Add arm_intrinsics.json for intrinsic-test and stdarch-verify
#1427
Conversation
This involves moving from the ACLE intrinsic definitions (which aren't available for SVE at this point) to a JSON file. This was derived from ARM's documentation[^1], and then relicensed under `MIT OR Apache-2.0` for use in this repository. [^1]: https://developer.arm.com/architectures/instruction-sets/intrinsics
|
r? @Amanieu (rustbot has picked a reviewer for you, use r? to override) |
| #vqshluq_n_s32 | ||
| #vqshluq_n_s64 | ||
| #vqshluq_n_s8 | ||
| #vqshlus_n_s32 |
There was a problem hiding this comment.
My understanding is that this issue is still present if intrinsic-test is changed to run the Rust code without --release. The problem is due to the way we generate LLVM IR: we emit the vector argument as a constant and then load it, but LLVM's intrinsic only works if the vector is passed as an immediate to the intrinsics.
There was a problem hiding this comment.
Yes, it's still there - hence why I've only commented these intrinsics out instead of removing them. We have a ticket to fix this in the future. The test tool currently runs in --release mode and I think it makes sense to extend coverage by including this set.
| vmul_lane_f32 | ||
| vmul_n_f32 | ||
| vmulq_lane_f32 | ||
| vmulq_n_f32 |
There was a problem hiding this comment.
Have you checked this by trying the C version without the -O2 flag?
There was a problem hiding this comment.
I have, everything passes on aarch64 and armv7 with Rust --release and clang -O0
|
Thanks! |
|
@Amanieu I've made a mistake and was about to push a fix that would fix CI! Should I open a new PR? |
|
Yes, please do! |
|
Thanks ❤️ |
This removes the ACLE submodule and ARM HTML scrape, resolving a few licensing issues. The JSON file, along with updates to QEMU and Clang in CI also increases the test coverage of Neon intrinsics.
cc: rust-lang/rust#90972 @pietroalbini