hardware_buffer.h

/*
 * Copyright 2017 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.
 */

/**
 * @file hardware_buffer.h
 * @brief API for native hardware buffers.
 */
/**
 * @defgroup AHardwareBuffer Native Hardware Buffer
 *
 * AHardwareBuffer objects represent chunks of memory that can be
 * accessed by various hardware components in the system. It can be
 * easily converted to the Java counterpart
 * android.hardware.HardwareBuffer and passed between processes using
 * Binder. All operations involving AHardwareBuffer and HardwareBuffer
 * are zero-copy, i.e., passing AHardwareBuffer to another process
 * creates a shared view of the same region of memory.
 *
 * AHardwareBuffers can be bound to EGL/OpenGL and Vulkan primitives.
 * For EGL, use the extension function eglGetNativeClientBufferANDROID
 * to obtain an EGLClientBuffer and pass it directly to
 * eglCreateImageKHR. Refer to the EGL extensions
 * EGL_ANDROID_get_native_client_buffer and
 * EGL_ANDROID_image_native_buffer for more information. In Vulkan,
 * the contents of the AHardwareBuffer can be accessed as external
 * memory. See the VK_ANDROID_external_memory_android_hardware_buffer
 * extension for details.
 *
 * @{
 */

#ifndef ANDROID_HARDWARE_BUFFER_H
#define ANDROID_HARDWARE_BUFFER_H

#include <inttypes.h>

#include <sys/cdefs.h>

#include <android/rect.h>

__BEGIN_DECLS

/**
 * Buffer pixel formats.
 */
enum AHardwareBuffer_Format {
    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_R8G8B8A8_UNORM
     *   OpenGL ES: GL_RGBA8
     */
    AHARDWAREBUFFER_FORMAT_R8G8B8A8_UNORM           = 1,

    /**
     * 32 bits per pixel, 8 bits per channel format where alpha values are
     * ignored (always opaque).
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_R8G8B8A8_UNORM
     *   OpenGL ES: GL_RGB8
     */
    AHARDWAREBUFFER_FORMAT_R8G8B8X8_UNORM           = 2,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_R8G8B8_UNORM
     *   OpenGL ES: GL_RGB8
     */
    AHARDWAREBUFFER_FORMAT_R8G8B8_UNORM             = 3,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_R5G6B5_UNORM_PACK16
     *   OpenGL ES: GL_RGB565
     */
    AHARDWAREBUFFER_FORMAT_R5G6B5_UNORM             = 4,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_R16G16B16A16_SFLOAT
     *   OpenGL ES: GL_RGBA16F
     */
    AHARDWAREBUFFER_FORMAT_R16G16B16A16_FLOAT       = 0x16,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_A2B10G10R10_UNORM_PACK32
     *   OpenGL ES: GL_RGB10_A2
     */
    AHARDWAREBUFFER_FORMAT_R10G10B10A2_UNORM        = 0x2b,

    /**
     * Opaque binary blob format.
     * Must have height 1 and one layer, with width equal to the buffer
     * size in bytes. Corresponds to Vulkan buffers and OpenGL buffer
     * objects. Can be bound to the latter using GL_EXT_external_buffer.
     */
    AHARDWAREBUFFER_FORMAT_BLOB                     = 0x21,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_D16_UNORM
     *   OpenGL ES: GL_DEPTH_COMPONENT16
     */
    AHARDWAREBUFFER_FORMAT_D16_UNORM                = 0x30,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_X8_D24_UNORM_PACK32
     *   OpenGL ES: GL_DEPTH_COMPONENT24
     */
    AHARDWAREBUFFER_FORMAT_D24_UNORM                = 0x31,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_D24_UNORM_S8_UINT
     *   OpenGL ES: GL_DEPTH24_STENCIL8
     */
    AHARDWAREBUFFER_FORMAT_D24_UNORM_S8_UINT        = 0x32,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_D32_SFLOAT
     *   OpenGL ES: GL_DEPTH_COMPONENT32F
     */
    AHARDWAREBUFFER_FORMAT_D32_FLOAT                = 0x33,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_D32_SFLOAT_S8_UINT
     *   OpenGL ES: GL_DEPTH32F_STENCIL8
     */
    AHARDWAREBUFFER_FORMAT_D32_FLOAT_S8_UINT        = 0x34,

    /**
     * Corresponding formats:
     *   Vulkan: VK_FORMAT_S8_UINT
     *   OpenGL ES: GL_STENCIL_INDEX8
     */
    AHARDWAREBUFFER_FORMAT_S8_UINT                  = 0x35,

    /**
     * YUV 420 888 format.
     * Must have an even width and height. Can be accessed in OpenGL
     * shaders through an external sampler. Does not support mip-maps
     * cube-maps or multi-layered textures.
     */
    AHARDWAREBUFFER_FORMAT_Y8Cb8Cr8_420             = 0x23,
};

/**
 * Buffer usage flags, specifying how the buffer will be accessed.
 */
enum AHardwareBuffer_UsageFlags {
    /// The buffer will never be locked for direct CPU reads using the
    /// AHardwareBuffer_lock() function. Note that reading the buffer
    /// using OpenGL or Vulkan functions or memory mappings is still
    /// allowed.
    AHARDWAREBUFFER_USAGE_CPU_READ_NEVER        = 0UL,
    /// The buffer will sometimes be locked for direct CPU reads using
    /// the AHardwareBuffer_lock() function. Note that reading the
    /// buffer using OpenGL or Vulkan functions or memory mappings
    /// does not require the presence of this flag.
    AHARDWAREBUFFER_USAGE_CPU_READ_RARELY       = 2UL,
    /// The buffer will often be locked for direct CPU reads using
    /// the AHardwareBuffer_lock() function. Note that reading the
    /// buffer using OpenGL or Vulkan functions or memory mappings
    /// does not require the presence of this flag.
    AHARDWAREBUFFER_USAGE_CPU_READ_OFTEN        = 3UL,
    /// CPU read value mask.
    AHARDWAREBUFFER_USAGE_CPU_READ_MASK         = 0xFUL,

    /// The buffer will never be locked for direct CPU writes using the
    /// AHardwareBuffer_lock() function. Note that writing the buffer
    /// using OpenGL or Vulkan functions or memory mappings is still
    /// allowed.
    AHARDWAREBUFFER_USAGE_CPU_WRITE_NEVER       = 0UL << 4,
    /// The buffer will sometimes be locked for direct CPU writes using
    /// the AHardwareBuffer_lock() function. Note that writing the
    /// buffer using OpenGL or Vulkan functions or memory mappings
    /// does not require the presence of this flag.
    AHARDWAREBUFFER_USAGE_CPU_WRITE_RARELY      = 2UL << 4,
    /// The buffer will often be locked for direct CPU writes using
    /// the AHardwareBuffer_lock() function. Note that writing the
    /// buffer using OpenGL or Vulkan functions or memory mappings
    /// does not require the presence of this flag.
    AHARDWAREBUFFER_USAGE_CPU_WRITE_OFTEN       = 3UL << 4,
    /// CPU write value mask.
    AHARDWAREBUFFER_USAGE_CPU_WRITE_MASK        = 0xFUL << 4,

    /// The buffer will be read from by the GPU as a texture.
    AHARDWAREBUFFER_USAGE_GPU_SAMPLED_IMAGE      = 1UL << 8,
    /// The buffer will be written to by the GPU as a framebuffer attachment.
    AHARDWAREBUFFER_USAGE_GPU_FRAMEBUFFER        = 1UL << 9,
    /**
     * The buffer will be written to by the GPU as a framebuffer
     * attachment.
     *
     * Note that the name of this flag is somewhat misleading: it does
     * not imply that the buffer contains a color format. A buffer with
     * depth or stencil format that will be used as a framebuffer
     * attachment should also have this flag. Use the equivalent flag
     * AHARDWAREBUFFER_USAGE_GPU_FRAMEBUFFER to avoid this confusion.
     */
    AHARDWAREBUFFER_USAGE_GPU_COLOR_OUTPUT       = AHARDWAREBUFFER_USAGE_GPU_FRAMEBUFFER,
    /**
     * The buffer will be used as a composer HAL overlay layer.
     *
     * This flag is currently only needed when using ASurfaceTransaction_setBuffer
     * to set a buffer. In all other cases, the framework adds this flag
     * internally to buffers that could be presented in a composer overlay.
     * ASurfaceTransaction_setBuffer is special because it uses buffers allocated
     * directly through AHardwareBuffer_allocate instead of buffers allocated
     * by the framework.
     */
    AHARDWAREBUFFER_USAGE_COMPOSER_OVERLAY       = 1ULL << 11,
    /**
     * The buffer is protected from direct CPU access or being read by
     * non-secure hardware, such as video encoders.
     *
     * This flag is incompatible with CPU read and write flags. It is
     * mainly used when handling DRM video. Refer to the EGL extension
     * EGL_EXT_protected_content and GL extension
     * GL_EXT_protected_textures for more information on how these
     * buffers are expected to behave.
     */
    AHARDWAREBUFFER_USAGE_PROTECTED_CONTENT      = 1UL << 14,
    /// The buffer will be read by a hardware video encoder.
    AHARDWAREBUFFER_USAGE_VIDEO_ENCODE           = 1UL << 16,
    /**
     * The buffer will be used for direct writes from sensors.
     * When this flag is present, the format must be AHARDWAREBUFFER_FORMAT_BLOB.
     */
    AHARDWAREBUFFER_USAGE_SENSOR_DIRECT_DATA     = 1UL << 23,
    /**
     * The buffer will be used as a shader storage or uniform buffer object.
     * When this flag is present, the format must be AHARDWAREBUFFER_FORMAT_BLOB.
     */
    AHARDWAREBUFFER_USAGE_GPU_DATA_BUFFER        = 1UL << 24,
    /**
     * The buffer will be used as a cube map texture.
     * When this flag is present, the buffer must have a layer count
     * that is a multiple of 6. Note that buffers with this flag must be
     * bound to OpenGL textures using the extension
     * GL_EXT_EGL_image_storage instead of GL_KHR_EGL_image.
     */
    AHARDWAREBUFFER_USAGE_GPU_CUBE_MAP               = 1UL << 25,
    /**
     * The buffer contains a complete mipmap hierarchy.
     * Note that buffers with this flag must be bound to OpenGL textures using
     * the extension GL_EXT_EGL_image_storage instead of GL_KHR_EGL_image.
     */
    AHARDWAREBUFFER_USAGE_GPU_MIPMAP_COMPLETE        = 1UL << 26,

    AHARDWAREBUFFER_USAGE_VENDOR_0  = 1ULL << 28,
    AHARDWAREBUFFER_USAGE_VENDOR_1  = 1ULL << 29,
    AHARDWAREBUFFER_USAGE_VENDOR_2  = 1ULL << 30,
    AHARDWAREBUFFER_USAGE_VENDOR_3  = 1ULL << 31,
    AHARDWAREBUFFER_USAGE_VENDOR_4  = 1ULL << 48,
    AHARDWAREBUFFER_USAGE_VENDOR_5  = 1ULL << 49,
    AHARDWAREBUFFER_USAGE_VENDOR_6  = 1ULL << 50,
    AHARDWAREBUFFER_USAGE_VENDOR_7  = 1ULL << 51,
    AHARDWAREBUFFER_USAGE_VENDOR_8  = 1ULL << 52,
    AHARDWAREBUFFER_USAGE_VENDOR_9  = 1ULL << 53,
    AHARDWAREBUFFER_USAGE_VENDOR_10 = 1ULL << 54,
    AHARDWAREBUFFER_USAGE_VENDOR_11 = 1ULL << 55,
    AHARDWAREBUFFER_USAGE_VENDOR_12 = 1ULL << 56,
    AHARDWAREBUFFER_USAGE_VENDOR_13 = 1ULL << 57,
    AHARDWAREBUFFER_USAGE_VENDOR_14 = 1ULL << 58,
    AHARDWAREBUFFER_USAGE_VENDOR_15 = 1ULL << 59,
    AHARDWAREBUFFER_USAGE_VENDOR_16 = 1ULL << 60,
    AHARDWAREBUFFER_USAGE_VENDOR_17 = 1ULL << 61,
    AHARDWAREBUFFER_USAGE_VENDOR_18 = 1ULL << 62,
    AHARDWAREBUFFER_USAGE_VENDOR_19 = 1ULL << 63,
};

/**
 * Buffer description. Used for allocating new buffers and querying
 * parameters of existing ones.
 */
typedef struct AHardwareBuffer_Desc {
    uint32_t    width;      ///< Width in pixels.
    uint32_t    height;     ///< Height in pixels.
    /**
     * Number of images in an image array. AHardwareBuffers with one
     * layer correspond to regular 2D textures. AHardwareBuffers with
     * more than layer correspond to texture arrays. If the layer count
     * is a multiple of 6 and the usage flag
     * AHARDWAREBUFFER_USAGE_GPU_CUBE_MAP is present, the buffer is
     * a cube map or a cube map array.
     */
    uint32_t    layers;
    uint32_t    format;     ///< One of AHardwareBuffer_Format.
    uint64_t    usage;      ///< Combination of AHardwareBuffer_UsageFlags.
    uint32_t    stride;     ///< Row stride in pixels, ignored for AHardwareBuffer_allocate()
    uint32_t    rfu0;       ///< Initialize to zero, reserved for future use.
    uint64_t    rfu1;       ///< Initialize to zero, reserved for future use.
} AHardwareBuffer_Desc;

/**
 * Holds data for a single image plane.
 */
typedef struct AHardwareBuffer_Plane {
    void*       data;        ///< Points to first byte in plane
    uint32_t    pixelStride; ///< Distance in bytes from the color channel of one pixel to the next
    uint32_t    rowStride;   ///< Distance in bytes from the first value of one row of the image to
                             ///  the first value of the next row.
} AHardwareBuffer_Plane;

/**
 * Holds all image planes that contain the pixel data.
 */
typedef struct AHardwareBuffer_Planes {
    uint32_t               planeCount; ///< Number of distinct planes
    AHardwareBuffer_Plane  planes[4];     ///< Array of image planes
} AHardwareBuffer_Planes;

/**
 * Opaque handle for a native hardware buffer.
 */
typedef struct AHardwareBuffer AHardwareBuffer;

#if __ANDROID_API__ >= 26

/**
 * Allocates a buffer that matches the passed AHardwareBuffer_Desc.
 *
 * If allocation succeeds, the buffer can be used according to the
 * usage flags specified in its description. If a buffer is used in ways
 * not compatible with its usage flags, the results are undefined and
 * may include program termination.
 *
 * Available since API level 26.
 *
 * \return 0 on success, or an error number of the allocation fails for
 * any reason. The returned buffer has a reference count of 1.
 */
int AHardwareBuffer_allocate(const AHardwareBuffer_Desc* desc,
        AHardwareBuffer** outBuffer) __INTRODUCED_IN(26);
/**
 * Acquire a reference on the given AHardwareBuffer object.
 *
 * This prevents the object from being deleted until the last reference
 * is removed.
 *
 * Available since API level 26.
 */
void AHardwareBuffer_acquire(AHardwareBuffer* buffer) __INTRODUCED_IN(26);

/**
 * Remove a reference that was previously acquired with
 * AHardwareBuffer_acquire() or AHardwareBuffer_allocate().
 *
 * Available since API level 26.
 */
void AHardwareBuffer_release(AHardwareBuffer* buffer) __INTRODUCED_IN(26);

/**
 * Return a description of the AHardwareBuffer in the passed
 * AHardwareBuffer_Desc struct.
 *
 * Available since API level 26.
 */
void AHardwareBuffer_describe(const AHardwareBuffer* buffer,
        AHardwareBuffer_Desc* outDesc) __INTRODUCED_IN(26);

/**
 * Lock the AHardwareBuffer for direct CPU access.
 *
 * This function can lock the buffer for either reading or writing.
 * It may block if the hardware needs to finish rendering, if CPU caches
 * need to be synchronized, or possibly for other implementation-
 * specific reasons.
 *
 * The passed AHardwareBuffer must have one layer, otherwise the call
 * will fail.
 *
 * If \a fence is not negative, it specifies a fence file descriptor on
 * which to wait before locking the buffer. If it's negative, the caller
 * is responsible for ensuring that writes to the buffer have completed
 * before calling this function.  Using this parameter is more efficient
 * than waiting on the fence and then calling this function.
 *
 * The \a usage parameter may only specify AHARDWAREBUFFER_USAGE_CPU_*.
 * If set, then outVirtualAddress is filled with the address of the
 * buffer in virtual memory. The flags must also be compatible with
 * usage flags specified at buffer creation: if a read flag is passed,
 * the buffer must have been created with
 * AHARDWAREBUFFER_USAGE_CPU_READ_RARELY or
 * AHARDWAREBUFFER_USAGE_CPU_READ_OFTEN. If a write flag is passed, it
 * must have been created with AHARDWAREBUFFER_USAGE_CPU_WRITE_RARELY or
 * AHARDWAREBUFFER_USAGE_CPU_WRITE_OFTEN.
 *
 * If \a rect is not NULL, the caller promises to modify only data in
 * the area specified by rect. If rect is NULL, the caller may modify
 * the contents of the entire buffer. The content of the buffer outside
 * of the specified rect is NOT modified by this call.
 *
 * It is legal for several different threads to lock a buffer for read
 * access; none of the threads are blocked.
 *
 * Locking a buffer simultaneously for write or read/write is undefined,
 * but will neither terminate the process nor block the caller.
 * AHardwareBuffer_lock may return an error or leave the buffer's
 * content in an indeterminate state.
 *
 * If the buffer has AHARDWAREBUFFER_FORMAT_BLOB, it is legal lock it
 * for reading and writing in multiple threads and/or processes
 * simultaneously, and the contents of the buffer behave like shared
 * memory.
 *
 * Available since API level 26.
 *
 * \return 0 on success. -EINVAL if \a buffer is NULL, the usage flags
 * are not a combination of AHARDWAREBUFFER_USAGE_CPU_*, or the buffer
 * has more than one layer. Error number if the lock fails for any other
 * reason.
 */
int AHardwareBuffer_lock(AHardwareBuffer* buffer, uint64_t usage,
        int32_t fence, const ARect* rect, void** outVirtualAddress) __INTRODUCED_IN(26);

/**
 * Lock a potentially multi-planar AHardwareBuffer for direct CPU access.
 *
 * This function is similar to AHardwareBuffer_lock, but can lock multi-planar
 * formats. The locked planes are returned in the \a outPlanes argument. Note,
 * that multi-planar should not be confused with multi-layer images, which this
 * locking function does not support.
 *
 * YUV formats are always represented by three separate planes of data, one for
 * each color plane. The order of planes in the array is guaranteed such that
 * plane #0 is always Y, plane #1 is always U (Cb), and plane #2 is always V
 * (Cr). All other formats are represented by a single plane.
 *
 * Additional information always accompanies the buffers, describing the row
 * stride and the pixel stride for each plane.
 *
 * In case the buffer cannot be locked, \a outPlanes will contain zero planes.
 *
 * See the AHardwareBuffer_lock documentation for all other locking semantics.
 *
 * Available since API level 29.
 *
 * \return 0 on success. -EINVAL if \a buffer is NULL, the usage flags
 * are not a combination of AHARDWAREBUFFER_USAGE_CPU_*, or the buffer
 * has more than one layer. Error number if the lock fails for any other
 * reason.
 */
int AHardwareBuffer_lockPlanes(AHardwareBuffer* buffer, uint64_t usage,
        int32_t fence, const ARect* rect, AHardwareBuffer_Planes* outPlanes) __INTRODUCED_IN(29);

/**
 * Unlock the AHardwareBuffer from direct CPU access.
 *
 * Must be called after all changes to the buffer are completed by the
 * caller.  If \a fence is NULL, the function will block until all work
 * is completed.  Otherwise, \a fence will be set either to a valid file
 * descriptor or to -1.  The file descriptor will become signaled once
 * the unlocking is complete and buffer contents are updated.
 * The caller is responsible for closing the file descriptor once it's
 * no longer needed.  The value -1 indicates that unlocking has already
 * completed before the function returned and no further operations are
 * necessary.
 *
 * Available since API level 26.
 *
 * \return 0 on success. -EINVAL if \a buffer is NULL. Error number if
 * the unlock fails for any reason.
 */
int AHardwareBuffer_unlock(AHardwareBuffer* buffer, int32_t* fence) __INTRODUCED_IN(26);

/**
 * Send the AHardwareBuffer to an AF_UNIX socket.
 *
 * Available since API level 26.
 *
 * \return 0 on success, -EINVAL if \a buffer is NULL, or an error
 * number if the operation fails for any reason.
 */
int AHardwareBuffer_sendHandleToUnixSocket(const AHardwareBuffer* buffer, int socketFd) __INTRODUCED_IN(26);

/**
 * Receive an AHardwareBuffer from an AF_UNIX socket.
 *
 * Available since API level 26.
 *
 * \return 0 on success, -EINVAL if \a outBuffer is NULL, or an error
 * number if the operation fails for any reason.
 */
int AHardwareBuffer_recvHandleFromUnixSocket(int socketFd, AHardwareBuffer** outBuffer) __INTRODUCED_IN(26);

#endif // __ANDROID_API__ >= 26

#if __ANDROID_API__ >= 29

/**
 * Test whether the given format and usage flag combination is
 * allocatable.
 *
 * If this function returns true, it means that a buffer with the given
 * description can be allocated on this implementation, unless resource
 * exhaustion occurs. If this function returns false, it means that the
 * allocation of the given description will never succeed.
 *
 * The return value of this function may depend on all fields in the
 * description, except stride, which is always ignored. For example,
 * some implementations have implementation-defined limits on texture
 * size and layer count.
 *
 * Available since API level 29.
 *
 * \return 1 if the format and usage flag combination is allocatable,
 *     0 otherwise.
 */
int AHardwareBuffer_isSupported(const AHardwareBuffer_Desc* desc) __INTRODUCED_IN(29);

/**
 * Lock an AHardwareBuffer for direct CPU access.
 *
 * This function is the same as the above lock function, but passes back
 * additional information about the bytes per pixel and the bytes per stride
 * of the locked buffer.  If the bytes per pixel or bytes per stride are unknown
 * or variable, or if the underlying mapper implementation does not support returning
 * additional information, then this call will fail with INVALID_OPERATION
 *
 * Available since API level 29.
 */
int AHardwareBuffer_lockAndGetInfo(AHardwareBuffer* buffer, uint64_t usage,
        int32_t fence, const ARect* rect, void** outVirtualAddress,
        int32_t* outBytesPerPixel, int32_t* outBytesPerStride) __INTRODUCED_IN(29);
#endif // __ANDROID_API__ >= 29

__END_DECLS

#endif // ANDROID_HARDWARE_BUFFER_H

/** @} */