Skip to content

Add blocked convolution #768

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 79 commits into
base: master
Choose a base branch
from
Open

Add blocked convolution #768

wants to merge 79 commits into from

Conversation

FedyuninV
Copy link
Contributor

No description provided.

Valeriy Fedyunin added 30 commits August 26, 2022 14:23
Fix current progress
89574a9 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Fix a couple more bugs
f1e664b 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Organize tests in a better way
75bccb1 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Fix Trivial/PaddingHeight test
05ac4d9 
Signed-off-by: Valeriy Fedyunin <[email protected]>
HELLL YEAHHHH It passes all the given test!!!
c52427d 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Add tests from yolox
5f68ebe 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Add simple performance check
93550c2 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Use COMPUTE_BLOCK macro
affcfe1 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Move ClearBlock to macro
96b189c 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Move ProcessFilterCountN to macro
497175b 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Move PROCESS_OUTPUT_COUNT_N to macro
40601bd 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Optimize PackData and UnpackData functions
4bc7630 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Fix convDesc leak
5d43103 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Refactor measurements
6b019b2 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Add temp code
ecf9efc 
Signed-off-by: Valeriy Fedyunin <[email protected]>
DEBUG (expand ISimdMathEngine interface)
21eab6e 
Signed-off-by: Valeriy Fedyunin <[email protected]>
DEBUG add AvxTestDesktop
f76fbf0 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Test commit
8d3cda2 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Add blocked convolution with macro to NeoMathEngineAvx
3c4b03d 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Fix compilation warning in NeoMathEngine test
6ae35b5 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Add blocked convolution test for NeoMathEngineAvx
aa50778 
Signed-off-by: Valeriy Fedyunin <[email protected]>
EXPERIMENTAL: start moving to JIT
6af3f0d 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Use single generator for all of the broadcast values
9aac8e6 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Remove unused code
d20c990 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Pass arguments to JIT via struct
69e8d29 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Move kernelHeight and kernelWidth loops to JIT
0082a87 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Move CLEAR_BLOCK macro to JIT
2299071 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Move postProcessing to JIT
76a7519 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Use loops during JIT
b730e7d 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Even less lines
f56f27c 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Valeriy Fedyunin added 27 commits October 29, 2022 00:23
Start cleaning up this mess...
067e1c8 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Use calcOutputPad for width
011332d 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Don't depend on the order of fields in CBlockedConvGen::CParams
64bcd1f 
Signed-off-by: Valeriy Fedyunin <[email protected]>
More parameter renamings and optimizations
1d3d647 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Fix comment
247521c 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Use more clear indexes for Ymm accumulators
a8d9b79 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Clarify bias Ymms
9621ba2 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Clarify a bit more
9fa5a15 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Improve naming
572e277 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Optimize passing arguments runConv
3963962 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Add multithreading
cf1917d 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Remove fixed TODO comment
2385013 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Turn on all the heuristics
008a9fa 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Use NeoML terms
f75dffb 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Reorder fields
7090124 
Signed-off-by: Valeriy Fedyunin <[email protected]>
More renamings
6c5700f 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Add more comments
63fe462 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Fix some naming and add some comments
9a9272e 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Minor optimization and even more comments
d0b54ca 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Fix test skip condition
857d06c 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Fix non-MSVC compilation
ede4132 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Reduce diff size
25fc832 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Refactor tests
af96c8b 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Add more heuristics for blocked convolution
591a047 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Grammar fix
75aa7ac 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Disable AvxTestDesktop everywhere but Windows
ab0638c 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Merge branch 'master' into BlockedConvExperiments
f48de7c 
Signed-off-by: Valeriy Fedyunin <[email protected]>
@FedyuninV FedyuninV marked this pull request as ready for review December 10, 2022 00:51
Valeriy Fedyunin added 2 commits December 10, 2022 02:01
Fix compilation
69dc1f1 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Disable blocked conv for AVX512
1db000d 
Signed-off-by: Valeriy Fedyunin <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@FedyuninV