allocators.hpp

Go to the documentation of this file.

/*
Copyright (c) 2014-2016, Scott Zuyderduyn
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice, this
   list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice,
   this list of conditions and the following disclaimer in the documentation
   and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

The views and conclusions contained in the software and documentation are those
of the authors and should not be interpreted as representing official policies,
either expressed or implied, of the FreeBSD Project.
*/

//----------------------------------------------------------------------------
// allocators.hpp
//
// STL-compatible memory allocators using CUDA memory allocation routines.
//
// Author: Scott D. Zuyderduyn, Ph.D. (scott.zuyderduyn@utoronto.ca)
//----------------------------------------------------------------------------

#pragma once
#ifndef ECUDA_ALLOCATORS_HPP
#define ECUDA_ALLOCATORS_HPP

#include <limits>
#include <stdexcept>

#include "global.hpp"
#include "type_traits.hpp"
#include "ptr/common.hpp"
#include "ptr/padded_ptr.hpp"

namespace ecuda {

template<typename T,unsigned Flags=cudaHostAllocDefault>
class host_allocator
{

public:
    typedef T                                                   value_type;      
    typedef typename ecuda::add_pointer<T>::type                pointer;         
    typedef typename ecuda::add_lvalue_reference<T>::type       reference;       
    typedef typename make_const<pointer>::type                  const_pointer;   
    typedef typename ecuda::add_lvalue_reference<const T>::type const_reference; 
    typedef std::size_t                                         size_type;       
    typedef std::ptrdiff_t                                      difference_type; 
    template<typename U> struct rebind { typedef host_allocator<U> other; }; 

public:
    host_allocator() throw() {}

    host_allocator( const host_allocator& alloc ) throw() {}

    template<typename U>
    host_allocator( const host_allocator<U>& alloc ) throw() {}

    ~host_allocator() throw() {}

    inline pointer address( reference x ) { return &x; }

    inline const_pointer address( const_reference x ) const { return &x; }

    pointer allocate( size_type n, std::allocator<void>::const_pointer hint = 0 )
    {
        pointer ptr = NULL;
        const cudaError_t result = cudaHostAlloc( reinterpret_cast<void**>(&ptr), n*sizeof(T), Flags );
        if( result != cudaSuccess ) throw std::bad_alloc();
        return ptr;
    }

    inline void deallocate( pointer ptr, size_type )
    {
        typedef typename ecuda::add_pointer<value_type>::type raw_pointer_type;
        default_host_delete<value_type>()( naked_cast<raw_pointer_type>(ptr) );
    }

    inline size_type max_size() const throw() { return std::numeric_limits<size_type>::max(); }

    inline void construct( pointer ptr, const_reference val ) { new ((void*)ptr) value_type (val);  }

    inline void destroy( pointer ptr ) { ptr->~value_type(); }

};

template<typename T>
class device_allocator
{

public:
    typedef T                                                   value_type;      
    typedef typename ecuda::add_pointer<T>::type                pointer;         
    typedef typename ecuda::add_lvalue_reference<T>::type       reference;       
    typedef typename make_const<pointer>::type                  const_pointer;   
    typedef typename ecuda::add_lvalue_reference<const T>::type const_reference; 
    typedef std::size_t                                         size_type;       
    typedef std::ptrdiff_t                                      difference_type; 
    template<typename U> struct rebind { typedef device_allocator<U> other; }; 

public:
    __HOST__ __DEVICE__ device_allocator() throw() {}

    __HOST__ __DEVICE__ device_allocator( const device_allocator& alloc ) throw() {}

    template<typename U>
    __HOST__ __DEVICE__ device_allocator( const device_allocator<U>& alloc ) throw() {}

    __HOST__ __DEVICE__ ~device_allocator() throw() {}

    __HOST__ __DEVICE__ inline pointer address( reference x ) { return &x; }

    __HOST__ __DEVICE__ inline const_pointer address( const_reference x ) const { return &x; }

    __HOST__ pointer allocate( size_type n, std::allocator<void>::const_pointer hint = 0 )
    {
        pointer ptr = NULL;
        const cudaError_t result = cudaMalloc( reinterpret_cast<void**>(&ptr), n*sizeof(T) );
        if( result != cudaSuccess ) throw std::bad_alloc();
        return ptr;
    }

    __HOST__ inline void deallocate( pointer ptr, size_type )
    {
        typedef typename ecuda::add_pointer<value_type>::type raw_pointer_type;
        default_device_delete<value_type>()( naked_cast<raw_pointer_type>(ptr) );
    }

    __HOST__ __DEVICE__ inline size_type max_size() const throw() { return std::numeric_limits<size_type>::max(); }

    __DEVICE__ inline void construct( pointer ptr, const_reference val ); // not supported on device

    __DEVICE__ inline void destroy( pointer ptr ); // not supported on device

};

template<typename T>
class device_pitch_allocator
{

public:
    typedef T                                                   value_type;      
    typedef padded_ptr<T,typename ecuda::add_pointer<T>::type>  pointer;         
    typedef typename ecuda::add_lvalue_reference<T>::type       reference;       
    typedef typename make_const<pointer>::type                  const_pointer;   
    typedef typename ecuda::add_lvalue_reference<const T>::type const_reference; 
    typedef std::size_t                                         size_type;       
    typedef std::ptrdiff_t                                      difference_type; 
    template<typename U> struct rebind { typedef device_allocator<U> other; }; 

private:
    template<typename U> struct char_cast;
    template<typename U> struct char_cast<U*>       { char* type;       };
    template<typename U> struct char_cast<const U*> { const char* type; };

public:
    __HOST__ __DEVICE__ device_pitch_allocator() throw() {}

    __HOST__ __DEVICE__ device_pitch_allocator( const device_pitch_allocator& alloc ) throw() {}

    template<typename U>
    __HOST__ __DEVICE__ device_pitch_allocator( const device_pitch_allocator<U>& alloc ) throw() {}

    __HOST__ __DEVICE__ ~device_pitch_allocator() throw() {}

    __HOST__ __DEVICE__ inline pointer address( reference x ) { return &x; }

    __HOST__ __DEVICE__ inline const_pointer address( const_reference x ) const { return &x; }

    __HOST__ pointer allocate( size_type w, size_type h, std::allocator<void>::const_pointer hint = 0 )
    {
        typename ecuda::add_pointer<value_type>::type ptr = NULL;
        size_type pitch;
        const cudaError_t result = cudaMallocPitch( reinterpret_cast<void**>(&ptr), &pitch, w*sizeof(value_type), h );
        if( result != cudaSuccess ) throw std::bad_alloc();
        return pointer( ptr, pitch );
    }

    __HOST__ inline void deallocate( pointer ptr, size_type )
    {
        typedef typename ecuda::add_pointer<value_type>::type raw_pointer_type;
        default_device_delete<value_type>()( naked_cast<raw_pointer_type>(ptr) );
    }

    __HOST__ __DEVICE__ inline size_type max_size() const throw() { return std::numeric_limits<size_type>::max(); }

    __DEVICE__ inline void construct( pointer ptr, const_reference val ); // not supported on device

    __DEVICE__ inline void destroy( pointer ptr ); // not supported on device

    __HOST__ __DEVICE__ inline const_pointer address( const_pointer ptr, size_type x, size_type y, size_type pitch ) const
    {
        return reinterpret_cast<const_pointer>( naked_cast<const char*>(ptr) + x*pitch + y*sizeof(value_type) );
    }

    __HOST__ __DEVICE__ inline pointer address( pointer ptr, size_type x, size_type y )
    {
        // TODO: this is not general if this is padded_ptr<T,[some other specialized class]>
        typedef typename ecuda::add_pointer<value_type>::type raw_pointer;
        raw_pointer p = naked_cast<raw_pointer>(ptr);
        typedef typename char_cast<raw_pointer>::type char_pointer;
        char_pointer p2 = reinterpret_cast<char_pointer>(p);
        p2 += ptr.get_pitch() * x;
        p = p2;
        p += y;
        return pointer( p, ptr.get_pitch() );
    }

};

} // namespace ecuda

#endif