Include C cache preloading in profile output

Also don't do it if there's no C compilation

Author: ricardoV94

pytensor/compile/function/types.py

@@ -1474,15 +1474,20 @@ def __init__(
         if getattr(mode, "profile", None):
             raise TypeError("profile passed via 'mode'. This isn't supported anymore")
         self.profile = profile
-        if profile:
+        if profile and config.cxx:
             # This is very important:
             # 1) We preload the cache here to not have its timing
             #    included with the rewrites.
             # 2) Do not refresh the cache here by default. It cause
             #    too much execution time during testing as we compile
             #    much more functions then the number of compile c
             #    module.
+            start_get_cache = time.perf_counter()
             pytensor.link.c.basic.get_module_cache().refresh()
+            get_cache_time = time.perf_counter() - start_get_cache
+            self.profile.linker_time += get_cache_time
+            self.profile.preload_cache_time += get_cache_time
+
         # Handle the case where inputs and/or outputs is a single
         # Variable (not in a list)
         unpack_single = False
@@ -1722,7 +1727,8 @@ def orig_function(
 
     """
 
-    t1 = time.perf_counter()
+    if profile:
+        t1 = time.perf_counter()
     mode = pytensor.compile.mode.get_mode(mode)
 
     inputs = list(map(convert_function_input, inputs))
@@ -1755,8 +1761,8 @@ def orig_function(
         with config.change_flags(compute_test_value="off"):
             fn = m.create(defaults)
     finally:
-        t2 = time.perf_counter()
-        if fn and profile:
+        if profile and fn:
+            t2 = time.perf_counter()
             profile.compile_time += t2 - t1
             # TODO: append
             profile.nb_nodes = len(fn.maker.fgraph.apply_nodes)

pytensor/compile/profiling.py

@@ -269,6 +269,9 @@ def reset(self):
     linker_time: float = 0.0
     # time spent linking graph (FunctionMaker.create)
 
+    preload_cache_time: float = 0.0
+    # time spent preloading the cache, so it does not affect rewrites profiling
+
     import_time: float = 0.0
     # time spent in importing compiled python module.
 
@@ -811,6 +814,7 @@ def summary_function(self, file):
             ),
             file=file,
         )
+        print(f"       C-cache preloading {self.preload_cache_time:e}s", file=file)
         print(f"       Import time {self.import_time:e}s", file=file)
         print(
             f"       Node make_thunk time {self.linker_node_make_thunks:e}s", file=file