[freetype2] GSoC-2023-Ahmet 355314817: detecting outliers

[Werner Lemberg]

branch: GSoC-2023-Ahmet
commit 3553148174ceb9c01eec1ca660aa77d5f0bd0b0e
Author: goksu <25721443+goeksu@users.noreply.github.com>
Commit: goksu <25721443+goeksu@users.noreply.github.com>

    detecting outliers
---
 builds/testing.mk               |  2 +-
 src/tools/ftbench/ftbench.c     | 71 ++++++++++++++++++++---------
 src/tools/ftbench/src/tohtml.py | 99 +++++++++++++++++++++++++++++++++++++++--
 3 files changed, 147 insertions(+), 25 deletions(-)

diff --git a/builds/testing.mk b/builds/testing.mk
index b5e79b07a..55288e4af 100644
--- a/builds/testing.mk
+++ b/builds/testing.mk
@@ -3,7 +3,7 @@ FTBENCH_DIR = $(TOP_DIR)/src/tools/ftbench
 FTBENCH_SRC = $(FTBENCH_DIR)/ftbench.c
 FTBENCH_OBJ = $(OBJ_DIR)/bench.$(SO)
 FTBENCH_BIN = $(OBJ_DIR)/bench$E
-FTBENCH_FLAG ?= -c 1000 -w 100
+FTBENCH_FLAG ?= -c 550 -w 50
 INCLUDES = $(TOP_DIR)/include
 FONTS = $(wildcard $(FTBENCH_DIR)/fonts/*.ttf)
 BASELINE_DIR = $(OBJ_DIR)/baseline/
diff --git a/src/tools/ftbench/ftbench.c b/src/tools/ftbench/ftbench.c
index c7126ae9f..8b2188131 100644
--- a/src/tools/ftbench/ftbench.c
+++ b/src/tools/ftbench/ftbench.c
@@ -260,11 +260,18 @@
 #define TIMER_GET( timer )    ( timer )->total
 #define TIMER_RESET( timer )  ( timer )->total = 0
 
+#define CHUNK_SIZE 100
 
+int compare(const void* a, const void* b) {
+    if (*(double*)a > *(double*)b) return 1;
+    if (*(double*)a < *(double*)b) return -1;
+    return 0;
+}
   /*
    * Bench code
    */
 
+
   static void
   benchmark( FT_Face   face,
              btest_t*  test,
@@ -276,6 +283,8 @@
     int       n, done;
     btimer_t  timer, elapsed;
    
+    int NUM_CHUNKS = max_iter / CHUNK_SIZE;
+    double results[NUM_CHUNKS];
 
     if ( test->cache_first )
     {
@@ -283,39 +292,59 @@
       test->bench( &timer, face, test->user_data );
     }
 
+    // Initial warm-up
+    for (n = 0; n < warmup; n++) {
+        test->bench(&timer, face, test->user_data);
+    }
+
     printf( "  %-25s ", test->title );
     fflush( stdout );
 
-    TIMER_RESET( &timer );
-    TIMER_RESET( &elapsed );
-
-    int is_warmup = 1;
 
-    for ( n = 0, done = 0; !max_iter || n < max_iter; n++ )
-    {
-      if ( is_warmup && n == warmup ){
-        is_warmup = 0;
+    for (int chunk = 0; chunk < NUM_CHUNKS; chunk++) {
         TIMER_RESET( &timer );
         TIMER_RESET( &elapsed );
-      }
-    
 
-      TIMER_START( &elapsed );
+        // Execute a chunk of iterations
+        for (n = 0, done = 0; n < CHUNK_SIZE; n++) {
+            TIMER_START( &elapsed );
+            done += test->bench( &timer, face, test->user_data );
+            TIMER_STOP( &elapsed );
 
-      done += test->bench( &timer, face, test->user_data );
+           
+        }
+         if (TIMER_GET( &elapsed ) > 1E6 * max_time) {
+                //break;
+          }
+        results[chunk] = TIMER_GET( &timer );
+    }
 
-      TIMER_STOP( &elapsed );
+    // Sort results for IQR calculation
+    qsort(results, NUM_CHUNKS, sizeof(double), compare);
 
-      
-      if (!is_warmup && TIMER_GET( &elapsed ) > 1E6 * max_time )
-        break;
+    double q1 = results[NUM_CHUNKS / 4];
+    double q3 = results[3 * NUM_CHUNKS / 4];
+    double iqr = q3 - q1;
+    double lower_bound = q1 - 1.5 * iqr;
+    double upper_bound = q3 + 1.5 * iqr;
+
+    double total_time = 0.0;
+    int valid_chunks = 0;
+
+    for (int chunk = 0; chunk < NUM_CHUNKS; chunk++) {
+        if (results[chunk] >= lower_bound && results[chunk] <= upper_bound) {
+            total_time += results[chunk];
+            valid_chunks++;
+        }
     }
 
-    if ( done )
-      printf( "%10.1f microseconds %10d done\n",
-              TIMER_GET( &timer ), done );
-    else
-      printf( "no error-free calls\n" );
+    double average_time = total_time / valid_chunks;
+
+
+    
+    printf( "%10.1f microseconds %10d done\n",
+              average_time, done );
+      
   }
 
 
diff --git a/src/tools/ftbench/src/tohtml.py b/src/tools/ftbench/src/tohtml.py
index 3fc01db60..61205a671 100644
--- a/src/tools/ftbench/src/tohtml.py
+++ b/src/tools/ftbench/src/tohtml.py
@@ -55,6 +55,9 @@ def main():
 
         generate_info_table(html_file, baseline_info, benchmark_info)
 
+        # Generate total results table
+        generate_total_results_table(html_file, BASELINE_DIR, BENCHMARK_DIR)
+        
         # Generate results tables
         for filename in os.listdir(BASELINE_DIR):
             if filename.endswith(".txt") and not filename == "info.txt":
@@ -64,6 +67,8 @@ def main():
                 generate_results_table(
                     html_file, baseline_results, benchmark_results, filename
                 )
+       
+                
         write_to_html(html_file, "<center>Freetype Benchmark</center>\n")
         write_to_html(html_file, "</body>\n</html>\n")
 
@@ -103,7 +108,95 @@ def generate_info_table(html_file, baseline_info, 
benchmark_info):
             ),
         )
     write_to_html(html_file, "</table><br/>")
-    write_to_html(html_file, "* Cumulative time for iterations which is better 
in smaller values<br/>\n")
+    write_to_html(html_file, "<p>* Average time for all iterations. Smaller 
values are better.</p>")
+    write_to_html(html_file, "<p>** N count in (x | y) format is for showing 
baseline and benchmark N counts seperately when they differs.</p>")
+    
+
+def generate_total_results_table(html_file, baseline_dir, benchmark_dir):
+    """Prepare total results table for html"""
+    
+    # This dictionary will store aggregated results.
+    test_results = {test: {"baseline": 0, "benchmark": 0, "n_baseline": 0, 
"n_benchmark": 0} for test in [
+        "Load", "Load_Advances (Normal)", "Load_Advances (Fast)", 
"Load_Advances (Unscaled)", "Render",
+        "Get_Glyph", "Get_Char_Index", "Iterate CMap", "New_Face", "Embolden", 
"Stroke", "Get_BBox",
+        "Get_CBox", "New_Face & load glyph(s)"
+    ]}
+    
+    for filename in os.listdir(baseline_dir):
+        if filename.endswith(".txt") and not filename == "info.txt":
+            
+            baseline_results = read_file(os.path.join(baseline_dir, filename))
+            benchmark_results = read_file(os.path.join(benchmark_dir, 
filename))
+            
+            for baseline_line, benchmark_line in zip(baseline_results, 
benchmark_results):
+                if baseline_line.startswith("  "):
+                    baseline_match = 
re.match(r"\s+(.*?)\s+(\d+\.\d+)\s+microseconds\s+(\d+)\s", baseline_line)
+                    benchmark_match = 
re.match(r"\s+(.*?)\s+(\d+\.\d+)\s+microseconds\s+(\d+)\s", benchmark_line)
+                    
+                    if baseline_match and benchmark_match:
+                        test = baseline_match.group(1).strip()
+                        baseline_value = float(baseline_match.group(2))
+                        benchmark_value = float(benchmark_match.group(2))
+                        baseline_n = int(baseline_match.group(3))
+                        benchmark_n = int(benchmark_match.group(3))
+                        
+                        # Aggregate the results
+                        if test in test_results:
+                            test_results[test]["baseline"] += baseline_value
+                            test_results[test]["benchmark"] += benchmark_value
+                            test_results[test]["n_baseline"] += baseline_n
+                            test_results[test]["n_benchmark"] += benchmark_n
+    
+    # Writing to HTML
+    write_to_html(html_file, "<h2>Total Results</h2>\n")
+    write_to_html(html_file, '<table border="1">\n')
+    write_to_html(
+        html_file,
+        '<tr><th>Test</th><th>N</th><th>Baseline (&#181;s)</th>\
+        <th>Benchmark (&#181;s)</th><th>Difference (%)</th></tr>\n'
+    )
+
+    total_baseline = total_benchmark = total_diff = total_n_baseline = 
total_n_benchmark = 0
+    
+    for test, values in test_results.items():
+        baseline = values["baseline"]
+        benchmark = values["benchmark"]
+        n_baseline = values["n_baseline"]
+        n_benchmark = values["n_benchmark"]
+        
+        n_display = f"{n_baseline} | {n_benchmark}" if n_baseline != 
n_benchmark else str(n_baseline)
+        
+        diff = ((baseline - benchmark) / baseline) * 100
+
+        # Calculate for total row
+        total_baseline += baseline
+        total_benchmark += benchmark
+        total_n_baseline += n_baseline
+        total_n_benchmark += n_benchmark
+        
+        # Check which value is smaller for color highlighting
+        baseline_color = "highlight" if baseline <= benchmark else ""
+        benchmark_color = "highlight" if benchmark <= baseline else ""
+
+        write_to_html(
+            html_file,
+            f'<tr><td class="col1">{test}</td><td>{n_display}</td>\
+            <td class="{baseline_color}">{baseline:.0f}</td>\
+            <td 
class="{benchmark_color}">{benchmark:.0f}</td><td>{diff:.1f}</td></tr>\n'
+        )
+
+    total_diff = ((total_baseline - total_benchmark) / total_baseline) * 100
+    total_n_display = f"{total_n_baseline} | {total_n_benchmark}" if 
total_n_baseline != total_n_benchmark else str(total_n_baseline)
+    
+    write_to_html(
+        html_file,
+        f'<tr><td class="col1">TOTAL</td><td 
class="col1">{total_n_display}</td>\
+        <td class="col1">{total_baseline:.0f}</td><td 
class="col1">{total_benchmark:.0f}</td>\
+        <td class="col1">{total_diff:.1f}</td></tr>\n'
+    )
+    
+    write_to_html(html_file, "</table><br/>\n")
+
 
 
 def generate_results_table(html_file, baseline_results, benchmark_results, 
filename):
@@ -114,7 +207,7 @@ def generate_results_table(html_file, baseline_results, 
benchmark_results, filen
         if line.startswith("ftbench results for font")
     ][0]
 
-    write_to_html(html_file, "<h2>Results for {}</h2>\n".format(fontname))
+    write_to_html(html_file, "<h3>Results for {}</h2>\n".format(fontname))
     write_to_html(html_file, '<table border="1">\n')
     write_to_html(
         html_file,
@@ -185,7 +278,7 @@ def generate_results_table(html_file, baseline_results, 
benchmark_results, filen
                             percentage_diff,
                         ),
                     )
-
+                
     write_to_html(
         html_file,
         '<tr><td class="col1">TOTAL</td><td class="col1">{}</td>\