt)
GCC 4.3.4 20090220 (prerelease) (x64.generic.Komisar), fprofiled, -march=core2

Patches used:

x264_hrd_pulldown.11_interlace.diff
x264_win_zone_parse_fix_05.diff



P.S.: I put md5 checksums in the 7-Zip files this time ;)

[Trahald]

@techouse Your x64 link links to the x86 file and vice versa

[techouse]

@techouse Your x64 link links to the x86 file and vice versaThanx! I am in a hurry cause I've got a train to catch.... ttyl ;)

[bob0r]

Sure, you can use mine :) I build&profile them on Vista x64, but i'm not sure how you want me to host the builds on my servers.

Cheers ;)

Very simple: Build x264 64bit (gpac/pthreads 64bit also) from git only, no patches... fprofiled.
And then put the files on any http host, as long as you create the dir: revision1130 with inside x264.exe and x264.md5

so example:
http://techouse.project357.com/builds/revision1130/x264.exe
http://techouse.project357.com/builds/revision1130/x264.md5
Then all i have to do is edit:
fixedurl=http://techouse.project357.com/builds/revision in my script and the files get mirrored to x264.nl.

You can also sent me a private http link if bandwidth is an issue (or ill create a host for you, but i think we have no problems with bandwidth :))

[techouse]

Very simple: Build x264 64bit (gpac/pthreads 64bit also) from git only, no patches... fprofiled.
And then put the files on any http host, as long as you create the dir: revision1130 with inside x264.exe and x264.md5

so example:
http://techouse.project357.com/builds/revision1130/x264.exe
http://techouse.project357.com/builds/revision1130/x264.md5
Then all i have to do is edit:
fixedurl=http://techouse.project357.com/builds/revision in my script and the files get mirrored to x264.nl.

You can also sent me a private http link if bandwidth is an issue (or ill create a host for you, but i think we have no problems with bandwidth :))OK, I'll do it as soon as I get to the campus ;) (that's tomorrow somewhere around 23:00 CET)

P.S.: Do you only need 64bit or both?

[bob0r]

...

P.S.: Do you only need 64bit or both?


Only 64bit, the x264.nl compile system is 32bit.

[techouse]

Only 64bit, the x264.nl compile system is 32bit.Sorry it took so long, but I had some work with tax reports. Anyway, I'll be on the campus around 12:00 CEST, so expect the build somewhere around 13:00 CEST.

Here you go:

http://techouse.project357.com/builds/revision1134/x264.exe
http://techouse.project357.com/builds/revision1134/x264.md5

(unpatched generic & fprofiled 64bit build, GCC 4.3.4 20090220 (prerelease) (x64.generic.Komisar))

[techouse]

x264_x86_r1134_techouse (http://techouse.project357.com/builds/x264_x86_r1134_techouse.7z) | INFO (http://techouse.project357.com/nfo/x264_x86_r1134_techouse.txt)
GCC 4.3.3, fprofiled, -march=core2

x264_x64_r1134_techouse (http://techouse.project357.com/builds/x264_x64_r1134_techouse.7z) | INFO (http://techouse.project357.com/nfo/x264_x64_r1134_techouse.txt)
GCC 4.3.4 20090220 (prerelease) (x64.generic.Komisar), fprofiled, -march=core2

Patches used:

x264_hrd_pulldown.11_interlace.diff
x264_win_zone_parse_fix_05.diff

[Betsy25]

Didn't x264 use all cores by default then ? :confused:

[LoRd_MuldeR]

Didn't x264 use all cores by default then ? :confused:

No, it doesn't and it never did! The default setting is "--threads 1" (that is: multi-threading = OFF).
To use multiple cores, you must either set "--threads auto" or manually specify "--threads n", where n = cores * 3/2.

[ajp_anton]

To use multiple cores, you must either set "--threads auto" or manually specify "--threads n", where n = cores * 3/2.

"--threads auto" = "--threads 0" = "--threads [cores*3/2]"
However, you can specify whatever number you want (you make it sound you *need* cores*3/2 =))

[Betsy25]

"--threads auto" = "--threads 0" = "--threads [cores*3/2]"
However, you can specify whatever number you want (you make it sound you *need* cores*3/2 =))

So, you actually replace an application that could please everyone with some that puts a lot of people in the bare cold ?

[Astrophizz]

I would guess that anyone who wouldn't know about the command line settings would be using a frontend, which would most likely set threads to auto.

[LoRd_MuldeR]

I would guess that anyone who wouldn't know about the command line settings would be using a frontend, which would most likely set threads to auto.

Exactly. If people use x264 from the commandline, they should be intelligent enough to type "x264 --help" and read :sly:

For the average user, who doesn't want to learn about CLI parameters, we have dozens of GUI's available...

[Betsy25]

Exactly. If people use x264 from the commandline, they should be intelligent enough to type "x264 --help" and read :sly:

For the average user, who doesn't want to learn about CLI parameters, we have dozens of GUI's available...

Is there actually anything positive about this change ? By removing the --threads option, does the program run faster ? Anything we don't know ? Or is it just some april fools release ?

[Dark Shikari]

Is there actually anything positive about this change ? By removing the --threads option, does the program run faster ? Anything we don't know ? Or is it just some april fools release ?I have no idea what you're talking about.

[LoRd_MuldeR]

Is there actually anything positive about this change ? By removing the --threads option, does the program run faster ? Anything we don't know ? Or is it just some april fools release ?

What change are you talking about? The "--threads n" parameter works like described above ever since x264 got the ability to do multi-threading.

At least it didn't change in the last ~2 years. And of course that parameter was not removed recently...

[nm]

Didn't x264 use all cores by default then ? :confused:
Are you referring to "-march=core2" in techouse's post? That GCC switch enables compiler optimizations for Intel Core 2 processor architecture. It has nothing to do with multithreading.

[Betsy25]

Are you referring to "-march=core2" in techouse's post? That GCC switch enables compiler optimizations for Intel Core 2 processor architecture. It has nothing to do with multithreading.

Exactly, I was talking about that one. Sorry for the confusion, oh I'm just a newb.:(

[LoRd_MuldeR]

Exactly, I was talking about that one. Sorry for the confusion, oh I'm just a newb.:(

Then please don't mix up compiler switches and x264 switches:
Compiler switches are used when you are building x264, while x264 switches are used when you are running x264.
As a normal user you only need to worry about the latter ones ;)

Some more info can be found here:
http://sites.google.com/site/linuxencoding/x264-ffmpeg-mapping

[skystrife]

x264 r1134 x64 (unpatched) (http://www.mediafire.com/?ntkgqriintz) - Alternate Download (http://skystrife.com/x264/revision1134/x264.exe)
gcc 4.3.4 fprofiled build.

-------------------------

x264.1134M.x86.exe (http://www.mediafire.com/?ygwnybz3gy3) - Alternate Download (http://skystrife.com/x264/x264.1134M.x86.exe) / x264.1134M.x64.exe (http://www.mediafire.com/?wydznnmomma) - Alternate Download (http://skystrife.com/x264/x264.1134M.x64.exe)
gcc 3.4.5 fprofiled build with -march=pentium2. / gcc 4.3.4 fprofiled build.

Patches used:

x264_hrd_pulldown.11_interlace.diff
x264_win_zone_parse_fix_05.diff

[ajp_anton]

Then please don't mix up compiler switches and x264 switches:
Compiler switches are used when you are building x264, while x264 switches are used when you are running x264.
As a normal user you only need to worry about the latter ones ;)As a normal used he doesn't need to worry about either of them =)

BTW... maybe it was just a dream, but wasn't --thread-input made automatically active some time ago? And if not, is there any reason not to?

[techouse]

I think --thread-input is automatically active if you use --threads auto. Correct me if I'm wrong.


P.S.: LoL @ that Core2 confusion :D

[techouse]

Can anyone please run a stability test this build on AMD64 and Core2? Thanx!

http://techouse.project357.com/builds/revision1136/x264.exe
http://techouse.project357.com/builds/revision1136/x264.md5

(unpatched generic & fprofiled 64bit build, GCC 4.3.4 20090328 (prerelease) (x64.generic.Komisar))

[Sharktooth]

I think --thread-input is automatically active if you use --threads auto. Correct me if I'm wrong.
true.

[Egh]

No, it doesn't and it never did! The default setting is "--threads 1" (that is: multi-threading = OFF).
To use multiple cores, you must either set "--threads auto" or manually specify "--threads n", where n = cores * 3/2.

any explanation regarding why the best parameter is 1.5 threads per core? Seems totally random to me, though I discovered that myself (still seems pathetic to me when 2 threads x264 takes only about 70% of two-core cpu and that's with high priority ;)

[LoRd_MuldeR]

any explanation regarding why the best parameter is 1.5 threads per core? Seems totally random to me, though I discovered that myself (still seems pathetic to me when 2 threads x264 takes only about 70% of two-core cpu and that's with high priority ;)

Because testing showed that "threads = 3/2 * cores" works best. And that finding doesn't seem to be random at all ;)

The intuitive solution "threads = cores" would only give optimal performance if no thread ever becomes idle (needs to wait for another thread).