# This file is part of Xpra. # Copyright (C) 2013-2018 Antoine Martin # Xpra is released under the terms of the GNU GPL v2, or, at your option, any # later version. See the file COPYING for details. #cython: auto_pickle=False, cdivision=True, language_level=3 from __future__ import absolute_import import binascii import os import sys import numpy import array from collections import deque, OrderedDict from pycuda import driver from xpra.os_util import WIN32, OSX, LINUX, PYTHON3, bytestostr, strtobytes from xpra.util import AtomicInteger, engs, csv, pver, envint, envbool, first_time from xpra.codecs.cuda_common.cuda_context import init_all_devices, get_devices, select_device, get_device_info, get_device_name, \ get_cuda_info, get_pycuda_info, device_info, reset_state, \ get_CUDA_function, record_device_failure, record_device_success, CUDA_ERRORS_INFO from xpra.codecs.codec_constants import video_spec, TransientCodecException from xpra.codecs.image_wrapper import ImageWrapper from xpra.codecs.nv_util import get_nvidia_module_version, get_license_keys, validate_driver_yuv444lossless, get_cards from xpra.log import Logger log = Logger("encoder", "nvenc") import ctypes from ctypes import cdll as loader, POINTER from libc.stdint cimport uintptr_t, uint8_t, uint16_t, uint32_t, int32_t, uint64_t from xpra.monotonic_time cimport monotonic_time CLIENT_KEYS_STR = get_license_keys(NVENCAPI_MAJOR_VERSION) + get_license_keys() DESIRED_PRESET = os.environ.get("XPRA_NVENC_PRESET", "") #NVENC requires compute capability value 0x30 or above: cdef int MIN_COMPUTE = 0x30 cdef int YUV444_THRESHOLD = envint("XPRA_NVENC_YUV444_THRESHOLD", 85) cdef int LOSSLESS_THRESHOLD = envint("XPRA_NVENC_LOSSLESS_THRESHOLD", 100) cdef int NATIVE_RGB = int(not WIN32) NATIVE_RGB = envbool("XPRA_NVENC_NATIVE_RGB", NATIVE_RGB) cdef int LOSSLESS_ENABLED = envbool("XPRA_NVENC_LOSSLESS", True) cdef int YUV420_ENABLED = envbool("XPRA_NVENC_YUV420P", True) cdef int YUV444_ENABLED = envbool("XPRA_NVENC_YUV444P", True) cdef int DEBUG_API = envbool("XPRA_NVENC_DEBUG_API", False) cdef int GPU_MEMCOPY = envbool("XPRA_NVENC_GPU_MEMCOPY", True) cdef int QP_MAX_VALUE = 51 #newer versions of ffmpeg can decode up to 63 YUV444_CODEC_SUPPORT = { "h264" : False, "h265" : False, } LOSSLESS_CODEC_SUPPORT = {} cdef inline int MIN(int a, int b): if a<=b: return a return b cdef inline int MAX(int a, int b): if a>=b: return a return b cdef extern from "string.h": void* memset(void * ptr, int value, size_t num) void* memcpy(void * destination, void * source, size_t num) cdef extern from "stdlib.h": void* malloc(size_t __size) void free(void* mem) CUresult = ctypes.c_int CUcontext = ctypes.c_void_p cdef extern from "nvEncodeAPI.h": ctypedef int NVENCSTATUS ctypedef void* NV_ENC_INPUT_PTR ctypedef void* NV_ENC_OUTPUT_PTR ctypedef void* NV_ENC_REGISTERED_PTR #not available with driver version 367.35 #NVENCSTATUS NvEncodeAPIGetMaxSupportedVersion(uint32_t* version) ctypedef enum NV_ENC_CAPS: NV_ENC_CAPS_NUM_MAX_BFRAMES NV_ENC_CAPS_SUPPORTED_RATECONTROL_MODES NV_ENC_CAPS_SUPPORT_FIELD_ENCODING NV_ENC_CAPS_SUPPORT_MONOCHROME NV_ENC_CAPS_SUPPORT_FMO NV_ENC_CAPS_SUPPORT_QPELMV NV_ENC_CAPS_SUPPORT_BDIRECT_MODE NV_ENC_CAPS_SUPPORT_CABAC NV_ENC_CAPS_SUPPORT_ADAPTIVE_TRANSFORM NV_ENC_CAPS_NUM_MAX_TEMPORAL_LAYERS NV_ENC_CAPS_SUPPORT_HIERARCHICAL_PFRAMES NV_ENC_CAPS_SUPPORT_HIERARCHICAL_BFRAMES NV_ENC_CAPS_LEVEL_MAX NV_ENC_CAPS_LEVEL_MIN NV_ENC_CAPS_SEPARATE_COLOUR_PLANE NV_ENC_CAPS_WIDTH_MAX NV_ENC_CAPS_HEIGHT_MAX NV_ENC_CAPS_SUPPORT_TEMPORAL_SVC NV_ENC_CAPS_SUPPORT_DYN_RES_CHANGE NV_ENC_CAPS_SUPPORT_DYN_BITRATE_CHANGE NV_ENC_CAPS_SUPPORT_DYN_FORCE_CONSTQP NV_ENC_CAPS_SUPPORT_DYN_RCMODE_CHANGE NV_ENC_CAPS_SUPPORT_SUBFRAME_READBACK NV_ENC_CAPS_SUPPORT_CONSTRAINED_ENCODING NV_ENC_CAPS_SUPPORT_INTRA_REFRESH NV_ENC_CAPS_SUPPORT_CUSTOM_VBV_BUF_SIZE NV_ENC_CAPS_SUPPORT_DYNAMIC_SLICE_MODE NV_ENC_CAPS_SUPPORT_REF_PIC_INVALIDATION NV_ENC_CAPS_PREPROC_SUPPORT NV_ENC_CAPS_ASYNC_ENCODE_SUPPORT NV_ENC_CAPS_MB_NUM_MAX NV_ENC_CAPS_EXPOSED_COUNT NV_ENC_CAPS_SUPPORT_YUV444_ENCODE NV_ENC_CAPS_SUPPORT_LOSSLESS_ENCODE NV_ENC_CAPS_SUPPORT_SAO NV_ENC_CAPS_SUPPORT_MEONLY_MODE NV_ENC_CAPS_SUPPORT_LOOKAHEAD NV_ENC_CAPS_SUPPORT_TEMPORAL_AQ NV_ENC_CAPS_SUPPORT_10BIT_ENCODE NV_ENC_CAPS_NUM_MAX_LTR_FRAMES NV_ENC_CAPS_SUPPORT_WEIGHTED_PREDICTION NV_ENC_CAPS_DYNAMIC_QUERY_ENCODER_CAPACITY NV_ENC_CAPS_SUPPORT_BFRAME_REF_MODE NV_ENC_CAPS_SUPPORT_EMPHASIS_LEVEL_MAP, ctypedef enum NV_ENC_DEVICE_TYPE: NV_ENC_DEVICE_TYPE_DIRECTX NV_ENC_DEVICE_TYPE_CUDA NV_ENC_DEVICE_TYPE_OPENGL ctypedef enum NV_ENC_INPUT_RESOURCE_TYPE: NV_ENC_INPUT_RESOURCE_TYPE_DIRECTX NV_ENC_INPUT_RESOURCE_TYPE_CUDADEVICEPTR NV_ENC_INPUT_RESOURCE_TYPE_CUDAARRAY NV_ENC_INPUT_RESOURCE_TYPE_OPENGL_TEX ctypedef enum NV_ENC_MEMORY_HEAP: NV_ENC_MEMORY_HEAP_AUTOSELECT NV_ENC_MEMORY_HEAP_VID NV_ENC_MEMORY_HEAP_SYSMEM_CACHED NV_ENC_MEMORY_HEAP_SYSMEM_UNCACHED ctypedef enum NV_ENC_H264_ENTROPY_CODING_MODE: NV_ENC_H264_ENTROPY_CODING_MODE_AUTOSELECT NV_ENC_H264_ENTROPY_CODING_MODE_CABAC NV_ENC_H264_ENTROPY_CODING_MODE_CAVLC ctypedef enum NV_ENC_STEREO_PACKING_MODE: NV_ENC_STEREO_PACKING_MODE_NONE NV_ENC_STEREO_PACKING_MODE_CHECKERBOARD NV_ENC_STEREO_PACKING_MODE_COLINTERLEAVE NV_ENC_STEREO_PACKING_MODE_ROWINTERLEAVE NV_ENC_STEREO_PACKING_MODE_SIDEBYSIDE NV_ENC_STEREO_PACKING_MODE_TOPBOTTOM NV_ENC_STEREO_PACKING_MODE_FRAMESEQ ctypedef enum NV_ENC_H264_FMO_MODE: NV_ENC_H264_FMO_AUTOSELECT NV_ENC_H264_FMO_ENABLE NV_ENC_H264_FMO_DISABLE ctypedef enum NV_ENC_H264_BDIRECT_MODE: NV_ENC_H264_BDIRECT_MODE_AUTOSELECT NV_ENC_H264_BDIRECT_MODE_DISABLE NV_ENC_H264_BDIRECT_MODE_TEMPORAL NV_ENC_H264_BDIRECT_MODE_SPATIAL ctypedef enum NV_ENC_H264_ADAPTIVE_TRANSFORM_MODE: NV_ENC_H264_ADAPTIVE_TRANSFORM_AUTOSELECT NV_ENC_H264_ADAPTIVE_TRANSFORM_DISABLE NV_ENC_H264_ADAPTIVE_TRANSFORM_ENABLE ctypedef enum NV_ENC_PARAMS_FRAME_FIELD_MODE: NV_ENC_PARAMS_FRAME_FIELD_MODE_FRAME NV_ENC_PARAMS_FRAME_FIELD_MODE_FIELD NV_ENC_PARAMS_FRAME_FIELD_MODE_MBAFF ctypedef enum NV_ENC_BUFFER_FORMAT: NV_ENC_BUFFER_FORMAT_UNDEFINED NV_ENC_BUFFER_FORMAT_NV12 NV_ENC_BUFFER_FORMAT_YV12 NV_ENC_BUFFER_FORMAT_IYUV NV_ENC_BUFFER_FORMAT_YUV444 NV_ENC_BUFFER_FORMAT_YUV420_10BIT NV_ENC_BUFFER_FORMAT_YUV444_10BIT NV_ENC_BUFFER_FORMAT_ARGB NV_ENC_BUFFER_FORMAT_ARGB10 NV_ENC_BUFFER_FORMAT_AYUV NV_ENC_BUFFER_FORMAT_ABGR NV_ENC_BUFFER_FORMAT_ABGR10 ctypedef enum NV_ENC_PIC_FLAGS: NV_ENC_PIC_FLAG_FORCEINTRA NV_ENC_PIC_FLAG_FORCEIDR NV_ENC_PIC_FLAG_OUTPUT_SPSPPS NV_ENC_PIC_FLAG_EOS ctypedef enum NV_ENC_PIC_STRUCT: NV_ENC_PIC_STRUCT_FRAME NV_ENC_PIC_STRUCT_FIELD_TOP_BOTTOM NV_ENC_PIC_STRUCT_FIELD_BOTTOM_TOP ctypedef enum NV_ENC_PIC_TYPE: NV_ENC_PIC_TYPE_P NV_ENC_PIC_TYPE_B NV_ENC_PIC_TYPE_I NV_ENC_PIC_TYPE_IDR NV_ENC_PIC_TYPE_BI NV_ENC_PIC_TYPE_SKIPPED NV_ENC_PIC_TYPE_INTRA_REFRESH NV_ENC_PIC_TYPE_UNKNOWN ctypedef enum NV_ENC_SLICE_TYPE: NV_ENC_SLICE_TYPE_DEFAULT NV_ENC_SLICE_TYPE_I NV_ENC_SLICE_TYPE_UNKNOWN ctypedef enum NV_ENC_MV_PRECISION: NV_ENC_MV_PRECISION_FULL_PEL NV_ENC_MV_PRECISION_HALF_PEL NV_ENC_MV_PRECISION_QUARTER_PEL ctypedef enum NV_ENC_LEVEL: NV_ENC_LEVEL_AUTOSELECT NV_ENC_LEVEL_H264_1 NV_ENC_LEVEL_H264_1b NV_ENC_LEVEL_H264_11 NV_ENC_LEVEL_H264_12 NV_ENC_LEVEL_H264_13 NV_ENC_LEVEL_H264_2 NV_ENC_LEVEL_H264_21 NV_ENC_LEVEL_H264_22 NV_ENC_LEVEL_H264_3 NV_ENC_LEVEL_H264_31 NV_ENC_LEVEL_H264_32 NV_ENC_LEVEL_H264_4 NV_ENC_LEVEL_H264_41 NV_ENC_LEVEL_H264_42 NV_ENC_LEVEL_H264_5 NV_ENC_LEVEL_H264_51 NV_ENC_LEVEL_H264_52 NV_ENC_LEVEL_HEVC_1 NV_ENC_LEVEL_HEVC_2 NV_ENC_LEVEL_HEVC_21 NV_ENC_LEVEL_HEVC_3 NV_ENC_LEVEL_HEVC_31 NV_ENC_LEVEL_HEVC_4 NV_ENC_LEVEL_HEVC_41 NV_ENC_LEVEL_HEVC_5 NV_ENC_LEVEL_HEVC_51 NV_ENC_LEVEL_HEVC_52 NV_ENC_LEVEL_HEVC_6 NV_ENC_LEVEL_HEVC_61 NV_ENC_LEVEL_HEVC_62 ctypedef enum NV_ENC_PARAMS_RC_MODE: NV_ENC_PARAMS_RC_CONSTQP #Constant QP mode NV_ENC_PARAMS_RC_VBR #Variable bitrate mode NV_ENC_PARAMS_RC_CBR #Constant bitrate mode NV_ENC_PARAMS_RC_CBR_LOWDELAY_HQ #low-delay CBR, high quality NV_ENC_PARAMS_RC_CBR_HQ #CBR, high quality (slower) NV_ENC_PARAMS_RC_VBR_HQ #SDK 7 names (deprecated): NV_ENC_PARAMS_RC_VBR_MINQP #Variable bitrate mode with MinQP NV_ENC_PARAMS_RC_2_PASS_QUALITY #Multi pass encoding optimized for image quality and works only with low latency mode NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP #Multi pass encoding optimized for maintaining frame size and works only with low latency mode NV_ENC_PARAMS_RC_2_PASS_VBR #Multi pass VBR NV_ENC_PARAMS_RC_CBR2 #(deprecated) ctypedef enum NV_ENC_HEVC_CUSIZE: NV_ENC_HEVC_CUSIZE_AUTOSELECT NV_ENC_HEVC_CUSIZE_8x8 NV_ENC_HEVC_CUSIZE_16x16 NV_ENC_HEVC_CUSIZE_32x32 NV_ENC_HEVC_CUSIZE_64x64 ctypedef struct NV_ENC_LOCK_BITSTREAM: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_LOCK_BITSTREAM_VER. uint32_t doNotWait #[in]: If this flag is set, the NvEncodeAPI interface will return buffer pointer even if operation is not completed. If not set, the call will block until operation completes. uint32_t ltrFrame #[out]: Flag indicating this frame is marked as LTR frame uint32_t reservedBitFields #[in]: Reserved bit fields and must be set to 0 void* outputBitstream #[in]: Pointer to the bitstream buffer being locked. uint32_t* sliceOffsets #[in,out]: Array which receives the slice offsets. Currently used only when NV_ENC_CONFIG_H264::sliceMode == 3. Array size must be equal to NV_ENC_CONFIG_H264::sliceModeData. uint32_t frameIdx #[out]: Frame no. for which the bitstream is being retrieved. uint32_t hwEncodeStatus #[out]: The NvEncodeAPI interface status for the locked picture. uint32_t numSlices #[out]: Number of slices in the encoded picture. Will be reported only if NV_ENC_INITIALIZE_PARAMS::reportSliceOffsets set to 1. uint32_t bitstreamSizeInBytes#[out]: Actual number of bytes generated and copied to the memory pointed by bitstreamBufferPtr. uint64_t outputTimeStamp #[out]: Presentation timestamp associated with the encoded output. uint64_t outputDuration #[out]: Presentation duration associates with the encoded output. void* bitstreamBufferPtr #[out]: Pointer to the generated output bitstream. Client should allocate sufficiently large buffer to hold the encoded output. Client is responsible for managing this memory. NV_ENC_PIC_TYPE pictureType #[out]: Picture type of the encoded picture. NV_ENC_PIC_STRUCT pictureStruct #[out]: Structure of the generated output picture. uint32_t frameAvgQP #[out]: Average QP of the frame. uint32_t frameSatd #[out]: Total SATD cost for whole frame. uint32_t ltrFrameIdx #[out]: Frame index associated with this LTR frame. uint32_t ltrFrameBitmap #[out]: Bitmap of LTR frames indices which were used for encoding this frame. Value of 0 if no LTR frames were used. uint32_t reserved[236] #[in]: Reserved and must be set to 0 void* reserved2[64] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_LOCK_INPUT_BUFFER: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_LOCK_INPUT_BUFFER_VER. uint32_t doNotWait #[in]: Set to 1 to make ::NvEncLockInputBuffer() a unblocking call. If the encoding is not completed, driver will return ::NV_ENC_ERR_ENCODER_BUSY error code. uint32_t reservedBitFields #[in]: Reserved bitfields and must be set to 0 NV_ENC_INPUT_PTR inputBuffer #[in]: Pointer to the input buffer to be locked, client should pass the pointer obtained from ::NvEncCreateInputBuffer() or ::NvEncMapInputResource API. void* bufferDataPtr #[out]: Pointed to the locked input buffer data. Client can only access input buffer using the \p bufferDataPtr. uint32_t pitch #[out]: Pitch of the locked input buffer. uint32_t reserved1[251] #[in]: Reserved and must be set to 0 void* reserved2[64] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_STAT: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_STAT_VER. uint32_t reserved #[in]: Reserved and must be set to 0 NV_ENC_OUTPUT_PTR outputBitStream #[out]: Specifies the pointer to output bitstream. uint32_t bitStreamSize #[out]: Size of generated bitstream in bytes. uint32_t picType #[out]: Picture type of encoded picture. See ::NV_ENC_PIC_TYPE. uint32_t lastValidByteOffset #[out]: Offset of last valid bytes of completed bitstream uint32_t sliceOffsets[16] #[out]: Offsets of each slice uint32_t picIdx #[out]: Picture number uint32_t reserved1[233] #[in]: Reserved and must be set to 0 void* reserved2[64] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_SEQUENCE_PARAM_PAYLOAD: pass ctypedef struct NV_ENC_EVENT_PARAMS: pass ctypedef struct NV_ENC_MAP_INPUT_RESOURCE: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_MAP_INPUT_RESOURCE_VER. uint32_t subResourceIndex #[in]: Deprecated. Do not use. void* inputResource #[in]: Deprecated. Do not use. NV_ENC_REGISTERED_PTR registeredResource #[in]: The Registered resource handle obtained by calling NvEncRegisterInputResource. NV_ENC_INPUT_PTR mappedResource #[out]: Mapped pointer corresponding to the registeredResource. This pointer must be used in NV_ENC_PIC_PARAMS::inputBuffer parameter in ::NvEncEncodePicture() API. NV_ENC_BUFFER_FORMAT mappedBufferFmt #[out]: Buffer format of the outputResource. This buffer format must be used in NV_ENC_PIC_PARAMS::bufferFmt if client using the above mapped resource pointer. uint32_t reserved1[251] #[in]: Reserved and must be set to 0. void* reserved2[63] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_REGISTER_RESOURCE: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_REGISTER_RESOURCE_VER. NV_ENC_INPUT_RESOURCE_TYPE resourceType #[in]: Specifies the type of resource to be registered. Supported values are ::NV_ENC_INPUT_RESOURCE_TYPE_DIRECTX, ::NV_ENC_INPUT_RESOURCE_TYPE_CUDADEVICEPTR. uint32_t width #[in]: Input buffer Width. uint32_t height #[in]: Input buffer Height. uint32_t pitch #[in]: Input buffer Pitch. uint32_t subResourceIndex #[in]: Subresource Index of the DirectX resource to be registered. Should eb set to 0 for other interfaces. void* resourceToRegister #[in]: Handle to the resource that is being registered. NV_ENC_REGISTERED_PTR registeredResource #[out]: Registered resource handle. This should be used in future interactions with the Nvidia Video Encoder Interface. NV_ENC_BUFFER_FORMAT bufferFormat #[in]: Buffer format of resource to be registered. uint32_t reserved1[248] #[in]: Reserved and must be set to 0. void* reserved2[62] #[in]: Reserved and must be set to NULL. ctypedef struct GUID: uint32_t Data1 uint16_t Data2 uint16_t Data3 uint8_t Data4[8] #Encode Codec GUIDS supported by the NvEncodeAPI interface. GUID NV_ENC_CODEC_H264_GUID GUID NV_ENC_CODEC_HEVC_GUID #Profiles: GUID NV_ENC_CODEC_PROFILE_AUTOSELECT_GUID GUID NV_ENC_H264_PROFILE_BASELINE_GUID GUID NV_ENC_H264_PROFILE_MAIN_GUID GUID NV_ENC_H264_PROFILE_HIGH_GUID GUID NV_ENC_H264_PROFILE_HIGH_444_GUID GUID NV_ENC_H264_PROFILE_STEREO_GUID GUID NV_ENC_H264_PROFILE_SVC_TEMPORAL_SCALABILTY GUID NV_ENC_H264_PROFILE_PROGRESSIVE_HIGH_GUID GUID NV_ENC_H264_PROFILE_CONSTRAINED_HIGH_GUID GUID NV_ENC_HEVC_PROFILE_MAIN_GUID GUID NV_ENC_HEVC_PROFILE_MAIN10_GUID GUID NV_ENC_HEVC_PROFILE_FREXT_GUID #Presets: GUID NV_ENC_PRESET_DEFAULT_GUID GUID NV_ENC_PRESET_HP_GUID GUID NV_ENC_PRESET_HQ_GUID GUID NV_ENC_PRESET_BD_GUID #V3 ONLY PRESETS: GUID NV_ENC_PRESET_LOW_LATENCY_DEFAULT_GUID GUID NV_ENC_PRESET_LOW_LATENCY_HQ_GUID GUID NV_ENC_PRESET_LOW_LATENCY_HP_GUID #V4 ONLY PRESETS: GUID NV_ENC_PRESET_LOSSLESS_DEFAULT_GUID GUID NV_ENC_PRESET_LOSSLESS_HP_GUID ctypedef struct NV_ENC_CAPS_PARAM: uint32_t version uint32_t capsToQuery uint32_t reserved[62] ctypedef struct NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS_VER. NV_ENC_DEVICE_TYPE deviceType #[in]: (NV_ENC_DEVICE_TYPE) Specified the device Type void *device #[in]: Pointer to client device. GUID *reserved #[in]: Pointer to a GUID key issued to the client. uint32_t apiVersion #[in]: API version. Should be set to NVENCAPI_VERSION. uint32_t reserved1[253] #[in]: Reserved and must be set to 0 void *reserved2[64] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_CREATE_INPUT_BUFFER: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_CREATE_INPUT_BUFFER_VER uint32_t width #[in]: Input buffer width uint32_t height #[in]: Input buffer width NV_ENC_MEMORY_HEAP memoryHeap #[in]: Deprecated. Do not use NV_ENC_BUFFER_FORMAT bufferFmt #[in]: Input buffer format uint32_t reserved #[in]: Reserved and must be set to 0 void *inputBuffer #[out]: Pointer to input buffer void *pSysMemBuffer #[in]: Pointer to existing sysmem buffer uint32_t reserved1[57] #[in]: Reserved and must be set to 0 void *reserved2[63] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_CREATE_BITSTREAM_BUFFER: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_CREATE_BITSTREAM_BUFFER_VER uint32_t size #[in]: Size of the bitstream buffer to be created NV_ENC_MEMORY_HEAP memoryHeap #[in]: Deprecated. Do not use uint32_t reserved #[in]: Reserved and must be set to 0 void *bitstreamBuffer#[out]: Pointer to the output bitstream buffer void *bitstreamBufferPtr #[out]: Reserved and should not be used uint32_t reserved1[58] #[in]: Reserved and should be set to 0 void* reserved2[64] #[in]: Reserved and should be set to NULL ctypedef struct NV_ENC_QP: uint32_t qpInterP uint32_t qpInterB uint32_t qpIntra ctypedef struct NV_ENC_CONFIG_HEVC_VUI_PARAMETERS: uint32_t overscanInfoPresentFlag #[in]: if set to 1 , it specifies that the overscanInfo is present uint32_t overscanInfo #[in]: Specifies the overscan info(as defined in Annex E of the ITU-T Specification). uint32_t videoSignalTypePresentFlag #[in]: If set to 1, it specifies that the videoFormat, videoFullRangeFlag and colourDescriptionPresentFlag are present. */ uint32_t videoFormat #[in]: Specifies the source video format(as defined in Annex E of the ITU-T Specification).*/ uint32_t videoFullRangeFlag #[in]: Specifies the output range of the luma and chroma samples(as defined in Annex E of the ITU-T Specification). */ uint32_t colourDescriptionPresentFlag #[in]: If set to 1, it specifies that the colourPrimaries, transferCharacteristics and colourMatrix are present. */ uint32_t colourPrimaries #[in]: Specifies color primaries for converting to RGB(as defined in Annex E of the ITU-T Specification) */ uint32_t transferCharacteristics #[in]: Specifies the opto-electronic transfer characteristics to use (as defined in Annex E of the ITU-T Specification) */ uint32_t colourMatrix #[in]: Specifies the matrix coefficients used in deriving the luma and chroma from the RGB primaries (as defined in Annex E of the ITU-T Specification). */ uint32_t chromaSampleLocationFlag #[in]: if set to 1 , it specifies that the chromaSampleLocationTop and chromaSampleLocationBot are present.*/ uint32_t chromaSampleLocationTop #[in]: Specifies the chroma sample location for top field(as defined in Annex E of the ITU-T Specification) */ uint32_t chromaSampleLocationBot #[in]: Specifies the chroma sample location for bottom field(as defined in Annex E of the ITU-T Specification) */ uint32_t bitstreamRestrictionFlag #[in]: if set to 1, it specifies the bitstream restriction parameters are present in the bitstream.*/ uint32_t reserved[15] ctypedef struct NV_ENC_CONFIG_H264_VUI_PARAMETERS: uint32_t overscanInfoPresentFlag #[in]: if set to 1 , it specifies that the overscanInfo is present uint32_t overscanInfo #[in]: Specifies the overscan info(as defined in Annex E of the ITU-T Specification). uint32_t videoSignalTypePresentFlag #[in]: If set to 1, it specifies that the videoFormat, videoFullRangeFlag and colourDescriptionPresentFlag are present. uint32_t videoFormat #[in]: Specifies the source video format(as defined in Annex E of the ITU-T Specification). uint32_t videoFullRangeFlag #[in]: Specifies the output range of the luma and chroma samples(as defined in Annex E of the ITU-T Specification). uint32_t colourDescriptionPresentFlag #[in]: If set to 1, it specifies that the colourPrimaries, transferCharacteristics and colourMatrix are present. uint32_t colourPrimaries #[in]: Specifies color primaries for converting to RGB(as defined in Annex E of the ITU-T Specification) uint32_t transferCharacteristics #[in]: Specifies the opto-electronic transfer characteristics to use (as defined in Annex E of the ITU-T Specification) uint32_t colourMatrix #[in]: Specifies the matrix coefficients used in deriving the luma and chroma from the RGB primaries (as defined in Annex E of the ITU-T Specification). uint32_t chromaSampleLocationFlag #[in]: if set to 1 , it specifies that thechromaSampleLocationTop and chromaSampleLocationBot are present. uint32_t chromaSampleLocationTop #[in]: Specifies the chroma sample location for top field(as defined in Annex E of the ITU-T Specification) uint32_t chromaSampleLocationBot #[in]: Specifies the chroma sample location for bottom field(as defined in Annex E of the ITU-T Specification) uint32_t bitstreamRestrictionFlag #[in]: if set to 1, it speficies the bitstream restriction parameters are present in the bitstream. uint32_t reserved[15] ctypedef struct NV_ENC_CONFIG_H264: uint32_t enableTemporalSVC #[in]: Set to 1 to enable SVC temporal uint32_t enableStereoMVC #[in]: Set to 1 to enable stereo MVC uint32_t hierarchicalPFrames #[in]: Set to 1 to enable hierarchical PFrames uint32_t hierarchicalBFrames #[in]: Set to 1 to enable hierarchical BFrames uint32_t outputBufferingPeriodSEI #[in]: Set to 1 to write SEI buffering period syntax in the bitstream uint32_t outputPictureTimingSEI #[in]: Set to 1 to write SEI picture timing syntax in the bitstream uint32_t outputAUD #[in]: Set to 1 to write access unit delimiter syntax in bitstream uint32_t disableSPSPPS #[in]: Set to 1 to disable writing of Sequence and Picture parameter info in bitstream uint32_t outputFramePackingSEI #[in]: Set to 1 to enable writing of frame packing arrangement SEI messages to bitstream uint32_t outputRecoveryPointSEI #[in]: Set to 1 to enable writing of recovery point SEI message uint32_t enableIntraRefresh #[in]: Set to 1 to enable gradual decoder refresh or intra refresh. If the GOP structure uses B frames this will be ignored uint32_t enableConstrainedEncoding #[in]: Set this to 1 to enable constrainedFrame encoding where each slice in the constarined picture is independent of other slices #Check support for constrained encoding using ::NV_ENC_CAPS_SUPPORT_CONSTRAINED_ENCODING caps. uint32_t repeatSPSPPS #[in]: Set to 1 to enable writing of Sequence and Picture parameter for every IDR frame uint32_t enableVFR #[in]: Set to 1 to enable variable frame rate. uint32_t enableLTR #[in]: Currently this feature is not available and must be set to 0. Set to 1 to enable LTR support and auto-mark the first uint32_t qpPrimeYZeroTransformBypassFlag #[in] To enable lossless encode set this to 1, set QP to 0 and RC_mode to NV_ENC_PARAMS_RC_CONSTQP and profile to HIGH_444_PREDICTIVE_PROFILE #Check support for lossless encoding using ::NV_ENC_CAPS_SUPPORT_LOSSLESS_ENCODE caps. uint32_t useConstrainedIntraPred #[in]: Set 1 to enable constrained intra prediction. uint32_t reservedBitFields[15] #[in]: Reserved bitfields and must be set to 0 uint32_t level #[in]: Specifies the encoding level. Client is recommended to set this to NV_ENC_LEVEL_AUTOSELECT in order to enable the NvEncodeAPI interface to select the correct level. uint32_t idrPeriod #[in]: Specifies the IDR interval. If not set, this is made equal to gopLength in NV_ENC_CONFIG.Low latency application client can set IDR interval to NVENC_INFINITE_GOPLENGTH so that IDR frames are not inserted automatically. uint32_t separateColourPlaneFlag #[in]: Set to 1 to enable 4:4:4 separate colour planes uint32_t disableDeblockingFilterIDC #[in]: Specifies the deblocking filter mode. Permissible value range: [0,2] uint32_t numTemporalLayers #[in]: Specifies max temporal layers to be used for hierarchical coding. Valid value range is [1,::NV_ENC_CAPS_NUM_MAX_TEMPORAL_LAYERS] uint32_t spsId #[in]: Specifies the SPS id of the sequence header. uint32_t ppsId #[in]: Specifies the PPS id of the picture header. NV_ENC_H264_ADAPTIVE_TRANSFORM_MODE adaptiveTransformMode #[in]: Specifies the AdaptiveTransform Mode. Check support for AdaptiveTransform mode using ::NV_ENC_CAPS_SUPPORT_ADAPTIVE_TRANSFORM caps. NV_ENC_H264_FMO_MODE fmoMode #[in]: Specified the FMO Mode. Check support for FMO using ::NV_ENC_CAPS_SUPPORT_FMO caps. NV_ENC_H264_BDIRECT_MODE bdirectMode #[in]: Specifies the BDirect mode. Check support for BDirect mode using ::NV_ENC_CAPS_SUPPORT_BDIRECT_MODE caps. NV_ENC_H264_ENTROPY_CODING_MODE entropyCodingMode #[in]: Specifies the entropy coding mode. Check support for CABAC mode using ::NV_ENC_CAPS_SUPPORT_CABAC caps. NV_ENC_STEREO_PACKING_MODE stereoMode #[in]: Specifies the stereo frame packing mode which is to be signalled in frame packing arrangement SEI uint32_t intraRefreshPeriod #[in]: Specifies the interval between successive intra refresh if enableIntrarefresh is set. Requires enableIntraRefresh to be set. #Will be disabled if NV_ENC_CONFIG::gopLength is not set to NVENC_INFINITE_GOPLENGTH. uint32_t intraRefreshCnt #[in]: Specifies the length of intra refresh in number of frames for periodic intra refresh. This value should be smaller than intraRefreshPeriod uint32_t maxNumRefFrames #[in]: Specifies the DPB size used for encoding. Setting it to 0 will let driver use the default dpb size. #The low latency application which wants to invalidate reference frame as an error resilience tool #is recommended to use a large DPB size so that the encoder can keep old reference frames which can be used if recent #frames are invalidated. uint32_t sliceMode #[in]: This parameter in conjunction with sliceModeData specifies the way in which the picture is divided into slices #sliceMode = 0 MB based slices, sliceMode = 1 Byte based slices, sliceMode = 2 MB row based slices, sliceMode = 3, numSlices in Picture #When forceIntraRefreshWithFrameCnt is set it will have priority over sliceMode setting #When sliceMode == 0 and sliceModeData == 0 whole picture will be coded with one slice uint32_t sliceModeData #[in]: Specifies the parameter needed for sliceMode. For: #sliceMode = 0, sliceModeData specifies # of MBs in each slice (except last slice) #sliceMode = 1, sliceModeData specifies maximum # of bytes in each slice (except last slice) #sliceMode = 2, sliceModeData specifies # of MB rows in each slice (except last slice) #sliceMode = 3, sliceModeData specifies number of slices in the picture. Driver will divide picture into slices optimally NV_ENC_CONFIG_H264_VUI_PARAMETERS h264VUIParameters #[in]: Specifies the H264 video usability info pamameters uint32_t ltrNumFrames #[in]: Specifies the number of LTR frames used. Additionally, encoder will mark the first numLTRFrames base layer reference frames within each IDR interval as LTR uint32_t ltrTrustMode #[in]: Specifies the LTR operating mode. Set to 0 to disallow encoding using LTR frames until later specified. Set to 1 to allow encoding using LTR frames unless later invalidated. uint32_t chromaFormatIDC #[in]: Specifies the chroma format. Should be set to 1 for yuv420 input, 3 for yuv444 input. #Check support for YUV444 encoding using ::NV_ENC_CAPS_SUPPORT_YUV444_ENCODE caps. uint32_t maxTemporalLayers #[in]: Specifies the max temporal layer used for hierarchical coding. uint32_t reserved1[270] #[in]: Reserved and must be set to 0 void *reserved2[64] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_CONFIG_HEVC: uint32_t level #[in]: Specifies the level of the encoded bitstream. uint32_t tier #[in]: Specifies the level tier of the encoded bitstream. NV_ENC_HEVC_CUSIZE minCUSize #[in]: Specifies the minimum size of luma coding unit. NV_ENC_HEVC_CUSIZE maxCUSize #[in]: Specifies the maximum size of luma coding unit. Currently NVENC SDK only supports maxCUSize equal to NV_ENC_HEVC_CUSIZE_32x32. uint32_t useConstrainedIntraPred #[in]: Set 1 to enable constrained intra prediction. uint32_t disableDeblockAcrossSliceBoundary #[in]: Set 1 to disable in loop filtering across slice boundary. uint32_t outputBufferingPeriodSEI #[in]: Set 1 to write SEI buffering period syntax in the bitstream uint32_t outputPictureTimingSEI #[in]: Set 1 to write SEI picture timing syntax in the bitstream uint32_t outputAUD #[in]: Set 1 to write Access Unit Delimiter syntax. uint32_t enableLTR #[in]: Set 1 to enable use of long term reference pictures for inter prediction. uint32_t disableSPSPPS #[in]: Set 1 to disable VPS,SPS and PPS signalling in the bitstream. uint32_t repeatSPSPPS #[in]: Set 1 to output VPS,SPS and PPS for every IDR frame. uint32_t enableIntraRefresh #[in]: Set 1 to enable gradual decoder refresh or intra refresh. If the GOP structure uses B frames this will be ignored uint32_t chromaFormatIDC #[in]: Specifies the chroma format. Should be set to 1 for yuv420 input, 3 for yuv444 input. uint32_t pixelBitDepthMinus8 #[in]: Specifies pixel bit depth minus 8. Should be set to 0 for 8 bit input, 2 for 10 bit input. uint32_t reserved #[in]: Reserved bitfields. uint32_t idrPeriod #[in]: Specifies the IDR interval. If not set, this is made equal to gopLength in NV_ENC_CONFIG.Low latency application client can set IDR interval to NVENC_INFINITE_GOPLENGTH so that IDR frames are not inserted automatically. uint32_t intraRefreshPeriod #[in]: Specifies the interval between successive intra refresh if enableIntrarefresh is set. Requires enableIntraRefresh to be set. #Will be disabled if NV_ENC_CONFIG::gopLength is not set to NVENC_INFINITE_GOPLENGTH. uint32_t intraRefreshCnt #[in]: Specifies the length of intra refresh in number of frames for periodic intra refresh. This value should be smaller than intraRefreshPeriod uint32_t maxNumRefFramesInDPB #[in]: Specifies the maximum number of references frames in the DPB. uint32_t ltrNumFrames #[in]: Specifies the maximum number of long term references can be used for prediction uint32_t vpsId #[in]: Specifies the VPS id of the video parameter set. Currently reserved and must be set to 0. uint32_t spsId #[in]: Specifies the SPS id of the sequence header. Currently reserved and must be set to 0. uint32_t ppsId #[in]: Specifies the PPS id of the picture header. Currently reserved and must be set to 0. uint32_t sliceMode #[in]: This parameter in conjunction with sliceModeData specifies the way in which the picture is divided into slices #sliceMode = 0 CTU based slices, sliceMode = 1 Byte based slices, sliceMode = 2 CTU row based slices, sliceMode = 3, numSlices in Picture #When sliceMode == 0 and sliceModeData == 0 whole picture will be coded with one slice uint32_t sliceModeData #[in]: Specifies the parameter needed for sliceMode. For: #sliceMode = 0, sliceModeData specifies # of CTUs in each slice (except last slice) #sliceMode = 1, sliceModeData specifies maximum # of bytes in each slice (except last slice) #sliceMode = 2, sliceModeData specifies # of CTU rows in each slice (except last slice) #sliceMode = 3, sliceModeData specifies number of slices in the picture. Driver will divide picture into slices optimally uint32_t maxTemporalLayersMinus1 #[in]: Specifies the max temporal layer used for hierarchical coding. NV_ENC_CONFIG_HEVC_VUI_PARAMETERS hevcVUIParameters #Specifies the HEVC video usability info pamameters uint32_t reserved1[218] #[in]: Reserved and must be set to 0. void* reserved2[64] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_CODEC_CONFIG: NV_ENC_CONFIG_H264 h264Config #[in]: Specifies the H.264-specific encoder configuration NV_ENC_CONFIG_HEVC hevcConfig #[in]: Specifies the HEVC-specific encoder configuration. Currently unsupported and must not to be used. uint32_t reserved[256] #[in]: Reserved and must be set to 0 ctypedef struct NV_ENC_RC_PARAMS: uint32_t version NV_ENC_PARAMS_RC_MODE rateControlMode #[in]: Specifies the rate control mode. Check support for various rate control modes using ::NV_ENC_CAPS_SUPPORTED_RATECONTROL_MODES caps. NV_ENC_QP constQP #[in]: Specifies the initial QP to be used for encoding, these values would be used for all frames if in Constant QP mode. uint32_t averageBitRate #[in]: Specifies the average bitrate(in bits/sec) used for encoding. uint32_t maxBitRate #[in]: Specifies the maximum bitrate for the encoded output. This is used for VBR and ignored for CBR mode. uint32_t vbvBufferSize #[in]: Specifies the VBV(HRD) buffer size. in bits. Set 0 to use the default VBV buffer size. uint32_t vbvInitialDelay #[in]: Specifies the VBV(HRD) initial delay in bits. Set 0 to use the default VBV initial delay uint32_t enableMinQP #[in]: Set this to 1 if minimum QP used for rate control. uint32_t enableMaxQP #[in]: Set this to 1 if maximum QP used for rate control. uint32_t enableInitialRCQP #[in]: Set this to 1 if user suppplied initial QP is used for rate control. uint32_t enableAQ #[in]: Set this to 1 to enable adaptive quantization. uint32_t reservedBitField1 #[in]: Reserved bitfields and must be set to 0 uint32_t reservedBitFields[27] #[in]: Reserved bitfields and must be set to 0 NV_ENC_QP minQP #[in]: Specifies the minimum QP used for rate control. Client must set NV_ENC_CONFIG::enableMinQP to 1. NV_ENC_QP maxQP #[in]: Specifies the maximum QP used for rate control. Client must set NV_ENC_CONFIG::enableMaxQP to 1. NV_ENC_QP initialRCQP #[in]: Specifies the initial QP used for rate control. Client must set NV_ENC_CONFIG::enableInitialRCQP to 1. uint32_t temporallayerIdxMask#[in]: Specifies the temporal layers (as a bitmask) whose QPs have changed. Valid max bitmask is [2^NV_ENC_CAPS_NUM_MAX_TEMPORAL_LAYERS - 1] uint8_t temporalLayerQP[8] #[in]: Specifies the temporal layer QPs used for rate control. Temporal layer index is used as as the array index uint8_t targetQuality #[in]: Target CQ (Constant Quality) level for VBR mode (range 0-51 with 0-automatic) uint8_t targetQualityLSB #[in]: Fractional part of target quality (as 8.8 fixed point format) uint16_t lookaheadDepth #[in]: Maximum depth of lookahead with range 0-32 (only used if enableLookahead=1) uint32_t reserved[9] ctypedef struct NV_ENC_CONFIG: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_CONFIG_VER. GUID profileGUID #[in]: Specifies the codec profile guid. If client specifies \p NV_ENC_CODEC_PROFILE_AUTOSELECT_GUID the NvEncodeAPI interface will select the appropriate codec profile. uint32_t gopLength #[in]: Specifies the number of pictures in one GOP. Low latency application client can set goplength to NVENC_INFINITE_GOPLENGTH so that keyframes are not inserted automatically. int32_t frameIntervalP #[in]: Specifies the GOP pattern as follows: \p frameIntervalP = 0: I, 1: IPP, 2: IBP, 3: IBBP If goplength is set to NVENC_INFINITE_GOPLENGTH \p frameIntervalP should be set to 1. uint32_t monoChromeEncoding #[in]: Set this to 1 to enable monochrome encoding for this session. NV_ENC_PARAMS_FRAME_FIELD_MODE frameFieldMode #[in]: Specifies the frame/field mode. Check support for field encoding using ::NV_ENC_CAPS_SUPPORT_FIELD_ENCODING caps. NV_ENC_MV_PRECISION mvPrecision #[in]: Specifies the desired motion vector prediction precision. NV_ENC_RC_PARAMS rcParams #[in]: Specifies the rate control parameters for the current encoding session. NV_ENC_CODEC_CONFIG encodeCodecConfig #[in]: Specifies the codec specific config parameters through this union. uint32_t reserved[278] #[in]: Reserved and must be set to 0 void *reserved2[64] #[in]: Reserved and must be set to NULL ctypedef struct NVENC_EXTERNAL_ME_HINT_COUNTS_PER_BLOCKTYPE: uint32_t numCandsPerBlk16x16 #[in]: Specifies the number of candidates per 16x16 block. uint32_t numCandsPerBlk16x8 #[in]: Specifies the number of candidates per 16x8 block. uint32_t numCandsPerBlk8x16 #[in]: Specifies the number of candidates per 8x16 block. uint32_t numCandsPerBlk8x8 #[in]: Specifies the number of candidates per 8x8 block. uint32_t reserved #[in]: Reserved for padding. uint32_t reserved1[3] #[in]: Reserved for future use. ctypedef struct NV_ENC_INITIALIZE_PARAMS: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_INITIALIZE_PARAMS_VER. GUID encodeGUID #[in]: Specifies the Encode GUID for which the encoder is being created. ::NvEncInitializeEncoder() API will fail if this is not set, or set to unsupported value. GUID presetGUID #[in]: Specifies the preset for encoding. If the preset GUID is set then , the preset configuration will be applied before any other parameter. uint32_t encodeWidth #[in]: Specifies the encode width. If not set ::NvEncInitializeEncoder() API will fail. uint32_t encodeHeight #[in]: Specifies the encode height. If not set ::NvEncInitializeEncoder() API will fail. uint32_t darWidth #[in]: Specifies the display aspect ratio Width. uint32_t darHeight #[in]: Specifies the display aspect ratio height. uint32_t frameRateNum #[in]: Specifies the numerator for frame rate used for encoding in frames per second ( Frame rate = frameRateNum / frameRateDen ). uint32_t frameRateDen #[in]: Specifies the denominator for frame rate used for encoding in frames per second ( Frame rate = frameRateNum / frameRateDen ). uint32_t enableEncodeAsync #[in]: Set this to 1 to enable asynchronous mode and is expected to use events to get picture completion notification. uint32_t enablePTD #[in]: Set this to 1 to enable the Picture Type Decision is be taken by the NvEncodeAPI interface. uint32_t reportSliceOffsets #[in]: Set this to 1 to enable reporting slice offsets in ::_NV_ENC_LOCK_BITSTREAM. Currently supported only for H264. Client must set this to 0 if NV_ENC_CONFIG_H264::sliceMode is 1 uint32_t enableSubFrameWrite #[in]: Set this to 1 to write out available bitstream to memory at subframe intervals uint32_t enableExternalMEHints #[in]: Set to 1 to enable external ME hints for the current frame. Currently this feature is supported only if NV_ENC_INITIALIZE_PARAMS::enablePTD to 0 or\p frameIntervalP = 1 (i.e no B frames). uint32_t enableMEOnlyMode #[in] Set to 1 to enable ME Only Mode uint32_t reservedBitFields[28] #[in]: Reserved bitfields and must be set to 0 uint32_t privDataSize #[in]: Reserved private data buffer size and must be set to 0 void *privData #[in]: Reserved private data buffer and must be set to NULL NV_ENC_CONFIG *encodeConfig #[in]: Specifies the advanced codec specific structure. If client has sent a valid codec config structure, it will override parameters set by the NV_ENC_INITIALIZE_PARAMS::presetGUID parameter. If set to NULL the NvEncodeAPI interface will use the NV_ENC_INITIALIZE_PARAMS::presetGUID to set the codec specific parameters. #Client can also optionally query the NvEncodeAPI interface to get codec specific parameters for a presetGUID using ::NvEncGetEncodePresetConfig() API. It can then modify (if required) some of the codec config parameters and send down a custom config structure as part of ::_NV_ENC_INITIALIZE_PARAMS. #Even in this case client is recommended to pass the same preset guid it has used in ::NvEncGetEncodePresetConfig() API to query the config structure; as NV_ENC_INITIALIZE_PARAMS::presetGUID. This will not override the custom config structure but will be used to determine other Encoder HW specific parameters not exposed in the API. uint32_t maxEncodeWidth #[in]: Maximum encode width to be used for current Encode session. #Client should allocate output buffers according to this dimension for dynamic resolution change. If set to 0, Encoder will not allow dynamic resolution change. uint32_t maxEncodeHeight #[in]: Maximum encode height to be allowed for current Encode session. #Client should allocate output buffers according to this dimension for dynamic resolution change. If set to 0, Encode will not allow dynamic resolution change. NVENC_EXTERNAL_ME_HINT_COUNTS_PER_BLOCKTYPE maxMEHintCountsPerBlock[2] #[in]: If Client wants to pass external motion vectors in NV_ENC_PIC_PARAMS::meExternalHints buffer it must specify the maximum number of hint candidates per block per direction for the encode session. #The NV_ENC_INITIALIZE_PARAMS::maxMEHintCountsPerBlock[0] is for L0 predictors and NV_ENC_INITIALIZE_PARAMS::maxMEHintCountsPerBlock[1] is for L1 predictors. #This client must also set NV_ENC_INITIALIZE_PARAMS::enableExternalMEHints to 1. uint32_t reserved[289] #[in]: Reserved and must be set to 0 void *reserved2[64] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_RECONFIGURE_PARAMS: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_RECONFIGURE_PARAMS_VER. NV_ENC_INITIALIZE_PARAMS reInitEncodeParams uint32_t resetEncoder #[in]: This resets the rate control states and other internal encoder states. This should be used only with an IDR frame. #If NV_ENC_INITIALIZE_PARAMS::enablePTD is set to 1, encoder will force the frame type to IDR uint32_t forceIDR #[in]: Encode the current picture as an IDR picture. This flag is only valid when Picture type decision is taken by the Encoder #[_NV_ENC_INITIALIZE_PARAMS::enablePTD == 1]. uint32_t reserved ctypedef struct NV_ENC_PRESET_CONFIG: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_PRESET_CONFIG_VER. NV_ENC_CONFIG presetCfg #[out]: preset config returned by the Nvidia Video Encoder interface. uint32_t reserved1[255] #[in]: Reserved and must be set to 0 void* reserved2[64] #[in]: Reserved and must be set to NULL ctypedef struct NV_ENC_H264_SEI_PAYLOAD: uint32_t payloadSize #[in] SEI payload size in bytes. SEI payload must be byte aligned, as described in Annex D uint32_t payloadType #[in] SEI payload types and syntax can be found in Annex D of the H.264 Specification. uint8_t *payload #[in] pointer to user data ctypedef NV_ENC_H264_SEI_PAYLOAD NV_ENC_SEI_PAYLOAD ctypedef struct NV_ENC_PIC_PARAMS_H264: uint32_t displayPOCSyntax #[in]: Specifies the display POC syntax This is required to be set if client is handling the picture type decision. uint32_t reserved3 #[in]: Reserved and must be set to 0 uint32_t refPicFlag #[in]: Set to 1 for a reference picture. This is ignored if NV_ENC_INITIALIZE_PARAMS::enablePTD is set to 1. uint32_t colourPlaneId #[in]: Specifies the colour plane ID associated with the current input. uint32_t forceIntraRefreshWithFrameCnt #[in]: Forces an intra refresh with duration equal to intraRefreshFrameCnt. #When outputRecoveryPointSEI is set this is value is used for recovery_frame_cnt in recovery point SEI message #forceIntraRefreshWithFrameCnt cannot be used if B frames are used in the GOP structure specified uint32_t constrainedFrame #[in]: Set to 1 if client wants to encode this frame with each slice completely independent of other slices in the frame. #NV_ENC_INITIALIZE_PARAMS::enableConstrainedEncoding should be set to 1 uint32_t sliceModeDataUpdate #[in]: Set to 1 if client wants to change the sliceModeData field to speficy new sliceSize Parameter #When forceIntraRefreshWithFrameCnt is set it will have priority over sliceMode setting uint32_t ltrMarkFrame #[in]: Set to 1 if client wants to mark this frame as LTR uint32_t ltrUseFrames #[in]: Set to 1 if client allows encoding this frame using the LTR frames specified in ltrFrameBitmap uint32_t reservedBitFields #[in]: Reserved bit fields and must be set to 0 uint8_t* sliceTypeData #[in]: Array which specifies the slice type used to force intra slice for a particular slice. Currently supported only for NV_ENC_CONFIG_H264::sliceMode == 3. #Client should allocate array of size sliceModeData where sliceModeData is specified in field of ::_NV_ENC_CONFIG_H264 #Array element with index n corresponds to nth slice. To force a particular slice to intra client should set corresponding array element to NV_ENC_SLICE_TYPE_I #all other array elements should be set to NV_ENC_SLICE_TYPE_DEFAULT uint32_t sliceTypeArrayCnt #[in]: Client should set this to the number of elements allocated in sliceTypeData array. If sliceTypeData is NULL then this should be set to 0 uint32_t seiPayloadArrayCnt #[in]: Specifies the number of elements allocated in seiPayloadArray array. NV_ENC_SEI_PAYLOAD *seiPayloadArray #[in]: Array of SEI payloads which will be inserted for this frame. uint32_t sliceMode #[in]: This parameter in conjunction with sliceModeData specifies the way in which the picture is divided into slices #sliceMode = 0 MB based slices, sliceMode = 1 Byte based slices, sliceMode = 2 MB row based slices, sliceMode = 3, numSlices in Picture #When forceIntraRefreshWithFrameCnt is set it will have priority over sliceMode setting #When sliceMode == 0 and sliceModeData == 0 whole picture will be coded with one slice uint32_t sliceModeData #[in]: Specifies the parameter needed for sliceMode. For: #sliceMode = 0, sliceModeData specifies # of MBs in each slice (except last slice) #sliceMode = 1, sliceModeData specifies maximum # of bytes in each slice (except last slice) #sliceMode = 2, sliceModeData specifies # of MB rows in each slice (except last slice) #sliceMode = 3, sliceModeData specifies number of slices in the picture. Driver will divide picture into slices optimally uint32_t ltrMarkFrameIdx #[in]: Specifies the long term referenceframe index to use for marking this frame as LTR. uint32_t ltrUseFrameBitmap #[in]: Specifies the the associated bitmap of LTR frame indices when encoding this frame. uint32_t ltrUsageMode #[in]: Specifies additional usage constraints for encoding using LTR frames from this point further. 0: no constraints, 1: no short term refs older than current, no previous LTR frames. uint32_t reserved[243] #[in]: Reserved and must be set to 0. void* reserved2[62] #[in]: Reserved and must be set to NULL. ctypedef struct NV_ENC_PIC_PARAMS_HEVC: uint32_t displayPOCSyntax #[in]: Specifies the display POC syntax This is required to be set if client is handling the picture type decision. uint32_t refPicFlag #[in]: Set to 1 for a reference picture. This is ignored if NV_ENC_INITIALIZE_PARAMS::enablePTD is set to 1. uint32_t temporalId #[in]: Specifies the temporal id of the picture uint32_t forceIntraRefreshWithFrameCnt #[in]: Forces an intra refresh with duration equal to intraRefreshFrameCnt. #When outputRecoveryPointSEI is set this is value is used for recovery_frame_cnt in recovery point SEI message #forceIntraRefreshWithFrameCnt cannot be used if B frames are used in the GOP structure specified uint32_t constrainedFrame #[in]: Set to 1 if client wants to encode this frame with each slice completely independent of other slices in the frame #NV_ENC_INITIALIZE_PARAMS::enableConstrainedEncoding should be set to 1 uint32_t sliceModeDataUpdate #[in]: Set to 1 if client wants to change the sliceModeData field to specify new sliceSize Parameter #When forceIntraRefreshWithFrameCnt is set it will have priority over sliceMode setting uint32_t ltrMarkFrame #[in]: Set to 1 if client wants to mark this frame as LTR uint32_t ltrUseFrames #[in]: Set to 1 if client allows encoding this frame using the LTR frames specified in ltrFrameBitmap uint32_t reservedBitFields #[in]: Reserved bit fields and must be set to 0 uint8_t* sliceTypeData #[in]: Array which specifies the slice type used to force intra slice for a particular slice. Currently supported only for NV_ENC_CONFIG_H264::sliceMode == 3. #Client should allocate array of size sliceModeData where sliceModeData is specified in field of ::_NV_ENC_CONFIG_H264 #Array element with index n corresponds to nth slice. To force a particular slice to intra client should set corresponding array element to NV_ENC_SLICE_TYPE_I #all other array elements should be set to NV_ENC_SLICE_TYPE_DEFAULT uint32_t sliceTypeArrayCnt #[in]: Client should set this to the number of elements allocated in sliceTypeData array. If sliceTypeData is NULL then this should be set to 0 uint32_t sliceMode #[in]: This parameter in conjunction with sliceModeData specifies the way in which the picture is divided into slices #sliceMode = 0 CTU based slices, sliceMode = 1 Byte based slices, sliceMode = 2 CTU row based slices, sliceMode = 3, numSlices in Picture #When forceIntraRefreshWithFrameCnt is set it will have priority over sliceMode setting #When sliceMode == 0 and sliceModeData == 0 whole picture will be coded with one slice uint32_t sliceModeData #[in]: Specifies the parameter needed for sliceMode. For: #sliceMode = 0, sliceModeData specifies # of CTUs in each slice (except last slice) #sliceMode = 1, sliceModeData specifies maximum # of bytes in each slice (except last slice) #sliceMode = 2, sliceModeData specifies # of CTU rows in each slice (except last slice) #sliceMode = 3, sliceModeData specifies number of slices in the picture. Driver will divide picture into slices optimally uint32_t ltrMarkFrameIdx #[in]: Specifies the long term reference frame index to use for marking this frame as LTR. uint32_t ltrUseFrameBitmap #[in]: Specifies the associated bitmap of LTR frame indices to use when encoding this frame. uint32_t ltrUsageMode #[in]: Not supported. Reserved for future use and must be set to 0. uint32_t seiPayloadArrayCnt #[in]: Specifies the number of elements allocated in seiPayloadArray array. uint32_t reserved #[in]: Reserved and must be set to 0. NV_ENC_SEI_PAYLOAD* seiPayloadArray #[in]: Array of SEI payloads which will be inserted for this frame. uint32_t reserved2 [244] #[in]: Reserved and must be set to 0. void* reserved3[61] #[in]: Reserved and must be set to NULL. ctypedef union NV_ENC_CODEC_PIC_PARAMS: NV_ENC_PIC_PARAMS_H264 h264PicParams #[in]: H264 encode picture params. NV_ENC_PIC_PARAMS_HEVC hevcPicParams #[in]: HEVC encode picture params. uint32_t reserved[256] #[in]: Reserved and must be set to 0. ctypedef struct NV_ENC_MEONLY_PARAMS: uint32_t version #[in]: Struct version. Must be set to NV_ENC_MEONLY_PARAMS_VER. uint32_t inputWidth #[in]: Specifies the input buffer width uint32_t inputHeight #[in]: Specifies the input buffer height NV_ENC_INPUT_PTR inputBuffer #[in]: Specifies the input buffer pointer. Client must use a pointer obtained from NvEncCreateInputBuffer() or NvEncMapInputResource() APIs. NV_ENC_INPUT_PTR referenceFrame #[in]: Specifies the reference frame pointer NV_ENC_OUTPUT_PTR outputMV #[in,out]: Specifies the pointer to output motion vector data buffer allocated by NvEncCreateMVBuffer. NV_ENC_BUFFER_FORMAT bufferFmt #[in]: Specifies the input buffer format. uint32_t reserved1[252] #[in]: Reserved and must be set to 0 void* reserved2[61] #[in]: Reserved and must be set to NULL ctypedef struct NVENC_EXTERNAL_ME_HINT: int32_t mvx #[in]: Specifies the x component of integer pixel MV (relative to current MB) S12.0. int32_t mvy #[in]: Specifies the y component of integer pixel MV (relative to current MB) S10.0 int32_t refidx #[in]: Specifies the reference index (31=invalid). Current we support only 1 reference frame per direction for external hints, so \p refidx must be 0. int32_t dir #[in]: Specifies the direction of motion estimation . 0=L0 1=L1. int32_t partType #[in]: Specifies the bloack partition type.0=16x16 1=16x8 2=8x16 3=8x8 (blocks in partition must be consecutive). int32_t lastofPart #[in]: Set to 1 for the last MV of (sub) partition int32_t lastOfMB #[in]: Set to 1 for the last MV of macroblock. ctypedef struct NV_ENC_PIC_PARAMS: uint32_t version #[in]: Struct version. Must be set to ::NV_ENC_PIC_PARAMS_VER. uint32_t inputWidth #[in]: Specifies the input buffer width uint32_t inputHeight #[in]: Specifies the input buffer height uint32_t inputPitch #[in]: Specifies the input buffer pitch. If pitch value is not known, set this to inputWidth. uint32_t encodePicFlags #[in]: Specifies bit-wise OR`ed encode pic flags. See ::NV_ENC_PIC_FLAGS enum. uint32_t frameIdx #[in]: Specifies the frame index associated with the input frame [optional]. uint64_t inputTimeStamp #[in]: Specifies presentation timestamp associated with the input picture. uint64_t inputDuration #[in]: Specifies duration of the input picture NV_ENC_INPUT_PTR inputBuffer #[in]: Specifies the input buffer pointer. Client must use a pointer obtained from ::NvEncCreateInputBuffer() or ::NvEncMapInputResource() APIs. NV_ENC_OUTPUT_PTR outputBitstream #[in]: Specifies the pointer to output buffer. Client should use a pointer obtained from ::NvEncCreateBitstreamBuffer() API. void* completionEvent #[in]: Specifies an event to be signalled on completion of encoding of this Frame [only if operating in Asynchronous mode]. Each output buffer should be associated with a distinct event pointer. NV_ENC_BUFFER_FORMAT bufferFmt #[in]: Specifies the input buffer format. NV_ENC_PIC_STRUCT pictureStruct #[in]: Specifies structure of the input picture. NV_ENC_PIC_TYPE pictureType #[in]: Specifies input picture type. Client required to be set explicitly by the client if the client has not set NV_ENC_INITALIZE_PARAMS::enablePTD to 1 while calling NvInitializeEncoder. NV_ENC_CODEC_PIC_PARAMS codecPicParams #[in]: Specifies the codec specific per-picture encoding parameters. uint32_t newEncodeWidth #[in]: Specifies the new output width for current Encoding session, in case of dynamic resolution change. Client should only set this in combination with NV_ENC_PIC_FLAGS::NV_ENC_PIC_FLAG_DYN_RES_CHANGE. #Additionally, if Picture Type decision is handled by the Client [_NV_ENC_INITIALIZE_PARAMS::enablePTD == 0], the client should set the _NV_ENC_PIC_PARAMS::pictureType as ::NV_ENC_PIC_TYPE_IDR. #If _NV_ENC_INITIALIZE_PARAMS::enablePTD == 1, then the Encoder will generate an IDR frame corresponding to this input. uint32_t newEncodeHeight #[in]: Specifies the new output width for current Encoding session, in case of dynamic resolution change. Client should only set this in combination with NV_ENC_PIC_FLAGS::NV_ENC_PIC_FLAG_DYN_RES_CHANGE. #Additionally, if Picture Type decision is handled by the Client [_NV_ENC_INITIALIZE_PARAMS::enablePTD == 0], the client should set the _NV_ENC_PIC_PARAMS::pictureType as ::NV_ENC_PIC_TYPE_IDR. #If _NV_ENC_INITIALIZE_PARAMS::enablePTD == 1, then the Encoder will generate an IDR frame corresponding to this input. NV_ENC_RC_PARAMS rcParams #[in]: Specifies the rate control parameters for the current encoding session. NVENC_EXTERNAL_ME_HINT_COUNTS_PER_BLOCKTYPE meHintCountsPerBlock[2] #[in]: Specifies the number of hint candidates per block per direction for the current frame. meHintCountsPerBlock[0] is for L0 predictors and meHintCountsPerBlock[1] is for L1 predictors. #The candidate count in NV_ENC_PIC_PARAMS::meHintCountsPerBlock[lx] must never exceed NV_ENC_INITIALIZE_PARAMS::maxMEHintCountsPerBlock[lx] provided during encoder intialization. NVENC_EXTERNAL_ME_HINT *meExternalHints #[in]: Specifies the pointer to ME external hints for the current frame. The size of ME hint buffer should be equal to number of macroblocks multiplied by the total number of candidates per macroblock. #The total number of candidates per MB per direction = 1*meHintCountsPerBlock[Lx].numCandsPerBlk16x16 + 2*meHintCountsPerBlock[Lx].numCandsPerBlk16x8 + 2*meHintCountsPerBlock[Lx].numCandsPerBlk8x8 # + 4*meHintCountsPerBlock[Lx].numCandsPerBlk8x8. For frames using bidirectional ME , the total number of candidates for single macroblock is sum of total number of candidates per MB for each direction (L0 and L1) uint32_t newDarWidth #[in]: Specifies the new disalay aspect ratio width for current Encoding session, in case of dynamic resolution change. Client should only set this in combination with NV_ENC_PIC_FLAGS::NV_ENC_PIC_FLAG_DYN_RES_CHANGE. #Additionally, if Picture Type decision is handled by the Client [_NV_ENC_INITIALIZE_PARAMS::enablePTD == 0], the client should set the _NV_ENC_PIC_PARAMS::pictureType as ::NV_ENC_PIC_TYPE_IDR. #If _NV_ENC_INITIALIZE_PARAMS::enablePTD == 1, then the Encoder will generate an IDR frame corresponding to this input. uint32_t newDarHeight #[in]: Specifies the new disalay aspect ratio height for current Encoding session, in case of dynamic resolution change. Client should only set this in combination with NV_ENC_PIC_FLAGS::NV_ENC_PIC_FLAG_DYN_RES_CHANGE. #If _NV_ENC_INITIALIZE_PARAMS::enablePTD == 1, then the Encoder will generate an IDR frame corresponding to this input. uint32_t reserved1[259] #[in]: Reserved and must be set to 0 void* reserved2[63] #[in]: Reserved and must be set to NULL #NVENCSTATUS NvEncodeAPICreateInstance(NV_ENCODE_API_FUNCTION_LIST *functionList) ctypedef NVENCSTATUS (*PNVENCOPENENCODESESSION) (void* device, uint32_t deviceType, void** encoder) nogil ctypedef NVENCSTATUS (*PNVENCGETENCODEGUIDCOUNT) (void* encoder, uint32_t* encodeGUIDCount) nogil ctypedef NVENCSTATUS (*PNVENCGETENCODEGUIDS) (void* encoder, GUID* GUIDs, uint32_t guidArraySize, uint32_t* GUIDCount) nogil ctypedef NVENCSTATUS (*PNVENCGETENCODEPROFILEGUIDCOUNT) (void* encoder, GUID encodeGUID, uint32_t* encodeProfileGUIDCount) nogil ctypedef NVENCSTATUS (*PNVENCGETENCODEPROFILEGUIDS) (void* encoder, GUID encodeGUID, GUID* profileGUIDs, uint32_t guidArraySize, uint32_t* GUIDCount) nogil ctypedef NVENCSTATUS (*PNVENCGETINPUTFORMATCOUNT) (void* encoder, GUID encodeGUID, uint32_t* inputFmtCount) nogil ctypedef NVENCSTATUS (*PNVENCGETINPUTFORMATS) (void* encoder, GUID encodeGUID, NV_ENC_BUFFER_FORMAT* inputFmts, uint32_t inputFmtArraySize, uint32_t* inputFmtCount) nogil ctypedef NVENCSTATUS (*PNVENCGETENCODECAPS) (void* encoder, GUID encodeGUID, NV_ENC_CAPS_PARAM* capsParam, int* capsVal) nogil ctypedef NVENCSTATUS (*PNVENCGETENCODEPRESETCOUNT) (void* encoder, GUID encodeGUID, uint32_t* encodePresetGUIDCount) nogil ctypedef NVENCSTATUS (*PNVENCGETENCODEPRESETGUIDS) (void* encoder, GUID encodeGUID, GUID* presetGUIDs, uint32_t guidArraySize, uint32_t* encodePresetGUIDCount) nogil ctypedef NVENCSTATUS (*PNVENCGETENCODEPRESETCONFIG) (void* encoder, GUID encodeGUID, GUID presetGUID, NV_ENC_PRESET_CONFIG* presetConfig) nogil ctypedef NVENCSTATUS (*PNVENCINITIALIZEENCODER) (void* encoder, NV_ENC_INITIALIZE_PARAMS* createEncodeParams) nogil ctypedef NVENCSTATUS (*PNVENCCREATEINPUTBUFFER) (void* encoder, NV_ENC_CREATE_INPUT_BUFFER* createInputBufferParams) nogil ctypedef NVENCSTATUS (*PNVENCDESTROYINPUTBUFFER) (void* encoder, NV_ENC_INPUT_PTR inputBuffer) nogil ctypedef NVENCSTATUS (*PNVENCCREATEBITSTREAMBUFFER) (void* encoder, NV_ENC_CREATE_BITSTREAM_BUFFER* createBitstreamBufferParams) nogil ctypedef NVENCSTATUS (*PNVENCDESTROYBITSTREAMBUFFER) (void* encoder, NV_ENC_OUTPUT_PTR bitstreamBuffer) nogil ctypedef NVENCSTATUS (*PNVENCENCODEPICTURE) (void* encoder, NV_ENC_PIC_PARAMS* encodePicParams) nogil ctypedef NVENCSTATUS (*PNVENCLOCKBITSTREAM) (void* encoder, NV_ENC_LOCK_BITSTREAM* lockBitstreamBufferParams) nogil ctypedef NVENCSTATUS (*PNVENCUNLOCKBITSTREAM) (void* encoder, NV_ENC_OUTPUT_PTR bitstreamBuffer) nogil ctypedef NVENCSTATUS (*PNVENCLOCKINPUTBUFFER) (void* encoder, NV_ENC_LOCK_INPUT_BUFFER* lockInputBufferParams) nogil ctypedef NVENCSTATUS (*PNVENCUNLOCKINPUTBUFFER) (void* encoder, NV_ENC_INPUT_PTR inputBuffer) nogil ctypedef NVENCSTATUS (*PNVENCGETENCODESTATS) (void* encoder, NV_ENC_STAT* encodeStats) nogil ctypedef NVENCSTATUS (*PNVENCGETSEQUENCEPARAMS) (void* encoder, NV_ENC_SEQUENCE_PARAM_PAYLOAD* sequenceParamPayload) nogil ctypedef NVENCSTATUS (*PNVENCREGISTERASYNCEVENT) (void* encoder, NV_ENC_EVENT_PARAMS* eventParams) nogil ctypedef NVENCSTATUS (*PNVENCUNREGISTERASYNCEVENT) (void* encoder, NV_ENC_EVENT_PARAMS* eventParams) nogil ctypedef NVENCSTATUS (*PNVENCMAPINPUTRESOURCE) (void* encoder, NV_ENC_MAP_INPUT_RESOURCE* mapInputResParams) nogil ctypedef NVENCSTATUS (*PNVENCUNMAPINPUTRESOURCE) (void* encoder, NV_ENC_INPUT_PTR mappedInputBuffer) nogil ctypedef NVENCSTATUS (*PNVENCDESTROYENCODER) (void* encoder) nogil ctypedef NVENCSTATUS (*PNVENCINVALIDATEREFFRAMES) (void* encoder, uint64_t invalidRefFrameTimeStamp) nogil ctypedef NVENCSTATUS (*PNVENCOPENENCODESESSIONEX) (NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS *openSessionExParams, void** encoder) nogil ctypedef NVENCSTATUS (*PNVENCREGISTERRESOURCE) (void* encoder, NV_ENC_REGISTER_RESOURCE* registerResParams) nogil ctypedef NVENCSTATUS (*PNVENCUNREGISTERRESOURCE) (void* encoder, NV_ENC_REGISTERED_PTR registeredRes) nogil ctypedef NVENCSTATUS (*PNVENCRECONFIGUREENCODER) (void* encoder, NV_ENC_RECONFIGURE_PARAMS* reInitEncodeParams) nogil ctypedef struct NV_ENCODE_API_FUNCTION_LIST: uint32_t version #[in]: Client should pass NV_ENCODE_API_FUNCTION_LIST_VER. uint32_t reserved #[in]: Reserved and should be set to 0. PNVENCOPENENCODESESSION nvEncOpenEncodeSession PNVENCGETENCODEGUIDCOUNT nvEncGetEncodeGUIDCount PNVENCGETENCODEPROFILEGUIDCOUNT nvEncGetEncodeProfileGUIDCount PNVENCGETENCODEPROFILEGUIDS nvEncGetEncodeProfileGUIDs PNVENCGETENCODEGUIDS nvEncGetEncodeGUIDs PNVENCGETINPUTFORMATCOUNT nvEncGetInputFormatCount PNVENCGETINPUTFORMATS nvEncGetInputFormats PNVENCGETENCODECAPS nvEncGetEncodeCaps PNVENCGETENCODEPRESETCOUNT nvEncGetEncodePresetCount PNVENCGETENCODEPRESETGUIDS nvEncGetEncodePresetGUIDs PNVENCGETENCODEPRESETCONFIG nvEncGetEncodePresetConfig PNVENCINITIALIZEENCODER nvEncInitializeEncoder PNVENCCREATEINPUTBUFFER nvEncCreateInputBuffer PNVENCDESTROYINPUTBUFFER nvEncDestroyInputBuffer PNVENCCREATEBITSTREAMBUFFER nvEncCreateBitstreamBuffer PNVENCDESTROYBITSTREAMBUFFER nvEncDestroyBitstreamBuffer PNVENCENCODEPICTURE nvEncEncodePicture PNVENCLOCKBITSTREAM nvEncLockBitstream PNVENCUNLOCKBITSTREAM nvEncUnlockBitstream PNVENCLOCKINPUTBUFFER nvEncLockInputBuffer PNVENCUNLOCKINPUTBUFFER nvEncUnlockInputBuffer PNVENCGETENCODESTATS nvEncGetEncodeStats PNVENCGETSEQUENCEPARAMS nvEncGetSequenceParams PNVENCREGISTERASYNCEVENT nvEncRegisterAsyncEvent PNVENCUNREGISTERASYNCEVENT nvEncUnregisterAsyncEvent PNVENCMAPINPUTRESOURCE nvEncMapInputResource PNVENCUNMAPINPUTRESOURCE nvEncUnmapInputResource PNVENCDESTROYENCODER nvEncDestroyEncoder PNVENCINVALIDATEREFFRAMES nvEncInvalidateRefFrames PNVENCOPENENCODESESSIONEX nvEncOpenEncodeSessionEx PNVENCREGISTERRESOURCE nvEncRegisterResource PNVENCUNREGISTERRESOURCE nvEncUnregisterResource PNVENCRECONFIGUREENCODER nvEncReconfigureEncoder void* reserved2[285] #[in]: Reserved and must be set to NULL #constants: unsigned int NVENCAPI_MAJOR_VERSION unsigned int NVENCAPI_MINOR_VERSION uint32_t NVENCAPI_VERSION unsigned int NV_ENCODE_API_FUNCTION_LIST_VER unsigned int NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS_VER unsigned int NV_ENC_INITIALIZE_PARAMS_VER unsigned int NV_ENC_PRESET_CONFIG_VER unsigned int NV_ENC_CONFIG_VER unsigned int NV_ENC_CREATE_INPUT_BUFFER_VER unsigned int NV_ENC_CREATE_BITSTREAM_BUFFER_VER unsigned int NV_ENC_CAPS_PARAM_VER unsigned int NV_ENC_LOCK_INPUT_BUFFER_VER unsigned int NV_ENC_LOCK_BITSTREAM_VER unsigned int NV_ENC_PIC_PARAMS_VER unsigned int NV_ENC_RC_PARAMS_VER unsigned int NV_ENC_REGISTER_RESOURCE_VER unsigned int NV_ENC_MAP_INPUT_RESOURCE_VER unsigned int NVENC_INFINITE_GOPLENGTH unsigned int NV_ENC_SUCCESS unsigned int NV_ENC_ERR_NO_ENCODE_DEVICE unsigned int NV_ENC_ERR_UNSUPPORTED_DEVICE unsigned int NV_ENC_ERR_INVALID_ENCODERDEVICE unsigned int NV_ENC_ERR_INVALID_DEVICE unsigned int NV_ENC_ERR_DEVICE_NOT_EXIST unsigned int NV_ENC_ERR_INVALID_PTR unsigned int NV_ENC_ERR_INVALID_EVENT unsigned int NV_ENC_ERR_INVALID_PARAM unsigned int NV_ENC_ERR_INVALID_CALL unsigned int NV_ENC_ERR_OUT_OF_MEMORY unsigned int NV_ENC_ERR_ENCODER_NOT_INITIALIZED unsigned int NV_ENC_ERR_UNSUPPORTED_PARAM unsigned int NV_ENC_ERR_LOCK_BUSY unsigned int NV_ENC_ERR_NOT_ENOUGH_BUFFER unsigned int NV_ENC_ERR_INVALID_VERSION unsigned int NV_ENC_ERR_MAP_FAILED unsigned int NV_ENC_ERR_NEED_MORE_INPUT unsigned int NV_ENC_ERR_ENCODER_BUSY unsigned int NV_ENC_ERR_EVENT_NOT_REGISTERD unsigned int NV_ENC_ERR_GENERIC unsigned int NV_ENC_ERR_INCOMPATIBLE_CLIENT_KEY unsigned int NV_ENC_ERR_UNIMPLEMENTED unsigned int NV_ENC_ERR_RESOURCE_REGISTER_FAILED unsigned int NV_ENC_ERR_RESOURCE_NOT_REGISTERED unsigned int NV_ENC_ERR_RESOURCE_NOT_MAPPED unsigned int NV_ENC_CAPS_MB_PER_SEC_MAX unsigned int NV_ENC_RECONFIGURE_PARAMS_VER NV_ENC_STATUS_TXT = { NV_ENC_SUCCESS : "This indicates that API call returned with no errors.", NV_ENC_ERR_NO_ENCODE_DEVICE : "This indicates that no encode capable devices were detected", NV_ENC_ERR_UNSUPPORTED_DEVICE : "This indicates that devices pass by the client is not supported.", NV_ENC_ERR_INVALID_ENCODERDEVICE : "This indicates that the encoder device supplied by the client is not valid.", NV_ENC_ERR_INVALID_DEVICE : "This indicates that device passed to the API call is invalid.", NV_ENC_ERR_DEVICE_NOT_EXIST : """This indicates that device passed to the API call is no longer available and needs to be reinitialized. The clients need to destroy the current encoder session by freeing the allocated input output buffers and destroying the device and create a new encoding session.""", NV_ENC_ERR_INVALID_PTR : "This indicates that one or more of the pointers passed to the API call is invalid.", NV_ENC_ERR_INVALID_EVENT : "This indicates that completion event passed in ::NvEncEncodePicture() call is invalid.", NV_ENC_ERR_INVALID_PARAM : "This indicates that one or more of the parameter passed to the API call is invalid.", NV_ENC_ERR_INVALID_CALL : "This indicates that an API call was made in wrong sequence/order.", NV_ENC_ERR_OUT_OF_MEMORY : "This indicates that the API call failed because it was unable to allocate enough memory to perform the requested operation.", NV_ENC_ERR_ENCODER_NOT_INITIALIZED: """This indicates that the encoder has not been initialized with ::NvEncInitializeEncoder() or that initialization has failed. The client cannot allocate input or output buffers or do any encoding related operation before successfully initializing the encoder.""", NV_ENC_ERR_UNSUPPORTED_PARAM : "This indicates that an unsupported parameter was passed by the client.", NV_ENC_ERR_LOCK_BUSY : """This indicates that the ::NvEncLockBitstream() failed to lock the output buffer. This happens when the client makes a non blocking lock call to access the output bitstream by passing NV_ENC_LOCK_BITSTREAM::doNotWait flag. This is not a fatal error and client should retry the same operation after few milliseconds.""", NV_ENC_ERR_NOT_ENOUGH_BUFFER : "This indicates that the size of the user buffer passed by the client is insufficient for the requested operation.", NV_ENC_ERR_INVALID_VERSION : "This indicates that an invalid struct version was used by the client.", NV_ENC_ERR_MAP_FAILED : "This indicates that ::NvEncMapInputResource() API failed to map the client provided input resource.", NV_ENC_ERR_NEED_MORE_INPUT : """ This indicates encode driver requires more input buffers to produce an output bitstream. If this error is returned from ::NvEncEncodePicture() API, this is not a fatal error. If the client is encoding with B frames then, ::NvEncEncodePicture() API might be buffering the input frame for re-ordering. A client operating in synchronous mode cannot call ::NvEncLockBitstream() API on the output bitstream buffer if ::NvEncEncodePicture() returned the ::NV_ENC_ERR_NEED_MORE_INPUT error code. The client must continue providing input frames until encode driver returns ::NV_ENC_SUCCESS. After receiving ::NV_ENC_SUCCESS status the client can call ::NvEncLockBitstream() API on the output buffers in the same order in which it has called ::NvEncEncodePicture(). """, NV_ENC_ERR_ENCODER_BUSY : """This indicates that the HW encoder is busy encoding and is unable to encode the input. The client should call ::NvEncEncodePicture() again after few milliseconds.""", NV_ENC_ERR_EVENT_NOT_REGISTERD : """This indicates that the completion event passed in ::NvEncEncodePicture() API has not been registered with encoder driver using ::NvEncRegisterAsyncEvent().""", NV_ENC_ERR_GENERIC : "This indicates that an unknown internal error has occurred.", NV_ENC_ERR_INCOMPATIBLE_CLIENT_KEY : "This indicates that the client is attempting to use a feature that is not available for the license type for the current system.", NV_ENC_ERR_UNIMPLEMENTED : "This indicates that the client is attempting to use a feature that is not implemented for the current version.", NV_ENC_ERR_RESOURCE_REGISTER_FAILED : "This indicates that the ::NvEncRegisterResource API failed to register the resource.", NV_ENC_ERR_RESOURCE_NOT_REGISTERED : "This indicates that the client is attempting to unregister a resource that has not been successfuly registered.", NV_ENC_ERR_RESOURCE_NOT_MAPPED : "This indicates that the client is attempting to unmap a resource that has not been successfuly mapped.", } CAPS_NAMES = { NV_ENC_CAPS_NUM_MAX_BFRAMES : "NUM_MAX_BFRAMES", NV_ENC_CAPS_SUPPORTED_RATECONTROL_MODES : "SUPPORTED_RATECONTROL_MODES", NV_ENC_CAPS_SUPPORT_FIELD_ENCODING : "SUPPORT_FIELD_ENCODING", NV_ENC_CAPS_SUPPORT_MONOCHROME : "SUPPORT_MONOCHROME", NV_ENC_CAPS_SUPPORT_FMO : "SUPPORT_FMO", NV_ENC_CAPS_SUPPORT_QPELMV : "SUPPORT_QPELMV", NV_ENC_CAPS_SUPPORT_BDIRECT_MODE : "SUPPORT_BDIRECT_MODE", NV_ENC_CAPS_SUPPORT_CABAC : "SUPPORT_CABAC", NV_ENC_CAPS_SUPPORT_ADAPTIVE_TRANSFORM : "SUPPORT_ADAPTIVE_TRANSFORM", NV_ENC_CAPS_NUM_MAX_TEMPORAL_LAYERS : "NUM_MAX_TEMPORAL_LAYERS", NV_ENC_CAPS_SUPPORT_HIERARCHICAL_PFRAMES: "SUPPORT_HIERARCHICAL_PFRAMES", NV_ENC_CAPS_SUPPORT_HIERARCHICAL_BFRAMES: "SUPPORT_HIERARCHICAL_BFRAMES", NV_ENC_CAPS_LEVEL_MAX : "LEVEL_MAX", NV_ENC_CAPS_LEVEL_MIN : "LEVEL_MIN", NV_ENC_CAPS_SEPARATE_COLOUR_PLANE : "SEPARATE_COLOUR_PLANE", NV_ENC_CAPS_WIDTH_MAX : "WIDTH_MAX", NV_ENC_CAPS_HEIGHT_MAX : "HEIGHT_MAX", NV_ENC_CAPS_SUPPORT_TEMPORAL_SVC : "SUPPORT_TEMPORAL_SVC", NV_ENC_CAPS_SUPPORT_DYN_RES_CHANGE : "SUPPORT_DYN_RES_CHANGE", NV_ENC_CAPS_SUPPORT_DYN_BITRATE_CHANGE : "SUPPORT_DYN_BITRATE_CHANGE", NV_ENC_CAPS_SUPPORT_DYN_FORCE_CONSTQP : "SUPPORT_DYN_FORCE_CONSTQP", NV_ENC_CAPS_SUPPORT_DYN_RCMODE_CHANGE : "SUPPORT_DYN_RCMODE_CHANGE", NV_ENC_CAPS_SUPPORT_SUBFRAME_READBACK : "SUPPORT_SUBFRAME_READBACK", NV_ENC_CAPS_SUPPORT_CONSTRAINED_ENCODING: "SUPPORT_CONSTRAINED_ENCODING", NV_ENC_CAPS_SUPPORT_INTRA_REFRESH : "SUPPORT_INTRA_REFRESH", NV_ENC_CAPS_SUPPORT_CUSTOM_VBV_BUF_SIZE : "SUPPORT_CUSTOM_VBV_BUF_SIZE", NV_ENC_CAPS_SUPPORT_DYNAMIC_SLICE_MODE : "SUPPORT_DYNAMIC_SLICE_MODE", NV_ENC_CAPS_SUPPORT_REF_PIC_INVALIDATION: "SUPPORT_REF_PIC_INVALIDATION", NV_ENC_CAPS_PREPROC_SUPPORT : "PREPROC_SUPPORT", NV_ENC_CAPS_ASYNC_ENCODE_SUPPORT : "ASYNC_ENCODE_SUPPORT", NV_ENC_CAPS_MB_NUM_MAX : "MB_NUM_MAX", NV_ENC_CAPS_EXPOSED_COUNT : "EXPOSED_COUNT", NV_ENC_CAPS_SUPPORT_YUV444_ENCODE : "SUPPORT_YUV444_ENCODE", NV_ENC_CAPS_SUPPORT_LOSSLESS_ENCODE : "SUPPORT_LOSSLESS_ENCODE", NV_ENC_CAPS_SUPPORT_SAO : "SUPPORT_SAO", NV_ENC_CAPS_SUPPORT_MEONLY_MODE : "SUPPORT_MEONLY_MODE", NV_ENC_CAPS_SUPPORT_LOOKAHEAD : "SUPPORT_LOOKAHEAD", NV_ENC_CAPS_SUPPORT_TEMPORAL_AQ : "SUPPORT_TEMPORAL_AQ", NV_ENC_CAPS_SUPPORT_10BIT_ENCODE : "SUPPORT_10BIT_ENCODE", NV_ENC_CAPS_NUM_MAX_LTR_FRAMES : "NUM_MAX_LTR_FRAMES", NV_ENC_CAPS_SUPPORT_WEIGHTED_PREDICTION : "SUPPORT_WEIGHTED_PREDICTION", NV_ENC_CAPS_DYNAMIC_QUERY_ENCODER_CAPACITY : "DYNAMIC_QUERY_ENCODER_CAPACITY", NV_ENC_CAPS_SUPPORT_BFRAME_REF_MODE : "SUPPORT_BFRAME_REF_MODE", NV_ENC_CAPS_SUPPORT_EMPHASIS_LEVEL_MAP : "SUPPORT_EMPHASIS_LEVEL_MAP", } PIC_TYPES = { NV_ENC_PIC_TYPE_P : "P", NV_ENC_PIC_TYPE_B : "B", NV_ENC_PIC_TYPE_I : "I", NV_ENC_PIC_TYPE_IDR : "IDR", NV_ENC_PIC_TYPE_BI : "BI", NV_ENC_PIC_TYPE_SKIPPED : "SKIPPED", NV_ENC_PIC_TYPE_INTRA_REFRESH : "INTRA_REFRESH", NV_ENC_PIC_TYPE_UNKNOWN : "UNKNOWN", } NvEncodeAPICreateInstance = None cuCtxGetCurrent = None def init_nvencode_library(): global NvEncodeAPICreateInstance, cuCtxGetCurrent if WIN32: load = ctypes.WinDLL nvenc_libname = "nvencodeapi64.dll" cuda_libname = "nvcuda.dll" else: #assert os.name=="posix" load = ctypes.cdll.LoadLibrary nvenc_libname = "libnvidia-encode.so" cuda_libname = "libcuda.so" #CUDA: log("init_nvencode_library() will try to load %s", cuda_libname) try: x = load(cuda_libname) log("init_nvencode_library() %s(%s)=%s", load, cuda_libname, x) except Exception as e: log("failed to load '%s'", cuda_libname, exc_info=True) raise ImportError("nvenc: the required library %s cannot be loaded: %s" % (cuda_libname, e)) cuCtxGetCurrent = x.cuCtxGetCurrent cuCtxGetCurrent.restype = ctypes.c_int # CUresult == int cuCtxGetCurrent.argtypes = [POINTER(CUcontext)] # CUcontext *pctx log("init_nvencode_library() %s.cuCtxGetCurrent=%s", os.path.splitext(cuda_libname)[0], cuCtxGetCurrent) #nvidia-encode: log("init_nvencode_library() will try to load %s", nvenc_libname) try: x = load(nvenc_libname) log("init_nvencode_library() %s(%s)=%s", load, nvenc_libname, x) except Exception as e: log("failed to load '%s'", nvenc_libname, exc_info=True) raise ImportError("nvenc: the required library %s cannot be loaded: %s" % (nvenc_libname, e)) NvEncodeAPICreateInstance = x.NvEncodeAPICreateInstance NvEncodeAPICreateInstance.restype = ctypes.c_int NvEncodeAPICreateInstance.argtypes = [ctypes.c_void_p] log("init_nvencode_library() NvEncodeAPICreateInstance=%s", NvEncodeAPICreateInstance) #NVENCSTATUS NvEncodeAPICreateInstance(NV_ENCODE_API_FUNCTION_LIST *functionList) def atob(a): if PYTHON3: return a.tobytes() return a.tostring() cdef guidstr(GUID guid): #really ugly! (surely there's a way using struct.unpack ?) #is this even endian safe? do we care? (always on the same system) parts = [] for v, s in ((guid.Data1, 4), (guid.Data2, 2), (guid.Data3, 2)): b = array.array('B', [0 for _ in range(s)]) for j in range(s): b[s-j-1] = v % 256 v = v // 256 parts.append(atob(b)) parts.append(atob(array.array('B', guid.get("Data4")[:2]))) parts.append(atob(array.array('B', guid.get("Data4")[2:8]))) s = b"-".join(binascii.hexlify(b).upper() for b in parts) #log.info("guidstr(%s)=%s", guid, s) return bytestostr(s) cdef GUID c_parseguid(src) except *: #just as ugly as above - shoot me now #only this format is allowed: sample_guid = "CE788D20-AAA9-4318-92BB-AC7E858C8D36" if len(src)!=len(sample_guid): raise Exception("invalid GUID format: expected %s characters but got %s" % (len(sample_guid), len(src))) cdef int i, s src = bytestostr(src) for i in range(len(sample_guid)): if sample_guid[i]=="-": #dash must be in the same place: if src[i]!="-": raise Exception("invalid GUID format: character at position %s is not '-': %s" % (i, src[i])) else: #must be an hex number: c = src.upper()[i] if c not in ("0123456789ABCDEF"): raise Exception("invalid GUID format: character at position %s is not in hex: %s" % (i, c)) parts = src.split("-") #ie: ["CE788D20", "AAA9", ...] nparts = [] for i, s in (0, 4), (1, 2), (2, 2), (3, 2), (4, 6): b = atob(array.array('B', binascii.unhexlify(strtobytes(parts[i])))) v = 0 for j in range(s): c = b[j] if not PYTHON3: c = ord(c) v += c<<((s-j-1)*8) nparts.append(v) cdef GUID guid guid.Data1 = nparts[0] guid.Data2 = nparts[1] guid.Data3 = nparts[2] v = (nparts[3]<<48) + nparts[4] for i in range(8): guid.Data4[i] = (v>>((7-i)*8)) % 256 return guid def parseguid(s): return c_parseguid(s) def test_parse(): sample_guid = u"CE788D20-AAA9-4318-92BB-AC7E858C8D36" x = c_parseguid(sample_guid) v = guidstr(x) assert v==sample_guid, "expected %s but got %s" % (sample_guid, v) test_parse() cdef GUID CLIENT_KEY_GUID memset(&CLIENT_KEY_GUID, 0, sizeof(GUID)) if CLIENT_KEYS_STR: #if we have client keys, parse them and keep the ones that look valid validated = [] for x in CLIENT_KEYS_STR: if x: try: CLIENT_KEY_GUID = c_parseguid(x) validated.append(x) except Exception as e: log.error("invalid nvenc client key specified: '%s' (%s)", x, e) del e CLIENT_KEYS_STR = validated CODEC_GUIDS = { guidstr(NV_ENC_CODEC_H264_GUID) : "H264", guidstr(NV_ENC_CODEC_HEVC_GUID) : "HEVC", } cdef codecstr(GUID guid): s = guidstr(guid) return CODEC_GUIDS.get(s, s) CODEC_PROFILES_GUIDS = { guidstr(NV_ENC_CODEC_H264_GUID) : { guidstr(NV_ENC_CODEC_PROFILE_AUTOSELECT_GUID) : "auto", guidstr(NV_ENC_H264_PROFILE_BASELINE_GUID) : "baseline", guidstr(NV_ENC_H264_PROFILE_MAIN_GUID) : "main", guidstr(NV_ENC_H264_PROFILE_HIGH_GUID) : "high", guidstr(NV_ENC_H264_PROFILE_STEREO_GUID) : "stereo", guidstr(NV_ENC_H264_PROFILE_SVC_TEMPORAL_SCALABILTY): "temporal", guidstr(NV_ENC_H264_PROFILE_PROGRESSIVE_HIGH_GUID) : "progressive-high", guidstr(NV_ENC_H264_PROFILE_CONSTRAINED_HIGH_GUID) : "constrained-high", #new in SDK v4: guidstr(NV_ENC_H264_PROFILE_HIGH_444_GUID) : "high-444", }, guidstr(NV_ENC_CODEC_HEVC_GUID) : { guidstr(NV_ENC_HEVC_PROFILE_MAIN_GUID) : "main", guidstr(NV_ENC_HEVC_PROFILE_MAIN10_GUID) : "main10", guidstr(NV_ENC_HEVC_PROFILE_FREXT_GUID) : "frext", }, } #this one is not defined anywhere but in the OBS source #(I think they have access to information we do not have): #GUID NV_ENC_PRESET_STREAMING = c_parseguid("7ADD423D-D035-4F6F-AEA5-50885658643C") CODEC_PRESETS_GUIDS = { guidstr(NV_ENC_PRESET_DEFAULT_GUID) : "default", guidstr(NV_ENC_PRESET_HP_GUID) : "hp", guidstr(NV_ENC_PRESET_HQ_GUID) : "hq", guidstr(NV_ENC_PRESET_BD_GUID) : "bd", guidstr(NV_ENC_PRESET_LOW_LATENCY_DEFAULT_GUID) : "low-latency", guidstr(NV_ENC_PRESET_LOW_LATENCY_HQ_GUID) : "low-latency-hq", guidstr(NV_ENC_PRESET_LOW_LATENCY_HP_GUID) : "low-latency-hp", #new in SDK4: guidstr(NV_ENC_PRESET_LOSSLESS_DEFAULT_GUID) : "lossless", guidstr(NV_ENC_PRESET_LOSSLESS_HP_GUID) : "lossless-hp", "7ADD423D-D035-4F6F-AEA5-50885658643C" : "streaming", } YUV444_PRESETS = ("high-444", "lossless", "lossless-hp",) LOSSLESS_PRESETS = ("lossless", "lossless-hp",) cdef presetstr(GUID preset): s = guidstr(preset) return CODEC_PRESETS_GUIDS.get(s, s) #try to map preset names to a "speed" value: PRESET_SPEED = { "lossless" : 0, "lossless-hp" : 30, "bd" : 40, "hq" : 50, "default" : 50, "hp" : 60, "low-latency-hq": 70, "low-latency" : 80, "low-latency-hp": 100, "streaming" : -1000, #disabled for now } PRESET_QUALITY = { "lossless" : 100, "lossless-hp" : 100, "bd" : 80, "hq" : 70, "low-latency-hq": 60, "default" : 50, "hp" : 40, "low-latency" : 20, "low-latency-hp": 0, "streaming" : -1000, #disabled for now } BUFFER_FORMAT = { NV_ENC_BUFFER_FORMAT_UNDEFINED : "undefined", NV_ENC_BUFFER_FORMAT_NV12 : "NV12_PL", NV_ENC_BUFFER_FORMAT_YV12 : "YV12_PL", NV_ENC_BUFFER_FORMAT_IYUV : "IYUV_PL", NV_ENC_BUFFER_FORMAT_YUV444 : "YUV444_PL", NV_ENC_BUFFER_FORMAT_YUV420_10BIT : "YUV420_10BIT", NV_ENC_BUFFER_FORMAT_YUV444_10BIT : "YUV444_10BIT", NV_ENC_BUFFER_FORMAT_ARGB : "ARGB", NV_ENC_BUFFER_FORMAT_ARGB10 : "ARGB10", NV_ENC_BUFFER_FORMAT_AYUV : "AYUV", NV_ENC_BUFFER_FORMAT_ABGR : "ABGR", NV_ENC_BUFFER_FORMAT_ABGR10 : "ABGR10", } def get_COLORSPACES(encoding): global YUV420_ENABLED, YUV444_ENABLED, YUV444_CODEC_SUPPORT out_cs = [] if YUV420_ENABLED: out_cs.append("YUV420P") if YUV444_CODEC_SUPPORT.get(encoding.lower(), YUV444_ENABLED) or NATIVE_RGB: out_cs.append("YUV444P") COLORSPACES = { "BGRX" : out_cs, "BGRA" : out_cs, "XRGB" : out_cs, "ARGB" : out_cs, } return COLORSPACES def get_input_colorspaces(encoding): return list(get_COLORSPACES(encoding).keys()) def get_output_colorspaces(encoding, input_colorspace): cs = get_COLORSPACES(encoding) out = cs.get(input_colorspace) assert out, "invalid input colorspace %s for encoding %s (must be one of: %s)" % (input_colorspace, encoding, out) #the output will actually be in one of those two formats once decoded #because internally that's what we convert to before encoding #(well, NV12... which is equivallent to YUV420P here...) return out WIDTH_MASK = 0xFFFE HEIGHT_MASK = 0xFFFE #Note: these counters should be per-device, but #when we call get_runtime_factor(), we don't know which device is going to get used! #since we have load balancing, using an overall factor isn't too bad context_counter = AtomicInteger() context_gen_counter = AtomicInteger() cdef double last_context_failure = 0 def get_runtime_factor(): global last_context_failure, context_counter device_count = len(init_all_devices()) max_contexts = 32 * device_count cc = context_counter.get() #try to avoid using too many contexts #(usually, we can have up to 32 contexts per card) low_limit = 16 f = max(0, 1.0 - (max(0, cc-low_limit)/max(1, max_contexts-low_limit))) #if we have had errors recently, lower our chances further: cdef double failure_elapsed = monotonic_time()-last_context_failure #discount factor gradually for 1 minute: f /= 61-min(60, failure_elapsed) log("nvenc.get_runtime_factor()=%s", f) return f MAX_SIZE = {} def get_spec(encoding, colorspace): assert encoding in get_encodings(), "invalid format: %s (must be one of %s" % (encoding, get_encodings()) assert colorspace in get_COLORSPACES(encoding), "invalid colorspace: %s (must be one of %s)" % (colorspace, get_COLORSPACES(encoding)) #ratings: quality, speed, setup cost, cpu cost, gpu cost, latency, max_w, max_h #undocumented and found the hard way, see: #https://www.xpra.org/trac/ticket/1046#comment:6 #https://xpra.org/trac/ticket/1550#comment:13 min_w, min_h = 128, 128 #FIXME: we should probe this using WIDTH_MAX, HEIGHT_MAX! global MAX_SIZE max_w, max_h = MAX_SIZE.get(encoding, (4096, 4096)) has_lossless_mode = colorspace in ("XRGB", "ARGB", "BGRX", "BGRA" ) and encoding=="h264" cs = video_spec(encoding=encoding, output_colorspaces=get_COLORSPACES(encoding)[colorspace], has_lossless_mode=LOSSLESS_CODEC_SUPPORT.get(encoding, LOSSLESS_ENABLED), codec_class=Encoder, codec_type=get_type(), quality=60+has_lossless_mode*40, speed=100, setup_cost=80, cpu_cost=10, gpu_cost=100, #using a hardware encoder for something this small is silly: min_w=min_w, min_h=min_h, max_w=max_w, max_h=max_h, can_scale=True, width_mask=WIDTH_MASK, height_mask=HEIGHT_MASK) cs.get_runtime_factor = get_runtime_factor return cs #ie: NVENCAPI_VERSION=0x30 -> PRETTY_VERSION = [3, 0] PRETTY_VERSION = (int(NVENCAPI_MAJOR_VERSION), int(NVENCAPI_MINOR_VERSION)) def get_version(): return ".".join((str(x) for x in PRETTY_VERSION)) def get_type(): return "nvenc" def get_info(): global last_context_failure, context_counter, context_gen_counter info = { "version" : PRETTY_VERSION, "device_count" : len(get_devices() or []), "context_count" : context_counter.get(), "generation" : context_gen_counter.get(), } cards = get_cards() if cards: info["cards"] = cards #only show the version if we have it already (don't probe now) v = get_nvidia_module_version(False) if v: info["kernel_module_version"] = v if LINUX: from xpra.scripts.config import python_platform info["kernel_version"] = python_platform.uname()[2] if last_context_failure>0: info["last_failure"] = int(monotonic_time()-last_context_failure) return info ENCODINGS = [] def get_encodings(): global ENCODINGS return ENCODINGS cdef inline int roundup(int n, int m): return (n + m - 1) & ~(m - 1) cdef uintptr_t cmalloc(size_t size, what) except 0: cdef void *ptr = malloc(size) if ptr==NULL: raise Exception("failed to allocate %i bytes of memory for %s" % (size, what)) return ptr cdef nvencStatusInfo(NVENCSTATUS ret): if ret in NV_ENC_STATUS_TXT: return "%s: %s" % (ret, NV_ENC_STATUS_TXT[ret]) return str(ret) class NVENCException(Exception): def __init__(self, code, fn): self.function = fn self.code = code msg = "%s - returned %s" % (fn, nvencStatusInfo(code)) Exception.__init__(self, msg) cdef inline raiseNVENC(NVENCSTATUS ret, msg): if DEBUG_API: log("raiseNVENC(%i, %s)", ret, msg) if ret!=0: raise NVENCException(ret, msg) cdef class Encoder: cdef unsigned int width cdef unsigned int height cdef unsigned int input_width cdef unsigned int input_height cdef unsigned int encoder_width cdef unsigned int encoder_height cdef unsigned int b_frames cdef object encoding cdef object src_format cdef object dst_formats cdef object scaling cdef int speed cdef int quality cdef uint32_t target_bitrate cdef uint32_t max_bitrate #PyCUDA: cdef object driver cdef int cuda_device_id cdef object cuda_info cdef object pycuda_info cdef object cuda_device_info cdef object cuda_device cdef object cuda_context cdef void *cuda_context_ptr cdef object kernel cdef object kernel_name cdef object max_block_sizes cdef object max_grid_sizes cdef unsigned long max_threads_per_block cdef uint64_t free_memory cdef uint64_t total_memory #NVENC: cdef NV_ENCODE_API_FUNCTION_LIST *functionList #@DuplicatedSignature cdef void *context cdef GUID codec cdef NV_ENC_REGISTERED_PTR inputHandle cdef object inputBuffer cdef object cudaInputBuffer cdef object cudaOutputBuffer cdef unsigned int inputPitch #note: this isn't the pitch (aka rowstride) we actually use! #just the value returned from the allocation call cdef unsigned int outputPitch cdef void *bitstreamBuffer cdef NV_ENC_BUFFER_FORMAT bufferFmt cdef object codec_name cdef object preset_name cdef object pixel_format cdef uint8_t lossless #statistics, etc: cdef double time cdef uint64_t first_frame_timestamp cdef unsigned long frames cdef unsigned long index cdef object last_frame_times cdef uint64_t bytes_in cdef uint64_t bytes_out cdef object __weakref__ cdef GUID init_codec(self) except *: codecs = self.query_codecs() #codecs={'H264': {"guid" : '6BC82762-4E63-4CA4-AA85-1E50F321F6BF', .. } internal_name = {"H265" : "HEVC"}.get(self.codec_name.upper(), self.codec_name.upper()) guid_str = codecs.get(internal_name, {}).get("guid") assert guid_str, "%s not supported! (only available: %s)" % (self.codec_name, csv(codecs.keys())) self.codec = c_parseguid(guid_str) return self.codec cdef GUID get_codec(self): return self.codec cdef GUID get_preset(self, GUID codec) except *: presets = self.query_presets(codec) options = {} #if a preset was specified, give it the best score possible (-1): if DESIRED_PRESET: options[-1] = DESIRED_PRESET #add all presets ranked by how far they are from the target speed and quality: log("presets for %s: %s (pixel format=%s)", guidstr(codec), csv(presets.keys()), self.pixel_format) for name, x in presets.items(): preset_speed = PRESET_SPEED.get(name, 50) preset_quality = PRESET_QUALITY.get(name, 50) is_lossless = name in LOSSLESS_PRESETS log("preset %16s: speed=%5i, quality=%5i (lossless=%s - want lossless=%s)", name, preset_speed, preset_quality, is_lossless, bool(self.lossless)) if is_lossless and self.pixel_format!="YUV444P": continue if preset_speed>=0 and preset_quality>=0: #quality (3) weighs more than speed (2): v = 2 * abs(preset_speed-self.speed) + 3 * abs(preset_quality-self.quality) if self.lossless!=is_lossless: v -= 100 l = options.setdefault(v, []) if x not in l: l.append((name, x)) log("get_preset(%s) speed=%s, quality=%s, lossless=%s, pixel_format=%s, options=%s", codecstr(codec), self.speed, self.quality, bool(self.lossless), self.pixel_format, options) for score in sorted(options.keys()): for preset, preset_guid in options.get(score): if preset and (preset in presets.keys()): log("using preset '%s' for speed=%s, quality=%s, lossless=%s, pixel_format=%s", preset, self.speed, self.quality, self.lossless, self.pixel_format) return c_parseguid(preset_guid) raise Exception("no matching presets available for '%s' with speed=%i and quality=%i" % (self.codec_name, self.speed, self.quality)) def init_context(self, int width, int height, src_format, dst_formats, encoding, int quality, int speed, scaling, options={}): #@DuplicatedSignature assert NvEncodeAPICreateInstance is not None, "encoder module is not initialized" log("init_context%s", (width, height, src_format, dst_formats, encoding, quality, speed, scaling, options)) assert src_format in ("ARGB", "XRGB", "BGRA", "BGRX"), "invalid source format %s" % src_format assert "YUV420P" in dst_formats or "YUV444P" in dst_formats self.width = width self.height = height self.speed = speed self.quality = quality self.scaling = scaling or (1, 1) v, u = self.scaling self.input_width = roundup(width, 32) self.input_height = roundup(height, 32) self.encoder_width = roundup(width*v//u, 32) self.encoder_height = roundup(height*v//u, 32) self.src_format = src_format self.dst_formats = dst_formats self.encoding = encoding self.codec_name = encoding.upper() #ie: "H264" self.preset_name = None self.frames = 0 self.cuda_device = None self.cuda_context = None self.pixel_format = "" self.last_frame_times = deque(maxlen=200) self.update_bitrate() cdef double start = monotonic_time() #the pixel format we feed into the encoder self.pixel_format = self.get_target_pixel_format(self.quality) self.lossless = self.get_target_lossless(self.pixel_format, self.quality) self.select_cuda_device(options) log("using %s %s compression at %s%% quality with pixel format %s", ["lossy","lossless"][self.lossless], encoding, self.quality, self.pixel_format) try: self.init_cuda() try: self.init_cuda_kernel() finally: self.cuda_context.pop() #the example code accesses the cuda context after a context.pop() #(which is weird) self.init_nvenc() record_device_success(self.cuda_device_id) except Exception as e: log("init_cuda failed", exc_info=True) record_device_failure(self.cuda_device_id) raise cdef double end = monotonic_time() log("init_context%s took %1.fms", (width, height, src_format, quality, speed, options), (end-start)*1000.0) def select_cuda_device(self, options={}): self.cuda_device_id, self.cuda_device = select_device(options.get("cuda_device", -1), min_compute=MIN_COMPUTE) assert self.cuda_device_id>=0 and self.cuda_device, "failed to select a cuda device" log("selected device %s", device_info(self.cuda_device)) def get_target_pixel_format(self, quality): global NATIVE_RGB, YUV420_ENABLED, YUV444_ENABLED, LOSSLESS_ENABLED, YUV444_THRESHOLD, YUV444_CODEC_SUPPORT v = None hasyuv444 = YUV444_CODEC_SUPPORT.get(self.encoding, YUV444_ENABLED) and "YUV444P" in self.dst_formats nativergb = NATIVE_RGB and hasyuv444 if nativergb and self.src_format in ("BGRX", "BGRA"): v = "BGRX" else: x,y = self.scaling hasyuv420 = YUV420_ENABLED and "YUV420P" in self.dst_formats if hasyuv444: #NVENC and the client can handle it, #now check quality and scaling: #(don't use YUV444 is we're going to downscale or use low quality anyway) if (quality>=YUV444_THRESHOLD and x==1 and y==1) or not hasyuv420: v = "YUV444P" if not v: if hasyuv420: v = "NV12" else: raise Exception("no compatible formats found for quality=%i, scaling=%s, YUV420 support=%s, YUV444 support=%s, codec=%s, dst-formats=%s" % (quality, self.scaling, hasyuv420, hasyuv444, self.codec_name, self.dst_formats)) log("get_target_pixel_format(%i)=%s for encoding=%s, scaling=%s, NATIVE_RGB=%s, YUV444_CODEC_SUPPORT=%s, YUV420_ENABLED=%s, YUV444_ENABLED=%s, YUV444_THRESHOLD=%s, LOSSLESS_ENABLED=%s, src_format=%s, dst_formats=%s", quality, v, self.encoding, self.scaling, bool(NATIVE_RGB), YUV444_CODEC_SUPPORT, bool(YUV420_ENABLED), bool(YUV444_ENABLED), YUV444_THRESHOLD, bool(LOSSLESS_ENABLED), self.src_format, csv(self.dst_formats)) return v def get_target_lossless(self, pixel_format, quality): global LOSSLESS_ENABLED, LOSSLESS_CODEC_SUPPORT if pixel_format!="YUV444P": return False if not LOSSLESS_CODEC_SUPPORT.get(self.encoding, LOSSLESS_ENABLED): return False return quality>=LOSSLESS_THRESHOLD def init_cuda(self): cdef unsigned int plane_size_div cdef unsigned int max_input_stride cdef int result cdef uintptr_t context_pointer assert self.cuda_device_id>=0 and self.cuda_device, "no NVENC device found!" global last_context_failure d = self.cuda_device cf = driver.ctx_flags log("init_cuda() pixel format=%s, device_id=%s", self.pixel_format, self.cuda_device_id) try: log("init_cuda cuda_device=%s (%s)", d, device_info(d)) self.cuda_context = d.make_context(flags=cf.SCHED_AUTO | cf.MAP_HOST) assert self.cuda_context, "failed to create a CUDA context for device %s" % device_info(d) log("init_cuda cuda_context=%s", self.cuda_context) self.cuda_info = get_cuda_info() log("init_cuda cuda info=%s", self.cuda_info) self.pycuda_info = get_pycuda_info() log("init_cuda pycuda info=%s", self.pycuda_info) self.cuda_device_info = { "device" : { "name" : d.name(), "pci_bus_id" : d.pci_bus_id(), "memory" : int(d.total_memory()//1024//1024), }, "api_version" : self.cuda_context.get_api_version(), } #get the CUDA context (C pointer): #a bit of magic to pass a cython pointer to ctypes: context_pointer = (&self.cuda_context_ptr) result = cuCtxGetCurrent(ctypes.cast(context_pointer, POINTER(ctypes.c_void_p))) if DEBUG_API: log("cuCtxGetCurrent() returned %s, cuda context pointer=%#x", CUDA_ERRORS_INFO.get(result, result), self.cuda_context_ptr) assert result==0, "failed to get current cuda context, cuCtxGetCurrent returned %s" % CUDA_ERRORS_INFO.get(result, result) except driver.MemoryError as e: last_context_failure = monotonic_time() log("init_cuda %s", e) raise TransientCodecException("could not initialize cuda: %s" % e) cdef init_cuda_kernel(self): global YUV420_ENABLED, YUV444_ENABLED, YUV444_CODEC_SUPPORT, NATIVE_RGB assert self.cuda_device_id>=0 and self.cuda_device, "cuda device not set!" cdef unsigned int plane_size_div, wmult, hmult, max_input_stride d = self.cuda_device #use alias to make code easier to read: da = driver.device_attribute if self.pixel_format=="BGRX": assert NATIVE_RGB kernel_name = None self.bufferFmt = NV_ENC_BUFFER_FORMAT_ARGB plane_size_div= 1 wmult = 4 hmult = 1 #if supported (separate plane flag), use YUV444P: elif self.pixel_format=="YUV444P": assert YUV444_CODEC_SUPPORT.get(self.encoding, YUV444_ENABLED), "YUV444 is not enabled for %s" % self.encoding kernel_name = "%s_to_YUV444" % (self.src_format.replace("X", "A")) #ie: ARGB_to_YUV444 self.bufferFmt = NV_ENC_BUFFER_FORMAT_YUV444 #3 full planes: plane_size_div = 1 wmult = 1 hmult = 3 elif self.pixel_format=="NV12": assert YUV420_ENABLED kernel_name = "%s_to_NV12" % (self.src_format.replace("X", "A")) #ie: BGRA_to_NV12 self.bufferFmt = NV_ENC_BUFFER_FORMAT_NV12 #1 full Y plane and 2 U+V planes subsampled by 4: plane_size_div = 2 wmult = 1 hmult = 3 else: raise Exception("BUG: invalid dst format: %s" % self.pixel_format) #allocate CUDA "output" buffer (on device): #this is the buffer we feed into the encoder #the data may come from the CUDA kernel, #or it may be uploaded directly there (ie: BGRX) self.cudaOutputBuffer, self.outputPitch = driver.mem_alloc_pitch(self.encoder_width*wmult, self.encoder_height*hmult//plane_size_div, 16) log("CUDA Output Buffer=%#x, pitch=%s", int(self.cudaOutputBuffer), self.outputPitch) if kernel_name: #load the kernel: self.kernel = get_CUDA_function(self.cuda_device_id, kernel_name) self.kernel_name = kernel_name assert self.kernel, "failed to load %s for device %i" % (self.kernel_name, self.cuda_device_id) #allocate CUDA input buffer (on device) 32-bit RGBX #(and make it bigger just in case - subregions from XShm can have a huge rowstride) #(this is the buffer we feed into the kernel) max_input_stride = MAX(2560, self.input_width)*4 self.cudaInputBuffer, self.inputPitch = driver.mem_alloc_pitch(max_input_stride, self.input_height, 16) log("CUDA Input Buffer=%#x, pitch=%s", int(self.cudaInputBuffer), self.inputPitch) #CUDA self.max_block_sizes = d.get_attribute(da.MAX_BLOCK_DIM_X), d.get_attribute(da.MAX_BLOCK_DIM_Y), d.get_attribute(da.MAX_BLOCK_DIM_Z) self.max_grid_sizes = d.get_attribute(da.MAX_GRID_DIM_X), d.get_attribute(da.MAX_GRID_DIM_Y), d.get_attribute(da.MAX_GRID_DIM_Z) log("max_block_sizes=%s", self.max_block_sizes) log("max_grid_sizes=%s", self.max_grid_sizes) self.max_threads_per_block = self.kernel.get_attribute(driver.function_attribute.MAX_THREADS_PER_BLOCK) log("max_threads_per_block=%s", self.max_threads_per_block) else: #we don't use a CUDA kernel self.kernel_name = None self.kernel = None self.cudaInputBuffer = None self.inputPitch = self.outputPitch self.max_block_sizes = 0 self.max_grid_sizes = 0 self.max_threads_per_block = 0 #allocate input buffer on host: #this is the buffer we upload to the device self.inputBuffer = driver.pagelocked_zeros(self.inputPitch*self.input_height, dtype=numpy.byte) log("inputBuffer=%s (size=%s)", self.inputBuffer, self.inputPitch*self.input_height) def init_nvenc(self): self.open_encode_session() self.init_encoder() self.init_buffers() def init_encoder(self): cdef GUID codec = self.init_codec() cdef NV_ENC_INITIALIZE_PARAMS *params = NULL cdef NVENCSTATUS r params = cmalloc(sizeof(NV_ENC_INITIALIZE_PARAMS), "initialization params") assert memset(params, 0, sizeof(NV_ENC_INITIALIZE_PARAMS))!=NULL try: self.init_params(codec, params) if DEBUG_API: log("nvEncInitializeEncoder using encode=%s", codecstr(codec)) with nogil: r = self.functionList.nvEncInitializeEncoder(self.context, params) raiseNVENC(r, "initializing encoder") log("NVENC initialized with '%s' codec and '%s' preset" % (self.codec_name, self.preset_name)) self.dump_caps(self.codec_name, codec) finally: if params.encodeConfig!=NULL: free(params.encodeConfig) free(params) cdef dump_caps(self, codec_name, GUID codec): #test all caps: caps = {} for cap, descr in CAPS_NAMES.items(): if cap!=NV_ENC_CAPS_EXPOSED_COUNT: v = self.query_encoder_caps(codec, cap) caps[descr] = v log("caps(%s)=%s", codec_name, caps) cdef init_params(self, GUID codec, NV_ENC_INITIALIZE_PARAMS *params): #caller must free the config! cdef GUID preset cdef NV_ENC_CONFIG *config = NULL cdef NV_ENC_PRESET_CONFIG *presetConfig = NULL #@DuplicatedSignature assert self.context, "context is not initialized" preset = self.get_preset(self.codec) self.preset_name = CODEC_PRESETS_GUIDS.get(guidstr(preset), guidstr(preset)) log("init_params(%s) using preset=%s", codecstr(codec), presetstr(preset)) input_format = BUFFER_FORMAT[self.bufferFmt] input_formats = self.query_input_formats(codec) assert input_format in input_formats, "%s does not support %s (only: %s)" % (self.codec_name, input_format, input_formats) assert memset(params, 0, sizeof(NV_ENC_INITIALIZE_PARAMS))!=NULL config = cmalloc(sizeof(NV_ENC_CONFIG), "encoder config") presetConfig = self.get_preset_config(self.preset_name, codec, preset) assert presetConfig!=NULL, "could not find preset %s" % self.preset_name try: assert memcpy(config, &presetConfig.presetCfg, sizeof(NV_ENC_CONFIG))!=NULL params.version = NV_ENC_INITIALIZE_PARAMS_VER params.encodeGUID = codec params.presetGUID = preset params.encodeWidth = self.encoder_width params.encodeHeight = self.encoder_height params.maxEncodeWidth = self.encoder_width params.maxEncodeHeight = self.encoder_height params.darWidth = self.encoder_width params.darHeight = self.encoder_height params.enableEncodeAsync = 0 #not supported on Linux params.enablePTD = 1 #not supported in sync mode!? params.reportSliceOffsets = 0 params.enableSubFrameWrite = 0 params.frameRateNum = 30 params.frameRateDen = 1 params.encodeConfig = config #config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_VBR #FIXME: check NV_ENC_CAPS_SUPPORTED_RATECONTROL_MODES caps #config.rcParams.enableMinQP = 1 #config.rcParams.enableMaxQP = 1 config.gopLength = NVENC_INFINITE_GOPLENGTH config.frameIntervalP = 1 #0=max quality, 63 lowest quality qpmin = QP_MAX_VALUE-min(QP_MAX_VALUE, int(QP_MAX_VALUE*(self.quality)//100)) qpmax = QP_MAX_VALUE-max(0, int(QP_MAX_VALUE*self.quality//100)) config.frameFieldMode = NV_ENC_PARAMS_FRAME_FIELD_MODE_FRAME #config.mvPrecision = NV_ENC_MV_PRECISION_FULL_PEL if True: #const QP: config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_CONSTQP if self.preset_name.find("lossless")>=0: qp = 0 else: qp = min(QP_MAX_VALUE, max(0, (qpmin + qpmax)//2)) config.rcParams.constQP.qpInterP = qp config.rcParams.constQP.qpInterB = qp config.rcParams.constQP.qpIntra = qp else: config.rcParams.rateControlMode = NV_ENC_PARAMS_RC_CBR config.rcParams.averageBitRate = 500000 config.rcParams.maxBitRate = 600000 #config.rcParams.vbvBufferSize = 0 #config.rcParams.vbvInitialDelay = 0 #config.rcParams.enableInitialRCQP = 1 #config.rcParams.initialRCQP.qpInterP = qpmin #config.rcParams.initialRCQP.qpIntra = qpmin #config.rcParams.initialRCQP.qpInterB = qpmin if self.pixel_format=="BGRX": chromaFormatIDC = 3 elif self.pixel_format=="NV12": chromaFormatIDC = 1 elif self.pixel_format=="YUV444P": chromaFormatIDC = 3 else: raise Exception("unknown pixel format %s" % self.pixel_format) if self.codec_name=="H264": config.encodeCodecConfig.h264Config.chromaFormatIDC = chromaFormatIDC #config.encodeCodecConfig.h264Config.h264VUIParameters.colourDescriptionPresentFlag = 0 #config.encodeCodecConfig.h264Config.h264VUIParameters.videoSignalTypePresentFlag = 0 config.encodeCodecConfig.h264Config.idrPeriod = config.gopLength config.encodeCodecConfig.h264Config.enableIntraRefresh = 0 #config.encodeCodecConfig.h264Config.maxNumRefFrames = 16 #config.encodeCodecConfig.h264Config.h264VUIParameters.colourMatrix = 1 #AVCOL_SPC_BT709 ? #config.encodeCodecConfig.h264Config.h264VUIParameters.colourPrimaries = 1 #AVCOL_PRI_BT709 ? #config.encodeCodecConfig.h264Config.h264VUIParameters.transferCharacteristics = 1 #AVCOL_TRC_BT709 ? #config.encodeCodecConfig.h264Config.h264VUIParameters.videoFullRangeFlag = 1 else: assert self.codec_name=="H265" config.encodeCodecConfig.hevcConfig.chromaFormatIDC = chromaFormatIDC #config.encodeCodecConfig.hevcConfig.level = NV_ENC_LEVEL_HEVC_5 config.encodeCodecConfig.hevcConfig.idrPeriod = config.gopLength config.encodeCodecConfig.hevcConfig.enableIntraRefresh = 0 #config.encodeCodecConfig.hevcConfig.maxNumRefFramesInDPB = 16 #config.encodeCodecConfig.hevcConfig.hevcVUIParameters.videoFormat = ... finally: if presetConfig!=NULL: free(presetConfig) def init_buffers(self): cdef NV_ENC_REGISTER_RESOURCE registerResource cdef NV_ENC_CREATE_INPUT_BUFFER createInputBufferParams cdef NV_ENC_CREATE_BITSTREAM_BUFFER createBitstreamBufferParams cdef uintptr_t resource cdef Py_ssize_t size cdef unsigned char* cptr = NULL cdef NVENCSTATUS r # assert self.context, "context is not initialized" #register CUDA input buffer: memset(®isterResource, 0, sizeof(NV_ENC_REGISTER_RESOURCE)) registerResource.version = NV_ENC_REGISTER_RESOURCE_VER registerResource.resourceType = NV_ENC_INPUT_RESOURCE_TYPE_CUDADEVICEPTR resource = int(self.cudaOutputBuffer) registerResource.resourceToRegister = resource registerResource.width = self.encoder_width registerResource.height = self.encoder_height registerResource.pitch = self.outputPitch registerResource.bufferFormat = self.bufferFmt if DEBUG_API: log("nvEncRegisterResource(%#x)", ®isterResource) with nogil: r = self.functionList.nvEncRegisterResource(self.context, ®isterResource) raiseNVENC(r, "registering CUDA input buffer") self.inputHandle = registerResource.registeredResource log("input handle for CUDA buffer: %#x", self.inputHandle) #allocate output buffer: memset(&createBitstreamBufferParams, 0, sizeof(NV_ENC_CREATE_BITSTREAM_BUFFER)) createBitstreamBufferParams.version = NV_ENC_CREATE_BITSTREAM_BUFFER_VER #this is the uncompressed size - must be big enough for the compressed stream: createBitstreamBufferParams.size = min(1024*1024*2, self.encoder_width*self.encoder_height*3//2) createBitstreamBufferParams.memoryHeap = NV_ENC_MEMORY_HEAP_SYSMEM_CACHED if DEBUG_API: log("nvEncCreateBitstreamBuffer(%#x)", &createBitstreamBufferParams) with nogil: r = self.functionList.nvEncCreateBitstreamBuffer(self.context, &createBitstreamBufferParams) raiseNVENC(r, "creating output buffer") self.bitstreamBuffer = createBitstreamBufferParams.bitstreamBuffer log("output bitstream buffer=%#x", self.bitstreamBuffer) assert self.bitstreamBuffer!=NULL def get_info(self): #@DuplicatedSignature global YUV444_CODEC_SUPPORT, YUV444_ENABLED, LOSSLESS_CODEC_SUPPORT, LOSSLESS_ENABLED cdef double pps info = get_info() info.update({ "width" : self.width, "height" : self.height, "frames" : int(self.frames), "codec" : self.codec_name, "encoder_width" : self.encoder_width, "encoder_height" : self.encoder_height, "bitrate" : self.target_bitrate, "quality" : self.quality, "speed" : self.speed, "lossless" : { "" : self.lossless, "supported" : LOSSLESS_CODEC_SUPPORT.get(self.encoding, LOSSLESS_ENABLED), "threshold" : LOSSLESS_THRESHOLD }, "yuv444" : { "supported" : YUV444_CODEC_SUPPORT.get(self.encoding, YUV444_ENABLED), "threshold" : YUV444_THRESHOLD, }, "cuda-device" : self.cuda_device_info, "cuda" : self.cuda_info, "pycuda" : self.pycuda_info, }) if self.scaling!=(1,1): info.update({ "input_width" : self.input_width, "input_height" : self.input_height, "scaling" : self.scaling}) if self.src_format: info["src_format"] = self.src_format if self.pixel_format: info["pixel_format"] = self.pixel_format cdef unsigned long long b = self.bytes_in if b>0 and self.bytes_out>0: info.update({ "bytes_in" : self.bytes_in, "bytes_out" : self.bytes_out, "ratio_pct" : int(100 * self.bytes_out // b)}) if self.preset_name: info["preset"] = self.preset_name cdef double t = self.time info["total_time_ms"] = int(self.time*1000.0) if self.frames>0 and t>0: pps = float(self.width) * float(self.height) * float(self.frames) / t info["pixels_per_second"] = int(pps) info["free_memory"] = int(self.free_memory) info["total_memory"] = int(self.total_memory) cdef uint64_t m = self.total_memory if m>0: info["free_memory_pct"] = int(100.0*self.free_memory/m) #calculate fps: cdef int f = 0 cdef double now = monotonic_time() cdef double last_time = now cdef double cut_off = now-10.0 cdef double ms_per_frame = 0 for start,end in tuple(self.last_frame_times): if end>cut_off: f += 1 last_time = min(last_time, end) ms_per_frame += (end-start) if f>0 and last_time self.context) if self.cuda_context: self.cuda_context.push() try: self.cuda_clean() finally: self.cuda_context.pop() self.cuda_context.detach() self.cuda_context = None self.width = 0 self.height = 0 self.input_width = 0 self.input_height = 0 self.encoder_width = 0 self.encoder_height = 0 self.src_format = "" self.dst_formats = [] self.scaling = None self.speed = 0 self.quality = 0 #PyCUDA: self.driver = 0 self.cuda_device_id = -1 self.cuda_info = None self.pycuda_info = None self.cuda_device_info = None self.cuda_device = None self.kernel = None self.kernel_name = None self.max_block_sizes = 0 self.max_grid_sizes = 0 self.max_threads_per_block = 0 self.free_memory = 0 self.total_memory = 0 #NVENC (mostly already cleaned up in cuda_clean): self.inputPitch = 0 self.outputPitch = 0 self.bitstreamBuffer = NULL self.bufferFmt = NV_ENC_BUFFER_FORMAT_UNDEFINED self.codec_name = "" self.preset_name = "" self.pixel_format = "" #statistics, etc: self.time = 0 self.frames = 0 self.first_frame_timestamp = 0 self.last_frame_times = [] self.bytes_in = 0 self.bytes_out = 0 cdef cuda_clean(self): cdef NVENCSTATUS r #@DuplicatedSignature if self.context!=NULL and self.frames>0: self.flushEncoder() if self.inputHandle!=NULL and self.context!=NULL: log("cuda_clean() unregistering CUDA output buffer input handle %#x", self.inputHandle) if DEBUG_API: log("nvEncUnregisterResource(%#x)", self.inputHandle) with nogil: r = self.functionList.nvEncUnregisterResource(self.context, self.inputHandle) raiseNVENC(r, "unregistering CUDA input buffer") self.inputHandle = NULL if self.inputBuffer is not None: log("cuda_clean() freeing CUDA host buffer %s", self.inputBuffer) self.inputBuffer = None if self.cudaInputBuffer is not None: log("cuda_clean() freeing CUDA input buffer %#x", int(self.cudaInputBuffer)) self.cudaInputBuffer.free() self.cudaInputBuffer = None if self.cudaOutputBuffer is not None: log("cuda_clean() freeing CUDA output buffer %#x", int(self.cudaOutputBuffer)) self.cudaOutputBuffer.free() self.cudaOutputBuffer = None if self.context!=NULL: if self.bitstreamBuffer!=NULL: log("cuda_clean() destroying output bitstream buffer %#x", self.bitstreamBuffer) if DEBUG_API: log("nvEncDestroyBitstreamBuffer(%#x)", self.bitstreamBuffer) with nogil: r = self.functionList.nvEncDestroyBitstreamBuffer(self.context, self.bitstreamBuffer) raiseNVENC(r, "destroying output buffer") self.bitstreamBuffer = NULL log("cuda_clean() destroying encoder %#x", self.context) if DEBUG_API: log("nvEncDestroyEncoder(%#x)", self.context) with nogil: r = self.functionList.nvEncDestroyEncoder(self.context) raiseNVENC(r, "destroying context") self.context = NULL global context_counter context_counter.decrease() log("cuda_clean() (still %s context%s in use)", context_counter, engs(context_counter)) self.cuda_context_ptr = 0 def get_width(self): return self.width def get_height(self): return self.height def get_type(self): #@DuplicatedSignature return "nvenc" def get_encoding(self): #@DuplicatedSignature return self.encoding def get_src_format(self): return self.src_format def set_encoding_speed(self, speed): if self.speed!=speed: self.speed = speed self.update_bitrate() def set_encoding_quality(self, quality): cdef NVENCSTATUS r #@DuplicatedSignature cdef NV_ENC_RECONFIGURE_PARAMS reconfigure_params assert self.context, "context is not initialized" if self.quality==quality: return log("set_encoding_quality(%s) current quality=%s", quality, self.quality) if quality 1Mbit/s #speed=50 -> 10Mbit/s #speed=90 -> 66Mbit/s #speed=100 -> 100Mbit/s MPixels = (self.encoder_width * self.encoder_height) / (1000.0 * 1000.0) if self.pixel_format=="NV12": #subsampling halves the input size: mult = 0.5 else: #yuv444p preserves it: mult = 1.0 lim = 100*1000000 self.target_bitrate = min(lim, max(1000000, int(((0.5+self.speed/200.0)**8)*lim*MPixels*mult))) self.max_bitrate = 2*self.target_bitrate cdef flushEncoder(self): cdef NV_ENC_PIC_PARAMS picParams cdef NVENCSTATUS r #@DuplicatedSignature assert self.context, "context is not initialized" memset(&picParams, 0, sizeof(NV_ENC_PIC_PARAMS)) picParams.version = NV_ENC_PIC_PARAMS_VER picParams.encodePicFlags = NV_ENC_PIC_FLAG_EOS if DEBUG_API: log("nvEncEncodePicture(%#x)", &picParams) with nogil: r = self.functionList.nvEncEncodePicture(self.context, &picParams) raiseNVENC(r, "flushing encoder buffer") def compress_image(self, image, quality=-1, speed=-1, options={}, retry=0): self.cuda_context.push() try: try: if quality>=0: self.set_encoding_quality(quality) if speed>=0: self.set_encoding_speed(speed) return self.do_compress_image(image, options) finally: self.cuda_context.pop() except driver.LogicError as e: if retry>0: raise log.warn("Warning: PyCUDA %s", e) del e self.clean() self.init_cuda() return self.convert_image(image, options, retry+1) cdef do_compress_image(self, image, options={}): cdef unsigned int stride, w, h assert self.context, "context is not initialized" assert self.context!=NULL, "context is not initialized" w = image.get_width() h = image.get_height() gpu_buffer = image.get_gpu_buffer() stride = image.get_rowstride() log("compress_image(%s, %s) kernel=%s, GPU buffer=%#x, stride=%i, input pitch=%i, output pitch=%i", image, options, self.kernel_name, int(gpu_buffer or 0), stride, self.inputPitch, self.outputPitch) assert image.get_planes()==ImageWrapper.PACKED, "invalid number of planes: %s" % image.get_planes() assert (w & WIDTH_MASK)<=self.input_width, "invalid width: %s" % w assert (h & HEIGHT_MASK)<=self.input_height, "invalid height: %s" % h assert self.inputBuffer is not None, "BUG: encoder is closed?" if self.frames==0: #first frame, record pts: self.first_frame_timestamp = image.get_timestamp() cdef unsigned long input_size if self.kernel: #copy to input buffer, CUDA kernel converts into output buffer: if GPU_MEMCOPY and gpu_buffer and stride<=self.inputPitch: driver.memcpy_dtod(self.cudaInputBuffer, int(gpu_buffer), stride*h) log("GPU memcopy %i bytes from %#x to %#x", stride*h, int(gpu_buffer), int(self.cudaInputBuffer)) else: stride = self.copy_image(image, self.inputBuffer, self.inputPitch, False) driver.memcpy_htod(self.cudaInputBuffer, self.inputBuffer) self.exec_kernel(w, h, stride) input_size = self.inputPitch * self.input_height else: #go direct to the CUDA "output" buffer: if GPU_MEMCOPY and gpu_buffer and stride<=self.outputPitch: driver.memcpy_dtod(self.cudaOutputBuffer, int(gpu_buffer), stride*h) log("GPU memcopy %i bytes from %#x to %#x", stride*h, int(gpu_buffer), int(self.cudaOutputBuffer)) else: stride = self.copy_image(image, self.inputBuffer, self.inputPitch, True) driver.memcpy_htod(self.cudaOutputBuffer, self.inputBuffer) input_size = stride * self.encoder_height self.bytes_in += input_size cdef NV_ENC_INPUT_PTR mappedResource = self.map_input_resource() assert mappedResource!=NULL try: return self.nvenc_compress(input_size, mappedResource, image.get_timestamp()) finally: self.unmap_input_resource(mappedResource) cdef unsigned int copy_image(self, image, target_buffer, unsigned int target_stride, int strict_stride) except -1: if DEBUG_API: log("copy_image(%s, %s, %i, %i)", image, target_buffer, target_stride, strict_stride) cdef unsigned int image_stride = image.get_rowstride() cdef unsigned int h = image.get_height() cdef unsigned int i, stride, min_stride pixels = image.get_pixels() assert pixels is not None, "failed to get pixels from %s" % image #copy to input buffer: cdef object buf if isinstance(pixels, (bytearray, bytes)): pixels = memoryview(pixels) if isinstance(pixels, memoryview): #copy memoryview to inputBuffer directly: buf = target_buffer else: #this is a numpy.ndarray type: buf = target_buffer.data cdef double start = monotonic_time() cdef unsigned long copy_len cdef unsigned long pix_len = len(pixels) assert pix_len>=(h*image_stride), "image pixel buffer is too small: expected at least %ix%i=%i bytes but got %i bytes" % (h, image_stride, h*image_stride, pix_len) if image_stride<=target_stride and not strict_stride: stride = image_stride copy_len = h*image_stride #assert pix_len<=input_size, "too many pixels (expected %s max, got %s) image: %sx%s stride=%s, input buffer: stride=%s, height=%s" % (input_size, pix_len, w, h, stride, self.inputPitch, self.input_height) log("copying %s bytes from %s into %s (len=%i), in one shot", pix_len, type(pixels), type(target_buffer), len(target_buffer)) #log("target: %s, %s, %s", buf.shape, buf.size, buf.dtype) if isinstance(pixels, memoryview): tmp = numpy.asarray(pixels, numpy.int8) else: tmp = numpy.frombuffer(pixels, numpy.int8) buf[:copy_len] = tmp[:copy_len] else: #ouch, we need to copy the source pixels into the smaller buffer #before uploading to the device... this is probably costly! stride = target_stride min_stride = min(stride, image_stride) log("copying %s bytes from %s into %s, %i stride at a time (from image stride=%i, target stride=%i)", stride*h, type(pixels), type(target_buffer), min_stride, image_stride, target_stride) copy_len = min_stride * h for i in range(h): x = i*stride y = i*image_stride buf[x:x+min_stride] = pixels[y:y+min_stride] cdef double end = monotonic_time() cdef double elapsed = end-start if elapsed==0: #mswindows monotonic time minimum precision is 1ms... elapsed = 0.0001 log("copy_image: %9i bytes uploaded in %3.1f ms: %5i MB/s", copy_len, int(1000*elapsed), copy_len/elapsed//1024//1024) return stride cdef exec_kernel(self, unsigned int w, unsigned int h, unsigned int stride): cdef uint8_t dx, dy if self.pixel_format=="NV12": #(these values are derived from the kernel code - which we should know nothing about here..) #divide each dimension by 2 since we process 4 pixels at a time: dx, dy = 2, 2 elif self.pixel_format=="YUV444P": #one pixel at a time: dx, dy = 1, 1 else: raise Exception("bug: invalid pixel format '%s'" % self.pixel_format) #FIXME: find better values and validate against max_block/max_grid: #calculate grids/blocks: #a block is a group of threads: (blockw * blockh) threads #a grid is a group of blocks: (gridw * gridh) blocks cdef uint32_t blockw = 32 cdef uint32_t blockh = 32 cdef uint32_t gridw = MAX(1, w//(blockw*dx)) cdef uint32_t gridh = MAX(1, h//(blockh*dy)) #if dx or dy made us round down, add one: if gridw*dx*blockw &mapInputResource, self.inputHandle) cdef NVENCSTATUS r = self.functionList.nvEncMapInputResource(self.context, &mapInputResource) raiseNVENC(r, "mapping input resource") cdef NV_ENC_INPUT_PTR mappedResource = mapInputResource.mappedResource if DEBUG_API: log("compress_image(..) device buffer mapped to %#x", mappedResource) return mappedResource cdef unmap_input_resource(self, NV_ENC_INPUT_PTR mappedResource): if DEBUG_API: log("nvEncUnmapInputResource(%#x)", mappedResource) cdef int r = self.functionList.nvEncUnmapInputResource(self.context, mappedResource) raiseNVENC(r, "unmapping input resource") cdef nvenc_compress(self, int input_size, NV_ENC_INPUT_PTR input, timestamp=0): cdef NVENCSTATUS r #@DuplicatedSignature cdef NV_ENC_PIC_PARAMS picParams #@DuplicatedSignature cdef NV_ENC_LOCK_BITSTREAM lockOutputBuffer assert input_size>0, "invalid input size %i" % input_size cdef double start = monotonic_time() if DEBUG_API: log("nvEncEncodePicture(%#x)", &picParams) memset(&picParams, 0, sizeof(NV_ENC_PIC_PARAMS)) picParams.version = NV_ENC_PIC_PARAMS_VER picParams.bufferFmt = self.bufferFmt picParams.pictureStruct = NV_ENC_PIC_STRUCT_FRAME picParams.inputWidth = self.encoder_width picParams.inputHeight = self.encoder_height picParams.inputPitch = self.outputPitch picParams.inputBuffer = input picParams.outputBitstream = self.bitstreamBuffer #picParams.pictureType: required when enablePTD is disabled if self.frames==0: #only the first frame needs to be IDR (as we never lose frames) picParams.pictureType = NV_ENC_PIC_TYPE_IDR picParams.encodePicFlags = NV_ENC_PIC_FLAG_OUTPUT_SPSPPS else: picParams.pictureType = NV_ENC_PIC_TYPE_P if self.encoding=="h264": picParams.codecPicParams.h264PicParams.displayPOCSyntax = 2*self.frames picParams.codecPicParams.h264PicParams.refPicFlag = self.frames==0 #this causes crashes with Pascal (ie GTX-1070): #picParams.codecPicParams.h264PicParams.sliceMode = 3 #sliceModeData specifies the number of slices #picParams.codecPicParams.h264PicParams.sliceModeData = 1 #1 slice! else: picParams.codecPicParams.hevcPicParams.displayPOCSyntax = 2*self.frames picParams.codecPicParams.hevcPicParams.refPicFlag = self.frames==0 picParams.frameIdx = self.frames picParams.inputTimeStamp = timestamp-self.first_frame_timestamp #inputDuration = 0 #FIXME: use frame delay? #picParams.rcParams.rateControlMode = NV_ENC_PARAMS_RC_VBR #FIXME: check NV_ENC_CAPS_SUPPORTED_RATECONTROL_MODES caps #picParams.rcParams.enableMinQP = 1 #picParams.rcParams.enableMaxQP = 1 #0=max quality, 63 lowest quality #qmin = QP_MAX_VALUE-min(QP_MAX_VALUE, int(QP_MAX_VALUE*(self.quality+20)/100)) #qmax = QP_MAX_VALUE-max(0, int(QP_MAX_VALUE*(self.quality-20)/100)) #picParams.rcParams.minQP.qpInterB = qmin #picParams.rcParams.minQP.qpInterP = qmin #picParams.rcParams.minQP.qpIntra = qmin #picParams.rcParams.maxQP.qpInterB = qmax #picParams.rcParams.maxQP.qpInterP = qmax #picParams.rcParams.maxQP.qpIntra = qmax #picParams.rcParams.averageBitRate = self.target_bitrate #picParams.rcParams.maxBitRate = self.max_bitrate with nogil: r = self.functionList.nvEncEncodePicture(self.context, &picParams) raiseNVENC(r, "error during picture encoding") memset(&lockOutputBuffer, 0, sizeof(NV_ENC_LOCK_BITSTREAM)) #lock output buffer: lockOutputBuffer.version = NV_ENC_LOCK_BITSTREAM_VER lockOutputBuffer.doNotWait = 0 lockOutputBuffer.outputBitstream = self.bitstreamBuffer if DEBUG_API: log("nvEncLockBitstream(%#x) bitstreamBuffer=%#x", &lockOutputBuffer, self.bitstreamBuffer) with nogil: r = self.functionList.nvEncLockBitstream(self.context, &lockOutputBuffer) raiseNVENC(r, "locking output buffer") assert lockOutputBuffer.bitstreamBufferPtr!=NULL #copy to python buffer: size = lockOutputBuffer.bitstreamSizeInBytes self.bytes_out += size data = ( lockOutputBuffer.bitstreamBufferPtr)[:size] if DEBUG_API: log("nvEncUnlockBitstream(%#x)", self.bitstreamBuffer) r = self.functionList.nvEncUnlockBitstream(self.context, self.bitstreamBuffer) raiseNVENC(r, "unlocking output buffer") #update info: self.free_memory, self.total_memory = driver.mem_get_info() client_options = { "frame" : int(self.frames), "pts" : int(timestamp-self.first_frame_timestamp), "speed" : self.speed, } if self.lossless: client_options["quality"] = 100 client_options["lossless"] = True else: client_options["quality"] = min(99, self.quality) #ensure we cap it at 99 because this is lossy if self.scaling!=(1,1): client_options["scaled_size"] = self.encoder_width, self.encoder_height cdef double end = monotonic_time() self.frames += 1 self.last_frame_times.append((start, end)) cdef double elapsed = end-start self.time += elapsed #log("memory: %iMB free, %iMB total", self.free_memory//1024//1024, self.total_memory//1024//1024) log("compress_image(..) %5s %3s returning %9s bytes (%.1f%%) for %4s %s-frame no %6i took %3.1fms", get_type(), get_version(), size, 100.0*size/input_size, self.encoding, PIC_TYPES.get(picParams.pictureType, picParams.pictureType), self.frames, 1000.0*elapsed) return data, client_options cdef NV_ENC_PRESET_CONFIG *get_preset_config(self, name, GUID encode_GUID, GUID preset_GUID) except *: """ you must free it after use! """ cdef NV_ENC_PRESET_CONFIG *presetConfig #@DuplicatedSignature cdef NVENCSTATUS r #@DuplicatedSignature assert self.context, "context is not initialized" presetConfig = cmalloc(sizeof(NV_ENC_PRESET_CONFIG), "preset config") memset(presetConfig, 0, sizeof(NV_ENC_PRESET_CONFIG)) presetConfig.version = NV_ENC_PRESET_CONFIG_VER presetConfig.presetCfg.version = NV_ENC_CONFIG_VER if DEBUG_API: log("nvEncGetEncodePresetConfig(%s, %s)", codecstr(encode_GUID), presetstr(preset_GUID)) r = self.functionList.nvEncGetEncodePresetConfig(self.context, encode_GUID, preset_GUID, presetConfig) if r!=0: log.warn("failed to get preset config for %s (%s / %s): %s", name, guidstr(encode_GUID), guidstr(preset_GUID), NV_ENC_STATUS_TXT.get(r, r)) return NULL return presetConfig cdef object query_presets(self, GUID encode_GUID): cdef uint32_t presetCount cdef uint32_t presetsRetCount cdef GUID* preset_GUIDs cdef GUID preset_GUID cdef NV_ENC_PRESET_CONFIG *presetConfig cdef NV_ENC_CONFIG *encConfig cdef NVENCSTATUS r #@DuplicatedSignature assert self.context, "context is not initialized" presets = {} if DEBUG_API: log("nvEncGetEncodePresetCount(%s, %#x)", codecstr(encode_GUID), &presetCount) with nogil: r = self.functionList.nvEncGetEncodePresetCount(self.context, encode_GUID, &presetCount) raiseNVENC(r, "getting preset count for %s" % guidstr(encode_GUID)) log("found %s preset%s:", presetCount, engs(presetCount)) assert presetCount<2**8 preset_GUIDs = cmalloc(sizeof(GUID) * presetCount, "preset GUIDs") try: if DEBUG_API: log("nvEncGetEncodePresetGUIDs(%s, %#x)", codecstr(encode_GUID), &presetCount) with nogil: r = self.functionList.nvEncGetEncodePresetGUIDs(self.context, encode_GUID, preset_GUIDs, presetCount, &presetsRetCount) raiseNVENC(r, "getting encode presets") assert presetsRetCount==presetCount unknowns = [] for x in range(presetCount): preset_GUID = preset_GUIDs[x] preset_name = CODEC_PRESETS_GUIDS.get(guidstr(preset_GUID)) if DEBUG_API: log("* %s : %s", guidstr(preset_GUID), preset_name or "unknown!") presetConfig = self.get_preset_config(preset_name, encode_GUID, preset_GUID) if presetConfig!=NULL: try: encConfig = &presetConfig.presetCfg if DEBUG_API: log("presetConfig.presetCfg=%s", encConfig) gop = {NVENC_INFINITE_GOPLENGTH : "infinite"}.get(encConfig.gopLength, encConfig.gopLength) log("* %-20s P frame interval=%i, gop length=%-10s", preset_name or "unknown!", encConfig.frameIntervalP, gop) finally: free(presetConfig) if preset_name is None: unknowns.append(guidstr(preset_GUID)) else: presets[preset_name] = guidstr(preset_GUID) if len(unknowns)>0: log.warn("Warning: found some unknown NVENC presets: %s", csv(unknowns)) finally: free(preset_GUIDs) if DEBUG_API: log("query_presets(%s)=%s", codecstr(encode_GUID), presets) return presets cdef object query_profiles(self, GUID encode_GUID): cdef uint32_t profileCount cdef uint32_t profilesRetCount cdef GUID* profile_GUIDs cdef GUID profile_GUID cdef NVENCSTATUS r #@DuplicatedSignature assert self.context, "context is not initialized" profiles = {} if DEBUG_API: log("nvEncGetEncodeProfileGUIDCount(%s, %#x)", codecstr(encode_GUID), &profileCount) with nogil: r = self.functionList.nvEncGetEncodeProfileGUIDCount(self.context, encode_GUID, &profileCount) raiseNVENC(r, "getting profile count") log("%s profiles:", profileCount) assert profileCount<2**8 profile_GUIDs = cmalloc(sizeof(GUID) * profileCount, "profile GUIDs") PROFILES_GUIDS = CODEC_PROFILES_GUIDS.get(guidstr(encode_GUID), {}) try: if DEBUG_API: log("nvEncGetEncodeProfileGUIDs(%s, %#x, %#x)", codecstr(encode_GUID), profile_GUIDs, &profileCount) with nogil: r = self.functionList.nvEncGetEncodeProfileGUIDs(self.context, encode_GUID, profile_GUIDs, profileCount, &profilesRetCount) raiseNVENC(r, "getting encode profiles") #(void* encoder, GUID encodeGUID, GUID* profileGUIDs, uint32_t guidArraySize, uint32_t* GUIDCount) assert profilesRetCount==profileCount for x in range(profileCount): profile_GUID = profile_GUIDs[x] profile_name = PROFILES_GUIDS.get(guidstr(profile_GUID)) log("* %s : %s", guidstr(profile_GUID), profile_name) profiles[profile_name] = guidstr(profile_GUID) finally: free(profile_GUIDs) return profiles cdef object query_input_formats(self, GUID encode_GUID): cdef uint32_t inputFmtCount cdef NV_ENC_BUFFER_FORMAT* inputFmts cdef uint32_t inputFmtsRetCount cdef NV_ENC_BUFFER_FORMAT inputFmt cdef NVENCSTATUS r #@DuplicatedSignature assert self.context, "context is not initialized" input_formats = {} if DEBUG_API: log("nvEncGetInputFormatCount(%s, %#x)", codecstr(encode_GUID), &inputFmtCount) with nogil: r = self.functionList.nvEncGetInputFormatCount(self.context, encode_GUID, &inputFmtCount) raiseNVENC(r, "getting input format count") log("%s input format type%s:", inputFmtCount, engs(inputFmtCount)) assert inputFmtCount>0 and inputFmtCount<2**8 inputFmts = cmalloc(sizeof(int) * inputFmtCount, "input formats") try: if DEBUG_API: log("nvEncGetInputFormats(%s, %#x, %i, %#x)", codecstr(encode_GUID), inputFmts, inputFmtCount, &inputFmtsRetCount) with nogil: r = self.functionList.nvEncGetInputFormats(self.context, encode_GUID, inputFmts, inputFmtCount, &inputFmtsRetCount) raiseNVENC(r, "getting input formats") assert inputFmtsRetCount==inputFmtCount for x in range(inputFmtCount): inputFmt = inputFmts[x] log("* %#x", inputFmt) for format_mask in sorted(BUFFER_FORMAT.keys()): if format_mask>0 and (format_mask & inputFmt)>0: format_name = BUFFER_FORMAT.get(format_mask) log(" + %#x : %s", format_mask, format_name) input_formats[format_name] = hex(format_mask) finally: free(inputFmts) return input_formats cdef int query_encoder_caps(self, GUID encode_GUID, NV_ENC_CAPS caps_type) except *: cdef int val cdef NV_ENC_CAPS_PARAM encCaps cdef NVENCSTATUS r #@DuplicatedSignature assert self.context, "context is not initialized" memset(&encCaps, 0, sizeof(NV_ENC_CAPS_PARAM)) encCaps.version = NV_ENC_CAPS_PARAM_VER encCaps.capsToQuery = caps_type with nogil: r = self.functionList.nvEncGetEncodeCaps(self.context, encode_GUID, &encCaps, &val) raiseNVENC(r, "getting encode caps for %s" % CAPS_NAMES.get(caps_type, caps_type)) if DEBUG_API: log("query_encoder_caps(%s, %s) %s=%s", codecstr(encode_GUID), caps_type, CAPS_NAMES.get(caps_type, caps_type), val) return val def query_codecs(self, full_query=False): cdef uint32_t GUIDCount cdef uint32_t GUIDRetCount cdef GUID* encode_GUIDs cdef GUID encode_GUID cdef NVENCSTATUS r #@DuplicatedSignature assert self.context, "context is not initialized" if DEBUG_API: log("nvEncGetEncodeGUIDCount(%#x, %#x)", self.context, &GUIDCount) with nogil: r = self.functionList.nvEncGetEncodeGUIDCount(self.context, &GUIDCount) raiseNVENC(r, "getting encoder count") log("found %i encoder%s:", GUIDCount, engs(GUIDCount)) assert GUIDCount<2**8 encode_GUIDs = cmalloc(sizeof(GUID) * GUIDCount, "encode GUIDs") codecs = {} try: if DEBUG_API: log("nvEncGetEncodeGUIDs(%#x, %i, %#x)", encode_GUIDs, GUIDCount, &GUIDRetCount) with nogil: r = self.functionList.nvEncGetEncodeGUIDs(self.context, encode_GUIDs, GUIDCount, &GUIDRetCount) raiseNVENC(r, "getting list of encode GUIDs") assert GUIDRetCount==GUIDCount, "expected %s items but got %s" % (GUIDCount, GUIDRetCount) for x in range(GUIDRetCount): encode_GUID = encode_GUIDs[x] codec_name = CODEC_GUIDS.get(guidstr(encode_GUID)) if not codec_name: log("[%s] unknown codec GUID: %s", x, guidstr(encode_GUID)) else: log("[%s] %s", x, codec_name) maxw = self.query_encoder_caps(encode_GUID, NV_ENC_CAPS_WIDTH_MAX) maxh = self.query_encoder_caps(encode_GUID, NV_ENC_CAPS_HEIGHT_MAX) async = self.query_encoder_caps(encode_GUID, NV_ENC_CAPS_ASYNC_ENCODE_SUPPORT) rate_control = self.query_encoder_caps(encode_GUID, NV_ENC_CAPS_SUPPORTED_RATECONTROL_MODES) codec = { "guid" : guidstr(encode_GUID), "name" : codec_name, "max-size" : (maxw, maxh), "async" : async, "rate-control" : rate_control } if full_query: presets = self.query_presets(encode_GUID) profiles = self.query_profiles(encode_GUID) input_formats = self.query_input_formats(encode_GUID) codec.update({ "presets" : presets, "profiles" : profiles, "input-formats" : input_formats, }) codecs[codec_name] = codec finally: free(encode_GUIDs) log("codecs=%s", csv(codecs.keys())) return codecs def open_encode_session(self): global context_counter, context_gen_counter, last_context_failure cdef NVENCSTATUS r #@DuplicatedSignature cdef int t cdef NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS params assert self.functionList is NULL, "session already active" assert self.context is NULL, "context already set" assert self.cuda_context and self.cuda_context_ptr!=NULL, "cuda context is not set" #params = malloc(sizeof(NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS)) log("open_encode_session() cuda_context=%s, cuda_context_ptr=%#x", self.cuda_context, self.cuda_context_ptr) self.functionList = cmalloc(sizeof(NV_ENCODE_API_FUNCTION_LIST), "function list") assert memset(self.functionList, 0, sizeof(NV_ENCODE_API_FUNCTION_LIST))!=NULL log("open_encode_session() functionList=%#x", self.functionList) #get NVENC function pointers: memset(self.functionList, 0, sizeof(NV_ENCODE_API_FUNCTION_LIST)) self.functionList.version = NV_ENCODE_API_FUNCTION_LIST_VER if DEBUG_API: log("NvEncodeAPICreateInstance(%#x)", self.functionList) r = NvEncodeAPICreateInstance( self.functionList) raiseNVENC(r, "getting API function list") assert self.functionList.nvEncOpenEncodeSessionEx!=NULL, "looks like NvEncodeAPICreateInstance failed!" #NVENC init: memset(¶ms, 0, sizeof(NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS)) params.version = NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS_VER params.deviceType = NV_ENC_DEVICE_TYPE_CUDA params.device = self.cuda_context_ptr params.reserved = &CLIENT_KEY_GUID params.apiVersion = NVENCAPI_VERSION cstr = ¶ms pstr = cstr[:sizeof(NV_ENC_OPEN_ENCODE_SESSION_EX_PARAMS)] if DEBUG_API: log("calling nvEncOpenEncodeSessionEx @ %#x", self.functionList.nvEncOpenEncodeSessionEx) self.context = NULL with nogil: r = self.functionList.nvEncOpenEncodeSessionEx(¶ms, &self.context) if r==NV_ENC_ERR_UNSUPPORTED_DEVICE: last_context_failure = monotonic_time() msg = "NV_ENC_ERR_UNSUPPORTED_DEVICE: could not open encode session (out of resources / no more codec contexts?)" log(msg) raise TransientCodecException(msg) if self.context==NULL: raise TransientCodecException("cannot open encoding session, context is NULL") raiseNVENC(r, "opening session") context_counter.increase() context_gen_counter.increase() log("success, encoder context=%#x (%s context%s in use)", self.context, context_counter, engs(context_counter)) _init_message = False def init_module(): log("nvenc.init_module()") #TODO: this should be a build time check: if NVENCAPI_MAJOR_VERSION<0x7: raise Exception("unsupported version of NVENC: %#x" % NVENCAPI_VERSION) log("NVENC encoder API version %s", ".".join([str(x) for x in PRETTY_VERSION])) cdef Encoder test_encoder cdef uint32_t max_version #cdef NVENCSTATUS r = NvEncodeAPIGetMaxSupportedVersion(&max_version) #raiseNVENC(r, "querying max version") #log(" maximum supported version: %s", max_version) #load the library / DLL: init_nvencode_library() #make sure we have devices we can use: devices = init_all_devices() if len(devices)==0: log("nvenc: no compatible devices found") return success = False valid_keys = [] failed_keys = [] try_keys = CLIENT_KEYS_STR or [None] TEST_ENCODINGS = ["h264", "h265"] FAILED_ENCODINGS = set() global YUV444_ENABLED, YUV444_CODEC_SUPPORT, LOSSLESS_ENABLED, ENCODINGS, MAX_SIZE if not validate_driver_yuv444lossless(): if YUV444_ENABLED: YUV444_ENABLED = False if LOSSLESS_ENABLED: LOSSLESS_ENABLED = False #check NVENC availibility by creating a context: device_warnings = OrderedDict() for client_key in try_keys: if client_key: #this will set the global key object used by all encoder contexts: log("init_module() testing with key '%s'", client_key) global CLIENT_KEY_GUID CLIENT_KEY_GUID = c_parseguid(client_key) for device_id in devices: log("testing encoder with device %s", device_id) options = {"cuda_device" : device_id} try: test_encoder = Encoder() test_encoder.select_cuda_device(options) test_encoder.init_cuda() log("test encoder=%s", test_encoder) test_encoder.open_encode_session() log("init_encoder() %s", test_encoder) codecs = test_encoder.query_codecs() log("device %i supports: %s", device_id, codecs) finally: test_encoder.clean() test_encoder = None test_encodings = [] for e in TEST_ENCODINGS: if e in FAILED_ENCODINGS: continue nvenc_encoding_name = { "h264" : "H264", "h265" : "HEVC", }.get(e, e) codec_query = codecs.get(nvenc_encoding_name) if not codec_query: wkey = "nvenc:%s-%s" % (device_id, nvenc_encoding_name) if first_time(wkey): log.warn("Warning: NVENC on device %s:", get_device_name(device_id) or device_id) log.warn(" does not support %s", nvenc_encoding_name) FAILED_ENCODINGS.add(e) continue #ensure MAX_SIZE is set: cmax = MAX_SIZE.get(e) qmax = codec_query.get("max-size") if qmax: #minimum of current value and value for this device: qmx, qmy = qmax cmx, cmy = cmax or qmax v = min(qmx, cmx), min(qmy, cmy) log("max-size(%s)=%s", e, v) MAX_SIZE[e] = v test_encodings.append(e) log("will test: %s", test_encodings) for encoding in test_encodings: colorspaces = get_input_colorspaces(encoding) assert colorspaces, "cannot use NVENC: no colorspaces available" src_format = colorspaces[0] dst_formats = get_output_colorspaces(encoding, src_format) test_encoder = None try: test_encoder = Encoder() test_encoder.init_context(1920, 1080, src_format, dst_formats, encoding, 50, 100, (1,1), options) success = True if client_key: log("the license key '%s' is valid", client_key) valid_keys.append(client_key) #check for YUV444 support yuv444_support = YUV444_ENABLED and test_encoder.query_encoder_caps(test_encoder.get_codec(), NV_ENC_CAPS_SUPPORT_YUV444_ENCODE) YUV444_CODEC_SUPPORT[encoding] = bool(yuv444_support) if YUV444_ENABLED and not yuv444_support: wkey = "nvenc:%s-%s-%s" % (device_id, encoding, "YUV444") if first_time(wkey): device_warnings.setdefault(device_id, OrderedDict()).setdefault(encoding, []).append("YUV444") log("no support for YUV444 with %s", encoding) log("%s YUV444 support: %s", encoding, YUV444_CODEC_SUPPORT.get(encoding, YUV444_ENABLED)) #check for lossless: lossless_support = yuv444_support and LOSSLESS_ENABLED and test_encoder.query_encoder_caps(test_encoder.get_codec(), NV_ENC_CAPS_SUPPORT_LOSSLESS_ENCODE) LOSSLESS_CODEC_SUPPORT[encoding] = lossless_support if LOSSLESS_ENABLED and not lossless_support: wkey = "nvenc:%s-%s-%s" % (device_id, encoding, "lossless") if first_time(wkey): device_warnings.setdefault(device_id, OrderedDict()).setdefault(encoding, []).append("lossless") log("no support for lossless mode with %s", encoding) log("%s lossless support: %s", encoding, LOSSLESS_CODEC_SUPPORT.get(encoding, LOSSLESS_ENABLED)) except NVENCException as e: log("encoder %s failed: %s", test_encoder, e) #special handling for license key issues: if e.code==NV_ENC_ERR_INCOMPATIBLE_CLIENT_KEY: if client_key: log("invalid license key '%s' (skipped)", client_key) failed_keys.append(client_key) else: log("a license key is required") elif e.code==NV_ENC_ERR_INVALID_VERSION: #we can bail out already: raise Exception("version mismatch, you need a newer/older codec build or newer/older drivers") else: #it seems that newer version will fail with #seemingly random errors when we supply the wrong key log.warn("error during NVENC encoder test: %s", e) if client_key: log(" license key '%s' may not be valid (skipped)", client_key) failed_keys.append(client_key) else: log(" a license key may be required") finally: if test_encoder: test_encoder.clean() if device_warnings: for device_id, encoding_warnings in device_warnings.items(): log.info("NVENC on device %s:", get_device_name(device_id) or device_id) for encoding, warnings in encoding_warnings.items(): log.info(" %s encoding does not support %s mode", encoding, " or ".join(warnings)) if success: #pick the first valid license key: if len(valid_keys)>0: x = valid_keys[0] log("using the license key '%s'", x) CLIENT_KEY_GUID = c_parseguid(x) else: log("no license keys are required") ENCODINGS[:] = [x for x in TEST_ENCODINGS if x not in FAILED_ENCODINGS] else: #we got license key error(s) if len(failed_keys)>0: raise Exception("the license %s specified may be invalid" % (["key", "keys"][len(failed_keys)>1])) else: raise Exception("you may need to provide a license key") global _init_message if ENCODINGS and not _init_message: log.info("NVENC v%i successfully initialized with codecs: %s", NVENCAPI_MAJOR_VERSION, csv(ENCODINGS)) _init_message = True def cleanup_module(): log("nvenc.cleanup_module()") reset_state() def selftest(full=False): v = get_nvidia_module_version(True) assert NVENCAPI_MAJOR_VERSION>=7, "unsupported NVENC version %i" % NVENCAPI_MAJOR_VERSION if v: NVENC_UNSUPPORTED_DRIVER_VERSION = envbool("XPRA_NVENC_UNSUPPORTED_DRIVER_VERSION", False) #SDK 7.0 requires version 367 or later #SDK 7.1 requires version 375 or later if v<(375, 0): if not NVENC_UNSUPPORTED_DRIVER_VERSION: raise ImportError("unsupported NVidia driver version %s\nuse XPRA_NVENC_UNSUPPORTED_DRIVER_VERSION=1 to force enable it" % pver(v)) #this is expensive, so don't run it unless "full" is set: if full: from xpra.codecs.codec_checks import get_encoder_max_sizes from xpra.codecs.nvenc import encoder init_module() log.info("%s max dimensions: %s", encoder, get_encoder_max_sizes(encoder))