Rebaseline code size tests

[kripken]

Ran into the need for this because a reversion PR (#15460) fails unexpectedly.

[kripken]

@sbc100 This fails even though I just updated the numbers using emsdk install tot. And I verified the tot has not changed since I created the PR - emsdk install tot locally has nothing new to download. Do I need to do anything else to update these?

[sbc100]

@sbc100 This fails even though I just updated the numbers using emsdk install tot. And I verified the tot has not changed since I created the PR - emsdk install tot locally has nothing new to download. Do I need to do anything else to update these?

No, I think that should work...

[kripken]

Very strange...

I guess I can update this file manually to the expectation. But I'm worried as that suggests different codegen is happening on my machine versus the bots.

If you have time can you see if rebaselining locally for you agrees with me or with the bots?

[sbc100]

Very strange...

I guess I can update this file manually to the expectation. But I'm worried as that suggests different codegen is happening on my machine versus the bots.

If you have time can you see if rebaselining locally for you agrees with me or with the bots?

Sure.

My process is normally to build llvm and binaryen from source at ToT before doing the rebaselining, but I think emsdk tot should work just as well.

[kripken]

I've tried building locally, I've tried verifying I get the exact same emsdk hash as CI, but I still get the same difference...

Anyhow, there's clearly something wrong on either my machine or on CI. When you get a chance to compare @sbc100 let's see which results your machine agrees with.

[kripken]

It seems like the problem might be that I had newest emscripten+llvm+binaryen, but had not manually run npm install, so I was on an older closure. Hopefully that fixes it...