- How do you verify that a floating-point algorithm is correct?- You double-check it.
-
- How do you verify that a floating-point algorithm is correct?
- You double-check it. -
- How do you verify that a floating-point algorithm is correct?
- You double-check it.Which is what I'm going to do with the yy's optimization.
-
- How do you verify that a floating-point algorithm is correct?
- You double-check it.@vitaut "Have you verified the IEEE 754 binary32 correctness?"
"I double checked it"
"Dude, you had a single job…" -
@vitaut "Have you verified the IEEE 754 binary32 correctness?"
"I double checked it"
"Dude, you had a single job…"@vitaut (by the way, what bignum doesn't want people to know: for reasonably fast algorithms, it's totally feasible to test every single 32 bit float on a desktop CPU. Don't even need to be fancy with multithreading. Don't even be clever about not storing results in RAM. It's only 16 GB. Your laptop will be fine. If float32 zmij is as fast as float64 zmij, then it'll take 51 s to test all floats.)
-
Which is what I'm going to do with the yy's optimization.
With today’s hardware you could even bruteforce it LOL
-
undefined oblomov@sociale.network shared this topic
-
@vitaut (by the way, what bignum doesn't want people to know: for reasonably fast algorithms, it's totally feasible to test every single 32 bit float on a desktop CPU. Don't even need to be fancy with multithreading. Don't even be clever about not storing results in RAM. It's only 16 GB. Your laptop will be fine. If float32 zmij is as fast as float64 zmij, then it'll take 51 s to test all floats.)