This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc 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: [RFC] Clean up SSE variable shifts


On 08/21/2010 03:27 PM, Lu, Hongjiu wrote:
>>
>>> What is the performance impact of extra function call
>>> vs. multiple copies of the same jump table?
>>
>> I don't know.  I would hope that an indirect call and return
>> would be just as predictable as an indirect jump and branch.
>>
>> Fancy helping me benchmark?
>>
> 
> Sure. Send me a patch. I will run some benchmarks.

And for good measure, here's a third patch which implements a
variable shift via pshufb.  Which means we avoid any control
mis-prediction in favor of a data load stall.  Whether this is 
better or worse given a large insn re-order buffer, I have
no idea.  Something else to test.


r~

Attachment: z
Description: Text document


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