[webkit-reviews] review granted: [Bug 219546] Improve vectorization in SincResampler : [Attachment 415442] Patch

bugzilla-daemon at webkit.org bugzilla-daemon at webkit.org
Fri Dec 4 12:24:01 PST 2020

Darin Adler <darin at apple.com> has granted Chris Dumez <cdumez at apple.com>'s
request for review:
Bug 219546: Improve vectorization in SincResampler

Attachment 415442: Patch


--- Comment #3 from Darin Adler <darin at apple.com> ---
Comment on attachment 415442
  --> https://bugs.webkit.org/attachment.cgi?id=415442

View in context: https://bugs.webkit.org/attachment.cgi?id=415442&action=review

How did you test all four versions?

> Source/WebCore/platform/audio/SincResampler.cpp:348
> +    vDSP_dotpr(inputP, 1, k1, 1, &sum1, kernelSize);
> +    vDSP_dotpr(inputP, 1, k2, 1, &sum2, kernelSize);

Is there a reason that we don’t write a 4 different implementations of a "dot
product" function instead of having the algorithm differ a level above that?

> Source/WebCore/platform/audio/SincResampler.cpp:351
> +    return static_cast<float>((1.0 - kernelInterpolationFactor) * sum1 +
kernelInterpolationFactor * sum2);

Is it important do do this as double and then convert back to float? I notice
we do that in both versions.

More information about the webkit-reviews mailing list