rustc: Improve compile time of platform intrinsics #32236
Conversation
|
r? @arielb1 (rust_highfive has picked a reviewer for you, use r? to override) |
|
cc @huonw |
alexcrichton
force-pushed
the
better-compile-intrinsics
branch
from
6e2956b to
0c8b1ba
Compare
| Aggregate(bool, Vec<Type>), | ||
| Pointer(&'static Type, Option<&'static Type>, /* const */ bool), | ||
| Vector(&'static Type, Option<&'static Type>, u8), | ||
| Aggregate(bool, &'static &'static [&'static Type]), |
There was a problem hiding this comment.
Is this a &&[] so that it is just pointer-sized, not slice sized?
There was a problem hiding this comment.
That does seem a little odd, actually. Since it's not going to change the overall size of the type, the alignment means that Pointer and Vector are 4 words big anyway.
There was a problem hiding this comment.
Ah this was actually because the generated static values had the type &'static [T] and that couldn't be used by value (only by reference). I realize now though that all of the statics can have type [T; N] and then we can use references and coercions to get the right type, so no there's no longer double indirection.
alexcrichton
force-pushed
the
better-compile-intrinsics
branch
from
0c8b1ba to
d92b717
Compare
|
cc me |
|
Why not store everything in rodata and just do a binary search (instead of a |
|
That's it, generate code of the form pub static INTRINSICS: &'static [Intrinsic] = &[
Intrinsic {
name: "movemask_ps",
inputs: &[&::F32x4],
output: &::I32,
definition: Named("llvm.x86.sse.movmsk.ps")
},
Intrinsic {
name: "storeu_ps",
inputs: &[&Type::Pointer(&::F32, Some(&::I8), false), &::F32x4],
output: &::VOID,
definition: Named("llvm.x86.sse.storeu.ps")
},
]; |
|
@arielb1 yeah that would probably work, I don't think it would improve compile times that much, more, however. |
|
I think it would be cleaner |
|
@alexcrichton last time (quite a while ago) I was dealing with matches of |
|
Yeah that would be nice to implement. OTOH, I think that being data-driven is the right thing to do here. |
|
We could also use something phf-like for static tables. I’m pretty sure it wouldn’t be too hard to do something smart about encoding tries into rodata either. Regardless, this PR stands very well on its own and we could see if there’s anything more to gain here independently. |
|
I'll go with what we have. @bors r+ |
|
📌 Commit d92b717 has been approved by |
|
Yeah I was actually thinking this'd be a neat case for phf, which gave me more inspiration to keep pushing on rustbuild :) |
|
⌛ Testing commit d92b717 with merge 4f8ab7d... |
rustc: Improve compile time of platform intrinsics This commit improves the compile time of `rustc_platform_intrinsics` from 23s to 3.6s if compiling with `-O` and from 77s to 17s if compiling with `-O -g`. The compiled rlib size also drops from 3.1M to 1.2M. The wins here were gained by removing the destructors associated with `Type` by removing the internal `Box` and `Vec` indirections. These destructors meant that a lot of landing pads and extra code were generated to manage the runtime representations. Instead everything can basically be statically computed and shoved into rodata, so all we need is a giant string compare to lookup what's what. Closes #28273
|
💔 Test failed - auto-linux-64-opt |
This commit improves the compile time of `rustc_platform_intrinsics` from 23s to 3.6s if compiling with `-O` and from 77s to 17s if compiling with `-O -g`. The compiled rlib size also drops from 3.1M to 1.2M. The wins here were gained by removing the destructors associated with `Type` by removing the internal `Box` and `Vec` indirections. These destructors meant that a lot of landing pads and extra code were generated to manage the runtime representations. Instead everything can basically be statically computed and shoved into rodata, so all we need is a giant string compare to lookup what's what. Closes rust-lang#28273
alexcrichton
force-pushed
the
better-compile-intrinsics
branch
from
d92b717 to
87ede2d
Compare
rustc: Improve compile time of platform intrinsics This commit improves the compile time of `rustc_platform_intrinsics` from 23s to 3.6s if compiling with `-O` and from 77s to 17s if compiling with `-O -g`. The compiled rlib size also drops from 3.1M to 1.2M. The wins here were gained by removing the destructors associated with `Type` by removing the internal `Box` and `Vec` indirections. These destructors meant that a lot of landing pads and extra code were generated to manage the runtime representations. Instead everything can basically be statically computed and shoved into rodata, so all we need is a giant string compare to lookup what's what. Closes #28273
This commit improves the compile time of
rustc_platform_intrinsicsfrom 23s to3.6s if compiling with
-Oand from 77s to 17s if compiling with-O -g. Thecompiled rlib size also drops from 3.1M to 1.2M.
The wins here were gained by removing the destructors associated with
Typebyremoving the internal
BoxandVecindirections. These destructors meant thata lot of landing pads and extra code were generated to manage the runtime
representations. Instead everything can basically be statically computed and
shoved into rodata, so all we need is a giant string compare to lookup what's
what.
Closes #28273