Repository navigation

Add managed-memory advise, prefetch, and discard-prefetch free functions - #1775

Merged

rparolin merged 101 commits into

NVIDIA:mainfrom

rparolin:rparolin/managed_mem_advise_prefetch

May 20, 2026

Merged

Add managed-memory advise, prefetch, and discard-prefetch free functions#1775
rparolin merged 101 commits into
NVIDIA:mainfrom
rparolin:rparolin/managed_mem_advise_prefetch

Commits on Mar 17, 2026
- wip
  rparolin
  committed
  abdec47
- wip
  rparolin
  committed
  c418050
- fixing ci compiler errors
  rparolin
  committed
  b879fa5
- skipping tests that aren't supported
  rparolin
  committed
  04ee3de
- cu12 support
  rparolin
  committed
  9ab3f46
- Merge branch 'main' into rparolin/managed_mem_advise_prefetch
  rparolin
  committed
  bd75bc3
- Merge branch 'main' into rparolin/managed_mem_advise_prefetch
  rparolin
  committed
  1b1343b
- Moving to function from Buffer class methods to free standing functions in the cuda.core.managed_memory namespace
  rparolin
  committed
  a948066
- precommit format
  rparolin
  committed
  1457599
Commits on Mar 18, 2026
- iterating on implementation
  rparolin
  committed
  acb4024
- Simplify managed-memory helpers: remove long-form aliases, cache lookups, fix docs
  
  rparolin
  and
  claude
  committed
  d10ab07
- Merge branch 'main' into rparolin/managed_mem_advise_prefetch
  rparolin
  committed
  ae1de36
- fix(test): reset _V2_BINDINGS cache so legacy-signature tests take the legacy path
  
  rparolin
  and
  claude
  committed
  c250c92
- fix(test): require concurrent_managed_access for advise tests that hit real hardware
  
  rparolin
  and
  claude
  committed
  89329d9
- fix: validate managed buffer before checking discard_prefetch bindings support
  
  rparolin
  and
  claude
  committed
  8a75d1b
- refactor: extract managed memory ops into dedicated _managed_memory_ops module
  
  rparolin
  and
  claude
  committed
  9e9b1e0
- pre-commit fix
  rparolin
  committed
  90f0711
Commits on Mar 19, 2026
- Removing blank file
  rparolin
  committed
  b4d252c
- wip
  rparolin
  committed
  faaa1d8
Commits on Apr 6, 2026
- Merge branch 'main' into rparolin/managed_mem_advise_prefetch
  rparolin
  committed
  18786be
Commits on Apr 27, 2026
- Merge remote-tracking branch 'upstream/main' into rparolin/managed_mem_advise_prefetch
  rparolin
  committed
  9766ddc
- fix(cuda.core): update binding_version import after upstream merge
  
  rparolin
  and
  claude
  committed
  cf2f20d
- revert: drop managed_memory shim in cuda.core.experimental
  
  rparolin
  and
  claude
  committed
  db3bac2
- feat(cuda.core): add Location dataclass for managed memory
  
  rparolin
  and
  claude
  committed
  20d036e
Commits on Apr 28, 2026
- feat(cuda.core): add _coerce_location helper
  
  rparolin
  and
  claude
  committed
  c2dae53
- test(cuda.core): update monkeypatch target after binding_version rename
  
  rparolin
  and
  claude
  committed
  935c8ba
- refactor(cuda.core): tighten memory-attr query
  
  rparolin
  and
  claude
  committed
  dc46535
- feat(cuda.core): unified 1..N managed_memory.prefetch with cydriver
  
  rparolin
  and
  claude
  committed
  818f5d2
- feat(cuda.core): add managed_memory.discard
  
  rparolin
  and
  claude
  committed
  e296e72
- feat(cuda.core): unified 1..N managed_memory.discard_prefetch with cydriver
  
  rparolin
  and
  claude
  committed
  e697131
- feat(cuda.core): unified 1..N managed_memory.advise + drop legacy apparatus
  
  rparolin
  and
  claude
  committed
  3bc1021
- refactor(cuda.core): use Buffer.is_managed property in managed_memory ops
  
  rparolin
  and
  claude
  committed
  fa23869
- docs(cuda.core): document Location, discard, and 1..N managed_memory ops
  
  rparolin
  and
  claude
  committed
  68bdd14
- chore(cuda.core): drop narrative comments and tighten _coerce_location docstring
  
  rparolin
  and
  claude
  committed
  b4d9cbf
- chore(cuda.core): satisfy pre-commit hooks
  
  rparolin
  and
  claude
  committed
  ee96758
- refactor(cuda.core): move managed_memory ops to cuda.core.utils
  
  rparolin
  and
  claude
  committed
  d6f60f2
- chore(cuda.core): use __all__ in utils instead of per-import noqa
  
  rparolin
  and
  claude
  committed
  3176271
- chore(cuda.core): collapse nested if in Location.__post_init__ (SIM102)
  
  rparolin
  and
  claude
  committed
  782f6a9
- test(cuda.core): share one DummyUnifiedMemoryResource per batched test
  
  rparolin
  and
  claude
  committed
  0789bf6
- test(cuda.core): query all buffers before closing in test_batched_same_advice
  
  rparolin
  and
  claude
  committed
  e0c782a
Commits on Apr 30, 2026
Commits on May 1, 2026
- refactor(cuda.core): use libcpp.vector for batched-op C arrays (R14)
  
  rparolin
  and
  claude
  committed
  2151e61
- fix(cuda.core): restore CUDA_ERROR_NOT_INITIALIZED auto-init in _query_memory_attrs (R4)
  
  rparolin
  and
  claude
  committed
  5c6d054
- refactor(cuda.core): make Host a plain class instead of a dataclass (R1)
  
  rparolin
  and
  claude
  committed
  47d5609
- feat(cuda.core)!: drop int location shorthand from managed-memory ops (R6, R8)
  
  rparolin
  and
  claude
  committed
  a40bb81
- docs(cuda.core): add AccessedBySet to api_private.rst (R5)
  
  rparolin
  and
  claude
  committed
  c43e81e
- docs(cuda.core): note the legacy NUMA round-trip limitation on preferred_location (R2, R7)
  
  rparolin
  and
  claude
  committed
  71e9daa
- refactor(cuda.core): use collections.abc.Sequence for input checks (R12, R13)
  
  rparolin
  and
  claude
  committed
  df928a0
- refactor(cuda.core): narrow Buffer.from_handle to Buffer-only (R3)
  
  rparolin
  and
  claude
  committed
  f522916
- refactor(cuda.core): single API surface per operation (R9, R10, R11)
  
  rparolin
  and
  claude
  committed
  6204c57
- refactor(cuda.core): build advise reverse-lookup eagerly at module load (N4)
  
  rparolin
  and
  claude
  committed
  36012fd
- refactor(cuda.core): factor shared body of _do_batch_{prefetch,discard_prefetch} (N2)
  
  rparolin
  and
  claude
  committed
  067fb15
- test(cuda.core): reuse production _get_int_attr in managed-memory tests (N6)
  
  rparolin
  and
  claude
  committed
  a9cd713
- feat(cuda.core): cu12 fallback for prefetch_batch (N3)
  
  rparolin
  and
  claude
  committed
  d75a7bd
- test(cuda.core): cover AccessedBySet read methods (N7)
  
  rparolin
  and
  claude
  committed
  0af5bd4
- feat(cuda.core): cu13 NUMA round-trip for ManagedBuffer.preferred_location (N8)
  
  rparolin
  and
  claude
  committed
  b0d1a21
- docs(cuda.core): replace stale utils autosummary entries
  rparolin
  committed
  4c228eb
Commits on May 4, 2026
- feat(cuda.core): make Host a singleton class
  rparolin
  committed
  5743e05
- refactor(cuda.core): rename AccessedBySet -> AccessedBySetProxy
  rparolin
  committed
  7126324
- fix(cuda.core): silence ruff lints on Host singleton
  rparolin
  committed
  238cb14
Commits on May 5, 2026
- fix(cuda.core): reject bool as Host(numa_id=...)
  rparolin
  committed
  d0b6621
- fix(cuda.core): hoist managed-buffer check in _advise_one
  rparolin
  committed
  d0f9c7e
- fix(cuda.core): clarify CUDA 12 NUMA-host error message
  rparolin
  committed
  191f29d
- fix(cuda.core): reject Host(numa_id=...) up-front on CUDA 12
  rparolin
  committed
  bcc056b
- fix(cuda.core): make ManagedBuffer.accessed_by setter atomic
  rparolin
  committed
  1b66367
- style(cuda.core): apply ruff format
  rparolin
  committed
  5efbe4e
- Merge remote-tracking branch 'upstream/main' into rparolin/managed_mem_advise_prefetch
  rparolin
  committed
  5e2c051
- Skip NUMA-aware Host coerce tests on CUDA 12 builds
  rparolin
  committed
  8c35376
Commits on May 6, 2026
- Merge remote-tracking branch 'upstream/main' into rparolin/managed_mem_advise_prefetch
  rparolin
  committed
  29235b9
Commits on May 8, 2026
- Merge remote-tracking branch 'upstream/main' into rparolin/managed_mem_advise_prefetch
  rparolin
  committed
  274fb95
Commits on May 11, 2026
- cuda.core: address open managed-memory review feedback (PR #1775)
  rparolin
  committed
  22385ac
Commits on May 15, 2026
Commits on May 19, 2026

Commits

Commits on Mar 17, 2026

Commits on Mar 18, 2026

Commits on Mar 19, 2026

Commits on Apr 6, 2026

Commits on Apr 27, 2026

Commits on Apr 28, 2026

Commits on Apr 30, 2026

Commits on May 1, 2026

Commits on May 4, 2026

Commits on May 5, 2026

Commits on May 6, 2026

Commits on May 8, 2026

Commits on May 11, 2026

Commits on May 15, 2026

Commits on May 19, 2026