[LLVMdev] help: bypass a type of loop optimization in LLVM front end

[hilbert Wang]

Tim and guys,

let's look at the  following example:
============
unsigned char f;
void main()
{
    for(unsigned char i = 0; i < f;i ++)
        asm("nop");
}
================

the optimized LLVM code was as follows:

====================

1.define void @main() nounwind {

2.entry:

3. %0 = load i8* @f, align 1, !tbaa !0

4.  %cmp4 = icmp eq i8 %0, 0

5.  br i1 %cmp4, label %for.end, label %for.body

6

7.for.body:                                         ; preds = %entry,
%for.body

8.  %i.05 = phi i8 [ %inc, %for.body ], [ 0, %entry ]

9. tail call void asm sideeffect "nop", ""() nounwind, !srcloc !2

10.  %inc = add i8 %i.05, 1

11.  %1 = load i8* @f, align 1, !tbaa !0

12. %cmp = icmp ult i8 %inc, %1

13.  br i1 %cmp, label %for.body, label %for.end

14.

15.for.end:                                          ; preds = %for.body,
%entry

16. ret void

17.}

===============


>From line 3 to 5, whether f is equal to 0 is checked first.


because the profit of the optimization did not make sense for the target.

why it did not make sense in our target was out the scope of the post.


I was asking for help, can I prevent this optimization in command line or
which function  i need to work on to skip such optimization in front end?




thanks,


cheers,


hilbert
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140426/41e432d7/attachment.html>