onnxruntime
Implement experimental intermediate cross CPU EP allocation
#24371
Merged
Merged
Implement experimental intermediate cross CPU EP allocation #24371
yuslepukhin
merged
29 commits into
main
from
yuslepukhin/qnn_copy_fix
Implement experimental intermediate cross CPU EP allocation
faf10b6a
yuslepukhin
commented
on 2025-04-10
Implement experimental intermediate cross CPU EP allocation
e23ac429
Address build failures
3d4f853a
Merge
fe843163
Address build issues
97c1b535
Re-work CPU output device override
396b0e7a
Adjust QNN EP for 4K CPU Allocator
fb9d3cb8
Make sure CPU 4K allocator is not created with arena
5d8c7246
Make QNN EP return QTP when rpcmem is present, otherwise just plain C…
a13fccd9
Query memory from EPs
c7b8096f
Merge branch 'main' into yuslepukhin/qnn_copy_fix
67fa917b
Apply code review changes
36886fa6
Make device comparision more specific
b15b24e6
Merge branch 'main' into yuslepukhin/qnn_copy_fix
96fd5571
Address review comments
3c7f21b0
Remove extra quailifier
fc48a456
Merge branch 'main' into yuslepukhin/qnn_copy_fix
1e781e7c
Resolve mlas deps
1ff6086a
Address review comments
35e455ca
Fix linking. Enforce default alignment only for CPU devices as before.
39346823
yuslepukhin
commented
on 2025-04-23
yuslepukhin
commented
on 2025-04-23
Undo some changes
b691e7b0
remove header from HTP shared memory allocations. replace with global…
fb8b6685
call QnnContextMemHandleManager::Unregister with shared memory addres…
5c15fdb8
Fix shared allocation detection
c0d1de42
skottmckay
commented
on 2025-04-24
edgchen1
commented
on 2025-04-24
edgchen1
commented
on 2025-04-24
add comments about AllocationTracker. add gsl include.
2e4b46c6
Remove alignment detection in qnn model
c2e3f99d
For CPU based memory:
b2b949b1
skottmckay
commented
on 2025-04-24
Extract common logic for input/output device deduction
6da69946
Disable Qnn Qtp shared allocator for intermediate values
f8635f7e
skottmckay
approved these changes
on 2025-04-25
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub