Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

Commit f456ebf

Browse filesBrowse files
bjorn3bjorn3
committed
Merge branch 'fix_ptx_kernel_abi' into HEAD
2 parents e1f75dd + 0a44f67 commit f456ebf
Copy full SHA for f456ebf

File tree

Expand file treeCollapse file tree

3 files changed

+49
-44
lines changed
Filter options
Expand file treeCollapse file tree

3 files changed

+49
-44
lines changed

‎compiler/rustc_target/src/callconv/nvptx64.rs

Copy file name to clipboardExpand all lines: compiler/rustc_target/src/callconv/nvptx64.rs
+30-14Lines changed: 30 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
use super::{ArgAttribute, ArgAttributes, ArgExtension, CastTarget};
2-
use crate::abi::call::{ArgAbi, FnAbi, PassMode, Reg, Size, Uniform};
2+
use crate::abi::call::{ArgAbi, FnAbi, Reg, Size, Uniform};
33
use crate::abi::{HasDataLayout, TyAbiInterface};
44

55
fn classify_ret<Ty>(ret: &mut ArgAbi<'_, Ty>) {
@@ -53,21 +53,37 @@ where
5353
Ty: TyAbiInterface<'a, C> + Copy,
5454
C: HasDataLayout,
5555
{
56-
if matches!(arg.mode, PassMode::Pair(..)) && (arg.layout.is_adt() || arg.layout.is_tuple()) {
57-
let align_bytes = arg.layout.align.abi.bytes();
56+
match arg.mode {
57+
super::PassMode::Ignore | super::PassMode::Direct(_) => return,
58+
super::PassMode::Pair(_, _) => {}
59+
super::PassMode::Cast { .. } => unreachable!(),
60+
super::PassMode::Indirect { .. } => {}
61+
}
62+
63+
// FIXME only allow structs and wide pointers here
64+
// panic!(
65+
// "`extern \"ptx-kernel\"` doesn't allow passing types other than primitives and structs"
66+
// );
67+
68+
let align_bytes = arg.layout.align.abi.bytes();
5869

59-
let unit = match align_bytes {
60-
1 => Reg::i8(),
61-
2 => Reg::i16(),
62-
4 => Reg::i32(),
63-
8 => Reg::i64(),
64-
16 => Reg::i128(),
65-
_ => unreachable!("Align is given as power of 2 no larger than 16 bytes"),
66-
};
67-
arg.cast_to(Uniform::new(unit, Size::from_bytes(2 * align_bytes)));
70+
let unit = match align_bytes {
71+
1 => Reg::i8(),
72+
2 => Reg::i16(),
73+
4 => Reg::i32(),
74+
8 => Reg::i64(),
75+
16 => Reg::i128(),
76+
_ => unreachable!("Align is given as power of 2 no larger than 16 bytes"),
77+
};
78+
if arg.layout.size.bytes() / align_bytes == 1 {
79+
// Make sure we pass the struct as array at the LLVM IR level and not as a single integer.
80+
arg.cast_to(CastTarget {
81+
prefix: [Some(unit), None, None, None, None, None, None, None],
82+
rest: Uniform::new(unit, Size::ZERO),
83+
attrs: ArgAttributes::new(),
84+
});
6885
} else {
69-
// FIXME: find a better way to do this. See https://github.com/rust-lang/rust/issues/117271.
70-
arg.make_direct_deprecated();
86+
arg.cast_to(Uniform::new(unit, arg.layout.size));
7187
}
7288
}
7389

‎compiler/rustc_ty_utils/src/abi.rs

Copy file name to clipboardExpand all lines: compiler/rustc_ty_utils/src/abi.rs
+18-13Lines changed: 18 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -473,20 +473,25 @@ fn fn_abi_sanity_check<'tcx>(
473473
// This really shouldn't happen even for sized aggregates, since
474474
// `immediate_llvm_type` will use `layout.fields` to turn this Rust type into an
475475
// LLVM type. This means all sorts of Rust type details leak into the ABI.
476-
// However wasm sadly *does* currently use this mode so we have to allow it --
477-
// but we absolutely shouldn't let any more targets do that.
478-
// (Also see <https://github.com/rust-lang/rust/issues/115666>.)
476+
// However wasm sadly *does* currently use this mode for it's "C" ABI so we
477+
// have to allow it -- but we absolutely shouldn't let any more targets do
478+
// that. (Also see <https://github.com/rust-lang/rust/issues/115666>.)
479479
//
480-
// The unstable abi `PtxKernel` also uses Direct for now.
481-
// It needs to switch to something else before stabilization can happen.
482-
// (See issue: https://github.com/rust-lang/rust/issues/117271)
483-
assert!(
484-
matches!(&*tcx.sess.target.arch, "wasm32" | "wasm64")
485-
|| matches!(spec_abi, ExternAbi::PtxKernel | ExternAbi::Unadjusted),
486-
"`PassMode::Direct` for aggregates only allowed for \"unadjusted\" and \"ptx-kernel\" functions and on wasm\n\
487-
Problematic type: {:#?}",
488-
arg.layout,
489-
);
480+
// The unadjusted ABI also uses Direct for all args and is ill-specified,
481+
// but unfortunately we need it for calling certain LLVM intrinsics.
482+
483+
match spec_abi {
484+
ExternAbi::Unadjusted => {}
485+
ExternAbi::C { unwind: _ }
486+
if matches!(&*tcx.sess.target.arch, "wasm32" | "wasm64") => {}
487+
_ => {
488+
panic!(
489+
"`PassMode::Direct` for aggregates only allowed for \"unadjusted\" functions and on wasm\n\
490+
Problematic type: {:#?}",
491+
arg.layout,
492+
);
493+
}
494+
}
490495
}
491496
}
492497
}

‎tests/assembly/nvptx-kernel-abi/nvptx-kernel-args-abi-v7.rs

Copy file name to clipboardExpand all lines: tests/assembly/nvptx-kernel-abi/nvptx-kernel-args-abi-v7.rs
+1-17Lines changed: 1 addition & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -242,22 +242,6 @@ pub unsafe extern "ptx-kernel" fn f_float_array_arg(_a: [f32; 5]) {}
242242
//pub unsafe extern "ptx-kernel" fn f_u128_array_arg(_a: [u128; 5]) {}
243243

244244
// CHECK: .visible .entry f_u32_slice_arg(
245-
// CHECK: .param .u64 f_u32_slice_arg_param_0
246-
// CHECK: .param .u64 f_u32_slice_arg_param_1
245+
// CHECK: .param .align 8 .b8 f_u32_slice_arg_param_0[16]
247246
#[no_mangle]
248247
pub unsafe extern "ptx-kernel" fn f_u32_slice_arg(_a: &[u32]) {}
249-
250-
// CHECK: .visible .entry f_tuple_u8_u8_arg(
251-
// CHECK: .param .align 1 .b8 f_tuple_u8_u8_arg_param_0[2]
252-
#[no_mangle]
253-
pub unsafe extern "ptx-kernel" fn f_tuple_u8_u8_arg(_a: (u8, u8)) {}
254-
255-
// CHECK: .visible .entry f_tuple_u32_u32_arg(
256-
// CHECK: .param .align 4 .b8 f_tuple_u32_u32_arg_param_0[8]
257-
#[no_mangle]
258-
pub unsafe extern "ptx-kernel" fn f_tuple_u32_u32_arg(_a: (u32, u32)) {}
259-
260-
// CHECK: .visible .entry f_tuple_u8_u8_u32_arg(
261-
// CHECK: .param .align 4 .b8 f_tuple_u8_u8_u32_arg_param_0[8]
262-
#[no_mangle]
263-
pub unsafe extern "ptx-kernel" fn f_tuple_u8_u8_u32_arg(_a: (u8, u8, u32)) {}

0 commit comments

Comments
0 (0)
Morty Proxy This is a proxified and sanitized view of the page, visit original site.