[Development] State of x86 SIMD in Qt

[lars.knoll at nokia.com]

On 12/30/11 10:26 PM, "ext Thiago Macieira" <thiago.macieira at intel.com>
wrote:

>On Thursday, 29 de December de 2011 15.16.19, jason.barron at nokia.com
>wrote:
>> Hi Thiago,
>> 
>> >  I can get started, but I need to know: how is this code tested? How
>>can I
>> >  verify that things are working?
>> The best way to test this is to run the lancelot autotests to verify
>>that
>> the output of the raster engine is the same before and after. The code
>>for
>> the baseline server is in Qt, but the baselines we are currently using
>> inside Nokia  are not (for lack of public infrastructure I guess) so
>>you'll
>> have to generate your own baseline (before the patch) and then test it
>> (after the patch). The tests are located here:
>> 
>>     qtbase/tests/auto/other/lancelot
>> 
>> the baselineserver is here:
>> 
>>     qtbase/tests/baselineserver
>> 
>> The draw helpers are mostly used in blending code so these lancelot
>>tests
>> (eg: porter_duff.qps) probably offer the best coverage in this area.
>
>Thanks Jason
>
>I'll run the regression tests next year.
>
>Are there benchmarks for this code? After my work of yesterday and today,
>we've got a full set of draw helpers compiled with AVX, so I'd like to
>give it 
>a go on my Sandy Bridge laptop.
>
>I'd also like to compare the previous SSE2 to the current SSE2 code.

There's unfortunately nothing complete. Some benchmarks are in
tests/benchmarks/gui/painting, but I'm not sure they cover these.

Cheers,
Lars