Compilation errors with CUDA 3.1

I’m using OpenCL with a GTX 280 under Fedora 10. I’ve just upgraded to CUDA 3.1, and several of my kernels no longer compile. Here is the error log:

[codebox]ptxas application ptx input, line 492; fatal : Parsing error near ‘PHINODE’: syntax error

ptxas fatal : Ptx assembly aborted due to errors

error : Ptx compilation failed: gpu=‘sm_13’, device code=‘cuModuleLoadDataEx_5’

: Considering profile ‘compute_13’ for gpu=‘sm_13’ in ‘cuModuleLoadDataEx_5’

: Retrieving binary for ‘cuModuleLoadDataEx_5’, for gpu=‘sm_13’, usage mode=’ ’

: Considering profile ‘compute_13’ for gpu=‘sm_13’ in ‘cuModuleLoadDataEx_5’

: Control flags for ‘cuModuleLoadDataEx_5’ disable search path

: Ptx binary found for ‘cuModuleLoadDataEx_5’, architecture=‘compute_13’

: Ptx compilation for ‘cuModuleLoadDataEx_5’, for gpu=‘sm_13’, ocg options=’ '[/codebox]

The string “PHINODE” does not appear anywhere in my kernel, so that seems to be something internal to the compiler. Any idea what could be causing this, or what I could do about it?


This appears to be related to the compiler flag -cl-denorms-are-zero. If I remove that, the compilatiion error goes away.