Struct MultiGpuParams 

pub struct MultiGpuParams<'a> {

    pub input: Bytes,
    pub rungs: &'a [Rung],
    pub header: DemuxHeader,
    pub source_color_metadata: ColorMetadata,
    pub source_pixel_format: PixelFormat,
    pub tonemap_to_sdr: bool,
    pub output_color_metadata: ColorMetadata,
    pub output_pixel_format: PixelFormat,
    pub needs_downsample: bool,
    pub frame_rate: f64,
    pub gpu_pool: Arc<GpuPool>,
    pub gpu_indices: Vec<u32>,
    pub decode_gpu: Option<u32>,
    pub output_root: PathBuf,
    pub timescale: u32,
    pub per_frame_ticks: u32,
    pub keyframe_interval: u32,
    pub segment_target_ticks: u64,
    pub total_input_frames: u64,
    pub constant_qp: bool,
}

Inputs to run_multigpu_hls.

Fields

input: Bytes

rungs: &'a [Rung]

header: DemuxHeader

source_color_metadata: ColorMetadata

source_pixel_format: PixelFormat

tonemap_to_sdr: bool

Whether the decode pump tonemaps HDR→SDR (from the spec’s ColorPolicy).

output_color_metadata: ColorMetadata

Resolved output color metadata + pixel format the encoders target (from OutputSpec::resolve_output).

output_pixel_format: PixelFormat

needs_downsample: bool

frame_rate: f64

gpu_pool: Arc<GpuPool>

gpu_indices: Vec<u32>

GPU indices the encode policy selected, in detection order. The decode pump pins to these (round-robin for per-rung pumps) so decode honors the same Family / SingleGpu / AllGpus constraint as encode. Empty ⇒ the decoder dispatch auto-selects (legacy behavior).

decode_gpu: Option<u32>

Explicit decode-pump GPU override. Some(i) forces every decode pump onto GPU i regardless of gpu_indices; None follows the policy.

output_root: PathBuf

timescale: u32

per_frame_ticks: u32

keyframe_interval: u32

segment_target_ticks: u64

total_input_frames: u64

constant_qp: bool

Force constant-QP chunk encoding (single-file ChunkSeamMode::ParallelConstQp) so stitched chunk seams are quality-flat. false for HLS (segments are independent) and the default Parallel single-file mode.