Heavy instancing and multithread parsing times

[Jahred]

Back in time, multithread parsing was much faster than monothread but that is no longer true. Test renders below with current corona daily and a forest object, time difference -7%

This 2016 topic concern the same problem : https://forum.corona-renderer.com/index.php?topic=12201.msg78706#msg78706

With Corona 6 (7?) and the "rewriting scene parsing" (trello road map), could you take a second look to optimize 'heavy instancing' ?

I struggle with heavy scene (many Railclone and Forest Pro objects, tiles, grass, gravel, etc and large surfaces) with 2 to 5 minutes of parsing, a little frustrating to make multiple test render to check my full project with many cameras.

Ironically the "grainy" test render itself take often less time to tell me if it's ok, Corona stay very fast for pure raytracing.

[rowmanns]

Hi,

Thanks for reporting this. We will investigate.

Rowan

(Internal ID=410588911)

[Ondra]

Did you try current v5 RC? There are some optimizations, in our heaviest scene, the parsing time was halved

[Jahred]

Hi Rowan, Ondra,

Tested with the last RC, it's strange.

First batch, the multithread parsing is faster, with a gain time compared to the first test (with +25% instances). For the second batch (*_02), the multi is slower than the monothread and the total gain time is lost compared the the 1st test.

Performance bug with Forest Pro ? The real problem remains the same, the difference between mono and multi is small.

[rowmanns]

Hi,

Are you using the latest version of ForestPack?

Rowan

[Jahred]

Last version yes, Forest Pack 6.2.2

[rowmanns]

Hi,

Thanks for the information.

I have spoken to our dev team, basically FP scatter objects are only parsed in a single thread so this is why you are seeing no difference between having multithreaded parsing enabled and disabled.
Secondly we will however investigate why is is taking so long to parse these FP instances as for example with CoronaScatter it takes much less time, especially in the latest daily build.

Cheers,

Rowan

[Jahred]

I am glad to hear you'll investigate.

When rewrite of parsing will occur, the dev team should take a second look at "multithread capability for one scatter object" as well. I repeat myself :)

Anyway, good work on general optimisation, I see good improvement in many scenes.