Files
UnrealEngine/Engine/Source/ThirdParty/NVIDIA/GPUDirect/include/DVPAPI.h
2025-05-18 13:04:45 +08:00

972 lines
42 KiB
C

/*
* Copyright 1993-2018 NVIDIA Corporation. All rights reserved.
*
* NOTICE TO LICENSEE:
*
* This source code and/or documentation ("Licensed Deliverables") are
* subject to NVIDIA intellectual property rights under U.S. and
* international Copyright laws.
*
* These Licensed Deliverables contained herein is PROPRIETARY and
* CONFIDENTIAL to NVIDIA and is being provided under the terms and
* conditions of a form of NVIDIA software license agreement by and
* between NVIDIA and Licensee ("License Agreement") or electronically
* accepted by Licensee. Notwithstanding any terms or conditions to
* the contrary in the License Agreement, reproduction or disclosure
* of the Licensed Deliverables to any third party without the express
* written consent of NVIDIA is prohibited.
*
* NOTWITHSTANDING ANY TERMS OR CONDITIONS TO THE CONTRARY IN THE
* LICENSE AGREEMENT, NVIDIA MAKES NO REPRESENTATION ABOUT THE
* SUITABILITY OF THESE LICENSED DELIVERABLES FOR ANY PURPOSE. IT IS
* PROVIDED "AS IS" WITHOUT EXPRESS OR IMPLIED WARRANTY OF ANY KIND.
* NVIDIA DISCLAIMS ALL WARRANTIES WITH REGARD TO THESE LICENSED
* DELIVERABLES, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY,
* NONINFRINGEMENT, AND FITNESS FOR A PARTICULAR PURPOSE.
* NOTWITHSTANDING ANY TERMS OR CONDITIONS TO THE CONTRARY IN THE
* LICENSE AGREEMENT, IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY
* SPECIAL, INDIRECT, INCIDENTAL, OR CONSEQUENTIAL DAMAGES, OR ANY
* DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
* WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
* ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
* OF THESE LICENSED DELIVERABLES.
*
* U.S. Government End Users. These Licensed Deliverables are a
* "commercial item" as that term is defined at 48 C.F.R. 2.101 (OCT
* 1995), consisting of "commercial computer software" and "commercial
* computer software documentation" as such terms are used in 48
* C.F.R. 12.212 (SEPT 1995) and is provided to the U.S. Government
* only as a commercial end item. Consistent with 48 C.F.R.12.212 and
* 48 C.F.R. 227.7202-1 through 227.7202-4 (JUNE 1995), all
* U.S. Government End Users acquire the Licensed Deliverables with
* only those rights set forth herein.
*
* Any use of the Licensed Deliverables in individual and commercial
* software must include, in the user documentation and internal
* comments to the code, the above Disclaimer and U.S. Government End
* Users Notice.
*/
#ifndef __DVAPI_H_
#define __DVAPI_H_
#include <stdint.h>
//------------------------------------------------------------------------
//
// Introduction
//
// The Digital Video Pipeline (DVP) library is a set of functions that
// provide fast path access into graphics/compute APIs. This is the basis
// library for the https://developer.nvidia.com/gpudirectforvideo, which
// is used by HW partners in broadcast to pass video streams (and other
// data) in a high bandwidth / low latency fashion into and out of the GPU.
//
//------------------------------------------------------------------------
//------------------------------------------------------------------------
//
// Errata
//
// A list of corrections to the API header:
//
// 1.) dvpMemcpy2D incorrectly described the parameters units as
// bytes. This has been removed, the parameters units are
// absolute pixel positions in the surface. Fixed in version 1.76.
//
//------------------------------------------------------------------------
#if defined(WIN32)
#define DVPAPI_INTERFACE __declspec( dllexport ) DVPStatus
#if defined(DVP_STATIC)
// If using a static build, force the linker to include the
// DVP CRT call back, so the DVP can properly cleanup thread
// data. For dynamic library linkage, this is handled by Dllmain.
// Dynamic linkage is recommended.
#ifdef _WIN64
#pragma comment(linker, "/INCLUDE:_tls_used")
#pragma comment(linker, "/INCLUDE:dvp_callback")
#else // _WIN64
#pragma comment(linker, "/INCLUDE:__tls_used")
#pragma comment(linker, "/INCLUDE:_dvp_callback")
#endif // _WIN64
#endif
#else
#define DVPAPI_INTERFACE extern DVPStatus
#endif
#define DVP_MAJOR_VERSION 2
#define DVP_MINOR_VERSION 15
typedef uint64_t DVPBufferHandle;
typedef uint64_t DVPSyncObjectHandle;
typedef enum
{
DVP_STATUS_OK = 0,
DVP_STATUS_INVALID_PARAMETER = 1,
DVP_STATUS_UNSUPPORTED = 2,
DVP_STATUS_END_ENUMERATION = 3,
DVP_STATUS_INVALID_DEVICE = 4,
DVP_STATUS_OUT_OF_MEMORY = 5,
DVP_STATUS_INVALID_OPERATION = 6,
DVP_STATUS_TIMEOUT = 7,
DVP_STATUS_INVALID_CONTEXT = 8,
DVP_STATUS_INVALID_RESOURCE_TYPE = 9,
DVP_STATUS_INVALID_FORMAT_OR_TYPE = 10,
DVP_STATUS_DEVICE_UNINITIALIZED = 11,
DVP_STATUS_UNSIGNALED = 12,
DVP_STATUS_SYNC_ERROR = 13,
DVP_STATUS_SYNC_STILL_BOUND = 14,
DVP_STATUS_ERROR = -1
} DVPStatus;
// Pixel component formats stored in the system memory buffer
// analogous to those defined in the OpenGL API, except for
// DVP_BUFFER and the DVP_CUDA_* types. DVP_BUFFER provides
// an unspecified format type to allow for general interpretation
// of the bytes at a later stage (in GPU shader). Note that not
// all paths will achieve optimal speeds due to lack of HW support
// for the transformation. The CUDA types are to be used when
// copying to/from a system memory buffer from-to a CUDA array, as the
// CUDA array implies a memory layout that matches the array.
typedef enum
{
DVP_BUFFER, // Buffer treated as a raw buffer
// and copied directly into GPU buffer
// without any interpretation of the
// stored bytes.
DVP_DEPTH_COMPONENT,
DVP_RGBA,
DVP_BGRA,
DVP_RED,
DVP_GREEN,
DVP_BLUE,
DVP_ALPHA,
DVP_RGB,
DVP_BGR,
DVP_RG,
DVP_LUMINANCE,
DVP_LUMINANCE_ALPHA,
DVP_CUDA_1_CHANNEL,
DVP_CUDA_2_CHANNELS,
DVP_CUDA_4_CHANNELS,
DVP_RGBA_INTEGER,
DVP_BGRA_INTEGER,
DVP_RED_INTEGER,
DVP_GREEN_INTEGER,
DVP_BLUE_INTEGER,
DVP_ALPHA_INTEGER,
DVP_RGB_INTEGER,
DVP_BGR_INTEGER,
DVP_RG_INTEGER,
DVP_LUMINANCE_INTEGER,
DVP_LUMINANCE_ALPHA_INTEGER
} DVPBufferFormats;
// Possible pixel component storage types for system memory buffers
typedef enum
{
DVP_UNSIGNED_BYTE,
DVP_BYTE,
DVP_UNSIGNED_SHORT,
DVP_SHORT,
DVP_UNSIGNED_INT,
DVP_INT,
DVP_FLOAT,
DVP_HALF_FLOAT,
DVP_UNSIGNED_BYTE_3_3_2,
DVP_UNSIGNED_BYTE_2_3_3_REV,
DVP_UNSIGNED_SHORT_5_6_5,
DVP_UNSIGNED_SHORT_5_6_5_REV,
DVP_UNSIGNED_SHORT_4_4_4_4,
DVP_UNSIGNED_SHORT_4_4_4_4_REV,
DVP_UNSIGNED_SHORT_5_5_5_1,
DVP_UNSIGNED_SHORT_1_5_5_5_REV,
DVP_UNSIGNED_INT_8_8_8_8,
DVP_UNSIGNED_INT_8_8_8_8_REV,
DVP_UNSIGNED_INT_10_10_10_2,
DVP_UNSIGNED_INT_2_10_10_10_REV
} DVPBufferTypes;
// Possible external memory handle types
typedef enum
{
DVP_OPAQUE_FD,
DVP_OPAQUE_WIN32,
DVP_OPAQUE_WIN32_KMT
} DVPExternalHandleTypes;
// System memory descriptor describing the size and storage formats
// of the buffer
typedef struct DVPSysmemBufferDescRec {
uint32_t width; // Buffer Width
uint32_t height; // Buffer Height
uint32_t stride; // Stride
uint32_t size; // Specifies the surface size if
// format == DVP_BUFFER
DVPBufferFormats format; // see enum above
DVPBufferTypes type; // see enum above
void *bufAddr; // Buffer memory address
} DVPSysmemBufferDesc;
// GPU external resource descriptor describing the size and storage formats
// of the external resource
typedef struct DVPGpuExternaResourceDescRec {
uint32_t width; // Image Width
uint32_t height; // Image Height
uint32_t size; // Specifies the memory size if
// format == DVP_BUFFER
DVPBufferFormats format; // see enum above
DVPBufferTypes type; // see enum above
DVPExternalHandleTypes handleType; // see enum above
union {
void* handle;
int fd;
} external;
} DVPGpuExternalResourceDesc;
// Flags specified at sync object creation:
// ----------------------------------------
// Tells the implementation to use events wherever
// possible instead of software spin loops. Note if HW
// wait operations are supported by the implementation
// then events will not be used in the dvpMemcpy*
// functions. In such a case, events may still be used
// in dvpSyncObjClientWait* functions.
#define DVP_SYNC_OBJECT_FLAGS_USE_EVENTS 0x00000001
typedef struct DVPSyncObjectDescRec {
uint32_t *sem; // Location to write semaphore value
uint32_t flags; // See above DVP_SYNC_OBJECT_FLAGS_* bits
DVPStatus (*externalClientWaitFunc) (DVPSyncObjectHandle sync,
uint32_t value,
bool GEQ, // If true then the function should wait for the sync value to be
// greater than or equal to the value parameter. Otherwise just a
// straight forward equality comparison should be performed.
uint64_t timeout);
// If non-null, externalClientWaitFunc allows the DVP library
// to call the application to wait for a sync object to be
// released. This allows the application to create events,
// which can be triggered on device interrupts instead of
// using spin loops inside the DVP library. Upon succeeding
// the function must return DVP_STATUS_OK, non-zero for failure
// and DVP_STATUS_TIMEOUT on timeout. The externalClientWaitFunc should
// not alter the current GL or CUDA context state
} DVPSyncObjectDesc;
// Time used when event timeouts should be ignored
#define DVP_TIMEOUT_IGNORED 0xFFFFFFFFFFFFFFFFull
// Flags supplied to the dvpInit* functions:
//
// DVP_DEVICE_FLAGS_SHARE_APP_CONTEXT is only supported for OpenGL
// contexts and is the only supported flag for CUDA. It allows for
// certain cases to be optimized by sharing the context
// of the application for the DVP operations. This removes the
// need to do certain synchronizations. See issue 5 for parallel
// issues. When used, the app's GL context must be current for all calls
// to the DVP library.
// the DVP library.
#define DVP_DEVICE_FLAGS_SHARE_APP_CONTEXT 0x000000001
//------------------------------------------------------------------------
// Function: dvpInit*
//
// To be called before any DVP resources are allocated.
// This call allows for specification of flags that may
// change the way DVP operations are performed. See above
// for the list of flags.
//
// If OpenGL or CUDA is used, the OpenGL/CUDA context
// must be current at time of call.
//
// Parameters: flags[IN] - Buffer description structure
// pDeviceUUID[IN] - Device UUID (Vulkan only)
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
//------------------------------------------------------------------------
// Function: dvpClose*
//
// Function to be called when app closes to allow freeing
// of any DVP library allocated resources.
//
// If OpenGL or CUDA is used, the OpenGL/CUDA context
// must be current at time of call.
//
// Parameters: none
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
//------------------------------------------------------------------------
// Function: dvpBegin
//
// Description: dvpBegin must be called before any combination of DVP
// function calls dvpMemCpy*, dvpMapBufferWaitDVP,
// dvpSyncObjClientWait*, and dvpMapBufferEndDVP. After
// the last of these functions has been called is dvpEnd
// must be called. This allows for more efficient batched
// DVP operations.
//
// Parameters: none
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpBegin();
//------------------------------------------------------------------------
// Function: dvpEnd
//
// Description: dvpEnd signals the end of a batch of DVP function calls
// that began with dvpBegin
//
// Parameters: none
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpEnd();
//------------------------------------------------------------------------
// Function: dvpCreateBuffer
//
// Description: Create a DVP buffer using system memory, wrapping a user
// passed pointer. The pointer must be aligned
// to values returned by dvpGetRequiredAlignments*
//
// Parameters: desc[IN] - Buffer description structure
// hBuf[OUT] - DVP Buffer handle
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpCreateBuffer(DVPSysmemBufferDesc *desc,
DVPBufferHandle *hBuf);
//------------------------------------------------------------------------
// Function: dvpDestroyBuffer
//
// Description: Destroy a previously created DVP buffer.
//
// Parameters: hBuf[IN] - DVP Buffer handle
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpDestroyBuffer(DVPBufferHandle hBuf);
//------------------------------------------------------------------------
// Function: dvpBindTo*
//
// Description: Bind a previously created system memory buffer to an
// OpenGL context, CUDA context or D3D[9,10,11] device.
// For OpenGL and CUDA the context must be current at
// the time of bind.
//
// Parameters: hBuf[IN] - DVP Buffer handle
// device[IN] - D3D device if used
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
//------------------------------------------------------------------------
// Function: dvpUnbindFrom*
//
// Description: Unbind a previously bound buffer from an API context.
// For OpenGL and CUDA the context must be current at
// the time of bind.
//
// Parameters: hBuf[IN] - DVP Buffer handle
// device[IN] - D3D device if used
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
//------------------------------------------------------------------------
// Function: dvpCreateGPU*
//
// Description: dvpCreateGPU* creates a DVP buffer reference from a named
// [buffer,texture,resource]. If OpenGL or CUDA is used, the
// OpenGL/CUDA context must be current at time of call.
//
// For D3D resources, as soon as dvpCreateGPUD3D* is called
// the resource is mapped for use by the DVP library.
// dvpMapBufferWaitAPI must be called before the resource
// is used.
//
// Parameters: bufferID[IN] - [buffer,texture,resource] to import
// bufferHandle[OUT] - DVP Buffer handle
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
//------------------------------------------------------------------------
// Function: dvpCreateGPUExternalResource*
//
// Description: dvpCreateGPUExternalResource creates a DVP buffer
// reference from a an external [buffer,image,resource].
//
// dvpMapBufferWaitAPI must be called before the resource
// is used.
//
// Parameters: device[IN] - Vk/D3D device
// desc[IN] - External resource description structure
// bufferHandle[OUT] - DVP Buffer handle
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
//------------------------------------------------------------------------
// Function: dvpFreeBuffer
//
// Description: dvpFreeBuffer frees the DVP buffer reference
//
// Parameters: gpuBufferHandle[IN] - DVP Buffer handle
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpFreeBuffer(DVPBufferHandle gpuBufferHandle);
//------------------------------------------------------------------------
// Function: dvpGetRequiredConstants*
//
// Description: dvpGetRequiredConstants* returns, for the associated
// API context or device:
// - Alignments required for the start address of the
// buffer.
// - An alignment recommendation for the stride (pitch)
// of the buffer, as the performance might vary based
// on the stride alignment. This is a conservative value
// so smaller values may still give optimal performance.
// Please refer to the SDK for more details.
// - Semaphore Addr alignment
// - Semaphore allocation size
// - Semaphore payload offset
// - Semaphore payload size in bytes
//
// If OpenGL or CUDA is used, the OpenGL/CUDA context
// must be current at time of call.
//
// On Windows Vista and later dvpGetRequiredConstants*
// may return DVP_STATUS_OUT_OF_MEMORY when the system
// limit for graphics contexts has been reached. To avoid
// this, spurious graphics/compute contexts should be avoided.
//
// Parameters: bufferAddrAlignment[OUT]
// bufferGPUStrideAlignment[OUT]
// semaphoreAddrAlignment[OUT]
// semaphoreAllocSize[OUT]
// semaphorePayloadOffset[OUT]
// semaphorePayloadSize[OUT]
// device[IN] - D3D device if used
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
// DVP_STATUS_OUT_OF_MEMORY
//------------------------------------------------------------------------
//------------------------------------------------------------------------
// Function: dvpMemcpy* general comments that apply to all the the
// memcpy functions.
//
// dvpMemcpy* acquires the srcSync and then initiates
// a DMA transfer of the given size from the
// srcBuffer to the dstBuffer at a given position. For
// dvpMemcpy* functions that do not have a source AND
// destination value, the copy is peformed at the source
// starting position (as the function parameter specifies)
// for both the srcBuffer and dstBuffer. For functions where
// both source and destination starting parameters are
// present, the copy is performed at different starting
// positions.
//
// Upon completion of a transfer, a release operation is
// performed on dstSync. The buffer is only usable once the
// release operation is complete.
//
// For OpenGL and CUDA sources or destinations the correct
// context must be current.
//
// When possible the implementation will use HW based
// synchronization and in these cases the timeout parameter
// is not supported, as the call will return immediately and
// the HW will perform the wait. When SW synchronization is
// in use, the wait will timeout after the specified time.
// The actual implementation time resolution may differ from
// the nanoseconds resolution of the parameter. To ignore
// the time out use the value DVP_TIMEOUT_IGNORED.
//
// If the timeout is hit, the corresponding copy is not
// performed.
//------------------------------------------------------------------------
//------------------------------------------------------------------------
// Function: dvpMemcpyLined
//
// Description: dvpMemcpyLined provides buffer copies between a
// DVP sysmem buffer and a graphics API texture (as opposed to
// a buffer type). Other buffer types (such
// as graphics API buffers) return DVP_STATUS_INVALID_PARAMETER.
//
// In addition, see "dvpMemcpy* general comments" above.
//
// Parameters: srcBuffer[IN] - src buffer handle
// srcSync[IN] - sync to acquire on before transfer
// srcAcquireValue[IN] - value to acquire on before transfer
// timeout[IN] - time out value in nanoseconds.
// dstBuffer[IN] - src buffer handle
// dstSync[IN] - sync to release on transfer completion
// dstReleaseValue[IN] - value to release on completion
// Either:
// startingLine[IN] - starting line of buffer
// or:
// srcStartingLine[IN] - starting line of src buffer
// dstStartingLine[IN] - starting line of dst buffer
// numberOfLines[IN] - number of lines to copy
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//
// GL state effected: The following GL state may be altered by this
// function (not relevant if no GL source or destination
// is used):
// -GL_PACK_SKIP_ROWS, GL_PACK_SKIP_PIXELS,
// GL_PACK_ROW_LENGTH
// -The buffer bound to GL_PIXEL_PACK_BUFFER
// -The current bound framebuffer (GL_FRAMEBUFFER_EXT)
// -GL_UNPACK_SKIP_ROWS, GL_UNPACK_SKIP_PIXELS,
// GL_UNPACK_ROW_LENGTH
// -The buffer bound to GL_PIXEL_UNPACK_BUFFER
// -The texture bound to GL_TEXTURE_2D
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpMemcpyLined(DVPBufferHandle srcBuffer,
DVPSyncObjectHandle srcSync,
uint32_t srcAcquireValue,
uint64_t timeout,
DVPBufferHandle dstBuffer,
DVPSyncObjectHandle dstSync,
uint32_t dstReleaseValue,
uint32_t startingLine,
uint32_t numberOfLines);
DVPAPI_INTERFACE
dvpMemcpyLined(DVPBufferHandle srcBuffer,
DVPSyncObjectHandle srcSync,
uint32_t srcAcquireValue,
uint64_t timeout,
DVPBufferHandle dstBuffer,
DVPSyncObjectHandle dstSync,
uint32_t dstReleaseValue,
uint32_t srcStartingLine,
uint32_t dstStartingLine,
uint32_t numberOfLines);
//------------------------------------------------------------------------
// Function: dvpMemcpy
//
// Description: dvpMemcpy provides buffer copies between a
// DVP sysmem buffer and a graphics API pure buffer (as
// opposed to a texture type). Other buffer types (such
// as graphics API textures) return
// DVP_STATUS_INVALID_PARAMETER.
//
// The start address of the srcBuffer is given by srcOffset
// and the dstBuffer start address is given by dstOffset.
//
// In addition, see "dvpMemcpy* general comments" above.
//
// Parameters: srcBuffer[IN] - src buffer handle
// srcSync[IN] - sync to acquire on before transfer
// srcAcquireValue[IN] - value to acquire on before transfer
// timeout[IN] - time out value in nanoseconds.
// dstBuffer[IN] - src buffer handle
// dstSync[IN] - sync to release on completion
// dstReleaseValue[IN] - value to release on completion
// srcOffset[IN] - byte offset of srcBuffer
// dstOffset[IN] - byte offset of dstBuffer
// count[IN] - number of bytes to copy
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//
// GL state effected: The following GL state may be altered by this
// function (not relevant if no GL source or destination
// is used):
// - The buffer bound to GL_COPY_WRITE_BUFFER
// - The buffer bound to GL_COPY_READ_BUFFER
//
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpMemcpy(DVPBufferHandle srcBuffer,
DVPSyncObjectHandle srcSync,
uint32_t srcAcquireValue,
uint64_t timeout,
DVPBufferHandle dstBuffer,
DVPSyncObjectHandle dstSync,
uint32_t dstReleaseValue,
uint32_t srcOffset,
uint32_t dstOffset,
uint32_t count);
//------------------------------------------------------------------------
// Function: dvpMemcpy2D
//
// Description: dvpMemcpy2D provides buffer copies between a
// DVP sysmem buffer and a graphics API texture (as opposed
// to a buffer type). Other buffer types (such
// as graphics API buffers) return
// DVP_STATUS_INVALID_PARAMETER.
//
// The start address of the copy is calculated as
// startPtr = baseAddress + pitch*startY + startX*elementSize,
// where the baseAdress and pitch are taken from the src
// and dst buffer properties.
//
// In addition, see "dvpMemcpy* general comments" above.
//
// Parameters: srcBuffer[IN] - src buffer handle
// srcSync[IN] - sync to acquire before transfer
// srcAcquireValue[IN] - value to acquire before transfer
// timeout[IN] - time out value in nanoseconds.
// dstBuffer[IN] - src buffer handle
// dstSync[IN] - sync to release on completion
// dstReleaseValue[IN] - value to release on completion
// Either:
// startY[IN] - Y position of src rectangle
// startX[IN] - starting X position
// Or:
// srcY[IN] - starting Y position of src rectangle
// srcX[IN] - X position of src rectangle
// dstY[IN] - starting Y position in dst rectangle
// dstX[IN] - X position of dst rectangle
// height[IN] - height of sub rectangle
// width[IN] - width of sub rectangle
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//
// GL state effected: The following GL state may be altered by this
// function (not relevant if no GL source or destination
// is used):
// -GL_PACK_SKIP_ROWS, GL_PACK_SKIP_PIXELS,
// GL_PACK_ROW_LENGTH
// -The buffer bound to GL_PIXEL_PACK_BUFFER
// -The current bound framebuffer (GL_FRAMEBUFFER_EXT)
// -GL_UNPACK_SKIP_ROWS, GL_UNPACK_SKIP_PIXELS,
// GL_UNPACK_ROW_LENGTH
// -The buffer bound to GL_PIXEL_UNPACK_BUFFER
// -The texture bound to GL_TEXTURE_2D
//
// CUDA notes: The DVP memcpy2D function differs from CUDA cuMemcpy2D which
// uses byte units for some of the copy parameters.
// For DVP, all parameters are in pixels and the the byte sizes
// of the pixels are assumed to match the those of the source
// or destination DVP sysmem buffer.
//
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpMemcpy2D(DVPBufferHandle srcBuffer,
DVPSyncObjectHandle srcSync,
uint32_t srcAcquireValue,
uint64_t timeout,
DVPBufferHandle dstBuffer,
DVPSyncObjectHandle dstSync,
uint32_t dstReleaseValue,
uint32_t startY,
uint32_t startX,
uint32_t height,
uint32_t width);
DVPAPI_INTERFACE
dvpMemcpy2D(DVPBufferHandle srcBuffer,
DVPSyncObjectHandle srcSync,
uint32_t srcAcquireValue,
uint64_t timeout,
DVPBufferHandle dstBuffer,
DVPSyncObjectHandle dstSync,
uint32_t dstReleaseValue,
uint32_t srcY,
uint32_t srcX,
uint32_t dstY,
uint32_t dstX,
uint32_t height,
uint32_t width);
//------------------------------------------------------------------------
// Function: dvpMapBufferEndAPI
//
// Description: Tells DVP to setup a signal for this buffer in the
// callers API context or device. The signal follows all
// previous API operations up to this point and, thus,
// allows subsequent DVP calls to know when then this buffer
// is ready for use within the DVP library. This function
// would be followed by a call to dvpMapBufferWaitDVP to
// synchronize rendering in the API stream and the DVP
// stream.
//
// If OpenGL or CUDA is used, the OpenGL/CUDA context
// must be current at time of call.
//
// The use of dvpMapBufferEndAPI is NOT recommended for
// CUDA synchronisation, as it is more optimal to use a
// applcation CUDA stream in conjunction with
// dvpMapBufferEndCUDAStream. This allows the driver to
// do optimisations, such as parllelise the copy operations
// and compute.
//
// This must be called outside the dvpBegin/dvpEnd pair. In
// addition, this call is not thread safe and must be called
// from or fenced against the rendering thread associated with
// the context or device.
//
// Parameters: gpuBufferHandle[IN] - buffer to track
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
// DVP_STATUS_UNSIGNALED - returned if the API is
// unable to place a signal in the API context queue
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpMapBufferEndAPI(DVPBufferHandle gpuBufferHandle);
//------------------------------------------------------------------------
// Function: dvpMapBufferEndDVP
//
// Description: Tells DVP to setup a signal for this buffer after
// DVP operations are complete. The signal allows
// the API to know when then this buffer is
// ready for use within a API stream. This function would
// be followed by a call to dvpMapBufferWaitAPI to
// synchronize copies in the DVP stream and the API
// rendering stream.
//
// This must be called inside the dvpBegin/dvpEnd pair.
//
// Parameters: gpuBufferHandle[IN] - buffer to track
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpMapBufferEndDVP(DVPBufferHandle gpuBufferHandle);
//------------------------------------------------------------------------
// Function: dvpMapBufferWaitAPI
//
// Description: Tells DVP to make the current API context or device to
// wait for a previous signal triggered by a
// dvpMapBufferEndDVP call.
//
// The use of dvpMapBufferWaitCUDAStream is NOT recommended for
// CUDA synchronisation, as it is more optimal to use a
// applcation CUDA stream in conjunction with
// dvpMapBufferEndCUDAStream. This allows the driver to
// do optimisations, such as parllelise the copy operations
// and compute.
//
// If OpenGL or CUDA is used, the OpenGL/CUDA context
// must be current at time of call.
//
// This must be called outside the dvpBegin/dvpEnd pair. In
// addition, this call is not thread safe and must be called
// from or fenced against the rendering thread associated with
// the context or device.
//
// Parameters: gpuBufferHandle[IN] - buffer to track
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpMapBufferWaitAPI(DVPBufferHandle gpuBufferHandle);
//------------------------------------------------------------------------
// Function: dvpMapBufferWaitDVP
//
// Description: Tells DVP to make the DVP stream wait for a previous
// signal triggered by a dvpMapBufferEndAPI call.
//
// This must be called inside the dvpBegin/dvpEnd pair.
//
// Parameters: gpuBufferHandle[IN] - buffer to track
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpMapBufferWaitDVP(DVPBufferHandle gpuBufferHandle);
//------------------------------------------------------------------------
// Function: dvpSyncObjCompletion
//
// Description: dvpSyncObjCompletion returns the time of completion
// of the last completed GPU release operation. This time is
// nanoseconds from and is in the same time domain as that
// defined in GL_ARB_timer_query.
//
// Parameters: syncObject[IN] - syncObject
// timeStamp[OUT] - return time in nanoseconds
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpSyncObjCompletion(DVPSyncObjectHandle syncObject,
uint64_t *timeStamp);
//------------------------------------------------------------------------
// Function: dvpImportSyncObject
//
// Description: dvpImportSyncObject creates a DVPSyncObject from the
// DVPSyncObjectDesc. Note that a sync object is not
// supported for copy operations targeting different APIs.
// This means, for example, it is illegal to call dvpMemCpy*
// for source or target GL texture with sync object A and
// then later use that same sync object in dvpMemCpy*
// operation for a source or target CUDA buffer. The same
// semaphore memory can still be used for two different sync
// objects.
//
// Parameters: desc[IN] - data describing the sync object
// syncObject[OUT] - handle to sync object
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpImportSyncObject(DVPSyncObjectDesc *desc,
DVPSyncObjectHandle *syncObject);
//------------------------------------------------------------------------
// Function: dvpFreeSyncObject
//
// Description: dvpFreeSyncObject waits for any outstanding releases on
// this sync object before freeing the resources allocated for
// the specified sync object. The application must make sure
// any outstanding acquire operations have already been
// completed.
//
// If OpenGL is being used and the app's GL context is being
// shared (via the DVP_DEVICE_FLAGS_SHARE_APP_CONTEXT flag),
// then dvpFreeSyncObject needs to be called while each context,
// on which the sync object was used, is current. If
// DVP_DEVICE_FLAGS_SHARE_APP_CONTEXT is used and there are out
// standing contexts from which this sync object must be free'd
// then dvpFreeSyncObject will return DVP_STATUS_SYNC_STILL_BOUND.
//
// Parameters: syncObject[IN] - handle to sync object to be free'd
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_ERROR
// DVP_STATUS_SYNC_STILL_BOUND
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpFreeSyncObject(DVPSyncObjectHandle syncObject);
//------------------------------------------------------------------------
// Function: dvpSyncObjClientWaitComplete
//
// Description: dvpSyncObjClientWaitComplete blocks until the
// corresponding sync object's semaphore is greater than
// or equal to the last release value issued by the
// DVP library. If the sync object has not yet been
// released, then the return value will be
// DVP_STATUS_INVALID_OPERATION.
//
// Function definition changes with library version 1.30:
// This must be called inside a dvpBegin/dvpEnd pair.
//
// Parameters: syncObject[IN] - handle to sync object
// timeout[IN] - time out value in nanoseconds.
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_INVALID_OPERATION
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpSyncObjClientWaitComplete(DVPSyncObjectHandle syncObject,
uint64_t timeout);
//------------------------------------------------------------------------
// Function: dvpSyncObjClientWaitPartial
//
// Description: dvpSyncObjClientWaitPartial blocks until the
// corresponding sync object's semaphore greater than
// or equal to the supplied value. If the sync object
// has not yet been released, then the return value
// will be DVP_STATUS_INVALID_OPERATION.
//
// Parameters: syncObject[IN] - handle to sync object
// value[IN] - value to wait for
// timeout[IN] - time out value in nanoseconds.
//
// Returns: DVP_STATUS_OK
// DVP_STATUS_INVALID_PARAMETER
// DVP_STATUS_INVALID_OPERATION
// DVP_STATUS_ERROR
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpSyncObjClientWaitPartial(DVPSyncObjectHandle syncObject,
uint32_t value,
uint64_t timeout);
//------------------------------------------------------------------------
// Function: dvpGetLibrayVersion
//
// Description: Returns the current version of the library
//
// Parameters: major[OUT] - returned major version
// minor[OUT] - returned minor version
//
// Returns: DVP_STATUS_OK
//------------------------------------------------------------------------
DVPAPI_INTERFACE
dvpGetLibrayVersion(uint32_t *major,
uint32_t *minor);
#endif // __DVAPI_H_