Merge pull request #4961 from nke001/ccw4102_cpy

theano/compile/profilemode.py

@@ -693,6 +693,23 @@ Test them first, as they are not guaranteed to always provide a speedup.""")
                           " generator supported on the GPU.")
                 break
 
+        # tip 6
+        import theano.sandbox.cuda as cuda
+        from theano.tensor.nnet import LogSoftmax
+        import theano.tensor.signal.pool as pool
+        import theano.gpuarray
+
+        for a, t in iteritems(apply_time):
+            node = a[1]
+            if (isinstance(node.op, pool.Pool)):
+                if (not cuda.dnn.dnn_available() and not theano.gpuarray.dnn.dnn_present()):
+                    print("Install CuDNN to do pooling faster"
+                          "this allows the operation to run on GPU")
+            if (isinstance(node.op, LogSoftmax)):
+                if (not cuda.dnn.dnn_available() and not theano.gpuarray.dnn.dnn_present()):
+                    print("Install CuDNN to do LogSoftmax faster"
+                          "this allows the operation to run on GPU")
+
         if not printed_tip:
             print("  Sorry, no tip for today.")
 

theano/compile/profiling.py

@@ -123,14 +123,14 @@ class ProfileStats(object):
     """
     def reset(self):
         """ Ignore previous function call"""
-        #self.compile_time = 0.
+        # self.compile_time = 0.
         self.fct_call_time = 0.
         self.fct_callcount = 0
         self.vm_call_time = 0.
         self.apply_time = {}
         self.apply_callcount = {}
         # self.apply_cimpl = None
-        #self.messge = None
+        # self.messge = None
     #
     # Note on implementation:
     # Class variables are used here so that each one can be
@@ -1072,6 +1072,7 @@ class ProfileStats(object):
 
             # Store the max of some stats by any function in this profile.
             max_sum_size = max(max_sum_size, sum_size)
+            
             def compute_max_stats(running_memory, stats):
                 (max_node_memory_size,
                  max_running_max_memory_size,
@@ -1261,8 +1262,9 @@ class ProfileStats(object):
     def print_tips(self, file):
         print("""Here are tips to potentially make your code run faster
                  (if you think of new ones, suggest them on the mailing list).
-                 Test them first, as they are not guaranteed to always provide a speedup.""", file = file)
+                 Test them first, as they are not guaranteed to always provide a speedup.""", file=file)
 
+        import theano
         RandomFunction = theano.tensor.raw_random.RandomFunction
         scal = theano.scalar
         T = theano.tensor
@@ -1310,7 +1312,7 @@ class ProfileStats(object):
                         return True
                     elif s_op.__class__ not in scalar_op_amdlibm_no_speed_up:
                         print("We don't know if amdlibm will accelerate "
-                              "this scalar op.", s_op , file = file)
+                              "this scalar op.", s_op, file=file)
                 return False
 
         def exp_float32_op(op):
@@ -1323,7 +1325,7 @@ class ProfileStats(object):
         printed_tip = False
         # tip 1
         if config.floatX == 'float64':
-            print("  - Try the Theano flag floatX=float32", file = file)
+            print("  - Try the Theano flag floatX=float32", file=file)
             printed_tip = True
 
         # tip 2
@@ -1331,7 +1333,7 @@ class ProfileStats(object):
                                            in self.apply_time]):
             print("  - Try installing amdlibm and set the Theano flag "
                   "lib.amdlibm=True. This speeds up only some Elemwise "
-                  "operation.", file = file)
+                  "operation.", file=file)
             printed_tip = True
 
         # tip 3
@@ -1340,7 +1342,7 @@ class ProfileStats(object):
                                            for a in self.apply_time]):
             print("  - With the default gcc libm, exp in float32 is slower "
                   "than in float64! Try Theano flag floatX=float64, or "
-                  "install amdlibm and set the theano flags lib.amdlibm=True", file = file)
+                  "install amdlibm and set the theano flags lib.amdlibm=True", file=file)
             printed_tip = True
 
         # tip 4
@@ -1353,7 +1355,7 @@ class ProfileStats(object):
                       " dot22 (which is faster). Make sure the inputs are "
                       "float32 or float64, and are the same for both inputs. "
                       "Currently they are: %s" %
-                      [i.type for i in node.inputs], file = file)
+                      [i.type for i in node.inputs], file=file)
                 printed_tip = True
 
         # tip 5
@@ -1364,24 +1366,42 @@ class ProfileStats(object):
                 print("  - Replace the default random number generator by "
                       "'from theano.sandbox.rng_mrg import MRG_RandomStreams "
                       "as RandomStreams', as this is is faster. It is still "
-                      "experimental, but seems to work correctly.", file = file)
+                      "experimental, but seems to work correctly.", file=file)
                 if config.device.startswith("gpu"):
                     print("     - MRG_RandomStreams is the only random number"
-                          " generator supported on the GPU.", file = file)
+                          " generator supported on the GPU.", file=file)
                 break
 
         # tip 6
         for a in self.apply_time:
             node = a
-            if (isinstance(node.op, T.Dot) and
-                len(set(i.dtype for i in node.inputs)) != 1):
+            if (isinstance(node.op, T.Dot) and len(set(i.dtype for i in node.inputs)) != 1):
                 print("  - You have a dot operation that has different dtype "
                       " for inputs (%s). Make sure that the inputs have same "
-                      " dtype." % [i.type for i in node.inputs], file = file)
+                      " dtype." % [i.type for i in node.inputs], file=file)
+                printed_tip = True
+
+        # tip 7
+        import theano.sandbox.cuda as cuda
+        from theano.tensor.nnet import LogSoftmax
+        import theano.tensor.signal.pool as pool
+        import theano.gpuarray
+
+        for a in self.apply_time:
+            node = a
+            if (isinstance(node.op, pool.Pool)):
+                if (not cuda.dnn.dnn_available() and not theano.gpuarray.dnn.dnn_present()):
+                    print("Install CuDNN to do pooling faster"
+                          "this allows the operation to run on GPU")
+                    printed_tip = True
+            if (isinstance(node.op, LogSoftmax)):
+                if (not cuda.dnn.dnn_available() and not theano.gpuarray.dnn.dnn_present()):
+                    print("Install CuDNN to do LogSoftmax faster"
+                          "this allows the operation to run on GPU")
                     printed_tip = True
 
         if not printed_tip:
-            print("  Sorry, no tip for today.", file = file)
+            print("  Sorry, no tip for today.", file=file)
 
 
 class ScanProfileStats(ProfileStats):