Rebase. Regenerate with a hacked script to show the real stack offsets for reviewing. I'll regenerate before commiting so the next person who runs the script won't get a surprise.

In D72582#1826859, @RKSimon wrote:

Is there a bz for this?

In D72906#1826061, @uweigand wrote:

The constrained fcmp intrinsics don't allow the TRUE/FALSE predicates.

Hmm, maybe they should then? The only reason I didn't add them initially was that I wasn't sure they were useful for anything; if they are, it should be straightforward to add them back.

I was able to move the other version of this code from TargetLowering back to LegalizeDAG. So just need to add the new type to it.

The title of this review is misleading. It should at least mention FPEnv, constrained intrinsics, or strict fp or something. Right now it sounds like FP_CONTRACT isn't supported at all.

LGTM

LGTM

In D72665#1818592, @pengfei wrote:

Yeah, it's useful and clear to use a STRICT_FP_EXTEND. Anyway, I think we can also achieve it by getting the function's attribute or propagating the fpexcept to FP_EXTEND.

This FP_EXTEND will be turned into a SSE store and an X87 load. That X87 load may generate an exception so we should have the fpexcept property on it. That requires a STRICT_FP_EXTEND.

Committed a version of this in ddfcd82bdc219dd2dc04d6826c417cea3da65d12.

LGTM

LGTM

LGTM

Rebase after the uint_to_fp v4i64->v4f32 custom type legalization change D72368.

Add a LowerOperationWrapper helper that updates a Result vector. It looks a lot like X86's override of TargetLowering::LowerOperationWrapper except the bool result. The TargetLowering version doesn't allow for the possibility of keeping the existing node since it only called by the type legalizer today. I think in the future we could try to merge this together, but it will require changes to other targets.

LGTM

LGTM