The garbage collector code is iterating the mSharedGarbageList from
begin to end and if the garbage object is completed, we destroy it and
erase it from the vector. This is done one by one. The problem with that
is each erase call end up moving the remaining vector to ahead by one,
and this process is repeated for each completed element. This is O(N^2)
problem and really magnifies when you have many garbage objects. Since
we only add garbage when it is GPU pending, so most of these garbages
are in sort of fifo order anyway. This CL changes the garbage list from
vector to queue so that erase becomes pop and this becomes a O(N)
problem. This reduces CPU overhead as measured with minimize-gpu-work
for gardenscapes from 10ms to 2ms due to suballocation generates lots of
garbages. But this CL is a general improvement for the garbage
collection performance.

Bug: b/215768827
Change-Id: I002aadf75f958d8b79eb6281a2608597776e908d
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3414354


Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>