[llvm-dev] RFC: We should stop merging allocas in the inliner

[Bruce Hoult via llvm-dev]

On Wed, Aug 3, 2016 at 3:43 AM, Chandler Carruth via llvm-dev <
llvm-dev at lists.llvm.org> wrote:

> On Tue, Aug 2, 2016 at 5:39 PM David Majnemer via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
>> On Tue, Aug 2, 2016 at 5:33 PM, Chandler Carruth via llvm-dev <
>> llvm-dev at lists.llvm.org> wrote:
>>
>>> FYI, I ran some initial measurements just for sanity on the test-suite
>>> on x86-64:
>>>
>>> Total stack frame size: 13362520 -> 13365864 (+0.02%)
>>> Average stack frame size: 581.814 -> 581.96 (+0.02%)
>>> Average compile time: 5.71178 -> 5.67256 (within noise)
>>>
>>
>> Could you share the median stack frame sizes?
>>
> 24 bytes
>

 That's about what I would expect -- maybe even smaller than I'd expect
with 64 bit ints/ptrs.

But the average is 582 bytes? That's pretty surprising. There must be some
really huge frames! Is that because of locally allocated speed-optimized
classes with big embedded arrays/buffers?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20160803/a39796f6/attachment-0001.html>