Re:High-Performance GPU Computing in the Julia Programming L...
by llukas on Monday October 28, 2019 @05:30AM
(#21375462)
Did you compare the performance of nvrtc vs offline nvcc compiler?
Re:High-Performance GPU Computing in the Julia Programming L...
by Athas on Monday October 28, 2019 @06:30AM
(#21375718)
No; the code we would need to generate would be rather different. Would you expect a significant difference? When we did research on nvrtc before implementing this, we couldn't find any concrete information that nvrtc should generate slower code.
In fact, I have run tens of thousands of lines of essentially equivalent CUDA and OpenCL code (automatically generated) on the same hardware, and performance was in all cases very similar[0]. If anything, CUDA was actually slower than average (but in the cases I investigated, this was down to arbitrary differences like the CUDA compiler not unrolling some loops as aggressively and such).
[0]: https://futhark-lang.org/blog/2019-02-08-futhark-0.9.1-relea...