Re:PyPy JIT for 64-bit ARM
by comicjk on Sunday July 28, 2019 @03:29PM
(#20549381)
I'm surprised and humbled that a project of this importance doesn't have their own physical ARM machine for benchmarking. Compare that to the computer and human time saved across all the places it could be used!
[ Parent
| on HN ]
The problem is that ARM hardware with enough RAM for their needs is extremely rare.
This work was sponsored by ARM itself, though, so it is a bit surprising they didn't supply suitable hardware from one of their licensees (they must have some to do benchmarks themselves).
PyPy is a little bit of a pathological case of their benchmark suite requires that much DRAM though
SolidRun has a MiniITX board supporting up to 2x32GB for around $1000 [0]. They also have a preproduction units for another board for $550 supporting as much RAM [1].
They probably had their reasons, but I'm just showing what's available.
https://www.solid-run.com/product/SRLX216S00D00GE064C13CH/
https://www.solid-run.com/nxp-lx2160a-family/honeycomb-works...
Why does the benchmark suite need so much RAM ?
I would expect benchmarks to match typical usage, if PyPy can't run in a couple of GB then choose a better language.
Compiling pypy is memory intensive, I don't believe it can be done in 2GB which is where most of the really affordable ARM devices seem to cap out.
But why do you have to compile PyPy on the same machine you’re testing it on?
Well cross-compiling (aka "cross translation) is made complicated because PyPy doesn't use a normal toolchain, but that's not the problem here. The PyPy benchmark suite itself takes that much memory to run.
You can build OpenJDK with the Hotspot JIT on a 2GB ARM64 system.
So the use cases for this project must also be extremely rare?
Don't want to be rude (nor rehash old topics) but is pypy used that much to be considered important? People with performance goals go to C, whether its a library or some Cython.
Don't get me wrong, i love pypy and the rpython jit.