This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] x86: Disable jump tables when retpolines are used (PR target/86952).


On 3/8/19 2:58 PM, Jan Hubicka wrote:
>> On 3/8/19 1:44 PM, Jan Hubicka wrote:
>>>> Hi.
>>>>
>>>> Thanks to Intel guys, we've done some re-measurement in PR86952
>>>> about usage of jump tables when retpolines are used.
>>>> Numbers prove that disabling of JT should be the best for now.
>>>>
>>>> Patch can bootstrap on x86_64-linux-gnu and survives regression tests.
>>>>
>>>> Ready to be installed?
>>>
>>> OK, thanks!
>>> I wonder if there is some threshold for extremely large jumptables where
>>> branchy sequence will loose, but I think it is better to disable them
>>> than what we have right now.
>>
>> I tested switch statements up to 4096 and it was still slower ;)
> 
> Well, we have switch statements with this many cases in
> insn-attrtab/latencytab tables.  I suppose what kind of code you compile
> with retpolines, but I would expect even bigger switch statements to
> appear in real code (large DFA automatons and such)

Question is whether you'll meet such huge switch statements in Linux kernel?

Martin

> 
> Honza
>>
>> Martin
>>
>>>
>>> Honza
>>>
>>


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]