-
Notifications
You must be signed in to change notification settings - Fork 12.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
2.pow(n) does not optimize well #47234
Comments
LLVM doesn’t optimise even a
so I don’t see it being able to optimise the more complicated algorithm used currently. |
Sooo... that means that there should be a MIR optimisation that checks whether the function is Or is this rather solved on the llvm side by having a pow intrinsic that works on integers? |
A MIR pass would be a pretty big hammer. We'd have to make this random library function specially recognized by the compiler. Also the implementation effort wouldn't be justified for a one-off optimization. I think the closest thing to a practical solution would be changing the library code that implements |
As of 1.53.0 and today's nightly the bug still occurs: https://godbolt.org/z/Kjn4odTdh |
As of 1.74.0, Rust still isn't able to optimize If the leading zeroes of the base are removed before exponentiation with a right shift and then added back with a left shift after exponentiation, then the Here is an example. This wasn't thoroughly checked for logic errors: Compiler Explorer link |
#114390 is the fix for this issue, but it doesn't have recent activity. |
We can fix that now that we have |
…li-obk Replacement of rust-lang#114390: Add new intrinsic `is_var_statically_known` and optimize pow for powers of two This adds a new intrinsic `is_val_statically_known` that lowers to [``@llvm.is.constant.*`](https://llvm.org/docs/LangRef.html#llvm-is-constant-intrinsic).` It also applies the intrinsic in the int_pow methods to recognize and optimize the idiom `2isize.pow(x)`. See rust-lang#114390 for more discussion. While I have extended the scope of the power of two optimization from rust-lang#114390, I haven't added any new uses for the intrinsic. That can be done in later pull requests. Note: When testing or using the library, be sure to use `--stage 1` or higher. Otherwise, the intrinsic will be a noop and the doctests will be skipped. If you are trying out edits, you may be interested in [`--keep-stage 0`](https://rustc-dev-guide.rust-lang.org/building/suggested.html#faster-builds-with---keep-stage). Fixes rust-lang#47234 Resolves rust-lang#114390 `@Centri3`
…li-obk Replacement of rust-lang#114390: Add new intrinsic `is_var_statically_known` and optimize pow for powers of two This adds a new intrinsic `is_val_statically_known` that lowers to [``@llvm.is.constant.*`](https://llvm.org/docs/LangRef.html#llvm-is-constant-intrinsic).` It also applies the intrinsic in the int_pow methods to recognize and optimize the idiom `2isize.pow(x)`. See rust-lang#114390 for more discussion. While I have extended the scope of the power of two optimization from rust-lang#114390, I haven't added any new uses for the intrinsic. That can be done in later pull requests. Note: When testing or using the library, be sure to use `--stage 1` or higher. Otherwise, the intrinsic will be a noop and the doctests will be skipped. If you are trying out edits, you may be interested in [`--keep-stage 0`](https://rustc-dev-guide.rust-lang.org/building/suggested.html#faster-builds-with---keep-stage). Fixes rust-lang#47234 Resolves rust-lang#114390 `@Centri3`
…li-obk Replacement of rust-lang#114390: Add new intrinsic `is_var_statically_known` and optimize pow for powers of two This adds a new intrinsic `is_val_statically_known` that lowers to [``@llvm.is.constant.*`](https://llvm.org/docs/LangRef.html#llvm-is-constant-intrinsic).` It also applies the intrinsic in the int_pow methods to recognize and optimize the idiom `2isize.pow(x)`. See rust-lang#114390 for more discussion. While I have extended the scope of the power of two optimization from rust-lang#114390, I haven't added any new uses for the intrinsic. That can be done in later pull requests. Note: When testing or using the library, be sure to use `--stage 1` or higher. Otherwise, the intrinsic will be a noop and the doctests will be skipped. If you are trying out edits, you may be interested in [`--keep-stage 0`](https://rustc-dev-guide.rust-lang.org/building/suggested.html#faster-builds-with---keep-stage). Fixes rust-lang#47234 Resolves rust-lang#114390 `@Centri3`
…li-obk Replacement of rust-lang#114390: Add new intrinsic `is_var_statically_known` and optimize pow for powers of two This adds a new intrinsic `is_val_statically_known` that lowers to [``@llvm.is.constant.*`](https://llvm.org/docs/LangRef.html#llvm-is-constant-intrinsic).` It also applies the intrinsic in the int_pow methods to recognize and optimize the idiom `2isize.pow(x)`. See rust-lang#114390 for more discussion. While I have extended the scope of the power of two optimization from rust-lang#114390, I haven't added any new uses for the intrinsic. That can be done in later pull requests. Note: When testing or using the library, be sure to use `--stage 1` or higher. Otherwise, the intrinsic will be a noop and the doctests will be skipped. If you are trying out edits, you may be interested in [`--keep-stage 0`](https://rustc-dev-guide.rust-lang.org/building/suggested.html#faster-builds-with---keep-stage). Fixes rust-lang#47234 Resolves rust-lang#114390 `@Centri3`
A mibbit user noted in
#rust-beginners
that the integerpow
method with a base of two generates far worse code than a shift: https://godbolt.org/g/mE1wjH (the shift is not equivalent for n >= 64, but you can see that it would still be better code if you accounted for that)Perhaps we can special case the base 2 and use a shift in that case? One complication is that
pow
is subject to overflow checks: Whether it panics or returns 0 for excessive exponents depends on whether overflow checks are enabled, and I don't know of a way to branch depending on whether they are enabled (which is required for this optimization to not change behavior).Alternatively, better LLVM optimizations could help, but this might be a bit too much to ask.
The text was updated successfully, but these errors were encountered: