25 Feb
2006
25 Feb
'06
1:16 p.m.
bulat.ziganshin@gmail.com wrote:
...
as you can see here, gcc isn't unrolled the loop and retained all the stack<->register moves. why? the short answer is what gcc just don't know that data pointed by the Sp pointer is non-volatile - they can't be changed in middle of the program by some async computation or exception and don't required to be updated immediately. may be there is some way to tell gcc this? this would immediately SUBSTANTIALLY improve all ghc code generation
Would the C99 restrict keyword help in this situation?