execute()

template<ComputeData InputType = std::vector<Kakshya::DataVariant>, ComputeData OutputType = InputType>

virtual output_type MayaFlux::Yantra::GpuExecutionContext< InputType, OutputType >::execute ( const input_type &  input, const ExecutionContext &  ctx  )

inlinevirtual

Dispatch to GPU and reconstruct a typed output Datum.

Routes to dispatch_core_chained for CHAINED mode; all other modes use dispatch_core. Both paths are defined in GpuDispatchCore.cpp.

Parameters

input	Input Datum to process.
ctx	ExecutionContext; CHAINED mode requires 'pass_count' and 'pc_updater' in execution_metadata.

Exceptions

std::runtime_error

If GPU initialisation fails.

Reimplemented in MayaFlux::Yantra::ShaderExecutionContext< InputType, OutputType >.

Definition at line 60 of file GpuExecutionContext.hpp.

    {
        if (!ensure_gpu_ready()) {
            error<std::runtime_error>(
                Journal::Component::Yantra,
                Journal::Context::BufferProcessing,
                std::source_location::current(),
                "GpuExecutionContext: GPU initialisation failed");
        }
 
        auto [ch_copies, structure_info] = extract_inputs(input);
 
        const GpuChannelResult raw = (ctx.mode == ExecutionMode::CHAINED)
            ? dispatch_core_chained(ch_copies, structure_info, ctx)
            : dispatch_core(ch_copies, structure_info);
 
        return collect_gpu_outputs(raw, ch_copies, structure_info);
    }

References MayaFlux::Yantra::ExecutionContext::mode.