 |
Post Aoad4lAXmngo8mMBjU by [email protected] |
 |
More posts by [email protected] |
|
Post #AoZ5bUrrp7KxXX6F5E by [email protected] |
|
0 likes, 1 repeats |
|
|
|
if you look at what LLVM actually spends its time doing (and I assume probably … |
|
|
|
Post #AoZ5bV3Z7ch87oZaJk by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@regehr As usual, all the discussion of "how do you get it on the GPU"… |
|
|
|
Post #AoZ5bVBMecvuW0DoTQ by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@TomF @regehr it's the legacy of serial programming. we need to learn to th… |
|
|
|
Post #Aoad4lAXmngo8mMBjU by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@regehr I wonder if Cranelift is better about the pointer chasing. @cfallin an… |
|
|
|
Post #Aoad4lLB9GCEflKgJE by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@djc @regehr interesting question -- I've never tried to do a pointwise com… |
|
|
|
Post #ApAwdA0ooqmvr7Ty3k by [email protected] |
|
0 likes, 0 repeats |
|
|
|
here's the code for this all-on-the-GPU compiler https://github.com/Snektro… |
|
|
|
Post #ApAwdAEzy88AZ67IA4 by [email protected] |
|
0 likes, 0 repeats |
|
|
|
the disappointing thing about their performance results is that register alloca… |
|
|
|
Post #ApAwdAQhGdUL9NadOa by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@regehr the worse news is that that's like a 10x improvement for big inputs… |
|
|
|
Post #ApAwdAc2aSYviYth4q by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@malwareminigun I guess the hope is that for huge compiles,. maybe the GPU vers… |
|
|
|
Post #ApAwdAlG2Bw2B9D3RY by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@regehr @malwareminigun Assuming you can actually turn the compilation into som… |
|
|
|
Post #ApAwdAu7VF1YcdM8G0 by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@saagar we’re comparing with a real implementation here. Given how traditiona… |
