Added comments

llvm/projects/hpvm-tensor-rt/code_autogenerators/source_code_autogenerator.py

@@ -11,7 +11,18 @@ import os
 import re
 
 class KnobConfiguration:
+    '''
+    Stores the configurations as well as other useful information for each knob configuration
+    Stores: id (may factor out if ids are guaranteed to start at 0/1 and be consecutive)
+            original function name
+            modified function name
+            new function parameters (knobs)
+            new function call (modified function name(knobs)) 
+    '''
     def __init__(self, raw_config):
+        '''
+        Args: raw_config = line of configuration file to parse
+        '''
         line_as_lst = raw_config.strip().split()
 
         self.id = int(line_as_lst[0])
@@ -22,56 +33,69 @@ class KnobConfiguration:
         # Build the new function call
         self.new_func_call = "%s(%s)" % (self.modified_func_name, ", ".join(self.params)) 
         
-    def __repr__(self):
-        return "%s %s %s %s %s" % (self.id, self.modified_func_name, self.params, self.orig_func_name, self.new_func_call)
 
-
-# Assumption: Original function call = the new function without the <approx><conv>
-# Ex: tensorConvPerfCuda replaces tensorConv  
-# TODO might be better if we just create a col in the table for this ...
-def generate_source_code(table, filename, dir_name, source_name):
+def generate_source_code(table, dir_name, filename, source_name):
+    '''
+    Generates source code for all configurations in the table for one original source 
+    Args
+        table: List of KnobConfigurations
+        dir_name: Directory new sources should be placed in
+        filename: Filename of original source
+        source_name: Filename without the file extension (ex: foo/blah.cc --> blah)
+    '''
     source_file = open(filename, "r") 
     for knob_config in table:
-        print("KNOB CONFIG: ", knob_config)
         source_file.seek(0, 0)
        
         new_file_contents = []
 
-        curr_line = ""
+        # Store complete line to handle cases where one line of code is split into two lines
+        complete_line = ""
         for line in source_file:
             # Handles case where 1 actual line of code is split into 2 lines 
             if len(line) >= 2 and not line.startswith("#") and \
                         line.find(";") == -1: # Last char is always \n
-                curr_line += line
+                complete_line += line
                 continue
 
-            curr_line += line
-            orig_func_ind = curr_line.find(knob_config.orig_func_name)
+            complete_line += line
+            orig_func_ind = complete_line.find(knob_config.orig_func_name)
 
             if orig_func_ind != -1:
                 # Replace from an instance of origOp to first instance of ) after origOp
+                # with the new function call
                 new_line = []
                 line_start_ind = 0 
                 last_ind = 0
+
                 while orig_func_ind != -1:
-                    new_line.append(curr_line[line_start_ind : orig_func_ind])
+                    new_line.append(complete_line[line_start_ind : orig_func_ind])
                     new_line.append(knob_config.new_func_call)
-                    line_start_ind = curr_line.find(")", orig_func_ind) + 1
-                    orig_func_ind = curr_line.find(knob_config.orig_func_name, line_start_ind)
-                new_line.append(curr_line[line_start_ind : ])
+                    line_start_ind = complete_line.find(")", orig_func_ind) + 1
+                    orig_func_ind = complete_line.find(knob_config.orig_func_name, line_start_ind)
+
+                new_line.append(complete_line[line_start_ind : ])
                 new_file_contents.append(''.join(new_line))
             else:
-                new_file_contents.append(curr_line)
-            curr_line = ""
+                new_file_contents.append(complete_line)
+            complete_line = ""
 
         new_filename = os.path.join(dir_name, "%s_%s.cc" % (source_name, knob_config.id))
         new_file = open(new_filename, "w")
         new_file.write(''.join(new_file_contents))
         new_file.close()
+
     source_file.close()
 
 
 def generate_all_sources(table, orig_files_filename):
+    '''
+    Generates directories and source code for all original sources for all knob configurations
+    Args:
+        table: List of KnobConfiguration objects
+        orig_files_filename: Filename of file containing all original source names to generate new
+               sources for 
+    '''
     orig_files = open(orig_files_filename, "r")
     for orig_filename in orig_files:
         orig_filename = orig_filename.strip()
@@ -97,12 +121,15 @@ def generate_all_sources(table, orig_files_filename):
             print("GENERATING DIRECTORY: %s" % dir_name)
             os.makedirs(dir_name)
             
-        generate_source_code(table, orig_filename, dir_name, source_name)
+        generate_source_code(table, dir_name, orig_filename, source_name)
     orig_files.close()
 
 
 def parse_table(table_filename):
-    # can we assume that the ids always start at 1 --> if so, can index by knobs 
+    '''
+    Given the filename of a table, parses the table into a list of KnobConfigurations 
+    '''
+    # Can we assume that the ids always start at 1 --> if so, can index by knobs 
     # else: need to use a dict
     table = []
     table_file = open(table_filename, "r")
@@ -123,10 +150,5 @@ if __name__ == "__main__":
             print("Run with --usage flag for more detailed information")
         exit(1)
 
-    # How do we know what function names correspond to what files?
-    # Assume that for the given filenames, we replace all function names with the corresp func name --> then we can't run this for different functions, need to replace the modified func name
-
-    # Parse the table first to avoid having to reparse it for each file to change 
-    
     table = parse_table(sys.argv[1])
     generate_all_sources(table, sys.argv[2])