TensorMap: GPU support #3
Open
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
kttian
pushed a commit
that referenced
this pull request
Aug 12, 2020
…n of ADD/SUB reference kernel operators Imported from GitHub PR tensorflow#35985 This PR is one of steps to extend 8-bit quantization to support symmetric 16-bit activations. Each activation is of type int16 and symmetric around zero. The weight tensor precision remains at 8-bit signed values. The bias is set to int64 precision. In this PR we introduce implementation and tests for ADD/SUB kernel reference function. The specification of this operator: SUB Input 0: data_type : int16 range : [-32768, 32767] granularity: per-tensor, zero_point=0 Input 1: data_type : int16 range : [-32768, 32767] granularity: per-tensor, zero_point=0 Output 0: data_type : int16 range : [-32768, 32767] granularity: per-tensor, zero_point=0 ADD Input 0: data_type : int16 range : [-32768, 32767] granularity: per-tensor, zero_point=0 Input 1: data_type : int16 range : [-32768, 32767] granularity: per-tensor, zero_point=0 Output 0: data_type : int16 range : [-32768, 32767] granularity: per-tensor, zero_point=0 Copybara import of the project: -- b94cb47 by Elena Zhelezina <elena.zhelezina@arm.com>: Added 16-bit version of ADD/SUB operators. Broadcasting is included. -- 924d0b7 by Elena Zhelezina <elena.zhelezina@arm.com>: Addressed reviewer comments. -- dd0d9e8 by Elena Zhelezina <elena.zhelezina@arm.com>: Added versioning to ADD/SUB + some rework of the existing code. -- abae3fd by Elena Zhelezina <elena.zhelezina@arm.com>: Added versioning for ADD/SUB with new option in the schema.fbs schema_generated.h is edited manually. -- 24f3f55 by Elena Zhelezina <elena.zhelezina@arm.com>: Fix for broken build. -- d252fe1 by Elena Zhelezina <elena.zhelezina@arm.com>: Fix for the failing internal test for NN delegates. -- 2223a5c by Elena Zhelezina <elena.zhelezina@arm.com>: Fix for asan failures. Change-Id: I2cf421ddda7f9e802202239136ab062bcd63b4aa -- 3c219a4 by Elena Zhelezina <elena.zhelezina@arm.com>: Added broadcast params to addsub structure. Change-Id: I61d7d4a94087d052a782890799211031f6ed3015 -- 9131a38 by Elena Zhelezina <elena.zhelezina@arm.com>: Corrected defaults. Change-Id: I9ea50c75014cc03ac91fdef0f5b4fe11395f7074 PiperOrigin-RevId: 324865496
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.
Registered the TensorMap op kernels with GPU.