Perf plot

flopcounts/x86/variant_inlining_7_SoA_AVX2.py

@@ -0,0 +1,41 @@
+# const int n_p_pad = (n_p + 3) & ~3;
+
+# flops per iteration (one pass of the outer time loop):
+# ADDs:   n_p * 12  +  n_p * n_p_pad * 15  +  n_p * board.n_obstacles * 12  +  n_p * 7
+# MULTs:  n_p * 4   +  n_p * n_p_pad * 24  +  n_p * board.n_obstacles * 19  +  n_p * 16
+# DIVs:              n_p * n_p_pad * 1  +  n_p * board.n_obstacles  +  n_p * 2
+# SQRTs:            n_p * n_p_pad * 3  +  n_p * board.n_obstacles  +  n_p * 3
+# EXPs:  n_p * n_p_pad * 1  +  n_p * board.n_obstacles
+
+
+def adds(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 3) & ~3
+  return int(n_iter *
+             (n_p * 12 + n_p * n_p_pad * 15 + n_p * n_obs * 12 + n_p * 7))
+
+
+def mults(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 3) & ~3
+  return int(n_iter *
+             (n_p * 4 + n_p * n_p_pad * 24 + n_p * n_obs * 19 + n_p * 16))
+
+
+def divs(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 3) & ~3
+  return int(n_iter * (n_p * n_p_pad * 1 + n_p * n_obs + n_p * 2))
+
+
+def sqrts(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 3) & ~3
+  return int(n_iter * (n_p * n_p_pad * 3 + n_p * n_obs + n_p * 3))
+
+
+def exps(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 3) & ~3
+  return int(n_iter * (n_p * n_p_pad * 1 + n_p * n_obs))
+
+
+def total(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  return adds(n_iter, n_p, n_obs) + mults(n_iter, n_p, n_obs) + divs(
+    n_iter, n_p, n_obs) + sqrts(n_iter, n_p,
+                                n_obs) + exps(n_iter, n_p, n_obs) * 30