No implementation of function Function(<built-in function mul>) found for signature: >>> mul(array(float32, 1d, C), array(float32, 1d, C))

[vrunm]

I am trying to optimize my face recognition code, so that it executes on GPU. There is a cosine_similarity function in scikit-learn, which was throwing error with numba, so then I created a separate function.

@cuda.jit
def cosine_similarity_numba(u:np.ndarray, v:np.ndarray):
    assert(u.shape[0] == v.shape[0])
    print(u.shape[0])
    print(v.shape[0])
    l = 0
    m = 0
    n = 0
    for i in range(u.shape[0]):
        print(u[i])
        print(v[i])
        l += u[i]*v[i]
        m += u[i]*u[i]
        n += v[i]*v[i]
    cos_theta = 1
    if m!=0 and n!=0:
        cos_theta = l/np.sqrt(m*n)
    return cos_theta

but numba is still throwing error.
The error is:-

Failed in cuda mode pipeline (step: nopython frontend)
No implementation of function Function(<built-in function mul>) found for signature:
 
 >>> mul(array(float32, 1d, C), array(float32, 1d, C))
 
There are 12 candidate implementations:
      - Of which 10 did not match due to:
      Overload of function 'mul': File: <numerous>: Line N/A.
        With argument(s): '(array(float32, 1d, C), array(float32, 1d, C))':
       No match.
      - Of which 2 did not match due to:
      Operator Overload in function 'mul': File: unknown: Line unknown.
        With argument(s): '(array(float32, 1d, C), array(float32, 1d, C))':
       No match for registered cases:
        * (int64, int64) -> int64
        * (int64, uint64) -> int64
        * (uint64, int64) -> int64
        * (uint64, uint64) -> uint64
        * (float32, float32) -> float32
        * (float64, float64) -> float64
        * (complex64, complex64) -> complex64
        * (complex128, complex128) -> complex128

During: typing of intrinsic-call at /home/proeffico/Desktop/ProlificCUDA/face_rec.py (92)

File "face_rec.py", line 92:
def cosine_similarity_numba(u:np.ndarray, v:np.ndarray):
    <source elided>
        print(v[i])
        l += u[i]*v[i]
        ^

During: resolving callee type: type(CUDADispatcher(<function cosine_similarity_numba at 0x7f622ce37700>))
During: typing of call at /home/proeffico/Desktop/ProlificCUDA/face_rec.py (59)


File "face_rec.py", line 59:
def ml_search_algorithm(names,feature_column,test_vector,thresh=0.5):
    <source elided>
    # step-3: Cal. cosine similarity
    similar = cosine_similarity_numba(x,test_vector.reshape(1,-1))
    ^
`

Is there any way around this?

[gmarkall]

Array operations need to be rewritten such that they are performed elementwise for the CUDA target, with elements assigned to threads according to your judgment. For your case you probably want to use the thread index to index into the arrays, as in the Vector Addition example.

[guilhermeleobas]

Closing this issue as it seems to be resolved. If this is not the case please re-open with a comment about any item that appears to be unresolved. Many thanks.