Merge pull request #5652 from lamblin/fix_small_tests

.jenkins/jenkins_test2.sh

@@ -75,6 +75,7 @@ THEANO_GPUARRAY_TESTS="theano/gpuarray/tests \
                        theano/sandbox/tests/test_rng_mrg.py:test_consistency_GPUA_serial \
                        theano/sandbox/tests/test_rng_mrg.py:test_consistency_GPUA_parallel \
                        theano/sandbox/tests/test_rng_mrg.py:test_GPUA_full_fill \
-                       theano/scan_module/tests/test_scan.py:T_Scan_Gpuarray"
+                       theano/scan_module/tests/test_scan.py:T_Scan_Gpuarray \
+                       theano/scan_module/tests/test_scan_checkpoints.py:TestScanCheckpoint.test_memory"
 FLAGS="init_gpu_device=$DEVICE,gpuarray.preallocate=1000,mode=FAST_RUN,on_opt_error=raise,on_shape_error=raise,cmodule.age_thresh_use=604800"
 THEANO_FLAGS=${FLAGS} time nosetests --with-xunit --xunit-file=theanogpuarray_tests.xml ${THEANO_GPUARRAY_TESTS}

theano/compile/profiling.py

@@ -251,8 +251,9 @@ class ProfileStats(object):
 
     # param is called flag_time_thunks because most other attributes with time
     # in the name are times *of* something, rather than configuration flags.
-    def __init__(self, atexit_print=True, flag_time_thunks=None, **kwargs):
-        if (config.profile and
+    def __init__(self, atexit_print=True, flag_time_thunks=None,
+                 gpu_checks=True, **kwargs):
+        if (config.profile and gpu_checks and
                 ((hasattr(theano, 'sandbox') and
                   hasattr(theano.sandbox, 'cuda') and
                   theano.sandbox.cuda.cuda_enabled) or (
@@ -266,7 +267,7 @@ class ProfileStats(object):
                     " You must set the environment variable"
                     " CUDA_LAUNCH_BLOCKING to 1 to tell the CUDA driver to"
                     " synchronize the execution to get a meaningful profile.")
-        if (config.profile and
+        if (config.profile and gpu_checks and
                 hasattr(theano, 'gpuarray') and
                 theano.gpuarray.pygpu_activated and
                 not config.profiling.ignore_first_call):

theano/compile/tests/test_profiling.py

@@ -35,7 +35,7 @@ class Test_profiling(unittest.TestCase):
             z += [T.outer(x[i], x[i + 1]).sum(axis=1) for i in range(len(x) - 1)]
             z += [x[i] + x[i + 1] for i in range(len(x) - 1)]
 
-            p = theano.ProfileStats(False)
+            p = theano.ProfileStats(False, gpu_checks=False)
 
             if theano.config.mode in ["DebugMode", "DEBUG_MODE", "FAST_COMPILE"]:
                 m = "FAST_RUN"
@@ -87,7 +87,7 @@ class Test_profiling(unittest.TestCase):
 
             z = ifelse(T.lt(a, b), x * 2, y * 2)
 
-            p = theano.ProfileStats(False)
+            p = theano.ProfileStats(False, gpu_checks=False)
 
             if theano.config.mode in ["DebugMode", "DEBUG_MODE", "FAST_COMPILE"]:
                 m = "FAST_RUN"

theano/gof/op.py

@@ -19,7 +19,7 @@ import theano
 from theano import config
 
 import theano.gof.cc
-from six import itervalues
+from six import itervalues, PY3
 from theano.gof import graph
 from theano.gof import utils
 from theano.gof.cmodule import GCC_compiler
@@ -35,6 +35,17 @@ __docformat__ = "restructuredtext en"
 _logger = logging.getLogger('theano.gof.op.Op')
 
 
+# Open file in "universal newline mode".
+# In Python 2, this is done by calling open(..., 'U'), but this is
+# deprected in Python 3 (where we would need to pass "newline=None",
+# which is the default).
+if PY3:
+    _open_u = open
+else:
+    def _open_u(file):
+        return open(file, 'U')
+
+
 class CLinkerObject(object):
     """
     Standard elements of an Op or Type used with the CLinker.
@@ -1297,7 +1308,7 @@ class COp(Op):
         self.func_codes = []
         for func_file in func_files:
             # U (universal) will convert all new lines format to \n.
-            with open(func_file, 'U') as f:
+            with _open_u(func_file) as f:
                 self.func_codes.append(f.read())
 
         # If both the old section markers and the new section markers are

theano/gof/vm.py

@@ -10,7 +10,6 @@ from __future__ import absolute_import, print_function, division
 from . import link
 from collections import defaultdict
 import logging
-import os
 import sys
 import time
 import warnings
@@ -757,21 +756,6 @@ class VM_Linker(link.LocalLinker):
         associated to self, else, a new VM_Linker associated to fgraph.
 
         """
-        if ((config.profile or config.print_global_stats) and
-                ((hasattr(theano, 'sandbox') and
-                  hasattr(theano.sandbox, 'cuda') and
-                  theano.sandbox.cuda.cuda_enabled) or
-                 (hasattr(theano, 'gpuarray') and
-                  theano.gpuarray.pygpu_activated))):
-            if os.environ.get('CUDA_LAUNCH_BLOCKING', '0') != '1':
-                raise Exception(
-                    "You are running the Theano profiler with CUDA enabled."
-                    " Theano GPU ops execution is asynchronous by default."
-                    " So by default, the profile is useless."
-                    " You must set the environment variable"
-                    " CUDA_LAUNCH_BLOCKING to 1 to tell the CUDA driver to"
-                    " synchronize the execution to get a meaningful profile.")
-
         if no_recycling is None:
             no_recycling = []
         if self.fgraph is not None and self.fgraph is not fgraph:

theano/gpuarray/tests/test_dnn.py

@@ -511,6 +511,8 @@ def test_pooling_opt_arbitrary_dimensions():
 
 
 def test_pooling_empty_batch():
+    if not dnn.dnn_available(test_ctx_name):
+        raise SkipTest(dnn.dnn_available.msg)
     img_shp = (0, 5, 6, 8)
     img = T.ftensor4('img')
 

theano/printing.py

@@ -863,7 +863,7 @@ def pydotprint(fct, outfile=None,
         if profile:
             time = profile.apply_time.get(node, 0)
             # second, %fct time in profiler
-            if profile.fct_callcount == 0:
+            if profile.fct_callcount == 0 or profile.fct_call_time == 0:
                 pf = 0
             else:
                 pf = time * 100 / profile.fct_call_time

theano/scan_module/tests/test_scan_checkpoints.py

@@ -53,12 +53,13 @@ class TestScanCheckpoint(unittest.TestCase):
         """Test that scan_checkpoint reduces memory usage."""
         if None not in theano.gpuarray.type.list_contexts():
             return unittest.SkipTest('Requires gpuarray backend.')
+        from theano.gpuarray.tests.config import mode_with_gpu  # noqa
         f = theano.function(inputs=[self.A, self.k],
-                            outputs=self.grad_A)
+                            outputs=self.grad_A, mode=mode_with_gpu)
         f_check = theano.function(inputs=[self.A, self.k],
-                                  outputs=self.grad_A_check)
+                                  outputs=self.grad_A_check, mode=mode_with_gpu)
         free_gmem = theano.gpuarray.type._context_reg[None].free_gmem
-        data = numpy.ones(free_gmem / 3000, dtype=numpy.float32)
+        data = numpy.ones(free_gmem // 3000, dtype=numpy.float32)
         # Check that it works with the checkpoints
         f_check(data, 1000)
         # Check that the basic scan fails in that case

theano/tests/test_printing.py

@@ -91,7 +91,7 @@ def test_pydotprint_profile():
         raise SkipTest('pydot not available')
 
     A = tensor.matrix()
-    prof = theano.compile.ProfileStats(atexit_print=False)
+    prof = theano.compile.ProfileStats(atexit_print=False, gpu_checks=False)
     f = theano.function([A], A + 1, profile=prof)
     theano.printing.pydotprint(f, print_output_file=False)
     f([[1]])