[patch] DAG nodes IR ordering code re-factoring

[Chris Lattner]

On May 23, 2013, at 3:40 PM, Andrew Trick <atrick at apple.com> wrote:

> On May 21, 2013, at 3:58 PM, Chris Lattner <clattner at apple.com> wrote:
>> On May 21, 2013, at 11:43 AM, "Guo, Xiaoyi" <Xiaoyi.Guo at amd.com> wrote:
>> 
>>> Hi Tom,
>>> 
>>> Here's the email thread where I explained my motivation and responses from other people.
>>> http://lists.cs.uiuc.edu/pipermail/llvmdev/2013-April/061728.html
>> 
>> I'm not thrilled about making SDNode larger.  Have you quantified the memory impact of this change?
> 
> The IROrder field fits in padding just before debugLoc. I verified that the bumpPtrAllocator stats don't change when the field is placed optimally.

Ok.

> Incidentally, we already allocate 256 bytes for every SDNode! That's huge compared to IR values and MI instructions. Most of that space is unused for common SDNode types.

That is crazy!

-Chris

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130525/53bd11ac/attachment.html>