Integrate CUDA workspace with activation memory patterns #32071
chilo-ms
wants to merge
39 commits into
main
from
chilo/static-workspace-preallocation
Report committed workspace estimates after graph partitioning
3afd9306
Combine profiled and estimated workspace accounting
37560cf0
Rename size-based resource accountant
a5af386b
Document pending resource accounting state
6dbbd9b0
Clarify initial resource cost calculation
edbd5755
Resolve workspace estimate with resource cost
c8068bce
Unify workspace selection state
da899f32
Report workspace estimate discrepancies
e9e0e3d4
Account for MatMulNBits prepack memory
661940ae
Preserve partial profile accounting behavior
a2c9de7e
Address workspace accounting review feedback
79248e39
Avoid GCC optional initialization warning
2c33db55
Avoid double-counting offline-prepacked GPU weights
44d524ea
Clarify Level-1 initialization accounting contract
00d62c9b
Account for MatMulNBits tactic profiler scratch
5d7d98e2
Align CUDA memory estimation with session config
d3cff998
Clarify Level-1 prepack memory accounting
8a78fb2c
Fix CUDA plugin config key compilation
a0932465
Verify Level-2 workspace reservations
1948ce23
Preallocate declared CUDA workspace
30da4dd0
Plan CUDA workspace with activations
f867bdde
Test planned workspace across MatMulNBits chain
84409b97
Bridge planned workspace access to CUDA
01165aab
Avoid workspace declaration for MatMulNBits GEMV
b5236e8e
Test CUDA arena savings from planned workspace
f4e45cbc
Merge main into chilo/level2-workspace-verification
c15cd49d
Limit strict workspace verification to overruns
9538baf3
Preserve workspace reservations for fused nodes
7d183380
Merge current main into workspace verification
e70994f5
Merge current main into static workspace preallocation
28e723b6
Merge Level-2 workspace verification into static workspace preallocation
3a02309b
Preserve reservations through graph replacements
393dee76
Track workspace reservations through Level-2 fusions
88c98a54
Fix workspace integration after branch merge
19c8b900
Exclude workspace verification from minimal builds
254ddd98
Track workspace reservations across Level-2 mutations
b9660e51
Merge remote-tracking branch 'origin/main' into copilot/pr32189-comments
f860ca8b
Track remaining workspace reservation mutations
e5460120
Merge latest Level-2 workspace verification updates
3258464a
Base automatically changed from
chilo/level2-workspace-verification
to
main
3 days ago
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub