[LLVMbugs] [Bug 708] NEW: Instcombine slow down

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.cs.uiuc.edu/bugs/show_bug.cgi?id=708

           Summary: Instcombine slow down
           Product: libraries
           Version: trunk
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: Backend: Alpha
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: alenhar2 at uiuc.edu


ulong %foo(ulong %y) {
entry:
        %tmp = and ulong %y,  65535
        %tmp2 = shr ulong %tmp,  ubyte 3
        ret ulong %tmp2
}


compiles to a 3 instruction sequence without instcombine
        zapnot $16,3,$0
        srl $0,3,$0
        ret $31,($26),1
 
After instcombine you get
ulong %foo(ulong %y) {
entry:
        %tmp = shr ulong %y, ubyte 3            ; <ulong> [#uses=1]
        %tmp2 = and ulong %tmp, 8191            ; <ulong> [#uses=1]
        ret ulong %tmp2
}

which compiles to
        lda $0,8191($31)
        srl $16,3,$1
        and $1,$0,$0
        ret $31,($26),1


Larger constants will overflow lda and cause even more instructions to load the
constant.  Probably need some shift then mask patterns to reverse the order of
things like that.



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.