added ProfileMode to sandbox/wraplinker

theano/sandbox/wraplinker.py

 from __future__ import absolute_import
+import time
 import numpy
 
 from ..gof.link import WrapLinker
-from ..gradient import numeric_grad
+from ..compile.mode import Mode
 
 class Todo(Exception): """todo"""
 
 #WrapLinker wrappers
-def cmp_outputs(i, node, *thunks):
+if 0:
+    from ..gradient import numeric_grad
+    def cmp_outputs(i, node, *thunks):
         """WrapLinker wrapper: raise an exception if outputs are different
 
         numpy.ndarrays of floating point types are compared approximately, rather
@@ -98,3 +101,51 @@ def WrapLinkerMany(linkers, wrappers):
 def DualLinker(linkers):
     return WrapLinkerMany(linkers, [run_all, cmp_outputs])
 
+
+class ProfileMode(Mode):
+    def __init__(self, local_linker, optimizer=None):
+        local_time = [0.0]
+        apply_time = {}
+        op_time = {}
+
+        def blah(i, node, *thunks):
+            t0 = time.time() 
+            for th in thunks:
+                th()
+            dt = time.time() - t0
+            local_time[0] += dt
+            apply_time[(i,node.op)] = apply_time.get((i,node.op), 0.0) + dt
+            op_time[node.op] = op_time.get(node.op, 0.0) + dt
+
+        self.local_time = local_time
+        self.apply_time = apply_time
+        self.op_time = op_time
+
+        linker = WrapLinkerMany([local_linker], [blah])
+        if optimizer:
+            Mode.__init__(self, linker, optimizer)
+        else:
+            Mode.__init__(self, linker)
+
+    def print_summary(self):
+        local_time = self.local_time[0]
+        apply_time = self.apply_time
+        op_time = self.op_time
+
+        print 'local_time', local_time
+        print 'apply-wise times'
+        atimes = [(t/local_time, (a[0], str(a[1]))) for a, t in apply_time.items()]
+        atimes.sort()
+        atimes.reverse()
+        for t,a in atimes[:15]:
+            print '  ', t, a
+        print '   ...'  #show that we are ignoring applies that don't take much time
+        print 'op-wise times'
+        otimes = [(t/local_time, a) for a, t in op_time.items()]
+        otimes.sort()
+        otimes.reverse()
+        for t,a in otimes[:15]:
+            print '  ', t, a
+        print '   ...'  #show that we are ignoring applies that don't take much time
+        print sum(t for a,t in op_time.items())
+