[LLVMbugs] [Bug 5284] New: llvm.bswap.i{16, 32} is not constant folded

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=5284

           Summary: llvm.bswap.i{16,32} is not constant folded
           Product: new-bugs
           Version: unspecified
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: new bugs
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: edwintorok at gmail.com
                CC: llvmbugs at cs.uiuc.edu


In the testcase the llvm.bswap.i{16,32} calls should be constant-folded:

%call.i.i = tail call zeroext i1 @is_bigendian() nounwind readnone ; <i1>
[#uses=2]
%call1.i = select i1 %call.i.i, i32 83886080, i32 5 ; <i32> [#uses=5]
..
br i1 %call.i.i, label %cond.true.i.i, label %cli_readint16.exit.thread

cond.true.i.i:                                    ; preds = %entry
  %trunc = trunc i32 %call1.i to i16              ; <i16> [#uses=1]
  %or.i.i = tail call i16 @llvm.bswap.i16(i16 %trunc) nounwind ; <i16>
[#uses=1]
  %conv = zext i16 %or.i.i to i32                 ; <i32> [#uses=1]
  %tmp1.i.i = tail call i32 @llvm.bswap.i32(i32 %call1.i) nounwind ; <i32>
[#uses=1]
  br label %cli_readint32.exit

^
Instcombine should realise that on cond.true.i.i the value of cal1.i is i32
83886080, and constant propagate that.

Also in the attached testcase the bswaps can be completely eliminated, since
br i1 %call.i.i, label %cond.true.i.i10, label %cli_writeint32.exit.thread

%cond.true.i.i10 unconditionally ranches to cli_readint16.exit, which has:
%a.val2 = phi i32 [ %tmp1.i.i, %cond.true.i.i10 ], [ %v,
%cli_writeint32.exit.thread ] ; <i32> [#uses=2]
....
br i1 %call.i.i, label %cond.true.i.i7, label %cli_readint32.exit

cond.true.i.i7:                                   ; preds = %cli_readint16.exit
  %tmp1.i.i6 = tail call i32 @llvm.bswap.i32(i32 %a.val2) nounwind ; <i32>
[#uses=1]
  br label %cli_readint32.exit


Now cond.true.i.i7 is reached if %call.i.i is true, which means that a PHI
translation should tell us that %a.val2 is actuall %tmp1.i.i in that case, so
we could propagate:
  %tmp1.i.i6 = tail call i32 @llvm.bswap.i32(i32 %tmp1.i.i) nounwind ; <i32>
[#uses=1]

And this is an identity transform bswap(bswap(%v)) == %v, so we should have:
%tmp1.i.i16 = %v

Similarly this PHI:
%call48 = phi i32 [ %tmp1.i.i6, %cond.true.i.i7 ], [ %a.val2,
%cli_readint16.exit ] 

should be simplified to
%call48 = %v (Since %a.val2 is %v on the other branch)


-- 
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.