Merge pull request #634 from pascanur/fix_scan_debugmode

theano/scan_module/scan.py

@@ -949,7 +949,6 @@ def scan(fn,
     info['truncate_gradient'] = truncate_gradient
     info['name'] = name
     info['mode'] = mode
-    info['inplace'] = -1
     info['destroy_map'] = {}
     info['gpu'] = False
     info['as_while'] = as_while

theano/scan_module/scan_op.py

@@ -528,6 +528,10 @@ class Scan(PureOp):
                                                     dtype='int32')
             cython_vector_outs = numpy.asarray(self.vector_outs,
                                                     dtype='int32')
+            cython_destroy_map = [x in self.destroy_map
+                                  for x in xrange(len(node.outputs))]
+            cython_destroy_map = numpy.asarray(cython_destroy_map,
+                                               dtype='int32')
             import scan_perform_ext
             p = lambda node, args, outs:\
                     scan_perform_ext.perform(
@@ -549,7 +553,7 @@ class Scan(PureOp):
                         cython_mit_mot_out_nslices,
                         self.fn.fn,
                         self.fn,
-                        self.inplace,
+                        cython_destroy_map,
                         args,
                         outs,
                         self)
@@ -782,7 +786,7 @@ class Scan(PureOp):
                          in xrange(self.n_outs + self.n_nit_sot)]
         # 2.1 Create storage space for outputs
         for idx in xrange(self.n_outs):
-            if self.inplace:
+            if idx in self.destroy_map:
                 # ^ Case 1. Outputs should be computed inplace of their
                 # initial state
                 outs[idx][0] = args[self.seqs_arg_offset + idx]
@@ -1451,7 +1455,6 @@ class Scan(PureOp):
         else:
             info['name'] = None
         info['mode'] = self.mode
-        info['inplace'] = False
         n_mit_sot = 0
         n_sit_sot = 0
 
@@ -1552,7 +1555,6 @@ class Scan(PureOp):
         else:
             info['name'] = None
         info['mode'] = self.mode
-        info['inplace'] = False
         info['mit_mot_out_slices'] = self.mit_mot_out_slices * 2
         new_tap_array = []
         b = 0

theano/scan_module/scan_opt.py

@@ -867,7 +867,6 @@ class ScanMerge(gof.Optimizer):
         info['truncate_gradient'] = nodes[0].op.truncate_gradient
         info['name'] = '&'.join([nd.op.name for nd in nodes])
         info['mode'] = nodes[0].op.mode
-        info['inplace'] = False
         info['gpu'] = False
         info['as_while'] = as_while
         info['profile'] = nodes[0].op.profile

theano/scan_module/scan_perform.pyx

@@ -63,7 +63,7 @@ from theano.sandbox import cuda
 
 
 def get_version():
-    return 0.266
+    return 0.276
 
 @cython.boundscheck(False)
 def perform(
@@ -85,7 +85,7 @@ def perform(
             numpy.ndarray[numpy.int32_t,ndim=1] mit_mot_out_nslices,
             fn,
             fnct,
-            bint inplace,
+            numpy.ndarray[numpy.int32_t,ndim=1] destroy_map,
             args,
             outs,
             self):
@@ -145,9 +145,8 @@ def perform(
     fnct: python object
         Only used to attach some timings for the profile mode ( can be
         skiped if we don't care about Theano's profile mode)
-    inplace
-        Boolean that says if things should be computed inplace or if they
-        should not.
+    destroy_map
+        Array of boolean saying if an output is computed inplace
     args: list of ndarrays (and random states)
         The inputs of scan in a given order ( n_steps, sequences, mit_mot,
         mit_sot, sit_sot, nit_sot, shared_outs, other_args)
@@ -230,7 +229,7 @@ def perform(
 
     # 2.1 Create storage space for outputs
     for idx in range(n_outs):
-        if inplace:
+        if destroy_map[idx] != 0:
             # ^ Case 1. Outputs should be computed inplace of their
             # initial state
             outs[idx][0] = args[ <unsigned int>(1+ n_seqs + idx)]

theano/scan_module/scan_perform_ext.py

@@ -14,7 +14,7 @@ logging.basicConfig(level=logging.DEBUG)
 if config.compiledir not in sys.path:
     sys.path.append(config.compiledir)
 
-version = 0.266  # must match constant returned in function get_version()
+version = 0.276  # must match constant returned in function get_version()
 
 need_reload = False
 try:

theano/scan_module/scan_utils.py

@@ -592,7 +592,6 @@ def compress_outs(op, not_required, inputs):
     info['n_nit_sot'] = 0
     info['truncate_gradient'] = op.info['truncate_gradient']
     info['name'] = op.info['name']
-    info['inplace'] = op.info['inplace']
     info['gpu'] = op.info['gpu']
     info['mode'] = op.info['mode']
     info['as_while'] = op.info['as_while']
@@ -857,8 +856,9 @@ class scan_args(object):
         q += n_shared_outs
 
         self.other_info = dict()
-        for k in ('truncate_gradient', 'name', 'mode', 'inplace',
+        for k in ('truncate_gradient', 'name', 'mode', 'destroy_map',
                   'gpu', 'as_while', 'profile'):
+            if k in info:
                 self.other_info[k] = info[k]
 
     inner_inputs = property(lambda self: (self.inner_in_seqs +