Hi Christian,
thanks for the feedback. Your observation is inline with most others. In
general users see a slight go noticeable speed improvement by OpenCL in
Marketsteijn demosaic. So it seems reasonable to leave Markesteijn
OpenCL on by default.
Best wishes
Ulrich
Am 08.02.2016 um 20:00
Hi Ulrich,
sorry for the delay for testing (i not very concerned as i don't have
x-trans sensor
but as i'm reading result form ML i saw only nvidia card, so i have make
the test for you ... :
for me, it's nearly unusable wihout opencl (even with my 8 core), it's too
slow
But with the opencl
Hi Ulrich,
sorry for the delay for testing (i not very concerned as i don't have
x-trans sensor
but as i'm reading result form ML i saw only nvidia card, so i have make
the test for you ... :
for me, it's nearly unusable wihout opencl (even with my 8 core), it's too
slow
But with the opencl
On 22/01/16 13:54, Robert William Hutton wrote:
OpenCL markesteijn-1: 0.281 secs
OpenCL markesteijn-3: 0.492 secs
CPUmarkesteijn-1: 0.544 secs
CPUmarkesteijn-3: 1.273 secs
Whoops, log attached.
Regards,
Rob
___
d
On 20/01/16 07:42, Ulrich Pegelow wrote:
1) with OpenCL enabled in preferences
2) with OpenCL disabled in preferences
a) exporting an X-Trans image with default history stack
b) as above but now with demosaic set to Markesteijn-3
It appears that exports are about twice as fast with OpenCL enabl
Hi,
thanks. So despite the quite modern NVIDIA GPU the OpenCL code is only
marginally faster than the CPU.
Ulrich
Am 20.01.2016 um 14:35 schrieb Marc Cousin:
Hi,
Here are my numbers… first thing to say is that it «feels» faster.
System is CPU: i7-4790K, GPU : GeForce GTX 960 (2GB Ram)
_
Hi,
that looks about the same as my system.
Thanks
Ulrich
Am 20.01.2016 um 15:22 schrieb Christian Kanzian:
Hi,
Sorry, I managed to cut off the details. :-(
Rerun again and details are attached now.
Christian
___
dar
Hi,
Sorry, I managed to cut off the details. :-(
Rerun again and details are attached now.
Christian
Am Mittwoch, 20. Januar 2016, 15:01:44 schrieb Christian Kanzian:
> Hi,
>
> Here are my figures for the sample raw:
>
> CPU: i7-2600 CPU @ 3.40GHz
> GPU: Radeon R9 270X 4GB GDDR5
> GLOBAL_ME
Hi,
Here are my figures for the sample raw:
CPU: i7-2600 CPU @ 3.40GHz
GPU: Radeon R9 270X 4GB GDDR5
GLOBAL_MEM_SIZE: 3072MB
DRIVER_VERSION: 1526.3 (VM)
RAM: 16 GB
Debian Jessie 64-bit
Summary:
no openCL Markesteijn-1
[dev_process_export] pixel pipeline processing took 1,030 secs (6,852
CPU)
Hi,
Here are my numbers… first thing to say is that it «feels» faster.
System is CPU: i7-4790K, GPU : GeForce GTX 960 (2GB Ram)
With HEAD:
==
With OpenCL:
1 pass, no color smoothing
[dev] took 0.000 secs (0.000 CPU) to load the image.
[export] creating pixelpipe took 0.009 s
BTW here is an example how it could look with OpenCL:
[pixelpipe_process] [export] using device 1
[dev_pixelpipe] took 0,000 secs (0,000 CPU) initing base buffer [export]
[dev_pixelpipe] took 0,016 secs (0,001 CPU) processed
`Raw-Schwarz-/Weißpunkt' on GPU, blended on GPU [export]
[dev_pixelpipe
Thanks!
Looking at your attachment it seems that OpenCL has not been used for
the demosaic step in any of the test cases. E.g.:
[dev_pixelpipe] took 0,736 secs (2,524 CPU) processed `Entrastern' on
CPU, blended on CPU [export]
That's the behavior I would expect for darktable prior to my rec
Am 19.01.2016 um 17:18 schrieb David Vincent-Jones:
Will the merge now permit X-Trans raw images to be used in the HDR module?
Nope. This is unrelated.
Ulrich
David
___
darktable developer mailing list
to unsubscribe
Definitely this doesn't touch the HDR module. Calculating an HDR image
from x-trans data may even work with the existing Bayer math. The
challenge then being to write the results in some form which darktable
can turn around and read. Perhaps this would mean teaching dt to write
DNGs with the x-tran
Hi,
did some tests:
Intel® Core™ i5-2500 CPU @ 3.30GHz × 4
GeForce GTX 750 Ti, 2GB,128-bit
Ubuntu 15.10 64-Bit, 16GB
opencl - m1
[dev_process_export] pixel pipeline processing took 1,075 secs (2,872 CPU)
opencl - m3
[dev_process_export] pixel pipeline processing took 2,001 secs (6,548 CPU)
no-
It's my understanding that the HDR module relies on Bayer pattern data, so no.
-R
On 20/01/16 03:18, David Vincent-Jones wrote:
Will the merge now permit X-Trans raw images to be used in the HDR module?
David
On 01/19/2016 08:42 PM, Ulrich Pegelow wrote:
Hi,
I just merged my OpenCL implemen
Will the merge now permit X-Trans raw images to be used in the HDR module?
David
On 01/19/2016 08:42 PM, Ulrich Pegelow wrote:
> Hi,
>
> I just merged my OpenCL implementation of the Markesteijn demosaicing
> algorithm into master. Markesteijn with one or three passes ("-1" and
> "-3", respective
Hi,
I just merged my OpenCL implementation of the Markesteijn demosaicing
algorithm into master. Markesteijn with one or three passes ("-1" and
"-3", respectively) is darktable's preferred method for demosaicing
images of cameras with Fuji's X-Trans sensor.
The algorithm is rather complex an
18 matches
Mail list logo
