ludus::graphics::rhi::NativeComputePipelineDescription¶

Struct

Native compute adapter for libraries using a dedicated queue on the same device. RHI owns the queue and orders shared-buffer producers/consumers with native events. Owner-thread callbacks must not reenter RHI, wait for GPU completion or use an unprepared queue. Only Metal storage-buffer bindings are currently admitted. Vulkan/browser reject explicitly.

Declared in modules/graphics/rhi/include/ludus/graphics/rhi/native_compute.h

#include <ludus/graphics/rhi/native_compute.h>

struct ludus::graphics::rhi::NativeComputePipelineDescription final

Fields¶

Bindings¶

BindingSetHandle ludus::graphics::rhi::NativeComputePipelineDescription::Bindings {}

Ready, immutable, non-dynamic Compute storage-only snapshot.

Related types: BindingSetHandle

Declared in modules/graphics/rhi/include/ludus/graphics/rhi/native_compute.h

ContextOrNull¶

void* ludus::graphics::rhi::NativeComputePipelineDescription::ContextOrNull = nullptr

Caller context; retained on successful creation through GPU retirement.

Declared in modules/graphics/rhi/include/ludus/graphics/rhi/native_compute.h

PrepareOrNull¶

RasterStatus(*) ludus::graphics::rhi::NativeComputePipelineDescription::PrepareOrNull(void *, const NativeDeviceIdentity &, foundation::BorrowedView< const NativeComputeBuffer >) noexcept(void *, const NativeDeviceIdentity &, foundation::BorrowedView< const NativeComputeBuffer >) noexcept = nullptr

Prepare/cache native objects between frames. Buffers appear in layout order. The array view expires on return; copied descriptors remain valid through pipeline retirement. Pending retains context and bindings and polls Prepare again without blocking. Private setup may run on a worker; bound buffers must remain untouched until Submit. Drain must join pending private preparation during device shutdown.

Related types: RasterStatus

Declared in modules/graphics/rhi/include/ludus/graphics/rhi/native_compute.h

SubmitOrNull¶

RasterStatus(*) ludus::graphics::rhi::NativeComputePipelineDescription::SubmitOrNull(void *) noexcept(void *) noexcept = nullptr

Enqueue GPU work on the exact prepared private queue. RHI orders the prior producer before it, and subsequent consumers/completion after it with native events.

Related types: RasterStatus

Declared in modules/graphics/rhi/include/ludus/graphics/rhi/native_compute.h

RetainOrNull¶

void(*) ludus::graphics::rhi::NativeComputePipelineDescription::RetainOrNull(void *) noexcept(void *) noexcept = nullptr

Retain the context without failure or allocation.

Declared in modules/graphics/rhi/include/ludus/graphics/rhi/native_compute.h

ReleaseOrNull¶

void(*) ludus::graphics::rhi::NativeComputePipelineDescription::ReleaseOrNull(void *) noexcept(void *) noexcept = nullptr

Release after the last retained GPU use, also on shutdown.

Declared in modules/graphics/rhi/include/ludus/graphics/rhi/native_compute.h

DrainOrNull¶

void(*) ludus::graphics::rhi::NativeComputePipelineDescription::DrainOrNull(void *) noexcept(void *) noexcept = nullptr

Shutdown-only fence for pending private preparation, called before native buffers/queue retire. Must drain even if the caller still retains other context references; must not reenter RHI.

Declared in modules/graphics/rhi/include/ludus/graphics/rhi/native_compute.h