Re: [patch x64 SEH]: Support xdata/pdata for link-once code

[Kai Tietz <ktietz70 at googlemail dot com>]

2010/9/14 Kai Tietz <ktietz70@googlemail.com>:
> 2010/9/14 Richard Henderson <rth@redhat.com>:
>> On 09/14/2010 10:06 AM, Kai Tietz wrote:
>>> Hmm, is there such a mass of different code-segments to be expected
>>> within one file that we will see here a real gain using hash-tables?
>>
>> With a large c++ file, with either -ffunction-sections or templates
>> (which are going to use link-once sections), you may well have on
>> the order of tens of thousands of code segments. ?And each one of
>> those code segments will be used exactly once.
>>
>> Each will result in a failing search through the linked list. ?Better
>> to fail earlier if possible, either through C1*O(log N) tree or C2*O(1)
>> hash table. ?I don't recall the relative magnitude of C1 and C2 in this
>> case, so I'd choose the one that's easier to use.
>>
>> Of course, there's the other common case of all functions in .text.
>> If you use the hash table, it might be worth-while to not store the
>> standard text section in the table, but in a separate variable.
>
> Yeah, I see. This was the reason why I add new section always to first
> element. So in the case of -ffunction-sections for seek. The issue is
> the initial find for an new section. Ok, I'll check what here the
> easiest algorithm will be.
>
> Kai
>
>
> --
> |? (\_/) This is Bunny. Copy and paste
> | (='.'=) Bunny into your signature to help
> | (")_(") him gain world domination
>

Reworked patch using hash table.

Tested for x86_64-w64-mingw32. Ok for apply?

Regards,
Kai

-- 
|? (\_/) This is Bunny. Copy and paste
| (='.'=) Bunny into your signature to help
| (")_(") him gain world domination

Attachment: discardable_seh.diff
Description: Binary data