Displaying 1 result from an estimated 1 matches for "squenece".
Did you mean:
squence
2017 May 19
4
memcmp code fragment
...then19
if.then19: ; preds = %if.end
%cmp22 = icmp ugt i8 %2, %3
br label %return
..
..
<repeats 12 times>
This code sequence can be collapsed into call to memcmp and we can get rid
of basic blocks. I have written a small peephole optimization for squenece
of instructions that identifies
branch termiantor, compare, load, gep etc and converts them to a call to
memcmp. This small pass gave me improvement of 67% on SPEC2000 bzip2 on X86.
Is there a better idea, other than small peephole pass on IR to optimize
this code?
Sirish
-------------- next part...