done with the current change

theano/sandbox/cuda/opt.py

@@ -2275,11 +2275,10 @@ def gpuScanOptimization(node):
             return outputs
     return False
 
-# en attente de tests et de correction
+
 @register_opt()
 @local_optimizer([tensor.AllocEmpty, gpu_from_host])
 def local_gpu_allocempty(node):
-    #print 'kikou'
     if (isinstance(node.op, tensor.AllocEmpty) and
         node.op.dtype=="float32"):
         return [host_from_gpu(GpuAllocEmpty()(*node.inputs))]

theano/sandbox/cuda/tests/test_basic_ops.py

@@ -380,6 +380,17 @@ def test_alloc_empty():
     assert out.shape == (2, 3)
     assert out.dtype == 'float32'
 
+    # Test that we merge them.
+    f = theano.function([], [cuda.basic_ops.gpu_alloc_empty(2, 3),
+                             cuda.basic_ops.gpu_alloc_empty(2, 3)])
+    out = f()
+    assert out[0].shape == (2, 3)
+    assert out[0].dtype == 'float32'
+    assert out[1].shape == (2, 3)
+    assert out[1].dtype == 'float32'
+    assert len([node for node in f.maker.fgraph.apply_nodes
+                if isinstance(node.op, cuda.basic_ops.GpuAllocEmpty)]) == 1
+
 
 def test_elemwise_empty():
     # test with 0 element

theano/sandbox/cuda/tests/test_opt.py

@@ -153,6 +153,7 @@ def test_gpualloc():
     l = f.maker.fgraph.toposort()
     assert numpy.any([isinstance(x.op, cuda.GpuAlloc) for x in l])
 
+
 def test_gpuallocempty():
 
 

theano/tensor/basic.py

@@ -5550,8 +5550,7 @@ class AllocEmpty(gof.Op):
         return [node.inputs]
 
     def c_code_cache_version(self):
-        return None
-        return (2,)
+        return (3,)
 
     def do_constant_folding(self, node):
         return False

theano/tensor/blas.py

@@ -2043,7 +2043,7 @@ def local_dot22_to_dot22scalar(node):
         assert not a.type.ndim
 
         z = T.AllocEmpty(d.owner.inputs[0].dtype)(d.owner.inputs[0].shape[0],
-                                  d.owner.inputs[1].shape[1])
+                                                  d.owner.inputs[1].shape[1])
         zero = T.as_tensor_variable(numpy.asarray(0, dtype=a.dtype))
         dot = gemm(z, a, d.owner.inputs[0], d.owner.inputs[1], zero)
         

theano/tensor/tests/test_blas.py

@@ -875,7 +875,7 @@ def test_dot22scalar():
                     cst = theano.tensor.basic.constant(.2, dtype=dtype4)
                     cst2 = theano.tensor.basic.constant(.1, dtype=dtype4)
 
-                    def check_dot22scalar(func, len_topo_scalar=-1):
+                    def check_dot22scalar_gemm(func, len_topo_scalar=-1):
                         topo = func.maker.fgraph.toposort()
                         ops = [x.op for x in topo]
                         classes = [type(x.op) for x in topo]
@@ -920,7 +920,7 @@ def test_dot22scalar():
                             f = theano.function([a, b], cst * T.dot(a, b),
                                                 mode=mode_blas_opt)
                             topo = f.maker.fgraph.toposort()
-                            check_dot22scalar(f, 1)
+                            check_dot22scalar_gemm(f, 1)
 
                             f(av, bv)
 
@@ -929,7 +929,7 @@ def test_dot22scalar():
                                                 cst * c * T.dot(a, b),
                                                 mode=mode_blas_opt)
                             topo = f.maker.fgraph.toposort()
-                            check_dot22scalar(f, 5)
+                            check_dot22scalar_gemm(f, 5)
                             #print (av.dtype, bv.dtype, cv.dtype)
 
                             f(av, bv, cv)
@@ -938,7 +938,7 @@ def test_dot22scalar():
                                             c * cst * T.dot(a, b),
                                             mode=mode_blas_opt)
                         topo = f.maker.fgraph.toposort()
-                        check_dot22scalar(f, 5)
+                        check_dot22scalar_gemm(f, 5)
                         f(av, bv, cv)
 
                         # Here, canonicalize also seems needed
@@ -948,7 +948,7 @@ def test_dot22scalar():
                                             cst2 * c * cst * T.dot(a, b),
                                             mode=m2)
                         topo = f.maker.fgraph.toposort()
-                        check_dot22scalar(f, 5)
+                        check_dot22scalar_gemm(f, 5)
                         f(av, bv, cv)
 
                         if dtype1 == dtype2 == dtype3:
@@ -956,7 +956,7 @@ def test_dot22scalar():
                                                 c * cst * a * T.dot(a, b),
                                                 mode=m2)
                             topo = f.maker.fgraph.toposort()
-                            check_dot22scalar(f, 5)
+                            check_dot22scalar_gemm(f, 5)
                             f(sv, sv, sv)
 
                             f = theano.function([a, b, c],
@@ -979,7 +979,7 @@ def test_dot22scalar():
                                                 c * a * cst * T.dot(a, b),
                                                 mode=m2)
                             topo = f.maker.fgraph.toposort()
-                            check_dot22scalar(f, 5)
+                            check_dot22scalar_gemm(f, 5)
                             f(sv, sv, sv)
 
                     cmp((3, 4), (4, 5), (3, 5))