[LLVMbugs] [Bug 4914] New: llvm-gcc doesn't do range analysis on induction variable

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=4914

           Summary: llvm-gcc doesn't do range analysis on induction variable
           Product: libraries
           Version: trunk
          Platform: PC
               URL: http://codepad.org/n6dGCqje
        OS/Version: All
            Status: NEW
          Severity: normal
          Priority: P2
         Component: Loop Optimizer
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: rjmccall at apple.com
                CC: llvmbugs at cs.uiuc.edu


This is a forwarded complaint by "bearophile" on LLVM IRC, who unfortunately
never files bugs.  I am willing to act as a stand-in in this case.

The code in the URL produces worse code on (an unspecified version of) llvm-gcc
than (an unspecified version of) gcc does;  assembly output from both is inline
in the test.

The difference seems to involve three things:

(1) gcc knows that i is never < 0, and so elides the first check when computing
sat(i).
(2) gcc knows that -i is always <= 0, and so infers that sat(-i) is always 0.
(3) gcc hoists a constant load outside of the loop.


-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.