RE: [libunwind] backtrace on EM64T

[Yeshurun, Meir]

Hi Arun,

Thanks for asking, and apologies for the late reply.

Yes, I added an ip -> dwarf_reg_state_t cache. It seems to work fine. I 
implemented it in a very ugly way, but it's good to know that it can be made to 
work without much effort. There was a big improvement in performance, but I 
didn't make any exact measurements yet.

The cache insertion and retrieval are done in the function

dwarf_find_save_locs (struct dwarf_cursor *c)

The only tricky part was to add

c->ret_addr_column = 16;

as the first statement in the function.

Regarding your patch, I didn't look at it. What is the purpose of this patch?


Regards,
Meir


-----Original Message-----
From: Arun Sharma [mailto:address@hidden 
Sent: Tuesday, December 06, 2005 11:43 PM
To: Yeshurun, Meir
Cc: David Mosberger-Tang
Subject: Re: [libunwind] backtrace on EM64T

Arun Sharma (अरुण) wrote:
> David Mosberger-Tang wrote:
> 
>> You really should be looking at the latest sources if you want to work
>> on x86-64---there were a number of changes there.
>>
> 
> [ Not sending the patch publicly because I might not be using the latest
>   sources. But wanted to share the patch anyway, for comments ]

Meir, were you able to make any progress on implementing the caching? 
What did you think about the patch?