GitLab

compute max count from buffer size for bitmap shader
zero-init paint in constructor to account for padding so that our == works correctly using memcmp

yields ~10% overall speedup
also, refresh misc fixes in freetype and antipath from trunk

remove remaining special-case neon functions that are (no longer) faster than the portable ones

* changes:
  enable arm flags in makefile and by including cpu-features.h

original code submitted by ARM LIMITED 2009

minor refresh on freetype and blitrow.h from skia/trunk

vector out to those from our blitrow factory

Merge commit 'fda848cd'

* commit 'fda848cd':
  only upscale to 565 from palette if the colortable is opaque

* changes:
  push upscaling directly into png, rather than having it in the baseclass, where it can re-invoke an allocator, which breaks ashmem (and possibly other allocator subclasses)

re-invoke an allocator, which breaks ashmem (and possibly other allocator subclasses)

am 421adab1: (-s ours) Do not merge support upscaling directly in png codec to preferred config remove copyTo post-upscaling in base imagedecoder, since that breaks our Allocator model for custom allocators (like ashmem), which can\'t be called twice on two different configs.

Merge commit '421adab1'

* commit '421adab1':
  Do not merge

support upscaling directly in png codec to preferred config
remove copyTo post-upscaling in base imagedecoder, since that breaks our Allocator model for custom
allocators (like ashmem), which can't be called twice on two different configs.

* changes:
  refresh from trunk check-point for lcd text rendering

check-point for lcd text rendering

fix degenerate/NaN cases for bad cubics

Merge commit '0ac4d481'

* commit '0ac4d481':
  Do not merge

check if we can promote to the preferred config

first stage of neon optimizations

am da7fbe20: If there is no sampling request in the decode request then use the width and height infomation from the header directly.

Merge commit 'da7fbe20'

* commit 'da7fbe20':
  If there is no sampling request in the decode request then use

the width and height infomation from the header directly.

use expand16 format to speed up blend
fix typo bug in loops for 4444

Merge commit 'ee38d710'

* commit 'ee38d710':
  catch underflow in restore()