Fix numpy_ops gemm output semantics when BLIS is used

The `out` keyword argument of the `gemm` op specifies an output array,
However, the semantics were different depending on whether BLIS is used:

* `use_blis==False`: the values of out array are overwritten.
* `use_blis==True`: the values of out are added to the result.

With this change, the values of `out` are also overwritten with
`use_blis=True`.

thinc/backends/numpy_ops.pyx

@@ -88,7 +88,7 @@ class NumpyOps(Ops):
         y = self.as_contig(y)
         if out is not None:
             out = self.as_contig(out)
-        return blis.py.gemm(x, y, out=out, trans1=trans1, trans2=trans2)
+        return blis.py.gemm(x, y, out=out, trans1=trans1, trans2=trans2, beta=0.)
 
     def relu(self, np.ndarray X, inplace=False):
         cdef np.ndarray out = X if inplace else X.copy()

thinc/tests/backends/test_ops.py

@@ -275,6 +275,14 @@ def test_gemm_computes_correctly(cpu_ops):
     cpu_ops.gemm(X, W, trans1=True, out=Y)
 
 
+@pytest.mark.parametrize("cpu_ops", [*CPU_OPS, BLIS_OPS])
+def test_gemm_out_used(cpu_ops):
+    a = b = numpy.zeros((2, 2), dtype="f")
+    c = numpy.ones((2, 2), dtype="f")
+    cpu_ops.gemm(a, b, out=c)
+    assert numpy.array_equal(c, numpy.zeros((2, 2)))
+
+
 @pytest.mark.parametrize("cpu_ops", CPU_OPS)
 @settings(max_examples=MAX_EXAMPLES, deadline=None)
 @given(X=strategies.arrays_BI())