[ FSU ] Enabls Asynchronos FSU for forwarding #2813
Conversation
jijoongmoon
requested review from
myungjoo,
again4you,
jaeyun-jung,
leemgs,
wooksong,
gichan-jang,
anyj0527,
lhs8928,
songgot,
jihochu,
DonghakPark,
SeoHyungjun and
baek2sm
as code owners
|
📝 TAOS-CI Version: 1.5.20200925. Thank you for submitting PR #2813. Please a submit 1commit/1PR (one commit per one PR) policy to get comments quickly from reviewers. Your PR must pass all verificiation processes of cibot before starting a review process from reviewers. If you are new member to join this project, please read manuals in documentation folder and wiki page. In order to monitor a progress status of your PR in more detail, visit http://ci.nnstreamer.ai/. |
jijoongmoon
requested review from
skykongkong8,
djeong20 and
EunjuYang
as code owners
There was a problem hiding this comment.
@jijoongmoon, 💯 All CI checkers are successfully verified. Thanks.
Applications/SimpleFC/jni/main.cpp
Outdated
| @@ -0,0 +1,275 @@ | |||
| // SPDX-License-Identifier: Apache-2.0 | |||
| /** | |||
| * Copyright (C) 2020 Jihoon Lee <[email protected]> | |||
There was a problem hiding this comment.
fixed. Thanks.
Applications/SimpleFC/jni/main.cpp
Outdated
| @@ -0,0 +1,275 @@ | |||
| // SPDX-License-Identifier: Apache-2.0 | |||
| /** | |||
| * Copyright (C) 2020 Jihoon Lee <[email protected]> | |||
There was a problem hiding this comment.
fixed. Thanks.
Applications/SimpleFC/jni/main.cpp
Outdated
| } | ||
|
|
||
| /** | ||
| * @brief Create resnet 18 |
There was a problem hiding this comment.
fixed. Thanks.
| @@ -0,0 +1,28 @@ | |||
| resnet_sources = [ | |||
There was a problem hiding this comment.
fixed. Thanks.
nntrainer/models/neuralnet.cpp
Outdated
| } else { | ||
| NNTR_THROW_IF(((mode == ExecutionMode::INFERENCE) && | ||
| (exec_mode == ExecutionMode::TRAIN)), | ||
| std::invalid_argument) |
There was a problem hiding this comment.
It looks like the if statement already checks mode == ExecutionMode::INFERENCE, so it wouldn't throw an exception. can we remove it?
There was a problem hiding this comment.
fixed. Thanks.
| << std::endl; | ||
| ml_logd("request load tensor for %d", f + 1); | ||
| model_graph.LoadTensors((f / (lookahead + 1) + 1) * | ||
| (lookahead + 1)); |
There was a problem hiding this comment.
could you explain this part of the code please?
There was a problem hiding this comment.
I left the comments in code.
nntrainer/tensor/manager.cpp
Outdated
| } | ||
| async_load_tensor.erase(order); | ||
| ml_logd("wait and completed %d", order); | ||
| ; |
There was a problem hiding this comment.
thansk. fixed.
Describe a commit content (Until 80 colums per line) in detail ASAP. **Changes proposed in this PR:** - Added TOC generator for README.md Resolves: **Self evaluation:** 1. Build test: [X]Passed [ ]Failed [ ]Skipped 2. Run test: [X]Passed [ ]Failed [ ]Skipped Signed-off-by: jijoong.moon <[email protected]>
jijoongmoon
force-pushed
the
asynch
branch
2 times, most recently
from
010c1f5 to
da64f66
Compare
This PR enables asynchronos mode for FSU (flash storage utilization) for better performance. It splits the load and unload tensors which make difficult to handle. Also fix the inference execution order when it is in INFERENCE mode and change the trainable option to false when it calls the request weights and tensors. Add the new function to load and unload tensors as well as check load complete. It also considers weight pool and tensor pool differenetly according to the ExecutionMode. It is not use FSU mode for tensor pool for the INFERENCE Mode. Resolves: **Self evaluation:** 1. Build test: [X]Passed [ ]Failed [ ]Skipped 2. Run test: [X]Passed [ ]Failed [ ]Skipped Signed-off-by: jijoong.moon <[email protected]>
|
|
||
| Step 5. Try to release the weights which has execution order less then f. | ||
|
|
||
| Step n. repeat next layer starting with checking the tenosrs are loaded, |
There was a problem hiding this comment.
Is this a typo?
| Step n. repeat next layer starting with checking the tenosrs are loaded, | |
| Step n. repeat next layer starting with checking the tensors are loaded, |
This PR enables asynchronos mode for FSU (flash storage utilization)
for better performance.
It splits the load and unload tensors which make difficult to handle.
Also fix the inference execution order when it is in INFERENCE mode
and change the trainable option to false when it calls the request
weights and tensors.
Add the new function to load and unload tensors as well as check load
complete.
It also considers weight pool and tensor pool differenetly according
to the ExecutionMode. It is not use FSU mode for tensor pool for the
INFERENCE Mode.
Resolves:
Self evaluation:
Signed-off-by: jijoong.moon [email protected]