Performance Signal Detected by TorchBench CI on '1.12.0.dev20220328+cu113'

[github-actions[bot]]

TorchBench CI has detected a performance signal.

Base PyTorch version: 1.12.0.dev20220323+cu113

Base PyTorch commit: 65329f4fac8fb22318b7a3eb115e9da207d8d41a

Affected PyTorch version: 1.12.0.dev20220328+cu113

Affected PyTorch commit: 1c5a8125798392f8d7c57e88735f43a14ae0beca

Affected Tests:

• test_train[Super_SloMo-cuda-jit]: -16.92249%
• test_eval[Super_SloMo-cuda-jit]: +12.22803%
• test_train[Super_SloMo-cuda-eager]: -16.85775%
• test_eval[Super_SloMo-cuda-eager]: +12.09915%
• test_eval[alexnet-cuda-jit]: -14.82252%
• test_eval[alexnet-cuda-eager]: -14.13340%
• test_train[densenet121-cuda-jit]: -7.82857%
• test_train[densenet121-cuda-eager]: -7.64626%
• test_train[maml-cuda-eager]: -12.12154%
• test_eval[maml-cuda-eager]: -13.42175%
• test_eval[maml_omniglot-cuda-eager]: -7.97607%
• test_eval[mnasnet1_0-cuda-jit]: -95.44156%
• test_train[mobilenet_v2-cuda-jit]: -10.52894%
• test_eval[mobilenet_v2-cuda-jit]: -14.17461%
• test_train[mobilenet_v2-cuda-eager]: -10.54444%
• test_eval[mobilenet_v2-cuda-eager]: -10.08228%
• test_eval[mobilenet_v3_large-cuda-jit]: -23.09441%
• test_train[moco-cuda-jit]: -9.83372%
• test_train[moco-cuda-eager]: -9.84092%
• test_eval[opacus_cifar10-cuda-eager]: -27.95975%
• test_eval[pytorch_stargan-cuda-jit]: -30.42362%
• test_eval[pytorch_stargan-cuda-eager]: -30.45168%
• test_train[resnet18-cuda-jit]: -17.28054%
• test_train[resnet18-cuda-eager]: -17.31519%
• test_train[resnet50-cuda-jit]: -16.86644%
• test_eval[resnet50-cuda-jit]: -20.43296%
• test_train[resnet50-cuda-eager]: -16.85397%
• test_eval[resnet50-cuda-eager]: -8.91445%
• test_train[resnet50_quantized_qat-cuda-eager]: -11.99613%
• test_train[resnext50_32x4d-cuda-jit]: -18.83313%
• test_train[resnext50_32x4d-cuda-eager]: -18.96089%
• test_eval[resnext50_32x4d-cuda-eager]: -12.01397%
• test_eval[shufflenet_v2_x1_0-cuda-jit]: +8.22455%
• test_train[speech_transformer-cuda-eager]: -8.09863%
• test_train[squeezenet1_1-cuda-jit]: -10.53504%
• test_eval[squeezenet1_1-cuda-jit]: -9.60593%
• test_train[squeezenet1_1-cuda-eager]: -10.52895%
• test_eval[squeezenet1_1-cuda-eager]: -10.20297%
• test_train[timm_resnest-cuda-jit]: -35.27120%
• test_eval[timm_resnest-cuda-jit]: -8.42915%
• test_train[timm_resnest-cuda-eager]: -35.60262%
• test_eval[timm_resnest-cuda-eager]: -10.86361%
• test_train[vgg16-cuda-jit]: -9.51113%
• test_train[vgg16-cuda-eager]: -9.53890%

cc @xuzhao9

[xuzhao9]

Speedup caused by cudnn version upgrade.