Simpler iteration timing/prediction code

Include libraries, allow for multiple compile and link flags for the
environment, always include debug symbols, and misc changes.

make.py

@@ -17,7 +17,8 @@ COMPILE_FLAGS = [
     "-pedantic-errors",
     "-fopenmp",
 ]
-LINK_FLAGS = []
+LINK_FLAGS = ["-fopenmp"]
+LIBRARIES = ["-lm"]
 OBJ_DIR = os.path.join("obj", "release" if ("RELEASE" in os.environ) else "debug")
 SRC_DIR = "src"
 EXE_NAME = "main." + ("release" if ("RELEASE" in os.environ) else "debug") + ".exe"
@@ -28,11 +29,12 @@ if "RELEASE" in os.environ:
     COMPILE_FLAGS[COMPILE_FLAGS.index("-O0")] = "-O2"
     COMPILE_FLAGS[COMPILE_FLAGS.index("-g3")] = "-g"
     COMPILE_FLAGS.append("-DRELEASE")
-    COMPILE_FLAGS.append("-g")
     LINK_FLAGS.append("-DRELEASE")
-else:
-    COMPILE_FLAGS.append("-fsanitize=address,undefined")
-    LINK_FLAGS.append("-fsanitize=address,undefined")
+
+if "COMPILE_FLAGS" in os.environ:
+    COMPILE_FLAGS.extend(os.environ["COMPILE_FLAGS"].split(";"))
+if "LINK_FLAGS" in os.environ:
+    LINK_FLAGS.extend(os.environ["LINK_FLAGS"].split(";"))
 
 
 def as_object(source_path):
@@ -88,7 +90,7 @@ def link():
     obj_files = glob(
         os.path.join(OBJ_DIR, "**", "*.obj").replace("\\", "/"), recursive=True
     )
-    sp.run([COMPILER, *LINK_FLAGS, "-o", EXE_NAME, *obj_files])
+    sp.run([COMPILER, *LINK_FLAGS, "-o", EXE_NAME, *obj_files, *LIBRARIES])
 
 
 @recipe(info="Removes all compiled objects.")

src/main.c

@@ -1,4 +1,5 @@
 #include <alloca.h>
+#include <math.h>
 #include <stdbool.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -26,10 +27,13 @@ static void ReportProgress(uintmax_t current, uintmax_t total, clock_t now) {
     if (current >= _timing_last_count) {
       float inferred_rate = (float)(current - _timing_last_count) /
                             ((float)(now - _timing_last_time) / CLOCKS_PER_SEC);
-      _timing_rate = _timing_rate * (1. - _timing_exp_backoff) +
-                     inferred_rate * _timing_exp_backoff;
-      if (_timing_exp_backoff > 0.0001) {
-        _timing_exp_backoff *= 0.99;
+      if (inferred_rate < _timing_rate) {
+        _timing_rate = inferred_rate;
+        _timing_exp_backoff = 0.3;
+      } else {
+        _timing_rate = _timing_rate * (1. - _timing_exp_backoff) +
+                       inferred_rate * _timing_exp_backoff;
+        _timing_exp_backoff = fmax(0.99 * _timing_exp_backoff, 0.001);
       }
       _timing_last_count = current;
       _timing_last_time = now;
@@ -256,8 +260,14 @@ int main(int argc, char *argv[]) {
       PermutationReset(b_out_perms + i);
     }
 
-#pragma omp for schedule(guided)
+#pragma omp for schedule(dynamic)
     for (size_t lhs_i = 0; lhs_i < row_count - 1; lhs_i++) {
+      {
+        clock_t now = clock();
+        uintmax_t tested_now = tested;
+        ReportProgress(tested_now, total, now);
+      }
+
       if (seen[lhs_i]) {
         const uintmax_t remaining = (uintmax_t)row_count * perm_iterations;
 #pragma omp atomic update
@@ -351,12 +361,6 @@ int main(int argc, char *argv[]) {
       skip_permutations:;
 #pragma omp atomic update
         tested += perm_iterations;
-        {
-          clock_t now = clock();
-#pragma omp task default(none) shared(tested, total) firstprivate(now)
-          ReportProgress(tested, total, now);
-        }
-
       }  // For loop over rhs_i
     }    // For loop over lhs_i