Fix an issue when fusing BatchNorm in backprop #48893
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
kaixih
force-pushed
the
bn_act_grad_double_bn_upstream
branch
2 times, most recently
from
71c16ee
to
289dd53
Compare
sanjoy
approved these changes
May 4, 2021
google-ml-butler
bot
added
kokoro:force-run
gbaned
added
ready to pull
copybara-service bot
pushed a commit
that referenced
this pull request
May 6, 2021
Still hits "unsupported cudnn config" error PiperOrigin-RevId: 372364589 Change-Id: Ice2e722179b46090d28a26fb1d3346bfcc131ebf
kaixih
referenced
this pull request
May 14, 2021
…ctivation Still hits "unsupported cudnn" error To repro, run (from tensorflow_models repository): ``` //tensorflow_models/official/benchmark:keras_imagenet_benchmark -- --logtostderr --benchmarks=Resnet50KerasBenchmarkSynth.benchmark_xla_1_gpu_fp16\$ ``` PiperOrigin-RevId: 372380247 Change-Id: Iae61fa8095b0c60716cfdf147544d8f7b3e3f493
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Previous PR #48063 might hit CUDNN_STATUS_NOT_SUPPORTED error if the Add node has both inputs to be the BatchNorm nodes, in which case we will mistakenly mark both BatchNormGrad as valid for fusion. Though it is not possible that both BatchNormGrad in the backward pass will be fused, there is chance that the forward pass and backward pass select different branch to fuse, causing the unexpected reserve space sizes.
This PR fixes this issue by checking if the BatchNormGrad's directly connected forward BatchNorm matches the to-be-fused forward BatchNorm. In addition, the unit tests are updated to test the double BN inputs of add op.
cc. @nluehr