OpenCL issue with 195.36.15

[grimm]

My OpenCL programs that used to run no longer run under the new driver. The problem appears to be with the clCreateContextFromType function that fails with a -32 error code or "CL_INVALID_PLATFORM". I'm calling the function with the device type of "CL_DEVICE_TYPE_GPU". Thanks,

Grimm

[danix]

Hi grimm,

Thanks for reporting this bug. Can you provide a test program (if not your program itself, then at least a simple test case that highlights the problem) and an nvidia-bug-report.log please?

[grimm]

No problem, I'm using the Ruby bindings to OpenCL. I have attached a sample program that came with the bindings that just lists the device info. This program used to work with the previous drivers and now gives the invalid device error. I also attached the bug report log as well. Hmm, the forum software would not allow me to upload the Ruby program file so I will include it here in full as it's not long.

device_query.rb

require "opencl"

context = OpenCL::Context.create_from_type(OpenCL:evice::T YPE_GPU)
context.devices.each_with_index do |device,i|
print "Device#{i}\n"
%w(type_names single_fp_config_names execution_capability_names queue_property
_names global_mem_cache_type_name local_mem_type_name max_work_item_sizes vendor
_id max_compute_units max_work_item_dimensions preferred_vector_width_char prefe
rred_vector_width_short preferred_vector_width_long preferred_vector_width_float
preferred_vector_width_double max_clock_frequency address_bits max_read_image_a
rgs max_write_image_args max_samplers mem_base_addr_align min_data_type_align_si
ze global_mem_cacheline_size max_constant_args max_mem_alloc_size global_mem_cac
he_size global_mem_size max_constant_buffer_size local_mem_size image_support er
ror_correction_support endian_little available compiler_available name vendor pr
ofile version extensions max_work_group_size image2D_max_width image2D_max_heigh
t image3D_max_width image3D_max_height image3D_max_depth max_parameter_size prof
iling_timer_resolution).each do |qn|
print " #{qn}: #{device.send(qn).inspect}\n"
end
end

Grimm

[danix]

Thanks, I'll file a bug for this internally.

What was the last version you used that did work?

[grimm]

Thanks, the last version I was using was I think 190.29.

Grimm

[grimm]

I have also confirmed that this is a problem with the C++ bindings on my machine as well. I have reverted back to the previous driver and everything is working again.

Thanks,

Grimm

[grimm]

The ruby bindings programmer has updated the bindings to work with the new 3.0 version. Everything is working great now. Thank you very much for your help.

Grimm