aboutsummaryrefslogtreecommitdiff
path: root/pl/math/atan_common.h
diff options
context:
space:
mode:
authorElliott Hughes <enh@google.com>2023-01-25 21:31:42 +0000
committerAutomerger Merge Worker <android-build-automerger-merge-worker@system.gserviceaccount.com>2023-01-25 21:31:42 +0000
commit8a481eb48cf7ace04026cbe776d65492f89a6d7d (patch)
tree945eb1832687ad5eb97c433013617dbe107718e5 /pl/math/atan_common.h
parent0bc581b0e4f509c96bffa7af329087ae2b4bfb97 (diff)
parent8783f524beaad825ac1bddeb93cef35b5f793513 (diff)
downloadarm-optimized-routines-8a481eb48cf7ace04026cbe776d65492f89a6d7d.tar.gz
Upgrade ARM-software/optimized-routines to v23.01 am: 62662f115a am: 8783f524be
Original change: https://android-review.googlesource.com/c/platform/external/arm-optimized-routines/+/2402215 Change-Id: Ief5438cf89d86c43e5f52d5e48b882bacb945170 Signed-off-by: Automerger Merge Worker <android-build-automerger-merge-worker@system.gserviceaccount.com>
Diffstat (limited to 'pl/math/atan_common.h')
-rw-r--r--pl/math/atan_common.h49
1 files changed, 49 insertions, 0 deletions
diff --git a/pl/math/atan_common.h b/pl/math/atan_common.h
new file mode 100644
index 0000000..da0da64
--- /dev/null
+++ b/pl/math/atan_common.h
@@ -0,0 +1,49 @@
+/*
+ * Double-precision polynomial evaluation function for scalar and vector atan(x)
+ * and atan2(y,x).
+ *
+ * Copyright (c) 2021-2023, Arm Limited.
+ * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
+ */
+
+#include "math_config.h"
+#include "estrin.h"
+
+#if V_SUPPORTED
+
+#include "v_math.h"
+
+#define DBL_T v_f64_t
+#define P(i) v_f64 (__atan_poly_data.poly[i])
+
+#else
+
+#define DBL_T double
+#define P(i) __atan_poly_data.poly[i]
+
+#endif
+
+/* Polynomial used in fast atan(x) and atan2(y,x) implementations
+ The order 19 polynomial P approximates (atan(sqrt(x))-sqrt(x))/x^(3/2). */
+static inline DBL_T
+eval_poly (DBL_T z, DBL_T az, DBL_T shift)
+{
+ /* Use split Estrin scheme for P(z^2) with deg(P)=19. Use split instead of
+ full scheme to avoid underflow in x^16. */
+ DBL_T z2 = z * z;
+ DBL_T x2 = z2 * z2;
+ DBL_T x4 = x2 * x2;
+ DBL_T x8 = x4 * x4;
+ DBL_T y
+ = FMA (ESTRIN_11_ (z2, x2, x4, x8, P, 8), x8, ESTRIN_7 (z2, x2, x4, P));
+
+ /* Finalize. y = shift + z + z^3 * P(z^2). */
+ y = FMA (y, z2 * az, az);
+ y = y + shift;
+
+ return y;
+}
+
+#undef DBL_T
+#undef FMA
+#undef P