in llvm-aie documentation for the AMD NPUs:
> scheduler support for negative operand latencies (i.e. an instruction writing to a register may be scheduled after a corresponding use);
> scheduler support for negative operand latencies (i.e. an instruction writing to a register may be scheduled after a corresponding use);
November 12, 2025 at 2:36 PM
in llvm-aie documentation for the AMD NPUs:
> scheduler support for negative operand latencies (i.e. an instruction writing to a register may be scheduled after a corresponding use);
> scheduler support for negative operand latencies (i.e. an instruction writing to a register may be scheduled after a corresponding use);
so originally i wasn't necessarily hired to be that technical, but when you're the first ops person at a compiler company, you have to learn pretty quickly about what clang and llvm are and what an opengl canvas is and why uikit is great on ios but hell on android where the aspect ratios are nuts.
November 12, 2025 at 12:20 PM
so originally i wasn't necessarily hired to be that technical, but when you're the first ops person at a compiler company, you have to learn pretty quickly about what clang and llvm are and what an opengl canvas is and why uikit is great on ios but hell on android where the aspect ratios are nuts.
I’m at this point strongly considering just writing a SuperH backend for LLVM, the instruction set is relatively small so might be feasible for me
November 12, 2025 at 7:24 AM
I’m at this point strongly considering just writing a SuperH backend for LLVM, the instruction set is relatively small so might be feasible for me
Heads up for maintainers of kernels in pmaports: we are migrating all kernel packages to build with LLVM where feasible.
"Adopting LLVM for our kernel builds comes with several advantages. It will let us enable ThinLTO for improved performance without compromising too much on compile times and […]
"Adopting LLVM for our kernel builds comes with several advantages. It will let us enable ThinLTO for improved performance without compromising too much on compile times and […]
Original post on social.treehouse.systems
social.treehouse.systems
November 12, 2025 at 5:54 AM
Heads up for maintainers of kernels in pmaports: we are migrating all kernel packages to build with LLVM where feasible.
"Adopting LLVM for our kernel builds comes with several advantages. It will let us enable ThinLTO for improved performance without compromising too much on compile times and […]
"Adopting LLVM for our kernel builds comes with several advantages. It will let us enable ThinLTO for improved performance without compromising too much on compile times and […]
vast-and-endless-sea language family frontend for llvm
November 11, 2025 at 11:58 PM
vast-and-endless-sea language family frontend for llvm
meow meow meow meow meow meow meow meow meow meow meow meow meow meow meow meow meow wav codec meow meow meow meow meow meow meow meow meow meow meow moew moew meow emeo emeo meow lexer.cpp meow meow parser.cpp meow codegen.cpp meow meow include "llvm/IR/LLVMContext.h" meow meow meow meow meow nyan?
November 11, 2025 at 5:13 PM
meow meow meow meow meow meow meow meow meow meow meow meow meow meow meow meow meow wav codec meow meow meow meow meow meow meow meow meow meow meow moew moew meow emeo emeo meow lexer.cpp meow meow parser.cpp meow codegen.cpp meow meow include "llvm/IR/LLVMContext.h" meow meow meow meow meow nyan?
vscode is slow as fuck when i work on llvm how am i supposed to get any real compiler stuff done bruh
November 11, 2025 at 3:17 PM
vscode is slow as fuck when i work on llvm how am i supposed to get any real compiler stuff done bruh
https://www.phoronix.com/news/Linux-6.19-Patch-Would-MS-Ext
LinuxカーネルがMicrosoft C Extensionsの有効化に向けて動いています。
GCCとLLVM/ClangでMicrosoftのC拡張を利用できるようになります。
Linux 6.19でのマージが検討されており、より良いコードが書けるようになる可能性があります。
LinuxカーネルがMicrosoft C Extensionsの有効化に向けて動いています。
GCCとLLVM/ClangでMicrosoftのC拡張を利用できるようになります。
Linux 6.19でのマージが検討されており、より良いコードが書けるようになる可能性があります。
The Linux Kernel Looks To "Bite The Bullet" In Enabling Microsoft C Extensions - Phoronix
www.phoronix.com
November 11, 2025 at 3:54 AM
https://www.phoronix.com/news/Linux-6.19-Patch-Would-MS-Ext
LinuxカーネルがMicrosoft C Extensionsの有効化に向けて動いています。
GCCとLLVM/ClangでMicrosoftのC拡張を利用できるようになります。
Linux 6.19でのマージが検討されており、より良いコードが書けるようになる可能性があります。
LinuxカーネルがMicrosoft C Extensionsの有効化に向けて動いています。
GCCとLLVM/ClangでMicrosoftのC拡張を利用できるようになります。
Linux 6.19でのマージが検討されており、より良いコードが書けるようになる可能性があります。
More performance is always a good thing, right? Our colleague Mikhail’s RISC-V Summit talk “Unlocking 15% More Performance: A Case Study in LLVM Optimization for RISC-V” is now available from the Summit’s YouTube channel! www.youtube.com/watch?v=cWzu...
Unlocking 15% More Performance: A Case Study in LLVM Optimization for RISC-V - Mikhail R. Gadelha
YouTube video by RISC-V International
www.youtube.com
November 10, 2025 at 8:42 PM
More performance is always a good thing, right? Our colleague Mikhail’s RISC-V Summit talk “Unlocking 15% More Performance: A Case Study in LLVM Optimization for RISC-V” is now available from the Summit’s YouTube channel! www.youtube.com/watch?v=cWzu...
LLVM Weekly - #619, November 10th 2025. DFAJumpThreading on by default for CoreMark perf boost, further CAS development, MLIR pygments lexer improvements, Arm contributions to LLVM 21, and more llvmweekly.org/issue/619
LLVM Weekly - #619, November 10th 2025
llvmweekly.org
November 10, 2025 at 8:11 PM
LLVM Weekly - #619, November 10th 2025. DFAJumpThreading on by default for CoreMark perf boost, further CAS development, MLIR pygments lexer improvements, Arm contributions to LLVM 21, and more llvmweekly.org/issue/619
It's also possible I'm mixing up GCC'isms and LLVM, so my apologies if I maybe give false hope.
That said, a number of embedded platforms use so called "none" or "embed" variants CPU specific compilers, then provide everything needed at the link stage to target specific devices.
That said, a number of embedded platforms use so called "none" or "embed" variants CPU specific compilers, then provide everything needed at the link stage to target specific devices.
November 10, 2025 at 6:42 PM
It's also possible I'm mixing up GCC'isms and LLVM, so my apologies if I maybe give false hope.
That said, a number of embedded platforms use so called "none" or "embed" variants CPU specific compilers, then provide everything needed at the link stage to target specific devices.
That said, a number of embedded platforms use so called "none" or "embed" variants CPU specific compilers, then provide everything needed at the link stage to target specific devices.
At some point i might snap and just write an SH4 LLVM backend 🥲
November 10, 2025 at 6:16 PM
At some point i might snap and just write an SH4 LLVM backend 🥲
It even includes support for JIT builds.
If you pass in --jit to a build, every-python will attempt to detect the correct version of LLVM needed to build from LLVM_VERSION specified in Tools/jit/_llvm.py at the time of the commit and use it!
Makes bisecting JIT things so much simpler 🥲
If you pass in --jit to a build, every-python will attempt to detect the correct version of LLVM needed to build from LLVM_VERSION specified in Tools/jit/_llvm.py at the time of the commit and use it!
Makes bisecting JIT things so much simpler 🥲
November 10, 2025 at 5:58 PM
It even includes support for JIT builds.
If you pass in --jit to a build, every-python will attempt to detect the correct version of LLVM needed to build from LLVM_VERSION specified in Tools/jit/_llvm.py at the time of the commit and use it!
Makes bisecting JIT things so much simpler 🥲
If you pass in --jit to a build, every-python will attempt to detect the correct version of LLVM needed to build from LLVM_VERSION specified in Tools/jit/_llvm.py at the time of the commit and use it!
Makes bisecting JIT things so much simpler 🥲
i dont understand why some languages would build their own VM unless its for a game engine. it is obvi not gonna be better than LLVM, or JVM. it is gonna be a complete dumpster fire of a time and more work..
November 10, 2025 at 12:24 AM
i dont understand why some languages would build their own VM unless its for a game engine. it is obvi not gonna be better than LLVM, or JVM. it is gonna be a complete dumpster fire of a time and more work..
New links: small scripts, build systems, LLVM instruction scheduling, Java exceptions compilation, recursive macros in C, UI framework, reading code
archiloque.net/weekly/25-11...
archiloque.net/weekly/25-11...
Links - 9th November 2025
Small scripts, build systems, LLVM instruction scheduling, Java exceptions compilation, recursive macros in C, UI framework, reading code
archiloque.net
November 9, 2025 at 8:58 PM
New links: small scripts, build systems, LLVM instruction scheduling, Java exceptions compilation, recursive macros in C, UI framework, reading code
archiloque.net/weekly/25-11...
archiloque.net/weekly/25-11...
Open-source contributions are a golden ticket. Working on projects like LLVM or GCC provides invaluable experience, showcasing practical skills and a commitment that stands out in a competitive field. #OpenSource 5/6
November 9, 2025 at 11:00 AM
Open-source contributions are a golden ticket. Working on projects like LLVM or GCC provides invaluable experience, showcasing practical skills and a commitment that stands out in a competitive field. #OpenSource 5/6
Gonna start looking at the TurboGrafx-16 now, using llvm-mos again if I can. I think that'll be the last port on the checklist I made myself. Or at least, it'll be the last port of the year.
November 9, 2025 at 4:47 AM
Gonna start looking at the TurboGrafx-16 now, using llvm-mos again if I can. I think that'll be the last port on the checklist I made myself. Or at least, it'll be the last port of the year.
they also pretend that other compilers are immune to bugs. no, they definitely aren't. llvm and the gcc break stuff regularly, just like rustc lol
November 8, 2025 at 9:56 PM
they also pretend that other compilers are immune to bugs. no, they definitely aren't. llvm and the gcc break stuff regularly, just like rustc lol
Doo: A Simple, Fast Programming Language Built on Rust and LLVM
C: https://news.ycombinator.com/item?id=45797044
posted on 2025.11.03 at 03:37:48 (c=3, p=10)
C: https://news.ycombinator.com/item?id=45797044
posted on 2025.11.03 at 03:37:48 (c=3, p=10)
November 8, 2025 at 2:33 PM
Doo: A Simple, Fast Programming Language Built on Rust and LLVM
C: https://news.ycombinator.com/item?id=45797044
posted on 2025.11.03 at 03:37:48 (c=3, p=10)
C: https://news.ycombinator.com/item?id=45797044
posted on 2025.11.03 at 03:37:48 (c=3, p=10)
WebAssembly ランタイムの wasmtime が Rust で書かれてて、そのバックエンドが cranelift と呼ばれているんだけど、rustc のバックエンドとして LLVM の代わりに cranelift を使うことができたはずで、コード生成が速くなったりしないか？
November 8, 2025 at 8:56 AM
WebAssembly ランタイムの wasmtime が Rust で書かれてて、そのバックエンドが cranelift と呼ばれているんだけど、rustc のバックエンドとして LLVM の代わりに cranelift を使うことができたはずで、コード生成が速くなったりしないか？
youtu.be/iN2XUUqw1YE
Why LLVM is slow; inst select, one pass, unrolling. gcc vs clang. PGO. Array init IR, try catch finally, throws, unwind tables, deopt, exceptional control flow. interior pointers. Stack unwinding vs branching checks
Why LLVM is slow; inst select, one pass, unrolling. gcc vs clang. PGO. Array init IR, try catch finally, throws, unwind tables, deopt, exceptional control flow. interior pointers. Stack unwinding vs branching checks
Coffee Compiler Club, 2025_11_07
YouTube video by Cliff Click
youtu.be
November 8, 2025 at 6:43 AM
youtu.be/iN2XUUqw1YE
Why LLVM is slow; inst select, one pass, unrolling. gcc vs clang. PGO. Array init IR, try catch finally, throws, unwind tables, deopt, exceptional control flow. interior pointers. Stack unwinding vs branching checks
Why LLVM is slow; inst select, one pass, unrolling. gcc vs clang. PGO. Array init IR, try catch finally, throws, unwind tables, deopt, exceptional control flow. interior pointers. Stack unwinding vs branching checks
man i love
i love fucking around with llvm dag patterns
i love fucking around with llvm dag patterns
November 8, 2025 at 6:31 AM
man i love
i love fucking around with llvm dag patterns
i love fucking around with llvm dag patterns
(as in the llvm tutorial language: https://llvm.org/docs/tutorial/ with the semi-famous rust compiler backend: https://cranelift.dev/)
and then maybe after that clean up the interpreter i made for my bachelor's thesis?? (i could start with that but i want something more......guided rn)
and then maybe after that clean up the interpreter i made for my bachelor's thesis?? (i could start with that but i want something more......guided rn)
LLVM Tutorial: Table of Contents — LLVM 22.0.0git documentation
llvm.org
November 8, 2025 at 2:44 AM
(as in the llvm tutorial language: https://llvm.org/docs/tutorial/ with the semi-famous rust compiler backend: https://cranelift.dev/)
and then maybe after that clean up the interpreter i made for my bachelor's thesis?? (i could start with that but i want something more......guided rn)
and then maybe after that clean up the interpreter i made for my bachelor's thesis?? (i could start with that but i want something more......guided rn)
yes!!! WebAssembly linker settings are now secure-by-default: https:// github.com/llvm/llvm-project/i ssues/151015
Interest | Match | Feed
Interest | Match | Feed
Origin
mastodon.social
November 8, 2025 at 12:47 AM
yes!!! WebAssembly linker settings are now secure-by-default: github.com/llvm/llvm-pr...
[lld][WebAssembly] `--stack-first` should be the default for wasm-ld · Issue #151015 · llvm/llvm-project
There are several arguments in favor of doing so: With stack protectors disabled by default, no dynamic stack overflow checks, a default stack size of 64K (the smallest of any non-bare-metal platfo...
github.com
November 8, 2025 at 12:47 AM
yes!!! WebAssembly linker settings are now secure-by-default: github.com/llvm/llvm-pr...