[Beignet] Kaby lake perf worse than Skylake?

[Yang, Rong R]

> -----Original Message-----
> From: Michael Gratton [mailto:mike at vee.net]
> Sent: Thursday, July 20, 2017 12:16
> To: Yang, Rong R <rong.r.yang at intel.com>
> Cc: beignet at lists.freedesktop.org
> Subject: Re: [Beignet] Kaby lake perf worse than Skylake?
> 
> Hi,
> 
> On Thu, Jul 20, 2017 at 1:30 PM, Yang, Rong R <rong.r.yang at intel.com>
> wrote:
> > Is the quantity of infrared data they process same? Is it affected by
> > fps? Has you try other method, such as the profiling time. Actually,
> > in beignet, the Kabylake is almost same as Skylake.
> 
> The sensor operates at a fixed frame rate (30 fps), so the data processed was
> the same in both cases.
> 
> I haven't tried profiling the process yet, but may look into that. Is there any
> tools available for profiling OpenCL code?

You could use the OpenCL api clGetEventProfilingInfo to get the OpenCL enqueue functions profiling information.

> 
> I expected Kabylake the performance would be about the same or better
> than Skylake, which is why I was surprised that it seems to perform not as
> well for this task.

To compare Kabylake and Skylake, you'd better fix them to the same  frequency, by setting the
/sys/kernel/debug/dri/0/i915_max_freq and /sys/kernel/debug/dri/0/i915_min_freq to the same value.

> 
> //Mike
> 
> --
> ⊨ Michael Gratton, Percept Wrangler.
> ⚙ <http://mjog.vee.net/>
>