"MAX_BLOCKS_PER_MULTIPROCESSOR" seems not visible from the interface

[Emerald01]

I am trying to query MAX_BLOCKS_PER_MULTIPROCESSOR from the pycuda.driver. This is an important attribute that I want to know before assigning blocks and threads. BTW, I am using Pycuda 2021.1

import pycuda.driver as cuda_driver

cuda_driver.Context.get_device().get_attributes()

But I did not see this field although the documentation has it. And I checked pycuda backend, it indeed has this field here in ./src/wrapper/wrap_cudadrv.cpp: , I guess written in boost

 .value("MAX_BLOCKS_PER_MULTIPROCESSOR", CU_DEVICE_ATTRIBUTE_MAX_BLOCKS_PER_MULTIPROCESSOR)

how can I query this from pycuda API then?

[inducer]

>>> import pycuda.autoinit
>>> import pycuda.driver as drv
>>> drv.Device(0).max_blocks_per_multiprocessor
32

[Emerald01]

I tried to use this way, but seems that it cannot identify this particular field

>>> cuda_driver.Device(0).max_blocks_per_multiprocessor

AttributeError: type object 'device_attribute' has no attribute 'MAX_BLOCKS_PER_MULTIPROCESSOR'

But for example, if I tried warp_size, it indeed works

>>> cuda_driver.Device(0).warp_size
32

The version I am using is 2021.1

>>>pycuda.VERSION
(2021, 1)

[inducer]

Likely you built PyCUDA against a header version less than CUDA 11.0. MAX_BLOCKS_PER_MULTIPROCESSOR was introduced in 11.0.