Log in

View Full Version : LAV Filters - DirectShow Media Splitter and Decoders


Pages : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 [172] 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508

nevcairiel
28th January 2012, 19:27
LAV Filters 0.45

LAV Splitter
- Improved playback of concatenated H.264 MOV/MP4 files
- Adjusted FLAC MediaType to avoid ReClock stealing the connection
- Improved playback of MPEG-TS with timestamp discontinuities

LAV Audio
- Fixed playback of very short PCM samples
- Improved COOK decoding without LAV Splitter

LAV Video
- New DXVA2 "copy back" decoder (see release notes)
- Updated Intel QuickSync decoder (r29)
- Multi-threaded decoding for RV30/40, UtVideo and DNxHD
- Support for the "FCC" color matrix
- Improved Dirac support
- CUVID support for MPEG-4 ASP (experimental)
- Support for Intel Indeo 4, MS Video 1, 8BPS, LOCO, ZMBV, VCR1, AASC


Download: Installer (both x86/x64) (http://files.1f0.de/lavf/LAVFilters-0.45.exe) -- Zips: 32-bit (http://files.1f0.de/lavf/LAVFilters-0.45.zip) & 64-bit (http://files.1f0.de/lavf/LAVFilters-0.45-x64.zip)

New and note-worthy? Well....

DXVA2 "copy back"
In our series of hardware decoders, the new DXVA2 decoder now features hardware decoding also for AMD/ATI.
Its a "copy back" (or "renderless") decoder - that means all frames are copied back from the GPU to normal memory.

Sadly, however, AMD/ATI cards suffer from a serious performance penalty, which prevents them from showing their full potential. The decoder is fully Blu-ray compliant, and even goes beyond that, however playing 60fps movies will most likely not work fluidly.
H.264, VC-1 and MPEG-2 are fully supported (progressive and interlaced), as long as your hardware can keep up. Very old AMD/ATI cards, like 3000 series and below, might not be able to use this decoder. 4000 and up seem to be working ok.

I do have plans to work on a "native" DXVA2 decoder, which does not need to copy the frames back from the GPU, avoiding this performance bottleneck - however that goes with the loss of the flexibility of course. No post-processor, only EVR, etc.

For everyone else, i recommend to stick to your native decoders (CUVID/QuickSync), as they generally work better and faster then DXVA2.

Known Issues:
- Interlaced H.264 on Intel can show image corruption
- Some MPEG-2 streams show image corruption (all GPUs)

There is also the potential for crashing the GPU driver. It happened mostly on VC-1 interlaced, and all cases i encountered i also fixed, but if it happens, just let me know, and i'll look into it.

MPEG-4 ASP in the CUVID decoder
There seems to be an issue in the latest NVIDIA drivers that cause the decoder to not work on most MPEG-4 ASP files!
Older drivers, like 270 series, seem to be working OK.

Its experimental, use it at your own risk, its off by default.

And the rest...
Everything else is pretty much standard changes, a few new video formats supported, a number of improvements inherited from ffmpeg, the usual number of bugs fixed.

I still didn't get around to re-writing the installer, but i hope i can get the motivation to do it for the next big release.
What else is up next? I haven't decided. There is some features on the table, audio mixing, dvd support, some subtitle work .. we'll see.

Take care.

wanezhiling
28th January 2012, 19:31
:pFirst

nevcairiel
28th January 2012, 19:33
:pFirst

Actually, you are #8554

fastplayer
28th January 2012, 19:46
4000 and up seem to be working ok.
Confirmed, at least on my HD4670. This is the first time I've used DXVA without having anything to complain about. Dunno if it's because the drivers have matured by now or nev just knows how it's done. I suspect the latter. :D
:thanks:

dukey
28th January 2012, 19:47
excellent work nevcairiel,
Copy back is good, but native DXVA would be even better :) What do you plan to do with subtitles ?

Sebastiii
28th January 2012, 20:00
Great :) Thanks :)

vinnytx
28th January 2012, 20:49
http://xhmikosr.1f0.de/index.php?folder=bGF2ZmlsdGVycw==

Thank you very much

clsid
28th January 2012, 20:57
The funny thing is, i don't even know why it works with LAV. I don't supply the palette, but for some reason it figures it out.

I did however add some code that should try to send the palette to the decoder.

Edit:
I also figured out why/how it works with LAV, hooray. :)
I can confirm it works great now. Just curious, why did it work with LAV?

clsid
28th January 2012, 21:03
What else is up next? I haven't decided. There is some features on the table, audio mixing, dvd support, some subtitle work .. we'll see.Issue 32 would be cool as well. Loading DirectVobSub after LAV Video. Should be needed only for WMP/WMC on Vista/7. Having that feature would mean one less reason to use Haali splitter and ffdshow when wanting subs.

nevcairiel
28th January 2012, 21:09
Just curious, why did it work with LAV?

Apparently the demuxers insert the palette data in a special format at the end of the frame data, which the ffmpeg decoders can then read, and every other normal decoder would just ignore it, because it already has all the data it needs. It seems to be working just fine. :)

ffdshow should also be able to understand that data, unless you broke that part in your libav somehow. :p

Issue 32 would be cool as well.
Since other people also have commented that they want some feature done more then the others

I work on what is fun to work on, not what would be useful! :)
Also, opinions on what feature is important are quite different amongst different people.

Anyhow, i don't take requests, is all. :)

aufkrawall
28th January 2012, 22:53
Hello nevcairiel,
I think it'd be a good idea to force RGB output for FRAPS video, if possible. This is required, the official FRAPS decoder does it too. Without it the video doesn't look right.
FRAPS RGB source LAV default output:
http://www.ld-host.de/uploads/thumbnails/48f239fd34598dc0782f3f2dd78ea060.png (http://www.ld-host.de/show/48f239fd34598dc0782f3f2dd78ea060.png)

LAV forced RGB output:
http://www.ld-host.de/uploads/thumbnails/de55107ab40558b7cd4183903313d0ab.png (http://www.ld-host.de/show/de55107ab40558b7cd4183903313d0ab.png)


FRAPS 4:2:0 source LAV default output:
http://www.ld-host.de/uploads/thumbnails/b33e31341e86df8b95671028b204ef84.png (http://www.ld-host.de/show/b33e31341e86df8b95671028b204ef84.png)

FRAPS 4:2:0 source LAV forced RGB output (it's really how it should look like, it's not too bright ;) ):
http://www.ld-host.de/uploads/thumbnails/7e70bdc94309f2d3da429f0af6c1d2bd.png (http://www.ld-host.de/show/7e70bdc94309f2d3da429f0af6c1d2bd.png)

FRAPS 4:2:0 source FRAPS decoder output:
http://www.ld-host.de/uploads/thumbnails/0e9943732a44bd813e1a50cd369a0396.png (http://www.ld-host.de/show/0e9943732a44bd813e1a50cd369a0396.png)

I'm going to suggest this also for ffdshow.

CruNcher
28th January 2012, 22:59
@Nev

stall issue (like known from the beyonce stream)

http://www.mediafire.com/?k3d0sum8cepydcj

jmone
28th January 2012, 23:38
FYI - Testing with J.River MC (cross post from the Beta Forum)
=========== PC: Tested on GTS450 with DXVA2 Option and madVR =========
- VC-1 and x264 at high framerate (eg 50/60i, 50p) works just fine and is fluid with no dropped frames
- MPEG SD (720x576i) works just fine and is fluid with no dropped frames
- MPEG HD / Non Square Pixel ?? (1440x1080) just presents a static Green Video output
- DTV plays fine (not Green Video on HD Content as well)

Works pretty well on this rig

=========== HTPC =========
Here are the more intersting / exciting results. The above results was on my PC, the following is from my HTPC where I have been using the Intel IGP X3000 (i7-2600K chip). It has been working well with SW decoding (no surprise given the 2600K's processing power) over Dec but I did had to change madVR's Luma Scaling algorithms to something like Mitchell-Netravali over anything with "Taps" as it would drop frames as render times would blow out.

So here is how the IGP X3000 goes with the two HW Accel Options:

Testing on X3000 with DXVA2 Option and madVR:
- VC-1 and x264 at high framerate (eg 50/60i, 50p) works just fine and is fluid with no dropped frames
- MPEG SD (720x576i) works just fine and is fluid with no dropped frames
- MPEG HD / Non Square Pixel (1440x1080) works just fine and is fluid with no dropped frames
- DTV crashed MC

Testing on X3000 with Intel Quicksync Option and madVR:
- VC-1 and x264 at high framerate (eg 50/60i, 50p) works just fine and is fluid with no dropped frames
- MPEG SD (720x576i) works just fine and is fluid with no dropped frames
- MPEG HD / Non Square Pixel (1440x1080) works just fine and is fluid with no dropped frames
- DTV works just fine and is fluid with no dropped frames

This is only some quick testing and:
1) I've not really tried testing the madVR scaling algorithms with the various HW Decoders but it seems to give more overhead to leave as it is. I did see dropped frames in 1440x1080/50i MPEG 2 with Lanczos (3-Taps) and DXVA2
2) I only played short 5-sec clips on my test files. I was looking at madVR's OSD to see any reported issues with dropped frames and long queue times and also half an eye on the playback to see an very obvious issues.
3) I used madVR to do the deinterlacing on both Quicksync and DXVA2. I saw some odd deinterlacing judder using YADIF in some combinations.

Couple of Q:
1) If there is a speed issue with decoding in LAVVideo, apart from seeing visual issues - will madVR report issues in the queues / dropped frames
2) Let me know if there is any specific combos that need testing.

Thanks for the great work and the IGP option for HTPC looks very promising --> one less item, lower heat, power and noise!

CruNcher
29th January 2012, 00:23
Your playback is complex you have to keep an eye on your GPU load as well CPU load alone isnt gonna work these days anymore, so you could transfer some load Deinterlacing to the CPU and keep more free for the GPU that way in Interlaced Scenarios and upscaling, though SB has nice scaling (Adaptive Lanczos4) that should top MadVR in Efficiency (being fixed function) ;)

The most efficient way on SB is trying to get every possible load on the fixed function and only absolute necessary load on the EUs GPU/CPU, though this is currently only doable with EVR or a custom Renderer that can make use of these functions (Cyberlink,Arcsoft)


@Nev
Auto AV sync correction (Lav Audio) fails in this special case (emergency weather warning system)

http://www.mediafire.com/?kzpt3pictktu6bh

ryrynz
29th January 2012, 01:12
Caught this during development builds and it hasn't been fixed in 0.45, there appears to be two issues.

If IntelQuickSyncDecoder.dll is not located or cannot be loaded then LAV falls back to standard decoding but doesn't display any message.
It would be nice if there was some indication about the state of the currently running video decoder which brings me to the second issue.

IntelQuickSyncDecoder 0.24 looks to be either broken or LAV isn't loading it properly and falling back to software mode.
If I install 0.45 set QuickSync and run GraphStudios decoder benchmark I get the same results as with software decoding ~1550FPS.
If I copy IntelQuickSyncDecoder 0.22 over and then benchmark I get ~2418 FPS.

Aspeh
29th January 2012, 01:44
Me and two people i house share with one who is studying Art and Design and the other Photography and Media just don't understand why people insist on running MadVR with LAV filters. ?

Tested on various screens and computers and even one laptop LAV filters looks really nice and natural but MadVR to us looks over processed to the point of looking fake like it's been airbrushed with photoshop with bad colour saturation and is to dark to watch.

Unless we are doing something wrong we just don't get it. ?

DragonQ
29th January 2012, 02:03
LAV Filters contains a splitter and various decoders. It reads the video file, finds the video and audio streams you want, decodes them and passes them to the renderer, which then displays the video and plays the audio. MadVR is generally considered the best quality software renderer. It has advanced algorithms for resizing/upscaling that can make low resolution videos look noticeably sharper than the standard alternative (EVR) can.

For example, the best way to watch SDTV is probably to use LAV Filters (with CUVID decoding and vector adaptive deinterlacing) and MadVR (with high quality settings). This should be able to best even some hardware deinterlacers and upscalers (e.g. what you'd find in a good quality HDTV).

(I think I have all that right. :p)

Fadeout
29th January 2012, 02:05
MadVR on its own doesn't touch colors or anything, it only slightly changes scaling, and it's barely noticeable.

CruNcher
29th January 2012, 02:13
@DragonQ
EVRs quality depends on the Hardware Scaling implementation underneath it you cant generalize it in any way to "EVR is bad" thats a pure myth that MadVR Avg joe users come up with, because the Hardware implementations in their card sucks. Also EVR Custom Renderer support shader based upscaling with different algorithms as well just recently MPC-HC experimental got Lanczos support. The biggest advantage of MadVR is its Playback stability :)

ajp_anton
29th January 2012, 02:30
Does LAV Audio work with stereo speakers (headphones) on its own, or is that what the future "mixer" is for? I'd like to get rid of ffdshow for good, which is still there only for the audio.

iSunrise
29th January 2012, 02:50
Does LAV Audio work with stereo speakers (headphones) on its own, or is that what the future "mixer" is for? I'd like to get rid of ffdshow for good, which is still there only for the audio.
Yeah, I would be interested in that, too. It would be great if nev could do a mixing matrix along with some "standard" presets, so it would make ffdshow´s mixing solution pretty much obsolete for me, too.

Andy o
29th January 2012, 03:34
@DragonQ
EVRs quality depends on the Hardware Scaling implementation underneath it you cant generalize it in any way to "EVR is bad" thats a pure myth that MadVR Avg joe users come up with, because the Hardware implementations in their card sucks. Also EVR Custom Renderer support shader based upscaling with different algorithms as well just recently MPC-HC experimental got Lanczos support. The biggest advantage of MadVR is its Playback stability :)

madVR's advantage with the level expansion and RGB conversion are very visible in test gradients (no banding), but on movies I can't tell the difference, though I haven't bothered to look for it very much since madVR just works for me. It's possible that on CG animation you could find a difference.

aufkrawall
29th January 2012, 04:29
madVR's advantage with the level expansion and RGB conversion are very visible in test gradients (no banding)

What about EVR custom + LAV with forced RGB output?

DrNein
29th January 2012, 06:08
Does LAV Audio work with stereo speakers (headphones) on its own, or is that what the future "mixer" is for? I'd like to get rid of ffdshow for good, which is still there only for the audio.

ffdshow headphone mode is "meh" and I would not expect any non-commercial filters to be any better. I use CyberLink Audio for Dolby Headphone but still must use ffdshow as an intermediate for HD audio. Would like to ditch ffdshow as well.

CruNcher
29th January 2012, 06:26
What about EVR custom + LAV with forced RGB output?

Indeed especially on SB it shows its more efficient to use the RGB32 conversion for chroma upsampling than most shaders (which higher GPU usage heavily, of course if you don't mind 15% more GPU usage and want to save the last drop of CPU in my tests it showed to be roughly 1% saving on SD HD 4% MPEG-2, so 3%/25% vs 2%/40% not a good deal imho)

madVR's advantage with the level expansion and RGB conversion are very visible in test gradients (no banding), but on movies I can't tell the difference, though I haven't bothered to look for it very much since madVR just works for me. It's possible that on CG animation you could find a difference.

Not only Chroma upsampling improvements can be seen also on a lot of TV Sets that use colorful lightning (I don't mean the TV @ your home but the Set in the TV Studio so Broadcast Content any Kind of Shows for example that have complex Stages and Lightning or Pop Concerts with a lot of effects)

mindbomb
29th January 2012, 06:35
i noticed this:
playing the ifo files in dvd video results in the total calculated duration of the video being incredibly off.

Using MPC VTS Reader with Lav Splitter.


What about EVR custom + LAV with forced RGB output?
Good, but outputting rgb32 will result in no deinterlacing from the renderer iirc.

So, you should only use it for cuvid atm, or with yadif.

hissatsu
29th January 2012, 06:40
I posted this several weeks ago, but apparently this isn't a common problem, as I didn't get any responses.

I'm having an issue with seamless blu-ray playback, using every version of LAV Filters after 0.39 (though 0.45). When it switches from one m2ts in the mpls playlist to the next (like in a TV series where each episode is a separate m2ts on the BD), the screen goes black, but audio continues playing normally. Hitting Ctrl-J to get statistics shows nothing, but right clicking brings up the menu. To get the video back, I have to do something that interacts with the splitter. If I seek or skip chapters, the video comes back. Switching audio tracks also brings back the video. Skipping chapters forward or back to the point where the next m2ts starts, the screen does not go black; it only happens during normal playback at the point of the switch.

I'm using MPC-HC with MadVR on Windows 7 x64 with a GTX 550Ti, and I've experienced this on two separate Windows installs with almost completely different hardware (Intel/AMD, different video cards). I've tried every version of LAV Filters through 0.45, with hardware acceleration (CUVID, DXVA2) and without; I've tried every version of MadVR for the past few months and many builds of MPC-HC, along with other renderers (EVR-CP, EVR Sync, Haali), and no setting makes any difference other than going back to LAV Filters 0.39 or earlier. I'd really like to use a newer version, but something appears to have changed with the splitter after that point.

If anyone has any ideas/suggestions to fix this or if anyone is having similar issues, I'd really like to know.

Andy o
29th January 2012, 06:47
I know with ffdshow's HQ RGB conversion + dithering I could achieve similar, but not the same, results with madVR's tests. I haven't tested those with LAV's RGB conversion. In any case, unless you have a reason not to use madVR (like underpowered hardware), why not? It's also the easiest way to do high-quality deinterlacing and IVTC. Pretty much the only way to get it done with AMD cards. With NV cards and CUVID, you still have to select 25/30p output manually (and change to 50/60p for regular deinterlacing) and apply an avisynth script for decimation, for it to work.

CruNcher
29th January 2012, 07:00
i noticed this:
playing the ifo files in dvd video results in the total calculated duration of the video being incredibly off.

Using MPC VTS Reader with Lav Splitter.



Good, but outputting rgb32 will result in no deinterlacing from the renderer iirc.

So, you should only use it for cuvid atm, or with yadif.

Not correct depends on the Hardware (Driver), on SB its not the case (EVR Custom)

nevcairiel
29th January 2012, 07:17
If IntelQuickSyncDecoder.dll is not located or cannot be loaded then LAV falls back to standard decoding but doesn't display any message.
It would be nice if there was some indication about the state of the currently running video decoder which brings me to the second issue.

The property page has a text field next to the dropdown, which will indicate if the DLL could be loaded properly. Feedback during playback would just be terribly annoying.
It'll say "Available" if you just open it and it can load the DLL (and some basic checks succeed), and it'll say "Active" if its actually being used during playback.


IntelQuickSyncDecoder 0.24 looks to be either broken or LAV isn't loading it properly and falling back to software mode.
If I install 0.45 set QuickSync and run GraphStudios decoder benchmark I get the same results as with software decoding ~1550FPS.
If I copy IntelQuickSyncDecoder 0.22 over and then benchmark I get ~2418 FPS.

Works just fine here.
Its known to be broken for non-SandyBridge GPUs, but thats not a bug in my code, but in the DLL, so not a bug report for this thread. :)

nevcairiel
29th January 2012, 07:33
@Nev
Auto AV sync correction (Lav Audio) fails in this special case (emergency weather warning system)

http://www.mediafire.com/?kzpt3pictktu6bh

That file has issues no matter what. When i seek after that announcement, its completely desync, which means that weather thingy completely screws up the timestamps of the audio.

I'll put that into the broken file category. :)

nevcairiel
29th January 2012, 07:59
@Nev

stall issue (like known from the beyonce stream)

http://www.mediafire.com/?k3d0sum8cepydcj

Fixed

ryrynz
29th January 2012, 08:12
The property page has a text field next to the dropdown, which will indicate if the DLL could be loaded properly. Feedback during playback would just be terribly annoying.
It'll say "Available" if you just open it and it can load the DLL (and some basic checks succeed), and it'll say "Active" if its actually being used during playback.


Thanks, I was thinking of a status bar icon or something but that 'active' I guess does the trick. It shows as active so it is loading 0.24 okay.
GraphStudio still benches it slow however, does yours bench the same with 0.22 vs 0.24? I'd like to confirm before posting the bug in the ffdshow forum.
For the record I have a Sandy 2600K, and the latest ffdshow runs QuickSync with the proper performance showing in GraphStudio like in 0.22 with LAV.
It would be odd if we're running the same CPU and your 0.24 benches the same or higher than using 0.22.

nevcairiel
29th January 2012, 08:23
To get your kind of numbers, it would have to be SD or something, wouldn't it?

I benchmarked a pretty default 1080p H264 MKV with LAV 0.45:

0.24:
~200 fps

0.22 (from LAV 0.44):
~170 fps

And a SD file

0.24
~800 fps

0.22:
~750 fps

Something to note:
When decoding very short files, so that the decoding process only takes a fraction of a second for the complete file, the numbers will most likely be very inaccurate.
In fact, the only way i ever see numbers like 1500 or 2400 is when the file is extremely short.

PS:
My version of the QS decoder is slightly newer then the one that ships in ffdshow at the moment, but not sure it would explain anything. :)
Technically, its all 0.23, the latest commit just increased the version to 0.24 without implementing any of the 0.24 features yet. :p

Aleksoid1978
29th January 2012, 09:06
Have issue with LAV Audio http://aleksoid.tosei.ru/Test/Sample/Matrix%201.avi
issue only with internal MPC-HC Avi splitter and system Avi splitter. Play fine with LAV Splitter.

nevcairiel
29th January 2012, 09:43
Have issue with LAV Audio http://aleksoid.tosei.ru/Test/Sample/Matrix%201.avi
issue only with internal MPC-HC Avi splitter and system Avi splitter. Play fine with LAV Splitter.

Fixed

ryrynz
29th January 2012, 09:45
@Nev It's was anime intro 704x480 at 1 min 12 seconds.

I took to a couple of ~1920x1080 Youtube trailers and benched.
I set to the run 5 times and average the result.

Snow White and the Huntsman 1.53 minutes
Normal LAV 650 FPS
QuickSync 0.22 476 FPS
QuickSync 0.24 361 FPS

Wrath of the Titans 2.05 minutes
Normal LAV 684 FPS
QuickSync 0.22 737 FPS
QuickSync 0.24 609 FPS

Same results, 0.22 beats 0.24.

The Snow White trailer leaves me a little confused I thought QuickSync was ALWAYS going to be faster..

Qaq
29th January 2012, 09:46
What about EVR custom + LAV with forced RGB output?
This way LAV adds dither noise after YUV>RGB conversion and EVR upscales that noise. madVR adds dither after upscaling at very final stage.

CruNcher
29th January 2012, 09:51
@Nev

Stream crashes the DXVA copy back Mpeg-2 decoder (currently the only one)

http://www.mediafire.com/?y2dbekgemoeo28m

nevcairiel
29th January 2012, 10:00
Same results, 0.22 beats 0.24.


It seems to depend mostly on the stream in question, i actually found one where 0.22 is faster as well now, but the ones i tested earlier still have 0.23 being faster.

I'll report my findings to Eric.

ryrynz
29th January 2012, 10:02
Awesome, thanks for confirming, I can upload all files I've tested with if Eric would like them.

Sebastiii
29th January 2012, 10:12
I posted this several weeks ago, but apparently this isn't a common problem, as I didn't get any responses.

I'm having an issue with seamless blu-ray playback, using every version of LAV Filters after 0.39 (though 0.45). When it switches from one m2ts in the mpls playlist to the next (like in a TV series where each episode is a separate m2ts on the BD), the screen goes black, but audio continues playing normally. Hitting Ctrl-J to get statistics shows nothing, but right clicking brings up the menu. To get the video back, I have to do something that interacts with the splitter. If I seek or skip chapters, the video comes back. Switching audio tracks also brings back the video. Skipping chapters forward or back to the point where the next m2ts starts, the screen does not go black; it only happens during normal playback at the point of the switch.

I'm using MPC-HC with MadVR on Windows 7 x64 with a GTX 550Ti, and I've experienced this on two separate Windows installs with almost completely different hardware (Intel/AMD, different video cards). I've tried every version of LAV Filters through 0.45, with hardware acceleration (CUVID, DXVA2) and without; I've tried every version of MadVR for the past few months and many builds of MPC-HC, along with other renderers (EVR-CP, EVR Sync, Haali), and no setting makes any difference other than going back to LAV Filters 0.39 or earlier. I'd really like to use a newer version, but something appears to have changed with the splitter after that point.

If anyone has any ideas/suggestions to fix this or if anyone is having similar issues, I'd really like to know.

Hi,

I have tested on Terminator Salvation (on MediaPortal with LAV Splitter ofc).
I didn't have your issue, i have keep running 2 switching on m2ts (quite simple on that BD, one is @ 53 seconds and the second @ 1 minute 56 secs.)
Running on NVidia with CUDA.
I will try to do the same test with my Avatar seamless BD and on MPC-HC.
Thanks

*EDIT* : On Avatar, it's ok too (on MePo)

nevcairiel
29th January 2012, 10:15
Awesome, thanks for confirming, I can upload all files I've tested with if Eric needs them.

I actually found the reason why ffdshow seems faster, and "fixed" that, still doesn't explain why 0.22 is faster then 0.24, i'll investigate a bit before i post.

ryrynz
29th January 2012, 10:31
Nice, you're on a bit of a fixing spree! I wasn't expecting a performance drop with 0.24 after reading about the improvements Eric made over 0.22 and certainly with it depending on the stream as much as it does, I just assumed it was broken. :D

Aspeh
29th January 2012, 10:36
MadVR on its own doesn't touch colors or anything, it only slightly changes scaling, and it's barely noticeable.

If you run LAV output side by side with MadVR it certainly looks alot different but that might be because we can't seem to work out between the 3 of us how to change Colour,Contrast and Brightness of MadVR to closer match that of the LAV output.

nevcairiel
29th January 2012, 10:41
@Nev

Stream crashes the DXVA copy back Mpeg-2 decoder (currently the only one)

http://www.mediafire.com/?y2dbekgemoeo28m

That one is fine with NVIDIA, only fails on Intel for me. I would suspect a driver bug or something like that...
I'll try to figure it out, maybe. :)

DragonQ
29th January 2012, 12:13
@DragonQ
EVRs quality depends on the Hardware Scaling implementation underneath it you cant generalize it in any way to "EVR is bad" thats a pure myth that MadVR Avg joe users come up with, because the Hardware implementations in their card sucks. Also EVR Custom Renderer support shader based upscaling with different algorithms as well just recently MPC-HC experimental got Lanczos support. The biggest advantage of MadVR is its Playback stability :)
Hmmm. EVR gives me pretty bad banding on a GT 430, even when supplied with RGB32. MadVR doesn't give me this effect, but I can't use it for live TV in MediaPortal unfortunately.

I also find it strange that EVR works instantly in MPC-HC whereas MadVR has some weird issues like the aspect ratio being wrong for a split second when switching to/from full screen, and a half second delay between raising/lowering volume and the OSD coming up.

Midzuki
29th January 2012, 12:55
Fixed

Would you mind telling us what was "fixed" exactly? :)

I don't intend to download a +600MB file, just to (TRY TO) find that out by myself. ;)

nevcairiel
29th January 2012, 13:08
Would you mind telling us what was "fixed" exactly? :)

The file used to not play properly, now it does - hence, fixed. :p
It was screwing up timestamps when the demuxer didn't bother to split the data into frames, and just delivered them in one big blob

You can always read the commit log if you want to know more. ;)

Midzuki
29th January 2012, 13:56
The file used to not play properly, now it does - hence, fixed. :p
It was screwing up timestamps when the demuxer didn't bother to split the data into frames, and just delivered them in one big blob

:thanks:

You can always read the commit log if you want to know more. ;)

But of course :o :stupid: