[patch] cse.c: Speed up cse_reg_info maintanance - Part 3

[Mike Stump]

On Feb 1, 2005, at 3:41 PM, Andrew Pinski wrote:
> On Feb 1, 2005, at 6:35 PM, Mike Stump wrote:
>> On Feb 1, 2005, at 3:12 PM, Kazu Hirata wrote:
>>> Attached is a patch to speed up cse_reg_info maintainance by
>>> dramatically simplifying the memory allocation.
>>> Tested on i686-pc-linux-gnu.  OK to apply?
>> This is exactly the type of patch that kills performance on darwin.
>> I have a counter proposal, why not free and then xmalloc instead of 
>> xrealloc, that way, we save the copy, which I suppose is the real 
>> cost you are trying to save.
> Did you look at his patch

Yes.

> (or did you look at it backwards)?

No.

> He is proposing to remove the xrealloc and change it to a xfree.

Yes, I saw.

> This is what he meant by simplifying the memory allocation.

I know.

The fact remains, this is exactly the kinda of patch that kills 
performance on darwin.

What we would like is imagine allocations that look like this:

1
2
3
4
5
6
7
8
9
10

we want ln(N) calls to xmalloc, no reallocs, and not N calls to xmalloc.