[inductor] Log precompilation time (pytorch#136395)

This has been useful for diagnosing the long compile time issues I've seen in the Triton CPU backend.

Pull Request resolved: pytorch#136395
Approved by: https://github.com/eellison

torch/_inductor/select_algorithm.py

@@ -1283,7 +1283,9 @@ def no_op(*args, **kwargs):
 
             def precompile_with_captured_stdout(choice):
                 with restore_stdout_stderr(initial_stdout, initial_stderr):
-                    return choice.precompile()
+                    start_time = time.time()
+                    choice.precompile()
+                    return time.time() - start_time
 
             executor = ThreadPoolExecutor(max_workers=num_workers)
 
@@ -1305,6 +1307,12 @@ def wait_on_futures():
                         log.error(
                             "Exception %s for benchmark choice %s", e, futures[future]
                         )
+                    else:
+                        log.info(
+                            "Precompiling benchmark choice %s took %.02fs",
+                            futures[future],
+                            future.result(),
+                        )
 
                 executor.shutdown(wait=True)