ai-techsystems/dnnc-operators

development and discussion of DNNC operators

dnnc-operators's Issues

Bool return type tensors (Equal operator) and multiple dtype tensor input in compute function (Dequantizelinear)

Equal operator

Can't return tensor<bool>

Error:

swig/python detected a memory leak of type 'dnnc::tensor< bool > *', no destructor found.

Unittest Error:

ERROR: test_Equal2D (__main__.EqualTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "Equal.py", line 40, in test_Equal1D
    np.testing.assert_allclose(npr, np.array(dcr.data()).astype(np.float32),
AttributeError: 'SwigPyObject' object has no attribute 'data'

Fix Equal, Greater, Less operators for <bool> return type

Transpose operator

When added Transpose operator to the dnnc interface, invokes the following issue of ambiguity

---

---

Numpy uses np.T to transpose a matrix, we can use same kind of naming convention if doing that only, solves the issue!

Fix operator "macros.h"

Add .astype() method like Numpy to convert dtypes

Fix tensor.h

Fix to_string method in tensor.h for multi-dimensional tensors.