Support FMA operation (#1354)

Adds a fused multiply-add operation to the Float module.

The following changes are made:
- configure: check math.h for the C99 fma() operation.
- fma declarations in float.ml[i] (stdlib/).
- C fma() call or emulation in runtime/floats.c.
- dedicated tests in testsuite/tests/fma.

2 files changed, 11 insertions, 0 deletions

diff --git a/stdlib/float.ml b/stdlib/float.ml
index f51275faf3..e5a4d52718 100644
--- a/stdlib/float.ml
+++ b/stdlib/float.ml
@@ -21,6 +21,8 @@ external mul : float -> float -> float = "%mulfloat"
 external div : float -> float -> float = "%divfloat"
 external rem : float -> float -> float = "caml_fmod_float" "fmod"
   [@@unboxed] [@@noalloc]
+external fma : float -> float -> float -> float = "caml_fma_float" "caml_fma"
+  [@@unboxed] [@@noalloc]
 external abs : float -> float = "%absfloat"
 
 let zero = 0.
diff --git a/stdlib/float.mli b/stdlib/float.mli
index b8a9094335..38386e0256 100644
--- a/stdlib/float.mli
+++ b/stdlib/float.mli
@@ -57,6 +57,15 @@ external mul : float -> float -> float = "%mulfloat"
 external div : float -> float -> float = "%divfloat"
 (** Floating-point division. *)
 
+external fma : float -> float -> float -> float =
+  "caml_fma_float" "caml_fma" [@@unboxed] [@@noalloc]
+(** [fma x y z] returns [x * y + z], with a best effort for computing
+   this expression with a single rounding, using either hardware
+   instructions (providing full IEEE compliance) or a software
+   emulation.  Note: since software emulation of the fma is costly,
+   make sure that you are using hardware fma support if performance
+   matters.  @since 4.08.0 *)
+
 external rem : float -> float -> float = "caml_fmod_float" "fmod"
 [@@unboxed] [@@noalloc]
 (** [rem a b] returns the remainder of [a] with respect to [b].  The returned