Closed stoklund closed 8 years ago
lgtm
@PeterJensen @sunfishcode
The changes to aobench.js are fairly large, so I just wanted to make sure that you verified that the test ran OK with the changes.
Thanks for doing this. Much appreciated!
Thanks, Peter!
I'll update the commits with the changes you suggest.
Regarding the changes to aobench.js: The updated test runs without issue on Firefox Nightly. It looks like initAobench()
does a basic sanity check of the computations by comparing the vector and scalar versions.
Do you think we need more verification than that?
Thanks Jacob,
The sanity check should be sufficient, I think. I'll merge the PR after you do the updates
Pushed updates patch series.
In sinx4.js: Use a separate variable for the Bool32x4 mask. In aobench.js: Use splat(0.0) and splat(1.0) for Float32x4 immediate splats.
With these changes, the benchmarks are running in Firefox Nightly.
Update the benchmarks to expect boolean vector results from comparisons.