Skip to content

Commit

Permalink
Auto merge of #125892 - workingjubilee:rollup-gytt1q7, r=workingjubilee
Browse files Browse the repository at this point in the history
Rollup of 3 pull requests

Successful merges:

 - #125311 (Make repr(packed) vectors work with SIMD intrinsics)
 - #125849 (Migrate `run-make/emit-named-files` to `rmake.rs`)
 - #125851 (Add some more specific checks to the MIR validator)

r? `@ghost`
`@rustbot` modify labels: rollup
  • Loading branch information
bors committed Jun 2, 2024
2 parents 8bec878 + 30dc2ba commit 5e6c2b6
Show file tree
Hide file tree
Showing 7 changed files with 222 additions and 57 deletions.
54 changes: 53 additions & 1 deletion compiler/rustc_codegen_llvm/src/intrinsic.rs
Original file line number Diff line number Diff line change
Expand Up @@ -482,8 +482,60 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> {
}

_ if name.as_str().starts_with("simd_") => {
// Unpack non-power-of-2 #[repr(packed, simd)] arguments.
// This gives them the expected layout of a regular #[repr(simd)] vector.
let mut loaded_args = Vec::new();
for (ty, arg) in arg_tys.iter().zip(args) {
loaded_args.push(
// #[repr(packed, simd)] vectors are passed like arrays (as references,
// with reduced alignment and no padding) rather than as immediates.
// We can use a vector load to fix the layout and turn the argument
// into an immediate.
if ty.is_simd()
&& let OperandValue::Ref(place) = arg.val
{
let (size, elem_ty) = ty.simd_size_and_type(self.tcx());
let elem_ll_ty = match elem_ty.kind() {
ty::Float(f) => self.type_float_from_ty(*f),
ty::Int(i) => self.type_int_from_ty(*i),
ty::Uint(u) => self.type_uint_from_ty(*u),
ty::RawPtr(_, _) => self.type_ptr(),
_ => unreachable!(),
};
let loaded =
self.load_from_place(self.type_vector(elem_ll_ty, size), place);
OperandRef::from_immediate_or_packed_pair(self, loaded, arg.layout)
} else {
*arg
},
);
}

let llret_ty = if ret_ty.is_simd()
&& let abi::Abi::Aggregate { .. } = self.layout_of(ret_ty).layout.abi
{
let (size, elem_ty) = ret_ty.simd_size_and_type(self.tcx());
let elem_ll_ty = match elem_ty.kind() {
ty::Float(f) => self.type_float_from_ty(*f),
ty::Int(i) => self.type_int_from_ty(*i),
ty::Uint(u) => self.type_uint_from_ty(*u),
ty::RawPtr(_, _) => self.type_ptr(),
_ => unreachable!(),
};
self.type_vector(elem_ll_ty, size)
} else {
llret_ty
};

match generic_simd_intrinsic(
self, name, callee_ty, fn_args, args, ret_ty, llret_ty, span,
self,
name,
callee_ty,
fn_args,
&loaded_args,
ret_ty,
llret_ty,
span,
) {
Ok(llval) => llval,
Err(()) => return Ok(()),
Expand Down
101 changes: 94 additions & 7 deletions compiler/rustc_mir_transform/src/validate.rs
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,7 @@ use rustc_middle::mir::coverage::CoverageKind;
use rustc_middle::mir::interpret::Scalar;
use rustc_middle::mir::visit::{NonUseContext, PlaceContext, Visitor};
use rustc_middle::mir::*;
use rustc_middle::ty::adjustment::PointerCoercion;
use rustc_middle::ty::{
self, CoroutineArgsExt, InstanceDef, ParamEnv, Ty, TyCtxt, TypeVisitableExt, Variance,
};
Expand Down Expand Up @@ -1134,9 +1135,76 @@ impl<'a, 'tcx> Visitor<'tcx> for TypeChecker<'a, 'tcx> {
// FIXME(dyn-star): make sure nothing needs to be done here.
}
// FIXME: Add Checks for these
CastKind::PointerWithExposedProvenance
| CastKind::PointerExposeProvenance
| CastKind::PointerCoercion(_) => {}
CastKind::PointerWithExposedProvenance | CastKind::PointerExposeProvenance => {}
CastKind::PointerCoercion(PointerCoercion::ReifyFnPointer) => {
// FIXME: check signature compatibility.
check_kinds!(
op_ty,
"CastKind::{kind:?} input must be a fn item, not {:?}",
ty::FnDef(..)
);
check_kinds!(
target_type,
"CastKind::{kind:?} output must be a fn pointer, not {:?}",
ty::FnPtr(..)
);
}
CastKind::PointerCoercion(PointerCoercion::UnsafeFnPointer) => {
// FIXME: check safety and signature compatibility.
check_kinds!(
op_ty,
"CastKind::{kind:?} input must be a fn pointer, not {:?}",
ty::FnPtr(..)
);
check_kinds!(
target_type,
"CastKind::{kind:?} output must be a fn pointer, not {:?}",
ty::FnPtr(..)
);
}
CastKind::PointerCoercion(PointerCoercion::ClosureFnPointer(..)) => {
// FIXME: check safety, captures, and signature compatibility.
check_kinds!(
op_ty,
"CastKind::{kind:?} input must be a closure, not {:?}",
ty::Closure(..)
);
check_kinds!(
target_type,
"CastKind::{kind:?} output must be a fn pointer, not {:?}",
ty::FnPtr(..)
);
}
CastKind::PointerCoercion(PointerCoercion::MutToConstPointer) => {
// FIXME: check same pointee?
check_kinds!(
op_ty,
"CastKind::{kind:?} input must be a raw mut pointer, not {:?}",
ty::RawPtr(_, Mutability::Mut)
);
check_kinds!(
target_type,
"CastKind::{kind:?} output must be a raw const pointer, not {:?}",
ty::RawPtr(_, Mutability::Not)
);
}
CastKind::PointerCoercion(PointerCoercion::ArrayToPointer) => {
// FIXME: Check pointee types
check_kinds!(
op_ty,
"CastKind::{kind:?} input must be a raw pointer, not {:?}",
ty::RawPtr(..)
);
check_kinds!(
target_type,
"CastKind::{kind:?} output must be a raw pointer, not {:?}",
ty::RawPtr(..)
);
}
CastKind::PointerCoercion(PointerCoercion::Unsize) => {
// This is used for all `CoerceUnsized` types,
// not just pointers/references, so is hard to check.
}
CastKind::IntToInt | CastKind::IntToFloat => {
let input_valid = op_ty.is_integral() || op_ty.is_char() || op_ty.is_bool();
let target_valid = target_type.is_numeric() || target_type.is_char();
Expand All @@ -1147,10 +1215,29 @@ impl<'a, 'tcx> Visitor<'tcx> for TypeChecker<'a, 'tcx> {
);
}
}
CastKind::FnPtrToPtr | CastKind::PtrToPtr => {
if !(op_ty.is_any_ptr() && target_type.is_unsafe_ptr()) {
self.fail(location, "Can't cast {op_ty} into 'Ptr'");
}
CastKind::FnPtrToPtr => {
check_kinds!(
op_ty,
"CastKind::{kind:?} input must be a fn pointer, not {:?}",
ty::FnPtr(..)
);
check_kinds!(
target_type,
"CastKind::{kind:?} output must be a raw pointer, not {:?}",
ty::RawPtr(..)
);
}
CastKind::PtrToPtr => {
check_kinds!(
op_ty,
"CastKind::{kind:?} input must be a raw pointer, not {:?}",
ty::RawPtr(..)
);
check_kinds!(
target_type,
"CastKind::{kind:?} output must be a raw pointer, not {:?}",
ty::RawPtr(..)
);
}
CastKind::FloatToFloat | CastKind::FloatToInt => {
if !op_ty.is_floating_point() || !target_type.is_numeric() {
Expand Down
1 change: 0 additions & 1 deletion src/tools/tidy/src/allowed_run_make_makefiles.txt
Original file line number Diff line number Diff line change
Expand Up @@ -37,7 +37,6 @@ run-make/dump-ice-to-disk/Makefile
run-make/dump-mono-stats/Makefile
run-make/duplicate-output-flavors/Makefile
run-make/dylib-chain/Makefile
run-make/emit-named-files/Makefile
run-make/emit-path-unhashed/Makefile
run-make/emit-shared-files/Makefile
run-make/emit-stack-sizes/Makefile
Expand Down
44 changes: 44 additions & 0 deletions tests/codegen/simd/packed-simd-alignment.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
//@ compile-flags: -Cno-prepopulate-passes

#![crate_type = "lib"]
#![feature(repr_simd, core_intrinsics)]
// make sure that codegen emits correctly-aligned loads and stores for repr(packed, simd) types
// the alignment of a load should be no less than T, and no more than the size of the vector type
use std::intrinsics::simd as intrinsics;

#[derive(Copy, Clone)]
#[repr(packed, simd)]
struct f32x3([f32; 3]);

#[derive(Copy, Clone)]
#[repr(packed, simd)]
struct f32x4([f32; 4]);

// CHECK-LABEL: load_f32x3
#[no_mangle]
pub fn load_f32x3(floats: &f32x3) -> f32x3 {
// FIXME: Is a memcpy really the best we can do?
// CHECK: @llvm.memcpy.{{.*}}ptr align 4 {{.*}}ptr align 4
*floats
}

// CHECK-LABEL: load_f32x4
#[no_mangle]
pub fn load_f32x4(floats: &f32x4) -> f32x4 {
// CHECK: load <4 x float>, ptr %{{[a-z0-9_]*}}, align {{4|8|16}}
*floats
}

// CHECK-LABEL: add_f32x3
#[no_mangle]
pub fn add_f32x3(x: f32x3, y: f32x3) -> f32x3 {
// CHECK: load <3 x float>, ptr %{{[a-z0-9_]*}}, align 4
unsafe { intrinsics::simd_add(x, y) }
}

// CHECK-LABEL: add_f32x4
#[no_mangle]
pub fn add_f32x4(x: f32x4, y: f32x4) -> f32x4 {
// CHECK: load <4 x float>, ptr %{{[a-z0-9_]*}}, align {{4|8|16}}
unsafe { intrinsics::simd_add(x, y) }
}
33 changes: 0 additions & 33 deletions tests/run-make/emit-named-files/Makefile

This file was deleted.

25 changes: 25 additions & 0 deletions tests/run-make/emit-named-files/rmake.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
use std::fs::create_dir;
use std::path::Path;

use run_make_support::{rustc, tmp_dir};

fn emit_and_check(out_dir: &Path, out_file: &str, format: &str) {
let out_file = out_dir.join(out_file);
rustc().input("foo.rs").emit(&format!("{format}={}", out_file.display())).run();
assert!(out_file.is_file());
}

fn main() {
let out_dir = tmp_dir().join("emit");

create_dir(&out_dir).unwrap();

emit_and_check(&out_dir, "libfoo.s", "asm");
emit_and_check(&out_dir, "libfoo.bc", "llvm-bc");
emit_and_check(&out_dir, "libfoo.ll", "llvm-ir");
emit_and_check(&out_dir, "libfoo.o", "obj");
emit_and_check(&out_dir, "libfoo.rmeta", "metadata");
emit_and_check(&out_dir, "libfoo.rlib", "link");
emit_and_check(&out_dir, "libfoo.d", "dep-info");
emit_and_check(&out_dir, "libfoo.mir", "mir");
}
21 changes: 6 additions & 15 deletions tests/ui/simd/repr_packed.rs
Original file line number Diff line number Diff line change
Expand Up @@ -6,9 +6,6 @@
#[repr(simd, packed)]
struct Simd<T, const N: usize>([T; N]);

#[repr(simd)]
struct FullSimd<T, const N: usize>([T; N]);

fn check_size_align<T, const N: usize>() {
use std::mem;
assert_eq!(mem::size_of::<Simd<T, N>>(), mem::size_of::<[T; N]>());
Expand Down Expand Up @@ -39,21 +36,15 @@ fn main() {
check_ty::<f64>();

unsafe {
// powers-of-two have no padding and work as usual
// powers-of-two have no padding and have the same layout as #[repr(simd)]
let x: Simd<f64, 4> =
simd_add(Simd::<f64, 4>([0., 1., 2., 3.]), Simd::<f64, 4>([2., 2., 2., 2.]));
assert_eq!(std::mem::transmute::<_, [f64; 4]>(x), [2., 3., 4., 5.]);

// non-powers-of-two have padding and need to be expanded to full vectors
fn load<T, const N: usize>(v: Simd<T, N>) -> FullSimd<T, N> {
unsafe {
let mut tmp = core::mem::MaybeUninit::<FullSimd<T, N>>::uninit();
std::ptr::copy_nonoverlapping(&v as *const _, tmp.as_mut_ptr().cast(), 1);
tmp.assume_init()
}
}
let x: FullSimd<f64, 3> =
simd_add(load(Simd::<f64, 3>([0., 1., 2.])), load(Simd::<f64, 3>([2., 2., 2.])));
assert_eq!(x.0, [2., 3., 4.]);
// non-powers-of-two should have padding (which is removed by #[repr(packed)]),
// but the intrinsic handles it
let x: Simd<f64, 3> = simd_add(Simd::<f64, 3>([0., 1., 2.]), Simd::<f64, 3>([2., 2., 2.]));
let arr: [f64; 3] = x.0;
assert_eq!(arr, [2., 3., 4.]);
}
}

0 comments on commit 5e6c2b6

Please sign in to comment.