bjope added a comment.
I tried running
/clang -cc1 -O3 -funroll-loops -S -emit-llvm pragma-do-while-unroll.cpp -o -
-mllvm -print-after-all
and I get this
...
!2 = distinct !{!2, !3}
!3 = !{!"llvm.loop.unroll.count", i32 3}
!4 = !{!5, !5, i64 0}
!5 = !{!"int", !6, i64 0}
!6 = !{!"omnipotent char", !7, i64 0}
!7 = !{!"Simple C/C++ TBAA"}
!8 = distinct !{!8, !9}
!9 = !{!"llvm.loop.unroll.count", i32 5}
*** IR Dump After Combine redundant instructions ***
; Function Attrs: nounwind
define i32 @test(i32* %a, i32 %n) local_unnamed_addr #0 {
entry:
br label %do.body, !llvm.loop !2
do.body: ; preds = %do.body, %entry
%i.0 = phi i32 [ 0, %entry ], [ %inc, %do.body ]
%sum.0 = phi i32 [ 0, %entry ], [ %add5, %do.body ]
%0 = zext i32 %i.0 to i64
%arrayidx = getelementptr inbounds i32, i32* %a, i64 %0
%1 = load i32, i32* %arrayidx, align 4, !tbaa !4
%add = add nsw i32 %1, 1
store i32 %add, i32* %arrayidx, align 4, !tbaa !4
%add5 = add nsw i32 %sum.0, %add
%inc = add nuw nsw i32 %i.0, 1
%cmp = icmp slt i32 %inc, %n
br i1 %cmp, label %do.body, label %do.end, !llvm.loop !2
do.end: ; preds = %do.body
br label %do.body6, !llvm.loop !8
do.body6: ; preds = %do.body6, %do.end
%i.1 = phi i32 [ 0, %do.end ], [ %inc15, %do.body6 ]
%sum.1 = phi i32 [ %add5, %do.end ], [ %add14, %do.body6 ]
%2 = zext i32 %i.1 to i64
%arrayidx8 = getelementptr inbounds i32, i32* %a, i64 %2
%3 = load i32, i32* %arrayidx8, align 4, !tbaa !4
%add9 = add nsw i32 %3, 1
store i32 %add9, i32* %arrayidx8, align 4, !tbaa !4
%add14 = add nsw i32 %sum.1, %add9
%inc15 = add nuw nsw i32 %i.1, 1
%cmp17 = icmp slt i32 %inc15, %n
br i1 %cmp17, label %do.body6, label %do.end18, !llvm.loop !8
do.end18: ; preds = %do.body6
ret i32 %add14
}
*** IR Dump After Simplify the CFG ***
; Function Attrs: nounwind
define i32 @test(i32* %a, i32 %n) local_unnamed_addr #0 {
entry:
br label %do.body, !llvm.loop !2
do.body: ; preds = %do.body, %entry
%i.0 = phi i32 [ 0, %entry ], [ %inc, %do.body ]
%sum.0 = phi i32 [ 0, %entry ], [ %add5, %do.body ]
%0 = zext i32 %i.0 to i64
%arrayidx = getelementptr inbounds i32, i32* %a, i64 %0
%1 = load i32, i32* %arrayidx, align 4, !tbaa !4
%add = add nsw i32 %1, 1
store i32 %add, i32* %arrayidx, align 4, !tbaa !4
%add5 = add nsw i32 %sum.0, %add
%inc = add nuw nsw i32 %i.0, 1
%cmp = icmp slt i32 %inc, %n
br i1 %cmp, label %do.body, label %do.body6, !llvm.loop !8
do.body6: ; preds = %do.body,
%do.body6
%i.1 = phi i32 [ %inc15, %do.body6 ], [ 0, %do.body ]
%sum.1 = phi i32 [ %add14, %do.body6 ], [ %add5, %do.body ]
%2 = zext i32 %i.1 to i64
%arrayidx8 = getelementptr inbounds i32, i32* %a, i64 %2
%3 = load i32, i32* %arrayidx8, align 4, !tbaa !4
%add9 = add nsw i32 %3, 1
store i32 %add9, i32* %arrayidx8, align 4, !tbaa !4
%add14 = add nsw i32 %sum.1, %add9
%inc15 = add nuw nsw i32 %i.1, 1
%cmp17 = icmp slt i32 %inc15, %n
br i1 %cmp17, label %do.body6, label %do.end18, !llvm.loop !8
do.end18: ; preds = %do.body6
ret i32 %add14
}
...
So up until simplifyCFG I think things look pretty good with different
loop-metadata for the two loops. But when simplifyCFG is tranforming
br i1 %cmp, label %do.body, label %do.end, !llvm.loop !2
do.end: ; preds = %do.body
br label %do.body6, !llvm.loop !8
into
br i1 %cmp, label %do.body, label %do.body6, !llvm.loop !8
we get incorrect metadata for one branch target.
Is the fault that the metadata only should be put on the back edge, not the
branch in the preheader?
Repository:
rC Clang
https://reviews.llvm.org/D48721
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
