Title: *Common Lisp go brrt: Achieving Near C/C++ Performance with Lupus* * *0:26**:* *Goal:* The speaker, Marco Heisig, aims to demonstrate the speed improvements possible in Common Lisp for numerical computation, specifically focusing on a Jacobi iteration example. * *1:44**:* *Initial Performance:* A standard Common Lisp implementation of a Jacobi iteration using loops achieves about 840 megaflops, comparable to JavaScript. * *2:57**:* *Optimization with "Lupus":* By replacing standard loops with a custom "Lupus" implementation, the performance jumps to 16 gigaflops, a ~20x speedup. * *3:39**:* *Key Takeaway:* The Lupus approach achieves near-C/C++ performance without resorting to complex code changes or sacrificing code clarity. It leverages AVX2 instructions for vectorized operations. * *4:37**:* *"Common Lisp go brrt":* The title and final demonstration playfully highlight the significant speed boost achieved, showcasing Common Lisp's potential for high-performance computing. * *0:00**:* *Context:* This was a lightning talk at the European Lisp Symposium 2022, demonstrating the Lupus project. I used gemini-1.5-pro-exp-0801 to summarize the transcript. Cost (if I didn't use the free tier): $0.02 Input tokens: 4650 Output tokens: 474
This is cool, however kind of disappointed that it seems to be based upon utilizing AVX, and which is not exactly common. This should really have been titled something like AVX accelerated Common Lisp if it wanted to be a bit more honest. Lot of us with just slightly older Intel, AMD, and Macs that this will not work.
Awesome.
Title: *Common Lisp go brrt: Achieving Near C/C++ Performance with Lupus*
* *0:26**:* *Goal:* The speaker, Marco Heisig, aims to demonstrate the speed improvements possible in Common Lisp for numerical computation, specifically focusing on a Jacobi iteration example.
* *1:44**:* *Initial Performance:* A standard Common Lisp implementation of a Jacobi iteration using loops achieves about 840 megaflops, comparable to JavaScript.
* *2:57**:* *Optimization with "Lupus":* By replacing standard loops with a custom "Lupus" implementation, the performance jumps to 16 gigaflops, a ~20x speedup.
* *3:39**:* *Key Takeaway:* The Lupus approach achieves near-C/C++ performance without resorting to complex code changes or sacrificing code clarity. It leverages AVX2 instructions for vectorized operations.
* *4:37**:* *"Common Lisp go brrt":* The title and final demonstration playfully highlight the significant speed boost achieved, showcasing Common Lisp's potential for high-performance computing.
* *0:00**:* *Context:* This was a lightning talk at the European Lisp Symposium 2022, demonstrating the Lupus project.
I used gemini-1.5-pro-exp-0801 to summarize the transcript.
Cost (if I didn't use the free tier): $0.02
Input tokens: 4650
Output tokens: 474
This is cool, however kind of disappointed that it seems to be based upon utilizing AVX, and which is not exactly common.
This should really have been titled something like AVX accelerated Common Lisp if it wanted to be a bit more honest.
Lot of us with just slightly older Intel, AMD, and Macs that this will not work.