Decrease output after PyArray_MatrixProduct2().

theano/tensor/blas_headers.py

@@ -1033,7 +1033,7 @@ def openblas_threads_text():
 
 def blas_header_version():
     # Version for the base header
-    version = (5,)
+    version = (6,)
     if detect_macos_sdot_bug():
         if detect_macos_sdot_bug.fix_works:
             # Version with fix

theano/tensor/c_code/alt_blas_template.c

@@ -307,6 +307,8 @@ void %(precision)sgemv_(
     PyArray_MatrixProduct2(vector_x, vector_y, dot_product);
     if (PyErr_Occurred())
         alt_fatal_error("NumPy %(precision)sdot_ implementation: unable to compute dot.");
+    // PyArray_MatrixProduct2 adds an extra reference to the output array.
+    Py_XDECREF(dot_product);
     // Get result.
     Py_XDECREF(dot_product);
     Py_XDECREF(vector_y);