Translate builtin_assume and unaligned SIMD intrinsics
#1132
Conversation
Unaligned vector types such as `__m128_u` or `__256_u` not not have any equivalents in Rust so we translate them to their aligned equivalents. This is unsound so we emit warning that the program may be mistranspiled.
thedataking
force-pushed
the
perl/builtin_assume
branch
from
8cc6c91 to
fb3add3
Compare
| // TODO: provide proper implementation, this is just to get stuff building | ||
| "__builtin_assume" => Ok(self.convert_expr(ctx.used(), args[0])?), |
There was a problem hiding this comment.
When you do implement this, note that Rust (as of 1.81.0) now has core::hint::assert_unchecked, which should be the equivalent of __builtin_assume.
There was a problem hiding this comment.
@kkysen do you want to add support for __builtin_assume into a separate PR with a test?
It seems my PR is trying to do two things and only one is blocked on what rust provides, so that's messy.
WDYT about landing the rest of this PR? Should the warning reference the PR folkert opened?
There was a problem hiding this comment.
do you want to add support for
__builtin_assumeinto a separate PR with a test?
We're still stuck on the old nightly, so adding support for newer features like this is tricky. That said, we could merge #1227, which would use stable for c2rust-transpile and let us start targeting newer Rust versions for just the transpiler. #1227 didn't work before, but now that CI is on GitHub Actions, it works without much modifications, so it wouldn't be too much work to review and merge that. If we do that, I can add support for it.
There was a problem hiding this comment.
There does seem to be some unaligned stuff in stdarch.
For example, _mm_storeu_si128.
Also, https://doc.rust-lang.org/std/arch/index.html?search=loadu.
| // Drop the `_u` suffix for unaligned SIMD types as they have | ||
| // no equivalents in Rust. Warn the user about possible probems. |
There was a problem hiding this comment.
Can we use this for unaligned SIMD types?
use core::arch::x86_64::__m256;
use core::mem;
#[unsafe(no_mangle)]
pub fn aligned() -> usize {
mem::align_of::<__m256>()
}
#[repr(packed)]
pub struct UnalignedM256(pub __m256);
#[unsafe(no_mangle)]
pub fn unaligned() -> usize {
mem::align_of::<UnalignedM256>()
}#[repr(packed)] lets you override the alignment and unalign a type.
Unaligned vector types such as
__m128_uor__256_unot not have any equivalents in Rust so we translate them to their aligned equivalents. This is unsound so we emit warning that the program may be mistranspiled. These fixes are required to transpile thezstdcompressor on modern x86 hosts.