[llvm-bugs] [Bug 35616] New: -callsite-splitting gives different result if dbg.value is present
via llvm-bugs
llvm-bugs at lists.llvm.org
Mon Dec 11 01:00:39 PST 2017
https://bugs.llvm.org/show_bug.cgi?id=35616
Bug ID: 35616
Summary: -callsite-splitting gives different result if
dbg.value is present
Product: new-bugs
Version: unspecified
Hardware: PC
OS: Linux
Status: NEW
Severity: enhancement
Priority: P
Component: new bugs
Assignee: unassignedbugs at nondot.org
Reporter: mikael.holmen at ericsson.com
CC: llvm-bugs at lists.llvm.org
Created attachment 19533
--> https://bugs.llvm.org/attachment.cgi?id=19533&action=edit
reproducer
Doing
opt -S -o - foo2.ll -callsite-splitting
and
opt -S -o - foo2.ll -strip-debug -callsite-splitting
on
define internal i16 @bar(i16 %p1, i16 %p2) {
%_tmp3 = mul i16 %p2, %p1
ret i16 %_tmp3
}
define i16 @foo(i16 %in) {
br label %bb1
bb1:
%_tmp0 = icmp ne i16 %in, 0
br i1 %_tmp0, label %bb2, label %bb3
bb2:
br label %bb3
bb3:
%_tmp1 = phi i16 [ 0, %bb1 ], [ 1, %bb2 ]
%c = phi i16 [ 2, %bb1 ], [ 3, %bb2 ]
call void @llvm.dbg.value(metadata i16 %c, metadata !7, metadata
!DIExpression()), !dbg !8
%_tmp2 = call i16 @bar(i16 %_tmp1, i16 5)
ret i16 %_tmp2
}
; Function Attrs: nounwind readnone speculatable
declare void @llvm.dbg.value(metadata, metadata, metadata) #0
attributes #0 = { nounwind readnone speculatable }
!llvm.dbg.cu = !{!0}
!llvm.module.flags = !{!2, !3}
!llvm.ident = !{!4}
!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "My
Compiler")
!1 = !DIFile(filename: "foo.c", directory: "/bar")
!2 = !{i32 2, !"Dwarf Version", i32 4}
!3 = !{i32 2, !"Debug Info Version", i32 3}
!4 = !{!"My Compiler"}
!5 = !DIBasicType(name: "int", size: 16, encoding: DW_ATE_signed)
!6 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 4, unit:
!0)
!7 = !DILocalVariable(name: "c", scope: !6, line: 5, type: !5)
!8 = !DILocation(line: 5, column: 7, scope: !6)
gives different results. With the llvm.dbg.value present the optimization
doesn't trigger, but if we remove it it does.
CallSiteSplitting.cpp does a couple of calls to getFirstNonPHI(). I suppose
using getFirstNonPHIOrDbg() could help ignoring the llvm.dbg.value but I have
no idea if more changes are needed.
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20171211/f17867f3/attachment-0001.html>
More information about the llvm-bugs
mailing list