Incorrect histogramdd behavior with masked arrays

Summing over the histogram should always give the number of non-masked records when no weighting is used. As shown below, the sum is reduced, but not by the correct amount. Calling the "compressed()" method and reshaping before running the histogram produces correct results.

The following uses numpy 1.8.1.

 x = ma.arange(300).reshape(100,3)

In [5]: print x[:,0].count()
100

In [6]: H,e = np.histogramdd(x)

In [7]: H.sum()
Out[7]: 100.0

In [8]: for i in range(10) : 
   ...:     x[i*i,:]=ma.masked
   ...:     

In [9]: print x[:,0].count()
90

In [10]: H,e = np.histogramdd(x)

In [11]: H.sum()
Out[11]: 98.0

x = x.compressed().reshape(90,3)

H,e = np.histogramdd(x)

H.sum()
Out[14]: 90.0

Start by reproducing the masked-array example with numpy.histogramdd and compare its result with x.compressed().reshape(90,3). Check how masked values are handled during histogramdd and review the existing histogramdd tests, if available. Done means an unweighted histogram sums to the number of non-masked records.