mikros/rust
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Use undef for partially-uninit constants up to 1024 bytes

Browse Source 
There needs to be some limit to avoid perf regressions on large arrays
with undef in each element (see comment in the code).
This commit is contained in:
Erik Desjardins 2022-02-18 15:57:10 -05:00
parent b8c56fa8c3
commit b7e5597491
5 changed files with 23 additions and 43 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
compiler/rustc_codegen_llvm/src/consts.rs
View File

@ -53,8 +53,10 @@ 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.
// Generating partially-uninit consts is limited to small allocations,
// 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 allow_partially_uninit =
match cx.sess().opts.debugging_opts.partially_uninit_const_threshold {
Some(max) => range.len() <= max,

4
compiler/rustc_session/src/options.rs
View File

@ -1341,9 +1341,9 @@ options! {
"panic strategy for panics in drops"),
parse_only: bool = (false, parse_bool, [UNTRACKED],
"parse only; do not compile, assemble, or link (default: no)"),
partially_uninit_const_threshold: Option<usize> = (None, parse_opt_number, [TRACKED],
partially_uninit_const_threshold: Option<usize> = (Some(1024), parse_opt_number, [TRACKED],
"allow generating const initializers with mixed init/uninit bytes, \
and set the maximum total size of a const allocation for which this is allowed (default: never)"),
and set the maximum total size of a const allocation for which this is allowed (default: 1024 bytes)"),
perf_stats: bool = (false, parse_bool, [UNTRACKED],
"print some performance-related statistics (default: no)"),
pick_stable_methods_before_any_unstable: bool = (true, parse_bool, [TRACKED],

4
src/test/codegen/consts.rs
View File

@ -43,7 +43,7 @@ pub fn inline_enum_const() -> E<i8, i16> {
#[no_mangle]
pub fn low_align_const() -> E<i16, [i16; 3]> {
// Check that low_align_const and high_align_const use the same constant
// CHECK: memcpy.p0i8.p0i8.i{{(32|64)}}(i8* align 2 %1, i8* align 2 getelementptr inbounds (<{ [8 x i8] }>, <{ [8 x i8] }>* [[LOW_HIGH]], i32 0, i32 0, i32 0), i{{(32|64)}} 8, i1 false)
// CHECK: memcpy.p0i8.p0i8.i{{(32|64)}}(i8* align 2 %1, i8* align 2 getelementptr inbounds (<{ [4 x i8], [4 x i8] }>, <{ [4 x i8], [4 x i8] }>* [[LOW_HIGH]], i32 0, i32 0, i32 0), i{{(32|64)}} 8, i1 false)
*&E::A(0)
}
@ -51,6 +51,6 @@ pub fn low_align_const() -> E<i16, [i16; 3]> {
#[no_mangle]
pub fn high_align_const() -> E<i16, i32> {
// Check that low_align_const and high_align_const use the same constant
// CHECK: memcpy.p0i8.p0i8.i{{(32|64)}}(i8* align 4 %1, i8* align 4 getelementptr inbounds (<{ [8 x i8] }>, <{ [8 x i8] }>* [[LOW_HIGH]], i32 0, i32 0, i32 0), i{{(32|64)}} 8, i1 false)
// CHECK: memcpy.p0i8.p0i8.i{{(32|64)}}(i8* align 4 %1, i8* align 4 getelementptr inbounds (<{ [4 x i8], [4 x i8] }>, <{ [4 x i8], [4 x i8] }>* [[LOW_HIGH]], i32 0, i32 0, i32 0), i{{(32|64)}} 8, i1 false)
*&E::A(0)
}

35
src/test/codegen/uninit-consts-allow-partially-uninit.rs
View File

@ -1,35 +0,0 @@
// compile-flags: -C no-prepopulate-passes -Z partially_uninit_const_threshold=1024
// Like uninit-consts.rs, but tests that we correctly generate partially-uninit consts
// when the (disabled by default) partially_uninit_const_threshold flag is used.
#![crate_type = "lib"]
use std::mem::MaybeUninit;
pub struct PartiallyUninit {
x: u32,
y: MaybeUninit<[u8; 10]>
}
// This should be partially undef.
// CHECK: [[PARTIALLY_UNINIT:@[0-9]+]] = private unnamed_addr constant <{ [4 x i8], [12 x i8] }> <{ [4 x i8] c"\EF\BE\AD\DE", [12 x i8] undef }>, align 4
// This shouldn't contain undef, since it's larger than the 1024 byte limit.
// CHECK: [[UNINIT_PADDING_HUGE:@[0-9]+]] = private unnamed_addr constant <{ [32768 x i8] }> <{ [32768 x i8] c"{{.+}}" }>, align 4
// CHECK-LABEL: @partially_uninit
#[no_mangle]
pub const fn partially_uninit() -> PartiallyUninit {
const X: PartiallyUninit = PartiallyUninit { x: 0xdeadbeef, y: MaybeUninit::uninit() };
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i{{(32|64)}}(i8* align 4 %1, i8* align 4 getelementptr inbounds (<{ [4 x i8], [12 x i8] }>, <{ [4 x i8], [12 x i8] }>* [[PARTIALLY_UNINIT]], i32 0, i32 0, i32 0), i{{(32|64)}} 16, i1 false)
X
}
// CHECK-LABEL: @uninit_padding_huge
#[no_mangle]
pub const fn uninit_padding_huge() -> [(u32, u8); 4096] {
const X: [(u32, u8); 4096] = [(123, 45); 4096];
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i{{(32|64)}}(i8* align 4 %1, i8* align 4 getelementptr inbounds (<{ [32768 x i8] }>, <{ [32768 x i8] }>* [[UNINIT_PADDING_HUGE]], i32 0, i32 0, i32 0), i{{(32|64)}} 32768, i1 false)
X
}

17
src/test/codegen/uninit-consts.rs
View File

@ -12,7 +12,12 @@ pub struct PartiallyUninit {
}
// CHECK: [[FULLY_UNINIT:@[0-9]+]] = private unnamed_addr constant <{ [10 x i8] }> undef
// CHECK: [[PARTIALLY_UNINIT:@[0-9]+]] = private unnamed_addr constant <{ [16 x i8] }> <{ [16 x i8] c"\EF\BE\AD\DE\00\00\00\00\00\00\00\00\00\00\00\00" }>, align 4
// CHECK: [[PARTIALLY_UNINIT:@[0-9]+]] = private unnamed_addr constant <{ [4 x i8], [12 x i8] }> <{ [4 x i8] c"\EF\BE\AD\DE", [12 x i8] undef }>, align 4
// This shouldn't contain undef, since it's larger than the 1024 byte limit.
// CHECK: [[UNINIT_PADDING_HUGE:@[0-9]+]] = private unnamed_addr constant <{ [32768 x i8] }> <{ [32768 x i8] c"{{.+}}" }>, align 4
// CHECK: [[FULLY_UNINIT_HUGE:@[0-9]+]] = private unnamed_addr constant <{ [16384 x i8] }> undef
// CHECK-LABEL: @fully_uninit
@ -27,7 +32,15 @@ pub const fn fully_uninit() -> MaybeUninit<[u8; 10]> {
#[no_mangle]
pub const fn partially_uninit() -> PartiallyUninit {
const X: PartiallyUninit = PartiallyUninit { x: 0xdeadbeef, y: MaybeUninit::uninit() };
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i{{(32|64)}}(i8* align 4 %1, i8* align 4 getelementptr inbounds (<{ [16 x i8] }>, <{ [16 x i8] }>* [[PARTIALLY_UNINIT]], i32 0, i32 0, i32 0), i{{(32|64)}} 16, i1 false)
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i{{(32|64)}}(i8* align 4 %1, i8* align 4 getelementptr inbounds (<{ [4 x i8], [12 x i8] }>, <{ [4 x i8], [12 x i8] }>* [[PARTIALLY_UNINIT]], i32 0, i32 0, i32 0), i{{(32|64)}} 16, i1 false)
X
}
// CHECK-LABEL: @uninit_padding_huge
#[no_mangle]
pub const fn uninit_padding_huge() -> [(u32, u8); 4096] {
const X: [(u32, u8); 4096] = [(123, 45); 4096];
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i{{(32|64)}}(i8* align 4 %1, i8* align 4 getelementptr inbounds (<{ [32768 x i8] }>, <{ [32768 x i8] }>* [[UNINIT_PADDING_HUGE]], i32 0, i32 0, i32 0), i{{(32|64)}} 32768, i1 false)
X
}