blob: 2997f6e3c94a80530fb7e252874732584988c015 [file] [log] [blame]
/*
* Copyright (C) 2016 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef C2BUFFER_H_
#define C2BUFFER_H_
#include <C2.h>
#include <C2BufferBase.h>
#include <C2Param.h> // for C2Info
#include <memory>
#include <vector>
#ifdef __ANDROID__
#include <android-C2Buffer.h>
#else
typedef void* C2Handle;
#endif
/// \defgroup buffer Buffers
/// @{
/// \defgroup buffer_sync Synchronization
/// @{
/**
* Synchronization is accomplished using event and fence objects.
*
* These are cross-process extensions of promise/future infrastructure.
* Events are analogous to std::promise<void>, whereas fences are to std::shared_future<void>.
*
* Fences and events are shareable/copyable.
*
* Fences are used in two scenarios, and all copied instances refer to the same event.
* \todo do events need to be copyable or should they be unique?
*
* acquire sync fence object: signaled when it is safe for the component or client to access
* (the contents of) an object.
*
* release sync fence object: \todo
*
* Fences can be backed by hardware. Hardware fences are guaranteed to signal NO MATTER WHAT within
* a short (platform specific) amount of time; this guarantee is usually less than 15 msecs.
*/
/**
* Fence object used by components and the framework.
*
* Implements the waiting for an event, analogous to a 'future'.
*
* To be implemented by vendors if using HW fences.
*/
class C2Fence {
public:
/**
* Waits for a fence to be signaled with a timeout.
*
* \todo a mechanism to cancel a wait - for now the only way to do this is to abandon the
* event, but fences are shared so canceling a wait will cancel all waits.
*
* \param timeoutNs the maximum time to wait in nsecs
*
* \retval C2_OK the fence has been signaled
* \retval C2_TIMED_OUT the fence has not been signaled within the timeout
* \retval C2_BAD_STATE the fence has been abandoned without being signaled (it will never
* be signaled)
* \retval C2_REFUSED no permission to wait for the fence (unexpected - system)
* \retval C2_CORRUPTED some unknown error prevented waiting for the fence (unexpected)
*/
c2_status_t wait(c2_nsecs_t timeoutNs);
/**
* Used to check if this fence is valid (if there is a chance for it to be signaled.)
* A fence becomes invalid if the controling event is destroyed without it signaling the fence.
*
* \return whether this fence is valid
*/
bool valid() const;
/**
* Used to check if this fence has been signaled (is ready).
*
* \return whether this fence has been signaled
*/
bool ready() const;
/**
* Returns a file descriptor that can be used to wait for this fence in a select system call.
* \note The returned file descriptor, if valid, must be closed by the caller.
*
* This can be used in e.g. poll() system calls. This file becomes readable (POLLIN) when the
* fence is signaled, and bad (POLLERR) if the fence is abandoned.
*
* \return a file descriptor representing this fence (with ownership), or -1 if the fence
* has already been signaled (\todo or abandoned).
*
* \todo this must be compatible with fences used by gralloc
*/
int fd() const;
/**
* Returns whether this fence is a hardware-backed fence.
* \return whether this is a hardware fence
*/
bool isHW() const;
/**
* Null-fence. A fence that has fired.
*/
constexpr C2Fence() : mImpl(nullptr) { }
private:
class Impl;
std::shared_ptr<Impl> mImpl;
C2Fence(std::shared_ptr<Impl> impl);
friend struct _C2FenceFactory;
};
/**
* Event object used by components and the framework.
*
* Implements the signaling of an event, analogous to a 'promise'.
*
* Hardware backed events do not go through this object, and must be exposed directly as fences
* by vendors.
*/
class C2Event {
public:
/**
* Returns a fence for this event.
*/
C2Fence fence() const;
/**
* Signals (all) associated fence(s).
* This has no effect no effect if the event was already signaled or abandoned.
*
* \retval C2_OK the fence(s) were successfully signaled
* \retval C2_BAD_STATE the fence(s) have already been abandoned or merged (caller error)
* \retval C2_DUPLICATE the fence(s) have already been signaled (caller error)
* \retval C2_REFUSED no permission to signal the fence (unexpected - system)
* \retval C2_CORRUPTED some unknown error prevented signaling the fence(s) (unexpected)
*/
c2_status_t fire();
/**
* Trigger this event from the merging of the supplied fences. This means that it will be
* abandoned if any of these fences have been abandoned, and it will be fired if all of these
* fences have been signaled.
*
* \retval C2_OK the merging was successfully done
* \retval C2_NO_MEMORY not enough memory to perform the merging
* \retval C2_DUPLICATE the fence have already been merged (caller error)
* \retval C2_BAD_STATE the fence have already been signaled or abandoned (caller error)
* \retval C2_REFUSED no permission to merge the fence (unexpected - system)
* \retval C2_CORRUPTED some unknown error prevented merging the fence(s) (unexpected)
*/
c2_status_t merge(std::vector<C2Fence> fences);
/**
* Abandons the event and any associated fence(s).
* \note Call this to explicitly abandon an event before it is destructed to avoid a warning.
*
* This has no effect no effect if the event was already signaled or abandoned.
*
* \retval C2_OK the fence(s) were successfully signaled
* \retval C2_BAD_STATE the fence(s) have already been signaled or merged (caller error)
* \retval C2_DUPLICATE the fence(s) have already been abandoned (caller error)
* \retval C2_REFUSED no permission to abandon the fence (unexpected - system)
* \retval C2_CORRUPTED some unknown error prevented signaling the fence(s) (unexpected)
*/
c2_status_t abandon();
private:
class Impl;
std::shared_ptr<Impl> mImpl;
};
/// \addtogroup buf_internal Internal
/// @{
/**
* Interface for objects that encapsulate an updatable status value.
*/
struct _C2InnateStatus {
inline c2_status_t status() const { return mStatus; }
protected:
_C2InnateStatus(c2_status_t status) : mStatus(status) { }
c2_status_t mStatus; // this status is updatable by the object
};
/// @}
/**
* This is a utility template for objects protected by an acquire fence, so that errors during
* acquiring the object are propagated to the object itself.
*/
template<typename T>
class C2Acquirable : public C2Fence {
public:
/**
* Acquires the object protected by an acquire fence. Any errors during the mapping will be
* passed to the object.
*
* \return acquired object potentially invalidated if waiting for the fence failed.
*/
T get() {
// TODO:
// wait();
return mT;
}
protected:
C2Acquirable(c2_status_t error, C2Fence fence, T t) : C2Fence(fence), mInitialError(error), mT(t) { }
private:
c2_status_t mInitialError;
T mT; // TODO: move instead of copy
};
/// @}
/// \defgroup linear Linear Data Blocks
/// @{
/**************************************************************************************************
LINEAR ASPECTS, BLOCKS AND VIEWS
**************************************************************************************************/
/**
* Basic segment math support.
*/
struct C2Segment {
uint32_t offset;
uint32_t size;
inline constexpr C2Segment(uint32_t offset_, uint32_t size_)
: offset(offset_),
size(size_) {
}
inline constexpr bool isEmpty() const {
return size == 0;
}
inline constexpr bool isValid() const {
return offset <= ~size;
}
inline constexpr operator bool() const {
return isValid() && !isEmpty();
}
inline constexpr bool operator!() const {
return !bool(*this);
}
C2_ALLOW_OVERFLOW
inline constexpr bool contains(const C2Segment &other) const {
if (!isValid() || !other.isValid()) {
return false;
} else {
return offset <= other.offset
&& offset + size >= other.offset + other.size;
}
}
inline constexpr bool operator==(const C2Segment &other) const {
if (!isValid()) {
return !other.isValid();
} else {
return offset == other.offset && size == other.size;
}
}
inline constexpr bool operator!=(const C2Segment &other) const {
return !operator==(other);
}
inline constexpr bool operator>=(const C2Segment &other) const {
return contains(other);
}
inline constexpr bool operator>(const C2Segment &other) const {
return contains(other) && !operator==(other);
}
inline constexpr bool operator<=(const C2Segment &other) const {
return other.contains(*this);
}
inline constexpr bool operator<(const C2Segment &other) const {
return other.contains(*this) && !operator==(other);
}
C2_ALLOW_OVERFLOW
inline constexpr uint32_t end() const {
return offset + size;
}
C2_ALLOW_OVERFLOW
inline constexpr C2Segment intersect(const C2Segment &other) const {
return C2Segment(c2_max(offset, other.offset),
c2_min(end(), other.end()) - c2_max(offset, other.offset));
}
/** clamps end to offset if it overflows */
inline constexpr C2Segment normalize() const {
return C2Segment(offset, c2_max(offset, end()) - offset);
}
/** clamps end to max if it overflows */
inline constexpr C2Segment saturate() const {
return C2Segment(offset, c2_min(size, ~offset));
}
};
/**
* Common aspect for all objects that have a linear capacity.
*/
class _C2LinearCapacityAspect {
/// \name Linear capacity interface
/// @{
public:
inline constexpr uint32_t capacity() const { return mCapacity; }
inline constexpr operator C2Segment() const {
return C2Segment(0, mCapacity);
}
protected:
#if UINTPTR_MAX == 0xffffffff
static_assert(sizeof(size_t) == sizeof(uint32_t), "size_t is too big");
#else
static_assert(sizeof(size_t) > sizeof(uint32_t), "size_t is too small");
// explicitly disable construction from size_t
inline explicit _C2LinearCapacityAspect(size_t capacity) = delete;
#endif
inline explicit constexpr _C2LinearCapacityAspect(uint32_t capacity)
: mCapacity(capacity) { }
inline explicit constexpr _C2LinearCapacityAspect(const _C2LinearCapacityAspect *parent)
: mCapacity(parent == nullptr ? 0 : parent->capacity()) { }
private:
uint32_t mCapacity;
/// @}
};
/**
* Aspect for objects that have a linear range inside a linear capacity.
*
* This class is copiable.
*/
class _C2LinearRangeAspect : public _C2LinearCapacityAspect {
/// \name Linear range interface
/// @{
public:
inline constexpr uint32_t offset() const { return mOffset; }
inline constexpr uint32_t endOffset() const { return mOffset + mSize; }
inline constexpr uint32_t size() const { return mSize; }
inline constexpr operator C2Segment() const {
return C2Segment(mOffset, mSize);
}
private:
// subrange of capacity [0, capacity] & [size, size + offset]
inline constexpr _C2LinearRangeAspect(uint32_t capacity_, size_t offset, size_t size)
: _C2LinearCapacityAspect(capacity_),
mOffset(c2_min(offset, capacity())),
mSize(c2_min(size, capacity() - mOffset)) {
}
protected:
// copy constructor (no error check)
inline constexpr _C2LinearRangeAspect(const _C2LinearRangeAspect &other)
: _C2LinearCapacityAspect(other.capacity()),
mOffset(other.offset()),
mSize(other.size()) {
}
// parent capacity range [0, capacity]
inline constexpr explicit _C2LinearRangeAspect(const _C2LinearCapacityAspect *parent)
: _C2LinearCapacityAspect(parent),
mOffset(0),
mSize(capacity()) {
}
// subrange of parent capacity [0, capacity] & [size, size + offset]
inline constexpr _C2LinearRangeAspect(const _C2LinearCapacityAspect *parent, size_t offset, size_t size)
: _C2LinearCapacityAspect(parent),
mOffset(c2_min(offset, capacity())),
mSize(c2_min(size, capacity() - mOffset)) {
}
// subsection of the parent's and [offset, offset + size] ranges
inline constexpr _C2LinearRangeAspect(const _C2LinearRangeAspect *parent, size_t offset, size_t size)
: _C2LinearCapacityAspect(parent),
mOffset(c2_min(c2_max(offset, parent == nullptr ? 0 : parent->offset()), capacity())),
mSize(std::min(c2_min(size, parent == nullptr ? 0 : parent->size()), capacity() - mOffset)) {
}
public:
inline constexpr _C2LinearRangeAspect childRange(size_t offset, size_t size) const {
return _C2LinearRangeAspect(
mSize,
c2_min(c2_max(offset, mOffset), capacity()) - mOffset,
c2_min(c2_min(size, mSize), capacity() - c2_min(c2_max(offset, mOffset), capacity())));
}
friend class _C2EditableLinearRangeAspect;
// invariants 0 <= mOffset <= mOffset + mSize <= capacity()
uint32_t mOffset;
uint32_t mSize;
/// @}
};
/**
* Utility class for safe range calculations using size_t-s.
*/
class C2LinearRange : public _C2LinearRangeAspect {
public:
inline constexpr C2LinearRange(const _C2LinearCapacityAspect &parent, size_t offset, size_t size)
: _C2LinearRangeAspect(&parent, offset, size) { }
inline constexpr C2LinearRange(const _C2LinearRangeAspect &parent, size_t offset, size_t size)
: _C2LinearRangeAspect(&parent, offset, size) { }
inline constexpr C2LinearRange intersect(size_t offset, size_t size) const {
return C2LinearRange(*this, offset, size);
}
};
/**
* Utility class for simple and safe capacity and range construction.
*/
class C2LinearCapacity : public _C2LinearCapacityAspect {
public:
inline constexpr explicit C2LinearCapacity(size_t capacity)
: _C2LinearCapacityAspect(c2_min(capacity, std::numeric_limits<uint32_t>::max())) { }
inline constexpr C2LinearRange range(size_t offset, size_t size) const {
return C2LinearRange(*this, offset, size);
}
};
/**
* Aspect for objects that have an editable linear range.
*
* This class is copiable.
*/
class _C2EditableLinearRangeAspect : public _C2LinearRangeAspect {
using _C2LinearRangeAspect::_C2LinearRangeAspect;
public:
/// \name Editable linear range interface
/// @{
/**
* Sets the offset to |offset|, while trying to keep the end of the buffer unchanged (e.g.
* size will grow if offset is decreased, and may shrink if offset is increased.) Returns
* true if successful, which is equivalent to if 0 <= |offset| <= capacity().
*
* Note: setting offset and size will yield different result depending on the order of the
* operations. Always set offset first to ensure proper size.
*/
inline bool setOffset(uint32_t offset) {
if (offset > capacity()) {
return false;
}
if (offset > mOffset + mSize) {
mSize = 0;
} else {
mSize = mOffset + mSize - offset;
}
mOffset = offset;
return true;
}
/**
* Sets the size to |size|. Returns true if successful, which is equivalent to
* if 0 <= |size| <= capacity() - offset().
*
* Note: setting offset and size will yield different result depending on the order of the
* operations. Always set offset first to ensure proper size.
*/
inline bool setSize(uint32_t size) {
if (size > capacity() - mOffset) {
return false;
} else {
mSize = size;
return true;
}
}
/**
* Sets the offset to |offset| with best effort. Same as setOffset() except that offset will
* be clamped to the buffer capacity.
*
* Note: setting offset and size (even using best effort) will yield different result depending
* on the order of the operations. Always set offset first to ensure proper size.
*/
inline void setOffset_be(uint32_t offset) {
(void)setOffset(c2_min(offset, capacity()));
}
/**
* Sets the size to |size| with best effort. Same as setSize() except that the selected region
* will be clamped to the buffer capacity (e.g. size is clamped to [0, capacity() - offset()]).
*
* Note: setting offset and size (even using best effort) will yield different result depending
* on the order of the operations. Always set offset first to ensure proper size.
*/
inline void setSize_be(uint32_t size) {
mSize = c2_min(size, capacity() - mOffset);
}
/// @}
};
/**************************************************************************************************
ALLOCATIONS
**************************************************************************************************/
/// \ingroup allocator Allocation and memory placement
/// @{
class C2LinearAllocation;
class C2GraphicAllocation;
/**
* Allocators are used by the framework to allocate memory (allocations) for buffers. They can
* support either 1D or 2D allocations.
*
* \note In theory they could support both, but in practice, we will use only one or the other.
*
* Never constructed on stack.
*
* Allocators are provided by vendors.
*/
class C2Allocator {
public:
/**
* Allocator ID type.
*/
typedef uint32_t id_t;
enum : id_t {
BAD_ID = 0xBADD, // invalid allocator ID
};
/**
* Allocation types. This is a bitmask and is used in C2Allocator::Info
* to list the supported allocation types of an allocator.
*/
enum type_t : uint32_t {
LINEAR = 1 << 0, //
GRAPHIC = 1 << 1,
};
/**
* Information about an allocator.
*
* Allocators don't have a query API so all queriable information is stored here.
*/
struct Traits {
C2String name; ///< allocator name
id_t id; ///< allocator ID
type_t supportedTypes; ///< supported allocation types
C2MemoryUsage minimumUsage; ///< usage that is minimally required for allocations
C2MemoryUsage maximumUsage; ///< usage that is maximally allowed for allocations
};
/**
* Returns the unique name of this allocator.
*
* This method MUST be "non-blocking" and return within 1ms.
*
* \return the name of this allocator.
* \retval an empty string if there was not enough memory to allocate the actual name.
*/
virtual C2String getName() const = 0;
/**
* Returns a unique ID for this allocator. This ID is used to get this allocator from the
* allocator store, and to identify this allocator across all processes.
*
* This method MUST be "non-blocking" and return within 1ms.
*
* \return a unique ID for this allocator.
*/
virtual id_t getId() const = 0;
/**
* Returns the allocator traits.
*
* This method MUST be "non-blocking" and return within 1ms.
*
* Allocators don't have a full-fledged query API, only this method.
*
* \return allocator information
*/
virtual std::shared_ptr<const Traits> getTraits() const = 0;
/**
* Allocates a 1D allocation of given |capacity| and |usage|. If successful, the allocation is
* stored in |allocation|. Otherwise, |allocation| is set to 'nullptr'.
*
* \param capacity the size of requested allocation (the allocation could be slightly
* larger, e.g. to account for any system-required alignment)
* \param usage the memory usage info for the requested allocation. \note that the
* returned allocation may be later used/mapped with different usage.
* The allocator should layout the buffer to be optimized for this usage,
* but must support any usage. One exception: protected buffers can
* only be used in a protected scenario.
* \param allocation pointer to where the allocation shall be stored on success. nullptr
* will be stored here on failure
*
* \retval C2_OK the allocation was successful
* \retval C2_NO_MEMORY not enough memory to complete the allocation
* \retval C2_TIMED_OUT the allocation timed out
* \retval C2_REFUSED no permission to complete the allocation
* \retval C2_BAD_VALUE capacity or usage are not supported (invalid) (caller error)
* \retval C2_OMITTED this allocator does not support 1D allocations
* \retval C2_CORRUPTED some unknown, unrecoverable error occured during allocation (unexpected)
*/
virtual c2_status_t newLinearAllocation(
uint32_t capacity __unused, C2MemoryUsage usage __unused,
std::shared_ptr<C2LinearAllocation> *allocation /* nonnull */) {
*allocation = nullptr;
return C2_OMITTED;
}
/**
* (Re)creates a 1D allocation from a native |handle|. If successful, the allocation is stored
* in |allocation|. Otherwise, |allocation| is set to 'nullptr'.
*
* \param handle the handle for the existing allocation. On success, the allocation will
* take ownership of |handle|.
* \param allocation pointer to where the allocation shall be stored on success. nullptr
* will be stored here on failure
*
* \retval C2_OK the allocation was recreated successfully
* \retval C2_NO_MEMORY not enough memory to recreate the allocation
* \retval C2_TIMED_OUT the recreation timed out (unexpected)
* \retval C2_REFUSED no permission to recreate the allocation
* \retval C2_BAD_VALUE invalid handle (caller error)
* \retval C2_OMITTED this allocator does not support 1D allocations
* \retval C2_CORRUPTED some unknown, unrecoverable error occured during allocation (unexpected)
*/
virtual c2_status_t priorLinearAllocation(
const C2Handle *handle __unused,
std::shared_ptr<C2LinearAllocation> *allocation /* nonnull */) {
*allocation = nullptr;
return C2_OMITTED;
}
/**
* Allocates a 2D allocation of given |width|, |height|, |format| and |usage|. If successful,
* the allocation is stored in |allocation|. Otherwise, |allocation| is set to 'nullptr'.
*
* \param width the width of requested allocation (the allocation could be slightly
* larger, e.g. to account for any system-required alignment)
* \param height the height of requested allocation (the allocation could be slightly
* larger, e.g. to account for any system-required alignment)
* \param format the pixel format of requested allocation. This could be a vendor
* specific format.
* \param usage the memory usage info for the requested allocation. \note that the
* returned allocation may be later used/mapped with different usage.
* The allocator should layout the buffer to be optimized for this usage,
* but must support any usage. One exception: protected buffers can
* only be used in a protected scenario.
* \param allocation pointer to where the allocation shall be stored on success. nullptr
* will be stored here on failure
*
* \retval C2_OK the allocation was successful
* \retval C2_NO_MEMORY not enough memory to complete the allocation
* \retval C2_TIMED_OUT the allocation timed out
* \retval C2_REFUSED no permission to complete the allocation
* \retval C2_BAD_VALUE width, height, format or usage are not supported (invalid) (caller error)
* \retval C2_OMITTED this allocator does not support 2D allocations
* \retval C2_CORRUPTED some unknown, unrecoverable error occured during allocation (unexpected)
*/
virtual c2_status_t newGraphicAllocation(
uint32_t width __unused, uint32_t height __unused, uint32_t format __unused,
C2MemoryUsage usage __unused,
std::shared_ptr<C2GraphicAllocation> *allocation /* nonnull */) {
*allocation = nullptr;
return C2_OMITTED;
}
/**
* (Re)creates a 2D allocation from a native handle. If successful, the allocation is stored
* in |allocation|. Otherwise, |allocation| is set to 'nullptr'.
*
* \param handle the handle for the existing allocation. On success, the allocation will
* take ownership of |handle|.
* \param allocation pointer to where the allocation shall be stored on success. nullptr
* will be stored here on failure
*
* \retval C2_OK the allocation was recreated successfully
* \retval C2_NO_MEMORY not enough memory to recreate the allocation
* \retval C2_TIMED_OUT the recreation timed out (unexpected)
* \retval C2_REFUSED no permission to recreate the allocation
* \retval C2_BAD_VALUE invalid handle (caller error)
* \retval C2_OMITTED this allocator does not support 2D allocations
* \retval C2_CORRUPTED some unknown, unrecoverable error occured during recreation (unexpected)
*/
virtual c2_status_t priorGraphicAllocation(
const C2Handle *handle __unused,
std::shared_ptr<C2GraphicAllocation> *allocation /* nonnull */) {
*allocation = nullptr;
return C2_OMITTED;
}
virtual ~C2Allocator() = default;
protected:
C2Allocator() = default;
};
/**
* \ingroup linear allocator
* 1D allocation interface.
*/
class C2LinearAllocation : public _C2LinearCapacityAspect {
public:
/**
* Maps a portion of an allocation starting from |offset| with |size| into local process memory.
* Stores the starting address into |addr|, or NULL if the operation was unsuccessful.
* |fence| will contain an acquire sync fence object. If it is already
* safe to access the buffer contents, then it will contain an empty (already fired) fence.
*
* \param offset starting position of the portion to be mapped (this does not have to
* be page aligned)
* \param size size of the portion to be mapped (this does not have to be page
* aligned)
* \param usage the desired usage. \todo this must be kSoftwareRead and/or
* kSoftwareWrite.
* \param fence a pointer to a fence object if an async mapping is requested. If
* not-null, and acquire fence will be stored here on success, or empty
* fence on failure. If null, the mapping will be synchronous.
* \param addr a pointer to where the starting address of the mapped portion will be
* stored. On failure, nullptr will be stored here.
*
* \todo Only one portion can be mapped at the same time - this is true for gralloc, but there
* is no need for this for 1D buffers.
* \todo Do we need to support sync operation as we could just wait for the fence?
*
* \retval C2_OK the operation was successful
* \retval C2_REFUSED no permission to map the portion
* \retval C2_TIMED_OUT the operation timed out
* \retval C2_DUPLICATE if the allocation is already mapped.
* \retval C2_NO_MEMORY not enough memory to complete the operation
* \retval C2_BAD_VALUE the parameters (offset/size) are invalid or outside the allocation, or
* the usage flags are invalid (caller error)
* \retval C2_CORRUPTED some unknown error prevented the operation from completing (unexpected)
*/
virtual c2_status_t map(
size_t offset, size_t size, C2MemoryUsage usage, C2Fence *fence /* nullable */,
void **addr /* nonnull */) = 0;
/**
* Unmaps a portion of an allocation at |addr| with |size|. These must be parameters previously
* passed to and returned by |map|; otherwise, this operation is a no-op.
*
* \param addr starting address of the mapped region
* \param size size of the mapped region
* \param fence a pointer to a fence object if an async unmapping is requested. If
* not-null, a release fence will be stored here on success, or empty fence
* on failure. This fence signals when the original allocation contains
* all changes that happened to the mapped region. If null, the unmapping
* will be synchronous.
*
* \retval C2_OK the operation was successful
* \retval C2_TIMED_OUT the operation timed out
* \retval C2_NOT_FOUND if the allocation was not mapped previously.
* \retval C2_BAD_VALUE the parameters (addr/size) do not correspond to previously mapped
* regions (caller error)
* \retval C2_CORRUPTED some unknown error prevented the operation from completing (unexpected)
* \retval C2_REFUSED no permission to unmap the portion (unexpected - system)
*/
virtual c2_status_t unmap(void *addr, size_t size, C2Fence *fence /* nullable */) = 0;
/**
* Returns the allocator ID for this allocation. This is useful to put the handle into context.
*/
virtual C2Allocator::id_t getAllocatorId() const = 0;
/**
* Returns a pointer to the allocation handle.
*/
virtual const C2Handle *handle() const = 0;
/**
* Returns true if this is the same allocation as |other|.
*/
virtual bool equals(const std::shared_ptr<C2LinearAllocation> &other) const = 0;
protected:
// \todo should we limit allocation directly?
C2LinearAllocation(size_t capacity) : _C2LinearCapacityAspect(c2_min(capacity, UINT32_MAX)) {}
virtual ~C2LinearAllocation() = default;
};
class C2CircularBlock;
class C2LinearBlock;
class C2GraphicBlock;
/**
* Block pools are used by components to obtain output buffers in an efficient way. They can
* support either linear (1D), circular (1D) or graphic (2D) blocks.
*
* Block pools decouple the recycling of memory/allocations from the components. They are meant to
* be an opaque service (there are no public APIs other than obtaining blocks) provided by the
* platform. Block pools are also meant to decouple allocations from memory used by buffers. This
* is accomplished by allowing pools to allot multiple memory 'blocks' on a single allocation. As
* their name suggest, block pools maintain a pool of memory blocks. When a component asks for
* a memory block, pools will try to return a free memory block already in the pool. If no such
* block exists, they will allocate memory using the backing allocator and allot a block on that
* allocation. When blocks are no longer used in the system, they are recycled back to the block
* pool and are available as free blocks.
*
* Never constructed on stack.
*/
class C2BlockPool {
public:
/**
* Block pool ID type.
*/
typedef uint64_t local_id_t;
enum : local_id_t {
BASIC_LINEAR = 0, ///< ID of basic (unoptimized) block pool for fetching 1D blocks
BASIC_GRAPHIC = 1, ///< ID of basic (unoptimized) block pool for fetching 2D blocks
PLATFORM_START = 0x10,
};
/**
* Returns the ID for this block pool. This ID is used to get this block pool from the platform.
* It is only valid in the current process.
*
* This method MUST be "non-blocking" and return within 1ms.
*
* \return a local ID for this block pool.
*/
virtual local_id_t getLocalId() const = 0;
/**
* Returns the ID of the backing allocator of this block pool.
*
* This method MUST be "non-blocking" and return within 1ms.
*
* \return the ID of the backing allocator of this block pool.
*/
virtual C2Allocator::id_t getAllocatorId() const = 0;
/**
* Obtains a linear writeable block of given |capacity| and |usage|. If successful, the
* block is stored in |block|. Otherwise, |block| is set to 'nullptr'.
*
* \param capacity the size of requested block.
* \param usage the memory usage info for the requested block. Returned blocks will be
* optimized for this usage, but may be used with any usage. One exception:
* protected blocks/buffers can only be used in a protected scenario.
* \param block pointer to where the obtained block shall be stored on success. nullptr will
* be stored here on failure
*
* \retval C2_OK the operation was successful
* \retval C2_NO_MEMORY not enough memory to complete any required allocation
* \retval C2_TIMED_OUT the operation timed out
* \retval C2_REFUSED no permission to complete any required allocation
* \retval C2_BAD_VALUE capacity or usage are not supported (invalid) (caller error)
* \retval C2_OMITTED this pool does not support linear blocks
* \retval C2_CORRUPTED some unknown, unrecoverable error occured during operation (unexpected)
*/
virtual c2_status_t fetchLinearBlock(
uint32_t capacity __unused, C2MemoryUsage usage __unused,
std::shared_ptr<C2LinearBlock> *block /* nonnull */) {
*block = nullptr;
return C2_OMITTED;
}
/**
* Obtains a circular writeable block of given |capacity| and |usage|. If successful, the
* block is stored in |block|. Otherwise, |block| is set to 'nullptr'.
*
* \param capacity the size of requested circular block. (note: the size of the obtained
* block could be slightly larger, e.g. to accommodate any system-required
* alignment)
* \param usage the memory usage info for the requested block. Returned blocks will be
* optimized for this usage, but may be used with any usage. One exception:
* protected blocks/buffers can only be used in a protected scenario.
* \param block pointer to where the obtained block shall be stored on success. nullptr
* will be stored here on failure
*
* \retval C2_OK the operation was successful
* \retval C2_NO_MEMORY not enough memory to complete any required allocation
* \retval C2_TIMED_OUT the operation timed out
* \retval C2_REFUSED no permission to complete any required allocation
* \retval C2_BAD_VALUE capacity or usage are not supported (invalid) (caller error)
* \retval C2_OMITTED this pool does not support circular blocks
* \retval C2_CORRUPTED some unknown, unrecoverable error occured during operation (unexpected)
*/
virtual c2_status_t fetchCircularBlock(
uint32_t capacity __unused, C2MemoryUsage usage __unused,
std::shared_ptr<C2CircularBlock> *block /* nonnull */) {
*block = nullptr;
return C2_OMITTED;
}
/**
* Obtains a 2D graphic block of given |width|, |height|, |format| and |usage|. If successful,
* the block is stored in |block|. Otherwise, |block| is set to 'nullptr'.
*
* \param width the width of requested block (the obtained block could be slightly larger, e.g.
* to accommodate any system-required alignment)
* \param height the height of requested block (the obtained block could be slightly larger,
* e.g. to accommodate any system-required alignment)
* \param format the pixel format of requested block. This could be a vendor specific format.
* \param usage the memory usage info for the requested block. Returned blocks will be
* optimized for this usage, but may be used with any usage. One exception:
* protected blocks/buffers can only be used in a protected scenario.
* \param block pointer to where the obtained block shall be stored on success. nullptr
* will be stored here on failure
*
* \retval C2_OK the operation was successful
* \retval C2_NO_MEMORY not enough memory to complete any required allocation
* \retval C2_TIMED_OUT the operation timed out
* \retval C2_REFUSED no permission to complete any required allocation
* \retval C2_BAD_VALUE width, height, format or usage are not supported (invalid) (caller
* error)
* \retval C2_OMITTED this pool does not support 2D blocks
* \retval C2_CORRUPTED some unknown, unrecoverable error occured during operation (unexpected)
*/
virtual c2_status_t fetchGraphicBlock(
uint32_t width __unused, uint32_t height __unused, uint32_t format __unused,
C2MemoryUsage usage __unused,
std::shared_ptr<C2GraphicBlock> *block /* nonnull */) {
*block = nullptr;
return C2_OMITTED;
}
virtual ~C2BlockPool() = default;
protected:
C2BlockPool() = default;
};
/// @}
// ================================================================================================
// BLOCKS
// ================================================================================================
/**
* Blocks are sections of allocations. They can be either 1D or 2D.
*/
class C2LinearAllocation;
/**
* A 1D block.
*
* \note capacity() is not meaningful for users of blocks; instead size() is the capacity of the
* usable portion. Use and offset() and size() if accessing the block directly through its handle
* to represent the allotted range of the underlying allocation to this block.
*/
class C2Block1D : public _C2LinearRangeAspect {
public:
/**
* Returns the underlying handle for this allocation.
*
* \note that the block and its block pool has shared ownership of the handle
* and if all references to the block are released, the underlying block
* allocation may get reused even if a client keeps a clone of this handle.
*/
const C2Handle *handle() const;
/**
* Returns the allocator's ID that created the underlying allocation for this block. This
* provides the context for understanding the handle.
*/
C2Allocator::id_t getAllocatorId() const;
protected:
class Impl;
/** construct a block. */
C2Block1D(std::shared_ptr<Impl> impl, const _C2LinearRangeAspect &range);
friend struct _C2BlockFactory;
std::shared_ptr<Impl> mImpl;
};
/**
* Read view provides read-only access for a linear memory segment.
*
* This class is copiable.
*/
class C2ReadView : public _C2LinearCapacityAspect {
public:
/**
* \return pointer to the start of the block or nullptr on error.
* This pointer is only valid during the lifetime of this view or until it is released.
*/
const uint8_t *data() const;
/**
* Returns a portion of this view.
*
* \param offset the start offset of the portion. \note This is clamped to the capacity of this
* view.
* \param size the size of the portion. \note This is clamped to the remaining data from offset.
*
* \return a read view containing a portion of this view
*/
C2ReadView subView(size_t offset, size_t size) const;
/**
* \return error during the creation/mapping of this view.
*/
c2_status_t error() const;
/**
* Releases this view. This sets error to C2_NO_INIT.
*/
//void release();
protected:
class Impl;
C2ReadView(std::shared_ptr<Impl> impl, uint32_t offset, uint32_t size);
explicit C2ReadView(c2_status_t error);
private:
friend struct _C2BlockFactory;
std::shared_ptr<Impl> mImpl;
uint32_t mOffset; /**< offset into the linear block backing this read view */
};
/**
* Write view provides read/write access for a linear memory segment.
*
* This class is copiable. \todo movable only?
*/
class C2WriteView : public _C2EditableLinearRangeAspect {
public:
/**
* Start of the block.
*
* \return pointer to the start of the block or nullptr on error.
* This pointer is only valid during the lifetime of this view or until it is released.
*/
uint8_t *base();
/**
* \return pointer to the block at the current offset or nullptr on error.
* This pointer is only valid during the lifetime of this view or until it is released.
*/
uint8_t *data();
/**
* \return error during the creation/mapping of this view.
*/
c2_status_t error() const;
/**
* Releases this view. This sets error to C2_NO_INIT.
*/
//void release();
protected:
class Impl;
C2WriteView(std::shared_ptr<Impl> impl);
explicit C2WriteView(c2_status_t error);
private:
friend struct _C2BlockFactory;
std::shared_ptr<Impl> mImpl;
};
/**
* A constant (read-only) linear block (portion of an allocation) with an acquire fence.
* Blocks are unmapped when created, and can be mapped into a read view on demand.
*
* This class is copiable and contains a reference to the allocation that it is based on.
*/
class C2ConstLinearBlock : public C2Block1D {
public:
/**
* Maps this block into memory and returns a read view for it.
*
* \return a read view for this block.
*/
C2Acquirable<C2ReadView> map() const;
/**
* Returns a portion of this block.
*
* \param offset the start offset of the portion. \note This is clamped to the capacity of this
* block.
* \param size the size of the portion. \note This is clamped to the remaining data from offset.
*
* \return a constant linear block containing a portion of this block
*/
C2ConstLinearBlock subBlock(size_t offset, size_t size) const;
/**
* Returns the acquire fence for this block.
*
* \return a fence that must be waited on before reading the block.
*/
C2Fence fence() const { return mFence; }
protected:
C2ConstLinearBlock(std::shared_ptr<Impl> impl, const _C2LinearRangeAspect &range, C2Fence mFence);
private:
friend struct _C2BlockFactory;
C2Fence mFence;
};
/**
* Linear block is a writeable 1D block. Once written, it can be shared in whole or in parts with
* consumers/readers as read-only const linear block(s).
*/
class C2LinearBlock : public C2Block1D {
public:
/**
* Maps this block into memory and returns a write view for it.
*
* \return a write view for this block.
*/
C2Acquirable<C2WriteView> map();
/**
* Creates a read-only const linear block for a portion of this block; optionally protected
* by an acquire fence. There are two ways to use this:
*
* 1) share ready block after writing data into the block. In this case no fence shall be
* supplied, and the block shall not be modified after calling this method.
* 2) share block metadata before actually (finishing) writing the data into the block. In
* this case a fence must be supplied that will be triggered when the data is written.
* The block shall be modified only until firing the event for the fence.
*/
C2ConstLinearBlock share(size_t offset, size_t size, C2Fence fence);
protected:
C2LinearBlock(std::shared_ptr<Impl> impl, const _C2LinearRangeAspect &range);
friend struct _C2BlockFactory;
};
/// @}
/**************************************************************************************************
CIRCULAR BLOCKS AND VIEWS
**************************************************************************************************/
/// \defgroup circular Circular buffer support
/// @{
/**
* Circular blocks can be used to share data between a writer and a reader (and/or other consumers)-
* in a memory-efficient way by reusing a section of memory. Circular blocks are a bit more complex
* than single reader/single writer schemes to facilitate block-based consuming of data.
*
* They can operate in two modes:
*
* 1) one writer that creates blocks to be consumed (this model can be used by components)
*
* 2) one writer that writes continuously, and one reader that can creates blocks to be consumed
* by further recipients (this model is used by the framework, and cannot be used by components.)
*
* Circular blocks have four segments with running pointers:
* - reserved: data reserved and available for the writer
* - committed: data committed by the writer and available to the reader (if present)
* - used: data used by consumers (if present)
* - available: unused data available to be reserved
*/
class C2CircularBlock : public C2Block1D {
// TODO: add methods
private:
size_t mReserved __unused; // end of reserved section
size_t mCommitted __unused; // end of committed section
size_t mUsed __unused; // end of used section
size_t mFree __unused; // end of free section
};
class _C2CircularBlockSegment : public _C2LinearCapacityAspect {
public:
/**
* Returns the available size for this segment.
*
* \return currently available size for this segment
*/
size_t available() const;
/**
* Reserve some space for this segment from its current start.
*
* \param size desired space in bytes
* \param fence a pointer to an acquire fence. If non-null, the reservation is asynchronous and
* a fence will be stored here that will be signaled when the reservation is
* complete. If null, the reservation is synchronous.
*
* \retval C2_OK the space was successfully reserved
* \retval C2_NO_MEMORY the space requested cannot be reserved
* \retval C2_TIMED_OUT the reservation timed out \todo when?
* \retval C2_CORRUPTED some unknown error prevented reserving space. (unexpected)
*/
c2_status_t reserve(size_t size, C2Fence *fence /* nullable */);
/**
* Abandons a portion of this segment. This will move to the beginning of this segment.
*
* \note This methods is only allowed if this segment is producing blocks.
*
* \param size number of bytes to abandon
*
* \retval C2_OK the data was successfully abandoned
* \retval C2_TIMED_OUT the operation timed out (unexpected)
* \retval C2_CORRUPTED some unknown error prevented abandoning the data (unexpected)
*/
c2_status_t abandon(size_t size);
/**
* Share a portion as block(s) with consumers (these are moved to the used section).
*
* \note This methods is only allowed if this segment is producing blocks.
* \note Share does not move the beginning of the segment. (\todo add abandon/offset?)
*
* \param size number of bytes to share
* \param fence fence to be used for the section
* \param blocks vector where the blocks of the section are appended to
*
* \retval C2_OK the portion was successfully shared
* \retval C2_NO_MEMORY not enough memory to share the portion
* \retval C2_TIMED_OUT the operation timed out (unexpected)
* \retval C2_CORRUPTED some unknown error prevented sharing the data (unexpected)
*/
c2_status_t share(size_t size, C2Fence fence, std::vector<C2ConstLinearBlock> &blocks);
/**
* Returns the beginning offset of this segment from the start of this circular block.
*
* @return beginning offset
*/
size_t begin();
/**
* Returns the end offset of this segment from the start of this circular block.
*
* @return end offset
*/
size_t end();
};
/**
* A circular write-view is a dynamic mapped view for a segment of a circular block. Care must be
* taken when using this view so that only the section owned by the segment is modified.
*/
class C2CircularWriteView : public _C2LinearCapacityAspect {
public:
/**
* Start of the circular block.
* \note the segment does not own this pointer.
*
* \return pointer to the start of the circular block or nullptr on error.
*/
uint8_t *base();
/**
* \return error during the creation/mapping of this view.
*/
c2_status_t error() const;
};
/**
* The writer of a circular buffer.
*
* Can commit data to a reader (not supported for components) OR share data blocks directly with a
* consumer.
*
* If a component supports outputting data into circular buffers, it must allocate a circular
* block and use a circular writer.
*/
class C2CircularWriter : public _C2CircularBlockSegment {
public:
/**
* Commits a portion of this segment to the next segment. This moves the beginning of the
* segment.
*
* \param size number of bytes to commit to the next segment
* \param fence fence used for the commit (the fence must signal before the data is committed)
*/
c2_status_t commit(size_t size, C2Fence fence);
/**
* Maps this block into memory and returns a write view for it.
*
* \return a write view for this block.
*/
C2Acquirable<C2CircularWriteView> map();
};
/// @}
/// \defgroup graphic Graphic Data Blocks
/// @{
/**
* C2Rect: rectangle type with non-negative coordinates.
*
* \note This struct has public fields without getters/setters. All methods are inline.
*/
struct C2Rect {
// public:
uint32_t width;
uint32_t height;
uint32_t left;
uint32_t top;
constexpr inline C2Rect()
: C2Rect(0, 0, 0, 0) { }
constexpr inline C2Rect(uint32_t width_, uint32_t height_)
: C2Rect(width_, height_, 0, 0) { }
constexpr C2Rect inline at(uint32_t left_, uint32_t top_) const {
return C2Rect(width, height, left_, top_);
}
// utility methods
inline constexpr bool isEmpty() const {
return width == 0 || height == 0;
}
inline constexpr bool isValid() const {
return left <= ~width && top <= ~height;
}
inline constexpr operator bool() const {
return isValid() && !isEmpty();
}
inline constexpr bool operator!() const {
return !bool(*this);
}
C2_ALLOW_OVERFLOW
inline constexpr bool contains(const C2Rect &other) const {
if (!isValid() || !other.isValid()) {
return false;
} else {
return left <= other.left && top <= other.top
&& left + width >= other.left + other.width
&& top + height >= other.top + other.height;
}
}
inline constexpr bool operator==(const C2Rect &other) const {
if (!isValid()) {
return !other.isValid();
} else {
return left == other.left && top == other.top
&& width == other.width && height == other.height;
}
}
inline constexpr bool operator!=(const C2Rect &other) const {
return !operator==(other);
}
inline constexpr bool operator>=(const C2Rect &other) const {
return contains(other);
}
inline constexpr bool operator>(const C2Rect &other) const {
return contains(other) && !operator==(other);
}
inline constexpr bool operator<=(const C2Rect &other) const {
return other.contains(*this);
}
inline constexpr bool operator<(const C2Rect &other) const {
return other.contains(*this) && !operator==(other);
}
C2_ALLOW_OVERFLOW
inline constexpr uint32_t right() const {
return left + width;
}
C2_ALLOW_OVERFLOW
inline constexpr uint32_t bottom() const {
return top + height;
}
C2_ALLOW_OVERFLOW
inline constexpr C2Rect intersect(const C2Rect &other) const {
return C2Rect(c2_min(right(), other.right()) - c2_max(left, other.left),
c2_min(bottom(), other.bottom()) - c2_max(top, other.top),
c2_max(left, other.left),
c2_max(top, other.top));
}
/** clamps right and bottom to top, left if they overflow */
inline constexpr C2Rect normalize() const {
return C2Rect(c2_max(left, right()) - left, c2_max(top, bottom()) - top, left, top);
}
private:
/// note: potentially unusual argument order
constexpr inline C2Rect(uint32_t width_, uint32_t height_, uint32_t left_, uint32_t top_)
: width(width_),
height(height_),
left(left_),
top(top_) { }
};
/**
* Interface for objects that have a width and height (planar capacity).
*/
class _C2PlanarCapacityAspect {
/// \name Planar capacity interface
/// @{
public:
inline constexpr uint32_t width() const { return _mWidth; }
inline constexpr uint32_t height() const { return _mHeight; }
inline constexpr operator C2Rect() const {
return C2Rect(_mWidth, _mHeight);
}
protected:
inline constexpr _C2PlanarCapacityAspect(uint32_t width, uint32_t height)
: _mWidth(width), _mHeight(height) { }
inline explicit constexpr _C2PlanarCapacityAspect(const _C2PlanarCapacityAspect *parent)
: _mWidth(parent == nullptr ? 0 : parent->width()),
_mHeight(parent == nullptr ? 0 : parent->height()) { }
private:
uint32_t _mWidth;
uint32_t _mHeight;
/// @}
};
/**
* C2PlaneInfo: information on the layout of a singe flexible plane.
*
* Public fields without getters/setters.
*/
struct C2PlaneInfo {
//public:
enum channel_t : uint32_t {
CHANNEL_Y, ///< luma
CHANNEL_R, ///< red
CHANNEL_G, ///< green
CHANNEL_B, ///< blue
CHANNEL_A, ///< alpha
CHANNEL_CR, ///< Cr
CHANNEL_CB, ///< Cb
} channel;
int32_t colInc; ///< column increment in bytes. may be negative
int32_t rowInc; ///< row increment in bytes. may be negative
uint32_t colSampling; ///< subsampling compared to width (must be a power of 2)
uint32_t rowSampling; ///< subsampling compared to height (must be a power of 2)
uint32_t allocatedDepth; ///< size of each sample (must be a multiple of 8)
uint32_t bitDepth; ///< significant bits per sample
/**
* the right shift of the significant bits in the sample. E.g. if a 10-bit significant
* value is laid out in a 16-bit allocation aligned to LSB (values 0-1023), rightShift
* would be 0 as the 16-bit value read from the sample does not need to be right shifted
* and can be used as is (after applying a 10-bit mask of 0x3FF).
*
* +--------+--------+
* | VV|VVVVVVVV|
* +--------+--------+
* 15 8 7 0
*
* If the value is laid out aligned to MSB, rightShift would be 6, as the value read
* from the allocated sample must be right-shifted by 6 to get the actual sample value.
*
* +--------+--------+
* |VVVVVVVV|VV |
* +--------+--------+
* 15 8 7 0
*/
uint32_t rightShift;
enum endianness_t : uint32_t {
NATIVE,
LITTLE_END, // LITTLE_ENDIAN is reserved macro
BIG_END, // BIG_ENDIAN is a reserved macro
} endianness; ///< endianness of the samples
/**
* The following two fields define the relation between multiple planes. If multiple planes are
* interleaved, they share a root plane (whichever plane's start address is the lowest), and
* |offset| is the offset of this plane inside the root plane (in bytes). |rootIx| is the index
* of the root plane. If a plane is independent, rootIx is its index and offset is 0.
*/
uint32_t rootIx; ///< index of the root plane
uint32_t offset; ///< offset of this plane inside of the root plane
inline constexpr ssize_t minOffset(uint32_t width, uint32_t height) const {
ssize_t offs = 0;
if (width > 0 && colInc < 0) {
offs += colInc * (ssize_t)(width - 1);
}
if (height > 0 && rowInc < 0) {
offs += rowInc * (ssize_t)(height - 1);
}
return offs;
}
inline constexpr ssize_t maxOffset(uint32_t width, uint32_t height) const {
ssize_t offs = (allocatedDepth + 7) >> 3;
if (width > 0 && colInc > 0) {
offs += colInc * (ssize_t)(width - 1);
}
if (height > 0 && rowInc > 0) {
offs += rowInc * (ssize_t)(height - 1);
}
return offs;
}
} C2_PACK;
struct C2PlanarLayout {
//public:
enum type_t : uint32_t {
TYPE_UNKNOWN = 0,
TYPE_YUV = 0x100, ///< YUV image with 3 planes
TYPE_YUVA, ///< YUVA image with 4 planes
TYPE_RGB, ///< RGB image with 3 planes
TYPE_RGBA, ///< RBGA image with 4 planes
};
type_t type; // image type
uint32_t numPlanes; // number of component planes
uint32_t rootPlanes; // number of layout planes (root planes)
enum plane_index_t : uint32_t {
PLANE_Y = 0,
PLANE_U = 1,
PLANE_V = 2,
PLANE_R = 0,
PLANE_G = 1,
PLANE_B = 2,
PLANE_A = 3,
MAX_NUM_PLANES = 4,
};
C2PlaneInfo planes[MAX_NUM_PLANES];
};
/**
* Aspect for objects that have a planar section (crop rectangle).
*
* This class is copiable.
*/
class _C2PlanarSectionAspect : public _C2PlanarCapacityAspect {
/// \name Planar section interface
/// @{
private:
inline constexpr _C2PlanarSectionAspect(uint32_t width, uint32_t height, const C2Rect &crop)
: _C2PlanarCapacityAspect(width, height),
mCrop(C2Rect(std::min(width - std::min(crop.left, width), crop.width),
std::min(height - std::min(crop.top, height), crop.height)).at(
std::min(crop.left, width),
std::min(crop.height, height))) {
}
public:
// crop can be an empty rect, does not have to line up with subsampling
// NOTE: we do not support floating-point crop
inline constexpr C2Rect crop() const { return mCrop; }
/**
* Returns a child planar section for |crop|, where the capacity represents this section.
*/
inline constexpr _C2PlanarSectionAspect childSection(const C2Rect &crop) const {
return _C2PlanarSectionAspect(
mCrop.width, mCrop.height,
// crop and translate |crop| rect
C2Rect(c2_min(mCrop.right() - c2_clamp(mCrop.left, crop.left, mCrop.right()),
crop.width),
c2_min(mCrop.bottom() - c2_clamp(mCrop.top, crop.top, mCrop.bottom()),
crop.height))
.at(c2_clamp(mCrop.left, crop.left, mCrop.right()) - mCrop.left,
c2_clamp(mCrop.top, crop.top, mCrop.bottom()) - mCrop.top));
}
protected:
inline constexpr _C2PlanarSectionAspect(const _C2PlanarCapacityAspect *parent)
: _C2PlanarCapacityAspect(parent), mCrop(width(), height()) {}
inline constexpr _C2PlanarSectionAspect(const _C2PlanarCapacityAspect *parent, const C2Rect &crop)
: _C2PlanarCapacityAspect(parent),
mCrop(parent == nullptr ? C2Rect() : ((C2Rect)*parent).intersect(crop).normalize()) { }
inline constexpr _C2PlanarSectionAspect(const _C2PlanarSectionAspect *parent, const C2Rect &crop)
: _C2PlanarCapacityAspect(parent),
mCrop(parent == nullptr ? C2Rect() : parent->crop().intersect(crop).normalize()) { }
private:
friend class _C2EditablePlanarSectionAspect;
C2Rect mCrop;
/// @}
};
/**
* Aspect for objects that have an editable planar section (crop rectangle).
*
* This class is copiable.
*/
class _C2EditablePlanarSectionAspect : public _C2PlanarSectionAspect {
/// \name Planar section interface
/// @{
using _C2PlanarSectionAspect::_C2PlanarSectionAspect;
public:
// crop can be an empty rect, does not have to line up with subsampling
// NOTE: we do not support floating-point crop
inline constexpr C2Rect crop() const { return mCrop; }
/**
* Sets crop to crop intersected with [(0,0) .. (width, height)]
*/
inline void setCrop_be(const C2Rect &crop) {
mCrop.left = std::min(width(), crop.left);
mCrop.top = std::min(height(), crop.top);
// It's guaranteed that mCrop.left <= width() && mCrop.top <= height()
mCrop.width = std::min(width() - mCrop.left, crop.width);
mCrop.height = std::min(height() - mCrop.top, crop.height);
}
/**
* If crop is within the dimensions of this object, it sets crop to it.
*
* \return true iff crop is within the dimensions of this object
*/
inline bool setCrop(const C2Rect &crop) {
if (width() < crop.width || height() < crop.height
|| width() - crop.width < crop.left || height() - crop.height < crop.top) {
return false;
}
mCrop = crop;
return true;
}
/// @}
};
/**
* Utility class for safe range calculations using size_t-s.
*/
class C2PlanarSection : public _C2PlanarSectionAspect {
public:
inline constexpr C2PlanarSection(const _C2PlanarCapacityAspect &parent, const C2Rect &crop)
: _C2PlanarSectionAspect(&parent, crop) { }
inline constexpr C2PlanarSection(const _C2PlanarSectionAspect &parent, const C2Rect &crop)
: _C2PlanarSectionAspect(&parent, crop) { }
inline constexpr C2PlanarSection intersect(const C2Rect &crop) const {
return C2PlanarSection(*this, crop);
}
};
/**
* Utility class for simple and safe planar capacity and section construction.
*/
class C2PlanarCapacity : public _C2PlanarCapacityAspect {
public:
inline constexpr explicit C2PlanarCapacity(size_t width, size_t height)
: _C2PlanarCapacityAspect(c2_min(width, std::numeric_limits<uint32_t>::max()),
c2_min(height, std::numeric_limits<uint32_t>::max())) { }
inline constexpr C2PlanarSection section(const C2Rect &crop) const {
return C2PlanarSection(*this, crop);
}
};
/**
* \ingroup graphic allocator
* 2D allocation interface.
*/
class C2GraphicAllocation : public _C2PlanarCapacityAspect {
public:
/**
* Maps a rectangular section (as defined by |rect|) of a 2D allocation into local process
* memory for flexible access. On success, it fills out |layout| with the plane specifications
* and fills the |addr| array with pointers to the first byte of the top-left pixel of each
* plane used. Otherwise, it leaves |layout| and |addr| untouched. |fence| will contain
* an acquire sync fence object. If it is already safe to access the
* buffer contents, then it will be an empty (already fired) fence.
*
* Safe regions for the pointer addresses returned can be gotten via C2LayoutInfo.minOffset()/
* maxOffset().
*
* \param rect section to be mapped (this does not have to be aligned)
* \param usage the desired usage. \todo this must be kSoftwareRead and/or
* kSoftwareWrite.
* \param fence a pointer to a fence object if an async mapping is requested. If
* not-null, and acquire fence will be stored here on success, or empty
* fence on failure. If null, the mapping will be synchronous.
* \param layout a pointer to where the mapped planes' descriptors will be
* stored. On failure, nullptr will be stored here.
* \param addr pointer to an array with at least C2PlanarLayout::MAX_NUM_PLANES
* elements. Only layout.numPlanes elements will be modified on success.
*
* \retval C2_OK the operation was successful
* \retval C2_REFUSED no permission to map the section
* \retval C2_DUPLICATE there is already a mapped region and this allocation cannot support
* multi-mapping (caller error)
* \retval C2_TIMED_OUT the operation timed out
* \retval C2_NO_MEMORY not enough memory to complete the operation
* \retval C2_BAD_VALUE the parameters (rect) are invalid or outside the allocation, or the
* usage flags are invalid (caller error)
* \retval C2_CORRUPTED some unknown error prevented the operation from completing (unexpected)
*/
virtual c2_status_t map(
C2Rect rect, C2MemoryUsage usage, C2Fence *fence,
C2PlanarLayout *layout /* nonnull */, uint8_t **addr /* nonnull */) = 0;
/**
* Unmaps a section of an allocation at |addr| with |rect|. These must be parameters previously
* passed to and returned by |map|; otherwise, this operation is a no-op.
*
* \param addr pointer to an array with at least C2PlanarLayout::MAX_NUM_PLANES
* elements containing the starting addresses of the mapped layers
* \param rect boundaries of the mapped section
* \param fence a pointer to a fence object if an async unmapping is requested. If
* not-null, a release fence will be stored here on success, or empty fence
* on failure. This fence signals when the original allocation contains
* all changes that happened to the mapped section. If null, the unmapping
* will be synchronous.
*
* \retval C2_OK the operation was successful
* \retval C2_TIMED_OUT the operation timed out
* \retval C2_NOT_FOUND there is no such mapped region (caller error)
* \retval C2_CORRUPTED some unknown error prevented the operation from completing (unexpected)
* \retval C2_REFUSED no permission to unmap the section (unexpected - system)
*/
virtual c2_status_t unmap(
uint8_t **addr /* nonnull */, C2Rect rect, C2Fence *fence /* nullable */) = 0;
/**
* Returns the allocator ID for this allocation. This is useful to put the handle into context.
*/
virtual C2Allocator::id_t getAllocatorId() const = 0;
/**
* Returns a pointer to the allocation handle.
*/
virtual const C2Handle *handle() const = 0;
/**
* Returns true if this is the same allocation as |other|.
*/
virtual bool equals(const std::shared_ptr<const C2GraphicAllocation> &other) const = 0;
protected:
using _C2PlanarCapacityAspect::_C2PlanarCapacityAspect;
virtual ~C2GraphicAllocation() = default;
};
class C2GraphicAllocation;
/**
* A 2D block.
*
* \note width()/height() is not meaningful for users of blocks; instead, crop().width() and
* crop().height() is the capacity of the usable portion. Use and crop() if accessing the block
* directly through its handle to represent the allotted region of the underlying allocation to this
* block.
*/
class C2Block2D : public _C2PlanarSectionAspect {
public:
/**
* Returns the underlying handle for this allocation.
*
* \note that the block and its block pool has shared ownership of the handle
* and if all references to the block are released, the underlying block
* allocation may get reused even if a client keeps a clone of this handle.
*/
const C2Handle *handle() const;
/**
* Returns the allocator's ID that created the underlying allocation for this block. This
* provides the context for understanding the handle.
*/
C2Allocator::id_t getAllocatorId() const;
protected:
class Impl;
C2Block2D(std::shared_ptr<Impl> impl, const _C2PlanarSectionAspect &section);
friend struct _C2BlockFactory;
std::shared_ptr<Impl> mImpl;
};
/**
* Graphic view provides read or read-write access for a graphic block.
*
* This class is copiable.
*
* \note Due to the subsampling of graphic buffers, a read view must still contain a crop rectangle
* to ensure subsampling is followed. This results in nearly identical interface between read and
* write views, so C2GraphicView can encompass both of them.
*/
class C2GraphicView : public _C2EditablePlanarSectionAspect {
public:
/**
* \return array of pointers (of layout().numPlanes elements) to the start of the planes or
* nullptr on error. Regardless of crop rect, they always point to the top-left corner of each
* plane. Access outside of the crop rect results in an undefined behavior.
*/
const uint8_t *const *data() const;
/**
* \return array of pointers (of layout().numPlanes elements) to the start of the planes or
* nullptr on error. Regardless of crop rect, they always point to the top-left corner of each
* plane. Access outside of the crop rect results in an undefined behavior.
*/
uint8_t *const *data();
/**
* \return layout of the graphic block to interpret the returned data.
*/
const C2PlanarLayout layout() const;
/**
* Returns a section of this view.
*
* \param rect the dimension of the section. \note This is clamped to the crop of this view.
*
* \return a read view containing the requested section of this view
*/
const C2GraphicView subView(const C2Rect &rect) const;
C2GraphicView subView(const C2Rect &rect);
/**
* \return error during the creation/mapping of this view.
*/
c2_status_t error() const;
protected:
class Impl;
C2GraphicView(std::shared_ptr<Impl> impl, const _C2PlanarSectionAspect &section);
explicit C2GraphicView(c2_status_t error);
private:
friend struct _C2BlockFactory;
std::shared_ptr<Impl> mImpl;
};
/**
* A constant (read-only) graphic block (portion of an allocation) with an acquire fence.
* Blocks are unmapped when created, and can be mapped into a read view on demand.
*
* This class is copiable and contains a reference to the allocation that it is based on.
*/
class C2ConstGraphicBlock : public C2Block2D {
public:
/**
* Maps this block into memory and returns a read view for it.
*
* \return a read view for this block.
*/
C2Acquirable<const C2GraphicView> map() const;
/**
* Returns a section of this block.
*
* \param rect the coordinates of the section. \note This is clamped to the crop rectangle of
* this block.
*
* \return a constant graphic block containing a portion of this block
*/
C2ConstGraphicBlock subBlock(const C2Rect &rect) const;
/**
* Returns the acquire fence for this block.
*
* \return a fence that must be waited on before reading the block.
*/
C2Fence fence() const { return mFence; }
protected:
C2ConstGraphicBlock(
std::shared_ptr<Impl> impl, const _C2PlanarSectionAspect &section, C2Fence fence);
private:
friend struct _C2BlockFactory;
C2Fence mFence;
};
/**
* Graphic block is a writeable 2D block. Once written, it can be shared in whole or in part with
* consumers/readers as read-only const graphic block.
*/
class C2GraphicBlock : public C2Block2D {
public:
/**
* Maps this block into memory and returns a write view for it.
*
* \return a write view for this block.
*/
C2Acquirable<C2GraphicView> map();
/**
* Creates a read-only const linear block for a portion of this block; optionally protected
* by an acquire fence. There are two ways to use this:
*
* 1) share ready block after writing data into the block. In this case no fence shall be
* supplied, and the block shall not be modified after calling this method.
* 2) share block metadata before actually (finishing) writing the data into the block. In
* this case a fence must be supplied that will be triggered when the data is written.
* The block shall be modified only until firing the event for the fence.
*/
C2ConstGraphicBlock share(const C2Rect &crop, C2Fence fence);
protected:
C2GraphicBlock(std::shared_ptr<Impl> impl, const _C2PlanarSectionAspect &section);
friend struct _C2BlockFactory;
};
/// @}
/// \defgroup buffer_onj Buffer objects
/// @{
// ================================================================================================
// BUFFERS
// ================================================================================================
/// \todo: Do we still need this?
///
// There are 2 kinds of buffers: linear or graphic. Linear buffers can contain a single block, or
// a list of blocks (LINEAR_CHUNKS). Support for list of blocks is optional, and can allow consuming
// data from circular buffers or scattered data sources without extra memcpy. Currently, list of
// graphic blocks is not supported.
class C2LinearBuffer; // read-write buffer
class C2GraphicBuffer; // read-write buffer
class C2LinearChunksBuffer;
/**
* C2BufferData: the main, non-meta data of a buffer. A buffer can contain either linear blocks
* or graphic blocks, and can contain either a single block or multiple blocks. This is determined
* by its type.
*/
class C2BufferData {
public:
/**
* The type of buffer data.
*/
enum type_t : uint32_t {
INVALID, ///< invalid buffer type. Do not use.
LINEAR, ///< the buffer contains a single linear block
LINEAR_CHUNKS, ///< the buffer contains one or more linear blocks
GRAPHIC, ///< the buffer contains a single graphic block
GRAPHIC_CHUNKS, ///< the buffer contains one of more graphic blocks
};
typedef type_t Type; // deprecated
/**
* Gets the type of this buffer (data).
* \return the type of this buffer data.
*/
type_t type() const;
/**
* Gets the linear blocks of this buffer.
* \return a constant list of const linear blocks of this buffer.
* \retval empty list if this buffer does not contain linear block(s).
*/
const std::vector<C2ConstLinearBlock> linearBlocks() const;
/**
* Gets the graphic blocks of this buffer.
* \return a constant list of const graphic blocks of this buffer.
* \retval empty list if this buffer does not contain graphic block(s).
*/
const std::vector<C2ConstGraphicBlock> graphicBlocks() const;
private:
class Impl;
std::shared_ptr<Impl> mImpl;
protected:
// no public constructor
explicit C2BufferData(const std::vector<C2ConstLinearBlock> &blocks);
explicit C2BufferData(const std::vector<C2ConstGraphicBlock> &blocks);
};
/**
* C2Buffer: buffer base class. These are always used as shared_ptrs. Though the underlying buffer
* objects (native buffers, ion buffers, or dmabufs) are reference-counted by the system,
* C2Buffers hold only a single reference.
*
* These objects cannot be used on the stack.
*/
class C2Buffer {
public:
/**
* Gets the buffer's data.
*
* \return the buffer's data.
*/
const C2BufferData data() const;
/**
* These will still work if used in onDeathNotify.
*/
#if 0
inline std::shared_ptr<C2LinearBuffer> asLinearBuffer() const {
return mType == LINEAR ? std::shared_ptr::reinterpret_cast<C2LinearBuffer>(this) : nullptr;
}
inline std::shared_ptr<C2GraphicBuffer> asGraphicBuffer() const {
return mType == GRAPHIC ? std::shared_ptr::reinterpret_cast<C2GraphicBuffer>(this) : nullptr;
}
inline std::shared_ptr<C2CircularBuffer> asCircularBuffer() const {
return mType == CIRCULAR ? std::shared_ptr::reinterpret_cast<C2CircularBuffer>(this) : nullptr;
}
#endif
///@name Pre-destroy notification handling
///@{
/**
* Register for notification just prior to the destruction of this object.
*/
typedef void (*OnDestroyNotify) (const C2Buffer *buf, void *arg);
/**
* Registers for a pre-destroy notification. This is called just prior to the destruction of
* this buffer (when this buffer is no longer valid.)
*
* \param onDestroyNotify the notification callback
* \param arg an arbitrary parameter passed to the callback
*
* \retval C2_OK the registration was successful.
* \retval C2_DUPLICATE a notification was already registered for this callback and argument
* \retval C2_NO_MEMORY not enough memory to register for this callback
* \retval C2_CORRUPTED an unknown error prevented the registration (unexpected)
*/
c2_status_t registerOnDestroyNotify(OnDestroyNotify onDestroyNotify, void *arg = nullptr);
/**
* Unregisters a previously registered pre-destroy notification.
*
* \param onDestroyNotify the notification callback
* \param arg an arbitrary parameter passed to the callback
*
* \retval C2_OK the unregistration was successful.
* \retval C2_NOT_FOUND the notification was not found
* \retval C2_CORRUPTED an unknown error prevented the registration (unexpected)
*/
c2_status_t unregisterOnDestroyNotify(OnDestroyNotify onDestroyNotify, void *arg = nullptr);
///@}
virtual ~C2Buffer() = default;
///@name Buffer-specific arbitrary metadata handling
///@{
/**
* Gets the list of metadata associated with this buffer.
*
* \return a constant list of info objects associated with this buffer.
*/
const std::vector<std::shared_ptr<const C2Info>> info() const;
/**
* Attaches (or updates) an (existing) metadata for this buffer.
* If the metadata is stream specific, the stream information will be reset.
*
* \param info Metadata to update
*
* \retval C2_OK the metadata was successfully attached/updated.
* \retval C2_NO_MEMORY not enough memory to attach the metadata (this return value is not
* used if the same kind of metadata is already attached to the buffer).
*/
c2_status_t setInfo(const std::shared_ptr<C2Info> &info);
/**
* Checks if there is a certain type of metadata attached to this buffer.
*
* \param index the parameter type of the metadata
*
* \return true iff there is a metadata with the parameter type attached to this buffer.
*/
bool hasInfo(C2Param::Type index) const;
/**
* Checks if there is a certain type of metadata attached to this buffer, and returns a
* shared pointer to it if there is. Returns an empty shared pointer object (nullptr) if there
* is not.
*
* \param index the parameter type of the metadata
*
* \return shared pointer to the metadata.
*/
std::shared_ptr<const C2Info> getInfo(C2Param::Type index) const;
/**
* Removes a metadata from the buffer.
*/
std::shared_ptr<C2Info> removeInfo(C2Param::Type index);
///@}
/**
* Creates a buffer containing a single linear block.
*
* \param block the content of the buffer.
*
* \return shared pointer to the created buffer.
*/
static std::shared_ptr<C2Buffer> CreateLinearBuffer(const C2ConstLinearBlock &block);
/**
* Creates a buffer containing a single graphic block.
*
* \param block the content of the buffer.
*
* \return shared pointer to the created buffer.
*/
static std::shared_ptr<C2Buffer> CreateGraphicBuffer(const C2ConstGraphicBlock &block);
protected:
// no public constructor
explicit C2Buffer(const std::vector<C2ConstLinearBlock> &blocks);
explicit C2Buffer(const std::vector<C2ConstGraphicBlock> &blocks);
private:
class Impl;
std::shared_ptr<Impl> mImpl;
// Type _mType;
};
/**
* An extension of C2Info objects that can contain arbitrary buffer data.
*
* \note This object is not describable and contains opaque data.
*/
class C2InfoBuffer {
public:
/**
* Gets the index of this info object.
*
* \return the parameter index.
*/
const C2Param::Index index() const;
/**
* Gets the buffer's data.
*
* \return the buffer's data.
*/
const C2BufferData data() const;
};
/// @}
/// \cond INTERNAL
/// \todo These are no longer used
/// \addtogroup linear
/// @{
/** \deprecated */
class C2LinearBuffer
: public C2Buffer, public _C2LinearRangeAspect,
public std::enable_shared_from_this<C2LinearBuffer> {
public:
/** \todo what is this? */
const C2Handle *handle() const;
protected:
inline C2LinearBuffer(const C2ConstLinearBlock &block);
private:
class Impl;
Impl *mImpl;
};
class C2ReadCursor;
class C2WriteCursor {
public:
uint32_t remaining() const; // remaining data to be read
void commit(); // commits the current position. discard data before current position
void reset() const; // resets position to the last committed position
// slices off at most |size| bytes, and moves cursor ahead by the number of bytes
// sliced off.
C2ReadCursor slice(uint32_t size) const;
// slices off at most |size| bytes, and moves cursor ahead by the number of bytes
// sliced off.
C2WriteCursor reserve(uint32_t size);
// bool read(T&);
// bool write(T&);
C2Fence waitForSpace(uint32_t size);
};
/// @}
/// \addtogroup graphic
/// @{
struct C2ColorSpace {
//public:
enum Standard {
BT601,
BT709,
BT2020,
// TODO
};
enum Range {
LIMITED,
FULL,
// TODO
};
enum TransferFunction {
BT709Transfer,
BT2020Transfer,
HybridLogGamma2,
HybridLogGamma4,
// TODO
};
};
/** \deprecated */
class C2GraphicBuffer : public C2Buffer {
public:
// constant attributes
inline uint32_t width() const { return mWidth; }
inline uint32_t height() const { return mHeight; }
inline uint32_t format() const { return mFormat; }
inline const C2MemoryUsage usage() const { return mUsage; }
// modifiable attributes
virtual const C2ColorSpace colorSpace() const = 0;
// best effort
virtual void setColorSpace_be(const C2ColorSpace &colorSpace) = 0;
virtual bool setColorSpace(const C2ColorSpace &colorSpace) = 0;
const C2Handle *handle() const;
protected:
uint32_t mWidth;
uint32_t mHeight;
uint32_t mFormat;
C2MemoryUsage mUsage;
class Impl;
Impl *mImpl;
};
/// @}
/// \endcond
/// @}
#endif // C2BUFFER_H_