Stream: git-wasmtime

Topic: wasmtime / PR #8365 winch(arm64): fpu arithmetics (add, s...

Wasmtime GitHub notifications bot (Apr 14 2024 at 15:01):

evacchi opened PR #8365 from evacchi:winch-fpu to bytecodealliance:main:

• Add some of the instructions in #8321
• Introduces preliminary support to FPU registers to regalloc
• Introduces tests for the instructions above (ported over from x64)
• Fixes a few typos.

This is a draft with a handful of tests for float_add{32,64} (LGTM but another pair of eyes would help), I will add the remaining later.

Wasmtime GitHub notifications bot (Apr 14 2024 at 15:07):

github-actions[bot] commented on PR #8365:

Subscribe to Label Action

cc @saulecabrera

<details>
This issue or pull request has been labeled: "winch"

Thus the following users have been cc'd because of the following labels:

• saulecabrera: winch

To subscribe or unsubscribe from this label, edit the <code>.github/subscribe-to-label.json</code> configuration file.

Learn more.
</details>

Wasmtime GitHub notifications bot (Apr 15 2024 at 07:40):

evacchi updated PR #8365.

Wasmtime GitHub notifications bot (Apr 15 2024 at 07:43):

evacchi edited PR #8365:

• Add some of the instructions in #8321
• Introduces preliminary support to FPU registers to regalloc
• Introduces tests for the instructions above (ported over from x64)
• Fixes a few typos.

This is a draft ~with a handful of tests for float_add{32,64} (LGTM but another pair of eyes would help), I will add the remaining later~. Added the remaining tests.

Wasmtime GitHub notifications bot (Apr 15 2024 at 15:15):

saulecabrera submitted PR review:

Thanks for the PR; the general direction looks good to me. I left a couple of comments that I think need to be resolved before landing this work.

Wasmtime GitHub notifications bot (Apr 15 2024 at 15:15):

saulecabrera submitted PR review:

Thanks for the PR; the general direction looks good to me. I left a couple of comments that I think need to be resolved before landing this work.

Wasmtime GitHub notifications bot (Apr 15 2024 at 15:15):

saulecabrera created PR review comment:

I believe here we need to properly exclude the designated float scratch register?

pub(crate) const NON_ALLOCATABLE_FPR: u32 = 1 << float_scratch().hw_enc();

/// Bitmask to represent the available general purpose registers
pub(crate) const ALL_FPR: u32 = u32::MAX & !NON_ALLOCATABLE_FPR;

Wasmtime GitHub notifications bot (Apr 15 2024 at 15:15):

saulecabrera created PR review comment:

I was under the impression that we could emit fmov sd, sn (the non-64-bit case) according to https://developer.arm.com/documentation/ddi0602/2024-03/SIMD-FP-Instructions/FMOV--register---Floating-point-Move-register-without-conversion-?lang=en, but maybe we'd need to implement this in Cranelift to make it possible. Could you a TODO comment here? I'll follow-up on this one.

Wasmtime GitHub notifications bot (Apr 15 2024 at 15:15):

saulecabrera created PR review comment:

/// Bitmask to represent the available floating point registers.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:17):

evacchi submitted PR review.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:17):

evacchi created PR review comment:

whoops!

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:18):

evacchi submitted PR review.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:18):

evacchi created PR review comment:

oh good point!

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:27):

evacchi submitted PR review.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:27):

evacchi created PR review comment:

sure I'll add a TODO!

If I understand correctly, the optimizing backend always uses F64 unless it needs V128:

ISLE:

(decl fpu_move (Type Reg) Reg)
(rule (fpu_move _ src)
      (let ((dst WritableReg (temp_writable_reg $I8X16))
            (_ Unit (emit (MInst.FpuMove128 dst src))))
        dst))
(rule 1 (fpu_move (fits_in_64 _) src)
      (let ((dst WritableReg (temp_writable_reg $F64))
            (_ Unit (emit (MInst.FpuMove64 dst src))))
        dst))

generated code:

// Generated as internal constructor for term fpu_move.
pub fn constructor_fpu_move<C: Context>(ctx: &mut C, arg0: Type, arg1: Reg) -> Reg {
    let v7 = C::fits_in_64(ctx, arg0);
    if let Some(v8) = v7 {
        let v10 = C::temp_writable_reg(ctx, F64);
        let v11 = MInst::FpuMove64 { rd: v10, rn: arg1 };
        let v12 = C::emit(ctx, &v11);
        let v13 = C::writable_reg_to_reg(ctx, v10);
        // Rule at src/isa/aarch64/inst.isle line 2251.
        return v13;
    }
    ... // v128

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:30):

evacchi updated PR #8365.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:30):

evacchi updated PR #8365.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:32):

evacchi updated PR #8365.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:32):

evacchi has marked PR #8365 as ready for review.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:32):

evacchi requested wasmtime-core-reviewers for a review on PR #8365.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:32):

evacchi requested fitzgen for a review on PR #8365.

Wasmtime GitHub notifications bot (Apr 15 2024 at 16:32):

evacchi requested wasmtime-compiler-reviewers for a review on PR #8365.

Wasmtime GitHub notifications bot (Apr 15 2024 at 19:42):

saulecabrera submitted PR review.

Wasmtime GitHub notifications bot (Apr 15 2024 at 20:12):

saulecabrera merged PR #8365.

Last updated: Dec 23 2024 at 12:05 UTC