Add SIMD funnel shift and round-to-even intrinsics #142078
Conversation
|
r? @wesleywiser rustbot has assigned @wesleywiser. Use |
rustbot
added
A-LLVM
|
Some changes occurred to the intrinsics. Make sure the CTFE / Miri interpreter cc @rust-lang/miri, @RalfJung, @oli-obk, @lcnr Some changes occurred to the platform-builtins intrinsics. Make sure the cc @antoyo, @GuillaumeGomez, @bjorn3, @calebzulawski, @programmerjake |
rustbot
added
A-intrinsics
Please add Miri to the list. :) (Doesn't have to be in the first PR.) |
Both |
You should be able to have more or less the same implementation for cg_gcc by putting the code in this file. The complicated part is to find the equivalent intrinsic names for GCC. |
If you want to block this PR on submitting a T-libs-api ACP, you can do that, but I don't see why you would. |
I only found the intrinsic name for x86 as
I don't think this should blocked on adding these to int types, just thought this might be a nice place to discuss it. I will go ahead and submit an independent ACP for that |
It seems GCC doesn't have an arch-independent builtin for this, so I guess it's better for this PR to not implement the intrinsic for cg_gcc. |
|
ok thanks |
|
Some changes occurred in compiler/rustc_codegen_cranelift cc @bjorn3 Some changes occurred in compiler/rustc_codegen_gcc |
|
I have implemented |
|
The Miri subtree was changed cc @rust-lang/miri |
library/core/src/intrinsics/simd.rs
Outdated
| pub unsafe fn simd_fshl<T>(a: T, b: T, shift: T) -> T; | ||
|
|
||
| /// Funnel Shifts vector right elementwise, with UB on overflow. | ||
| /// | ||
| /// Concatenates `a` and `b` elementwise (with `a` in the most significant half), | ||
| /// creating a vector of the same length, but with each element being twice as | ||
| /// wide. Then shift this vector right elementwise by `shift`, shifting in zeros, | ||
| /// and extract the least significant half of each of the elements. If `a` and `b` | ||
| /// are the same, this is equivalent to an elementwise rotate right operation. | ||
| /// | ||
| /// `T` must be a vector of integers. | ||
| /// | ||
| /// # Safety | ||
| /// | ||
| /// Each element of `shift` must be less than `<int>::BITS`. | ||
| #[rustc_intrinsic] | ||
| #[rustc_nounwind] | ||
| pub unsafe fn simd_fshr<T>(a: T, b: T, shift: T) -> T; |
There was a problem hiding this comment.
So, I have been thinking and I can't get over the bit where the current mnemonic sounds like it has something to do with floats in my head. I think we might want to call out these names as simd_funnel_shl and simd_funnel_shr? I looked around and we get names like __funnelshift_lc and __funnelshift_rc for similar intrinsics (e.g. CUDA), and I know that some languages actually do support shifts on floats, with... sometimes interesting semantics.
There was a problem hiding this comment.
Yeah, simd_funnel_sh{l,r} sounds nice enough (with the scalar functions as funnel_sh{l,r}). I will change the name
There was a problem hiding this comment.
I have changed the names. Btw @workingjubilee as you seem to be interested, should I r? you?
sayantn
changed the title
simd_fsh{l,r} and simd_round_ties_even
This PR adds 3 new SIMD intrinsics
simd_funnel_shl- funnel shift leftsimd_funnel_shr- funnel shift rightsimd_round_ties_even(vector version ofround_ties_even_fN)TODO: implement
simd_fsh{l,r}in miri, cg_gcc and cg_clif#t-compiler > More SIMD intrinsics
@rustbot label T-compiler T-libs A-intrinsics F-core_intrinsics
cc @workingjubilee