[LLVMbugs] [Bug 4939] New: incorrect ARM code for SETOLE floating-point comparisons

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=4939

           Summary: incorrect ARM code for SETOLE floating-point comparisons
           Product: libraries
           Version: trunk
          Platform: Macintosh
        OS/Version: All
            Status: NEW
          Severity: normal
          Priority: P2
         Component: Backend: ARM
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: bob.wilson at apple.com
                CC: llvmbugs at cs.uiuc.edu


The FPCCToARMCC function in ARMISelLowering.cpp handles SETOLE comparisons by
translating to ARMCC::GT and indicating that "...the operands should be
inverted".  That comment is ambiguous and the code using this function appears
to interpret it in two different ways.  For SELECT_CC, the "true" and "false"
operands are swapped when FPCCToARMCC returns true.  For BR_CC, the values
being compared are swapped.  Unfortunately, neither interpretation is correct!

If the values being compared are swapped, SETOLE should be translated to SETOGE
or ARMCC::GE.

If the SELECT_CC result "true" and "false" result values are swapped, SETOLE
should be translated to SETUGT or ARMCC::HI.

Unless I'm missing something here, there's no reason to swap anything, since
SETOLE for floating-point values can be directly translated to ARMCC::LS.

This problem was exposed by a failure of the SPEC 450.soplex test.


-- 
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.