Make Pulley pass simd_f32x4_arith.wast
#9897
Conversation
github-actions
bot
added
cranelift
Subscribe to Label Actioncc @fitzgen
This issue or pull request has been labeled: "cranelift", "pulley"
Thus the following users have been cc'd because of the following labels:
To subscribe or unsubscribe from this label, edit the |
pulley/src/lib.rs
Outdated
| /// `low128(dst) = low128(src1) - low128(src2)` | ||
| vsub32x4 = Vsub32x4 { operands: BinaryOperands<VReg> }; |
There was a problem hiding this comment.
Mind using f32x4 in the opcodes (e.g. vsubf32x4) to clearly indicate that it's for floats and not integers?
|
@tyoeer I think this might need a rebase? |
Adds float SIMD instructions on 4 lanes (f32x4) for subtraction, multiplication, and negation. `vtrunc32x4` and `vmuli32x4` were used as basis on how to organize things.
To be exact: `spec_testsuite/simd_f32x4_arith.wast` has been removed from the should fail list for Pulley.
Specifically: - "vsub32x4" -> "vsubf32x4" - "vmul32x4" -> "vmulf32x4"
It has been rebased. Another PR has also already added |
|
Oh no worries! This falls into the "various technical reasons" category where this couldn't be merged due to a merge conflict. I'll go update those docs though to mention that, thanks for dropping the link! |
Indicate that a rebase is required if there's a merge conflict or CI issues for example. Inspired by discussion on bytecodealliance#9897
Indicate that a rebase is required if there's a merge conflict or CI issues for example. Inspired by discussion on #9897
Helping towards issue #9783.
This PR makes Pulley pass
spec_testsuite/simd_f32x4_arith.wast, by adding f32x4 instructions for subtraction, negation, multiplication,and division(division already got added by another PR).Random notes:
pulley/src/lib.rsare based on what is done withvtrunc32x4.vmuli32x4.ftrunc32,vtrunc32x4, andvtrunc64x2are together, butftrunc64isn't near there.fat the end, e.g.vtrunc32x4vs.vabsf32x4.