hkaiser changed the topic of #ste||ar to: STE||AR: Systems Technology, Emergent Parallelism, and Algorithm Research | stellar-group.org | HPX: A cure for performance impaired parallel applications | github.com/STEllAR-GROUP/hpx | This channel is logged: irclog.cct.lsu.edu
FunMiles has joined #ste||ar
FunMiles has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
Yorlik__ has joined #ste||ar
Yorlik_ has quit [Ping timeout: 256 seconds]
FunMiles has joined #ste||ar
K-ballo has quit [Quit: K-ballo]
zao has quit [Ping timeout: 240 seconds]
zao has joined #ste||ar
FunMiles has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<jedi18[m]>
If you checkout to 4f49820 you'll see that we have inner_product, but then if you checkout to the merge commit 85d43e5 it's not there anymore
<jedi18[m]>
So yeah it was a probably a mistake, I'll reintroduce the file? (and also adapt it to c++20 while I'm at it?)
<jedi18[m]>
Btw can those who are on libera chat see what someone has messaged when they're not online or will they have to manually go check the logs? (which I'm assuming not everyone does every time)
hkaiser has joined #ste||ar
<jedi18[m]>
hkaiser: So I ran inclusive scan on medusa on a single node. PAR is with the new partitioner code, OLD is the old one
<hkaiser>
jedi18[m]: yes, I think inner_product is the name used for the sequential algorithm
<hkaiser>
the parallel algorithm has lessened requirements on the operator, so they decided to use a different name
<hkaiser>
the parallel version is transform_reduce
<hkaiser>
transform_reduce requires op1 and op2 to be commutative and associative, but inner_product makes no such requirement
<hkaiser>
jedi18[m]: for the numbers - could you create a graph showing the two versions? (x axis array size, y-axis speedup/slowdown compared to seq execution)
<hkaiser>
also, I think you could do the same measurements with varying chunk sizes, I believe the speedup gained could be improved this way
<jedi18[m]>
Ok so we don’t need to add inner_product since the user can just use the stl version?
<hkaiser>
our transform_reduce has a sequential version, doesn't it?
<jedi18[m]>
Sure ok I’ll create the graph
<hkaiser>
for inner_product: yes, no need to add it
<hkaiser>
sorry for my confusion yesterday, I had to go back to my notes to remind myself what happened
<jedi18[m]>
Ok I’ll try varying the chunk_size and also try to optimise the implementation further if possible
<hkaiser>
jedi18[m]: nice
<hkaiser>
we might also want to measure other algorithms that rely on the scan partitioner
<jedi18[m]>
Yes transform reduce does have a sequential version
<hkaiser>
+1
<hkaiser>
so we're good
<jedi18[m]>
Oh ok yeah, I’ll measure the other algorithms too
<jedi18[m]>
Yep, thanks!
<hkaiser>
jedi18[m]: I removed inner_product from the various tickets
<jedi18[m]>
Ok thanks
<hkaiser>
jedi18[m]: I also merged the fix for mismatch et.al. - is there now something left for starts/ends_with ?
<hkaiser>
ahh - it has conflicts now (#5410)
<jedi18[m]>
I’m not sure, I think one starts_with test fail was unrelated to the mismatch fix, let me investigate it a bit more and I’ll let you know if it’s ready to be merged
<hkaiser>
ok, thanks
<jedi18[m]>
hkaiser any idea how I can use the --repeat-until-fail when running the test through VS?
<hkaiser>
jedi18[m]: you cannot as VS doesn't rely on ctest
<hkaiser>
you can do it on the command line only
<jedi18[m]>
Oh ok got it, thanks
hkaiser has quit [Quit: Bye!]
hkaiser has joined #ste||ar
FunMiles has joined #ste||ar
FunMiles has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
FunMiles has joined #ste||ar
FunMiles has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
FunMiles has joined #ste||ar
FunMiles has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
FunMiles has joined #ste||ar
FunMiles has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
FunMiles has joined #ste||ar
FunMiles has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]