Three functions (hMin(), hMax() and CalculateLuma()) were implemented with use of NEON intrinsics in order to improve performance of the computations.
Results from odroid-n2 (Mpx/s):
ST
MT
HST
HMT
ETC1
45.098
201.383
45.080
200.505
ETC2
41.162
179.972
60.623
318.686
ETC2 RGBA
26.102
140.651
33.674
172.449
ST - single thread
MT - multi thread
H - heuristic
General diff between images compressed using new and old compression mode selection mechanism (results obtained on ARM based hardware):
Detailed comparison of image kodim23.png. Image on the left (new) was compressed using new compression mode selector implemented with NEON intrinsics, while image on the right (old) was compressed with heuristic selector disabled (--disable-heuristics):
Description
Three functions (hMin(), hMax() and CalculateLuma()) were implemented with use of NEON intrinsics in order to improve performance of the computations.
Results from odroid-n2 (Mpx/s):
ST - single thread
MT - multi thread
H - heuristic
General diff between images compressed using new and old compression mode selection mechanism (results obtained on ARM based hardware):
Detailed comparison of image kodim23.png. Image on the left (new) was compressed using new compression mode selector implemented with NEON intrinsics, while image on the right (old) was compressed with heuristic selector disabled (--disable-heuristics):