Stream: git-wasmtime

Topic: wasmtime / PR #4460 cranelift: Implement scalar `fma` on x86

view this post on Zulip Wasmtime GitHub notifications bot (Jul 16 2022 at 16:15):

afonso360 opened PR #4460 from x86-scalar-fma to main:

:wave: Hey,

This PR Implements fma for scalar values.

x86 does not have dedicated instructions for scalar fma, so we lower to a libcall which seems to be what llvm does.

I tried to implement it for SIMD as well but got a bit lost, ill try again later.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 16 2022 at 16:18):

afonso360 edited PR #4460 from x86-scalar-fma to main:

:wave: Hey,

This PR Implements fma for scalar values.

x86 does not have dedicated instructions for scalar fma, so we lower to a libcall which is matches llvm behaviour.

I tried to implement it for SIMD as well but got a bit lost, ill try again later.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 16 2022 at 16:20):

afonso360 submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 16 2022 at 16:20):

afonso360 created PR review comment:

I think there is probably a better way to do this, but I'm still not too comfortable with isle.

Can we make this more generic?

view this post on Zulip Wasmtime GitHub notifications bot (Jul 16 2022 at 18:04):

afonso360 edited PR #4460 from x86-scalar-fma to main:

:wave: Hey,

This PR Implements fma for scalar values.

x86 does not have dedicated instructions for scalar fma, so we lower to a libcall which matches llvm's behaviour.

I tried to implement it for SIMD as well but got a bit lost, ill try again later.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 17 2022 at 10:09):

afonso360 updated PR #4460 from x86-scalar-fma to main.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 17 2022 at 10:09):

afonso360 edited PR #4460 from x86-scalar-fma to main:

:wave: Hey,

This PR Implements fma for scalar values.

x86 does not have dedicated instructions for scalar fma, so we lower to a libcall which matches llvm's behaviour.

I tried to implement it for SIMD as well but got a bit lost, ill try again later.

We can't enable fma.clif runtests until we merge #4453.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 01 2022 at 20:52):

afonso360 updated PR #4460 from x86-scalar-fma to main.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 02 2022 at 17:14):

cfallin submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 02 2022 at 17:14):

cfallin created PR review comment:

This will become a compile error once my remove-Baldrdash PR merges (it's in-queue now), sorry!

view this post on Zulip Wasmtime GitHub notifications bot (Aug 02 2022 at 17:14):

cfallin submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 02 2022 at 17:24):

afonso360 submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 02 2022 at 17:24):

afonso360 created PR review comment:

I can remove it now, no problem!

view this post on Zulip Wasmtime GitHub notifications bot (Aug 02 2022 at 17:35):

afonso360 edited PR review comment.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 02 2022 at 20:24):

afonso360 updated PR #4460 from x86-scalar-fma to main.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 03 2022 at 07:01):

afonso360 updated PR #4460 from x86-scalar-fma to main.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 03 2022 at 09:16):

afonso360 requested cfallin for a review on PR #4460.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 03 2022 at 17:29):

cfallin submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 03 2022 at 17:29):

cfallin merged PR #4460.

Last updated: Dec 23 2024 at 13:07 UTC