https://community.arm.com/processors/b/blog/posts/armv8-a-architecture-2016-additions http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0801g/hko1477562192868.html FJCVTZS instruction to convert floating points to 32 bit integers for JS use.

An Apple developer says he plans to add support for FJCVTZS in JavaScriptCore this week: https://bugs.webkit.org/show_bug.cgi?id=184023

At least plausibly we could also use this in implementing ToUnsignedInteger/ToSignedInteger in js/public/Conversions.h. We already have an ARM-specific definition for ToSignedInteger<int32_t>, and we could perhaps use it more. We could also do the same for mfbt/FloatingPoint.h's NumberIsInt32 and NumberEqualsInt32, with some care for -0 in the latter case. Making use of this instruction in any of those cases -- in code not generated at runtime -- would require some care, given we presumably want to support ARM versions that don't have this instruction. So it might not be worth it. But we should at least consider the possibility if it could be used reasonably efficiently somehow.

I'm aware of this. Our existing VIXL implementation doesn't support encoding it yet, so it will require upgrading to trunk (or backporting, but I'd rather upgrade).