bjoernager/rust
bjoernager/rust
1
Fork 0
Code Issues Pull requests Activity

Auto merge of #94130 - erikdesjardins:partially, r=nikic

Browse source 
Use undef for (some) partially-uninit constants

There needs to be some limit to avoid perf regressions on large arrays
with undef in each element (see comment in the code).

Fixes: #84565
Original PR: #83698

Depends on LLVM 14: #93577
This commit is contained in:
bors 2022-02-25 05:44:33 +00:00
commit ece55d416e
7 changed files with 44 additions and 66 deletions
Download patch file Download diff file Expand all files Collapse all files

41
compiler/rustc_codegen_llvm/src/consts.rs
View file

@ -2,6 +2,7 @@ use crate::base;
use crate::common::CodegenCx;
use crate::debuginfo;
use crate::llvm::{self, True};
use crate::llvm_util;
use crate::type_::Type;
use crate::type_of::LayoutLlvmExt;
use crate::value::Value;
@ -37,7 +38,7 @@ pub fn const_alloc_to_llvm<'ll>(cx: &CodegenCx<'ll, '_>, alloc: &Allocation) ->
alloc: &'a Allocation,
range: Range<usize>,
) {
let mut chunks = alloc
let chunks = alloc
.init_mask()
.range_as_init_chunks(Size::from_bytes(range.start), Size::from_bytes(range.end));
@ -53,30 +54,26 @@ pub fn const_alloc_to_llvm<'ll>(cx: &CodegenCx<'ll, '_>, alloc: &Allocation) ->
}
};
// Generating partially-uninit consts inhibits optimizations, so it is disabled by default.
// See https://github.com/rust-lang/rust/issues/84565.
let allow_partially_uninit =
match cx.sess().opts.debugging_opts.partially_uninit_const_threshold {
Some(max) => range.len() <= max,
None => false,
};
// Generating partially-uninit consts is limited to small numbers of chunks,
// to avoid the cost of generating large complex const expressions.
// For example, `[(u32, u8); 1024 * 1024]` contains uninit padding in each element,
// and would result in `{ [5 x i8] zeroinitializer, [3 x i8] undef, ...repeat 1M times... }`.
let max = if llvm_util::get_version() < (14, 0, 0) {
// Generating partially-uninit consts inhibits optimizations in LLVM < 14.
// See https://github.com/rust-lang/rust/issues/84565.
1
} else {
cx.sess().opts.debugging_opts.uninit_const_chunk_threshold
};
let allow_uninit_chunks = chunks.clone().take(max.saturating_add(1)).count() <= max;
if allow_partially_uninit {
if allow_uninit_chunks {
llvals.extend(chunks.map(chunk_to_llval));
} else {
let llval = match (chunks.next(), chunks.next()) {
(Some(chunk), None) => {
// exactly one chunk, either fully init or fully uninit
chunk_to_llval(chunk)
}
_ => {
// partially uninit, codegen as if it was initialized
// (using some arbitrary value for uninit bytes)
let bytes = alloc.inspect_with_uninit_and_ptr_outside_interpreter(range);
cx.const_bytes(bytes)
}
};
llvals.push(llval);
// If this allocation contains any uninit bytes, codegen as if it was initialized
// (using some arbitrary value for uninit bytes).
let bytes = alloc.inspect_with_uninit_and_ptr_outside_interpreter(range);
llvals.push(cx.const_bytes(bytes));
}
}