-- --
-- GNAT COMPILER COMPONENTS --
-- --
--- G N A T . S S E . V e c t o r _ T y p e s --
+-- G N A T . S S E . V E C T O R _ T Y P E S --
-- --
-- S p e c --
-- --
------------------------------------------------------------------------------
-- This unit exposes the Ada __m128 like data types to represent the contents
--- of SSE registers, for use by the SSE intrinsics.
+-- of SSE registers, for use by bindings to the SSE intrinsic operations.
+
+-- See GNAT.SSE for the list of targets where this facility is supported
package GNAT.SSE.Vector_Types is
- -- The reference guide states a few usage guidelines for the C types :
+ -- The reference guide states a few usage guidelines for the C types:
- -- << Since these new data types are not basic ANSI C data types, you
- -- must observe the following usage restrictions:
+ -- Since these new data types are not basic ANSI C data types, you
+ -- must observe the following usage restrictions:
--
-- * Use new data types only on either side of an assignment, as a
-- return value, or as a parameter. You cannot use it with other
-- access the byte elements and structures.
--
-- * Use new data types only with the respective intrinsics described
- -- in this documentation. >>
+ -- in this documentation.
- type M128 is private; -- SSE >= 1
- type M128d is private; -- SSE >= 2
- type M128i is private; -- SSE >= 2
+ type m128 is private; -- SSE >= 1
+ type m128d is private; -- SSE >= 2
+ type m128i is private; -- SSE >= 2
private
+ -- Each of the m128 types maps to a specific vector_type with an extra
+ -- "may_alias" attribute as in GCC's definitions for C, for instance in
+ -- xmmintrin.h:
- -- GCC'wise, vector operations operate on objects of vector modes,
- -- conveyed through vector types obtained by setting an attribute on what
- -- looks like a component typedef. For example, in C (xmmintrin.h):
- --
- -- typedef float __v4sf __attribute__ ((__vector_size__ (16)));
-
- -- We can obtain the same low level GCC effect in Ada with
- -- Machine_Attribute pragmas, as in
- --
- -- type Vf is new Float;
- -- pragma Machine_Attribute (Vf, "vector_size", 16);
- --
- -- which makes Vf a 16bytes long V4SFmode type for GCC. The effect on the
- -- type layout is not conveyed to the front-end, however, so the latter
- -- still sees "Vf" as a 4bytes long single float. This leads to numerous
- -- potential pitfalls if this type is directly exposed user land, so we
- -- add wrapper records with rep clauses to compensate.
+ -- /* The Intel API is flexible enough that we must allow aliasing
+ -- with other vector types, and their scalar components. */
+ -- typedef float __m128
+ -- __attribute__ ((__vector_size__ (16), __may_alias__));
- -- The wrapper records all have a single component of the twisted low
- -- level type, so they inherit the mode while the rep clauses convey the
- -- size and alignment information to the front-end.
+ -- /* Internal data types for implementing the intrinsics. */
+ -- typedef float __v4sf __attribute__ ((__vector_size__ (16)));
------------
- -- M128 --
+ -- m128 --
------------
- -- << The __m128 data type can hold four 32-bit floating-point values. >>
-
- type V4sf is new Float32;
- pragma Machine_Attribute (V4sf, "vector_size", VECTOR_BYTES);
+ -- The __m128 data type can hold four 32-bit floating-point values
- type M128 is record
- Value : V4sf;
- end record;
- for M128'Size use VECTOR_BYTES * 8;
- for M128'Alignment use VECTOR_ALIGN;
+ type m128 is array (1 .. 4) of Float32;
+ for m128'Alignment use VECTOR_ALIGN;
+ pragma Machine_Attribute (m128, "vector_type");
+ pragma Machine_Attribute (m128, "may_alias");
-------------
- -- M128d --
+ -- m128d --
-------------
- -- << The __m128d data type can hold two 64-bit floating-point values. >>
+ -- The __m128d data type can hold two 64-bit floating-point values
- type V2df is new Float64;
- pragma Machine_Attribute (V2df, "vector_size", VECTOR_BYTES);
-
- type M128d is record
- Value : V2df;
- end record;
- for M128d'Size use VECTOR_BYTES * 8;
- for M128d'Alignment use VECTOR_ALIGN;
+ type m128d is array (1 .. 2) of Float64;
+ for m128d'Alignment use VECTOR_ALIGN;
+ pragma Machine_Attribute (m128d, "vector_type");
+ pragma Machine_Attribute (m128d, "may_alias");
-------------
- -- M128i --
+ -- m128i --
-------------
- -- << The __m128i data type can hold sixteen 8-bit, eight 16-bit, four
- -- 32-bit, or two 64-bit integer values. >>
-
- type V2di is new Integer64;
- pragma Machine_Attribute (V2di, "vector_size", VECTOR_BYTES);
+ -- The __m128i data type can hold sixteen 8-bit, eight 16-bit, four 32-bit,
+ -- or two 64-bit integer values.
- type M128i is record
- Value : V2di;
- end record;
- for M128i'Size use VECTOR_BYTES * 8;
- for M128i'Alignment use VECTOR_ALIGN;
+ type m128i is array (1 .. 2) of Integer64;
+ for m128i'Alignment use VECTOR_ALIGN;
+ pragma Machine_Attribute (m128i, "vector_type");
+ pragma Machine_Attribute (m128i, "may_alias");
end GNAT.SSE.Vector_Types;
OSDN Git Service
