ffdshow tryouts project: Discussion & Development

[Inventive Software]

Yes, but just because it's enabled doesn't mean it's used. It says it'll use it if it's there, which it plainly isn't, and I have no idea what the inherent difference is between Intel's MMX and SSE, and AMD's 3DNow!

[Px]

Originally Posted by drevil_xxl

Intel E6300 doesn't have 3dnow! ext.
3dnow! ext can't be disabled...

Why? It's disabled because it unsupported, nothing wrong...

[Px]

Originally Posted by Inventive Software

Yes, but just because it's enabled doesn't mean it's used. It says it'll use it if it's there, which it plainly isn't, and I have no idea what the inherent difference is between Intel's MMX and SSE, and AMD's 3DNow!

That is religion question

[foxyshadis]

Originally Posted by Px

That is religion question

Until recently the AMD processors were always optimized much better for 3dnow than for SSE/2, but they gained rough parity in later revisions of the K8. There were also a few things you could do better in 3dnow anyway, but they've been incorporated into the SSE monster set over the years.

[_xxl]

It's disabled because it unsupported

Look again please.
I''ll try to fix this bug.

Yes, but just because it's enabled doesn't mean it's used

True but it will be used:
SSE3 in xvidcore 1.2.
SSSE3 in x264.
SSE4(2) maybe it will be used in x264.

[Kurtnoise]

Best patch of the year...

Index: bin/clear.bat
===================================================================
--- bin/clear.bat	(revision 1125)
+++ bin/clear.bat	(working copy)
@@ -1 +1 @@
-rm *.exe *.dll *.ilk *.ax *.vdf *.acm
+del *.exe *.dll *.ilk *.ax *.vdf *.acm

[Px]

Originally Posted by drevil_xxl

Look again please.
I''ll try to fix this bug.

I understand, you telling about lower part where user choose extensions, and I speaking about upper part, where shows actually used extensions
And I agree - checkboxes with unsupported extensions must be grayed

[Leak]

Here's yet another update of my AviSynth patch - sorry for the two-week delay caused mainly by my TFT monitor's backlight dying:

* Disable buffering when "Apply pulldown" is off and ahead and back buffers are set to 0 (i.e. the old AviSynth filter behaviour)
* Use original timestamps when the number of frames isn't changed

Patch (against rev. 1125)
Download (pure MSVC 8/GCC 3.4.4 build with ICL9 kerneldeint/tomsmocomp DLLs)

I'd consider this version feature-complete; all that's left is adding a help-file documenting the AviSynth filter settings and operation and, of course, priming the help button on the AviSynth page to bring it up - but I thought another release while I'm tackling that can't hurt...

np: Hug - Tactic Without Practice (Heroes)

[Piyoko]

I just wanted to thank you for your work on this, Leak, as I've been looking for a good on-the-fly IVTC solution to watching anime DVDs for a very long time now.

Your newest version however gives me a "System exception - Access Violation" error message whenever I add TDecimate() to the AviSynth script.

I'm also wondering, is there any way to make the subtitles look better when decoding DVDs with ffdshow (libmpeg2.) As it is now they look terrible as the colour information is completely out of whack, for example: ffdshow/libmpeg2 (bad), nVidia decoder (good).

[Leak]

Originally Posted by Piyoko

I just wanted to thank you for your work on this, Leak, as I've been looking for a good on-the-fly IVTC solution to watching anime DVDs for a very long time now.

*cough*

Same here...

Your newest version however gives me a "System exception - Access Violation" error message whenever I add TDecimate() to the AviSynth script.

Strange, it works fine for me - what's the exact script you're using, and what are your buffer settings? And what filter are you using for MPEG2 decoding?

EDIT: Hmmm... I just noticed a crash when I use one of AviSynth's resizing filters after TFM/TDecimate... that's strange.

I'll investigate that further tomorrow; it's a bit late now...

I'm also wondering, is there any way to make the subtitles look better when decoding DVDs with ffdshow (libmpeg2.) As it is now they look terrible as the colour information is completely out of whack, for example: ffdshow/libmpeg2 (bad), nVidia decoder (good).

Hmmm... I guess Haruhiko will have to answer that one, as I hardly know anything about the DVD subtitle part - except that there once was a bug in there that could crash ffdshow...

np: Jimi Tenor & Kabu Kabu - Hermetic Man (Joystone)

[haruhiko_yamagata]

Originally Posted by Piyoko

I'm also wondering, is there any way to make the subtitles look better when decoding DVDs with ffdshow (libmpeg2.) As it is now they look terrible as the colour information is completely out of whack, for example: ffdshow/libmpeg2 (bad), nVidia decoder (good).

OK, perhaps it's not DVD specific. ffdshow's color subtitle rendering is broken. I have a plan to fix this and make it ASS2 compatible, though it's hard...

[jidelite]

Originally Posted by Leak

Here's yet another update of my AviSynth patch - sorry for the two-week delay caused mainly by my TFT monitor's backlight dying:

* Disable buffering when "Apply pulldown" is off and ahead and back buffers are set to 0 (i.e. the old AviSynth filter behaviour)
* Use original timestamps when the number of frames isn't changed

Patch (against rev. 1125)
Download (pure MSVC 8/GCC 3.4.4 build with ICL9 kerneldeint/tomsmocomp DLLs)

I'd consider this version feature-complete; all that's left is adding a help-file documenting the AviSynth filter settings and operation and, of course, priming the help button on the AviSynth page to bring it up - but I thought another release while I'm tackling that can't hurt...

np: Hug - Tactic Without Practice (Heroes)

Leak,
Using your lastest version, I tried disabling buffering by unchecking "Apply pulldown" and seting ahead and back buffers to 0 but I get the same behavior as if buffering was on (slight stutter when the frame duration changes). I looked at the AviSynth OSD info and it still tells me I'm buffering ahead 1 (using the SeeSaw filter only). Should the AviSynth display look the same as if buffering were turned on when it is set to off?

Thanks,
jidelite

[Piyoko]

Originally Posted by haruhiko_yamagata

OK, perhaps it's not DVD specific. ffdshow's color subtitle rendering is broken. I have a plan to fix this and make it ASS2 compatible, though it's hard...

I really appreciate it, thanks.

Originally Posted by Leak

Strange, it works fine for me - what's the exact script you're using, and what are your buffer settings? And what filter are you using for MPEG2 decoding?

I used the exact same settings from 1103, which worked there. This is the script I used:

LoadPlugin("C:\TIVTC\TIVTC.dll")
TFM(order=1,mode=1,PP=0,chroma=false)
TDecimate(mode=0,hybrid=0,chroma=false)

Tried both using the internal ffdshow mpeg2 decoders, as well as an external (MPC's built-in) and got the same error message with all of them. With 1103 I used buffers 0 back/11 ahead, but again it didn't matter for 1125, got the same error no matter what.

[Leak]

Originally Posted by jidelite

Using your lastest version, I tried disabling buffering by unchecking "Apply pulldown" and seting ahead and back buffers to 0 but I get the same behavior as if buffering was on (slight stutter when the frame duration changes). I looked at the AviSynth OSD info and it still tells me I'm buffering ahead 1 (using the SeeSaw filter only). Should the AviSynth display look the same as if buffering were turned on when it is set to off?

Well, with those settings I just take a single frame, immediately run it through the AviSynth script and spit the frame out again with the same timestamps that went in, so I don't know where that stutter comes from - unless the timestamps are totally bogus (like the start and end timestamps of the current frame being before the end timestamp of the last frame), which will make it silently drop the frame or adjust the timestamps slightly if the difference is small enough.

As far as the OSD is concerned - it only shows the values that would be neccessary for proper operation, but the filter itself will use the values you set in the config page.

np: Pole - Like Rain (But Different) (Pole)

[Leak]

Originally Posted by Piyoko

Tried both using the internal ffdshow mpeg2 decoders, as well as an external (MPC's built-in) and got the same error message with all of them. With 1103 I used buffers 0 back/11 ahead, but again it didn't matter for 1125, got the same error no matter what.

Heh... turns out I was using TIVTC 1.0.0RC4; updating to 1.0.1 "fixed" things so I get the same error message as you now... *boggles*

I'll look into it.

EDIT: Okay, I found the problem - TIVTC starts requesting frames in TFM or TDecimate (can't remember which) as soon as the filter is constructed, which I didn't really expect, so a few variables weren't initialized at that time that should have been.

Of course, I really can't hand TFM/TDecimate a frame at that point (ffdshow didn't even produce one yet), but I guess simply handing it a blank frame is better than a crash...

Oh yeah, Haruhiko - that most probably was where your green frames came from...

I'm still hunting some other bug that messes up the last timestamp for static DVD menus if "Apply pulldown" is deactivated and causes the static image to not show up, but that should be it then.

np: Burnt Friedman & Jaki Liebezeit - Wirklich (Secret Rhythms)

[_xxl]

Please test my new builds.
http://rapidshare.com/files/28939785...70430.rar.html
I don't have a SSSE3 (Intel core2) cpu.I can't test if X264 is faster or not.

[wozio]

Originally Posted by Leak

I'm still hunting some other bug that messes up the last timestamp for static DVD menus if "Apply pulldown" is deactivated and causes the static image to not show up, but that should be it then.

I also have problems with static dvd menus but I haven't had time to track why it occurs. In cyberlink decoder it plays fine. If something can be done with it it would be great. I noticed the same thing in dscaler5 decoder so seems this is libmpeg2 problem.

BTW when apply pulldown is deactivated in your patched version it behavies exactly the same as normal ffdshow, right?

[Leak]

Originally Posted by wozio

I also have problems with static dvd menus but I haven't had time to track why it occurs. In cyberlink decoder it plays fine. If something can be done with it it would be great. I noticed the same thing in dscaler5 decoder so seems this is libmpeg2 problem.

Well, if the problem is that the menu image doesn't show up then yeah, that's it.

If the problem is that there's no highlights on static DVD menus, then that's something that needs to be fixed in the DVD subtitle part, probably by copying the last frame and overlaying the changed highlight on it - the problem is that during static DVD menus there's no new image being produced that a changed highlight could be overlaid on, which is what makes highlights work on animated menus.

BTW when apply pulldown is deactivated in your patched version it behavies exactly the same as normal ffdshow, right?

Not quite - you'll have to both deactivate "Apply pulldown" and set both buffer settings to 0.

[wozio]

When you mentioned it I tried and I have a lot of problems with dvd menus. First highlight doesn't show in animated menu, it shows after first button change. In static menus highlights don't show at all. And going back to menu from movie doesn't work, menu does not appear but works (I mean choosing button triggers correct action but it must be done blind).

I have ffdshow rev 1120 normal one.

[Kado]

@drevil_xxl
Used your rev1125 test build and GCC crashes as well on my Pentium D 930 and Windows XP.
SSE3 does not get used on my system when selected on the options and cannot test SSSE3 because my system does not support it.

EDIT
SSE3 is working, MPC must be reopened for changes to take effect, reopening the video is not enough!
EDIT/