Commit Graph

2238 Commits

Author SHA1 Message Date
Renault Fernandes 3a0dc55cad
Use cublasStatus_t instead of cudaError_t when error checking cublas calls (#7626)
Cublas success is indicated by CUBLAS_STATUS_SUCCESS. This must be used
to determine if a cublas function call was successful.

Although Cuda does not complain if we use cudaError_t, when bulding with
ROCm support it is absolutely necessary to use the correct enum
(cublasStatus_t), else ROCm will always interpret the returned
status as an error and crash.
2021-04-26 00:59:41 +03:00
deoksangkim c9f2c5f0e9
fix memory error in batchnorm layer (#7619) 2021-04-26 00:58:22 +03:00
Stefano Sinigardi 4f794aa152
almost revert last PR (#7634)
* [darknet] split pr/ci pipelines to use fast caching only when accessible

* restore original pipeline names

* restore original ccpp file name

* speed up scripts avoiding -wait flag

* restore fast cache in readonly

* clean up scripts removing references to github packages, which are often down/broken
2021-04-22 18:03:23 +03:00
Stefano Sinigardi 91efb7c7f6
Deploy CUDA 11.3 in CI (#7630)
* [nuget] use personal baget repo

* update cuda to 11.3 on windows

* add error messages during build to not have green badge when failed

* remove cygwin script, is untested now; use deploy-cuda.ps1 script also inside setup.ps1
2021-04-21 19:25:46 +03:00
Alexey 96971deaa7
Update README.md 2021-04-21 03:11:31 +03:00
Renault Fernandes 8bc512dca3
Fix no inference detections bug when cudnn is disabled. (#7617)
Fixes issue described here: https://github.com/AlexeyAB/darknet/issues/7616
2021-04-20 01:58:00 +03:00
Stefano Sinigardi b6cd759218 apply feedback from vcpkg ci 2021-04-18 16:49:54 +02:00
Stefano Sinigardi ba8f7bf756
improve build process even more (#7610)
* test for shared libs

* improve flag handling

* build so lib in ci

* clone vcpkg if not found

* fix

* improve clang handling

* bump vcpkg.json version

* gemm.c compatible with ARM

* remove unnecessary vcpkg.json field

* do not unnecessarily rebuild vcpkg in the CI build stage

* use alexeyab nuget cache

* enable self-removal of build folder

* add interactivity in build.ps1 if not opt-out

* spellcheck README

* fix another leftover spelling error

* also auto-update darknet if possible

* do not self-update darknet in CI
2021-04-18 16:11:36 +03:00
Alexey 1f4598aefa
Merge pull request #6402 from lineofbestgit/master
Small grammar change / sentence repetition fix
2021-04-16 22:48:16 +03:00
Stefano Sinigardi cfd466b554
Merge branch 'master' into master 2021-04-16 15:42:15 +02:00
Alexey d7c37b5616
Merge pull request #7279 from cenit/dev/cenit/fixes
move to opencv4.5 on vcpkg in ci, do not default to ninja on windows, update README
2021-04-16 02:05:05 +03:00
Stefano Sinigardi 580a5a6c9b improve readme more 2021-04-15 17:59:23 +02:00
Stefano Sinigardi b064668bbb improve readme wording about build.ps1 2021-04-15 14:53:58 +02:00
Stefano Sinigardi e1c02e02da circumvent current upstream bugs 2021-04-15 08:15:27 +02:00
Stefano Sinigardi 56cd9c6e8c enable opencv w/cuda in build.ps1 2021-04-14 17:15:36 +02:00
Stefano Sinigardi 0ee4ff36b2 download weights using a local script, do not rely on vcpkg 2021-04-14 15:19:06 +02:00
Stefano Sinigardi 013d5e676d use only build.ps1 2021-04-14 15:06:10 +02:00
Stefano Sinigardi 4a774e0836 find also vcpkg nested inside darknet folder 2021-04-14 13:40:42 +02:00
Stefano Sinigardi 42ecbfa4b4 improve non-windows handling in ps1 script 2021-04-14 13:30:50 +02:00
Stefano Sinigardi fa8d21dee5 use build.ps1 also in CI to better test it 2021-04-14 11:50:54 +02:00
Stefano Sinigardi b409569ee4 upgrade to run-vcpkg@v7 2021-04-11 19:23:49 +02:00
Stefano Sinigardi 0ee9c108cf
Merge branch 'master' into dev/cenit/fixes 2021-04-11 19:20:50 +02:00
Alexey e83d6526cd
Update README.md 2021-04-07 03:16:01 +03:00
AlexeyAB 1e3a616ed6 compile fix 2021-03-28 18:49:55 +03:00
AlexeyAB 4949b6cd84 minor fix 2021-03-28 18:13:43 +03:00
Alexey 730e5c1c6a
Update README.md 2021-03-28 17:51:02 +03:00
Alexey 22cb99e041
Merge pull request #7515 from kikaitech/cuda_floorf_ceilf
Cuda: floor -> floorf, ceil -> ceilf
2021-03-24 00:08:39 +03:00
hainq 0bd7e6713d cuda: floor -> floorf, ceil -> ceilf 2021-03-19 12:13:26 +07:00
AlexeyAB b8c9c9d457 minor fix 2021-03-01 05:44:04 +03:00
AlexeyAB f058d26a8a circle ci fix 2021-03-01 05:35:05 +03:00
AlexeyAB 1a874378e4 Try new docker image for circle-ci 2021-02-28 18:43:16 +03:00
AlexeyAB 92c92017b2 Try new docker image for circle-ci 2021-02-28 18:38:42 +03:00
AlexeyAB 7bcac1c84b Try new docker for circle-ci 2021-02-28 09:30:31 +03:00
AlexeyAB f8df52ad72 minor fix 2021-02-28 09:11:22 +03:00
AlexeyAB 53db0163eb Added CUDA graphs/streams. To speed up the inference use: [net] use_cuda_graph = 1 2021-02-28 08:57:44 +03:00
Stefano Sinigardi 94fdfa4590 use ninja whenever possible in build.ps1 2021-02-21 22:59:36 +01:00
Stefano Sinigardi 44c868baca add missing pubkey install step 2021-02-21 18:11:02 +01:00
Stefano Sinigardi 552bddc140 move to ubuntu 20.04 on github actions 2021-02-21 18:00:59 +01:00
Stefano Sinigardi 6a2cab982a use same opencv feature set independently of cuda 2021-02-21 12:47:05 +01:00
Stefano Sinigardi 44d29d5982 do not enable cudnn feature if cuda is not enabled 2021-02-21 10:31:29 +01:00
Stefano Sinigardi 8f3cd2840e fixes 2021-02-20 17:05:27 +01:00
Stefano Sinigardi fd6ea9ba60 use same logic for all enables: if on, then package is required 2021-02-20 11:22:23 +01:00
Stefano Sinigardi b9d296e308 improve manifest mode 2021-02-20 09:28:07 +01:00
Stefano Sinigardi 9284d02809 comment out windows-vcpkg, improve setup scripts 2021-02-18 22:22:39 +01:00
Stefano Sinigardi b5d852057e fix 2021-01-29 17:40:29 +01:00
Stefano Sinigardi 661e8ab26b add nasm dependency 2021-01-29 17:39:25 +01:00
Stefano Sinigardi dc71b14e77 fix interface included directory for opencv 2021-01-29 17:08:09 +01:00
Stefano Sinigardi 05bda64c92 Update vcpkg refs to OpenCV 4.5.1 2021-01-29 08:30:06 +01:00
Stefano Sinigardi 5f84f90909 revert to double backslash 2021-01-24 17:07:10 +01:00
Stefano Sinigardi e2a4253bb0 add cuda to path to github ci; improve diagnostic messages 2021-01-24 15:31:13 +01:00