Fix teardown

[hughperkins]

Fixes issue https://github.com/clMathLibraries/clBLAS/issues/159

What is the issue?

• teardown doesnt re-initialize the autogemm kernels to null

What is the effect of the issue?

• following sequence causes segfault:
  • clblassetup
  • gemm
  • clblasteardown
  • clblassetup
  • gemm

How does the fix work?

• AutoGemm now adds a new method to set the autogemm kernels to null
• init.c contains a call to this new method

(Note that this also incorporates pull request for 'array initialize must be an initializer list', otherwise gemm doesnt get very far on my machine :-P )

[hughperkins]

Note: example output for the same sourecode as in https://github.com/clMathLibraries/clBLAS/issues/159#issuecomment-150896488 :

$ LD_LIBRARY_PATH=../build/library/ ./test
i=0
got platformids
got deviceids
created context
created commandqueue
setup blas ok
calling sgemm....

clblasSgemmEx result:
11 12 13 
21 35720 36680 
31 50720 52080 
41 65720 67480 
finished ok :-)
i=1
got platformids
got deviceids
created context
created commandqueue
setup blas ok
calling sgemm....

clblasSgemmEx result:
11 12 13 
21 35720 36680 
31 50720 52080 
41 65720 67480 
finished ok :-)
i=2
got platformids
got deviceids
created context
created commandqueue
setup blas ok
calling sgemm....

clblasSgemmEx result:
11 12 13 
21 35720 36680 
31 50720 52080 
41 65720 67480 
finished ok :-)

[hughperkins]

(Updated commit to call clReleaseKernel on the kernel first)

[hughperkins]

(I think this new commit has addressed also issue https://github.com/clMathLibraries/clBLAS/issues/166 , at least in combination with the fix identified in https://github.com/clMathLibraries/clBLAS/issues/167 )

[guacamoleo]

Hugh's concern and fix are both legit. I checked out his commit and validated that it passes the gemm tests.

[hughperkins]

Thanks!

[hughperkins]

Hi. Just realized that this fails to build the first time, since include/AutoGemmIncludes/AutoGemmClKernels.cpp isnt present. And init.c is compiled before the generation step.