diff --git a/constant/__init__.py b/constant/__init__.py index 93a44a8..028fff5 100644 --- a/constant/__init__.py +++ b/constant/__init__.py @@ -41,7 +41,14 @@ EFFECT_TYPES = ( 'blur', # 模糊效果（预留） ) -# 统一视频编码参数（来自集成文档） +# 硬件加速类型 +HW_ACCEL_NONE = 'none' # 纯软件编解码 +HW_ACCEL_QSV = 'qsv' # Intel Quick Sync Video (核显/独显) +HW_ACCEL_CUDA = 'cuda' # NVIDIA NVENC/NVDEC + +HW_ACCEL_TYPES = (HW_ACCEL_NONE, HW_ACCEL_QSV, HW_ACCEL_CUDA) + +# 统一视频编码参数（软件编码，来自集成文档） VIDEO_ENCODE_PARAMS = { 'codec': 'libx264', 'preset': 'medium', @@ -51,6 +58,28 @@ VIDEO_ENCODE_PARAMS = { 'pix_fmt': 'yuv420p', } +# QSV 硬件加速视频编码参数（Intel Quick Sync） +VIDEO_ENCODE_PARAMS_QSV = { + 'codec': 'h264_qsv', + 'preset': 'medium', # QSV 支持: veryfast, faster, fast, medium, slow, slower, veryslow + 'profile': 'main', + 'level': '4.0', + 'global_quality': '23', # QSV 使用 global_quality 代替 crf（1-51，值越低质量越高） + 'look_ahead': '1', # 启用前瞻分析提升质量 + 'pix_fmt': 'nv12', # QSV 硬件表面格式 +} + +# CUDA 硬件加速视频编码参数（NVIDIA NVENC） +VIDEO_ENCODE_PARAMS_CUDA = { + 'codec': 'h264_nvenc', + 'preset': 'p4', # NVENC 预设 p1-p7（p1 最快，p7 最慢/质量最高），p4 ≈ medium + 'profile': 'main', + 'level': '4.0', + 'rc': 'vbr', # 码率控制模式：vbr 可变码率 + 'cq': '23', # 恒定质量模式的质量值（0-51） + 'pix_fmt': 'yuv420p', # NVENC 输入格式（会自动转换） +} + # 统一音频编码参数 AUDIO_ENCODE_PARAMS = { 'codec': 'aac', diff --git a/domain/config.py b/domain/config.py index 59b450e..8fbbdfd 100644 --- a/domain/config.py +++ b/domain/config.py @@ -9,6 +9,8 @@ import os from dataclasses import dataclass, field from typing import List, Optional +from constant import HW_ACCEL_NONE, HW_ACCEL_QSV, HW_ACCEL_CUDA, HW_ACCEL_TYPES + # 默认支持的任务类型 DEFAULT_CAPABILITIES = [ @@ -54,6 +56,9 @@ class WorkerConfig: download_timeout: int = 300 # 秒，下载超时 upload_timeout: int = 600 # 秒，上传超时 + # 硬件加速配置 + hw_accel: str = HW_ACCEL_NONE # 硬件加速类型: none, qsv, cuda + @classmethod def from_env(cls) -> 'WorkerConfig': """从环境变量创建配置""" @@ -98,6 +103,11 @@ class WorkerConfig: download_timeout = int(os.getenv('DOWNLOAD_TIMEOUT', '300')) upload_timeout = int(os.getenv('UPLOAD_TIMEOUT', '600')) + # 硬件加速配置 + hw_accel = os.getenv('HW_ACCEL', HW_ACCEL_NONE).lower() + if hw_accel not in HW_ACCEL_TYPES: + hw_accel = HW_ACCEL_NONE + return cls( api_endpoint=api_endpoint, access_key=access_key, @@ -110,7 +120,8 @@ class WorkerConfig: capabilities=capabilities, ffmpeg_timeout=ffmpeg_timeout, download_timeout=download_timeout, - upload_timeout=upload_timeout + upload_timeout=upload_timeout, + hw_accel=hw_accel ) def get_work_dir_path(self, task_id: str) -> str: @@ -120,3 +131,15 @@ class WorkerConfig: def ensure_temp_dir(self) -> None: """确保临时目录存在""" os.makedirs(self.temp_dir, exist_ok=True) + + def is_hw_accel_enabled(self) -> bool: + """是否启用了硬件加速""" + return self.hw_accel != HW_ACCEL_NONE + + def is_qsv(self) -> bool: + """是否使用 QSV 硬件加速""" + return self.hw_accel == HW_ACCEL_QSV + + def is_cuda(self) -> bool: + """是否使用 CUDA 硬件加速""" + return self.hw_accel == HW_ACCEL_CUDA diff --git a/handlers/base.py b/handlers/base.py index 0d8f4ab..2542096 100644 --- a/handlers/base.py +++ b/handlers/base.py @@ -19,6 +19,10 @@ from domain.task import Task from domain.result import TaskResult, ErrorCode from domain.config import WorkerConfig from services import storage +from constant import ( + HW_ACCEL_NONE, HW_ACCEL_QSV, HW_ACCEL_CUDA, + VIDEO_ENCODE_PARAMS, VIDEO_ENCODE_PARAMS_QSV, VIDEO_ENCODE_PARAMS_CUDA +) if TYPE_CHECKING: from services.api_client import APIClientV2 @@ -26,15 +30,94 @@ if TYPE_CHECKING: logger = logging.getLogger(__name__) -# v2 统一视频编码参数（来自集成文档） -VIDEO_ENCODE_ARGS = [ - '-c:v', 'libx264', - '-preset', 'medium', - '-profile:v', 'main', - '-level', '4.0', - '-crf', '23', - '-pix_fmt', 'yuv420p', -] +def get_video_encode_args(hw_accel: str = HW_ACCEL_NONE) -> List[str]: + """ + 根据硬件加速配置获取视频编码参数 + + Args: + hw_accel: 硬件加速类型 (none, qsv, cuda) + + Returns: + FFmpeg 视频编码参数列表 + """ + if hw_accel == HW_ACCEL_QSV: + params = VIDEO_ENCODE_PARAMS_QSV + return [ + '-c:v', params['codec'], + '-preset', params['preset'], + '-profile:v', params['profile'], + '-level', params['level'], + '-global_quality', params['global_quality'], + '-look_ahead', params['look_ahead'], + ] + elif hw_accel == HW_ACCEL_CUDA: + params = VIDEO_ENCODE_PARAMS_CUDA + return [ + '-c:v', params['codec'], + '-preset', params['preset'], + '-profile:v', params['profile'], + '-level', params['level'], + '-rc', params['rc'], + '-cq', params['cq'], + '-b:v', '0', # 配合 vbr 模式使用 cq + ] + else: + # 软件编码（默认） + params = VIDEO_ENCODE_PARAMS + return [ + '-c:v', params['codec'], + '-preset', params['preset'], + '-profile:v', params['profile'], + '-level', params['level'], + '-crf', params['crf'], + '-pix_fmt', params['pix_fmt'], + ] + + +def get_hwaccel_decode_args(hw_accel: str = HW_ACCEL_NONE) -> List[str]: + """ + 获取硬件加速解码参数（输入文件之前使用） + + Args: + hw_accel: 硬件加速类型 (none, qsv, cuda) + + Returns: + FFmpeg 硬件加速解码参数列表 + """ + if hw_accel == HW_ACCEL_CUDA: + # CUDA 硬件加速解码 + # 注意：使用 cuda 作为 hwaccel，但输出到系统内存以便 CPU 滤镜处理 + return ['-hwaccel', 'cuda', '-hwaccel_output_format', 'cuda'] + elif hw_accel == HW_ACCEL_QSV: + # QSV 硬件加速解码 + return ['-hwaccel', 'qsv', '-hwaccel_output_format', 'qsv'] + else: + return [] + + +def get_hwaccel_filter_prefix(hw_accel: str = HW_ACCEL_NONE) -> str: + """ + 获取硬件加速滤镜前缀（用于 hwdownload 从 GPU 到 CPU） + + 注意：由于大多数复杂滤镜（如 lut3d, overlay, crop 等）不支持硬件表面， + 我们需要在滤镜链开始时将硬件表面下载到系统内存。 + + Args: + hw_accel: 硬件加速类型 + + Returns: + 需要添加到滤镜链开头的 hwdownload 滤镜字符串 + """ + if hw_accel == HW_ACCEL_CUDA: + return 'hwdownload,format=nv12,' + elif hw_accel == HW_ACCEL_QSV: + return 'hwdownload,format=nv12,' + else: + return '' + + +# v2 统一视频编码参数（兼容旧代码，使用软件编码） +VIDEO_ENCODE_ARGS = get_video_encode_args(HW_ACCEL_NONE) # v2 统一音频编码参数 AUDIO_ENCODE_ARGS = [ @@ -178,6 +261,33 @@ class BaseHandler(TaskHandler, ABC): self.config = config self.api_client = api_client + def get_video_encode_args(self) -> List[str]: + """ + 获取当前配置的视频编码参数 + + Returns: + FFmpeg 视频编码参数列表 + """ + return get_video_encode_args(self.config.hw_accel) + + def get_hwaccel_decode_args(self) -> List[str]: + """ + 获取硬件加速解码参数（在输入文件之前使用） + + Returns: + FFmpeg 硬件加速解码参数列表 + """ + return get_hwaccel_decode_args(self.config.hw_accel) + + def get_hwaccel_filter_prefix(self) -> str: + """ + 获取硬件加速滤镜前缀 + + Returns: + 需要添加到滤镜链开头的 hwdownload 滤镜字符串 + """ + return get_hwaccel_filter_prefix(self.config.hw_accel) + def before_handle(self, task: Task) -> None: """处理前钩子""" logger.debug(f"[task:{task.task_id}] Before handle: {task.task_type.value}") diff --git a/handlers/compose_transition.py b/handlers/compose_transition.py index e6ea62e..cf8dba6 100644 --- a/handlers/compose_transition.py +++ b/handlers/compose_transition.py @@ -10,7 +10,7 @@ import os import logging from typing import List, Optional -from handlers.base import BaseHandler, VIDEO_ENCODE_ARGS +from handlers.base import BaseHandler from domain.task import Task, TaskType, TransitionConfig, TRANSITION_TYPES from domain.result import TaskResult, ErrorCode @@ -235,8 +235,8 @@ class ComposeTransitionHandler(BaseHandler): '-map', '[outv]', ] - # 编码参数（与片段视频一致） - cmd.extend(VIDEO_ENCODE_ARGS) + # 编码参数（根据硬件加速配置动态获取） + cmd.extend(self.get_video_encode_args()) # 帧率 fps = output_spec.fps diff --git a/handlers/render_video.py b/handlers/render_video.py index 6a76299..b9158a0 100644 --- a/handlers/render_video.py +++ b/handlers/render_video.py @@ -10,7 +10,7 @@ import os import logging from typing import List, Optional, Tuple -from handlers.base import BaseHandler, VIDEO_ENCODE_ARGS +from handlers.base import BaseHandler from domain.task import Task, TaskType, RenderSpec, OutputSpec, Effect from domain.result import TaskResult, ErrorCode @@ -170,6 +170,11 @@ class RenderSegmentVideoHandler(BaseHandler): """ cmd = ['ffmpeg', '-y', '-hide_banner'] + # 硬件加速解码参数（在输入文件之前） + hwaccel_args = self.get_hwaccel_decode_args() + if hwaccel_args: + cmd.extend(hwaccel_args) + # 输入文件 cmd.extend(['-i', input_file]) @@ -196,8 +201,8 @@ class RenderSegmentVideoHandler(BaseHandler): elif filters: cmd.extend(['-vf', filters]) - # 编码参数（v2 统一参数） - cmd.extend(VIDEO_ENCODE_ARGS) + # 编码参数（根据硬件加速配置动态获取） + cmd.extend(self.get_video_encode_args()) # 帧率 fps = output_spec.fps @@ -253,6 +258,12 @@ class RenderSegmentVideoHandler(BaseHandler): effects = render_spec.get_effects() has_camera_shot = any(e.effect_type == 'cameraShot' for e in effects) + # 硬件加速时需要先 hwdownload（将 GPU 表面下载到系统内存） + hwaccel_prefix = self.get_hwaccel_filter_prefix() + if hwaccel_prefix: + # 去掉末尾的逗号，作为第一个滤镜 + filters.append(hwaccel_prefix.rstrip(',')) + # 1. 变速处理 speed = float(render_spec.speed) if render_spec.speed else 1.0 if speed != 1.0 and speed > 0: @@ -304,7 +315,8 @@ class RenderSegmentVideoHandler(BaseHandler): fps=fps, has_overlay=has_overlay, overlap_head_ms=overlap_head_ms, - overlap_tail_ms=overlap_tail_ms + overlap_tail_ms=overlap_tail_ms, + use_hwdownload=bool(hwaccel_prefix) ) # 6. 帧冻结（tpad）- 用于转场 overlap 区域 @@ -337,7 +349,8 @@ class RenderSegmentVideoHandler(BaseHandler): fps: int, has_overlay: bool = False, overlap_head_ms: int = 0, - overlap_tail_ms: int = 0 + overlap_tail_ms: int = 0, + use_hwdownload: bool = False ) -> str: """ 构建包含特效的 filter_complex 滤镜图 @@ -351,6 +364,7 @@ class RenderSegmentVideoHandler(BaseHandler): has_overlay: 是否有叠加层 overlap_head_ms: 头部 overlap 时长 overlap_tail_ms: 尾部 overlap 时长 + use_hwdownload: 是否使用了硬件加速解码（已在 base_filters 中包含 hwdownload） Returns: filter_complex 格式的滤镜字符串 diff --git a/services/api_client.py b/services/api_client.py index a5615ed..7888240 100644 --- a/services/api_client.py +++ b/services/api_client.py @@ -12,6 +12,7 @@ from typing import Dict, List, Optional, Any from domain.task import Task from domain.config import WorkerConfig +from util.system import get_hw_accel_info_str logger = logging.getLogger(__name__) @@ -338,7 +339,9 @@ class APIClientV2: 'cpu': f"{psutil.cpu_count()} cores", 'memory': f"{psutil.virtual_memory().total // (1024**3)}GB", 'cpuUsage': f"{psutil.cpu_percent()}%", - 'memoryAvailable': f"{psutil.virtual_memory().available // (1024**3)}GB" + 'memoryAvailable': f"{psutil.virtual_memory().available // (1024**3)}GB", + 'hwAccelConfig': self.config.hw_accel, # 当前配置的硬件加速 + 'hwAccelSupport': get_hw_accel_info_str(), # 系统支持的硬件加速 } # 尝试获取 GPU 信息 diff --git a/util/system.py b/util/system.py index 433307d..efd24c1 100644 --- a/util/system.py +++ b/util/system.py @@ -8,10 +8,10 @@ import os import platform import subprocess -from typing import Optional +from typing import Optional, Dict, Any import psutil -from constant import SOFTWARE_VERSION, DEFAULT_CAPABILITIES +from constant import SOFTWARE_VERSION, DEFAULT_CAPABILITIES, HW_ACCEL_NONE, HW_ACCEL_QSV, HW_ACCEL_CUDA def get_sys_info(): @@ -101,3 +101,166 @@ def get_ffmpeg_version() -> str: pass return 'unknown' + + +def check_ffmpeg_encoder(encoder: str) -> bool: + """ + 检查 FFmpeg 是否支持指定的编码器 + + Args: + encoder: 编码器名称，如 'h264_nvenc', 'h264_qsv' + + Returns: + bool: 是否支持该编码器 + """ + try: + result = subprocess.run( + ['ffmpeg', '-hide_banner', '-encoders'], + capture_output=True, + text=True, + timeout=5 + ) + if result.returncode == 0: + return encoder in result.stdout + except Exception: + pass + + return False + + +def check_ffmpeg_decoder(decoder: str) -> bool: + """ + 检查 FFmpeg 是否支持指定的解码器 + + Args: + decoder: 解码器名称，如 'h264_cuvid', 'h264_qsv' + + Returns: + bool: 是否支持该解码器 + """ + try: + result = subprocess.run( + ['ffmpeg', '-hide_banner', '-decoders'], + capture_output=True, + text=True, + timeout=5 + ) + if result.returncode == 0: + return decoder in result.stdout + except Exception: + pass + + return False + + +def check_ffmpeg_hwaccel(hwaccel: str) -> bool: + """ + 检查 FFmpeg 是否支持指定的硬件加速方法 + + Args: + hwaccel: 硬件加速方法，如 'cuda', 'qsv', 'dxva2', 'd3d11va' + + Returns: + bool: 是否支持该硬件加速方法 + """ + try: + result = subprocess.run( + ['ffmpeg', '-hide_banner', '-hwaccels'], + capture_output=True, + text=True, + timeout=5 + ) + if result.returncode == 0: + return hwaccel in result.stdout + except Exception: + pass + + return False + + +def detect_hw_accel_support() -> Dict[str, Any]: + """ + 检测系统的硬件加速支持情况 + + Returns: + dict: 硬件加速支持信息 + { + 'cuda': { + 'available': bool, + 'gpu': str or None, + 'encoder': bool, # h264_nvenc + 'decoder': bool, # h264_cuvid + }, + 'qsv': { + 'available': bool, + 'encoder': bool, # h264_qsv + 'decoder': bool, # h264_qsv + }, + 'recommended': str # 推荐的加速方式: 'cuda', 'qsv', 'none' + } + """ + result = { + 'cuda': { + 'available': False, + 'gpu': None, + 'encoder': False, + 'decoder': False, + }, + 'qsv': { + 'available': False, + 'encoder': False, + 'decoder': False, + }, + 'recommended': HW_ACCEL_NONE + } + + # 检测 CUDA/NVENC 支持 + gpu_info = get_gpu_info() + if gpu_info: + result['cuda']['gpu'] = gpu_info + result['cuda']['available'] = check_ffmpeg_hwaccel('cuda') + result['cuda']['encoder'] = check_ffmpeg_encoder('h264_nvenc') + result['cuda']['decoder'] = check_ffmpeg_decoder('h264_cuvid') + + # 检测 QSV 支持 + result['qsv']['available'] = check_ffmpeg_hwaccel('qsv') + result['qsv']['encoder'] = check_ffmpeg_encoder('h264_qsv') + result['qsv']['decoder'] = check_ffmpeg_decoder('h264_qsv') + + # 推荐硬件加速方式（优先 CUDA，其次 QSV） + if result['cuda']['available'] and result['cuda']['encoder']: + result['recommended'] = HW_ACCEL_CUDA + elif result['qsv']['available'] and result['qsv']['encoder']: + result['recommended'] = HW_ACCEL_QSV + + return result + + +def get_hw_accel_info_str() -> str: + """ + 获取硬件加速支持信息的可读字符串 + + Returns: + str: 硬件加速支持信息描述 + """ + support = detect_hw_accel_support() + + parts = [] + + if support['cuda']['available']: + gpu = support['cuda']['gpu'] or 'Unknown GPU' + status = 'encoder+decoder' if support['cuda']['encoder'] and support['cuda']['decoder'] else ( + 'encoder only' if support['cuda']['encoder'] else 'decoder only' if support['cuda']['decoder'] else 'hwaccel only' + ) + parts.append(f"CUDA({gpu}, {status})") + + if support['qsv']['available']: + status = 'encoder+decoder' if support['qsv']['encoder'] and support['qsv']['decoder'] else ( + 'encoder only' if support['qsv']['encoder'] else 'decoder only' if support['qsv']['decoder'] else 'hwaccel only' + ) + parts.append(f"QSV({status})") + + if not parts: + return "No hardware acceleration available" + + return ', '.join(parts) + f" [recommended: {support['recommended']}]"