OpenCL: libmath functions explode for float64 #631
Description
Both bh107 and bohrium explode when applying libmath functions (sin, cos, tanh, exp, etc.) to float64 data in OpenCL mode.
In [1]: import bohrium as bh
In [2]: A = bh.random.random((10))
In [3]: A.dtype
Out[3]: dtype('float64')
In [4]: B = bh.tanh(A)
In [5]: bh.flush()
Error code: -11
Error building:
stringInput.cl:18:20: error: call to 'tanh' is ambiguous
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:34:20: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:133:19: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1457:21: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1458:21: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1459:21: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1460:21: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1461:22: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1462:20: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1463:20: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1464:20: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1465:20: note: candidate function
/usr/lib/x86_64-linux-gnu/beignet/include/ocl_math.h:1466:21: note: candidate function
terminate called after throwing an instance of 'cl::Error'
what(): clBuildProgram
Abortedbh107 does exactly the same. However, for float32 it works:
In [1]: import bohrium as bh
In [2]: A = bh.random.random((10)).astype(bh.float32)
In [3]: B = bh.tanh(A)
In [5]: bh.flush()
In [6]: print(B)
[0.7571309 0.37564066 0.19148484 0.27567866 0.7203832 0.5446466
0.28313637 0.7497538 0.64565045 0.31858957]