rank only, change formula

codeflash/benchmarking/function_ranker.py

@@ -14,19 +14,17 @@ if TYPE_CHECKING:
 
 
 class FunctionRanker:
-    """Ranks and filters functions for optimization based on profiling trace data using the ttX scoring method.
+    """Ranks and filters functions based on a ttX score derived from profiling data.
 
-    The FunctionRanker analyzes function-level timing statistics from a trace file and assigns a ttX score to each function:
+    The ttX score is calculated as:
+        ttX = own_time + (time_spent_in_callees / call_count)
 
-        ttX = own_time + (time_spent_in_callees x call_count)
+    This score prioritizes functions that are computationally heavy themselves (high `own_time`)
+    or that make expensive calls to other functions (high average `time_spent_in_callees`).
 
-    This scoring prioritizes functions that:
-      1. Consume significant time themselves (own_time)
-      2. Are called frequently and have expensive subcalls (time_spent_in_callees x call_count)
-
-    first, filters out functions whose own_time is less than a specified percentage (importance_threshold = minimum fraction of total runtime a function must account for to be considered important) of the total runtime, considering them unimportant for optimization.
-
-    The remaining functions are then ranked in descending order by their ttX score, prioritizing those most likely to yield performance improvements if optimized.
+    Functions are first filtered by an importance threshold based on their `own_time` as a
+    fraction of the total runtime. The remaining functions are then ranked by their ttX score
+    to identify the best candidates for optimization.
     """
 
     def __init__(self, trace_file_path: Path) -> None:
@@ -59,7 +57,7 @@ class FunctionRanker:
                 time_in_callees_ns = cumulative_time_ns - total_time_ns
 
                 # Calculate ttX score
-                ttx_score = own_time_ns + (time_in_callees_ns * call_count)
+                ttx_score = own_time_ns + (time_in_callees_ns / call_count)
 
                 function_key = f"{filename}:{qualified_name}"
                 self._function_stats[function_key] = {
@@ -99,11 +97,27 @@ class FunctionRanker:
         return stats["ttx_score"] if stats else 0.0
 
     def rank_functions(self, functions_to_optimize: list[FunctionToOptimize]) -> list[FunctionToOptimize]:
-        return sorted(functions_to_optimize, key=self.get_function_ttx_score, reverse=True)
+        ranked = sorted(functions_to_optimize, key=self.get_function_ttx_score, reverse=True)
+        logger.info(
+            f"Function ranking order: {[f'{func.function_name} (ttX={self.get_function_ttx_score(func):.2f})' for func in ranked]}"
+        )
+        return ranked
 
     def get_function_stats_summary(self, function_to_optimize: FunctionToOptimize) -> dict | None:
         return self._get_function_stats(function_to_optimize)
 
+    def rerank_functions(self, functions_to_optimize: list[FunctionToOptimize]) -> list[FunctionToOptimize]:
+        """Ranks functions based on their ttX score.
+
+        This method calculates the ttX score for each function and returns
+        the functions sorted in descending order of their ttX score.
+        """
+        if not self._function_stats:
+            logger.warning("No function stats available to rank functions.")
+            return []
+
+        return self.rank_functions(functions_to_optimize)
+
     def rerank_and_filter_functions(self, functions_to_optimize: list[FunctionToOptimize]) -> list[FunctionToOptimize]:
         """Reranks and filters functions based on their impact on total runtime.
 

codeflash/discovery/functions_to_optimize.py

@@ -218,7 +218,7 @@ def get_functions_to_optimize(
                 all_functions.extend(file_functions)
 
             if all_functions:
-                ranked_functions = ranker.rerank_and_filter_functions(all_functions)
+                ranked_functions = ranker.rank_functions(all_functions)
                 functions_count = len(ranked_functions)
 
                 ranked_dict = {}

tests/test_function_ranker.py

@@ -58,8 +58,8 @@ def test_load_function_stats(function_ranker):
     # Verify funcA specific values
     assert func_a_stats["function_name"] == "funcA"
     assert func_a_stats["call_count"] == 1
-    assert func_a_stats["own_time_ns"] == 153000
-    assert func_a_stats["cumulative_time_ns"] == 5960000
+    assert func_a_stats["own_time_ns"] == 63000
+    assert func_a_stats["cumulative_time_ns"] == 5443000
 
 
 def test_get_function_ttx_score(function_ranker, workload_functions):
@@ -72,9 +72,9 @@ def test_get_function_ttx_score(function_ranker, workload_functions):
     assert func_a is not None
     ttx_score = function_ranker.get_function_ttx_score(func_a)
     
-    # Expected ttX score: own_time + (time_in_callees * call_count)
-    # = 153000 + ((5960000 - 153000) * 1) = 5960000
-    assert ttx_score == 5960000
+    # Expected ttX score: own_time + (time_in_callees / call_count)
+    # = 63000 + ((5443000 - 63000) / 1) = 5443000
+    assert ttx_score == 5443000
 
 
 def test_rank_functions(function_ranker, workload_functions):
@@ -112,9 +112,9 @@ def test_get_function_stats_summary(function_ranker, workload_functions):
     
     assert stats is not None
     assert stats["function_name"] == "funcA"
-    assert stats["own_time_ns"] == 153000
-    assert stats["cumulative_time_ns"] == 5960000
-    assert stats["ttx_score"] == 5960000
+    assert stats["own_time_ns"] == 63000
+    assert stats["cumulative_time_ns"] == 5443000
+    assert stats["ttx_score"] == 5443000
 
 
 
@@ -134,8 +134,8 @@ def test_importance_calculation(function_ranker):
     assert func_a_stats is not None
     importance = func_a_stats["own_time_ns"] / total_program_time
     
-    # funcA importance should be approximately 1.0% (153000/15281000)
-    assert abs(importance - 0.01001) < 0.001
+    # funcA importance should be approximately 0.57% (63000/10968000)
+    assert abs(importance - 0.0057) < 0.001
 
 
 def test_simple_model_predict_stats(function_ranker, workload_functions):
@@ -152,15 +152,15 @@ def test_simple_model_predict_stats(function_ranker, workload_functions):
     assert stats is not None
     assert stats["function_name"] == "predict"
     assert stats["call_count"] == 1
-    assert stats["own_time_ns"] == 2368000
-    assert stats["cumulative_time_ns"] == 4103000
-    assert stats["ttx_score"] == 4103000
+    assert stats["own_time_ns"] == 2289000
+    assert stats["cumulative_time_ns"] == 4017000
+    assert stats["ttx_score"] == 4017000
     
     # Test ttX score calculation
     ttx_score = function_ranker.get_function_ttx_score(predict_func)
-    # Expected ttX score: own_time + (time_in_callees * call_count)
-    # = 2368000 + ((4103000 - 2368000) * 1) = 4103000
-    assert ttx_score == 4103000
+    # Expected ttX score: own_time + (time_in_callees / call_count)
+    # = 2289000 + ((4017000 - 2289000) / 1) = 4017000
+    assert ttx_score == 4017000
     
     # Test importance calculation for predict function
     total_program_time = sum(
@@ -168,5 +168,5 @@ def test_simple_model_predict_stats(function_ranker, workload_functions):
         if s.get("own_time_ns", 0) > 0
     )
     importance = stats["own_time_ns"] / total_program_time
-    # predict importance should be approximately 15.5% (2368000/15281000)
-    assert abs(importance - 0.155) < 0.01
+    # predict importance should be approximately 20.9% (2289000/10968000)
+    assert abs(importance - 0.209) < 0.01