https://llvm.org/bugs/show_bug.cgi?id=26636

            Bug ID: 26636
           Summary: [SimplifyCFG] branch weights are not preserved when
                    forming select
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Transformation Utilities
          Assignee: unassignedb...@nondot.org
          Reporter: spatel+l...@rotateright.com
                CC: llvm-bugs@lists.llvm.org
    Classification: Unclassified

The conditional branch has profile weight metadata:

target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.11.0"

define i32 @foo(i32 %x) #0 {
entry:
  %cmp = icmp sgt i32 %x, 4
  br i1 %cmp, label %if.then, label %if.end, !prof !0

if.then:
  %add = add nsw i32 %x, 7
  br label %return

if.end:
  %mul = mul nsw i32 %x, 17
  br label %return

return:
  %retval.0 = phi i32 [ %add, %if.then ], [ %mul, %if.end ]
  ret i32 %retval.0
}

!0 = !{!"branch_weights", i32 676, i32 207}

-------------------------------------------------------------------------------

But after -simplifycfg, the metadata is gone:

$ ./opt -simplifycfg selbrweight.ll -S
; ModuleID = 'selbrweight.ll'
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.11.0"

define i32 @foo(i32 %x) {
entry:
  %cmp = icmp sgt i32 %x, 4
  %add = add nsw i32 %x, 7
  %mul = mul nsw i32 %x, 17
  %retval.0 = select i1 %cmp, i32 %add, i32 %mul
  ret i32 %retval.0
}

-------------------------------------------------------------------------------

I don't know if weight metadata can currently be applied to selects, but the
select instruction should propagate the weight info that was on the branch.

This was noted by Benjamin Kramer in:
http://reviews.llvm.org/D16836

-- 
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs

Reply via email to