1
0
forked from Alepha/Alepha
Files
Alepha/Memory/Blob.h

526 lines
17 KiB
C++

static_assert( __cplusplus > 2020'99 );
#pragma once
#include <Alepha/Alepha.h>
#include <cassert>
#include <cstdlib>
#include <memory>
#include <Alepha/swappable.h>
#include <Alepha/fastRandom.h>
#include <Alepha/Exception.h>
#include <Alepha/error.h>
#include <Alepha/IOStreams/String.h>
#include <Alepha/Utility/evaluation_helpers.h>
#include "Buffer.h"
// TODO: Put this into the `Alepha::Memory::` namespace.
// TODO: Consider whether a "republish" into `Alepha::` namespace is a good idea.
namespace Alepha::Hydrogen ::detail:: Blob_m
{
inline namespace exports
{
class Blob;
class DataCarveTooLargeError;
class DataCarveOutOfRangeError;
}
namespace C
{
const auto doubleBlobPointerOption= "ALEPHA_USE_DOUBLE_BLOB_POINTERS";
const int storageSplitRandomBitDepth= 12; // TODO: Environment tunable? Global tunable?
const bool debug= false;
const bool debugLifecycle= false or C::debug;
const bool debugCtors= false or C::debugLifecycle or C::debug;
const bool debugAssignment= false or C::debugLifecycle or C::debug;
const bool debugSwap= false or C::debugLifecycle or C::debug;
const bool debugSplitSharing= false or C::debug;
const bool debugInteriorCarve= false or C::debug;
}
using std::begin, std::end;
using IOStreams::stringify;
class exports::DataCarveTooLargeError
: public virtual Buffer_m::OutOfRangeError
{
public:
explicit
DataCarveTooLargeError( const void *const location, const std::size_t request, const std::size_t available )
: std::out_of_range( "Tried to carve " + stringify( request ) + " bytes from `Blob` object at location "
+ stringify( location ) + " which only has " + stringify( available ) + " bytes allocated." ),
OutOfRangeError( location, request, available )
{}
};
class exports::DataCarveOutOfRangeError
: public virtual Buffer_m::OutOfRangeError
{
public:
explicit
DataCarveOutOfRangeError( const void *const location, const std::size_t request, const std::size_t available )
: std::out_of_range( "Tried to carve " + stringify( request ) + " bytes from `Blob` object at location "
+ stringify( location ) + " which only has " + stringify( available ) + " bytes allocated." ),
OutOfRangeError( location, request, available )
{}
};
class exports::Blob
: public BufferModel< Blob >, public swappable
{
private:
using IndirectStorage= std::shared_ptr< std::shared_ptr< Blob > >;
IndirectStorage storage; // If this is empty, then this `Blob` object doesn't share ownership. This references the shared pool.
Buffer< Mutable > buffer;
std::size_t viewLimit= 0; // TODO: Consider allowing for unrooted sub-buffer views?
// Potentially rollover the internal storage, during a blob sharing operation.
void
maybeRollover()
{
if( not ::getenv( C::doubleBlobPointerOption ) ) return;
if( Alepha::fastRandomBits( C::storageSplitRandomBitDepth ) ) return;
// The inner pointer gets incremented.
std::shared_ptr< Blob > inner= *storage;
// Capture the new inner into the outer.
storage= std::make_shared< std::shared_ptr< Blob > >( std::move( inner ) );
// Post rollover, there should only be one reference to start.
assert( storage.use_count() == 1 );
}
// Takeover a portion of a shared block
explicit
Blob( IndirectStorage storage, Buffer< Mutable > buffer ) noexcept
: storage( std::move( storage ) ), buffer( buffer ), viewLimit( buffer.size() )
{}
public:
~Blob() { reset(); }
using StorageReservation= IndirectStorage;
const StorageReservation &reservation() const { return storage; }
auto
swap_lens() noexcept
{
if( C::debugSwap ) error() << "Swap lens called." << std::endl;
return swap_magma( storage, buffer, viewLimit );
}
/*!
* Allocate a new arena of specified size and release the old arena.
*
* This function has the strong guarantee -- if the allocation fails, the
* old arena is still allocated.
*
* @param size The size of the new arena to allocate.
*
* @note: No data are copied.
*/
void
reset() noexcept
{
if( not storage ) delete [] buffer.byte_data();
else storage.reset();
buffer= {};
viewLimit= 0;
}
/*!
* Allocate a new arena of specified size using the specified allocator and release the old arena.
*
* This function has the strong guarantee -- if the allocation fails, the
* old arena is still allocated.
*
* @param size The size of the new arena to allocate.
* @param newAlloc The allocator to use for the replacement arena (and for future allocations as well).
*
* @note: No data are copied.
*/
void
reset( const std::size_t size )
{
Blob tmp{ size };
swap( tmp, *this );
}
// Copy deep copies the data.
Blob( const Blob &copy )
: buffer( new std::byte[ copy.size() ], copy.size() ),
viewLimit( copy.size() )
{
if( C::debugCtors ) error() << "Blob copy invoked." << std::endl;
copyData( *this, copy );
}
Blob( Blob &&orig ) noexcept { swap( *this, orig ); }
template< typename ByteIterator >
explicit
Blob( ByteIterator first, ByteIterator last )
: Blob( std::distance( first, last ) )
{
std::copy( first, last, byte_data() );
}
// Move assignment
Blob &
operator= ( Blob &&orig ) noexcept
{
Blob temp= std::move( orig );
swap( *this, temp );
return *this;
}
Blob &
operator= ( const Blob &source )
{
if( buffer.size() < source.size() ) reset( source.size() );
else viewLimit= source.size();
copyData( *this, source );
return *this;
}
void
setSize( const std::size_t size )
{
if( size > buffer.size() ) throw std::runtime_error( "Cannot size `Blob` to a larger size than its allocated buffer." );
viewLimit= size;
}
explicit
Blob( const std::size_t amount )
: buffer( new std::byte[ amount ]{}, amount ), // The data must be 0'ed upon allocation.
viewLimit( amount )
{}
explicit
Blob( const Buffer< Const > b )
: Blob( b.size() )
{
copyData( buffer, b );
}
Blob() noexcept= default;
// Buffer Model adaptors:
constexpr operator Buffer< Mutable > () noexcept { return { buffer, viewLimit }; }
constexpr operator Buffer< Const > () const noexcept { return { buffer, viewLimit }; }
// Carving functions:
/*!
* Carve the head off of a `Blob` object.
*
* "Carving" a `Blob` object splits it into two different `Blob` objects, each sharing and keeping alive
* the original physical memory backing the source `Blob` object. The return value of a "carve"
* operation is a new `Blob` object of the requested size. When the original `Blob` is "carved", it
* will shrink itself down by the requested number of bytes.
*
* Carving is very useful to maintain a large number of `Blob` objects referring to small chunks of data
* inside a large single physical backing. This helps maintain zero-copy semantics.
*
* @param amount The amount of data to carve off.
* @param alignment The size alignment that the new base should be at (the extra padding is
* considered part of the resulting `Blob`).
* @return A new `Blob` object referring to the same physical data, scoped to `amount` bytes (with
* possible extra space, due to alignment).
*/
Blob
carveHead( const std::size_t amount )
{
if( amount > size() ) throw DataCarveTooLargeError( data(), amount, size() );
if( not storage )
{
// If there's no two-layer scheme, we have to start the sharing...
storage= std::make_shared< std::shared_ptr< Blob > >( std::make_shared< Blob >( std::move( *this ) ) );
// Now that it's shared, we repoint ourselves at the invisible `Blob` above.
buffer= (*storage)->buffer;
viewLimit= (*storage)->viewLimit;
}
assert( storage );
// Now we assume that there's a two-layer scheme, so we operate based upon that.
maybeRollover();
Blob rv{ storage, Buffer< Mutable >{ buffer, amount } };
buffer= buffer + amount;
viewLimit-= amount;
if( size() == 0 ) *this= Blob{};
return rv;
}
/*!
* Carve the tail off of a `Blob` object.
*
* @see `Blob::carveTail`
*
* @param amount The amount of data to carve off.
* @return A new `Blob` object referring to the same physical data, scoped to `amount` bytes.
*/
Blob
carveTail( const std::size_t amount )
{
if( amount > this->size() ) throw DataCarveTooLargeError( data(), amount, size() );
Blob temp= carveHead( size() - amount );
swap( *this, temp );
return temp;
}
// Assorted helpers:
template< typename T > void operator []( T ) const= delete;
template< typename T > void operator []( T )= delete;
constexpr std::size_t capacity() const noexcept { return buffer.size(); }
bool
isContiguousWith( const Blob &other ) const & noexcept
{
return
(
storage != nullptr
and
*storage == *other.storage
and
byte_data() + size() == other.byte_data()
);
}
/*!
* This function returns a proof that two `Blob` objects are contiguous.
*
* The proof object can be checked to prove that two `Blob`s are contiguous,
* and the `compose` operation on the proof object can be used to
* actually compose them -- that will cause the `other` `Blob` to be moved
* from, thus leaving it empty.
*/
auto
isContiguousWith( Blob &&other ) & noexcept
{
class ContiguousProof
{
private:
bool result;
Blob &self;
Blob &other;
friend Blob;
explicit constexpr
ContiguousProof( const bool result, Blob &self, Blob &other )
: result( result ), self( self ), other( other ) {}
public:
constexpr operator bool () const noexcept { return result; }
void
compose() const noexcept
{
assert( result );
self= Blob{ self.storage, Buffer< Mutable >{ self.data(), self.size() + other.size() } };
other.reset();
}
};
return ContiguousProof{ std::as_const( *this ).isContiguousWith( other ), *this, other };
}
constexpr friend std::size_t mailboxWeight( const Blob &b ) noexcept { return b.size(); }
/*!
* Determine whether some data can be appended to this `Blob` object.
*
* When this function returns `true`, a call to `concatenate` will return an empty `Buffer`.
*
* Because `Blob` objects can have unused capacity, sometimes it's possible to copy more data into that
* area without having to reallocate. This function returns true when that is possible and false
* otherwise.
*
* @param buffer The data buffer to check if it will fit.
* @return `true` if the data will fit and `false` otherwise.
*/
bool
couldConcatenate( const Buffer< Const > buffer ) const noexcept
{
return buffer.size() <= ( capacity() - size() );
}
/*!
* Determine whether some `Blob` can be appended to this `Blob` object.
*
* When this function returns `true`, a call to `concatenate` will return an empty `Blob` object.
*
* Because `Blob` objects can have unused capacity, sometimes it's possible to copy more data into that
* area without having to reallocate. Additionally, `Blob` objects created by carving can reference
* contiguous parts of the same buffer, so concatenation can be accomplished by shifting ownership
* instead of copying. This function returns true when a non-allocating append is possible and false
* otherwise.
*
* @param data The data buffer to check if it will fit.
* @return `true` if the data will fit and `false` otherwise.
*/
bool
couldConcatenate( const Blob &data ) const noexcept
{
return isContiguousWith( data ) or couldConcatenate( Buffer< Const >{ data } );
}
/*!
* Append, without reallocation, as much data as possible from the argument.
*
* Because `Blob` objects can have unused capacity, this unused space can be used to store more data
* without reallocation. This function copies as much data as will fit, but it will not allocate more
* storage. The `Buffer` which is returned (as `[[nodiscard]]`) refers to the uncopied range of
* data from the input. The returned `Buffer` has the same access level (`Constness`) as the
* parameter.
*
* @note When `couldConcatenate` is true, the returned `Buffer` is always empty.
*
* @param data The data buffer to copy from.
* @return A `Buffer` over the remaining uncopied data.
*/
template< Constness constness >
[[nodiscard]] Buffer< constness >
concatenate( const Buffer< constness > data ) noexcept
{
const auto amount= std::min( capacity() - size(), data.size() );
const Buffer< Const > fitted{ data, amount };
copyData( buffer + size(), fitted );
setSize( size() + amount );
return data + amount;
}
/*!
* Append, without reallocation, as much data as possible from the argument.
*
* Because `Blob` objects can have unused capacity, this unused space can be used to store more data
* without reallocation. Further, `Blob` objects created by carving can reference contiguous parts of
* the same buffer, so concatenation can be accomplished by shifting ownership instead of copying. This
* function attempts to compose two `Blob` objects if contiguous, otherwise it copies as much data as
* will fit, but it will not allocate more storage. The `Blob` object which is returned (as
* `[[nodiscard]]`) refers to the uncopied range of data from the input.
*
* @note When `couldConcatenate` is true, the returned `Blob` object is always empty.
* @note The `Blob` to be appended must be passed by `std::move`.
*
* @param blob The `Blob` object to append.
* @return A `Blob` object owning the uncopied portion.
*/
[[nodiscard]] Blob
concatenate( Blob &&blob ) noexcept
{
if( const auto proof= isContiguousWith( std::move( blob ) ) )
{
proof.compose();
return Blob{};
}
else
{
const auto amount= concatenate( Buffer< Const >{ blob } ).size();
const auto rv= blob.carveTail( amount );
blob.reset();
return rv;
}
}
/*!
* Append some data, reallocating if necessary.
*
* Copies all data from the specified `Buffer`, `data`, to the end of this `Blob` object,
* reallocating if necessary. The specified `requested` is a suggested minimum allocation size.
* The amount allocated will be at least that much, but may be more, if more is needed. This function
* does not attempt to amortize reallocation and copy across multiple calls. When working with `Blob`
* objects, it is the programmer's responsibility to minimize reallocation and copy overhead.
*
* @param data The data to append.
* @param requested The suggested size to allocate -- the amount allocated will be at least this
* much. If omitted, then allocations will be sized to fit.
*
* @note Specifying a `requested` in excess of the combined size may cause reallocation when it
* would otherwise not occur.
*/
void
combine( const Buffer< Const > data, const std::size_t requested= 0 )
{
const std::size_t needed= std::max( requested, size() + data.size() );
if( couldConcatenate( data ) and needed >= requested )
{
std::ignore= concatenate( data );
return;
}
Blob tmp{ needed };
copyData( tmp, *this );
copyData( tmp + size(), data );
tmp.setSize( size() + data.size() );
using std::swap;
swap( *this, tmp );
}
/*!
* Append some data, reallocating if necessary.
*
* Copies or composes all data from the specified `Blob` object, `blob`, to the end of this `Blob`
* object, reallocating if necessary. The specified `requested` is a suggested minimum allocation
* size. This `Blob` object's allocated will be at least that much, but may be more, if more is needed.
* This function does not attempt to amortize reallocation and copy across multiple calls. When working
* with `Blob` objects, it is the programmer's responsibility to minimize reallocation and copy
* overhead.
*
* @note The `Blob` to be appended must be passed by `std::move`.
*
* @param blob The `Blob` object to append.
* @param requested The suggested size to allocate -- the amount allocated will be at least this
* much. If omitted, then allocations will be sized to fit.
*
* @note Specifying a `requested` in excess of the combined size may cause reallocation when it
* would otherwise not occur.
*/
void
combine( Blob &&blob, const std::size_t requested= 0 )
{
const std::size_t needed= std::max( requested, size() + blob.size() );
if( couldConcatenate( blob ) and needed >= requested )
{
std::ignore= concatenate( std::move( blob ) );
}
else
{
combine( Buffer< Const >{ blob }, requested );
}
blob.reset();
}
};
//static_assert( Capability< Blob, swappable > );
//static_assert( detail::swaps::SwapLensable< Blob > );
}
namespace Alepha::Hydrogen::inline exports::inline Blob_m
{
using namespace detail::Blob_m::exports;
}