saagarjha

9 ms·

211.
by saagarjha 5mo ago
Because Fil-C is not a serious project
212.
by saagarjha 5mo ago
They’re using Claude lmao
213.
by saagarjha 5mo ago
Nobody is suing the maintainer for support here so this is completely irrelevant.
214.
by saagarjha 5mo ago
Because the idea that something this obvious occurred to the CTO first is very, very unlikely. What is more probable was that leadership ignored people who disagreed until the CTO convinced himself it was a good idea and went ahead with it.
215.
by saagarjha 5mo ago
Why didn’t it occur to someone that this would be fun to do within the CTO having to realize this
216.
by saagarjha 5mo ago
I hope you don’t take this the wrong way but this entire blog post seems a little like what I see happens when someone asks an LLM how to optimize their code. There are like half a dozen, or more, different things that were tried: AI is rea
217.
by saagarjha 5mo ago
Doing this in general is really hard. Unfortunately the blog post doesn't really go into detail of how hard, though it does mention some cases. For example, if you run your agent in a VM with network access, it can come across somethin
218.
by saagarjha 5mo ago
I mean it is barely getting any updates and the website doesn’t work
219.
by saagarjha 5mo ago
Not quite related, but I figure the audience might have some overlap: what is going on with Keystone?
220.
by saagarjha 5mo ago
Atherton does not have a Caltrain station anymore.
221.
by saagarjha 5mo ago
I do not wish my Amazon delivery driver to show up in my living room.
222.
by saagarjha 5mo ago
It’s real, you can measure it yourself on modern Nvidia hardware
223.
by saagarjha 5mo ago
Please point to the things being described in that comment and explain what the average person in 1930 would have felt about them.
224.
by saagarjha 5mo ago
Ok but you do realize that he was alive in the 1900s, not 1000 BC.
225.
by saagarjha 5mo ago
Comparing against torch.compile is not particularly impressive
226.
by saagarjha 5mo ago
Guys who have only written CUTLASS GEMM epilogue fusions, seeing their second kernel: Getting a lot of "GEMM epilogue fusion" vibes from this
227.
by saagarjha 5mo ago
The lack of fast GPU kernels written by AI does not lend credence to your theory.
228.
by saagarjha 5mo ago
Source for "compilers basically always did volatile since everything was always spilled".
229.
by saagarjha 5mo ago
Compilers do not surprise all that often which is why it is extra surprising when it does happen
230.
by saagarjha 5mo ago
Yes, but usually you don't want this. You think you do, but you don't: you can't always eliminate these, and often eliminating the extra accesses is not the most efficient thing to do either. Sometimes it's faster to hav
231.
by saagarjha 5mo ago
Compilers elide loads all the time this is one of the more basic optimizations a compiler can do. We just mostly think those are "good" optimizations.
232.
by saagarjha 5mo ago
Yeah most modern languages settle on having fewer types that are used most of the time and don't require as many casts.
233.
by saagarjha 5mo ago
Ooh, fun! Good luck!
234.
by saagarjha 5mo ago
If you are using SIMD instructions on x86 you probably want the unaligned ones
235.
by saagarjha 5mo ago
Yeah on average. On some paths it's almost free
236.
by saagarjha 5mo ago
This would be a regression
237.
by saagarjha 5mo ago
I don’t deny that LLMs can detect some attacks. I just don’t think they can be made to do so reliably.
238.
by saagarjha 5mo ago
I made the weaker claim that x can be removed. This is something I could prove with compiler output but I would have to find a compiler willing to make this optimization which is not something I can guarantee.
239.
by saagarjha 5mo ago
And your reason for believing this is…
240.
by saagarjha 5mo ago
I meant reading the uninitialized variable

More ›