IMX6Q is x6 times slower on Android 8.0.0 than on Android 4.4.3!
I use MCIMX6Q-SDB sabresd board and tinymembench tool for memory benchmarking.
And here are the results:
Tinymembench Operation | Android 4.4.3 Linux kernel 3.10.53-88098, MB/s | Android 8.0.0 Linux kernel 4.9.17-04736, MB/s |
---|---|---|
C fill | 2121.2 | 353.8 |
standard memset | 2118.7 | 353.6 |
NEON fill | 2105.5 | 353.7 |
ARM fill (STRD) | 2120.9 | 353.9 |
As you can see only specific operations type are affected - "fill/set".
No such huge performance difference with other operation like copy-read:
Tinymembench Operation | Android 4.4.3 Linux kernel 3.10.53-88098, MB/s | Android 8.0.0 Linux kernel 4.9.17-04736, MB/s |
---|---|---|
C copy | 318.0 | 334.6 |
standard memcpy | 358.7 | 370.8 |
NEON copy | 300.5 | 310.8 |
ARM copy prefetched (incr pld) | 402.6 | 382.8 |
Other benchmark utils like Geekbench or Antutu show the same situation with memory performance: Android 8.0.0 is much slower than Android 4.4.3.
Why does Android 8.0.0 have much lower numbers than Android 4.4.3?
What should I do to bring Android 4.4.3 performance to the Android 8.0.0 (port some drivers patches... etc.) ?
Solved! Go to Solution.
may be useful
https://community.nxp.com/thread/371109
also performance issues and additional custom optimizations may be supported using
NXP Professional Services | NXP
Best regards
igor
Hi denkraev
issue may be caused by arm errata (in uboot/include/configs :
define CONFIG_ARM_ERRATA)
mx6_common.h\configs\include - uboot-imx - i.MX U-Boot
Also please check that correct image was used (from nxp official repository)
Best regards
igor
-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------
Official images are in use.
Errata is enabled.
may be useful
https://community.nxp.com/thread/371109
also performance issues and additional custom optimizations may be supported using
NXP Professional Services | NXP
Best regards
igor