Vluzacn's ZHLT v34 Update

[Phantomas]

it uses the brightness level of the point directly underneath the player

Nope .


PS:
Vaqtincha want make shadows exactly as you say (Solokiller):

[FreeSlave]

2) The number of CPU threads cannot be automatically detected. The user should always specify '-threads #' through command line, otherwise it's default to 1 thread.

Besides sysconf(_SC_NPROCESSORS_ONLN) there's linux specific API (i.e. not available on other unices) that can give you the number of CPU: http://linux.die.net/man/3/cpu_isset

See how it can be used: https://bitbucket.org/FreeSlave/vhlt...hreads.cpp-564

[amckern]

Started a branch for AVX2, don't know enough about AVX2-512 - and don't have a CPU that supports it either - so AVX2 at the moment:

http://downloads.ammahls.com/zhlt/VHLT34_avx2_0.1.zip (64 Bit build only)

Control Map - bigcave2 (http://scmapdb.com/map:big-cave-2)

I would consider gains >5s are an improvement, as you can see the most build is 2 seconds slower then my original version, even though i have not changed any code. The CPP check build was based around an optimization recommendation in the vis.cpp file - it ended up not being much faster (at around 200 milliseconds)




Feedback and comments?

[amckern]

Please test brinks - this build had a lot of adjustments made within brink.cpp

https://cdn.discordapp.com/attachmen...07.03.2018.rar

[Hezus]

Please test brinks - this build had a lot of adjustments made within brink.cpp

Gave this build a shot but it seems it doesn't support maps on large grid scale. Got a whole bunch of errors like the one below, and then VIS and RAD crashed.

Warning: Entity 41 (classname "trigger_numericdisplay"): origin outside +/-4096: (-5024,-3080,264)

[amckern]

Warning: Entity 41 (classname "trigger_numericdisplay"): origin outside +/-4096: (-5024,-3080,264)

I used the source code found in the first post (v34): https://forums.svencoop.com/attachme...7&d=1439879328

Do you have trouble with it? If you compile it your self, and attempt to run a map compile do you have trouble as well?

[Hezus]

I'm using the custom build of V34 from the SC SDK (hlcsg v3.4-VL34-SC5.0 64-bit). I think this one includes the grid fix where the original VL34 doesn't.

[amckern]

I'll see if i can merge the fix with this build then

[AdamR]

This just makes me realise how old my Intel i7-920 is getting, even if it does run at 3.8 GHz instead of 2.67 GHz.

[Hezus]

You're going to need to ask someone else to test it for you, since I just realised that my i5-2500K doesn't support AVX2.

@AdamR: Somehow I feel that CPU's don't age that fast anymore. My i5 is nearly 7 years old and still runs smoothly with pretty much all Windows 10 applications. In comparisment I had a 286 in 1992 and a Pentium II in 1999. Now there's a difference

[AdamR]

Heh yeah, especially as my one before this was an AMD Athlon 64 x2 4800+. Also considering mine runs faster than the top model in the line (i7-995X Extreme Edition) it's just getting an even longer life. Jumping from 2 threads to 8 was also very significant.

PassMark: AMD Athlon 64 x2 4800+, Intel i7-920, and Intel i7-995X compared

Mine probably consumes quite a bit more power than a modern equivalent though. My work PC has an i7-6700, and that runs faster using significantly less power.

[GiGaBiTe]

Mine probably consumes quite a bit more power than a modern equivalent though.

Not "probably", it does consume significantly more power, especially overclocked. i7-920 is a 130W TDP part normally with a max power draw of 230W under heavy load. It pulls about 45% more power when overclocked to 3.8 GHz, or ~330W max under heavy load.

I have an i7-6700k which runs at 4 GHz stock with a 91W TDP.

[amckern]

I have been tinkering with the newer stuff for speeding the compile times within the SvenCoop tools source code, at the moment i wont be publishing anything here as we need to wait 5.21 to be released with support for the new default chop size in CSG.

[GiGaBiTe]

The compile tools are pretty efficient already, even an old quad can compile a large and complex Sven map in just a few minutes, assuming they're optimized properly.

I'd be cool if the compile tools had MPI for distributed compiling. We have the old netvis.exe from eons ago which allowed distributed compiles for at least the vis process, but it never really developed beyond the "let's see if this can work" phase. Distributed compiling for lighting would have been more useful, but I think that would have been a whole lot more difficult to code vs. sorting leaf visibility in a BSP tree.