The cl_event return by clEnqueueNDRange in 197.45 on windows seems to be broken.
If I try to get the get command queue with clGetEventInfo, the cl_command_queue isn’t valid (the address seems to be shifted by few bytes) and a clRetainCommandQueue on it crash.
It’s the same problem with 257.15 beta …
I got bitten by the same thing.
Event created with clEnqueueTask()
clGetEventInfo returns another address than sent into clEnqueueTask,
myapp crashes there so I don’t know what happens with clRetainCommandQueue.
Drivers: 258.96
Windows 7 64 (with a 32 bits application.)
Geforce 250 and Intel I7
The same code works on linux ubuntu 10.4 (32 bits) (with nvidia drivers supplied by ubuntu)
on the same hardware.
/Dan
I got bitten by the same thing.
Event created with clEnqueueTask()
clGetEventInfo returns another address than sent into clEnqueueTask,
myapp crashes there so I don’t know what happens with clRetainCommandQueue.
Drivers: 258.96
Windows 7 64 (with a 32 bits application.)
Geforce 250 and Intel I7
The same code works on linux ubuntu 10.4 (32 bits) (with nvidia drivers supplied by ubuntu)
on the same hardware.
/Dan
I already know the command queue of any events and do not need to do this, since I force 1 command queue per context & have an array of contexts. I would suggest the following, if you are not already doing:
clEnqueueNDRange
clFlush, or some implicit clFlush
clGetEventInfo
If you look at the last paragraph of the clFlush man page, it implies events associated with command queues need to be flushed before use.
I already know the command queue of any events and do not need to do this, since I force 1 command queue per context & have an array of contexts. I would suggest the following, if you are not already doing:
clEnqueueNDRange
clFlush, or some implicit clFlush
clGetEventInfo
If you look at the last paragraph of the clFlush man page, it implies events associated with command queues need to be flushed before use.