DescriptionOilpan: Scale up the number of marking threads according to available processors
Currently we use 2 marking threads regardless of the number of processors available.
Given that the marking phase takes a substantial amount of time,
this CL changes the number of marking threads to min(the number of processors, 4).
This CL also decreases the block size of each CallStack from 200 to 128
because load balancing is not good in some benchmarks such as LargeDistributionWithoutLayout. This CL improves LargeDistributionWithoutLayout by 10% for example.
BUG=420515
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=183613
Patch Set 1 #Patch Set 2 : #Patch Set 3 : #Patch Set 4 : #Patch Set 5 : #Patch Set 6 : #
Messages
Total messages: 8 (3 generated)
|