Asher
banner
ashermancinelli.bsky.social
Asher
@ashermancinelli.bsky.social
compilers and coffee and lifting

Compiler engineer at nvidia
These views are exclusively my own

https://www.ashermancinelli.com/about.html
The Flang documentation and source code is up there too! Glad you liked the episode!
September 21, 2025 at 4:39 PM
Thanks! Michael Wolfe’s racecar book might be a fun read: www.goodreads.com/book/show/33...
High-Performance Compilers for Parallel Computing
This work covers everything necessary to build a compet…
www.goodreads.com
September 21, 2025 at 4:38 PM
Thanks a bunch!
July 25, 2025 at 8:56 PM
Thanks for the read! I know next to nothing about rust so I’d love to revise it
July 25, 2025 at 4:27 PM
There are some warts but a huge development in compiler tech imo
July 25, 2025 at 4:23 PM
You might think of it as 500 seperate IRs or progressively lowering one mega ir, but principle is the same. Nicer than the typical micropass arch imo. Things can move up or down in the pipeline as needed which isnt always the case w micropasses
July 25, 2025 at 4:22 PM
Lol ic ic. One of the nice features of mlir is its heterogeneity for this reason- all your IRs still use the same substrate+infrastructure and you can progressively lower from higher to lower level semantics
July 25, 2025 at 4:20 PM
Interesting. Whoms your intended user? What kinda code are they writing?
July 25, 2025 at 4:16 PM
Oh gotcha! I’d love to chat more about it haha
July 25, 2025 at 4:14 PM
That was sorta the premise of this blog post for anyone interested www.ashermancinelli.com/csblog/2025-...
My Ideal Array Language - Notes
www.ashermancinelli.com
July 25, 2025 at 4:13 PM
Excited to see what y’all figure out! I think it’s really cool to have unbufferized values at the ssa stage - most compilers choose between the two but when you have both, it’s like youve forced the src lang to be pure functional, so its easier to optimize
July 25, 2025 at 4:12 PM
So the hir isnt ssa?
July 25, 2025 at 3:26 PM
The llvm flang compiler has some cool unbufferized array SSA types at the highest level IR, wonder if theres space/use for that in HIR
July 25, 2025 at 2:58 PM
What semantics are you representing in the higher level ir that you couldnt/didnt at the lower ones?
July 25, 2025 at 2:24 PM
What’s julia’s memory model for arrays? It’s got an llvm based jit, right?
July 23, 2025 at 12:46 PM
I’m all for julia but there’s some super neat stuff in fortran! Also, the llvm fortran compiler is super fancy- maybe the first general purpose mlir based compiler (as in, not ai related)
July 23, 2025 at 12:45 PM
What was your fortran xp like?
July 23, 2025 at 1:09 AM
Yoooo
July 23, 2025 at 1:09 AM
Hey welcome!
November 25, 2024 at 3:18 PM
Hey thank you for this!
November 25, 2024 at 3:13 PM
Please make a starter pack if you do find more🙏
November 21, 2024 at 2:17 AM