languages/rust
languages/rust
1
0
Fork 0
mirror of https://github.com/rust-lang/rust synced 2024-07-20 05:56:41 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Auto merge of #42771 - arielb1:no-inline-unwind, r=nagisa

Browse source 
mark calls in the unwind path as !noinline

The unwind path is always cold, so that should not have bad performance
implications.  This avoids catastrophic exponential inlining, and also
decreases the size of librustc.so by 1.5% (OTOH, the size of `libstd.so`
increased by 0.5% for some reason).

Fixes #41696.

r? @nagisa
This commit is contained in:
bors 2017-06-21 21:29:45 +00:00
parent 622e7e6487 0b93798959
commit 6de26f42de
2 changed files with 7 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
src/librustc_trans/mir/block.rs
View file

@ -146,6 +146,13 @@ fn trans_terminator(&mut self,
} else {
let llret = bcx.call(fn_ptr, &llargs, cleanup_bundle);
fn_ty.apply_attrs_callsite(llret);
if this.mir[bb].is_cleanup {
// Cleanup is always the cold path. Don't inline
// drop glue. Also, when there is a deeply-nested
// struct, there are "symmetry" issues that cause
// exponential inlining - see issue #41696.
llvm::Attribute::NoInline.apply_callsite(llvm::AttributePlace::Function, llret);
}
if let Some((ret_dest, ret_ty, target)) = destination {
let op = OperandRef {

2
src/test/run-pass/issue-41696.rs
View file

@ -9,8 +9,6 @@
// except according to those terms.
// this used to cause exponential code-size blowup during LLVM passes.
// ignore-test FIXME #41696
// min-llvm-version 3.9
#![feature(test)]