fix tensor size limit resulted from shared memory size

[liufengwei0103]

No description provided.

[hengzi]

我用这个分支试了下，input 大一点还是有问题，例如 flow.ones(1024, 128)，但报错信息不一样了

Check failed: cudaMallocHost(&ptr, 1 << granularity) : an illegal memory access was encountered (700) 

[liufengwei0103]

cudaMallocHost(&ptr, 1 << granularity)

好的，我再看一下。

[liufengwei0103]

cudaMallocHost(&ptr, 1 << granularity)

好的，我再看一下。

更新代码之后，没有问题了。

[github-actions]

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/8289/

[github-actions]

Speed stats:

GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 130.5ms (= 13045.6ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 142.6ms (= 14263.2ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.09 (= 142.6ms / 130.5ms)

OneFlow resnet50 time: 78.9ms (= 7895.0ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 87.8ms (= 8778.7ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.11 (= 87.8ms / 78.9ms)

OneFlow resnet50 time: 53.8ms (= 10764.0ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 62.6ms (= 12510.5ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.16 (= 62.6ms / 53.8ms)

OneFlow resnet50 time: 41.2ms (= 8243.0ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 43.7ms (= 8741.1ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.06 (= 43.7ms / 41.2ms)

OneFlow resnet50 time: 37.5ms (= 7497.5ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 40.1ms (= 8028.9ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.07 (= 40.1ms / 37.5ms)

OneFlow swin dataloader time: 0.242s (= 48.309s / 200, num_workers=1)
PyTorch swin dataloader time: 0.152s (= 30.351s / 200, num_workers=1)
Relative speed: 0.628 (= 0.152s / 0.242s)

OneFlow swin dataloader time: 0.064s (= 12.789s / 200, num_workers=4)
PyTorch swin dataloader time: 0.041s (= 8.135s / 200, num_workers=4)
Relative speed: 0.636 (= 0.041s / 0.064s)

OneFlow swin dataloader time: 0.058s (= 11.579s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.427s / 200, num_workers=8)
Relative speed: 0.382 (= 0.022s / 0.058s)

❌ OneFlow resnet50 time: 148.2ms (= 14816.0ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 172.3ms (= 17231.4ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.16 (= 172.3ms / 148.2ms)

OneFlow resnet50 time: 99.4ms (= 9935.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 111.7ms (= 11170.5ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.12 (= 111.7ms / 99.4ms)

OneFlow resnet50 time: 74.0ms (= 14805.4ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 86.9ms (= 17380.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.17 (= 86.9ms / 74.0ms)

OneFlow resnet50 time: 61.8ms (= 12366.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 72.9ms (= 14570.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.18 (= 72.9ms / 61.8ms)

OneFlow resnet50 time: 55.3ms (= 11054.4ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 77.6ms (= 15524.1ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.40 (= 77.6ms / 55.3ms)

[github-actions]

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

[github-actions]

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.

[github-actions]

Speed stats:

[github-actions]

Speed stats:

[github-actions]

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/8289/