1 files changed, 68 insertions, 0 deletions
diff --git a/math/aarch64/sve/tanhf.c b/math/aarch64/sve/tanhf.c
new file mode 100644
index 000000000000..9007e7badb0d
--- /dev/null
+++ b/math/aarch64/sve/tanhf.c
@@ -0,0 +1,68 @@
+/*
+ * Single-precision SVE tanh(x) function.
+ *
+ * Copyright (c) 2023-2024, Arm Limited.
+ * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
+ */
+
+#include "sv_math.h"
+#include "test_sig.h"
+#include "test_defs.h"
+#include "sv_expm1f_inline.h"
+
+/* Largest value of x for which tanhf(x) rounds to 1 (or -1 for negative).  */
+#define BoringBound 0x1.205966p+3f
+
+static const struct data
+{
+  struct sv_expm1f_data expm1f_consts;
+  uint32_t onef, special_bound;
+  float boring_bound;
+} data = {
+  .expm1f_consts = SV_EXPM1F_DATA,
+  .onef = 0x3f800000,
+  .special_bound = 0x7f800000,
+  .boring_bound = BoringBound,
+};
+
+static svfloat32_t NOINLINE
+special_case (svfloat32_t x, svbool_t pg, svbool_t is_boring,
+	      svfloat32_t boring, svfloat32_t q, svbool_t special)
+{
+  svfloat32_t y
+      = svsel_f32 (is_boring, boring, svdiv_x (pg, q, svadd_x (pg, q, 2.0)));
+  return sv_call_f32 (tanhf, x, y, special);
+}
+
+/* Approximation for single-precision SVE tanh(x), using a simplified
+   version of expm1f. The maximum error is 2.57 ULP:
+   _ZGVsMxv_tanhf (0x1.fc1832p-5) got 0x1.fb71a4p-5
+				 want 0x1.fb71aap-5.  */
+svfloat32_t SV_NAME_F1 (tanh) (svfloat32_t x, const svbool_t pg)
+{
+  const struct data *d = ptr_barrier (&data);
+
+  svfloat32_t ax = svabs_x (pg, x);
+  svuint32_t iax = svreinterpret_u32 (ax);
+  svuint32_t sign = sveor_x (pg, svreinterpret_u32 (x), iax);
+  svfloat32_t boring = svreinterpret_f32 (svorr_x (pg, sign, d->onef));
+  svbool_t special = svcmpgt (pg, iax, d->special_bound);
+  svbool_t is_boring = svacgt (pg, x, d->boring_bound);
+
+  /* tanh(x) = (e^2x - 1) / (e^2x + 1).  */
+  svfloat32_t q = expm1f_inline (svmul_x (svptrue_b32 (), x, 2.0), pg,
+				 &d->expm1f_consts);
+
+  if (unlikely (svptest_any (pg, special)))
+    return special_case (x, pg, is_boring, boring, q, special);
+  svfloat32_t y = svdiv_x (pg, q, svadd_x (pg, q, 2.0));
+  return svsel_f32 (is_boring, boring, y);
+}
+
+TEST_SIG (SV, F, 1, tanh, -10.0, 10.0)
+TEST_ULP (SV_NAME_F1 (tanh), 2.07)
+TEST_DISABLE_FENV (SV_NAME_F1 (tanh))
+TEST_SYM_INTERVAL (SV_NAME_F1 (tanh), 0, 0x1p-23, 1000)
+TEST_SYM_INTERVAL (SV_NAME_F1 (tanh), 0x1p-23, BoringBound, 100000)
+TEST_SYM_INTERVAL (SV_NAME_F1 (tanh), BoringBound, inf, 100)
+CLOSE_SVE_ATTR