Reclock's Resampler version 0.1.7 ICL11

[LINUS]

now would be the perfect time for a new ReClock release

Yes, expiration date is getting closer.

Thanks in advance (though I do not have any problems with the current version!)

 

[yesgrey3]

Can you link to the final, agreed, optimised, compiled version of the Resampler?

The link is on the first post of this thread. But remember is not yet the final. After James release his build, let's compare it with the one linked here and if Jame's build is faster I will remove my link. Please post here the compare results.

Could you post the current sources?

James, here it is:
View attachment Resampler_0.1.5pre2_source.zip
Remember this is not the official release, is still 0.1.5pre2. I will contact the author to ask him if we could consider this the final or not.
Another thing, to save you some trouble... You will need a compiler which supports 1999 ISO Standard C. The ICL10.1 does.

 

[James]

Another thing, to save you some trouble... You will need a compiler which supports 1999 ISO Standard C. The ICL10.1 does.

Uh-oh. I'm not sure if VS2005 does.

 

[James]

Uh-oh. I'm not sure if VS2005 does.

Just tried, it doesn't.

 

[yesgrey3]

Just tried, it doesn't.

Let me know what you want to do. Just keep my build linked in this thread and keep the 0.1.4 version with your reclock installer, or remove my build and just forget about 0.1.5...

 

[James]

Let me know what you want to do. Just keep my build linked in this thread and keep the 0.1.4 version with your reclock installer, or remove my build and just forget about 0.1.5...

I changed the source a little and it compiles with VS2005 (ignoring some warnings). No big deal. I'll release ReClock 1.8.3.1 beta in a second...

 

[leeperry]

I changed the source a little and it compiles with VS2005 (ignoring some warnings). No big deal. I'll release ReClock 1.8.3.1 beta in a second...

awesome, muchas gracias

After James release his build, let's compare it with the one linked here and if Jame's build is faster I will remove my link. Please post here the compare results.

I'll post my results in a few 8)

 

[James]

After James release his build, let's compare it with the one linked here and if Jame's build is faster I will remove my link. Please post here the compare results.

I am quite sure your build is faster (using a better compiler). Nevertheless, I'm quite happy with my build.

 

[leeperry]

ok so I've picked 3 movies, basically yesgrey3's generic O3 build's faster for 6 channels, and it's very close for stereo....maybe James build's a bit faster in that particular case.

but for very high bitrate movies(non h264), yesgrey3's build seems less CPU intensive(better MT management?)...even for stereo

***5.1 32float 96KHz ***

movie A / 1.5mbit dts
yesgrey3 : max 25 / sits 22.8
James : max 26.9 / sits 24.9

movie B / 384kb ac3
yesgrey3 : max 24.4 / sits 23
James : max 26 / sits 24

movie C / 1.5mbit dts
yesgrey3 : max 24.5 / sits 23
James : max 26.6 / sits 24.3

***stereo 32float 96KHz ***

movie A / 1.5mbit dts
yesgrey3 : max 16.1 / sits 14
James : max 15.3 / sits 14

movie B / 384kb ac3
yesgrey3 : max 13.7 / sits 13
James : max 14.1 / sits 12.6

movie C / 1.5mbit dts
yesgrey3 : max 15.3 / sits 13.5
James : max 15.1 / sits 13.7

 

[yesgrey3]

I am quite sure your build is faster (using a better compiler). Nevertheless, I'm quite happy with my build.

I have said this because with 0.1.4 I found that your build was a little faster than mine, but with 0.1.5 it appears that's not the case anymore. No problem, I'm not in a race, I am just looking for the better option.

In the next days I will build a new version using ICL11... Let's see if I can defeat myself...
leeperry, stay tuned, once again you'll be the judge...

 

[James]

ok so I've picked 3 movies, basically yesgrey3's generic O3 build's faster for 6 channels, and it's very close for stereo....maybe James build's a bit faster in that particular case.

but for very high bitrate movies(non h264), yesgrey3's build less CPU intensive(better MT management?)...even for stereo

With MT you mean multi threading/tasking? No way. Better optimized code, that's all.

 

[James]

I have said this because with 0.1.4 I found that your build was a little faster than mine,

Really? That's a little unexpected.

but with 0.1.5 it appears that's not the case anymore. No problem, I'm not in a race, I am just looking for the better option.

In the next days I will build a new version using ICL11... Let's see if I can defeat myself...
leeperry, stay tuned, once again you'll be the judge...

Are you using any optimization options, like MMX/SSE/SSE2/...? I don't, because I am afraid of broken filters written in assembler, which might stumble if some CPU registers have changed.
Ogo saves/restores all FPU registers, which is very ... scary. It is even mentioned in his changelog.
That aside, I tried the global optimizer and SSE2 in VS2005, and resampler was getting slower. Really weird.

 

[yesgrey3]

Really? That's a little unexpected.

Not so. My build was with O2 optimizations, and have found with this one that O3 is better. Maybe the previous also was better with O3, but now it's not worth trying it.

Are you using any optimization options, like MMX/SSE/SSE2/...?
That aside, I tried the global optimizer and SSE2 in VS2005, and resampler was getting slower. Really weird.

Haven't you read this thread from start (the SSSE3 version included all SIMDs up to SSSE3)? I have build several versions and the faster was the generic build... I don't think it's weird, maybe it's the code that doesn't need the SIMDs, or that the assembly generated is of poor quality to this kind of code...

 

[James]

Haven't you read this thread from start (the SSSE3 version included all SIMDs up to SSSE3)?

Obviously not... or I've forgotten... shame on me...

 

[BurnerHEAD]

Quote: Originally Posted by yesgrey3
Another thing, to save you some trouble... You will need a compiler which supports 1999 ISO Standard C. The ICL10.1 does.

Quote: Originally Posted by James
Uh-oh. I'm not sure if VS2005 does.

Just, a note for you James, you ever thought about using VS2008?
IMHO it's a lot better.

Best Regards,
BurnerHEAD

 

[James]

Just, a note for you James, you ever thought about using VS2008?
IMHO it's a lot better.

No, as executables compiled with VS2008 won't run on Windows 9x/ME and NT4.
Define "a lot better".

 

[yesgrey3]

Obviously not... or I've forgotten... shame on me...

No problem.
But the result was funny. It showed us that all this SIMDs obsession should be refrained. It's not a solution for everything, only for the things that it was aimed for...

 

[Jong]

Overall this build is now fantastic. No problem at all using "excellent". I would almost say it should be "recommended" for all but those with antiquated hardware.

One small thing though. Its much less important now, since this build is so much more efficient, but the original ICL10 build (0.1.4?) was clearly multi-core capable. It spread itself nicely across the cores. This build seems to all run on the first core. I liked the old way, although I appreciate it may depend on your CPUs and workload.

 

[yesgrey3]

but the original ICL10 build (0.1.4?) was clearly multi-core capable.

Can you tell me the date and size (in bytes) of the file?
It's strange, because the code is not multithreaded. The compiler has the possibility of creating some multithreaded parts, but I have disabled all, because, as we have tested, the cpu load was higher.
Remember that MT is not a miracle solution. For some things is great, for others is not. In a quad core cpu, having the resampler MT, even if he uses a little more cpu power, could be good, but in a dual core cpu, is better a faster ST resampler, because the video decoder could run on the other...

 

[James]

One small thing though. Its much less important now, since this build is so much more efficient, but the original ICL10 build (0.1.4?) was clearly multi-core capable.

No. It may look like this, depending which thread calls ReClock's "RenderSample", and on which core this thread is scheduled.