OpenCL: internal implementation tests in MQL5 - page 46

[Andrey Dik]

I need a test to calculate double precision. Somebody give me a hint, I'm a bit confused...

[Vladimir Gomonov]

joo:
I need a test to calculate double precision. Somebody chime in, I'm getting a little dumb...

Didn't you throw in an indicator? Or do you want a speed race again?

[Andrey Dik]

MetaDriver:
The indicator was throwing.

Nah, something as simple as the adjective 'door'.

[Ilyas]

The new build has changed the CLContextCreate function parameter, int device is now used instead of bool use_gpu.

if device>=0 is used as the number of OpenCL device in sequence
when device==OPENCL_DEVICE_ANY (-1) is used to automatically select from all available devices
when device==OPENCL_DEVICE_GPU (-2) is used to automatically select from all available GPU devices

Also, the function long CLGetInfoInteger(int handle,ENUM_OPENCL_PROPERTY_INTEGER prop_id) that allows getting properties of OpenCL objects

Only one property OPENCL_DEVICE_COUNT is available yet (the first parameter is ignored when requesting this property)

[Andrey Dik]

mql5:
The new build has changed the CLContextCreate function parameter, int device is now used instead of bool use_gpu.

device>=0 is used as OpenCL device number in order
If device==OPENCL_DEVICE_ANY (-1), then automatic selection among all available devices is used
when device==OPENCL_DEVICE_GPU (-2), automatically chooses from all available GPU devices

Also, the long CLGetInfoInteger(int handle,ENUM_OPENCL_PROPERTY_INTEGER prop_id) function has been added which allows to obtain properties of OpenCL objects

Only one OPENCL_DEVICE_COUNT property is available yet (the first parameter is ignored when requesting this property)

Does it already work in the current build or is it planned for the next build?

[Dmitriy Parfenovich]

joo:
Does it already work in the current build or is it planned for the next one?

just checked, it works in this build.

[Ilyas]

Sorry, I didn't indicate if I meant the 619 build. These changes have been implemented since build 619.
All EX5s using OpenCL and built with an earlier build of the compiler (<619) need to be rebuilt as stated in the EA log.

from build 619 onwards, the ENUM_OPENCL_HANDLE_TYPE CLHandleType(int handle) function is also available; it allows to get an OpenCL object type from a handle

and returns one of the following values:

OPENCL_INVALID
OPENCL_CONTEXT
OPENCL_PROGRAM
OPENCL_KERNEL
OPENCL_BUFFER

[Konstantin Grebenshikov]

http://www.3dnews.ru/news/626616 Victory for AMD. I wonder! How will this affect this project???

[Renat Fatkhullin]

GKS:
http://www.3dnews.ru/news/626616 Victory for AMD. I wonder! How will this affect this project???

AMD has taken the lead in mass product computing power by leaps and bounds, while NVIDIA has been betting on the unreasonably expensive, narrowly specialised Tesla.

[Konstantin Grebenshikov]

Renat:
AMD is way ahead on computational power in mass product while NVIDIA was betting on unreasonably expensive highly specialized Tesla.

Yes, that's for sure. Besides if you take flagships AMD has more "hungry bees" and calculations are better.

P.S. For me the choice of my next graphics card is obvious as I'm going to participate in the project.