badkins has quit [Remote host closed the connection]
SGASAU has quit [Remote host closed the connection]
badkins has joined #racket
SGASAU has joined #racket
dddddd has joined #racket
badkins has quit [Remote host closed the connection]
badkins has joined #racket
jellie has quit [Quit: WeeChat 2.8]
badkins has quit [Ping timeout: 256 seconds]
endformationage has joined #racket
orivej_ has quit [Ping timeout: 260 seconds]
badkins has joined #racket
badkins has quit [Ping timeout: 256 seconds]
TCZ has quit [Quit: Leaving]
efm has joined #racket
TCZ has joined #racket
Codaraxis has joined #racket
TCZ has quit [Quit: Leaving]
nullcone has quit [Quit: Connection closed for inactivity]
badkins has joined #racket
badkins has quit [Ping timeout: 256 seconds]
narimiran has quit [Quit: leaving]
sauvin has quit [Read error: Connection reset by peer]
badkins has joined #racket
<dzoe>
samth: I finally sorted the mess in my ~/racket-something/ directories and turns out, CS is underperforming 6-10x to BC, so my bug hunt turns to performance loss hunt ...
<dzoe>
Good news is that both 64bit and 32bit Linux builds (latest master) have no apparent problems with my workloads.
<samth>
dzoe: I think CS underperforming BC on futures by 2x wouldn't surprise me, but 6-10x is surprising
<dzoe>
But that is not the case.
<dzoe>
Looking at the processor stats makes me think that futures work better in CS.
<dzoe>
I get to sligtly over 500% with BC and almost 600% with CS
<dzoe>
So the futures-scheduling part is definitely better.
<samth>
Oh interesting
<dzoe>
But the actual operations are way slower.
<dzoe>
I will write some benchmarks that allow me to isolate the problem.
<samth>
is this all about flonums?
<samth>
in which case, it's sadly expected
<samth>
because all Chez flonums are boxed
<dzoe>
Well... the current workload is flonums-heavy.
<dzoe>
That is true.
<dzoe>
But it is in my backlog to test fixnums-only implementation as well.
<dzoe>
Problem is I finally implemented parallelized perspective-correct texture mapping with z-buffer and that is two flonum divisions per pixel :)
<dzoe>
But if the problem is with boxed values, it means even z-buffer would have terrible performance impact.
<dzoe>
(As it is 2560x1440 flonums each frame ... linear interpolation + comparisons)
<dzoe>
The thing is it should be faster with flonums since Pentium-I ...
<dzoe>
(25 years?)
deeglaze has joined #racket
<dzoe>
On the other hand ... with 64bit values it should be possible to switch to fixed-point arithmetics without major problems.
<dzoe>
I've done this on 80486 and 32bit values using 16 higher bits for the integer part and lower 16 bits for the fractional part.
<dzoe>
It will also be interesting to see how this performs on 4-core arm.
<dzoe>
Maybe I should start looking for an open PhD position so that I have a motivation to really dedicate some long-term time to it :)
lavaflow has quit [Ping timeout: 260 seconds]
hoplaahei has joined #racket
<hoplaahei>
hi. I'd like to use Racket to autogenerate some config files from templates. Should I do it in pure Racket, use a pre-processor, or would a dsl like Scribble be easier?
vraid has quit [Ping timeout: 240 seconds]
<dzoe>
hoplaahei: I mostly use #lang at-exp racket for this and just interleave @-expressions with regular code, it is usually sufficient.
<hoplaahei>
dzoe: will take a look thanks. I'm not sure I need all the document generations features of scribble, so your suggestion might be easier.
badkins has quit [Remote host closed the connection]