Changeset 891Tweet

Timestamp:

Sep 2, 2011, 1:36:42 PM (12 years ago)

Author:

sam

Message:

optim: split the Taylor series calculation into two separate values.

This is at the cost of one additional multiply, but performance increases
by more than 11%, because the PS3 pipeline is a lot happier now.

File:

• trunk/src/trig.cpp (modified) (2 diffs)

trunk/src/trig.cpp

@@ -169 +169 @@
     double absx = lol_fabs(x * INV_PI);
     double sign = lol_fsel(x, PI, NEG_PI);
+
+    /* To compute sin(x) we build a Taylor series for |x|/pi wrapped to
+     * the range [-1, 1]. We also switch the result sign if the number
+     * of cycles is odd. */
 #if defined __CELLOS_LV2__
     double num_cycles = lol_round(absx);
@@ -185 +189 @@
 #endif
     double norm_x = absx - num_cycles;
-    double y = norm_x * norm_x;
-    double taylor = (((((((SC[7] * y + SC[6]) * y + SC[5])
-                                 * y + SC[4]) * y + SC[3])
-                                 * y + SC[2]) * y + SC[1])
-                                 * y + SC[0]) * y + ONE;
+
+    /* Computing x^4 is one multiplication too many we do, but it helps
+     * interleave the Taylor series operations a lot better. */
+    double x2 = norm_x * norm_x;
+    double x4 = x2 * x2;
+    double sub1 = ((SC[7] * x4 + SC[5]) * x4 + SC[3]) * x4 + SC[1];
+    double sub2 = ((SC[6] * x4 + SC[4]) * x4 + SC[2]) * x4 + SC[0];
+    double taylor = (sub1 * x2 + sub2) * x2 + ONE;
+
     double result = norm_x * taylor;
     return result * sign;