small_data_threshold.rs: Adapt to LLVM head changes

[TimNN]

When compiled against LLVM head, small_data_threshold.rs fails with:

/.../small_data_threshold.rs:61:12: error: RISCV: expected string not found in input
--
  | //@ RISCV: .section .sdata,
  | ^
  | /.../small_data_threshold.s:1:1: note: scanning from here
  | .text
  | ^
  | /.../small_data_threshold.s:6:2: note: possible intended match here
  | .section .sdata.U,"aw",@progbits
  | ^

I don't know how exactly the current output looks like, or if there was a specific reason for including the trailing comma on the first line.

I only saw a failure for RISCV, but it seemed sensible to adjust MIPS as well.

CI passes with this patch applied: https://buildkite.com/llvm-project/rust-llvm-integrate-prototype/builds/31053

@rustbot label: +llvm-main

cc @paulmenage

[TimNN]

@rustbot author

[TimNN]

@rustbot ready

r? @nikic

[paulmenage]

The current (expected) output looks like

.section .sdata,"aw",@progbits

The new LLVM output appears to include the variable name as part of the section name, e.g. .sdata.U for variable U.

For now (dealing with both current and future LLVM behaviour) this change makes sense, but once the new LLVM is part of the regular build (when will that likely be?) we should update the test to match the full section names, something like

//@ RISCV: .section .sdata.U,

(and the other similar match lines).

That's better than was possible with the old LLVM version, as it ensures that we're matching against the .sdata section specifically for the variable U.

[nikic]

when will that likely be?

In about half a year.

We could use .section .sdata{{(\.U)?}}, here, but not sure it's really better than just dropping the comma.

[nikic]

@bors r+ rollup

[bors]

📋 Looks like this PR is still in progress, ignoring approval.

Hint: Remove WIP from this PR's title when it is ready for review.

[nikic]

@bors r+ rollup

[bors]

📌 Commit cf006f3 has been approved by nikic

It is now in the queue for this repository.