601716 - Should we add Vector<Boolean> specialization?

Status: RESOLVED WONTFIX

(Tamarin Graveyard :: Virtual Machine, enhancement)

Description

[Steven Johnson]

Currently we implement Vector<Boolean> as Vector<*>, which uses 32 bits per entry. A simple specialization could have 1 byte per entry trivially (1 bit per entry with a bit more work). Not sure if this is used often enough to be worth specializing though.

Comment 1

[Werner Sharp]

Attachment: brightspot dump of vector names

Of 16k SWFs, 71 instances of Vector.<Boolean>

Comment 2

[Lars T Hansen]

On 64-bit systems we would use 64 bits per entry for Vector.<Boolean> I think (Atom representation, ie, uintptr_t).  And it's not just the size; atom writes go through the slowest write path (atomWriteBarrier).  Even after optimizations (see eg 601817 and earlier) that's about 10 times slower than writing to Vector.<int>.

A quick fix which could be a significant improvement might be to specialize for Vector.<Boolean> but to map it onto the Vector.<int> representation.

Not high priority given the scant use of Vector.<Boolean> probably.