pl/math: Add scalar & vector/Neon cbrt

New routines use the same algorithm, with simplified argument
reduction and recombination in the vector variant. Both are accurate
to 2 ULP.

1 files changed, 20 insertions, 0 deletions

diff --git a/pl/math/tools/cbrt.sollya b/pl/math/tools/cbrt.sollya
new file mode 100644
index 0000000..7f179eb
--- /dev/null
+++ b/pl/math/tools/cbrt.sollya
@@ -0,0 +1,20 @@
+// polynomial for approximating cbrt(x) in double precision
+//
+// Copyright (c) 2022, Arm Limited.
+// SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
+
+deg = 3;
+
+a = 0.5;
+b = 1;
+
+
+f = x^(1/3);
+
+poly = fpminimax(f, deg, [|double ...|], [a;b]);
+
+display = hexadecimal;
+print("rel error:", accurateinfnorm(1-poly(x)/f(x), [a;b], 30));
+print("in [",a,b,"]");
+print("coeffs:");
+for i from 0 to deg do round(coeff(poly,i), D, RN);