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

support_register_op_arange_fp16 cpu #10019

Merged
merged 5 commits into from
Mar 27, 2023
Merged

support_register_op_arange_fp16 cpu #10019

merged 5 commits into from
Mar 27, 2023

Conversation

ccssu
Copy link
Contributor

@ccssu ccssu commented Mar 22, 2023 •
edited
Loading

此pr为arange op cpu kernrel 注册 fp16。

cuda已经有注册在 Regist arange fp16 (#9202)

  • oneflow/user/kernels/arange_kernel.cpp 中 cpu kernrel 注册 fp16 类型 。
  • 添加 fp16 测试 在 python/oneflow/test/modules/test_arange.py

Reference

@ccssu ccssu requested review from BBuf and daquexian as code owners March 22, 2023 03:12
@CLAassistant
Copy link

CLAassistant commented Mar 22, 2023 •
edited
Loading

CLA assistant check
All committers have signed the CLA.

@ccssu ccssu requested a review from oneflow-ci-bot March 22, 2023 10:36
@github-actions
Copy link
Contributor

CI failed when running job: cpu-module. PR label automerge has been removed

@github-actions
Copy link
Contributor

Speed stats:

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 141.0ms (= 14096.3ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 142.8ms (= 14282.2ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.01 (= 142.8ms / 141.0ms)

OneFlow resnet50 time: 80.4ms (= 8038.6ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 83.6ms (= 8356.7ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.04 (= 83.6ms / 80.4ms)

OneFlow resnet50 time: 48.9ms (= 9772.5ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 58.9ms (= 11780.1ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.21 (= 58.9ms / 48.9ms)

OneFlow resnet50 time: 32.2ms (= 6437.7ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 44.3ms (= 8868.6ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.38 (= 44.3ms / 32.2ms)

OneFlow resnet50 time: 25.3ms (= 5053.3ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 40.9ms (= 8174.6ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.62 (= 40.9ms / 25.3ms)

OneFlow swin dataloader time: 0.238s (= 47.552s / 200, num_workers=1)
PyTorch swin dataloader time: 0.145s (= 29.066s / 200, num_workers=1)
Relative speed: 0.611 (= 0.145s / 0.238s)

OneFlow swin dataloader time: 0.071s (= 14.251s / 200, num_workers=4)
PyTorch swin dataloader time: 0.043s (= 8.538s / 200, num_workers=4)
Relative speed: 0.599 (= 0.043s / 0.071s)

OneFlow swin dataloader time: 0.042s (= 8.447s / 200, num_workers=8)
PyTorch swin dataloader time: 0.023s (= 4.604s / 200, num_workers=8)
Relative speed: 0.545 (= 0.023s / 0.042s)

❌ OneFlow resnet50 time: 153.0ms (= 15295.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 161.9ms (= 16185.0ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.06 (= 161.9ms / 153.0ms)

OneFlow resnet50 time: 91.0ms (= 9098.2ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 106.4ms (= 10643.1ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.17 (= 106.4ms / 91.0ms)

OneFlow resnet50 time: 59.1ms (= 11811.8ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.0ms (= 15592.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.32 (= 78.0ms / 59.1ms)

OneFlow resnet50 time: 41.8ms (= 8362.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 69.7ms (= 13944.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.67 (= 69.7ms / 41.8ms)

OneFlow resnet50 time: 35.6ms (= 7118.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 73.2ms (= 14636.5ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 2.06 (= 73.2ms / 35.6ms)

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/10019/

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 140.8ms (= 14080.5ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 142.9ms (= 14287.5ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.01 (= 142.9ms / 140.8ms)

OneFlow resnet50 time: 80.5ms (= 8051.4ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 84.4ms (= 8442.6ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.05 (= 84.4ms / 80.5ms)

OneFlow resnet50 time: 49.2ms (= 9847.0ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 56.6ms (= 11321.1ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.15 (= 56.6ms / 49.2ms)

OneFlow resnet50 time: 32.6ms (= 6520.8ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 43.2ms (= 8630.9ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.32 (= 43.2ms / 32.6ms)

OneFlow resnet50 time: 24.9ms (= 4975.4ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 36.7ms (= 7343.6ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.48 (= 36.7ms / 24.9ms)

OneFlow swin dataloader time: 0.234s (= 46.852s / 200, num_workers=1)
PyTorch swin dataloader time: 0.151s (= 30.104s / 200, num_workers=1)
Relative speed: 0.643 (= 0.151s / 0.234s)

OneFlow swin dataloader time: 0.070s (= 13.908s / 200, num_workers=4)
PyTorch swin dataloader time: 0.042s (= 8.430s / 200, num_workers=4)
Relative speed: 0.606 (= 0.042s / 0.070s)

OneFlow swin dataloader time: 0.042s (= 8.404s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.330s / 200, num_workers=8)
Relative speed: 0.515 (= 0.022s / 0.042s)

❌ OneFlow resnet50 time: 153.0ms (= 15304.7ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 166.6ms (= 16660.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.09 (= 166.6ms / 153.0ms)

OneFlow resnet50 time: 91.3ms (= 9125.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 101.3ms (= 10126.4ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.11 (= 101.3ms / 91.3ms)

OneFlow resnet50 time: 59.6ms (= 11912.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.7ms (= 15746.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.32 (= 78.7ms / 59.6ms)

OneFlow resnet50 time: 42.5ms (= 8508.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 74.3ms (= 14856.4ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.75 (= 74.3ms / 42.5ms)

OneFlow resnet50 time: 36.4ms (= 7274.5ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 67.1ms (= 13415.3ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.84 (= 67.1ms / 36.4ms)

@ccssu ccssu requested review from oneflow-ci-bot and removed request for oneflow-ci-bot March 23, 2023 13:24
@github-actions
Copy link
Contributor

Speed stats:

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 140.9ms (= 14089.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 142.3ms (= 14228.7ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.01 (= 142.3ms / 140.9ms)

OneFlow resnet50 time: 80.4ms (= 8043.5ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 85.3ms (= 8530.0ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.06 (= 85.3ms / 80.4ms)

OneFlow resnet50 time: 49.3ms (= 9852.4ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 57.4ms (= 11488.5ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.17 (= 57.4ms / 49.3ms)

OneFlow resnet50 time: 32.3ms (= 6454.3ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 42.6ms (= 8529.2ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.32 (= 42.6ms / 32.3ms)

OneFlow resnet50 time: 25.1ms (= 5015.4ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 41.5ms (= 8309.7ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.66 (= 41.5ms / 25.1ms)

OneFlow swin dataloader time: 0.235s (= 46.978s / 200, num_workers=1)
PyTorch swin dataloader time: 0.152s (= 30.333s / 200, num_workers=1)
Relative speed: 0.646 (= 0.152s / 0.235s)

OneFlow swin dataloader time: 0.070s (= 14.035s / 200, num_workers=4)
PyTorch swin dataloader time: 0.041s (= 8.133s / 200, num_workers=4)
Relative speed: 0.579 (= 0.041s / 0.070s)

OneFlow swin dataloader time: 0.043s (= 8.594s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.486s / 200, num_workers=8)
Relative speed: 0.522 (= 0.022s / 0.043s)

❌ OneFlow resnet50 time: 152.7ms (= 15265.7ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 161.8ms (= 16180.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.06 (= 161.8ms / 152.7ms)

OneFlow resnet50 time: 90.9ms (= 9094.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 103.8ms (= 10384.1ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.14 (= 103.8ms / 90.9ms)

OneFlow resnet50 time: 59.1ms (= 11828.4ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.7ms (= 15736.0ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.33 (= 78.7ms / 59.1ms)

OneFlow resnet50 time: 41.8ms (= 8366.1ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 71.2ms (= 14248.5ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.70 (= 71.2ms / 41.8ms)

OneFlow resnet50 time: 36.6ms (= 7324.4ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 70.9ms (= 14175.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.94 (= 70.9ms / 36.6ms)

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/10019/

@ccssu ccssu enabled auto-merge (squash) March 27, 2023 02:36
@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 141.1ms (= 14108.6ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 143.3ms (= 14326.0ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.02 (= 143.3ms / 141.1ms)

OneFlow resnet50 time: 80.7ms (= 8069.1ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 84.0ms (= 8403.9ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.04 (= 84.0ms / 80.7ms)

OneFlow resnet50 time: 49.6ms (= 9914.8ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 57.9ms (= 11582.1ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.17 (= 57.9ms / 49.6ms)

OneFlow resnet50 time: 33.0ms (= 6593.4ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 41.6ms (= 8317.4ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.26 (= 41.6ms / 33.0ms)

OneFlow resnet50 time: 25.1ms (= 5014.0ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 35.9ms (= 7187.2ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.43 (= 35.9ms / 25.1ms)

OneFlow swin dataloader time: 0.235s (= 47.047s / 200, num_workers=1)
PyTorch swin dataloader time: 0.147s (= 29.478s / 200, num_workers=1)
Relative speed: 0.627 (= 0.147s / 0.235s)

OneFlow swin dataloader time: 0.068s (= 13.625s / 200, num_workers=4)
PyTorch swin dataloader time: 0.040s (= 8.004s / 200, num_workers=4)
Relative speed: 0.587 (= 0.040s / 0.068s)

OneFlow swin dataloader time: 0.041s (= 8.226s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.352s / 200, num_workers=8)
Relative speed: 0.529 (= 0.022s / 0.041s)

❌ OneFlow resnet50 time: 152.8ms (= 15283.3ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 163.3ms (= 16334.1ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.07 (= 163.3ms / 152.8ms)

OneFlow resnet50 time: 91.4ms (= 9135.1ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 108.3ms (= 10827.5ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.19 (= 108.3ms / 91.4ms)

OneFlow resnet50 time: 59.6ms (= 11912.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 81.4ms (= 16270.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.37 (= 81.4ms / 59.6ms)

OneFlow resnet50 time: 42.3ms (= 8462.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 68.9ms (= 13776.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.63 (= 68.9ms / 42.3ms)

OneFlow resnet50 time: 37.0ms (= 7407.9ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 66.7ms (= 13345.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.80 (= 66.7ms / 37.0ms)

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 141.1ms (= 14110.0ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 144.1ms (= 14408.7ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.02 (= 144.1ms / 141.1ms)

OneFlow resnet50 time: 81.2ms (= 8120.7ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 85.4ms (= 8537.7ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.05 (= 85.4ms / 81.2ms)

OneFlow resnet50 time: 50.6ms (= 10124.3ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 57.0ms (= 11391.5ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.13 (= 57.0ms / 50.6ms)

OneFlow resnet50 time: 33.6ms (= 6724.7ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 42.8ms (= 8551.6ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.27 (= 42.8ms / 33.6ms)

OneFlow resnet50 time: 26.4ms (= 5278.9ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 39.8ms (= 7957.4ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.51 (= 39.8ms / 26.4ms)

OneFlow swin dataloader time: 0.240s (= 47.957s / 200, num_workers=1)
PyTorch swin dataloader time: 0.148s (= 29.600s / 200, num_workers=1)
Relative speed: 0.617 (= 0.148s / 0.240s)

OneFlow swin dataloader time: 0.070s (= 14.047s / 200, num_workers=4)
PyTorch swin dataloader time: 0.043s (= 8.540s / 200, num_workers=4)
Relative speed: 0.608 (= 0.043s / 0.070s)

OneFlow swin dataloader time: 0.043s (= 8.531s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.437s / 200, num_workers=8)
Relative speed: 0.520 (= 0.022s / 0.043s)

❌ OneFlow resnet50 time: 152.8ms (= 15284.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 167.0ms (= 16695.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.09 (= 167.0ms / 152.8ms)

OneFlow resnet50 time: 92.1ms (= 9210.2ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 105.9ms (= 10587.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.15 (= 105.9ms / 92.1ms)

OneFlow resnet50 time: 60.7ms (= 12135.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 83.4ms (= 16689.4ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.38 (= 83.4ms / 60.7ms)

OneFlow resnet50 time: 42.1ms (= 8410.5ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 70.7ms (= 14144.0ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.68 (= 70.7ms / 42.1ms)

OneFlow resnet50 time: 36.6ms (= 7325.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 71.5ms (= 14295.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.95 (= 71.5ms / 36.6ms)

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/10019/

@ccssu ccssu merged commit 7f1a02a into master Mar 27, 2023
@ccssu ccssu deleted the support_register_op_arange_fp16 branch March 27, 2023 09:14
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@BBuf BBuf BBuf approved these changes

@marigoold marigoold marigoold approved these changes

@daquexian daquexian Awaiting requested review from daquexian

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

Assignees
No one assigned
Labels
api feature global ir
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@ccssu @CLAassistant @BBuf @marigoold