Vladimir Pantelic vladoman@gmail.com writes:
On 06/24/2011 08:13 PM, Woodruff, Richard wrote:
This suggests to me that a simple drop-in of libjpeg-turbo might be actually easy to do, and that there is probably a significant performance benefit to be achieved. One thing to keep in mind is that this code still supports armv6, so we'd probably want to preserve that.
You have to be a little careful as ARM did drop some bad NEON errata in A9 which require care to work around. Some CPU versions are impacted while others are not, tanstaafl.
AFAIK ARM has a tool that can scan object code for "bad" neon sequences, maybe they are willing to make that generally available?
I asked them about the that tool, and they confirmed its existence, but expressed some reluctance about releasing it, mainly due to lack of internal procedures for doing so. If more people asked about it, perhaps they could be persuaded to release it.