Vectorize InstanceNormalization and BatchNormalization #469
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
This enables printing a debug representation of a vector implementing
`Simd` via `println!("{:?}", simd_vec.to_array())`.
Add two vectorized functions that will be useful as part of the InstanceNormalization operation. - `vec_sum_square_sub` is like `vec_sum_square` but subtracts a constant from each element before squaring. - `vec_shift_scale_bias` subtracts a constant from each element and then shifts and scales the result.
…tion
InstanceNormalization has three steps:
1. Compute channel mean
2. Compute channel variance
3. Shift and scale result to normalize the mean and variance, and then
apply a per-channel scale and bias
Previously only step 1 was vectorized. This vectorizes steps 2 and 3 as well.
Tested using the wav2vec example on x64, this made InstanceNormalization ~3x
faster (~26ms -> ~8ms per run).
BatchNormalization and InstanceNormalization are very similar, except BatchNormalization uses pre-computed mean and variance statistics while InstanceNormalization computes them dynamically. Extract the vectorized normalization step from `instance_normalization` and re-use it for `batch_normalization`.
robertknight
changed the title
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Outdated suggestions cannot be applied.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.
InstanceNormalization has two steps:
BatchNormalization is similar but uses precomputed values for step 1. Previously part of step 1 was vectorized, but not the variance calculation. Add a vectorized kernel for step 2 and use it for BatchNormalization and InstanceNormalization.
Tested using the wav2vec example on x64, this made InstanceNormalization ~3x faster (~26ms -> ~8ms per run).