Trying to run the parameter tuner example

[labv]

I have compiled the matrix parameter tuner example supplied with source successfully, but I do receive the following execution-time error:

ViennaCL: FATAL ERROR: Could not find kernel 'trans_lower_triangular_substitute_inplace'
Number of kernels in program: 22
terminate called after throwing an instance of 'char const*'

If I comment most of the tests defined in the file matrix.cpp, I can make the the example run without any errors, but it ends up optimising very few kernels.

Looking the kernels declaration files, I really can't find any kernel named, for example, "trans_lower_triangular_substitute_inplace". This makes me suspect that maybe the matrix parameter tunning example is needing some update in order to follow the library changes.

I have tried to fix the (supposedly broken) code, but had no success. Curiously, even though the vector parameter tunning example have the same kind of problem, the sparse parameter tunning example works just fine without any modification.

Is such code really broken or am I doing something wrong?

Some additional info:
ViennaCL 1.4.2
ATI HD5770
Ubuntu 11.04 x64
Catalyst 12.1
CAL 1.4.1741

[Karl Rupp]

Hi labv,

double-checking the matrix parameter tuner showed that this is indeed broken - thanks for reporting. I created an issue here:
https://github.com/viennacl/viennacl-dev/issues/27
so that it gets fixed with the next release.

Some background on that: We changed the internally used kernels for vectors and dense matrices recently, so the parameter tuner would need to use the new names - but still uses the old ones. If you really need to get the parameter tuner going, please let me know, I can provide a patch. Otherwise, please remain patient for the next release, where we expect to provide higher performance than what is possible with the existing parameter tuner.

Best regards,
Karli