Perf plot

flopcounts/x86/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt.py

@@ -0,0 +1,61 @@
+# const int n_p_pad = (n_p + 15) & ~15;
+
+# ADDs: n_iter * (
+#   n_p * 36 +
+#   n_p * n_p_pad * 29 +
+#   n_p * board.n_obstacles * 10 +
+#   n_p * 11
+# )
+
+# MULTs: n_iter * (
+#   n_p * 4 +
+#   n_p * n_p_pad * 40 +
+#   n_p * board.n_obstacles * 13 +
+#   n_p * 30
+# )
+
+# DIVs: n_iter * (
+#   n_p * n_p_pad +
+#   n_p * board.n_obstacles
+# )
+
+# SQRTs: n_iter * (
+#   n_p * n_p_pad * 3 +
+#   n_p * board.n_obstacles +
+#   n_p
+# )
+
+# EXPs: n_iter * n_p * board.n_obstacles
+
+
+def adds(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 15) & ~15
+  return int(n_iter *
+             (n_p * 36 + n_p * n_p_pad * 29 + n_p * n_obs * 10 + n_p * 11))
+
+
+def mults(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 15) & ~15
+  return int(n_iter *
+             (n_p * 4 + n_p * n_p_pad * 40 + n_p * n_obs * 13 + n_p * 30))
+
+
+def divs(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 15) & ~15
+  return int(n_iter * (n_p * n_p_pad + n_p * n_obs))
+
+
+def sqrts(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 15) & ~15
+  return int(n_iter * (n_p * n_p_pad * 3 + n_p * n_obs + n_p))
+
+
+def exps(n_iter: int, n_p: int, n_obs: int = 2) -> int:
+  n_p_pad = (n_p + 15) & ~15
+  return int(n_iter * 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