Groucho's Avisynth Stuff

[wonkey_monkey]

Does Avsmeter still do some kind of pre-scanning even when -o is specified? I've noticed it takes longer to start processing (pauses on "Frame (current | last): 0 | 10789199"), and updates less frequently, the more frames I give it. The pause and update interval are also dependent on the speed of the filters I'm running.

[Groucho2004]

Originally Posted by wonkey_monkey

Does Avsmeter still do some kind of pre-scanning even when -o is specified? I've noticed it takes longer to start processing (pauses on "Frame (current | last): 0 | 10789199"), and updates less frequently, the more frames I give it. The pause and update interval are also dependent on the speed of the filters I'm running.

Using the "-o" switch, the initial delay you see is entirely down to how fast Avisynth reads the first frame. With certain filters, this can take a while. I guess it has to do with how many frames a filter wants to read in advance and/or the general processing speed of that filter.

If you have a clip with a very large number of frames (such as your example above), AVSMeter changes the frame interval for measurements. The data for each measurement (fps, CPU usage, etc.) is stored in an array. Increasing the frame interval keeps the memory usage of that array reasonable.

[wonkey_monkey]

Using the "-o" switch, the initial delay you see is entirely down to how fast Avisynth reads the first frame.

VirtualDub doesn't exhibit the same delay. For example:

colorbars.loop(X).blur(1)

This loads instantly in VirtualDub with X=10, X=100, or X=1000, but AVSmeter gets progressively slower to start.

Take out the blur(1) and it's a lot faster, which indicates some frames are being fetched somewhere.

If you have a clip with a very large number of frames (such as your example above), AVSMeter changes the frame interval for measurements. The data for each measurement (fps, CPU usage, etc.) is stored in an array. Increasing the frame interval keeps the memory usage of that array low.

The update speed also changes depending on whether blur(1) is included or not, so it doesn't seem to be purely due to the number of frames.

(This is probably purely academic as 100,000,000 frame scripts are unlikely to come up very often)

[Groucho2004]

Originally Posted by wonkey_monkey

VirtualDub doesn't exhibit the same delay. For example:

colorbars.loop(X).blur(1)

This loads instantly in VirtualDub with X=10, X=100, or X=1000, but AVSmeter gets progressively slower to start.

You're right, it has to read the first interval of frames before it can display any meaningful numbers. So, in your example, with X = 10, the interval is 10, with X = 1000, the interval is 1000. You can verify this if you use the -log switch and look at the 'performance data' in the log file.

Originally Posted by wonkey_monkey

The update speed also changes depending on whether blur(1) is included or not, so it doesn't seem to be purely due to the number of frames.

I don't understand the logic here. 1000 frames with blur() is of course slower than 1000 frames without it.

[wonkey_monkey]

Ugh, never mind, I've finally got it through my thick head.

[Groucho2004]

avsr 0.2.0

• Fixed wrong AVISYNTH_INTERFACE_VERSION

[Groucho2004]

Added tritical's ColorMatrix (64 bit), see first post.

[hydra3333]

thank you

[StainlessS]

Lovely, thank you G2K4.

[Natty]

tyvm

[Groucho2004]

Added DeSpot 3.6.1.1 (x86 & x64)

[StainlessS]

Damn, nuther one for my archive.

Thanx Muchly

[Groucho2004]

Originally Posted by StainlessS

Damn, nuther one for my archive.

I'm terribly sorry.

Interesting fact - I removed all the SSE inline asm stuff and it turns out that it is even slightly faster by just using the C code. I guess the Intel compiler helps a lot (note to self - try MS compiler).

Edit - Intel compiler produces a DLL twice as fast as the one built with VC.

[StainlessS]

Intel pay for if not on Linux, yes ?
[mucho wonga]

[Groucho2004]

Originally Posted by StainlessS

Intel pay for if not on Linux, yes ?

Correct.

[VoodooFX]

How to read first test? Did it parsed clip in avs 2.6 test?

[Groucho2004]

Originally Posted by VoodooFX

How to read first test? Did it parsed clip in avs 2.6 test?

The error message seems pretty clear to me. You'll have to elaborate a bit as to what is unclear. Also, posting this in the AVSMeter thread would have made more sense.

[VoodooFX]

Originally Posted by Groucho2004

The error message seems pretty clear to me. You'll have to elaborate a bit as to what is unclear. Also, posting this in the AVSMeter thread would have made more sense.

Sorry I somehow ended in this tread.
I managed to benchmark script with avsr, can limit for very fast speed be removed in avsmeter, it can be still meaningful.

[ChaosKing]

Is the download page down?

[Groucho2004]

Originally Posted by ChaosKing

Is the download page down?

Works for me.