Skip to content
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

Running a specific convolution algorithm #61

Closed
aryamazaheri opened this issue Nov 7, 2018 · 3 comments
Closed

Running a specific convolution algorithm #61

aryamazaheri opened this issue Nov 7, 2018 · 3 comments

Comments

@aryamazaheri
Copy link

Is it possible to run a specific convolution algorithm without searching through all the algorithms? I noticed that running miopenConvolutionForwardGetWorkSpaceSize() is mandatory before calling miopenConvolutionForward but this prevents me from running a convolution with a specific algorithm that I have in mind. What is the alternative way?
@dagamayank
Copy link
Contributor

@aryamazaheri No, not right now. We are actively working to implement this feature. As an alternative you can try disabling some algorithms as mentioned here.

@aryamazaheri
Copy link
Author

Thanks. Apparently, it is not possible to solely keep Winograd algorithm enabled! :(
There is no flag devised for disabling GEMM conv.

@dagamayank
Copy link
Contributor

There is no flag devised for disabling GEMM conv.

Right, there is no flag for GEMM. However, you can follow a somewhat contrived method to disable GEMM completely. This requires you to build MIOpen from source and use the following options at the cmake configure step:
cmake -DCMAKE_DISABLE_FIND_PACKAGE_miopengemm=TRUE -DMIOPEN_USE_ROCBLAS=false..

Note: once the above is done you may notice some convolutions to fail as the default fallback, i.e., GEMM is now removed. Therefore, use it at your own risk.

Instructions on how to build MIOpen from source can be found in the README.

odellus pushed a commit to synpon/MIOpen that referenced this issue Nov 15, 2018
@ce1adon
Merge pull request ROCm#61 from AMDComputeLibraries/develop
4264619 
up2dev
@alexandraBara alexandraBara mentioned this issue Sep 11, 2020
Closed
cderb added a commit that referenced this issue Oct 5, 2022
@cderb
Squashed 'fin/' changes from 53d2563fe..e05dcb421
8875272 
e05dcb421 perf db validation fix (#68)
260d9465d Add INT8 as a data_type v2 (#67)
b6a5b2a77 sync with fin folder in miopen (#62)
0e03399ec prep for Palamida scan (#63)
e6bd05c33 Performance db testing (#61)
30d699b9e Perf Eval Update (#60)
3535b948c PerfCompile and PerfEval changes (#59)
de79468d2 remove unneccessary solution check, add check for previously modified kernel names (#56)
6924286a2 miopen hash update (#55)
530399575 Refactor googletest infra to align with MIOpen (#53)
71c50d146 Datatype fix for BN (#57)
8abe2f5c6 Perf Eval updates, Add find info (#51)
e1c1ef0f5 filter find compile by solver input (#54)
722feea66 sp/chk precomp kernel 264 (#41)
b9aba2034 Batch norm find compile (#50)
359f3da80 Fix missing link directives in fin binary (#48)
a4020c1ba Cache Miss Fixes (#46)
2ec7ef44d Enable google test and compiling fin in the CI (#47)
8b6b453bc Applicability support for batch norm (#45)
44323aae9 Perf compile/eval for fin (#42)
ebd9aa6bd update member name (#43)
d6d798efe add cu count (#39)
8e1989a9f Add find option for selecting only dynamic solvers (#38)
0e164bf66 setting json version (#37)
f3f7fed18 Remove function redefinition (#36)
e1de51a58 Performance DB de-serialize test (#34)
043cdcdaa Layout support in Fin (#33)
3a1d58236 Hotfix (#32)
ee3f0d543 4.4 Tuning Bugfixes (#31)
832dbe234 Tunability Reporting (#27)
a564a229f include gfx90a_110 (#28)

git-subtree-dir: fin
git-subtree-split: e05dcb42187f05fe0d0d1b05b822dc4b750f199e
junliume added a commit that referenced this issue Oct 6, 2022
@cderb @junliume
Perf DB cleanup + Fin pdb verification fix (#1810)
e7d3c60 
* remove datatype 0,1 from perf_db

* rm invalid fp16 entries from pdb

* Squashed 'fin/' changes from 53d2563fe..e05dcb421

e05dcb421 perf db validation fix (#68)
260d9465d Add INT8 as a data_type v2 (#67)
b6a5b2a77 sync with fin folder in miopen (#62)
0e03399ec prep for Palamida scan (#63)
e6bd05c33 Performance db testing (#61)
30d699b9e Perf Eval Update (#60)
3535b948c PerfCompile and PerfEval changes (#59)
de79468d2 remove unneccessary solution check, add check for previously modified kernel names (#56)
6924286a2 miopen hash update (#55)
530399575 Refactor googletest infra to align with MIOpen (#53)
71c50d146 Datatype fix for BN (#57)
8abe2f5c6 Perf Eval updates, Add find info (#51)
e1c1ef0f5 filter find compile by solver input (#54)
722feea66 sp/chk precomp kernel 264 (#41)
b9aba2034 Batch norm find compile (#50)
359f3da80 Fix missing link directives in fin binary (#48)
a4020c1ba Cache Miss Fixes (#46)
2ec7ef44d Enable google test and compiling fin in the CI (#47)
8b6b453bc Applicability support for batch norm (#45)
44323aae9 Perf compile/eval for fin (#42)
ebd9aa6bd update member name (#43)
d6d798efe add cu count (#39)
8e1989a9f Add find option for selecting only dynamic solvers (#38)
0e164bf66 setting json version (#37)
f3f7fed18 Remove function redefinition (#36)
e1de51a58 Performance DB de-serialize test (#34)
043cdcdaa Layout support in Fin (#33)
3a1d58236 Hotfix (#32)
ee3f0d543 4.4 Tuning Bugfixes (#31)
832dbe234 Tunability Reporting (#27)
a564a229f include gfx90a_110 (#28)

git-subtree-dir: fin
git-subtree-split: e05dcb42187f05fe0d0d1b05b822dc4b750f199e

* fix clang-format issue

Co-authored-by: Jun Liu <Liu.Jun@amd.com>
cderb added a commit that referenced this issue Nov 21, 2022
@cderb
Squashed 'fin/' changes from 53d2563fe..49e3e3a62
4042531 
49e3e3a62 clang format
db80b1777 update to using TestPerfCfgParams for pdb validity checks
e48a4fd3a format
a4f85842c exception for non-tunable solvers in params check
d58c42bbd Check params at end of perf tuning (#70)
1a3b47c7b Return status for failed compile commands (#69)
d59962752 out_layout -> in_layout
6ba7a8f3f Rename conv_mode to mode (#64)
513a3da1b [bg/LWPTUNA-173] (#65)
e05dcb421 perf db validation fix (#68)
260d9465d Add INT8 as a data_type v2 (#67)
b6a5b2a77 sync with fin folder in miopen (#62)
0e03399ec prep for Palamida scan (#63)
e6bd05c33 Performance db testing (#61)
30d699b9e Perf Eval Update (#60)
3535b948c PerfCompile and PerfEval changes (#59)
de79468d2 remove unneccessary solution check, add check for previously modified kernel names (#56)
6924286a2 miopen hash update (#55)
530399575 Refactor googletest infra to align with MIOpen (#53)
71c50d146 Datatype fix for BN (#57)
8abe2f5c6 Perf Eval updates, Add find info (#51)
e1c1ef0f5 filter find compile by solver input (#54)
722feea66 sp/chk precomp kernel 264 (#41)
b9aba2034 Batch norm find compile (#50)
359f3da80 Fix missing link directives in fin binary (#48)
a4020c1ba Cache Miss Fixes (#46)
2ec7ef44d Enable google test and compiling fin in the CI (#47)
8b6b453bc Applicability support for batch norm (#45)
44323aae9 Perf compile/eval for fin (#42)
ebd9aa6bd update member name (#43)
d6d798efe add cu count (#39)
8e1989a9f Add find option for selecting only dynamic solvers (#38)
0e164bf66 setting json version (#37)
f3f7fed18 Remove function redefinition (#36)
e1de51a58 Performance DB de-serialize test (#34)
043cdcdaa Layout support in Fin (#33)
3a1d58236 Hotfix (#32)
ee3f0d543 4.4 Tuning Bugfixes (#31)
832dbe234 Tunability Reporting (#27)
a564a229f include gfx90a_110 (#28)

git-subtree-dir: fin
git-subtree-split: 49e3e3a62a7cc54adacbeea95680d35f9a4685de
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@aryamazaheri @dagamayank