I have my fairly large game engine ish project which uses luajit that I
develop on my freetime (which is pretty much all the time) for personal
fun. I think it can be useful for doing some high level performance tests
but to figure out why something is slow I'd probably have to look at it
myself because of code complexity. I'm not a game engine expert or anything
either so I might be doing something stupid with opengl and whatever too.
Jit is essential here because a 3d scene can go from 200 fps to 10 fps by
turning it off. Right now I have to use a patch for some NYI case with
initializing unions found here on the mailing list (I think the patch is a
dirty hack). I use unions so a vec2 can be a size or position depending on
the context. This sounds like a preprocessor's job though. Without this
patch performance noticeably drops a lot. I had to raise the maxmcode size
too.
I feel like I'm bragging about how big and complex this is but I'm just
hoping it might be useful for testing. Obviously if you want to make a game
engine do it in C or something or use ue4 or unity.
the code can be found here: https://github.com/CapsAdmin/goluwa
On Fri, Aug 21, 2015 at 4:11 PM, malkia <malkia@xxxxxxxxx> wrote:
In short - possibly it might be better to reuse established tests,
although for a different system, rather than come up with completely
new ones, but I wouldn't bet my life on it :)
On Fri, Aug 21, 2015 at 7:10 AM, malkia <malkia@xxxxxxxxx> wrote:
I was just thinking out loud. I'm in no position to know much aboutwrote:
tests anyway (Did 15+ years of game development mostly, where
asserts() were prevalent and manual QA - for both tools & game).
So I was just thinking if there are some unknown bugs in luajit
(interpretter, jit, etc), maybe porting a system to lua might help
since that system might come with extensive tests what is expected for
given situations, and the code in lua does not follow it, crashes,
etc. But it's a long shot. Even sqlite with it's extensive testing
suite, still has bugs, but from my work with sqlite they seem to be
most of the time new features.
Or take some other relatively small project - like zlib, libpng, and
completely port it there, and then reuse their tests, even port their
asserts, or so.
Again this might help nothing about the system, but having luajit
tests is really important.
(Oh, the llvm, gcc compilers must have some good tests too, a lot of
these probably won't make sense for lua/luajit in general, but some
others would)
On Thu, Aug 20, 2015 at 6:13 PM, demetri <demetri.spanos@xxxxxxxxx>
Here is a crazy idea. If someone is to port a rather well established
project to lua [snip]
I think this would be a great addition to the test suite already in
preparation, but perhaps I have misunderstood you: as far as I can
tell this wouldn't involve any work on LuaJIT itself. To put it another
way, I don't see how doing this (valuable, I agree) project proves that
we can continue feature development on LuaJIT internals.
Do you mean, perhaps, working on some specific LuaJIT feature that
would, for example, help optimize code for such a port of (say) sqlite?
If so what features (from the issue tracker or otherwise) do you think
might be helpful in that regard?
Or if I've just misunderstood you entirely please feel free to
correct me :)
Cheers,
Demetri
--
Dimiter "malkia" Stanev,
ICQ: 21875894
malkia@xxxxxxx
malkia@xxxxxxxxx
--
Dimiter "malkia" Stanev,
ICQ: 21875894
malkia@xxxxxxx
malkia@xxxxxxxxx
