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

enable cuda graph support for Conv2dCutlassKernel #9761

Merged
merged 8 commits into from
Jan 17, 2023
Merged

enable cuda graph support for Conv2dCutlassKernel #9761

merged 8 commits into from
Jan 17, 2023

Conversation

liujuncheng
Copy link
Collaborator

启用cuda graph的情况下，tuning部分的代码不能执行在计算stream上，同时，tensor 也无法保证可用，所以重新分配

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 









❌ OneFlow resnet50 time: 139.3ms (= 13929.3ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 167.6ms (= 16757.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 167.6ms / 139.3ms)

OneFlow resnet50 time: 85.5ms (= 8553.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 101.9ms (= 10190.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.19 (= 101.9ms / 85.5ms)

OneFlow resnet50 time: 57.8ms (= 11555.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.2ms (= 15649.5ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.35 (= 78.2ms / 57.8ms)

OneFlow resnet50 time: 44.0ms (= 8803.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 81.1ms (= 16226.4ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.84 (= 81.1ms / 44.0ms)

OneFlow resnet50 time: 40.9ms (= 8172.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 72.7ms (= 14543.9ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.78 (= 72.7ms / 40.9ms)

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/9761/

@mergify mergify bot mentioned this pull request Jan 17, 2023
Merged
@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 









❌ OneFlow resnet50 time: 140.3ms (= 14026.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 160.9ms (= 16092.3ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.15 (= 160.9ms / 140.3ms)

OneFlow resnet50 time: 89.1ms (= 8908.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 103.0ms (= 10301.2ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.16 (= 103.0ms / 89.1ms)

OneFlow resnet50 time: 64.7ms (= 12943.5ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 87.9ms (= 17571.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.36 (= 87.9ms / 64.7ms)

OneFlow resnet50 time: 50.7ms (= 10133.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 80.4ms (= 16084.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.59 (= 80.4ms / 50.7ms)

OneFlow resnet50 time: 42.7ms (= 8544.2ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 67.3ms (= 13462.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.58 (= 67.3ms / 42.7ms)

@liujuncheng liujuncheng enabled auto-merge (squash) January 17, 2023 12:43
@github-actions
Copy link
Contributor

Code got formatted by CI. Please request CI again if you still want to have this PR merged. If the PR is from a forked repo, please download the patch files from the GitHub Actions web page and apply them locally.

@liujuncheng liujuncheng requested review from oneflow-ci-bot and removed request for oneflow-ci-bot January 17, 2023 14:06
@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 









❌ OneFlow resnet50 time: 139.7ms (= 13966.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 160.3ms (= 16031.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.15 (= 160.3ms / 139.7ms)

OneFlow resnet50 time: 87.9ms (= 8788.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 101.2ms (= 10124.5ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.15 (= 101.2ms / 87.9ms)

OneFlow resnet50 time: 63.7ms (= 12732.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 88.1ms (= 17614.9ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.38 (= 88.1ms / 63.7ms)

OneFlow resnet50 time: 46.4ms (= 9287.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 70.6ms (= 14111.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.52 (= 70.6ms / 46.4ms)

OneFlow resnet50 time: 41.8ms (= 8368.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 68.5ms (= 13695.3ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.64 (= 68.5ms / 41.8ms)

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/9761/

@liujuncheng liujuncheng merged commit d028365 into master Jan 17, 2023
@liujuncheng liujuncheng deleted the dev_cutlass_conv_cuda_graph branch January 17, 2023 19:20
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@leaves-zwx leaves-zwx leaves-zwx approved these changes

@howin98 howin98 howin98 approved these changes

@oneflow-ci-bot oneflow-ci-bot Awaiting requested review from oneflow-ci-bot

Assignees
No one assigned
Labels
automerge enhancement op
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@liujuncheng @leaves-zwx @howin98 @oneflow-ci-bot