Skip to content

Block SIMD in transmute_immediate; delete OperandValueKind #143410

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 3 commits into
base: master
Choose a base branch
from
+138 −214
Open

Block SIMD in transmute_immediate; delete OperandValueKind #143410

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
5292554
Block SIMD in transmute_immediate; delete `OperandValueKind`
scottmcm Jul 4, 2025
e0c54c3
Rename `transmute_immediate` → `transmute_scalar`
scottmcm Jul 4, 2025
4e61527
Address PR feedback
scottmcm Jul 4, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 10 additions & 10 deletions compiler/rustc_codegen_ssa/src/mir/operand.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@ use rustc_session::config::OptLevel;
use tracing::{debug, instrument};

use super::place::{PlaceRef, PlaceValue};
use super::rvalue::transmute_immediate;
use super::rvalue::transmute_scalar;
use super::{FunctionCx, LocalRef};
use crate::common::IntPredicate;
use crate::traits::*;
Expand Down Expand Up @@ -346,14 +346,16 @@ impl<'a, 'tcx, V: CodegenObject> OperandRef<'tcx, V> {

let val = if field.is_zst() {
OperandValue::ZeroSized
} else if let BackendRepr::SimdVector { .. } = self.layout.backend_repr {
// codegen_transmute_operand doesn't support SIMD, but since the previous
// check handled ZSTs, the only possible field access into something SIMD
// is to the `non_1zst_field` that's the same SIMD. (Other things, even
// just padding, would change the wrapper's representation type.)
assert_eq!(field.size, self.layout.size);
self.val
} else if field.size == self.layout.size {
assert_eq!(offset.bytes(), 0);
fx.codegen_transmute_operand(bx, *self, field).unwrap_or_else(|| {
bug!(
"Expected `codegen_transmute_operand` to handle equal-size \
field {i:?} projection from {self:?} to {field:?}"
)
})
fx.codegen_transmute_operand(bx, *self, field)
} else {
let (in_scalar, imm) = match (self.val, self.layout.backend_repr) {
// Extract a scalar component from a pair.
Expand Down Expand Up @@ -606,10 +608,8 @@ impl<'a, 'tcx, V: CodegenObject> OperandRef<'tcx, Result<V, abi::Scalar>> {
};

let mut update = |tgt: &mut Result<V, abi::Scalar>, src, from_scalar| {
let from_bty = bx.cx().type_from_scalar(from_scalar);
let to_scalar = tgt.unwrap_err();
let to_bty = bx.cx().type_from_scalar(to_scalar);
let imm = transmute_immediate(bx, src, from_scalar, from_bty, to_scalar, to_bty);
let imm = transmute_scalar(bx, src, from_scalar, to_scalar);
*tgt = Ok(imm);
};

Expand Down
256 changes: 101 additions & 155 deletions compiler/rustc_codegen_ssa/src/mir/rvalue.rs
View file
Open in desktop

Large diffs are not rendered by default.

11 changes: 7 additions & 4 deletions tests/codegen/intrinsics/transmute-x64.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,17 +9,20 @@ use std::mem::transmute;
// CHECK-LABEL: @check_sse_pair_to_avx(
#[no_mangle]
pub unsafe fn check_sse_pair_to_avx(x: (__m128i, __m128i)) -> __m256i {
// CHECK: start:
// CHECK-NOT: alloca
// CHECK: %0 = load <4 x i64>, ptr %x, align 16
// CHECK: store <4 x i64> %0, ptr %_0, align 32
// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 32 %_0, ptr align 16 %x, i64 32, i1 false)
// CHECK-NEXT: ret void
transmute(x)
}

// CHECK-LABEL: @check_sse_pair_from_avx(
#[no_mangle]
pub unsafe fn check_sse_pair_from_avx(x: __m256i) -> (__m128i, __m128i) {
// CHECK: start:
// CHECK-NOT: alloca
// CHECK: %0 = load <4 x i64>, ptr %x, align 32
// CHECK: store <4 x i64> %0, ptr %_0, align 16
// CHECK-NEXT: %[[TEMP:.+]] = load <4 x i64>, ptr %x, align 32
// CHECK-NEXT: store <4 x i64> %[[TEMP]], ptr %_0, align 16
// CHECK-NEXT: ret void
transmute(x)
}
14 changes: 7 additions & 7 deletions tests/codegen/intrinsics/transmute.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,28 +29,28 @@ pub struct Aggregate8(u8);
// CHECK-LABEL: @check_bigger_size(
#[no_mangle]
pub unsafe fn check_bigger_size(x: u16) -> u32 {
// CHECK: call void @llvm.trap
// CHECK: call void @llvm.assume(i1 false)
transmute_unchecked(x)
}

// CHECK-LABEL: @check_smaller_size(
#[no_mangle]
pub unsafe fn check_smaller_size(x: u32) -> u16 {
// CHECK: call void @llvm.trap
// CHECK: call void @llvm.assume(i1 false)
transmute_unchecked(x)
}

// CHECK-LABEL: @check_smaller_array(
#[no_mangle]
pub unsafe fn check_smaller_array(x: [u32; 7]) -> [u32; 3] {
// CHECK: call void @llvm.trap
// CHECK: call void @llvm.assume(i1 false)
transmute_unchecked(x)
}

// CHECK-LABEL: @check_bigger_array(
#[no_mangle]
pub unsafe fn check_bigger_array(x: [u32; 3]) -> [u32; 7] {
// CHECK: call void @llvm.trap
// CHECK: call void @llvm.assume(i1 false)
transmute_unchecked(x)
}

Expand All @@ -73,9 +73,9 @@ pub unsafe fn check_to_empty_array(x: [u32; 5]) -> [u32; 0] {
#[no_mangle]
#[custom_mir(dialect = "runtime", phase = "optimized")]
pub unsafe fn check_from_empty_array(x: [u32; 0]) -> [u32; 5] {
// CHECK-NOT: trap
// CHECK: call void @llvm.trap
// CHECK-NOT: trap
// CHECK-NOT: call
// CHECK: call void @llvm.assume(i1 false)
// CHECK-NOT: call
mir! {
{
RET = CastTransmute(x);
Expand Down
6 changes: 2 additions & 4 deletions tests/codegen/simd-intrinsic/simd-intrinsic-transmute-array.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,8 +40,7 @@ pub fn build_array_s(x: [f32; 4]) -> S<4> {
// CHECK-LABEL: @build_array_transmute_s
#[no_mangle]
pub fn build_array_transmute_s(x: [f32; 4]) -> S<4> {
// CHECK: %[[VAL:.+]] = load <4 x float>, ptr %x, align [[ARRAY_ALIGN]]
// CHECK: store <4 x float> %[[VAL:.+]], ptr %_0, align [[VECTOR_ALIGN]]
// CHECK: call void @llvm.memcpy.{{.+}}({{.*}} align [[VECTOR_ALIGN]] {{.*}} align [[ARRAY_ALIGN]] {{.*}}, [[USIZE]] 16, i1 false)
unsafe { std::mem::transmute(x) }
}

Expand All @@ -55,7 +54,6 @@ pub fn build_array_t(x: [f32; 4]) -> T {
// CHECK-LABEL: @build_array_transmute_t
#[no_mangle]
pub fn build_array_transmute_t(x: [f32; 4]) -> T {
// CHECK: %[[VAL:.+]] = load <4 x float>, ptr %x, align [[ARRAY_ALIGN]]
// CHECK: store <4 x float> %[[VAL:.+]], ptr %_0, align [[VECTOR_ALIGN]]
// CHECK: call void @llvm.memcpy.{{.+}}({{.*}} align [[VECTOR_ALIGN]] {{.*}} align [[ARRAY_ALIGN]] {{.*}}, [[USIZE]] 16, i1 false)
unsafe { std::mem::transmute(x) }
}
14 changes: 10 additions & 4 deletions tests/codegen/transmute-scalar.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,8 +111,11 @@ pub fn fake_bool_unsigned_to_bool(b: FakeBoolUnsigned) -> bool {
struct S([i64; 1]);

// CHECK-LABEL: define{{.*}}i64 @single_element_simd_to_scalar(<1 x i64> %b)
// CHECK: bitcast <1 x i64> %b to i64
// CHECK: ret i64
// CHECK-NEXT: start:
// CHECK-NEXT: %[[RET:.+]] = alloca [8 x i8]
// CHECK-NEXT: store <1 x i64> %b, ptr %[[RET]]
// CHECK-NEXT: %[[TEMP:.+]] = load i64, ptr %[[RET]]
// CHECK-NEXT: ret i64 %[[TEMP]]
#[no_mangle]
#[cfg_attr(target_family = "wasm", target_feature(enable = "simd128"))]
#[cfg_attr(target_arch = "arm", target_feature(enable = "neon"))]
Expand All @@ -124,8 +127,11 @@ pub extern "C" fn single_element_simd_to_scalar(b: S) -> i64 {
}

// CHECK-LABEL: define{{.*}}<1 x i64> @scalar_to_single_element_simd(i64 %b)
// CHECK: bitcast i64 %b to <1 x i64>
// CHECK: ret <1 x i64>
// CHECK-NEXT: start:
// CHECK-NEXT: %[[RET:.+]] = alloca [8 x i8]
// CHECK-NEXT: store i64 %b, ptr %[[RET]]
// CHECK-NEXT: %[[TEMP:.+]] = load <1 x i64>, ptr %[[RET]]
// CHECK-NEXT: ret <1 x i64> %[[TEMP]]
#[no_mangle]
#[cfg_attr(target_family = "wasm", target_feature(enable = "simd128"))]
#[cfg_attr(target_arch = "arm", target_feature(enable = "neon"))]
Expand Down
31 changes: 1 addition & 30 deletions tests/codegen/vec-in-place.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,9 +41,6 @@ pub fn vec_iterator_cast_primitive(vec: Vec<i8>) -> Vec<u8> {
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
vec.into_iter().map(|e| e as u8).collect()
}

Expand All @@ -55,9 +52,6 @@ pub fn vec_iterator_cast_wrapper(vec: Vec<u8>) -> Vec<Wrapper<u8>> {
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
vec.into_iter().map(|e| Wrapper(e)).collect()
}

Expand Down Expand Up @@ -86,9 +80,6 @@ pub fn vec_iterator_cast_unwrap(vec: Vec<Wrapper<u8>>) -> Vec<u8> {
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
vec.into_iter().map(|e| e.0).collect()
}

Expand All @@ -100,9 +91,6 @@ pub fn vec_iterator_cast_aggregate(vec: Vec<[u64; 4]>) -> Vec<Foo> {
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
vec.into_iter().map(|e| unsafe { std::mem::transmute(e) }).collect()
}

Expand All @@ -114,9 +102,6 @@ pub fn vec_iterator_cast_deaggregate_tra(vec: Vec<Bar>) -> Vec<[u64; 4]> {
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call

// Safety: For the purpose of this test we assume that Bar layout matches [u64; 4].
// This currently is not guaranteed for repr(Rust) types, but it happens to work here and
Expand All @@ -133,9 +118,6 @@ pub fn vec_iterator_cast_deaggregate_fold(vec: Vec<Baz>) -> Vec<[u64; 4]> {
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call
// CHECK: call{{.+}}void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: loop
// CHECK-NOT: call

// Safety: For the purpose of this test we assume that Bar layout matches [u64; 4].
// This currently is not guaranteed for repr(Rust) types, but it happens to work here and
Expand All @@ -156,12 +138,7 @@ pub fn vec_iterator_cast_unwrap_drop(vec: Vec<Wrapper<String>>) -> Vec<String> {
// CHECK-NOT: call
// CHECK-NOT: %{{.*}} = mul
// CHECK-NOT: %{{.*}} = udiv
// CHECK: call
// CHECK-SAME: void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: br i1 %{{.*}}, label %{{.*}}, label %{{.*}}
// CHECK-NOT: call
// CHECK-NOT: %{{.*}} = mul
// CHECK-NOT: %{{.*}} = udiv
// CHECK: ret void

vec.into_iter().map(|Wrapper(e)| e).collect()
}
Expand All @@ -178,12 +155,6 @@ pub fn vec_iterator_cast_wrap_drop(vec: Vec<String>) -> Vec<Wrapper<String>> {
// CHECK-NOT: call
// CHECK-NOT: %{{.*}} = mul
// CHECK-NOT: %{{.*}} = udiv
// CHECK: call
// CHECK-SAME: void @llvm.assume(i1 %{{.+}})
// CHECK-NOT: br i1 %{{.*}}, label %{{.*}}, label %{{.*}}
// CHECK-NOT: call
// CHECK-NOT: %{{.*}} = mul
// CHECK-NOT: %{{.*}} = udiv
// CHECK: ret void

vec.into_iter().map(Wrapper).collect()
Expand Down
Loading