Correct some wrong uses of LLVM intrinsics

[sayantn]

There were some small mistakes in the LLVM intrinsics, these were autoupgraded by LLVM so the builds passed. Also replaces all extern "C" LLVM intrinsics blocks with extern "unadjusted"

AArch64

cc @adamgemmell

• wrong type parameter in neon.vluti4q.lane
• missing type parameter in prefetch
• typo in sve.{and,bic,eor,orr,orn}.z
• extra .x in sve.fcmla.lane and sve.frecpx
• trailing whitespace in sve.expa.x
• one missing type parameter in sve.fmmla
• splitting svcvt in two parts, because same size operands need a different intrinsic

x86

• missing type parameter in prefetch

NVPTX

I don't know who to cc here, the file was last modified by @gnzlbg but he doesn't seem active here anymore

• deprecated intrinsic llvm.nvvm.barrier0 used

WASM32

cc @alexcrichton

• missing type parameter in llvm.wasm.memory.{grow,size}

r? @folkertdev

[folkertdev]

Out of curiosity, how did you find these?

[sayantn]

Out of curiosity, how did you find these?

A modified version of rust-lang/rust#157145

[folkertdev]

This looks fine to me, but maybe target maintainers have thoughts?

Also I wonder if this breaks with LLVM 21 at all?

View changes since this review

[alexcrichton]

wasm bits lgtm, thanks!

[adamgemmell]

Looks good, thanks you!

View changes since this review