Y halo thar

As a HL1 mapping/development community I thought you might be interested in a BETA build of ZHLT 3.4 that makes use of SSE2 instructions on processors that have them (Intel: Pentium 4 or later, AMD: FX lines or later and most 64-bit processors).

Amckern, who currently maintains ZHLT gave these to the Sven Co-op development team, and most of us have noticed roughly 15%-20% speed improvement on HLRAD after comparing compiles with and without the SSE2 builds.

These compilers WILL require a processor capable of SSE2 instructions, you can use CPU-Z to or something similar check if your processor has the SSE2 instruction set.

Downloads:

Mirrors for 32-bit (IA32) Windows: Amckern's, AdamR's
Mirrors for 64-bit (AMD64) Windows: Amckern's, AdamR's
Mirrors for all Windows compiles, contains latest non-SSE2 builds too: Amckern's, AdamR's

Feel free to mirror these - please ensure you tell users about the SSE2 requirement if you host them elsewhere, and the fact this is a BETA build.

About multi-threading / 64-bit:

The compilers DO multi-thread (and seem to do so very well), so dual/quad core processors could probably see a near double/quadruple performance increase. You can use the -threads X parameter to limit the compiler to a specific number of cores if you wish.

As for 64-bit, I've not done any benchmarking - however when I switched from XP Pro to XP Pro x64 I did notice a fair chunk of time difference, however I can't say for sure if it was half or anything that drastic.

Wow , very nice to see zhlt is still in development ( and who knows, maybe hammer 3.5 still is too )

AdamR, if you can get me a list of added parameters (if any), i can add upgrade my Compilator to support this new beta version of ZHLT 3.4.

Thanks.

I don't think there are any new parameters, I will check though

That looks awfully similar to Nem's Batch Compiler...

How cool! I'll do test out the new 64-bit tools and report back results on my machine. THANKS

)

edit: Cpuz says my cpu has SSE2 instruction sets, and i'm running x64, but my results are dismal using these upgraded tools compared to the latest regular ZHLT 3.4 x64 final...

VIS takes 4 times lonter and RAD time was doubled... I'll try restarting my comp, doing it again, and try some other tests tho... Is there something else i have to do?

Nah they just work better - no new flags

The x64, are optimized for AMD 64 - they might not work as well with Intel - i also don't own any 64 bit CPU's - so if your having issues, fall back to the 3.4 Final build that uses SSE1

AWW i was so excited! = (

Is this the real AMCKERN?! HOLY CRAP thanks for you work on the ZHLT tools over the years WE WORSHIP YOU HERE!

)

Interesting, gonna test them right now and see what happens.
Used the 32bit tools.
Unfortunately the compiling times increased. I tested them on a small map, perhaps i should try them on larger maps, that compile a lot longer.

I'd try them but i have a Intel Dual core.
they probably won't work for me from what i'm hearing.
And how nice of you to drop by Amckern!

I been hear for a while - i use a Pentium D (More or less a el cheapo dual core) - the only thing that wont run on more then one core is BSP, because the windows IO library that i use for BSP culling is only designed for one core.

I found when trying to reproduce my original compile times from old maps on my current system configuration, I wasn't able to(I recently fresh installed vista 64 on my machine, and fsr compiles are taking 4 times as long as my previous, stock installed os...troubleshooting now)

So in other words, it looks like i'm getting similar or identical performance between the 3.4 x64 FINAL tools, and the SSE2 upgraded ones.

Sorry 'bout the misinformation! I'll test more and report back if i find any more interesting info.. = )

So this amckern has a hand in ZHLT?

Sadly

GAH I was on POWER SAVING! = (
(that's why my compiles were taking 4 times as long(i was so scared!)

The power went out the other night, and i switched to power saving mode. Stupidly, I thought the Windows automatically would switch back to "high performance" when it started to recieve AC power againe....

So anyway, i'll do sum moar compile testing under "high performance" and post my actual results! SORRY!

Sadly?

Results for ZHLT 3.4 x64 Final vs. 3.4 x64 SSE2, on an intel-based x64 system:

On a medium map, vis took 30 seconds longer and RAD was 30 seconds shorter... so, no real increase or impediment from what i can see. Cool friggin update anyway, and very exciting to see an update on the tools regardless!

Sadly unhappy -

I dunno what you mean, you've made our lives so much easier by maintaining and updating the tools...WE LUV U!

)

I was joking

I've just compiled a large map on both x86 and x64 versions.

x86 --
CSG: 14.67 seconds elapsed
BSP: 6.34 seconds elapsed
VIS: 5550.31 seconds elapsed [1h 32m 30s]
RAD: 1257.22 seconds elapsed [20m 57s]

x64 --
CSG: 5.63 seconds elapsed
BSP: 3.95 seconds elapsed
VIS: 4421.16 seconds elapsed [1h 13m 41s]
RAD: 1018.92 seconds elapsed [16m 58s]

Looks like some improvement there

I've also been modifying the compilers, and so far successfully compiled a map with more leaves than the default 8192 limit (now 32767).

One query I have is why does compiling for x64 give me a warning "unknown option, ignoring: /ARCH:SSE2" when x86 doesn't... does x64 make use of SSE2 already or something?

[Edit] ^ never mind, discovered that the Microsoft C++ compiler can't compile a 64-bit program WITHOUT using SSE2. Makes sense seeing as pretty much any AMD64 based processor includes the SSE2 instruction set.

One thing i noticed is if i ran with the latest 64-bit tools, i had some lighting glitches, glitches that disappeared when i compiled with the 32-bit tools. I would have to test more to say for sure, but i thought i would throw it out there and see if anyone else noticed the same thing?

Too bad this was released on the snarkpit first, what a competitor we are, to be the second to get this amazing release, usually a good competitor get's the good shiz first if you know what i mean.

EDIT: Ignore drunken post.

So if you're on 32-bit, like myself, you're not going to get any benefits from the new built?

Probably will get some benefits muzz, as it uses the processer for this build and i assume the last one didn't, but i think he's saying 64 bit would just run a bit better...?

(No, i don't fully understand this release, don't quote me on this post)

EDIT:

ZHLT 3.4 that makes use of SSE2 instructions on processors that have them (Intel: Pentium 4 or later, AMD: FX lines or later and most 64-bit processors).

This is where i got my info from in the post.