Scale lang (basically) a superset of cuda for amd gpus #23
Description
Note i just paraphrase the docs and i don't expect anything other than the more you know. So maybe hvm3.X/4 support intel and amd.
And no expertise just digging rabbit holes
On ZLUDA return to development last year.
it supporting uhd intel and rdna amd and later for the end user not the developer thou but tests don't hurt, they say they'll be production ready in q3 (paraphrased)
From the gpu support alone i see that
for amd scale is fit and supports nvidia and because they say ROCm is the basis of expansion intel maybe supported.
And intel zulda 3 for end user so test and give no guarantee and zulda 4 on q3 withe same treatment
Why:
Cuda on non nvidia gpus (which is sooner or later under your interest of accessibility ) should easily work compiled by scale
or "copy-pasted" (idk the term) as it's almost a superset
And the earlier the easier
Note info (of the original issue) from 15 minute search so not the best
From official website
SCALE is a GPGPU programming toolkit that allows CUDA applications to be natively compiled for AMD GPUs
SCALE does not require the CUDA program or its build system to be modified.
And
SCALE accepts CUDA programs as-is. No need to port them to another language. This is true even if your program uses inline PTX asm.
The SCALE compiler accepts the same command-line options and CUDA dialect as nvcc, serving as a drop-in replacement.
"Impersonates" an installation of the NVIDIA CUDA Toolkit, so existing build tools and scripts like cmake just
So i think that is under your interest, sooner or later it'll be a must for accessibility except intel (arc gpus are good) and phone gpus
And iirc igpus have better cpu connection
PS just a theoretical no expertise whatsoever other than digging rabbit holes
PS 2 not a native English speaker (and the Punctuation is misused and bad because it is omit in Arabic ie. my native language)
Addition 1 from official website you can compile CUDA with scale
Addition 2 there's differences (idk how much, no expertise as i said) (doc page: https://docs.scale-lang.com/manual/differences/) but it's still kind of a superset
Thou Possible Incompatibility due to some unavailable functionality eg per-thread default stream behaviour