Linked by Thom Holwerda on Tue 8th Jan 2013 23:27 UTC
Thread beginning with comment 548497
To view parent comment, click here.
To read all comments associated with this story, please click here.
To view parent comment, click here.
To read all comments associated with this story, please click here.
So we can finally agree that the direct translation approach has inefficiencies that cannot be avoided without optimising the code for ARM. Whew, that took a while!
You say efficiency isn't important to you if but if an optimiser could significantly increase active runtime of the CPU between charges and has no negative trade-offs, then it would be significant to a lot of (and even most) ARM users.
So we can finally agree that the direct translation approach has inefficiencies that cannot be avoided without optimising the code for ARM.
As I wrote, I'd prefer just to re-schedule translated code to fit issue slots. I didn't understand why you're mentioned flag registers overhead. So you probably thought about x86 flag register update after every operation? Sorry I didn't stated that explicitly, but translation engine should trace flags usage for basic blocks. This is a baseline requirement.
You say efficiency isn't important to you if but if an optimiser could significantly increase active runtime
I'm not sure that you'll achieve a _much better_ results with complex optimizations.




Member since:
2005-08-09
I'm not trying to stop you =)
"the best is the enemy of the good"
I'm still very skeptical about the overall experience but we'll have to wait and see for ourselves.
Phones are around for significant period of time. Compatibility and performance are good enough.
I mentioned Eltech in my second post.
Edited 2013-01-12 13:03 UTC