Hello guys.
I can't find a way for creating point position attribute with 64 bit size.
I saw some samples in HDK and all of those use 32 bit size for point position.
I even try to cast a P attribute to 64 bit inside Houdini but after printing positions with python everything are still the same!
Is this true that Houdini can only accept point position attribute as float?
Thanks in advance.

Yes, AFAIK attributes are only 32 bit wide.

Thank you dear Enivob.
I find out that GA_Detail class takes an argument to set size of position:
GA_Detail Constructor [www.sidefx.com]
But the problem is how can I access that via GU_Detail class because constructors of this class don't take any GA_Storage to construct GA_Detail as we wish.
Another bad news is all setters in GA_Detail class are getting UT_Vector3 as their argument for set position attribute which is 32 bit in size!

not sure about Python, but since you can use Attrib Cast SOP to cast attributes to 64bit and some SOPs like Transform SOP allegedly use 64bit precision if P is 64bit I suppose there is a way to do it in HDK
VEX however is currently just 32bit

Thank you dear Tomas.
I just did a simple test in following image.
Seems P can't cast to 64 bit even if we explicitly tell Houdini!
If you notice there is difference between 32 and 64 bit new_pos which is an arbitrary attribute.

There is some info from Neil in this thread
https://www.sidefx.com/forum/topic/43740/ [www.sidefx.com]

but you are right, it doesn't seem that switching to 64bit does anything
here is a simple test with transforming geo to get rounding errors, both 32bit and 64bit versions look the same unfortunately

What a pity.
I'm wondering why SESI put a storage argument in GA_Detail.h to set size for position attributes!
Really Thanks Tomas.