On Fri, Sep 3, 2021 at 8:42 PM Jakub Jelinek via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> On Mon, Aug 02, 2021 at 02:31:12PM +0800, liuhongt via Gcc-patches wrote:
> > * doc/extend.texi (Half-Precision Floating Point): Documemt
> > _Float16 for x86.
>
> > --- a/gcc/doc/extend.texi
> > +++ b/gcc/doc/extend.texi
> > @@ -1102,6 +1102,7 @@ typedef _Complex float __attribute__((mode(IC)))
> > _Complex_ibm128;
> > @section Half-Precision Floating Point
> > @cindex half-precision floating point
> > @cindex @code{__fp16} data type
> > +@cindex @code{__Float16} data type
> >
> > On ARM and AArch64 targets, GCC supports half-precision (16-bit) floating
> > point via the @code{__fp16} type defined in the ARM C Language Extensions.
> > @@ -1150,6 +1151,18 @@ calls.
> > It is recommended that portable code use the @code{_Float16} type defined
> > by ISO/IEC TS 18661-3:2015. @xref{Floating Types}.
> >
> > +On x86 targets with @code{target("sse2")} and above, GCC supports
> > half-precision
> > +(16-bit) floating point via the @code{_Float16} type which is defined by
> > +18661-3:2015. For C++, x86 provide a builtin type named @code{_Float16}
> > +which contains same data format as C.
> > +
> > +Without @option{-mavx512fp16}, @code{_Float16} type is storage only, all
> > +operations will be emulated by software emulation and the @code{float}
> > +instructions. The default behavior for @code{FLT_EVAL_METHOD} is to keep
> > +the intermediate result of the operation as 32-bit precision. This may lead
> > +to inconsistent behavior between software emulation and AVX512-FP16
> > +instructions.
> > +
> > @node Decimal Float
> > @section Decimal Floating Types
> > @cindex decimal floating types
>
> Shouldn't there be more changes for this in doc/extend.texi?
>
> I'd say that x86 with -msse2 should be mentioned in
> The @code{_Float16} type is supported on AArch64
> systems by default, and on ARM systems when the IEEE format for 16-bit
> floating-point types is selected with @option{-mfp16-format=ieee}.
>
> and in
> @node Half-Precision
> I'd say that one sentence about the x86 support should go already in the
> first paragraph, perhaps after:
> On ARM and AArch64 targets, GCC supports half-precision (16-bit) floating
> point via the @code{__fp16} type defined in the ARM C Language Extensions.
> On ARM systems, you must enable this type explicitly with the
> @option{-mfp16-format} command-line option in order to use it.
> because users just won't scroll down to immediately find out that in the
> 10th/11th paragraph it talks about x86.
> Just mention there that on all 3 arches it is available using the _Float16
> type
> in C, on x86 in C++ too and then on ARM/AArch64 using __fp16.
>
> Jakub
>
How about this?
Adjust the wording for x86 _Float16 type.
gcc/ChangeLog:
* doc/extend.texi: (@node Floating Types): Adjust the wording.
(@node Half-Precision): Ditto.
1 file changed, 14 insertions(+), 13 deletions(-)
gcc/doc/extend.texi | 27 ++++++++++++++-------------
modified gcc/doc/extend.texi
@@ -1076,9 +1076,11 @@ systems where @code{__float128} is supported.
The @code{_Float32}
type is supported on all systems supporting IEEE binary32; the
@code{_Float64} and @code{_Float32x} types are supported on all systems
supporting IEEE binary64. The @code{_Float16} type is supported on AArch64
-systems by default, and on ARM systems when the IEEE format for 16-bit
-floating-point types is selected with @option{-mfp16-format=ieee}.
-GCC does not currently support @code{_Float128x} on any systems.
+systems by default, and also on x86 systems with @code{target("sse2")}
+for both C and C++.
+On ARM systems when the IEEE format for 16-bit floating-point types is
+selected with @option{-mfp16-format=ieee}. GCC does not currently support
+@code{_Float128x} on any systems.
On the i386, x86_64, IA-64, and HP-UX targets, you can declare complex
types using the corresponding internal complex type, @code{XCmode} for
@@ -1108,6 +1110,10 @@ On ARM and AArch64 targets, GCC supports
half-precision (16-bit) floating
point via the @code{__fp16} type defined in the ARM C Language Extensions.
On ARM systems, you must enable this type explicitly with the
@option{-mfp16-format} command-line option in order to use it.
+On x86 targets with @code{target("sse2")} and above, GCC supports
+half-precision (16-bit) floating point via the @code{_Float16} type.
+For C++, x86 provides a builtin type named @code{_Float16} which contains
+same data format as C.
ARM targets support two incompatible representations for half-precision
floating-point values. You must choose one of the representations and
@@ -1151,16 +1157,11 @@ calls.
It is recommended that portable code use the @code{_Float16} type defined
by ISO/IEC TS 18661-3:2015. @xref{Floating Types}.
-On x86 targets with @code{target("sse2")} and above, GCC supports
half-precision
-(16-bit) floating point via the @code{_Float16} type which is defined by
-18661-3:2015. For C++, x86 provide a builtin type named @code{_Float16}
-which contains same data format as C.
-
-Without @option{-mavx512fp16}, @code{_Float16} type is storage only, all
-operations will be emulated by software emulation and the @code{float}
-instructions. The default behavior for @code{FLT_EVAL_METHOD} is to keep
-the intermediate result of the operation as 32-bit precision. This may lead
-to inconsistent behavior between software emulation and AVX512-FP16
+On x86 targets, without @option{-mavx512fp16}, @code{_Float16} type is
+storage only, all operations will be emulated by software emulation and the
+@code{float} instructions. The default behavior for @code{FLT_EVAL_METHOD} is
+to keep the intermediate result of the operation as 32-bit precision. This may
+lead to inconsistent behavior between software emulation and AVX512-FP16
instructions.
@node Decimal Float
--
BR,
Hongtao
