[LLVMdev] Stack usage analysis using LLVM

[Jim Grosbach]

On Sep 5, 2013, at 12:04 PM, John Criswell <criswell at illinois.edu> wrote:

> On 9/3/13 1:27 PM, Snehasish Kumar wrote:
>> Hi
>> 
>> I was wondering if someone knows about any effort within the LLVM community to perform stack usage analysis per function similar to GCC's "-fstack-usage" option?
> 
> I am not familiar with the -fstack-usage option in GCC, but as far as I know, LLVM does not have a FunctionPass or MachineFunctionPass which performs the calculations described below.  You could, of course, quickly test this by running the clang -fstack-usage and see what happens.  LLVM and the DragonEgg plugin for GCC is more likely to support the feature.
> 
> That said, writing such as pass would be very easy to do.  If you're only concerned about the size of stack-allocated objects, you can write an LLVM pass that looks for alloca instructions, determine that they are not in loops, and then sums up the sizes of the allocatoed memory using the DataLayout pass.
> 
> If you want something that includes stack-spill slots and the like, then you'd need to write a MachineFunction Pass and examine the generated machine instructions.  Alternatively, there might be a way in a MachineFunctionPass to get a pointer to a MachineFrame object and to query its size.
> 

While not the same, you may find r183595 and the surrounding discussion useful, as it's very related.

-Jim


> -- John T.
> 
>> 
>> In short, with fstack-usage, gcc prints out the maximum stack usage per function (in bytes) which it can determine as a) static (no calls to alloca in source) b) bounded (calls to alloca with constants) c) unbounded (calls to alloca with           variables). A more detailed description can be found in this pdf.
>> 
>> 
>> Thanks,
>> Snehasish Kumar
>> 
>> 
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130906/ebf83445/attachment.html>