Shrink Crystal::System.print_error's output size
#15490
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Terrific 👍 I've been reusing the to_int_slice calls in Crystal.trace (https://github.com/crystal-lang/crystal/blob/master/src/crystal/tracing.cr) so we'll have to fix the calls there too. |
straight-shoota
approved these changes
Feb 21, 2025
kojix2
pushed a commit
to kojix2/crystal
that referenced
this pull request
Feb 23, 2025
`Crystal::System` is by far the single largest LLVM module when compiling a blank source file, even though all the module does by itself is defining `.print_error` and friends. On my machine, with debug information stripped, `C-rystal5858S-ystem.o0.bc` is 213.4 KiB big, compared to `_main.o0.bc`'s 101.7 KiB. Disassembling the bytecode back to LLVM IR produces a monstrosity with 33k lines. This PR brings the numbers down to 48.0 KiB and 6.1k lines, while slightly improving performance, using the following tricks: * The type of `.as?(T)` is always `T?` and does not perform intersection, so even simple types like `Int32` are upcast into the whole `Int::Primitive?`, leading to a lot of redundant downcasts later. A simple `is_a?` will suffice as a type filter in `read_arg`. (I believe this is mentioned somewhere but couldn't find it) * In `.to_int_slice`, the `num` variable is cast into an `Int32 | UInt32 | Int64 | UInt64`, and each subsequent line dispatches over that union. The fix here is to split the rest of the body into a separate method, and call it with each variant of the union. This form of dispatching is akin to rewriting `.to_int_slice` as an instance method on the integers. * `.to_int_slice` is now non-yielding, as the inlining added too much bloat. The caller is responsible for preparing a suitably sized buffer. Additionally, this reduces the time for the bytecode generation phase from an average of 0.35s down to 0.26s.
1 task
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Crystal::Systemis by far the single largest LLVM module when compiling a blank source file, even though all the module does by itself is defining.print_errorand friends. On my machine, with debug information stripped,C-rystal5858S-ystem.o0.bcis 213.4 KiB big, compared to_main.o0.bc's 101.7 KiB. Disassembling the bytecode back to LLVM IR produces a monstrosity with 33k lines. This PR brings the numbers down to 48.0 KiB and 6.1k lines, while slightly improving performance, using the following tricks:.as?(T)is alwaysT?and does not perform intersection, so even simple types likeInt32are upcast into the wholeInt::Primitive?, leading to a lot of redundant downcasts later. A simpleis_a?will suffice as a type filter inread_arg. (I believe this is mentioned somewhere but couldn't find it).to_int_slice, thenumvariable is cast into anInt32 | UInt32 | Int64 | UInt64, and each subsequent line dispatches over that union. The fix here is to split the rest of the body into a separate method, and call it with each variant of the union. This form of dispatching is akin to rewriting.to_int_sliceas an instance method on the integers..to_int_sliceis now non-yielding, as the inlining added too much bloat. The caller is responsible for preparing a suitably sized buffer.Additionally, this reduces the time for the bytecode generation phase from an average of 0.35s down to 0.26s.