Remove the workarounds for some resolved issues

[sayantn]

• Re-enable dox CI for armv7-unknown-linux-gnueabihf
• Remove workarounds for Buggy optimization of vfmaddcsh intrinsics llvm/llvm-project#98306

[rustbot]

r? @Amanieu

rustbot has assigned @Amanieu.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[sayantn]

There seems to be some instruction check failures (specifically vdup_n and vmov_n intrinsics both are generating vmov, and before this they both generated vdup). @Amanieu do I need to fix that or is that ok to have?

[Amanieu]

There seems to be some instruction check failures (specifically vdup_n and vmov_n intrinsics both are generating vmov, and before this they both generated vdup). @Amanieu do I need to fix that or is that ok to have?

I'd just remove the tests on those (set the assert_instr to nop or just remove it). They are data movement operations and we don't guarantee a specific instruction sequence for them.

[sayantn]

There are a few more changes in ASM, specifically

• vqdmlal_high_n_s{16,32} : sqdmlal2 -> sqdmlal
• vqdmlsl_high_n_s{16,32} : sqdmlsl2 -> sqdmlsl
• vqdmull_high_n_s{16,32} : sqdmull2 -> sqdmull

@Amanieu are these acceptable?

[Amanieu]

That looks like a regression on the LLVM side, it should be emitting the 2 variant of the instruction.

For now this can be accepted with a FIXME, but an issue should be opened in LLVM about it.

[sayantn]

Seems like rust-lang/rust#60637 is still causing some problems in LLVM optimizing for darwin and ios, so I am delegating that