Introduce deduced parameter attributes, and use them for deducing readonly on

indirect immutable freeze by-value function parameters.

Right now, `rustc` only examines function signatures and the platform ABI when
determining the LLVM attributes to apply to parameters. This results in missed
optimizations, because there are some attributes that can be determined via
analysis of the MIR making up the function body. In particular, `readonly`
could be applied to most indirectly-passed by-value function arguments
(specifically, those that are freeze and are observed not to be mutated), but
it currently is not.

This patch introduces the machinery that allows `rustc` to determine those
attributes. It consists of a query, `deduced_param_attrs`, that, when
evaluated, analyzes the MIR of the function to determine supplementary
attributes. The results of this query for each function are written into the
crate metadata so that the deduced parameter attributes can be applied to
cross-crate functions. In this patch, we simply check the parameter for
mutations to determine whether the `readonly` attribute should be applied to
parameters that are indirect immutable freeze by-value.  More attributes could
conceivably be deduced in the future: `nocapture` and `noalias` come to mind.

Adding `readonly` to indirect function parameters where applicable enables some
potential optimizations in LLVM that are discussed in [issue 103103] and [PR
103070] around avoiding stack-to-stack memory copies that appear in functions
like `core::fmt::Write::write_fmt` and `core::panicking::assert_failed`. These
functions pass a large structure unchanged by value to a subfunction that also
doesn't mutate it. Since the structure in this case is passed as an indirect
parameter, it's a pointer from LLVM's perspective. As a result, the
intermediate copy of the structure that our codegen emits could be optimized
away by LLVM's MemCpyOptimizer if it knew that the pointer is `readonly
nocapture noalias` in both the caller and callee. We already pass `nocapture
noalias`, but we're missing `readonly`, as we can't determine whether a
by-value parameter is mutated by examining the signature in Rust. I didn't have
much success with having LLVM infer the `readonly` attribute, even with fat
LTO; it seems that deducing it at the MIR level is necessary.

No large benefits should be expected from this optimization *now*; LLVM needs
some changes (discussed in [PR 103070]) to more aggressively use the `noalias
nocapture readonly` combination in its alias analysis. I have some LLVM patches
for these optimizations and have had them looked over. With all the patches
applied locally, I enabled LLVM to remove all the `memcpy`s from the following
code:

```rust
fn main() {
    println!("Hello {}", 3);
}
```

which is a significant codegen improvement over the status quo. I expect that
if this optimization kicks in in multiple places even for such a simple
program, then it will apply to Rust code all over the place.

[issue 103103]: https://github.com/rust-lang/rust/issues/103103

[PR 103070]: https://github.com/rust-lang/rust/pull/103070
This commit is contained in:
Patrick Walton 2022-10-17 19:42:15 -07:00
parent b1ab3b738a
commit da630ac79d
14 changed files with 393 additions and 9 deletions

View file

@ -4,6 +4,7 @@ use rustc_middle::ty::layout::{
fn_can_unwind, FnAbiError, HasParamEnv, HasTyCtxt, LayoutCx, LayoutOf, TyAndLayout,
};
use rustc_middle::ty::{self, Ty, TyCtxt};
use rustc_session::config::OptLevel;
use rustc_span::def_id::DefId;
use rustc_target::abi::call::{
ArgAbi, ArgAttribute, ArgAttributes, ArgExtension, Conv, FnAbi, PassMode, Reg, RegKind,
@ -384,7 +385,7 @@ fn fn_abi_new_uncached<'tcx>(
conv,
can_unwind: fn_can_unwind(cx.tcx(), fn_def_id, sig.abi),
};
fn_abi_adjust_for_abi(cx, &mut fn_abi, sig.abi)?;
fn_abi_adjust_for_abi(cx, &mut fn_abi, sig.abi, fn_def_id)?;
debug!("fn_abi_new_uncached = {:?}", fn_abi);
Ok(cx.tcx.arena.alloc(fn_abi))
}
@ -394,6 +395,7 @@ fn fn_abi_adjust_for_abi<'tcx>(
cx: &LayoutCx<'tcx, TyCtxt<'tcx>>,
fn_abi: &mut FnAbi<'tcx, Ty<'tcx>>,
abi: SpecAbi,
fn_def_id: Option<DefId>,
) -> Result<(), FnAbiError<'tcx>> {
if abi == SpecAbi::Unadjusted {
return Ok(());
@ -404,7 +406,18 @@ fn fn_abi_adjust_for_abi<'tcx>(
|| abi == SpecAbi::RustIntrinsic
|| abi == SpecAbi::PlatformIntrinsic
{
let fixup = |arg: &mut ArgAbi<'tcx, Ty<'tcx>>| {
// Look up the deduced parameter attributes for this function, if we have its def ID and
// we're optimizing in non-incremental mode. We'll tag its parameters with those attributes
// as appropriate.
let deduced_param_attrs = if cx.tcx.sess.opts.optimize != OptLevel::No
&& cx.tcx.sess.opts.incremental.is_none()
{
fn_def_id.map(|fn_def_id| cx.tcx.deduced_param_attrs(fn_def_id)).unwrap_or_default()
} else {
&[]
};
let fixup = |arg: &mut ArgAbi<'tcx, Ty<'tcx>>, arg_idx: Option<usize>| {
if arg.is_ignore() {
return;
}
@ -451,10 +464,30 @@ fn fn_abi_adjust_for_abi<'tcx>(
// so we pick an appropriately sized integer type instead.
arg.cast_to(Reg { kind: RegKind::Integer, size });
}
// If we deduced that this parameter was read-only, add that to the attribute list now.
//
// The `readonly` parameter only applies to pointers, so we can only do this if the
// argument was passed indirectly. (If the argument is passed directly, it's an SSA
// value, so it's implicitly immutable.)
if let (Some(arg_idx), &mut PassMode::Indirect { ref mut attrs, .. }) =
(arg_idx, &mut arg.mode)
{
// The `deduced_param_attrs` list could be empty if this is a type of function
// we can't deduce any parameters for, so make sure the argument index is in
// bounds.
if let Some(deduced_param_attrs) = deduced_param_attrs.get(arg_idx) {
if deduced_param_attrs.read_only {
attrs.regular.insert(ArgAttribute::ReadOnly);
debug!("added deduced read-only attribute");
}
}
}
};
fixup(&mut fn_abi.ret);
for arg in fn_abi.args.iter_mut() {
fixup(arg);
fixup(&mut fn_abi.ret, None);
for (arg_idx, arg) in fn_abi.args.iter_mut().enumerate() {
fixup(arg, Some(arg_idx));
}
} else {
fn_abi.adjust_for_foreign_abi(cx, abi)?;