array.hpp

Go to the documentation of this file.

#pragma once
#ifndef CUDA_API_WRAPPERS_ARRAY_HPP_
#define CUDA_API_WRAPPERS_ARRAY_HPP_

#include "context.hpp"
#include "error.hpp"

#ifndef CUDA_NO_HALF
#include <cuda_fp16.h>
#endif

namespace cuda {

class device_t;

template <typename T, dimensionality_t NumDimensions>
class array_t;

namespace array {

using handle_t = CUarray;

template <dimensionality_t NumDimensions>
using descriptor_t = typename ::std::conditional<NumDimensions == 2, CUDA_ARRAY_DESCRIPTOR, CUDA_ARRAY3D_DESCRIPTOR>::type;

template <typename T, dimensionality_t NumDimensions>
array_t<T, NumDimensions> wrap(
    device::id_t                 device_id,
    context::handle_t            context_handle,
    handle_t                     handle,
    dimensions_t<NumDimensions>  dimensions) noexcept;

namespace detail_ {

template <typename T> struct format_specifier {};

template <> struct format_specifier<uint8_t > { static constexpr const CUarray_format value = CU_AD_FORMAT_UNSIGNED_INT8;  };
template <> struct format_specifier<uint16_t> { static constexpr const CUarray_format value = CU_AD_FORMAT_UNSIGNED_INT16; };
template <> struct format_specifier<uint32_t> { static constexpr const CUarray_format value = CU_AD_FORMAT_UNSIGNED_INT32; };
template <> struct format_specifier<int8_t  > { static constexpr const CUarray_format value = CU_AD_FORMAT_SIGNED_INT8;    };
template <> struct format_specifier<int16_t > { static constexpr const CUarray_format value = CU_AD_FORMAT_SIGNED_INT16;   };
template <> struct format_specifier<int32_t > { static constexpr const CUarray_format value = CU_AD_FORMAT_SIGNED_INT32;   };
#ifndef CUDA_NO_HALF
template <> struct format_specifier<half    > { static constexpr const CUarray_format value = CU_AD_FORMAT_HALF;           };
#endif
template <> struct format_specifier<float   > { static constexpr const CUarray_format value = CU_AD_FORMAT_FLOAT;          };

template<typename T>
handle_t create_in_current_context(dimensions_t<3> dimensions)
{
    handle_t handle;
    CUDA_ARRAY3D_DESCRIPTOR descriptor;
    descriptor.Width = dimensions.width;
    descriptor.Height = dimensions.height;
    descriptor.Depth = dimensions.depth;
    descriptor.Format = format_specifier<T>::value;
    descriptor.NumChannels = 1;
        // We don't currently support an array of packed pairs or quadruplets; if you want this,
        // file an issue.
    descriptor.Flags = 0;

    auto status = cuArray3DCreate(&handle, &descriptor);
    throw_if_error_lazy(status, "failed allocating 3D CUDA array");
    return handle;
}

template<typename T>
handle_t create_in_current_context(dimensions_t<2> dimensions)
{
    CUDA_ARRAY_DESCRIPTOR descriptor;
    descriptor.Width = dimensions.width;
    descriptor.Height = dimensions.height;
    descriptor.Format = format_specifier<T>::value;
    descriptor.NumChannels = 1;
    handle_t handle;
    auto status = cuArrayCreate(&handle, &descriptor);
    throw_if_error_lazy(status, "failed allocating 2D CUDA array");
    return handle;
}

template <typename T, dimensionality_t NumDimensions>
handle_t create(context::handle_t context_handle, dimensions_t<NumDimensions> dimensions)
{
    CAW_SET_SCOPE_CONTEXT(context_handle);
    return create_in_current_context<T>(dimensions);
}

template <typename T, dimensionality_t NumDimensions>
handle_t create(const context_t& context, dimensions_t<NumDimensions> dimensions);

template <dimensionality_t NumDimensions>
descriptor_t<NumDimensions> get_descriptor_in_current_context(handle_t handle);

template <>
inline descriptor_t<2> get_descriptor_in_current_context<2>(handle_t handle)
{
    descriptor_t<2> result;
    auto status = cuArrayGetDescriptor(&result, handle);
    throw_if_error_lazy(status,
        ::std::string("Failed obtaining the descriptor of the CUDA 2D array at ")
        + cuda::detail_::ptr_as_hex(handle));
    return result;
}

template <>
inline descriptor_t<4> get_descriptor_in_current_context<3>(handle_t handle)
{
    descriptor_t<3> result;
    auto status = cuArray3DGetDescriptor(&result, handle);
    throw_if_error_lazy(status,
        ::std::string("Failed obtaining the descriptor of the CUDA 3D array at ")
        + cuda::detail_::ptr_as_hex(handle));
    return result;
}

template <dimensionality_t NumDimensions>
descriptor_t<NumDimensions> get_descriptor(context::handle_t context_handle, handle_t handle)
{
    CAW_SET_SCOPE_CONTEXT(context_handle);
    return get_descriptor_in_current_context<NumDimensions>(handle);
}

template <dimensionality_t NumDimensions>
dimensions_t<NumDimensions> dimensions_of(const descriptor_t<NumDimensions>& descriptor);

template <>
inline dimensions_t<3> dimensions_of(const descriptor_t<3>& descriptor)
{
    return { descriptor.Width, descriptor.Height, descriptor.Depth };
}

template <>
inline dimensions_t<2> dimensions_of(const descriptor_t<2>& descriptor)
{
    return { descriptor.Width, descriptor.Height };
}

template <dimensionality_t NumDimensions>
dimensions_t<NumDimensions> dimensions_of_in_current_context(handle_t handle_in_current_context)
{
    auto descriptor = get_descriptor_in_current_context<NumDimensions>(handle_in_current_context);
    return dimensions_of<NumDimensions>(descriptor);
}

template <dimensionality_t NumDimensions>
dimensions_t<NumDimensions> dimensions_of(context::handle_t context_handle, handle_t handle)
{
    CAW_SET_SCOPE_CONTEXT(context_handle);
    return dimensions_of_in_current_context<NumDimensions>(handle);
}

} // namespace detail_

} // namespace array

template <typename T, dimensionality_t NumDimensions>
class array_t {
    static_assert(NumDimensions == 2 or NumDimensions == 3, "CUDA only supports 2D and 3D arrays");

public:
    using handle_type = array::handle_t;
    using descriptor_type = array::descriptor_t<NumDimensions>;
    using dimensions_type = array::dimensions_t<NumDimensions>;

    array_t(device::id_t device_id, context::handle_t context_handle, handle_type handle, dimensions_type dimensions) :
        dimensions_(dimensions), device_id_(device_id), context_handle_(context_handle), handle_(handle)
    {
        assert(handle != nullptr);
    }

    array_t(const array_t& other) = delete;
    array_t(array_t&& other) noexcept : array_t(other.device_id_, other.context_handle_, other.handle_, other.dimensions_)
    {
        other.handle_ = nullptr;
    }

    ~array_t() noexcept(false)
    {
        CAW_SET_SCOPE_CONTEXT(context_handle_);
        if (handle_) {
            auto status = cuArrayDestroy(handle_);
            // Note: Throwing in a noexcept destructor; if the free'ing fails, the program
            // will likely terminate
            throw_if_error_lazy(status, "Failed destroying CUDA array " + cuda::detail_::ptr_as_hex(handle_));
        }
    }

    friend array_t array::wrap<T, NumDimensions>(device::id_t, context::handle_t, handle_type, dimensions_type) noexcept;

    handle_type get() const noexcept { return handle_; }
    device::id_t device_id() const noexcept { return device_id_; }
    context::handle_t context_handle() const noexcept { return context_handle_; }
    dimensions_type dimensions() const noexcept { return dimensions_; }
    device_t device() const noexcept;
    context_t context() const;

    ::std::size_t size() const noexcept { return dimensions().size(); }

    ::std::size_t size_bytes() const noexcept { return size() * sizeof(T); }

    descriptor_type descriptor() const  { return array::detail_::get_descriptor<NumDimensions>(context_handle_, handle_); }

protected:
    dimensions_type    dimensions_;
    device::id_t       device_id_;
    context::handle_t  context_handle_;
    handle_type        handle_;
};

namespace array {

template <typename T, dimensionality_t NumDimensions>
array_t<T, NumDimensions> wrap(
    device::id_t                 device_id,
    context::handle_t            context_handle,
    handle_t                     handle,
    dimensions_t<NumDimensions>  dimensions) noexcept
{
    return { device_id, context_handle, handle, dimensions };
}

template <typename T, dimensionality_t NumDimensions>
array_t<T,NumDimensions> create(
    const context_t&             context,
    dimensions_t<NumDimensions>  dimensions);

template <typename T, dimensionality_t NumDimensions>
array_t<T,NumDimensions> create(
    const device_t&              device,
    dimensions_t<NumDimensions>  dimensions);

} // namespace array

} // namespace cuda

#endif // CUDA_API_WRAPPERS_ARRAY_HPP_