llvm dev - Jun 2011 - [LLVMdev] Different Runtimes using clang-gcc and clang-opt-llc-gcc

Hello,

  I have trouble compiling programs using clang on Ubuntu-Natty
(getting some linker errors). So I generated a object file using "-c"
and then used gcc to generate the final executable. It worked fine.

I tried the following method also:

$ clang -O0 -c -flto foo.c
$ llc foo.o
$ gcc foo.o.s

It also worked. However the problem is I am getting different runtimes
using the two different methods. The executable generated using the
second method is running faster than the one generated using the first
method. I am confused as in both cases I am using -O0.

In general if I want measure runtimes which of the two methods is more accurate.

Thanks in advance.
-Suresh

PS: Is there any fix for the linker errors generated by clang on Natty.

Hi Suresh,
> $ clang -O0 -c -flto foo.c
> $ llc foo.o
^ llc defaults to -O2
> $ gcc foo.o.s
Ciao, Duncan.

On Jun 8, 2011, at 5:49 AM, Duncan Sands wrote:
> Hi Suresh,
> 
>> $ clang -O0 -c -flto foo.c
>> $ llc foo.o
> 
> ^ llc defaults to -O2
Yep. Exactly right. Specifically, the code generator and register allocator
behave differently at -O0, sacrificing generated code quality to get better
debugging information and compile-time performance. That said, running things
this way, most of the mid-level optimizations that would normally be applied at
-O2 still aren't as they're not part of llc.
> 
>> $ gcc foo.o.s
> 
> Ciao, Duncan.
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev