On 11/22/13 13:05, Steven Bosscher wrote:
On Fri, Nov 22, 2013 at 11:09 AM, Jeff Law wrote:
So we've long had an arbitrary restriction in the jump thread identification
code which made it very conservative when searching for threading
opportunities once it encountered a back edge.
You're adding a pretty complex piece of code, for something that's
been tricky in the past: Threading around loops makes control and data
flow harder to handle for later passes.
Is there any code you have in mind for which removing the conservatism
is a win? If not, then perhaps we should test first, whether your
patch doesn't hurt the loop and vect opts ability to perform their
job.
Or is this prep work for the FSA patch?
It was the last piece in the the FSA work. It should dramatically help
the coremark benchmark.
jeff
