Skip to content

Latest commit

 

History

History
154 lines (106 loc) · 5.77 KB

README.md

File metadata and controls

154 lines (106 loc) · 5.77 KB
Raw

typeutils

日本語版

aggregate.hpp

using namespace typu;
using T = type_t< agg_t< char, int, char, double >, 0 /* natural align */ >;

char* storage = reinterpret_cast< char* >(malloc(get<T, 0>::size));
*get<T, 0, 2>::addr(storage) = 'd';
*get<T, 0, 3>::addr(storage) = 3.141592;

/* convert double -> float */
template <> struct mapto< double > : public base_mapper< float >{};
using S = typename morph< T, mapto >::mapped;

A computable metatype: builidng and manipurating types like modifing struct/union.

GPU and other dedicated devices require data and its layout. This helps generating types which dedicate device requires raw-data and its descriptors from a CPU-friendly type in compile-time.

  • non-struct: compositing types provide meta-programmable type
  • alignment-safe: type_t(s) are compositable with strong/week alignment requirements
  • morph: generates homeomorphic type using map
  • typesafe: may avoid to prevent compiler's analyzing alias due to the application be less typecast

clang5.0 or later. or MSVC2017.

type_t<T, Align>

Contains a single type, and provide requirement of alignment. Align=0 means a natural alignment.
type_t will often contains agg_t or sel_t. Nesting type_t provides more strong/week alignment than container. This may put a padding before/after of containee.

Aggregation Type agg_t<Ts...>/sel_t<Ts...>

Contains any types.
agg_t/sel_t corresponds to struct/union. But these container (not containee) shall not be have dedicate constructors, destructors and virtual functions. For containee, no dedicate restructions but it should not be void.

Notation named_t<Name, T>

Gives a symbol, Name, as single containee T. The symbol will be used in lookup. It will have no effect for calculation of offset that a type does not have its index since it's not aggregation type, but will be a participant for map/fold's.

get<T, Index...>

Accesses member-type by indices.
In the following example, type will be float type indexed by <0, 4>. the first 0 means the directly containee agg_t.

using T = type_t< agg_t< int, int, int, int, float, double >, 0 >;
using S = typename get<T, 0, 4 >::type; /* float */

Nesting types requires higher dimension indices.
Also get has offset, addr from effective-address and leaf-index, index for the indexed type.

size_t ofs = get< T0, 0, 1, 0, 0 >::offset;
*get<T0, 0, 1, 0, 0>::addr(base) = 0xdeadbeef;

get<T, Index...>::index will be replaced to the indexed type by Index... where has its storage on 'T'. (TBD: When a sel_t's member is in a type, T, will be assumed or selected the largest type of members)

using T1 = type_t< agg_t< int, int, agg_t< bool, bool, bool >, float[16], agg_t<int, int, float, double>, char >, 0 >;
size_t idx = get<T2, 0, 4, 2 >::index; /* float's index = 8 */

morph<T, Fun...>

Gives the new type as map. In the following example, morph will convert float[16] and float[4] to correspond algebra type, matrix and vector. mapped will be mapped type.
Function's argument will be matched node types like as agg_t, sel_t or type_t.

using namespace typu;
template <> struct typu::mapto< float[16] > : public base_mapper< matrix44_t > {};
template <> struct typu::mapto< float[4] > : public base_mapper< vector4_t > {};

using T = type_t< agg_t< float[16], float[4], float[16] >, 16 >;
using S = typename morph< T, mapto >::mapped; /* type_t< agg_t< matrix44_t, vector4_t, matrix44_t >, 16 > */

Also morph is able to remove types by mapping them void. In the following example, it morphs int to void via a function fun(int)->void. A function is expressed in template's full specialization. Then void and types has no its storage will be sanitized.
If you need a full type you can access it with mapped_raw.

using namespace typu;
template <> struct typu::mapto< int > : public base_mapper< void > {};

using T = type_t< agg_t< int, named_t< "int"_hash, type_t< agg_t< int, named_t< 0, int > >, 16 > >, char >, 0 >;
using S = typename morph< T, mapto >::mapped; /* type_t< char, 0 > */

fold<T, Acc, Acc0, Fun... >

(TBD)

Applies Fun... to T's all containees and gives a value typed as Acc. A function's argument will be matched node types like agg_t, sel_t or type_t, morph as well.
Next example shows counting int of T.

using namespace typu;
template < typename Acc, Acc Acc0, typename In >
struct fold_fun { static const Acc value = Acc0; };
template <typename Acc, Acc Acc0 >
struct fold_fun< Acc, Acc0, int > { static const Acc value = Acc0 + 1; };

using T = type_t< agg_t< int, agg_t< int >, agg_t< float[16] >, char >, 0 >;
auto n = fold< T, int, 0, fold_fun >::value; // 2

Functions will be applied to sel_t's all containees.

lookup.hpp

using namespace typu;
using T = type_t< agg_t< char, int, char, named_t< 0xdeadbeef, double >, 0 /* natural align */ >;

static_assert(std::is_same< double, typename lu<T, 0xdeadbeef>::type >::value, "found type must be double");
printf("offset: %zu\n", lu<T, 0xdeadbeef>::offset);

named_t<Name, T> names T as unique symbol, Name. Looking the type up via lu<T, Symbol> is done at compile-time as well get or other functions do.

at.hpp

This header has utility to manipurate variadic template parameter.

using namespace typu;

/* get a type Index=2 */
static_assert(std::is_same< at_type< 2, char, short, int, long >::type, int >::value, "sorry"); 

/* get 2 types from a head */
static_assert(std::is_same< to_types< 2, char, short, int, long >::type, std::tuple< char, short > >::value, "sorry");

to_types<N, T...> has two variations. One is to_types<>::type, which uses std::tuple. Another is to_types<>::types which uses simple type_list<...>.