Merge pull request #6108 from abergeron/careduce_compile

theano/gof/params_type.py

@@ -655,7 +655,7 @@ class ParamsType(Type):
         return list(sorted(list(c_support_code_set))) + [final_struct_code]
 
     def c_code_cache_version(self):
-        return ((2,), tuple(t.c_code_cache_version() for t in self.types))
+        return ((3,), tuple(t.c_code_cache_version() for t in self.types))
 
     # As this struct has constructor and destructor, it could be instanciated on stack,
     # but current implementations of C ops will then pass the instance by value at functions,
@@ -684,6 +684,7 @@ class ParamsType(Type):
         /* Seems c_init() is not called for a op param. So I call `new` here. */
         %(name)s = new %(struct_name)s;
 
+        { // This need a separate namespace for Clinker
         const char* fields[] = {%(fields_list)s};
         if (py_%(name)s == Py_None) {
             PyErr_SetString(PyExc_ValueError, "ParamsType: expected an object, not None.");
@@ -703,5 +704,6 @@ class ParamsType(Type):
                 %(fail)s
             }
         }
+        }
         """ % dict(name=name, struct_name=self.name, length=self.length, fail=sub['fail'],
                    fields_list='"%s"' % '", "'.join(self.fields))

theano/gpuarray/tests/test_elemwise.py

@@ -7,7 +7,7 @@ import scipy.special
 
 import theano
 from theano import scalar, gof, tensor
-from theano.compile import DebugMode
+from theano.compile import DebugMode, Mode
 from theano.tests.unittest_tools import SkipTest, assert_allclose
 
 from theano.tensor.tests import test_elemwise
@@ -212,7 +212,9 @@ class test_GpuCAReduceCPY(test_elemwise.test_CAReduce):
     def test_perform(self):
         for dtype in self.dtypes + self.bin_dtypes:
             for op in self.reds:
-                self.with_linker(gof.PerformLinker(), op, dtype=dtype,
+                self.with_mode(Mode(linker='py',
+                                    optimizer=mode_with_gpu.optimizer),
+                               op, dtype=dtype,
                                pre_scalar_op=self.pre_scalar_op)
 
     def test_perform_nan(self):
@@ -220,14 +222,18 @@ class test_GpuCAReduceCPY(test_elemwise.test_CAReduce):
             if not dtype.startswith('float'):
                 continue
             for op in self.reds:
-                self.with_linker(gof.PerformLinker(), op, dtype=dtype,
+                self.with_mode(Mode(linker='py',
+                                    optimizer=mode_with_gpu.optimizer),
+                               op, dtype=dtype,
                                test_nan=True,
                                pre_scalar_op=self.pre_scalar_op)
 
     def test_c(self):
         for dtype in self.dtypes + self.bin_dtypes:
             for op in self.reds:
-                self.with_linker(gof.CLinker(), op, dtype=dtype,
+                self.with_mode(Mode(linker='c',
+                                    optimizer=mode_with_gpu.optimizer),
+                               op, dtype=dtype,
                                pre_scalar_op=self.pre_scalar_op)
 
     def test_c_nan(self):
@@ -235,7 +241,9 @@ class test_GpuCAReduceCPY(test_elemwise.test_CAReduce):
             if not dtype.startswith('float'):
                 continue
             for op in self.reds:
-                self.with_linker(gof.CLinker(), op, dtype=dtype,
+                self.with_mode(Mode(linker='c',
+                                    optimizer=mode_with_gpu.optimizer),
+                               op, dtype=dtype,
                                test_nan=True,
                                pre_scalar_op=self.pre_scalar_op)
 
@@ -334,6 +342,9 @@ class test_GpuCAReduceCuda(test_GpuCAReduceCPY):
             scalar.maximum, scalar.minimum]
     pre_scalar_op = None
 
+    def test_perform_noopt(self):
+        return
+
     def test_perform(self):
         return
 

theano/tensor/elemwise.py

@@ -1202,7 +1202,7 @@ second dimension
         return support_code
 
     def c_code_cache_version_apply(self, node):
-        version = [12]  # the version corresponding to the c code in this Op
+        version = [13]  # the version corresponding to the c code in this Op
 
         # now we insert versions for the ops on which we depend...
         scalar_node = Apply(
@@ -1622,7 +1622,7 @@ class CAReduce(Op):
 
     def c_code_cache_version_apply(self, node):
         # the version corresponding to the c code in this Op
-        version = [7]
+        version = [8]
 
         # now we insert versions for the ops on which we depend...
         scalar_node = Apply(

theano/tensor/elemwise_cgen.py

@@ -72,16 +72,12 @@ def make_checks(loop_orders, dtypes, sub):
                 %(var)s_n%(index)s = PyArray_DIMS(%(var)s)[%(index)s];
                 %(var)s_stride%(index)s = PyArray_STRIDES(%(var)s)[%(index)s] / sizeof(%(dtype)s);
                 %(var)s_jump%(index)s_%(j)s = %(jump)s;
-                //printf("%(var)s_jump%(index)s_%(j)s is:");
-                //std::cout << %(var)s_jump%(index)s_%(j)s << std::endl;
                 """ % locals()
                 adjust = "%(var)s_n%(index)s*%(var)s_stride%(index)s" % locals()
             else:
                 jump = "-(%s)" % adjust
                 init += """
                 %(var)s_jump%(index)s_%(j)s = %(jump)s;
-                //printf("%(var)s_jump%(index)s_%(j)s is:");
-                //std::cout << %(var)s_jump%(index)s_%(j)s << std::endl;
                 """ % locals()
                 adjust = "0"
     check = ""
@@ -169,6 +165,8 @@ def make_alloc(loop_orders, dtype, sub, fortran='0'):
                 PyErr_Clear();
                 Py_XDECREF(%(olv)s);
                 %(olv)s = (PyArrayObject*)PyArray_EMPTY(%(nd)s, dims, %(type)s, 0);
+            } else {
+                Py_DECREF(success);
             }
         }
         if (!%(olv)s) {

theano/tensor/tests/test_elemwise.py

@@ -15,7 +15,7 @@ from theano import gof, scalar, config
 
 from theano import tensor
 from theano.tensor import TensorType, as_tensor_variable
-from theano.compile.mode import get_default_mode
+from theano.compile.mode import get_default_mode, Mode
 from theano.tensor.elemwise import (CAReduce, Elemwise, DimShuffle,
                                     Prod, ProdWithoutZeros)
 from theano.tests import unittest_tools
@@ -365,7 +365,7 @@ class test_CAReduce(unittest_tools.InferShapeTester):
              ((), ())]
     type = TensorType
 
-    def with_linker(self, linker, scalar_op=scalar.add, dtype="floatX",
+    def with_mode(self, mode, scalar_op=scalar.add, dtype="floatX",
                   pre_scalar_op=None,
                   test_nan=False, tensor_op=None):
         for xsh, tosum in self.cases:
@@ -383,7 +383,7 @@ class test_CAReduce(unittest_tools.InferShapeTester):
             if tosum is None:
                 tosum = list(range(len(xsh)))
 
-            f = copy(linker).accept(FunctionGraph([x], [e])).make_function()
+            f = theano.function([x], e, mode=mode)
             xv = np.asarray(np.random.rand(*xsh))
 
             if dtype not in tensor.discrete_dtypes:
@@ -495,8 +495,7 @@ class test_CAReduce(unittest_tools.InferShapeTester):
                 e = tensor_op(x, axis=tosum)
             if tosum is None:
                 tosum = list(range(len(xsh)))
-            f = copy(linker).accept(FunctionGraph([x],
-                                                  [e.shape])).make_function()
+            f = theano.function([x], e.shape, mode=mode)
             if not(scalar_op in [scalar.maximum, scalar.minimum] and
                    ((xsh == () or np.prod(xsh) == 0))):
                 try:
@@ -505,69 +504,78 @@ class test_CAReduce(unittest_tools.InferShapeTester):
                     # GpuCAReduce don't implement all cases when size is 0
                     assert xv.size == 0
 
+    def test_perform_noopt(self):
+        self.with_mode(Mode(linker='py', optimizer=None), scalar.add, dtype='floatX')
+
     def test_perform(self):
         for dtype in ["bool", "floatX", "complex64", "complex128", "int8", "uint8"]:
-            self.with_linker(gof.PerformLinker(), scalar.add, dtype=dtype)
-            self.with_linker(gof.PerformLinker(), scalar.mul, dtype=dtype)
-            self.with_linker(gof.PerformLinker(), scalar.maximum, dtype=dtype)
-            self.with_linker(gof.PerformLinker(), scalar.minimum, dtype=dtype)
-            self.with_linker(gof.PerformLinker(), scalar.and_, dtype=dtype,
+            self.with_mode(Mode(linker='py'), scalar.add, dtype=dtype)
+            self.with_mode(Mode(linker='py'), scalar.mul, dtype=dtype)
+            self.with_mode(Mode(linker='py'), scalar.maximum, dtype=dtype)
+            self.with_mode(Mode(linker='py'), scalar.minimum, dtype=dtype)
+            self.with_mode(Mode(linker='py'), scalar.and_, dtype=dtype,
                            tensor_op=tensor.all)
-            self.with_linker(gof.PerformLinker(), scalar.or_, dtype=dtype,
+            self.with_mode(Mode(linker='py'), scalar.or_, dtype=dtype,
                            tensor_op=tensor.any)
         for dtype in ["int8", "uint8"]:
-            self.with_linker(gof.PerformLinker(), scalar.or_, dtype=dtype)
-            self.with_linker(gof.PerformLinker(), scalar.and_, dtype=dtype)
-            self.with_linker(gof.PerformLinker(), scalar.xor, dtype=dtype)
+            self.with_mode(Mode(linker='py'), scalar.or_, dtype=dtype)
+            self.with_mode(Mode(linker='py'), scalar.and_, dtype=dtype)
+            self.with_mode(Mode(linker='py'), scalar.xor, dtype=dtype)
 
     def test_perform_nan(self):
         for dtype in ["floatX", "complex64", "complex128"]:
-            self.with_linker(gof.PerformLinker(), scalar.add, dtype=dtype,
+            self.with_mode(Mode(linker='py'), scalar.add, dtype=dtype,
                            test_nan=True)
-            self.with_linker(gof.PerformLinker(), scalar.mul, dtype=dtype,
+            self.with_mode(Mode(linker='py'), scalar.mul, dtype=dtype,
                            test_nan=True)
-            self.with_linker(gof.PerformLinker(), scalar.maximum, dtype=dtype,
+            self.with_mode(Mode(linker='py'), scalar.maximum, dtype=dtype,
                            test_nan=True)
-            self.with_linker(gof.PerformLinker(), scalar.minimum, dtype=dtype,
+            self.with_mode(Mode(linker='py'), scalar.minimum, dtype=dtype,
                            test_nan=True)
-            self.with_linker(gof.PerformLinker(), scalar.or_, dtype=dtype,
+            self.with_mode(Mode(linker='py'), scalar.or_, dtype=dtype,
                            test_nan=True, tensor_op=tensor.any)
-            self.with_linker(gof.PerformLinker(), scalar.and_, dtype=dtype,
+            self.with_mode(Mode(linker='py'), scalar.and_, dtype=dtype,
                            test_nan=True, tensor_op=tensor.all)
 
+    def test_c_noopt(self):
+        # We need to make sure that we cover the corner cases that
+        # optimizations normally cover
+        if not theano.config.cxx:
+            raise SkipTest("G++ not available, so we need to skip this test.")
+        self.with_mode(Mode(linker='c', optimizer=None), scalar.add, dtype='floatX')
+
     @attr('slow')
     def test_c(self):
         if not theano.config.cxx:
             raise SkipTest("G++ not available, so we need to skip this test.")
-
         for dtype in ["bool", "floatX", "complex64", "complex128", "int8", "uint8"]:
-            self.with_linker(gof.CLinker(), scalar.add, dtype=dtype)
-            self.with_linker(gof.CLinker(), scalar.mul, dtype=dtype)
+            self.with_mode(Mode(linker='c'), scalar.add, dtype=dtype)
+            self.with_mode(Mode(linker='c'), scalar.mul, dtype=dtype)
         for dtype in ["bool", "floatX", "int8", "uint8"]:
-            self.with_linker(gof.CLinker(), scalar.minimum, dtype=dtype)
-            self.with_linker(gof.CLinker(), scalar.maximum, dtype=dtype)
-            self.with_linker(gof.CLinker(), scalar.and_, dtype=dtype,
+            self.with_mode(Mode(linker='c'), scalar.minimum, dtype=dtype)
+            self.with_mode(Mode(linker='c'), scalar.maximum, dtype=dtype)
+            self.with_mode(Mode(linker='c'), scalar.and_, dtype=dtype,
                            tensor_op=tensor.all)
-            self.with_linker(gof.CLinker(), scalar.or_, dtype=dtype,
+            self.with_mode(Mode(linker='c'), scalar.or_, dtype=dtype,
                            tensor_op=tensor.any)
         for dtype in ["bool", "int8", "uint8"]:
-            self.with_linker(gof.CLinker(), scalar.or_, dtype=dtype)
-            self.with_linker(gof.CLinker(), scalar.and_, dtype=dtype)
-            self.with_linker(gof.CLinker(), scalar.xor, dtype=dtype)
+            self.with_mode(Mode(linker='c'), scalar.or_, dtype=dtype)
+            self.with_mode(Mode(linker='c'), scalar.and_, dtype=dtype)
+            self.with_mode(Mode(linker='c'), scalar.xor, dtype=dtype)
 
     @attr('slow')
     def test_c_nan(self):
         if not theano.config.cxx:
             raise SkipTest("G++ not available, so we need to skip this test.")
         for dtype in ["floatX", "complex64", "complex128"]:
-            self.with_linker(gof.CLinker(), scalar.add, dtype=dtype,
+            self.with_mode(Mode(linker='c'), scalar.add, dtype=dtype,
                            test_nan=True)
-            self.with_linker(gof.CLinker(), scalar.mul, dtype=dtype,
+            self.with_mode(Mode(linker='c'), scalar.mul, dtype=dtype,
                            test_nan=True)
         for dtype in ["floatX"]:
-            self.with_linker(gof.CLinker(), scalar.minimum, dtype=dtype,
+            self.with_mode(Mode(linker='c'), scalar.minimum, dtype=dtype,
                            test_nan=True)
-            self.with_linker(gof.CLinker(), scalar.maximum, dtype=dtype,
+            self.with_mode(Mode(linker='c'), scalar.maximum, dtype=dtype,
                            test_nan=True)
 
     def test_infer_shape(self, dtype=None, pre_scalar_op=None):
@@ -1241,16 +1249,6 @@ def test_gt_grad():
     cost = (scores * (scores > 0)).sum()
     T.grad(cost, input_)
 
-"""
-if __name__ == '__main__':
-    #unittest.main()
-    suite = unittest.TestSuite([test_Prod('test_mul_without_zeros_zeros')])
-    #suite.addTest(test_Prod('test_verify_grad_with_zeros'))
-    #suite.addTest(test_Prod('test_prod_without_zeros'))
-    #suite.addTest(test_Prod('test_other_grad_tests'))
-    unittest.TextTestRunner().run(suite)
-"""
-
 
 def test_clip_grad():