qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH v3 4/5] qtest: precompute hex nibs


From: John Snow
Subject: Re: [Qemu-devel] [PATCH v3 4/5] qtest: precompute hex nibs
Date: Wed, 06 May 2015 12:18:41 -0400
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.6.0


On 05/06/2015 11:19 AM, Markus Armbruster wrote:
> John Snow <address@hidden> writes:
> 
>> On 05/06/2015 02:25 AM, Markus Armbruster wrote:
>>> John Snow <address@hidden> writes:
>>>
>>>> Instead of letting printf and friends do this for us
>>>> one byte at a time, fill a buffer ourselves and then
>>>> send the entire buffer in one go.
>>>>
>>>> This gives a moderate speed improvement over the old
>>>> method.
>>>
>>> Out of curiosity: how much of the improvement is due to doing our own
>>> buffering instead of printf()'s (assuming the stream is buffered), and
>>> how much is due to doing our own hex formatting instead of printf()'s?
>>>
>>
>> Out of ignorance: How would I measure?
> 
> Heh, well played!
> 
> The code before the series uses chr unbuffered:
> 
>         for (i = 0; i < len; i++) {
>             qtest_send(chr, "%02x", data[i]);
>         }
> 
> qtest_send() formats into two bytes, passes them to
> qemu_chr_fe_write_all(), which writes them to chr.
> 
> The chr are typically unbuffered, so this could well produce a series of
> two-byte write() system calls.
> 
> Adding some buffering will obviously make a difference for larger len.
> 
> Whether formatting hex digits by hands can make a difference is not
> obvious.
> 
> To find out, add just buffering.  Something like this in your patch
> instead of byte2hex():
> 
>          for (i = 0; i < len; i++) {
> -            qtest_sendf(chr, "%02x", data[i]);
> +            snprintf(&enc[i * 2], 2, "%02x", data[i]);
>          }
> 
> If the speedup is pretty much entirely due to buffering (which I
> suspect), then your commit message could use a bit of love :)
> 

When you're right, you're right. The difference may not be statistically
meaningful, but with today's current planetary alignment, using
sprintf() to batch the sends instead of my home-rolled nib computation
function, I can eke out a few more tenths of a second.

If there are no objections, I will stage patches 1-3 and 5, and resubmit
a quick v2 of just this single patch, unless you want to go ahead and
say that making the edit will be fine, then I will just edit it before
sending the pullreq.

--js



reply via email to

[Prev in Thread] Current Thread [Next in Thread]