Stream: git-wasmtime

Topic: wasmtime / PR #6657 riscv64: Implement various SIMD float...

Wasmtime GitHub notifications bot (Jun 28 2023 at 13:17):

afonso360 requested fitzgen for a review on PR #6657.

Wasmtime GitHub notifications bot (Jun 28 2023 at 13:17):

afonso360 requested wasmtime-compiler-reviewers for a review on PR #6657.

Wasmtime GitHub notifications bot (Jun 28 2023 at 13:17):

afonso360 requested wasmtime-default-reviewers for a review on PR #6657.

Wasmtime GitHub notifications bot (Jun 28 2023 at 13:17):

afonso360 opened PR #6657 from afonso360:riscv-simd-float-arith2 to bytecodealliance:main:

:wave: Hey,

This PR implements a few SIMD float ops. These are:

• fabs
• fcopysign
• fmin_pseudo
• fmax_pseudo
• fmin
• fmax

f{min,max} are slightly complicated since we don't have a instruction that exactly matches the required semantics. The implementation here builds a mask for the inputs that are NaN and returns the canonical NaN in these cases. This is the same lowering that V8 uses.

I've had to alter the simd-fmax-fmin tests since they were expecting the sign bit of the resulting NaN to be unchanged, however this seems not to be required as per our documentation. I've moved these tests to the nondeterministic testsuites that we have.

Wasmtime GitHub notifications bot (Jun 28 2023 at 15:22):

alexcrichton submitted PR review:

:+1:

Wasmtime GitHub notifications bot (Jun 28 2023 at 16:00):

alexcrichton merged PR #6657.

Last updated: Oct 23 2024 at 20:03 UTC