Ryanunderhill/cuda shared #7626
jywu-msft
merged
182 commits into
master
from
ryanunderhill/cuda_shared
First iteration of making cuda a shared provider.
6a54ed4a
Merge with master
7c74748c
Merge branch 'master' of https://github.com/microsoft/onnxruntime int…
c4b845bb
More cuda shared library refactoring
2df3996e
More cuda shared library refactoring
6b91423c
Merge with master changes
a6f7b582
More build options tested, converted the training ops over.
a841d17d
Merge with master
57591f5b
Fix merge breaks
20644043
Fix submodules
821f5789
Fix submodules
aace40f1
Fix submodules
68335442
Fix python
4cf4cf30
Fix compile errors
cae0af00
Duplicate symbol fix
698301e8
Merge with master
80cae233
Test fix for ROCM provider
fa577218
Another ROCM test workaround
ae38f058
ROCM Build Test
88f51f0e
ROCM build fix
6017a496
ROCM
ec1efa3d
ROCM
6662d8d6
ROCM
9c6cc88e
ROCM
af0a5d1c
ROCM
639fcb91
ROCM test
53dd09db
Reduce header dependencies
5205522c
Remove redundant namespace
470fdc59
Test fix for linux
2f7bb27f
Fix linux build
dcafa85a
Fix Eigen build error
d254c317
Fix unused parameter warning
5173d1c7
Test link error
9cfeff7f
Another linker test
7f6ad2e8
Linker test
8344a1cd
Linker test
81ad74d1
Another test
81599249
Another build test
03bdd1cf
Fix linux link error
7c3855d6
Build test
d9d2c3d5
Fix control flow ops to use common base class with core code
21a73486
Remove extra qualifiers
4e8eab65
Fix template syntax for linux
263ca763
Fix cuda memory leak
29e18e7d
Fix pybind
95fa78df
Test disabling cast
7bf00902
Cleanup
663ebee6
Merge branch 'master' of https://github.com/microsoft/onnxruntime int…
e74de85a
Restore cuda in test
9f1c2ed7
Remove more header dependencies
4cf9e0b9
Test not adding cuda provider to session
4b8fba4f
Make GetProviderInfo_CUDA throw
f2d14cd7
No-op cuda provider creation
0a20e40a
Fix some setup issues
a36dd5d5
Fix memory cleanup on unload
5c6910ed
Diagnostics
96fa0845
Don't unload library
fca6d30e
Add diagnostics
06eac846
Fix deleting registry at right time.
393286f4
Test disabling profiler
8c1b52cf
Merge with master
9405a9cc
Fix merge break
b89e9817
Revert profiler change
784743e1
Move unloading of shared providers into Environment
41144546
Free more global allocations before library unloads
99cc4418
Add more diagnostics
92be95d0
Move unloading back to the OrtEnv as there are multiple Environments …
42d744b5
Fix more cmake files
0605567e
ERROR -> WARNING
37174409
Fix python shutdown
5a3a8fe2
Test not using dml in pipeline
3e4199bf
Change python version and disable dml
2b650e64
Update python version
9df9325f
Test adding unload method for shared providers
c427b787
Disable DLL test
e901cb93
Python test
c7ec2cfe
Revert "Python test"
acba6779
Revert "Disable DLL test"
58a675ec
Revert "Test adding unload method for shared providers"
9287e602
Point to RyanWinGPU
edf0522b
Revert python version
bc7f8ad4
Fix id_to_allocator_map
1ff20025
Another python exit test
07684619
Remove extra debug messages
b2c202d9
Revert DllMain idea, it didn't work
401ef4a6
Merge with master
401751b4
Merge conflicts
0f3252e4
Merge with master
8ba8c262
Fix merge with master issues.
ac7679f6
Merge branch 'master' of https://github.com/microsoft/onnxruntime int…
ad0dde55
Comments
605508a0
Undo edit to file
e2856cfa
Cleanup + new training ops
d2f96e4c
Revert yml changes
192d2306
Fix another merge error
d74b1e87
ROCM fix
0853a58c
ROCM fix v2
3dfec5a3
Put back Linux hack, it is necessary
85e5fddd
Stupid fixes
b4b6c4fa
Fix submodule out of sync
be540afb
Merge branch 'master' of https://github.com/microsoft/onnxruntime int…
6878463b
ROCM fix 3
7d74a925
ROCM 4
4fc25692
Test java fix
a826b0ca
Fix typos
34511ff8
Merge branch 'master' of https://github.com/microsoft/onnxruntime int…
af3824ce
Java test on my VM
29288eb4
Fix build error
dda22150
Spotless fix
24eed8bd
Leave temp file around to load properly
9dfdafe6
Fix cleanup on exit
74b1016d
Fix break
981af7d0
Java comments
6ad01b90
Remove LongformerAttentionBase workaround
648679b3
Spotless fix
e02270e6
Switch yml back to regular build pool
be35fc2a
Craigacp
commented
on 2021-05-10
Craigacp
commented
on 2021-05-10
Revert "Switch yml back to regular build pool"
64c0deeb
Code review feedback
d308c577
Merge with master
1698c57f
Fix errors due to merge
11bec629
Spotless fix
a79227dd
Fix minimal build
49a8b5db
Java fix for non cuda case
912b61e4
Java fix for CPU build
12224c88
Fix Nuphar?
7cc57c11
Fix nuphar 2
cd2c891a
Fix formatting
56713323
Revert "Remove LongformerAttentionBase workaround"
7bdf68ff
Training fix
6076ef96
Another java fix
c419ff24
Formatting
8c5d7580
Formatting
a59cf4b0
For orttraining
149c037a
Last orttraining build fix...
dfc8dab1
training fixes
23c0011b
Fix test provider error
8e27b5d3
Missing pass command
02fbbdce
Removed in wrong spot
f5cd80af
Python typo
9b84d763
Python typos
c5b9d9cb
Python crash on exit, possibly due to unloading of libraries.
fcb98063
Remove test_execution_provider from training build
330339ca
Still can't unload providers in python, alas.
ff40ba07
Disable Nvtx temporarily
335562de
MPI Kernels for Training
66babe4e
MPI Kernels part 2
3b2bd5f5
Patch through INcclService
4ad3a41f
Oops, wrong CMakeLists
f4bb6f5c
Missing namespace
3022ed54
Fix missing ()
8d66d84c
Move INcclService::GetInstance around to link nicer
218176be
Missing }
66d5ac1f
Missing MPI libraries for Cuda
919c28a4
Add extra GetType functions used by MPI
3a9b3a64
Missing Nccl library
af6b373f
Remove LOGS statements as a test
b1924a93
Add in a couple more missing GetType methods
1a20b6d1
Update comments
3263241b
Missed a logging reference in mpi_context.h
06128b96
Sync with Master
86cb70a1
Convert aten_op to shared (due to marge with master)
741e09a8
Test moving DistributedRunContext instance into shared provider layer
0a59bc39
Test passed, now with fix
8b538e79
Missing static
374ae717
Oops, scope DistributedRunContext to just NCCL
5f62d4bb
edgchen1
commented
on 2021-05-14
edgchen1
commented
on 2021-05-14
edgchen1
commented
on 2021-05-14
pranavsharma
commented
on 2021-05-14
edgchen1
commented
on 2021-05-14
Merge with master
67adb7bf
Merge related issues and code review feedback.
9ddbb046
Merge error
8d911833
Bump to rel-1.9.1 (#7684)
3c1cf82f
edgchen1
commented
on 2021-05-17
edgchen1
commented
on 2021-05-17
jywu-msft
commented
on 2021-05-18
jywu-msft
commented
on 2021-05-18
pranavsharma
commented
on 2021-05-18
pranavsharma
commented
on 2021-05-18
Formatting
5da718a1
Merge with master
70f52251
pranavsharma
commented
on 2021-05-18
pranavsharma
dismissed these changes
on 2021-05-18
Craigacp
commented
on 2021-05-19
Code review feedback for Java build on non Windows
23cad068
Merge branch 'master' of https://github.com/microsoft/onnxruntime int…
c569dee8
snnn
commented
on 2021-05-19
snnn
commented
on 2021-05-19
Remove cupti library dependency from core library
6a889ee8
Test Java pipeline fix
9ef92f35
Linux build fix
a73a8115
Revert "Linux build fix"
d224e47e
Revert "Remove cupti library dependency from core library"
9c589a7b
Packaging pipeline fixes to copy cuda shared provider for tensorrt & …
066ee43e
Add cuda to Tensorrt nuget package
0e95c2b5
Merge with master
7b8b04a1
onnxruntime_common still has a cuda header dependency
5ed09a5a
jywu-msft
approved these changes
on 2021-05-20
Reviewers
Assignees
No one assigned
Labels
training
core runtime
Milestone
No milestone
Login to write a write a comment.
Login via GitHub