prepared_tensor_cache.h

namespace onnx_light

Alias that makes onnx-light headers compatible with code that references ONNX_LIGHT_NAMESPACE (the macro used in the standard onnx package).

Set to ONNX_LIGHT_NAMESPACE so both names resolve to the same namespace.

Symbol-visibility attribute for the public onnx-light C++ API.

Maps the upstream compatibility macro to onnx-light’s explicit proto ABI annotation. This keeps declarations from vendored ONNX headers visible when lib_onnx_proto uses hidden visibility by default.

Namespace alias so that ONNX C++ code (and consumers such as onnxruntime) that refers to the literal onnx namespace — rather than the ONNX_NAMESPACE macro — resolves to the onnx-light namespace. The standard onnx package lives in namespace onnx; onnx-light uses onnx_light (via ONNX_LIGHT_NAMESPACE), so this alias keeps onnx-light a true drop-in. It is only introduced when the onnx-light namespace differs from onnx.

namespace core

namespace runtime

Typedefs

using PreparedSourceLoader = std::function<std::vector<uint8_t>()>

using PreparedTensorPrepacker = std::function<std::vector<uint8_t>(const std::vector<uint8_t>&)>

using PreparedTensorPublisher = std::function<void(const std::vector<uint8_t>&)>

Enums

enum class PreparedCacheMissReason

Values:

enumerator kNone

enumerator kNotFound

enumerator kCorrupt

enumerator kStaleSource

enumerator kWrongArchitecture

enumerator kWrongIsa

enumerator kWrongRuntime

enumerator kWrongLayout

enumerator kWrongFormat

class PreparedTensorCache

#include <prepared_tensor_cache.h>

Loads compatible prepared tensors before their portable source and persists misses.

Public Functions

PreparedTensorCache() = default

~PreparedTensorCache()

PreparedTensorCache(const PreparedTensorCache&) = delete

PreparedTensorCache &operator=(const PreparedTensorCache&) = delete

PreparedTensorLoadResult LoadOrPrepare(const std::filesystem::path &cache_path, const PreparedTensorMetadata &required_metadata, const PreparedSourceLoader &load_source, const PreparedTensorPrepacker &prepack, const PreparedTensorPublisher &publish)

Publishes a compatible cached payload or prepares the portable source on a miss.

Cache persistence starts only after the complete prepared value is published.

void WaitForBackgroundWrites()

Waits for all pending atomic cache writes.

std::vector<std::string> TakePersistenceDiagnostics()

Waits for pending writes and returns their diagnostics.

Private Functions

void PersistInBackground(std::filesystem::path cache_path, PreparedTensorMetadata metadata, std::vector<uint8_t> payload)

void WriterLoop()

Private Members

std::mutex mutex_

std::condition_variable writer_ready_

std::deque<PersistenceRequest> pending_

std::thread writer_

size_t pending_writes_ = 0

bool stop_writer_ = false

std::vector<std::string> persistence_diagnostics_

struct PersistenceRequest

Public Members

std::filesystem::path cache_path

PreparedTensorMetadata metadata

std::vector<uint8_t> payload

struct PreparedTensorLoadResult

#include <prepared_tensor_cache.h>

Public Members

bool cache_hit = false

PreparedCacheMissReason miss_reason = PreparedCacheMissReason::kNone

std::string diagnostic

struct PreparedTensorMetadata

#include <prepared_tensor_cache.h>

Public Members

std::string source_digest

std::string architecture

platform::CpuFeatureSet required_isa

std::string runtime

std::string runtime_version

std::string kernel_layout

std::string format_version