Why do ARM chips have an instruction with Javascript in the name (FJCVTZS)?

156
votes

FJCVTZS is "Floating-point Javascript Convert to Signed fixed-point, rounding toward Zero". It is supported in Arm v8.3-A chips and later. Which is odd, because you don't expect to see JavaScript so close to the bare metal.

I can find explanations of what the instruction does, but not why it exists. This thread says "it exists as a single instruction is because JS's lack of an integer type means certain use cases need this operation obscenely often for no good algorithmic reason.". That's plausible but I would like a more detailed understanding.

1
javascriptassemblyarm
Supporting that quote: The JavaScript engine has to do this operation (which is called ToInt32 in the spec) whenver you apply a bitwise operator to a number and at various other times (unless the engine has been able to maintain the number as an integer as an optimization, but in many cases it cannot). – T.J. Crowder
Note that the crucial difference to the usual FCVTZS instruction provided for this job is that FJCVTZS has a different behaviour on overflow. Namely, you always get the least 32 bits whereas FCVTZS seems to do something different instead if the number doesn't fit. The correct behaviour of FJCVTZS seems to be slightly tricky to implement otherwise. – fuz
@Tim I agree with TJ, minutes of the meeting are off-topic (and I suppose they're also restricted then hardly you can have them) but to run JavaScript on an ARM-powered device IS a thing, see Building IoT devices with JavaScript – Adriano Repetti
@TimSmith Thank you for your cooperation. Let's see if we can get this question reopened. – fuz

1 Answers

140
votes

It is because JS uses double precision for the numbers, but if you want to perform operations with bits, the task is nontrivial, so a specific instruction to convert JS double into integer makes the thing easier.

This ARM link explains it very well: https://community.arm.com/processors/b/blog/posts/armv8-a-architecture-2016-additions

In order to add more information regarding fuz's comment, the differences between FCVTZS and FJCVTZS (both of them convert floating point to int) are that in case of overflow, FJCVTZS value will be 0x80000000 instead of overflowing. Furthermore, FJCVTZS can generate an exception in order to indicate how the conversion was (i.e. inexact).

FJCVTZS : http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0801g/hko1477562192868.html

FCVTZS : http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0802a/FCVTZS_float_int.html