Add blocked convolution#768

Open

FedyuninV wants to merge 79 commits intoneoml-lib:masterfrom

FedyuninV:BlockedConvExperiments

Commits on Aug 26, 2022

Fix current progress
Valeriy Fedyunin
committed
Fix a couple more bugs
Valeriy Fedyunin
committed
Organize tests in a better way
Valeriy Fedyunin
committed
Fix Trivial/PaddingHeight test
Valeriy Fedyunin
committed
HELLL YEAHHHH It passes all the given test!!!
Valeriy Fedyunin
committed
Add tests from yolox
Valeriy Fedyunin
committed

Commits on Aug 29, 2022

Add simple performance check
Valeriy Fedyunin
committed
Use COMPUTE_BLOCK macro
Valeriy Fedyunin
committed
Move ClearBlock to macro
Valeriy Fedyunin
committed
Move ProcessFilterCountN to macro
Valeriy Fedyunin
committed
Move PROCESS_OUTPUT_COUNT_N to macro
Valeriy Fedyunin
committed

Commits on Aug 31, 2022

Optimize PackData and UnpackData functions
Valeriy Fedyunin
committed
Fix convDesc leak
Valeriy Fedyunin
committed
Refactor measurements
Valeriy Fedyunin
committed

Commits on Sep 1, 2022

Add temp code
Valeriy Fedyunin
committed

Commits on Sep 24, 2022

DEBUG (expand ISimdMathEngine interface)
Valeriy Fedyunin
committed
DEBUG add AvxTestDesktop
Valeriy Fedyunin
committed

Commits on Sep 25, 2022

Test commit
Valeriy Fedyunin
committed

Commits on Sep 27, 2022

Add blocked convolution with macro to NeoMathEngineAvx
Valeriy Fedyunin
committed
Fix compilation warning in NeoMathEngine test
Valeriy Fedyunin
committed
Add blocked convolution test for NeoMathEngineAvx
Valeriy Fedyunin
committed

Commits on Oct 9, 2022

EXPERIMENTAL: start moving to JIT
Valeriy Fedyunin
committed

Commits on Oct 17, 2022

Use single generator for all of the broadcast values
Valeriy Fedyunin
committed
Remove unused code
Valeriy Fedyunin
committed

Commits on Oct 18, 2022

Pass arguments to JIT via struct
Valeriy Fedyunin
committed

Commits on Oct 19, 2022

Move kernelHeight and kernelWidth loops to JIT
Valeriy Fedyunin
committed
Move CLEAR_BLOCK macro to JIT
Valeriy Fedyunin
committed
Move postProcessing to JIT
Valeriy Fedyunin
committed

Commits on Oct 21, 2022

Use loops during JIT
Valeriy Fedyunin
committed
Even less lines
Valeriy Fedyunin
committed

Commits on Oct 22, 2022

Manually assign registers
Valeriy Fedyunin
committed

Commits on Oct 23, 2022

Use RSP instead of Param1
Valeriy Fedyunin
committed
Shorten the macro (a little bit)
Valeriy Fedyunin
committed
Wrap generator into C++ class (will be useful in future)
Valeriy Fedyunin
committed
Move filterCount and outputCount to generation parameters
Valeriy Fedyunin
committed
Move PROCESS_FILTER_COUNT_N to JIT
Valeriy Fedyunin
committed
Move all we need to JIT!!!
Valeriy Fedyunin
committed

Commits on Oct 24, 2022

Merge branch 'master' into BlockedConvExperiments
Valeriy Fedyunin
committed
Disable NeoAvxTestDesktop for 32-bit and for FineObjects
Valeriy Fedyunin
committed
Disable BlockedConv tests for FineObjects
Valeriy Fedyunin
committed
Use blocked conv when possible
Valeriy Fedyunin
committed

Commits on Oct 25, 2022

DEBUG: temporary disable blocked conv
Valeriy Fedyunin
committed
Extend test set
Valeriy Fedyunin
committed
Enable blocked convolution when heuristics are OK
Valeriy Fedyunin
committed

Commits on Oct 26, 2022

Add more statistics to BlockedConv Real tests
Valeriy Fedyunin
committed
Overflow protection
Valeriy Fedyunin
committed
Reduce the number of ZEROUPPER calls
Valeriy Fedyunin
committed
Better tuning
Valeriy Fedyunin
committed
Reduce memory consumption
Valeriy Fedyunin
committed
Add even more tests to NeoAvxTestDesktop
Valeriy Fedyunin
committed

Commits on Oct 28, 2022

Start cleaning up this mess...
Valeriy Fedyunin
committed
Use calcOutputPad for width
Valeriy Fedyunin
committed
Don't depend on the order of fields in CBlockedConvGen::CParams
Valeriy Fedyunin
committed
More parameter renamings and optimizations
Valeriy Fedyunin
committed
Fix comment
Valeriy Fedyunin
committed
Use more clear indexes for Ymm accumulators
Valeriy Fedyunin
committed
Clarify bias Ymms
Valeriy Fedyunin
committed

Commits on Oct 29, 2022

Clarify a bit more
Valeriy Fedyunin
committed
Improve naming
Valeriy Fedyunin
committed
Optimize passing arguments runConv
Valeriy Fedyunin
committed
Add multithreading
Valeriy Fedyunin
committed
Remove fixed TODO comment
Valeriy Fedyunin
committed

Commits on Oct 31, 2022

Turn on all the heuristics
Valeriy Fedyunin
committed
Use NeoML terms
Valeriy Fedyunin
committed
Reorder fields
Valeriy Fedyunin
committed
More renamings
Valeriy Fedyunin
committed
Add more comments
Valeriy Fedyunin
committed

Commits on Nov 14, 2022

Fix some naming and add some comments
Valeriy Fedyunin
committed
Minor optimization and even more comments
Valeriy Fedyunin
committed
Fix test skip condition
Valeriy Fedyunin
committed
Fix non-MSVC compilation
Valeriy Fedyunin
committed
Reduce diff size
Valeriy Fedyunin
committed

Commits on Nov 15, 2022

Refactor tests
Valeriy Fedyunin
committed
Add more heuristics for blocked convolution
Valeriy Fedyunin
committed
Grammar fix
Valeriy Fedyunin
committed
Disable AvxTestDesktop everywhere but Windows
Valeriy Fedyunin
committed

Commits on Dec 10, 2022

Merge branch 'master' into BlockedConvExperiments
Valeriy Fedyunin
committed
Fix compilation
Valeriy Fedyunin
committed

Commits on Dec 19, 2022

Disable blocked conv for AVX512
Valeriy Fedyunin
committed