diff options
| author | Pierre Roux | 2019-04-04 00:14:47 +0200 |
|---|---|---|
| committer | Pierre Roux | 2019-11-01 10:21:03 +0100 |
| commit | dca0135a263717b3a1a1d7c4f054f039dc08109e (patch) | |
| tree | 3f2ab5ea79e084a9c72e1376d5399ad4a62cb771 /kernel/float64.mli | |
| parent | 3e0db1b645a8653c62b8b5a4978e6d8fbbe9a9cc (diff) | |
Make primitive float work on x86_32
Flag -fexcess-precision=standard is not enough on x86_32
where -msse2 -mfpmath=sse is required (-msse is not enough)
to avoid double rounding issues in the VM.
Most floating-point operation are now implemented in C because OCaml
is suffering double rounding issues on x86_32 with 80 bits extended
precision registers used for floating-point values, causing double
rounding making floating-point arithmetic incorrect with respect to
its specification.
Add a runtime test for double roundings.
Diffstat (limited to 'kernel/float64.mli')
| -rw-r--r-- | kernel/float64.mli | 7 |
1 files changed, 0 insertions, 7 deletions
diff --git a/kernel/float64.mli b/kernel/float64.mli index acc3a556ab..78dc1a7bd7 100644 --- a/kernel/float64.mli +++ b/kernel/float64.mli @@ -46,19 +46,14 @@ val classify : t -> float_class [@@ocaml.inline always] val mul : t -> t -> t -[@@ocaml.inline always] val add : t -> t -> t -[@@ocaml.inline always] val sub : t -> t -> t -[@@ocaml.inline always] val div : t -> t -> t -[@@ocaml.inline always] val sqrt : t -> t -[@@ocaml.inline always] (** Link with integers *) val of_int63 : Uint63.t -> t @@ -77,10 +72,8 @@ val ldshiftexp : t -> Uint63.t -> t [@@ocaml.inline always] val next_up : t -> t -[@@ocaml.inline always] val next_down : t -> t -[@@ocaml.inline always] (** Return true if two floats are equal. * All NaN values are considered equal. *) |