Add a test to prevent the problem from re-occuring.

theano/sandbox/tests/test_rng_mrg.py

@@ -361,13 +361,9 @@ def test_GPU_nstreams_limit():
 
 
 def test_consistency_GPUA_serial():
-    """
-    Verify that the random numbers generated by GPUA_mrg_uniform, serially,
-    are the same as the reference (Java) implementation by L'Ecuyer et al.
-
-    """
-    from theano.gpuarray.tests.test_basic_ops import \
-        mode_with_gpu as mode
+    # Verify that the random numbers generated by GPUA_mrg_uniform, serially,
+    # are the same as the reference (Java) implementation by L'Ecuyer et al.
+    from theano.gpuarray.tests.test_basic_ops import mode_with_gpu as mode
     from theano.gpuarray.type import gpuarray_shared_constructor
 
     seed = 12345
@@ -415,14 +411,10 @@ def test_consistency_GPUA_serial():
 
 
 def test_consistency_GPUA_parallel():
-    """
-    Verify that the random numbers generated by GPUA_mrg_uniform, in
-    parallel, are the same as the reference (Java) implementation by
-    L'Ecuyer et al.
-
-    """
-    from theano.gpuarray.tests.test_basic_ops import \
-        mode_with_gpu as mode
+    # Verify that the random numbers generated by GPUA_mrg_uniform, in
+    # parallel, are the same as the reference (Java) implementation by
+    # L'Ecuyer et al.
+    from theano.gpuarray.tests.test_basic_ops import mode_with_gpu as mode
     from theano.gpuarray.type import gpuarray_shared_constructor
 
     seed = 12345
@@ -468,6 +460,30 @@ def test_consistency_GPUA_parallel():
     assert(numpy.allclose(samples, java_samples))
 
 
+def test_GPUA_full_fill():
+    # Make sure the whole sample buffer is filled.  Also make sure
+    # large samples are consistent with CPU results.
+    from theano.gpuarray.tests.test_basic_ops import (mode_with_gpu as mode,
+                                                      test_ctx_name)
+    from theano.gpuarray.type import gpuarray_shared_constructor
+
+    # This needs to be large to trigger the problem on GPU
+    size = (10, 1000)
+
+    R = MRG_RandomStreams(234, use_cuda=False)
+    uni = R.uniform(size, nstreams=60*256)
+    f_cpu = theano.function([], uni)
+
+    rstate_gpu = gpuarray_shared_constructor(R.state_updates[-1][0].get_value())
+    new_rstate, sample = rng_mrg.GPUA_mrg_uniform.new(rstate_gpu, ndim=None,
+                                                      dtype='float32',
+                                                      size=size)
+    rstate_gpu.default_update = new_rstate
+    f_gpu = theano.function([], sample)
+
+    utt.assert_allclose(f_cpu(), f_gpu())
+
+
 def basictest(f, steps, sample_size, prefix="", allow_01=False, inputs=None,
               target_avg=0.5, target_std=None, mean_rtol=0.01, std_tol=0.01):
     if inputs is None: