1525664 - Consider using a 32-bit field for BigInt length

Status: RESOLVED FIXED

(Core :: JavaScript Engine, enhancement, P2)

Description

[Robin Templeton [:terpri]]

Currently, the digit length of a BigInt value is stored in the first word of its cell, along with a sign bit and bits reserved by GC. Should it be stored instead in a fixed 32-bit field, similar to JSStrings? This would reduce the space left over for inline digits on some platforms, but might simplify JIT integration since code for strings could be reused.

Comment 1

[Andy Wingo [:wingo]]

Attachment: Bug 1525664 - BigInt has 32-bit digitLength field

Comment 2

[Andy Wingo [:wingo]]

To be clear the tradeoff here is that for 32-bit x86, arm32, and mips32, we go from 96 bits of inline storage -- meaning that we can store values in (-2^96, 2^96) without having to malloc external storage -- to 64 bits of inline storage. But probably most values are either in the 64-bit range or rather large; and if the just-over-64-bit space becomes important, we'd need to fix for 64-bit targets as well.

On 64-bit platforms (mips64, arm64, x86-64), we were already limited to 64 bits of inline storage. With this change the maximum possible bigint length decreases from a theoretical 2^64-1 digits, each digit being 64 bits, to 2^32-1 digits. But that's already like 32 GByte for a single value, so no big deal there.

What we gain is easier access to the length field, which makes JIT a bit easier.

Comment 3

[Jan de Mooij [:jandem]]

Attachment: Bug 1525664 part 1 - Move length/flags fields from JSString to a new CellWithLengthAndFlags class. r?tcampbell!

The next part will change BigInt to use the same base class.

Comment 4

[Jan de Mooij [:jandem]]

Attachment: Bug 1525664 part 2 - Change BigInt to use CellWithLengthAndFlags. r?tcampbell!,wingo!

Depends on D39318

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/5055a6b65938
part 1 - Move length/flags fields from JSString to a new CellWithLengthAndFlags class. r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/6aeaedaaa1af
part 2 - Change BigInt to use CellWithLengthAndFlags. r=tcampbell,wingo

Comment 6

[Cosmin Sabou [:CosminS]]

https://hg.mozilla.org/mozilla-central/rev/5055a6b65938
https://hg.mozilla.org/mozilla-central/rev/6aeaedaaa1af

Comment 7

[Jan de Mooij [:jandem]]

Attachment: Bug 1525664 follow-up - Fix GDB pretty printer for JSString. r?jonco!

I had to use try-except instead of "in" because if the value is a pointer type
(like JSString*) then "in" returned false.

I tested this with both 32-bit and 64-bit builds.

Comment 9

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/c44d035213c1
follow-up - Fix GDB pretty printer for JSString. r=jonco

Comment 10

[Razvan Maries]

https://hg.mozilla.org/mozilla-central/rev/c44d035213c1