hughleat

8 ms·

1. 1.▲
   by hughleat 2y ago

   Never let a computer scientist name anything :-)
2. 2.▲
   by hughleat 2y ago

   We plan to have a peer reviewed version of the paper where we will probably have more details on that. Otherwise we can't give anymore details than in the paper or post, etc. without going through legal which takes ages. Science is get
3. 3.▲
   by hughleat 2y ago

   Yes! An AI building a compiler by learning from a super-optimiser is something I have wanted to do for a while now :-)
4. 4.▲
   by hughleat 2y ago

   For the disassembler we round trip. x86 ->(via model) IR ->(via clang) x86. If they are identical then the IR is correct. Could be correct even if not identical, but then you need to check. For the auto-tuning, we suggest the best pas
5. 5.▲
   by hughleat 2y ago

   It can emulate the compiler (IR + passes -> IR or ASM). > What would be more interesting is training a large model on pure (code, assembly) pairs like a normal translation task. It is that. > Presumably a very generalized model wou
6. 6.▲
   by hughleat 2y ago

   There are two models. - foundation model is pretrained on asm and ir. Then it is trained to emulate the compiler (ir + passes -> ir or asm) - ftd model is fine tuned for solving phase ordering and disassembling FTD is there to demo capab
7. 7.▲
   by hughleat 2y ago

   Yep! No GCC on this one. And yep, that's not far off how the pretraining data was gathered - but with random optimisations to give it a bit of variety.
8. 8.▲
   by hughleat 2y ago

   Yep, we tried it :-) These were exactly the problems we had with it.