Skip to main content
tenferro_gpu::cuda surface inventory
- Items in the table: 292 (
#[doc(hidden)] lines under cuda/: 130; table rows carrying the mark: 130)
- Verdicts: private 149, public 101, workspace-internal 42.
- Regenerate:
python3 scripts/cuda_surface_inventory.py; verdicts live in docs/internals/cuda-surface-verdicts.json.
- Scope:
pub/pub(crate)/pub(super) items declared under crates/tenferro-gpu/src/cuda/** – modules, structs, enums, unions, traits, type aliases, functions, constants and statics, plus the methods of their impl blocks – excluding #[cfg(test)] items, #[cfg(test)] impls and test modules. Struct fields and enum variants are not rows; a #[doc(hidden)] on one still counts in the mark total.
- Reference counts: other workspace crates (excluding
tenferro-gpu), the crate’s own examples/, tests/ and benches/, and the user-facing guides (docs/guides, docs/getting-started, docs/reference, docs/api, docs/tutorials, docs/tutorial-code). Design notes, plans and worklogs are not counted: they say what someone once meant, not who uses the item.
- Counts are name-based. A module-level item counts only in files that name
tenferro_gpu; a method counts its qualified Type::name uses and its .name( receiver calls in files that name the owner type, so a same-named method on another type can still raise a count. Treat a count as a lead and the evidence column as the review.
- Evidence is a citation, not a verification: the script checks only that the cited file exists and that the line field is numeric. It does not read the line, so a
(cited at ...) site shows where a reviewer looked, not that the code there is the item on the row – a same-named method on another receiver passes. A row whose caller evidence nobody could confirm says so in its verdict reason instead of citing a site. | item | kind | vis | hidden | reachable from cuda | other crates | examples/tests | guides | verdict | evidence | |—|—|—|—|—|—|—|—|—|—| | crate::cuda::CudaBackend | struct | pub(crate) | yes | no | 0 — | 0 — | 0 — | private | the seam marker: nothing outside cuda/ names it; only the crate-private gpu::Buffer<CudaBackend> in storage.rs uses it, and the documentation hits are design notes from before the cutover | | tenferro_gpu::cuda::cuda_capabilities | fn | pub | no | yes | 0 — | 2 crates/tenferro-gpu/tests/seam_capability.rs:165, crates/tenferro-gpu/tests/seam_capability.rs:166 | 1 docs/guides/devices-and-gpu.md:322 | public | the already-documented capability table (issue #2298’s one documented symbol) | | tenferro_gpu::cuda::Contraction | struct | pub | yes | yes | 0 — | 11 crates/tenferro-gpu/tests/seam_contraction.rs:8, crates/tenferro-gpu/tests/seam_contraction.rs:51, crates/tenferro-gpu/tests/seam_contraction.rs:109 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::Cutensor | struct | pub | yes | yes | 0 — | 10 crates/tenferro-gpu/tests/seam_contraction.rs:8, crates/tenferro-gpu/tests/seam_contraction.rs:50, crates/tenferro-gpu/tests/seam_contraction.rs:108 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::Cutensor::cached_plans | fn | pub | no | yes | 0 — | 3 crates/tenferro-gpu/tests/seam_contraction.rs:140, crates/tenferro-gpu/tests/seam_contraction.rs:507, crates/tenferro-gpu/tests/seam_contraction.rs:556 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::Cutensor::cold_plans | fn | pub | no | yes | 0 — | 3 crates/tenferro-gpu/tests/seam_contraction.rs:90, crates/tenferro-gpu/tests/seam_contraction.rs:138, crates/tenferro-gpu/tests/seam_contraction.rs:506 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::Cutensor::contract | fn | pub | no | yes | 0 — | 10 crates/tenferro-gpu/tests/seam_contraction.rs:77, crates/tenferro-gpu/tests/seam_contraction.rs:136, crates/tenferro-gpu/tests/seam_contraction.rs:160 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::Cutensor::new | fn | pub | no | yes | 0 — | 9 crates/tenferro-gpu/tests/seam_contraction.rs:50, crates/tenferro-gpu/tests/seam_contraction.rs:108, crates/tenferro-gpu/tests/seam_contraction.rs:158 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::Cutensor::warm_hits | fn | pub | no | yes | 0 — | 2 crates/tenferro-gpu/tests/seam_contraction.rs:91, crates/tenferro-gpu/tests/seam_contraction.rs:139 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | crate::cuda::cutensor::MAX_TENSOR_MODES | const | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::Operand | struct | pub | yes | yes | 0 — | 33 crates/tenferro-gpu/tests/seam_contraction.rs:8, crates/tenferro-gpu/tests/seam_contraction.rs:53, crates/tenferro-gpu/tests/seam_contraction.rs:59 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | crate::cuda::device | mod | pub(crate) | no | no | 295 crates/tenferro-ad/src/eager.rs:1820, crates/tenferro-ad/src/eager.rs:1821, crates/tenferro-ad/src/eager.rs:1822 | 335 crates/tenferro-gpu/examples/cuda_quickstart.rs:12, crates/tenferro-gpu/examples/cuda_quickstart.rs:18, crates/tenferro-gpu/examples/cuda_quickstart.rs:22 | 1 docs/reference/libtorch.md:184 | private | pub(crate) module: the device types are public through tenferro_gpu::cuda:: re-exports, not through this path | | tenferro_gpu::cuda::CudaDeviceError | enum | pub | no | yes | 0 — | 0 — | 0 — | public | the error of cuda_devices and CudaEngine::open | | tenferro_gpu::cuda::CudaDeviceError::device | fn | pub | no | yes | 0 — | 0 — | 1 docs/reference/libtorch.md:184 | public | documented accessor (has a doctest) | | tenferro_gpu::cuda::CudaDeviceError::discovered | fn | pub | no | yes | 0 — | 0 — | 0 — | public | documented accessor (has a doctest) | | tenferro_gpu::cuda::CudaDeviceError::operation | fn | pub | no | yes | 0 — | 0 — | 0 — | public | documented accessor (has a doctest) | | tenferro_gpu::cuda::CudaDeviceError::requested | fn | pub | no | yes | 0 — | 0 — | 0 — | public | documented accessor (has a doctest) | | tenferro_gpu::cuda::CudaDeviceId | struct | pub | no | yes | 44 crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:7, crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:105, crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:148 | 82 crates/tenferro-gpu/tests/blas1_cuda.rs:21, crates/tenferro-gpu/tests/blas1_cuda.rs:44, crates/tenferro-gpu/tests/chunked_staging.rs:11 | 0 — | public | the device id every documented CUDA call takes; sibling crates and the guides name it | | tenferro_gpu::cuda::CudaDeviceId::from_ordinal | fn | pub | no | yes | 31 crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:105, crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:148, crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:189 | 49 crates/tenferro-gpu/tests/blas1_cuda.rs:44, crates/tenferro-gpu/tests/chunked_staging.rs:20, crates/tenferro-gpu/tests/coalesced_run.rs:17 | 0 — | public | the documented constructor of the device id; sibling crates and the guides use it | | tenferro_gpu::cuda::CudaDeviceId::ordinal | fn | pub | no | yes | 0 — | 0 — | 0 — | public | the only way back to the ordinal; tenferro-fft calls it | | tenferro_gpu::cuda::CudaDeviceInfo | struct | pub | no | yes | 0 — | 0 — | 0 — | public | the value cuda_devices returns; its fields are private | | tenferro_gpu::cuda::CudaDeviceInfo::compute_capability | fn | pub | no | yes | 0 — | 0 — | 0 — | public | accessor of a public value type with private fields | | tenferro_gpu::cuda::CudaDeviceInfo::id | fn | pub | no | yes | 0 — | 0 — | 10 docs/guides/devices-and-gpu.md:118, docs/guides/tenferro-fft.md:156, docs/guides/troubleshooting.md:50 | public | the accessor the quickstart uses to open an engine (device.id()) | | tenferro_gpu::cuda::CudaDeviceInfo::name | fn | pub | no | yes | 0 — | 0 — | 0 — | public | accessor of a public value type with private fields | | crate::cuda::device::CudaDeviceInfo::new | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaDeviceInfo::total_memory_bytes | fn | pub | no | yes | 0 — | 0 — | 0 — | public | accessor of a public value type with private fields | | tenferro_gpu::cuda::CudaDeviceInfo::uuid | fn | pub | no | yes | 0 — | 0 — | 0 — | public | accessor of a public value type with private fields | | crate::cuda::device::DiscoveryDriver | trait | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::cuda_devices | fn | pub | no | yes | 3 crates/tenferro-ad/src/extension/tests.rs:868, crates/tenferro-ad/src/extension/tests.rs:914, crates/tenferro-ad/src/extension/tests.rs:958 | 2 crates/tenferro-gpu/examples/cuda_quickstart.rs:18, crates/tenferro-gpu/examples/cuda_session_transfer.rs:51 | 12 docs/guides/devices-and-gpu.md:17, docs/guides/devices-and-gpu.md:98, docs/guides/devices-and-gpu.md:114 | public | documented device discovery: cuda_devices in the quickstart, the guides and a sibling test | | crate::cuda::device::discover_with | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::CudaEngine::dispatcher | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::arithmetic | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::binary | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::clamp | fn | pub(crate) | no | no | 0 — | 0 — | 4 docs/guides/devices-and-gpu.md:355, docs/guides/tensor-operations.md:145, docs/reference/libtorch.md:531 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::clamp_buffers | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::combined_strided | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::compare | fn | pub(crate) | no | no | 0 — | 0 — | 12 docs/guides/devices-and-gpu.md:353, docs/guides/tensor-operations.md:126, docs/guides/tensor-operations.md:144 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::contiguous_to_view | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::convert | fn | pub(crate) | no | no | 0 — | 0 — | 11 docs/guides/devices-and-gpu.md:392, docs/guides/eager-operations.md:84, docs/guides/tensor-operations.md:205 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::copy_exact | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::dynamic_slice | fn | pub(crate) | no | no | 0 — | 0 — | 5 docs/guides/devices-and-gpu.md:398, docs/guides/eager-operations.md:89, docs/reference/jax-stablehlo-needed.md:87 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::dynamic_update_slice | fn | pub(crate) | no | no | 0 — | 0 — | 3 docs/guides/devices-and-gpu.md:399, docs/reference/jax-stablehlo-needed.md:87, docs/reference/stablehlo-primitives.md:103 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::for_entries | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::gather | fn | pub(crate) | no | no | 0 — | 0 — | 10 docs/getting-started/api-migration.md:95, docs/guides/choosing-an-api.md:102, docs/guides/devices-and-gpu.md:393 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::in_scope | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::pad | fn | pub(crate) | no | no | 0 — | 0 — | 4 docs/guides/devices-and-gpu.md:395, docs/guides/eager-operations.md:89, docs/reference/jax-stablehlo-needed.md:86 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::prepare | fn | pub(crate) | no | no | 0 — | 0 — | 47 docs/getting-started/api-migration.md:24, docs/getting-started/core-concepts.md:169, docs/getting-started/index.md:218 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::reduce | fn | pub(crate) | no | no | 0 — | 0 — | 7 docs/reference/jax-stablehlo-needed.md:72, docs/reference/jax-stablehlo-needed.md:86, docs/reference/jax-stablehlo-needed.md:217 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::reduce_planned | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::run_planned | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::scatter_add | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::select | fn | pub(crate) | no | no | 0 — | 0 — | 10 docs/guides/devices-and-gpu.md:354, docs/reference/jax-stablehlo-needed.md:81, docs/reference/jax-stablehlo-needed.md:124 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::slice | fn | pub(crate) | no | no | 0 — | 0 — | 5 docs/guides/devices-and-gpu.md:395, docs/guides/eager-operations.md:89, docs/reference/jax-stablehlo-needed.md:87 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::strided_to_strided | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::triangular_view | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::unary | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::validate_float_indices | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::Dispatcher::validate_real_cast | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::IndexPointer | enum | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::PreparedDispatchers | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::dispatch::pod_arg | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaEngine | struct | pub | yes | yes | 196 crates/tenferro-ad/src/eager.rs:357, crates/tenferro-ad/src/eager_backend.rs:59, crates/tenferro-ad/src/extension/tests.rs:865 | 173 crates/tenferro-gpu/examples/cuda_quickstart.rs:3, crates/tenferro-gpu/examples/cuda_quickstart.rs:22, crates/tenferro-gpu/examples/cuda_session_transfer.rs:10 | 8 docs/getting-started/api-migration.md:22, docs/getting-started/core-concepts.md:16, docs/guides/choosing-a-backend.md:146 | public | the CUDA engine a user opens and drives; the quickstart and most sibling-crate files name it | | tenferro_gpu::cuda::CudaEngine::clear_vendor_resources | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | the redundancy audit (2026-11-04, item 9) records it as a deliberate downstream-facing hook; it stays public and #[doc(hidden)] | | crate::cuda::engine::CudaEngine::cutensor | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::engine::CudaEngine::device_id | fn | pub(crate) | no | no | 1 crates/tenferro-fft/src/cuda/mod.rs:85 | 1 crates/tenferro-gpu/tests/seam_storage.rs:70 | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::engine::CudaEngine::ensure_healthy | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaEngine::launch_completion_syncs | fn | pub | yes | yes | 8 crates/tenferro-ad/tests/integration/gpu_nn_composites.rs:707, crates/tenferro-ad/tests/integration/gpu_nn_composites.rs:718, crates/tenferro-ad/tests/integration/gpu_nn_composites.rs:734 | 2 crates/tenferro-gpu/tests/seam_full_reduction.rs:221, crates/tenferro-gpu/tests/seam_full_reduction.rs:224 | 0 — | workspace-internal | tenferro-ad and tenferro-linalg tests read the counter | | crate::cuda::engine::CudaEngine::note_staged_read | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::engine::CudaEngine::note_tiled_transpose | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaEngine::open | fn | pub | yes | yes | 35 crates/tenferro-ad/src/extension/tests.rs:869, crates/tenferro-ad/src/extension/tests.rs:915, crates/tenferro-ad/src/extension/tests.rs:961 | 50 crates/tenferro-gpu/examples/cuda_quickstart.rs:22, crates/tenferro-gpu/examples/cuda_session_transfer.rs:55, crates/tenferro-gpu/tests/blas1_cuda.rs:44 | 16 docs/getting-started/api-migration.md:22, docs/getting-started/index.md:28, docs/guides/choosing-a-backend.md:15 | public | the documented constructor: CudaEngine::open in the quickstart, the guides and sibling crates | | tenferro_gpu::cuda::CudaEngine::publish | fn | pub | yes | yes | 0 — | 9 crates/tenferro-gpu/tests/seam_publication.rs:61, crates/tenferro-gpu/tests/seam_publication.rs:80, crates/tenferro-gpu/tests/seam_publication.rs:106 | 0 — | public | the publication entry point; its return types are the publication surface below | | tenferro_gpu::cuda::CudaEngine::runtime | fn | pub | yes | yes | 149 crates/tenferro-ad/src/eager.rs:477, crates/tenferro-ad/src/eager.rs:579, crates/tenferro-ad/src/eager.rs:4427 | 320 crates/tenferro-gpu/examples/cuda_quickstart.rs:26, crates/tenferro-gpu/examples/cuda_quickstart.rs:27, crates/tenferro-gpu/examples/cuda_quickstart.rs:32 | 29 docs/guides/devices-and-gpu.md:88, docs/guides/devices-and-gpu.md:122, docs/guides/devices-and-gpu.md:123 | public | the documented accessor for the runtime: engine.runtime() in the quickstart, the guides and sibling crates | | tenferro_gpu::cuda::CudaEngine::runtime_engine_id | fn | pub | yes | yes | 6 crates/tenferro-einsum/tests/integration/cuda_eager_tensor.rs:70, crates/tenferro-fft/tests/cuda_fft/common.rs:404, crates/tenferro-linalg/tests/integration/deferred_completion.rs:52 | 0 — | 0 — | workspace-internal | tenferro-einsum, tenferro-fft and tenferro-linalg tests key extension modules by it | | tenferro_gpu::cuda::CudaEngine::runtime_identity | fn | pub | yes | yes | 5 crates/tenferro-fft/src/cuda/mod.rs:91, crates/tenferro-fft/tests/cuda_fft/common.rs:355, crates/tenferro-fft/tests/cuda_fft/common.rs:360 | 0 — | 0 — | workspace-internal | tenferro-fft keys its cuFFT plan cache by it (cited at crates/tenferro-fft/src/cuda/mod.rs:91) | | tenferro_gpu::cuda::CudaEngine::vendor_resource_inits | fn | pub | yes | yes | 2 crates/tenferro-linalg/tests/integration/deferred_completion.rs:100, crates/tenferro-linalg/tests/integration/deferred_completion.rs:124 | 0 — | 0 — | workspace-internal | tenferro-linalg’s deferred-completion test reads the counter | | tenferro_gpu::cuda::CudaEngine::vendor_resource_stats | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | the redundancy audit (2026-11-04, item 9) records it as a deliberate downstream-facing hook; it stays public and #[doc(hidden)] | | tenferro_gpu::cuda::CudaEngineError | enum | pub | yes | yes | 0 — | 0 — | 0 — | public | the error of CudaEngine::open, named by the module and runtime doctests | | crate::cuda::engine::DEFAULT_RUNTIME_ENGINE_ID | const | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::engine::default_runtime_engine_id | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::event_domain::CoalescedEventToken | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::event_domain::StreamEventDomainDriver | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::event_domain::StreamEventDomainDriver::new | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::event_domain::StreamEventToken | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::event_domain::event_domain_error | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::PreparedFusion | struct | pub | yes | yes | 0 — | 3 crates/tenferro-gpu/tests/seam_fusion.rs:8, crates/tenferro-gpu/tests/seam_fusion.rs:48, crates/tenferro-gpu/tests/seam_fusion.rs:67 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::PreparedFusion::dtype | fn | pub | no | yes | 0 — | 0 — | 19 docs/guides/einsum.md:246, docs/guides/einsum.md:247, docs/guides/tenferro-fft.md:175 | private | the count is the same-named method on other types; no other crate names PreparedFusion, so nothing outside cuda/ can call it | | tenferro_gpu::cuda::PreparedFusion::key | fn | pub | no | yes | 0 — | 0 — | 0 — | private | no code names it; only a design note mentions it, and the in-crate fusion::tests module reads it, so it is gated together with the seam tests | | tenferro_gpu::cuda::PreparedFusion::launch | fn | pub | no | yes | 0 — | 1 crates/tenferro-gpu/tests/seam_fusion.rs:52 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::PreparedFusion::prepare | fn | pub | no | yes | 0 — | 2 crates/tenferro-gpu/tests/seam_fusion.rs:48, crates/tenferro-gpu/tests/seam_fusion.rs:67 | 47 docs/getting-started/api-migration.md:24, docs/getting-started/core-concepts.md:169, docs/getting-started/index.md:218 | private | the count is the same-named method on other types; no other crate names PreparedFusion, so nothing outside cuda/ can call it | | tenferro_gpu::cuda::gpu_available | fn | pub | yes | yes | 83 crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:103, crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:147, crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:188 | 45 crates/tenferro-gpu/examples/cuda_quickstart.rs:10, crates/tenferro-gpu/examples/cuda_quickstart.rs:13, crates/tenferro-gpu/examples/cuda_session_transfer.rs:43 | 13 docs/guides/devices-and-gpu.md:15, docs/guides/devices-and-gpu.md:98, docs/guides/devices-and-gpu.md:106 | public | documented device probe; the guides, the quickstart and every GPU test call it, and issue #2298 asks for exactly this decision | | crate::cuda::identity | mod | pub(crate) | no | no | 30 crates/tenferro-ad/src/eager.rs:3139, crates/tenferro-fft/src/cuda/descriptor.rs:111, crates/tenferro-fft/src/cuda/descriptor.rs:144 | 35 crates/tenferro-gpu/tests/integration/session_contract.rs:17, crates/tenferro-gpu/tests/seam_contraction.rs:372, crates/tenferro-gpu/tests/seam_contraction.rs:426 | 0 — | private | pub(crate) module: the identity types are public through tenferro_gpu::cuda:: re-exports, not through this path | | tenferro_gpu::cuda::CudaComputeCapability | struct | pub | no | yes | 0 — | 0 — | 0 — | public | returned by CudaDeviceInfo::compute_capability | | tenferro_gpu::cuda::CudaDeviceUuid | struct | pub | no | yes | 0 — | 0 — | 0 — | public | documented device identity with doctests | | tenferro_gpu::cuda::CudaDeviceUuid::as_bytes | fn | pub | no | yes | 0 — | 0 — | 0 — | public | documented accessor with a doctest | | tenferro_gpu::cuda::CudaDeviceUuid::from_bytes | fn | pub | no | yes | 0 — | 0 — | 0 — | public | documented constructor with a doctest | | tenferro_gpu::cuda::CudaDeviceUuid::to_bytes | fn | pub | no | yes | 0 — | 0 — | 0 — | public | documented accessor with a doctest | | tenferro_gpu::cuda::GpuExtensionCapability | enum | pub | no | yes | 0 — | 0 — | 0 — | private | nobody constructs it and no public signature returns it; its rustdoc names CudaEngine::supports, which does not exist, so it is a leftover of the CubeCL-era seam to delete rather than document | | tenferro_gpu::cuda::KernelModules | struct | pub | yes | yes | 0 — | 6 crates/tenferro-gpu/tests/seam_launch.rs:14, crates/tenferro-gpu/tests/seam_launch.rs:57, crates/tenferro-gpu/tests/seam_launch.rs:72 | 0 — | workspace-internal | kernel-launch family kept workspace-internal by the #2298 decision: the extension-hook design says the RRQR kernels will need a module and launch surface, so it stays hidden pub until that hook lands (docs/design/cuda-surface.md section 4) | | tenferro_gpu::cuda::KernelModules::for_entries | fn | pub | yes | yes | 0 — | 5 crates/tenferro-gpu/tests/seam_launch.rs:57, crates/tenferro-gpu/tests/seam_launch.rs:72, crates/tenferro-gpu/tests/seam_launch.rs:89 | 0 — | workspace-internal | kernel-launch family kept workspace-internal by the #2298 decision: the extension-hook design says the RRQR kernels will need a module and launch surface, so it stays hidden pub until that hook lands (docs/design/cuda-surface.md section 4) | | tenferro_gpu::cuda::KernelModules::function | fn | pub | yes | yes | 0 — | 0 — | 0 — | workspace-internal | kernel-launch family kept workspace-internal by the #2298 decision: the extension-hook design says the RRQR kernels will need a module and launch surface, so it stays hidden pub until that hook lands (docs/design/cuda-surface.md section 4) | | crate::cuda::launch::KernelModules::in_scope | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::KernelModules::launch | fn | pub | yes | yes | 0 — | 5 crates/tenferro-gpu/tests/seam_launch.rs:97, crates/tenferro-gpu/tests/seam_launch.rs:138, crates/tenferro-gpu/tests/seam_launch.rs:187 | 0 — | workspace-internal | kernel-launch family kept workspace-internal by the #2298 decision: the extension-hook design says the RRQR kernels will need a module and launch surface, so it stays hidden pub until that hook lands (docs/design/cuda-surface.md section 4) | | tenferro_gpu::cuda::KernelModules::modules | fn | pub | yes | yes | 0 — | 1 crates/tenferro-gpu/tests/seam_launch.rs:59 | 0 — | workspace-internal | kernel-launch family kept workspace-internal by the #2298 decision: the extension-hook design says the RRQR kernels will need a module and launch surface, so it stays hidden pub until that hook lands (docs/design/cuda-surface.md section 4) | | crate::cuda::launch::KernelModules::prepare | fn | pub(crate) | no | no | 0 — | 0 — | 47 docs/getting-started/api-migration.md:24, docs/getting-started/core-concepts.md:169, docs/getting-started/index.md:218 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::device_buffer | fn | pub | yes | yes | 0 — | 8 crates/tenferro-gpu/tests/seam_launch.rs:101, crates/tenferro-gpu/tests/seam_launch.rs:102, crates/tenferro-gpu/tests/seam_launch.rs:142 | 0 — | workspace-internal | kernel-launch family kept workspace-internal by the #2298 decision: the extension-hook design says the RRQR kernels will need a module and launch surface, so it stays hidden pub until that hook lands (docs/design/cuda-surface.md section 4) | | tenferro_gpu::cuda::layout_arg | fn | pub | yes | yes | 0 — | 3 crates/tenferro-gpu/tests/seam_launch.rs:145, crates/tenferro-gpu/tests/seam_launch.rs:221, crates/tenferro-gpu/tests/seam_launch.rs:258 | 0 — | workspace-internal | kernel-launch family kept workspace-internal by the #2298 decision: the extension-hook design says the RRQR kernels will need a module and launch surface, so it stays hidden pub until that hook lands (docs/design/cuda-surface.md section 4) | | tenferro_gpu::cuda::PendingDownload | struct | pub | yes | yes | 0 — | 0 — | 0 — | public | the handle download_pending returns | | tenferro_gpu::cuda::PendingDownload::is_ready | fn | pub | yes | yes | 0 — | 0 — | 4 docs/guides/engines-and-transfers.md:122, docs/guides/engines-and-transfers.md:129, docs/guides/engines-and-transfers.md:155 | public | documented poll (guides name is_ready) | | crate::cuda::pending::PendingDownload::new | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::PendingDownload::wait | fn | pub | yes | yes | 0 — | 0 — | 7 docs/guides/engines-and-transfers.md:127, docs/guides/engines-and-transfers.md:129, docs/guides/engines-and-transfers.md:133 | public | documented wait (guides name wait) | | tenferro_gpu::cuda::PendingUpload | struct | pub | yes | yes | 0 — | 0 — | 0 — | public | the handle upload_pending returns | | tenferro_gpu::cuda::PendingUpload::is_ready | fn | pub | yes | yes | 0 — | 0 — | 4 docs/guides/engines-and-transfers.md:122, docs/guides/engines-and-transfers.md:129, docs/guides/engines-and-transfers.md:155 | public | documented poll (guides name is_ready) | | crate::cuda::pending::PendingUpload::new | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::PendingUpload::wait | fn | pub | yes | yes | 0 — | 0 — | 7 docs/guides/engines-and-transfers.md:127, docs/guides/engines-and-transfers.md:129, docs/guides/engines-and-transfers.md:133 | public | documented wait (guides name wait) | | tenferro_gpu::cuda::RejectedTransfer | struct | pub | yes | yes | 0 — | 0 — | 0 — | public | the value a rejected submission hands back inside TransferSubmitError | | tenferro_gpu::cuda::TransferSubmitError | enum | pub | yes | yes | 0 — | 12 crates/tenferro-gpu/tests/pending_download.rs:12, crates/tenferro-gpu/tests/pending_download.rs:195, crates/tenferro-gpu/tests/pending_download.rs:213 | 0 — | public | the documented error of the asynchronous transfers | | crate::cuda::pending::cuda_storage | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::PinnedHostBuffer | struct | pub | yes | yes | 0 — | 47 crates/tenferro-gpu/examples/cuda_session_transfer.rs:11, crates/tenferro-gpu/examples/cuda_session_transfer.rs:108, crates/tenferro-gpu/examples/cuda_session_transfer.rs:123 | 2 docs/guides/devices-and-gpu.md:69, docs/guides/engines-and-transfers.md:139 | public | documented pinned host staging buffer | | tenferro_gpu::cuda::PinnedHostBuffer::as_mut_slice | fn | pub | yes | yes | 0 — | 4 crates/tenferro-gpu/examples/cuda_session_transfer.rs:109, crates/tenferro-gpu/tests/pending_upload.rs:45, crates/tenferro-gpu/tests/seam_publication.rs:48 | 2 docs/guides/engines-and-transfers.md:118, docs/guides/engines-and-transfers.md:140 | public | documented accessor of the pinned buffer | | tenferro_gpu::cuda::PinnedHostBuffer::as_slice | fn | pub | yes | yes | 0 — | 29 crates/tenferro-gpu/examples/cuda_session_transfer.rs:125, crates/tenferro-gpu/examples/cuda_session_transfer.rs:126, crates/tenferro-gpu/examples/cuda_session_transfer.rs:129 | 93 docs/getting-started/api-migration.md:19, docs/getting-started/core-concepts.md:49, docs/getting-started/core-concepts.md:55 | public | documented accessor of the pinned buffer | | tenferro_gpu::cuda::PinnedHostBuffer::is_empty | fn | pub | yes | yes | 0 — | 2 crates/tenferro-gpu/tests/seam_publication.rs:170, crates/tenferro-gpu/tests/seam_publication.rs:191 | 2 docs/tutorials/eager-autodiff-pytorch-style.md:39, docs/tutorial-code/src/bin/eager_autodiff_pytorch_style.rs:28 | public | documented emptiness accessor | | tenferro_gpu::cuda::PinnedHostBuffer::len | fn | pub | yes | yes | 0 — | 29 crates/tenferro-gpu/examples/cuda_session_transfer.rs:18, crates/tenferro-gpu/examples/cuda_session_transfer.rs:97, crates/tenferro-gpu/examples/cuda_session_transfer.rs:108 | 48 docs/getting-started/core-concepts.md:171, docs/getting-started/index.md:146, docs/getting-started/index.md:201 | public | documented length accessor | | tenferro_gpu::cuda::PinnedHostBuffer::new | fn | pub | yes | yes | 0 — | 36 crates/tenferro-gpu/examples/cuda_session_transfer.rs:108, crates/tenferro-gpu/examples/cuda_session_transfer.rs:123, crates/tenferro-gpu/tests/pending_download.rs:60 | 3 docs/guides/engines-and-transfers.md:117, docs/guides/engines-and-transfers.md:132, docs/guides/engines-and-transfers.md:143 | public | documented constructor (PinnedHostBuffer::new in the transfers guide) | | tenferro_gpu::cuda::PublicationAdmissionError | enum | pub | yes | yes | 0 — | 3 crates/tenferro-gpu/tests/seam_publication.rs:9, crates/tenferro-gpu/tests/seam_publication.rs:168, crates/tenferro-gpu/tests/seam_publication.rs:189 | 0 — | public | the error CudaEngine::publish returns | | tenferro_gpu::cuda::PublicationDownload | struct | pub | yes | yes | 0 — | 0 — | 0 — | public | the handle SourcePublication::download_into returns | | tenferro_gpu::cuda::PublicationDownload::is_ready | fn | pub | yes | yes | 0 — | 0 — | 4 docs/guides/engines-and-transfers.md:122, docs/guides/engines-and-transfers.md:129, docs/guides/engines-and-transfers.md:155 | public | poll of the publication download | | tenferro_gpu::cuda::PublicationDownload::wait | fn | pub | yes | yes | 0 — | 0 — | 7 docs/guides/engines-and-transfers.md:127, docs/guides/engines-and-transfers.md:129, docs/guides/engines-and-transfers.md:133 | public | wait of the publication download (download_pending(..).wait()) | | tenferro_gpu::cuda::PublicationError | enum | pub | yes | yes | 0 — | 0 — | 0 — | public | the error of every publication step | | tenferro_gpu::cuda::RejectedPublication | struct | pub | yes | yes | 0 — | 0 — | 0 — | public | the value a refused publication hands back inside PublicationAdmissionError | | tenferro_gpu::cuda::SourcePublication | struct | pub | yes | yes | 0 — | 0 — | 0 — | public | returned by CudaEngine::publish; the issue names publication as public surface | | tenferro_gpu::cuda::SourcePublication::await_frontier | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | the documented wait before a download | | tenferro_gpu::cuda::SourcePublication::bytes | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | accessor of the published value | | tenferro_gpu::cuda::SourcePublication::download_into | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | the documented download of a published value | | tenferro_gpu::cuda::SourcePublication::frontier_ready | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | the documented poll before a download | | crate::cuda::publication::SourcePublication::new | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::raw | mod | pub | no | yes | 340 crates/tenferro-fft/src/cuda/plan.rs:6, crates/tenferro-fft/src/cuda/plan.rs:8, crates/tenferro-fft/src/cuda/plan.rs:10 | 24 crates/tenferro-gpu/tests/cuda_refused_host_reservation.rs:7, crates/tenferro-gpu/tests/cuda_refused_host_reservation.rs:19, crates/tenferro-gpu/tests/cuda_refused_host_reservation.rs:63 | 5 docs/guides/devices-and-gpu.md:172, docs/guides/external-linalg-interop.md:534, docs/guides/external-linalg-interop.md:575 | public | the documented raw extension scope (issue #1597): tenferro_gpu::cuda::raw is a public module whose items the guides name | | tenferro_gpu::cuda::CudaEngine::with_raw | fn | pub | yes | yes | 28 crates/tenferro-fft/src/cuda/plan.rs:617, crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:94, crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:276 | 8 crates/tenferro-gpu/tests/cuda_refused_host_reservation.rs:66, crates/tenferro-gpu/tests/pinned_staging_reuse.rs:30, crates/tenferro-gpu/tests/seam_raw.rs:29 | 6 docs/guides/devices-and-gpu.md:171, docs/guides/engines-and-transfers.md:183, docs/guides/external-linalg-interop.md:498 | public | the same hook, declared beside the raw scope (with_raw in raw.rs) (cited at crates/tenferro-fft/src/cuda/plan.rs:617) | | crate::cuda::raw::CudaEngineMarker | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::raw::CudaResourceGuard | struct | pub | no | yes | 0 — | 0 — | 0 — | public | returned by Session::resource; the extension crates hold it | | tenferro_gpu::cuda::raw::DeviceBytes | struct | pub | yes | yes | 2 crates/tenferro-fft/src/cuda/plan.rs:25, crates/tenferro-fft/src/cuda/plan.rs:344 | 0 — | 0 — | public | returned by Session::alloc_bytes, named in a tenferro-fft signature | | tenferro_gpu::cuda::raw::DeviceBytes::byte_len | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | caller evidence unverified: no byte_len call on a DeviceBytes exists outside cuda/ | | tenferro_gpu::cuda::raw::DeviceBytes::with_ptr | fn | pub | yes | yes | 1 crates/tenferro-fft/src/cuda/plan.rs:347 | 0 — | 0 — | public | called by tenferro-fft and tenferro-linalg on a raw device allocation (cited at crates/tenferro-fft/src/cuda/plan.rs:347) | | crate::cuda::raw::PinnedStaging | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::raw::PinnedStaging::bytes | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::raw::PinnedStaging::empty | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::raw::PinnedStaging::pointer | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::raw::Session | struct | pub | yes | yes | 0 — | 0 — | 0 — | public | the documented raw extension session (tenferro_gpu::cuda::raw) | | tenferro_gpu::cuda::raw::Session::alloc_bytes | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | called by tenferro-fft and tenferro-linalg inside the raw scope (raw.alloc_bytes) (cited at crates/tenferro-fft/src/cuda/plan.rs:625, crates/tenferro-linalg/src/gpu/linalg.rs:1465) | | tenferro_gpu::cuda::raw::Session::alloc_output | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | called by tenferro-linalg (raw.alloc_output) (cited at crates/tenferro-linalg/src/gpu/linalg.rs:1411, crates/tenferro-linalg/src/gpu/linalg.rs:1426) | | tenferro_gpu::cuda::raw::Session::copy_bytes | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | called by tenferro-linalg (raw.copy_bytes) (cited at crates/tenferro-linalg/src/gpu/linalg.rs:1433) | | tenferro_gpu::cuda::raw::Session::download_tensor | fn | pub | yes | yes | 0 — | 0 — | 8 docs/guides/devices-and-gpu.md:61, docs/guides/devices-and-gpu.md:66, docs/guides/engines-and-transfers.md:108 | public | documented raw download (cited at crates/tenferro-linalg/src/gpu/linalg.rs:1509) | | tenferro_gpu::cuda::raw::Session::resource | fn | pub | no | yes | 0 — | 0 — | 1 docs/tutorial-code/src/bin/cuda_vendor_interop.rs:6 | public | called by tenferro-linalg to register its vendor handles (raw.resource) (cited at crates/tenferro-linalg/src/gpu/linalg.rs:1419) | | tenferro_gpu::cuda::raw::Session::retain_tensor | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | called by tenferro-fft to keep an allocation alive (cited at crates/tenferro-fft/src/cuda/plan.rs:641) | | tenferro_gpu::cuda::raw::Session::stream | fn | pub | yes | yes | 0 — | 0 — | 3 docs/guides/external-linalg-interop.md:567, docs/guides/external-linalg-interop.md:626, docs/tutorial-code/src/bin/cuda_vendor_interop.rs:85 | public | called by tenferro-fft and tenferro-linalg (raw.stream()) (cited at crates/tenferro-fft/src/cuda/plan.rs:620) | | tenferro_gpu::cuda::raw::Session::synchronize | fn | pub | yes | yes | 2 crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:716, crates/tenferro-linalg/benches/incremental_householder_qr.rs:76 | 0 — | 19 docs/guides/devices-and-gpu.md:86, docs/guides/devices-and-gpu.md:88, docs/guides/engines-and-transfers.md:21 | public | called by the extension crates and their benches (cited at crates/tenferro-fft/src/cuda/plan.rs:684) | | tenferro_gpu::cuda::raw::Session::tensor | fn | pub | yes | yes | 13 crates/tenferro-linalg/src/cpu/tests/managed_cholesky.rs:227, crates/tenferro-linalg/src/cpu/tests/managed_cholesky.rs:228, crates/tenferro-linalg/src/cpu/tests/managed_cholesky.rs:230 | 0 — | 8 docs/getting-started/pytorch-jax-mapping.md:34, docs/getting-started/pytorch-jax-mapping.md:35, docs/guides/external-linalg-interop.md:558 | public | called by tenferro-fft and tenferro-linalg (raw.tensor(...)) (cited at crates/tenferro-fft/src/cuda/plan.rs:657) | | tenferro_gpu::cuda::raw::Session::tensor_mut | fn | pub | yes | yes | 0 — | 0 — | 4 docs/guides/external-linalg-interop.md:561, docs/guides/external-linalg-interop.md:631, docs/tutorial-code/src/bin/cuda_vendor_interop.rs:5 | public | called by tenferro-fft and tenferro-linalg (raw.tensor_mut(...)) (cited at crates/tenferro-fft/src/cuda/plan.rs:665) | | tenferro_gpu::cuda::raw::Session::upload_bytes | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | called by tenferro-linalg inside the raw scope (cited at crates/tenferro-linalg/src/gpu/linalg.rs:1665) | | crate::cuda::raw::StagingAccounting | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::raw::StagingAccounting::allocations | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::raw::StagingAccounting::limit | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::raw::StagingAccounting::new | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::raw::StagingAccounting::record_retained | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::raw::StagingAccounting::retained | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::raw::StagingAccounting::set_limit | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::raw::StreamRef | struct | pub | yes | yes | 0 — | 0 — | 0 — | public | returned by Session::stream | | tenferro_gpu::cuda::raw::StreamRef::raw_handle | fn | pub | yes | yes | 0 — | 0 — | 4 docs/guides/external-linalg-interop.md:568, docs/guides/external-linalg-interop.md:626, docs/tutorial-code/src/bin/cuda_vendor_interop.rs:7 | public | called by tenferro-fft and tenferro-linalg (raw.stream().raw_handle()) (cited at crates/tenferro-fft/src/cuda/plan.rs:620) | | tenferro_gpu::cuda::raw::TensorMut | struct | pub | yes | yes | 0 — | 1 crates/tenferro-gpu/tests/storage_provider_cuda.rs:73 | 0 — | public | returned by Session::tensor_mut | | tenferro_gpu::cuda::raw::TensorMut::byte_len | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | caller evidence unverified: no byte_len call on a TensorMut exists outside cuda/ | | tenferro_gpu::cuda::raw::TensorMut::raw_ptr | fn | pub | yes | yes | 0 — | 0 — | 6 docs/guides/external-linalg-interop.md:582, docs/guides/external-linalg-interop.md:584, docs/guides/external-linalg-interop.md:587 | public | called by the extension crates (raw_ptr on a raw tensor) (cited at crates/tenferro-fft/src/cuda/plan.rs:679) | | tenferro_gpu::cuda::raw::TensorMut::with_ptr | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | caller evidence unverified: the with_ptr calls in the extension crates are DeviceBytes::with_ptr; no TensorMut::with_ptr call exists outside cuda/ | | tenferro_gpu::cuda::raw::TensorRef | struct | pub | yes | yes | 0 — | 1 crates/tenferro-gpu/tests/storage_provider_cuda.rs:73 | 0 — | public | returned by Session::tensor | | tenferro_gpu::cuda::raw::TensorRef::byte_len | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | read by the extension crates on a raw tensor reference (cited at crates/tenferro-linalg/src/gpu/linalg.rs:1433) | | tenferro_gpu::cuda::raw::TensorRef::raw_ptr | fn | pub | yes | yes | 0 — | 0 — | 6 docs/guides/external-linalg-interop.md:582, docs/guides/external-linalg-interop.md:584, docs/guides/external-linalg-interop.md:587 | public | called by the extension crates (raw_ptr on a raw tensor) (cited at crates/tenferro-fft/src/cuda/plan.rs:678) | | tenferro_gpu::cuda::raw::TensorRef::with_ptr | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | caller evidence unverified: the with_ptr calls in the extension crates are DeviceBytes::with_ptr; no TensorRef::with_ptr call exists outside cuda/ | | tenferro_gpu::cuda::raw::with_cuda_engine | fn | pub | yes | yes | 8 crates/tenferro-fft/src/lib.rs:2069, crates/tenferro-fft/src/lib.rs:2070, crates/tenferro-fft/src/lib.rs:2666 | 1 crates/tenferro-gpu/tests/seam_raw.rs:105 | 0 — | public | the documented entry into the raw scope, used by tenferro-fft (cited at crates/tenferro-fft/src/lib.rs:2069) | | crate::cuda::runtime::ActivatedContext | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime::CompletionDeferral | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaRuntime | struct | pub | yes | yes | 8 crates/tenferro-fft/src/cuda/plan.rs:26, crates/tenferro-fft/src/cuda/plan.rs:59, crates/tenferro-fft/src/cuda/plan.rs:395 | 23 crates/tenferro-gpu/tests/chunked_staging.rs:11, crates/tenferro-gpu/tests/chunked_staging.rs:23, crates/tenferro-gpu/tests/pending_download.rs:11 | 2 docs/guides/choosing-a-backend.md:146, docs/guides/tenferro-fft.md:237 | public | the documented runtime: engine.runtime() returns it, tenferro-fft names it in a signature and the guides document its knobs | | crate::cuda::runtime::CudaRuntime::alloc_pinned | fn | pub(crate) | yes | no | 0 — | 0 — | 0 — | private | no reader outside cuda/; called from cuda/pinned.rs | | tenferro_gpu::cuda::CudaRuntime::allocation_domain_id | fn | pub | yes | yes | 0 — | 0 — | 0 — | workspace-internal | tenferro-linalg checks an operand’s domain against it (cited at crates/tenferro-linalg/src/gpu/linalg.rs:364) | | tenferro_gpu::cuda::CudaRuntime::artifact_cache_hits | fn | pub | no | yes | 0 — | 0 — | 0 — | private | no reader anywhere: nothing outside cuda/ names it; read only by the test-only PreparedFusion::counters; gated with its two sibling counters | | tenferro_gpu::cuda::CudaRuntime::artifact_dir_evictions | fn | pub | no | yes | 0 — | 0 — | 1 docs/guides/devices-and-gpu.md:292 | public | documented by the devices guide | | tenferro_gpu::cuda::CudaRuntime::artifact_dir_max_bytes | fn | pub | no | yes | 0 — | 0 — | 1 docs/guides/devices-and-gpu.md:290 | public | documented by the devices guide | | tenferro_gpu::cuda::CudaRuntime::artifact_dir_stats | fn | pub | no | yes | 0 — | 0 — | 1 docs/guides/devices-and-gpu.md:288 | public | documented by the devices guide | | tenferro_gpu::cuda::CudaRuntime::clear_artifact_dir | fn | pub | no | yes | 0 — | 0 — | 1 docs/guides/devices-and-gpu.md:291 | public | documented by the devices guide | | tenferro_gpu::cuda::CudaRuntime::clear_pinned_staging | fn | pub | no | yes | 0 — | 0 — | 1 docs/guides/engines-and-transfers.md:56 | public | documented by the transfers guide | | tenferro_gpu::cuda::CudaRuntime::compile_attempts | fn | pub | no | yes | 0 — | 2 crates/tenferro-gpu/tests/seam_fusion.rs:59, crates/tenferro-gpu/tests/seam_fusion.rs:69 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::CudaRuntime::completion_checkpoints | fn | pub | yes | yes | 0 — | 0 — | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | crate::cuda::runtime::CudaRuntime::completion_deferred | fn | pub(crate) | yes | no | 0 — | 0 — | 0 — | private | no reader outside cuda/; read by cuda/structural.rs | | tenferro_gpu::cuda::CudaRuntime::copy_to | fn | pub | yes | yes | 0 — | 0 — | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | crate::cuda::runtime::CudaRuntime::defer_completion | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime::CudaRuntime::detach | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaRuntime::device_id | fn | pub | yes | yes | 0 — | 0 — | 0 — | workspace-internal | tenferro-fft reads the device ordinal through it (cited at crates/tenferro-fft/src/cuda/mod.rs:85) | | tenferro_gpu::cuda::CudaRuntime::device_info | fn | pub | yes | yes | 0 — | 0 — | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::CudaRuntime::device_ordinal | fn | pub | yes | yes | 0 — | 0 — | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::CudaRuntime::download_bytes | fn | pub | no | yes | 0 — | 9 crates/tenferro-gpu/tests/seam_contraction.rs:78, crates/tenferro-gpu/tests/seam_contraction.rs:141, crates/tenferro-gpu/tests/seam_contraction.rs:187 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | crate::cuda::runtime::CudaRuntime::identity | fn | pub(crate) | yes | no | 0 — | 0 — | 0 — | private | CudaEngine::runtime_identity is what tenferro-fft keys its plan cache on; this accessor has no outside caller, only cuda/engine.rs and cuda/runtime.rs | | tenferro_gpu::cuda::CudaRuntime::module_loads | fn | pub | no | yes | 0 — | 0 — | 0 — | private | no code outside cuda/ names it; only the test-only PreparedFusion::counters reads it, so it is gated with its two sibling counters | | tenferro_gpu::cuda::CudaRuntime::new | fn | pub | yes | yes | 0 — | 0 — | 0 — | public | the runtime constructor (CudaRuntime::new(device)) | | crate::cuda::runtime::CudaRuntime::note_completion_checkpoint | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaRuntime::pending_transfer_retirements | fn | pub | yes | yes | 0 — | 23 crates/tenferro-gpu/tests/pending_download.rs:123, crates/tenferro-gpu/tests/pending_download.rs:127, crates/tenferro-gpu/tests/pending_download.rs:135 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::CudaRuntime::pinned_staging_allocations | fn | pub | yes | yes | 0 — | 0 — | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::CudaRuntime::pinned_staging_stats | fn | pub | no | yes | 0 — | 0 — | 1 docs/guides/engines-and-transfers.md:53 | public | documented by the transfers guide | | tenferro_gpu::cuda::CudaRuntime::pool_release_threshold_bytes | fn | pub | yes | yes | 0 — | 0 — | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | crate::cuda::runtime::CudaRuntime::retire_detached | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaRuntime::set_pinned_staging_limit_bytes | fn | pub | no | yes | 0 — | 0 — | 1 docs/guides/engines-and-transfers.md:55 | public | documented by the transfers guide | | tenferro_gpu::cuda::CudaRuntime::stats | fn | pub | yes | yes | 0 — | 5 crates/tenferro-gpu/tests/seam_staging.rs:30, crates/tenferro-gpu/tests/seam_staging.rs:33, crates/tenferro-gpu/tests/seam_staging.rs:80 | 0 — | private | the count is the same-named method on other types; no other crate names CudaRuntime, so nothing outside cuda/ can call it | | tenferro_gpu::cuda::CudaRuntime::stream_waits | fn | pub | yes | yes | 0 — | 0 — | 0 — | workspace-internal | tenferro-linalg’s deferred-completion test reads the counter (cited at crates/tenferro-linalg/tests/integration/deferred_completion.rs:108, crates/tenferro-linalg/tests/integration/deferred_completion.rs:114) | | crate::cuda::runtime::CudaRuntime::submit | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaRuntime::synchronize | fn | pub | yes | yes | 3 crates/tenferro-fft/src/cuda/plan.rs:684, crates/tenferro-fft/src/cuda/tests/source_contract.rs:69, crates/tenferro-linalg/benches/cuda_svd_driver.rs:354 | 10 crates/tenferro-gpu/tests/pending_download.rs:133, crates/tenferro-gpu/tests/pending_download.rs:170, crates/tenferro-gpu/tests/pending_upload.rs:106 | 19 docs/guides/devices-and-gpu.md:86, docs/guides/devices-and-gpu.md:88, docs/guides/engines-and-transfers.md:21 | public | documented barrier (engine.runtime().synchronize()) | | crate::cuda::runtime::CudaRuntime::synchronize_best_effort | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaRuntime::trim_pool | fn | pub | yes | yes | 0 — | 0 — | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::CudaRuntime::upload_bytes | fn | pub | yes | yes | 0 — | 43 crates/tenferro-gpu/tests/seam_contraction.rs:45, crates/tenferro-gpu/tests/seam_contraction.rs:46, crates/tenferro-gpu/tests/seam_contraction.rs:48 | 0 — | private | the count is Session::upload_bytes/raw.upload_bytes on the raw scope; no other crate names CudaRuntime::upload_bytes | | tenferro_gpu::cuda::CudaRuntime::with_config | fn | pub | yes | yes | 0 — | 1 crates/tenferro-gpu/tests/seam_staging.rs:103 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::CudaRuntime::with_current_context | fn | pub | yes | yes | 2 crates/tenferro-fft/src/cuda/plan.rs:65, crates/tenferro-fft/src/cuda/plan.rs:482 | 0 — | 0 — | workspace-internal | tenferro-fft activates the context around its cuFFT calls (cited at crates/tenferro-fft/src/cuda/plan.rs:482) | | crate::cuda::runtime::CudaRuntime::with_raw_locked | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaRuntimeIdentity | struct | pub | yes | yes | 4 crates/tenferro-fft/src/cuda/descriptor.rs:3, crates/tenferro-fft/src/cuda/descriptor.rs:113, crates/tenferro-fft/tests/cuda_fft/cache.rs:9 | 0 — | 0 — | workspace-internal | tenferro-fft names it as its plan-cache key type | | crate::cuda::runtime::CudaRuntimeState | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::DEFAULT_PINNED_STAGING_LIMIT_BYTES | const | pub | no | yes | 0 — | 0 — | 1 docs/guides/engines-and-transfers.md:51 | public | documented by the transfers guide | | crate::cuda::runtime::Detached | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::PinnedStagingStats | struct | pub | no | yes | 0 — | 0 — | 0 — | public | returned by CudaRuntime::pinned_staging_stats | | crate::cuda::runtime::write_best_effort_warning | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::CudaPreparedKind | enum | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::CudaPreparedOperation | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::PreparedCudaRegistrationIdentity | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::core_operation_name | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::cuda_input_placement | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::cuda_input_signature | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::cuda_operation_kind | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::cuda_provider_device_identity | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::cuda_runtime_engine_registration | fn | pub | no | yes | 9 crates/tenferro-ad/src/eager_backend.rs:61, crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:385, crates/tenferro-ad/tests/integration/gpu_nn_composites.rs:373 | 6 crates/tenferro-gpu/examples/cuda_session_transfer.rs:79, crates/tenferro-gpu/tests/coalesced_run.rs:20, crates/tenferro-gpu/tests/seam_runtime.rs:165 | 1 docs/guides/engines-and-transfers.md:66 | public | the documented registration path, used by tenferro-ad and sibling tests | | tenferro_gpu::cuda::cuda_runtime_hardware_class | fn | pub | no | yes | 0 — | 0 — | 0 — | public | has its own doctest; names the CUDA hardware class the registration carries | | crate::cuda::runtime_adapter::cuda_runtime_storage_class | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::prepare_cuda | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::runtime_adapter::prepare_cuda_registration_identity | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::storage::CudaStorage | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::storage::CudaStorage::allocation | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::storage::CudaStorage::bytes | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::storage::CudaStorage::device_ordinal | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::storage::CudaStorage::domain | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::storage::fresh_storage | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::CudaEngine::alloc_output | fn | pub | yes | yes | 36 crates/tenferro-fft/src/cuda/mod.rs:502, crates/tenferro-fft/src/cuda/mod.rs:514, crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:16 | 6 crates/tenferro-gpu/tests/cuda_refused_allocation_observer.rs:10, crates/tenferro-gpu/tests/cuda_refused_allocation_observer.rs:78, crates/tenferro-gpu/tests/integration/cuda_refused_allocation.rs:3 | 0 — | workspace-internal | declared in cuda::structural::CudaEngine::alloc_output; a sibling crate calls it today (cited at crates/tenferro-linalg/src/gpu/linalg.rs:827) | | tenferro_gpu::cuda::CudaEngine::ensure_gpu_resident | fn | pub | yes | yes | 2 crates/tenferro-fft/src/cuda/mod.rs:229, crates/tenferro-linalg/src/gpu/linalg.rs:338 | 0 — | 0 — | workspace-internal | declared in cuda::structural::CudaEngine::ensure_gpu_resident; a sibling crate calls it today (cited at crates/tenferro-linalg/src/gpu/linalg.rs:338) | | tenferro_gpu::cuda::CudaEngine::linalg_apply_reflector | fn | pub | yes | yes | 2 crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:167, crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:184 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:165 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_column_norms | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:159 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:77 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_extract_r | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:180 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:175 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_fill_one | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg.rs:3869 | 0 — | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_householder_assemble | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:498 | 0 — | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_householder_explicit_v | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:37 | 0 — | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_identity_shifted | fn | pub | yes | yes | 2 crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:19, crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:182 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:135 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_initialize_permutation | fn | pub | yes | yes | 2 crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:138, crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:156 | 0 — | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_lu_apply_pivots | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg.rs:3887 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:208 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_lu_extract_outputs | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg.rs:3829 | 0 — | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_lu_parity | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg.rs:3854 | 0 — | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_matrix_adjoint | fn | pub | yes | yes | 2 crates/tenferro-linalg/src/gpu/linalg.rs:2006, crates/tenferro-linalg/src/gpu/linalg.rs:2021 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:50 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine (cited at crates/tenferro-linalg/src/gpu/linalg.rs:2006) | | tenferro_gpu::cuda::CudaEngine::linalg_prefix_rank | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:188 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:189 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine (cited at crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:188) | | tenferro_gpu::cuda::CudaEngine::linalg_qr_apply_phase | fn | pub | yes | yes | 3 crates/tenferro-linalg/src/gpu/linalg.rs:2069, crates/tenferro-linalg/src/gpu/linalg.rs:2087, crates/tenferro-linalg/src/gpu/linalg.rs:2105 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:126 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_qr_phase | fn | pub | yes | yes | 3 crates/tenferro-linalg/src/gpu/linalg.rs:2068, crates/tenferro-linalg/src/gpu/linalg.rs:2086, crates/tenferro-linalg/src/gpu/linalg.rs:2104 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:115 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_reflector | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:162 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:150 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_select_pivot | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:160 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:93 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_slice | fn | pub | yes | yes | 2 crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:640, crates/tenferro-linalg/src/gpu/linalg.rs:3150 | 0 — | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_swap_columns | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg/rank_revealing_qr.rs:161 | 1 crates/tenferro-gpu/tests/seam_linalg_aux.rs:100 | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine | | tenferro_gpu::cuda::CudaEngine::linalg_triangular | fn | pub | yes | yes | 3 crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:646, crates/tenferro-linalg/src/gpu/linalg.rs:1517, crates/tenferro-linalg/src/gpu/linalg.rs:3152 | 0 — | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine (cited at crates/tenferro-linalg/src/gpu/linalg.rs:1517) | | tenferro_gpu::cuda::CudaEngine::linalg_upper_trapezoidal_violation | fn | pub | yes | yes | 1 crates/tenferro-linalg/src/gpu/linalg.rs:829 | 0 — | 0 — | workspace-internal | tenferro-linalg’s QR/LU decomposition paths call this auxiliary on the engine (cited at crates/tenferro-linalg/src/gpu/linalg.rs:829) | | tenferro_gpu::cuda::CudaEngine::scale_tensor_write | fn | pub | yes | yes | 1 crates/tenferro-fft/src/cuda/mod.rs:139 | 2 crates/tenferro-gpu/tests/seam_structural.rs:1656, crates/tenferro-gpu/tests/seam_structural.rs:1673 | 0 — | workspace-internal | declared in cuda::structural::CudaEngine::scale_tensor_write; a sibling crate calls it today (cited at crates/tenferro-fft/src/cuda/mod.rs:139) | | crate::cuda::structural::CudaEngine::to_contiguous | fn | pub(crate) | yes | no | 0 — | 0 — | 1 docs/getting-started/api-migration.md:22 | private | no reader outside cuda/; read by cuda/structural.rs. A one-line forward to TensorStructural::to_contiguous_read, and a second path to the same operation | | tenferro_gpu::cuda::CudaEngine::to_contiguous_view | fn | pub | yes | yes | 56 crates/tenferro-linalg/src/gpu/mod.rs:11, crates/tenferro-linalg/src/gpu/mod.rs:53, crates/tenferro-linalg/src/gpu/mod.rs:58 | 1 crates/tenferro-gpu/tests/seam_raw.rs:207 | 0 — | workspace-internal | declared in cuda::structural::CudaEngine::to_contiguous_view; a sibling crate calls it today (cited at crates/tenferro-linalg/src/gpu/mod.rs:53) | | crate::cuda::structural::DeviceOperand | trait | pub | yes | no | 0 — | 0 — | 0 — | workspace-internal | the operand bound on the linalg_* auxiliaries that tenferro-linalg calls; pub(crate) trips private_bounds on every one of them | | crate::cuda::structural::Region | struct | pub(super) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::structural::TiledTransposeGeometry | struct | pub(super) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::structural::activation | mod | pub(super) | no | no | 2 crates/tenferro-ad/tests/integration/gpu_nn_composites.rs:663, crates/tenferro-fft/src/cuda/plan.rs:76 | 3 crates/tenferro-gpu/tests/seam_raw.rs:116, crates/tenferro-gpu/tests/seam_raw.rs:128, crates/tenferro-gpu/tests/seam_raw.rs:130 | 0 — | private | pub(super) module: the fused-activation kernels stay inside cuda::structural | | crate::cuda::structural::activation::CudaEngine::fused_activation | fn | pub(super) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::structural::activation::FusedActivationKernels | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::structural::alloc_tensor | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::structural::element_count | fn | pub(crate) | no | no | 7 crates/tenferro-fft/tests/cuda_fft/common.rs:69, crates/tenferro-fft/tests/cuda_fft/common.rs:77, crates/tenferro-fft/tests/cuda_fft/common.rs:84 | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::structural::region_of | fn | pub(super) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::structural::tiled_transpose_geometry | fn | pub(super) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::download_pending | fn | pub | yes | yes | 0 — | 28 crates/tenferro-gpu/examples/cuda_session_transfer.rs:124, crates/tenferro-gpu/tests/pending_download.rs:61, crates/tenferro-gpu/tests/pending_download.rs:80 | 3 docs/guides/devices-and-gpu.md:68, docs/guides/engines-and-transfers.md:133, docs/guides/engines-and-transfers.md:150 | public | documented asynchronous transfer (guides) | | tenferro_gpu::cuda::download_tensor | fn | pub | yes | yes | 26 crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:466, crates/tenferro-ad/tests/integration/gpu_nn_composites.rs:467, crates/tenferro-ad/tests/integration/gpu_nn_composites.rs:544 | 122 crates/tenferro-gpu/examples/cuda_quickstart.rs:32, crates/tenferro-gpu/examples/cuda_quickstart.rs:41, crates/tenferro-gpu/tests/blas1_cuda.rs:131 | 21 docs/getting-started/pytorch-jax-mapping.md:27, docs/getting-started/pytorch-jax-mapping.md:73, docs/guides/devices-and-gpu.md:61 | public | documented transfer: download_tensor in the guides and the quickstart, and by sibling crates in their GPU tests | | crate::cuda::transfer::placement | fn | pub(crate) | no | no | 11 crates/tenferro-ad/src/eager_backend.rs:30, crates/tenferro-fft/tests/cuda_fft/eager.rs:175, crates/tenferro-fft/tests/cuda_fft/support.rs:56 | 0 — | 9 docs/guides/external-linalg-interop.md:159, docs/guides/external-linalg-interop.md:179, docs/guides/external-linalg-interop.md:326 | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | crate::cuda::transfer::submit_download | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::upload_pending | fn | pub | yes | yes | 0 — | 24 crates/tenferro-gpu/examples/cuda_session_transfer.rs:110, crates/tenferro-gpu/tests/pending_upload.rs:59, crates/tenferro-gpu/tests/pending_upload.rs:82 | 3 docs/guides/devices-and-gpu.md:68, docs/guides/engines-and-transfers.md:119, docs/guides/engines-and-transfers.md:147 | public | documented asynchronous transfer (guides) | | tenferro_gpu::cuda::upload_tensor | fn | pub | yes | yes | 51 crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:106, crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:111, crates/tenferro-ad/tests/integration/gpu_ad_tests.rs:149 | 201 crates/tenferro-gpu/examples/cuda_quickstart.rs:26, crates/tenferro-gpu/examples/cuda_quickstart.rs:27, crates/tenferro-gpu/examples/cuda_quickstart.rs:36 | 23 docs/getting-started/pytorch-jax-mapping.md:27, docs/getting-started/pytorch-jax-mapping.md:72, docs/guides/devices-and-gpu.md:44 | public | documented transfer: upload_tensor in the guides and the quickstart, and by sibling crates in their GPU tests | | crate::cuda::vendor | mod | pub(crate) | no | no | 20 crates/tenferro-fft/src/cuda/mod.rs:212, crates/tenferro-fft/src/cuda/plan.rs:75, crates/tenferro-fft/src/cuda/plan.rs:143 | 6 crates/tenferro-gpu/tests/integration/public_surface_contract.rs:52, crates/tenferro-gpu/tests/seam_raw.rs:4, crates/tenferro-gpu/tests/seam_raw.rs:76 | 0 — | private | pub(crate) module: Cublas/GemmRequest reach users through the cuda:: re-export, and this path is crate-private | | crate::cuda::vendor::ComplexCoefficient | struct | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ | | tenferro_gpu::cuda::Cublas | struct | pub | yes | yes | 0 — | 6 crates/tenferro-gpu/tests/seam_vendor.rs:8, crates/tenferro-gpu/tests/seam_vendor.rs:63, crates/tenferro-gpu/tests/seam_vendor.rs:99 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::Cublas::gemm | fn | pub | no | yes | 5 crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:192, crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:209, crates/tenferro-linalg/src/gpu/linalg/householder_qr.rs:226 | 5 crates/tenferro-gpu/tests/seam_vendor.rs:65, crates/tenferro-gpu/tests/seam_vendor.rs:101, crates/tenferro-gpu/tests/seam_vendor.rs:135 | 0 — | private | the count is .gemm( on tenferro-linalg’s own cuBLAS helper types, which share the name; only this crate’s tests/seam_vendor.rs calls this gemm, so removing pub means porting that suite first | | tenferro_gpu::cuda::Cublas::new | fn | pub | no | yes | 0 — | 5 crates/tenferro-gpu/tests/seam_vendor.rs:63, crates/tenferro-gpu/tests/seam_vendor.rs:99, crates/tenferro-gpu/tests/seam_vendor.rs:133 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | tenferro_gpu::cuda::GemmRequest | struct | pub | yes | yes | 0 — | 6 crates/tenferro-gpu/tests/seam_vendor.rs:8, crates/tenferro-gpu/tests/seam_vendor.rs:65, crates/tenferro-gpu/tests/seam_vendor.rs:101 | 0 — | private | only this crate’s own examples//tests/ name it, so removing pub means porting that suite first (docs/design/cuda-surface.md section 6) | | crate::cuda::vendor::narrow_i32 | fn | pub(crate) | no | no | 0 — | 0 — | 0 — | private | already pub(crate)/pub(super) or module-local: no reader outside cuda/ |