Code

Draft PR: testing#2628

Code

Closed

ddavis-2015 wants to merge 95 commits intotensorflow:maintensorflow/tflite-micro:mainfrom ddavis-2015:bq-compressionddavis-2015/tflite-micro:bq-compressionCopy head branch name to clipboard

+12,050-1,925

Commits

Commits on Aug 14, 2024

• testing

  ddavis-2015

  committed
  82a3b15

  Copy full SHA for 82a3b15

Commits on Aug 16, 2024

• Add DEPTHWISE_CONV kernel unit test.

  Show description for 9af664d
  ddavis-2015

  committed
  9af664d

  Copy full SHA for 9af664d

• cleanup

  ddavis-2015

  committed
  cc1a4a0

  Copy full SHA for cc1a4a0

Commits on Aug 17, 2024

• Add DEPTHWISE_CONV compressed kernel unit tests.

  Show description for 13c5f71
  ddavis-2015

  committed
  13c5f71

  Copy full SHA for 13c5f71

Commits on Aug 24, 2024

• Fix XTENSA compiler warnings.

  Show description for 8dbda0f
  ddavis-2015

  committed
  8dbda0f

  Copy full SHA for 8dbda0f

Commits on Aug 25, 2024

• Add FULLY_CONNECTED compression support for XTENSA.

  Show description for 56cb57e
  ddavis-2015

  committed
  56cb57e

  Copy full SHA for 56cb57e

• fix xtensa FULLY_CONNECTED copyright

  ddavis-2015

  committed
  37bec9a

  Copy full SHA for 37bec9a

• Add DEPTHWISE_CONV compression support to XTENSA.

  Show description for 3f1d113
  ddavis-2015

  committed
  3f1d113

  Copy full SHA for 3f1d113

Commits on Aug 27, 2024

• Change DecompressToBuffer API

  ddavis-2015

  committed
  120bdbd

  Copy full SHA for 120bdbd

• Add compression support for DEPTHWISE_CONV for XTENSA P6.

  ddavis-2015

  committed
  7c406d7

  Copy full SHA for 7c406d7

• Add compression support for CONV2D for XTENSA P6.

  ddavis-2015

  committed
  f1a49ea

  Copy full SHA for f1a49ea

• Add compression support for FULLY_CONNECTED for XTENSA P6.

  ddavis-2015

  committed
  404bd9c

  Copy full SHA for 404bd9c

• fix copyright

  ddavis-2015

  committed
  7fed169

  Copy full SHA for 7fed169

• Add TFLM compression testing to XTENSA CI

  ddavis-2015

  committed
  23d2111

  Copy full SHA for 23d2111

• update micro speech test arena size

  ddavis-2015

  committed
  26e53de

  Copy full SHA for 26e53de

Commits on Aug 30, 2024

• more viewer output

  Show description for d458447
  ddavis-2015

  committed
  d458447

  Copy full SHA for d458447

• update code style check exclusions

  ddavis-2015

  committed
  41dd3a3

  Copy full SHA for 41dd3a3

Commits on Sep 4, 2024

• Add profiling to decompression code using external context.

  Show description for ff63716
  ddavis-2015

  committed
  ff63716

  Copy full SHA for ff63716

Commits on Sep 6, 2024

• viewer works with more models

  ddavis-2015

  committed
  b392b6e

  Copy full SHA for b392b6e

• compile without optimized kernel directory (reference kernels only)

  ddavis-2015

  committed
  e85d9b7

  Copy full SHA for e85d9b7

Commits on Sep 7, 2024

• Update CONCATENATION kernel to more closely match TfLite.

  ddavis-2015

  committed
  7ac50a0

  Copy full SHA for 7ac50a0

Commits on Sep 8, 2024

• Add TFLM compression support for CONCATENATION.

  Show description for e099723
  ddavis-2015

  committed
  e099723

  Copy full SHA for e099723

• Allow compression script to handle value tables of length 1.

  Show description for ab3cc79
  ddavis-2015

  committed
  ab3cc79

  Copy full SHA for ab3cc79

Commits on Sep 9, 2024

• update copyright

  ddavis-2015

  committed
  37e8ee3

  Copy full SHA for 37e8ee3

• Update compression script to support buffer sharing between tensors in different subgraphs.

  ddavis-2015

  committed
  4c6e6ad

  Copy full SHA for 4c6e6ad

• Add TFLM compression support to the ASSIGN_VARIABLE kernel

  ddavis-2015

  committed
  a072400

  Copy full SHA for a072400

Commits on Sep 11, 2024

• model viewer also shows tensor is_variable value

  ddavis-2015

  committed
  c77d48e

  Copy full SHA for c77d48e

• make profilers static.

  Show description for 307430f
  ddavis-2015

  committed
  307430f

  Copy full SHA for 307430f

• Handle \ (backslash) appearing within the metadata

  ddavis-2015

  committed
  ac2054a

  Copy full SHA for ac2054a

Commits on Sep 19, 2024

• refactor bit width 4 decompression code.

  ddavis-2015

  committed
  5f26a6b

  Copy full SHA for 5f26a6b

• Add TFLM compression README

  ddavis-2015

  committed
  be37537

  Copy full SHA for be37537

Commits on Sep 20, 2024

• TFLM compression bitwidth 2 optimization

  ddavis-2015

  committed
  14bc841

  Copy full SHA for 14bc841

Commits on Sep 21, 2024

• TFLM compression bitwidth 2 improvements

  ddavis-2015

  committed
  7cf87b5

  Copy full SHA for 7cf87b5

• cleanup and fix issues with bitwidth 2 decompression optimization

  ddavis-2015

  committed
  2bac419

  Copy full SHA for 2bac419

Commits on Sep 24, 2024

• TFLM compression bitwidth 3 optimized

  ddavis-2015

  committed
  46e8fe6

  Copy full SHA for 46e8fe6

• compression document updates

  ddavis-2015

  committed
  d51a035

  Copy full SHA for d51a035

• refactor and further optimize bitwidth 4 decompression.

  ddavis-2015

  committed
  1cf9c78

  Copy full SHA for 1cf9c78

Commits on Sep 26, 2024

• Fix name of bitwidth 4 decompression method

  ddavis-2015

  committed
  e9b4621

  Copy full SHA for e9b4621

Commits on Sep 27, 2024

• refactoring

  ddavis-2015

  committed
  8c0a01c

  Copy full SHA for 8c0a01c

Commits on Sep 28, 2024

• refactoring

  ddavis-2015

  committed
  24c1152

  Copy full SHA for 24c1152

• add comment

  ddavis-2015

  committed
  3f901a7

  Copy full SHA for 3f901a7

Commits on Sep 29, 2024

• Merge branch 'main' into bq-compression

  ddavis-2015

  committed
  445278f

  Copy full SHA for 445278f

Commits on Oct 2, 2024

• Improve compression documentation

  ddavis-2015

  committed
  d1a281e

  Copy full SHA for d1a281e

Commits on Oct 4, 2024

• add xtensa bit width 4 decompression code

  ddavis-2015

  committed
  eb85180

  Copy full SHA for eb85180

• add xtensa any bit width decompression code

  ddavis-2015

  committed
  487c17a

  Copy full SHA for 487c17a

Commits on Oct 8, 2024

• testing

  ddavis-2015

  committed
  b84853c

  Copy full SHA for b84853c

Commits on Oct 11, 2024

• refactor decompression code into reference and platform specific

  Show description for 2388549
  ddavis-2015

  committed
  2388549

  Copy full SHA for 2388549

• revert to original Cadence bit width 4 code

  ddavis-2015

  committed
  99c6e35

  Copy full SHA for 99c6e35

Commits on Oct 13, 2024

• reduce HIFI5 decompression code size

  ddavis-2015

  committed
  ad2b1c3

  Copy full SHA for ad2b1c3

Commits on Oct 14, 2024

• align compressed tensor data as per schema

  ddavis-2015

  committed
  77bb05d

  Copy full SHA for 77bb05d

• cleanup

  ddavis-2015

  committed
  9bb2b63

  Copy full SHA for 9bb2b63

Commits on Oct 17, 2024

• add decompression unit test

  ddavis-2015

  committed
  b318421

  Copy full SHA for b318421

Commits on Oct 18, 2024

• decompression unit test improvements

  ddavis-2015

  committed
  81ecf2e

  Copy full SHA for 81ecf2e

• working decompression unit test

  ddavis-2015

  committed
  efedcc2

  Copy full SHA for efedcc2

• fix C++ bitwidth 6 & 7 decompression

  ddavis-2015

  committed
  a110e41

  Copy full SHA for a110e41

Commits on Oct 19, 2024

• pre-merge empty commit

  ddavis-2015

  committed
  4894265

  Copy full SHA for 4894265

• Squashed commit of the following:

  Show description for 3d765e6
  ddavis-2015

  committed
  3d765e6

  Copy full SHA for 3d765e6

• cleanup

  ddavis-2015

  committed
  4a02b22

  Copy full SHA for 4a02b22

Commits on Oct 20, 2024

• Cleanup header file usage.

  Show description for 821dfdf
  ddavis-2015

  committed
  821dfdf

  Copy full SHA for 821dfdf

Commits on Oct 21, 2024

• fix CI code style errors

  Show description for d96b614
  ddavis-2015

  committed
  d96b614

  Copy full SHA for d96b614

• add compression build/test to bazel default test script

  ddavis-2015

  committed
  122db20

  Copy full SHA for 122db20

• use kernel optimzer level -O3 and -LNO:simd for Xtensa HIFI5

  ddavis-2015

  committed
  459569a

  Copy full SHA for 459569a

• fix code style errors.

  Show description for 2d825e3
  ddavis-2015

  committed
  2d825e3

  Copy full SHA for 2d825e3

• fix code style errors.

  ddavis-2015

  committed
  b43c16c

  Copy full SHA for b43c16c

Commits on Oct 22, 2024

• update to latest Cadence decompression code.

  ddavis-2015

  committed
  7dc34a9

  Copy full SHA for 7dc34a9

Commits on Oct 23, 2024

• header file cleanup.

  ddavis-2015

  committed
  df29a4c

  Copy full SHA for df29a4c

Commits on Oct 30, 2024

• first cut at op relocation script.

  Show description for 8c53ee3
  ddavis-2015

  committed
  8c53ee3

  Copy full SHA for 8c53ee3

• additional operator relocation check.

  ddavis-2015

  committed
  fc4b473

  Copy full SHA for fc4b473

Commits on Oct 31, 2024

• cleanup

  ddavis-2015

  committed
  ac4a0a4

  Copy full SHA for ac4a0a4

• keep generic benchmark application binary size stable regardless of whether compression is enabled

  ddavis-2015

  committed
  4dca8e7

  Copy full SHA for 4dca8e7

Commits on Nov 2, 2024

• Fix MicroProfiler bug with ClearEvents().

  Show description for 0d889e0
  ddavis-2015

  committed
  0d889e0

  Copy full SHA for 0d889e0

Commits on Nov 4, 2024

• fix arena

  ddavis-2015

  committed
  40e7530

  Copy full SHA for 40e7530

Commits on Nov 5, 2024

• remove [[maybe_unused]]

  ddavis-2015

  committed
  7776cda

  Copy full SHA for 7776cda

Commits on Nov 10, 2024

• expand model_facade

  Show description for cfd9890
  ddavis-2015

  committed
  cfd9890

  Copy full SHA for cfd9890

• single pending ops queue

  Show description for f651c88
  ddavis-2015

  committed
  f651c88

  Copy full SHA for f651c88

Commits on Nov 11, 2024

• changes to make the memory planner debug output easier to interpret

  ddavis-2015

  committed
  5e1a1c9

  Copy full SHA for 5e1a1c9

Commits on Nov 18, 2024

• Implement alternate profiler for MicroInterpreter.

  Show description for ae6a207
  ddavis-2015

  committed
  ae6a207

  Copy full SHA for ae6a207

• Fix typo.

  ddavis-2015

  committed
  fddf003

  Copy full SHA for fddf003

• Revert FakeMicroContext changes for alternate profiler.

  Show description for 2f8cead
  ddavis-2015

  committed
  2f8cead

  Copy full SHA for 2f8cead

Commits on Nov 19, 2024

• cleanup

  ddavis-2015

  committed
  83dafce

  Copy full SHA for 83dafce

Commits on Nov 20, 2024

• Add input tensor CRC to Generic Benchmark application.

  Show description for 0a49b2a
  ddavis-2015

  committed
  0a49b2a

  Copy full SHA for 0a49b2a

• Update to latest Cadence code. Int8 any bitwidth on normal quant axis updated.

  ddavis-2015

  committed
  300751d

  Copy full SHA for 300751d

• Pre-rebase to main empty commit

  ddavis-2015

  committed
  a6dc3e0

  Copy full SHA for a6dc3e0

Commits on Nov 25, 2024

• Squashed commit of the following:

  Show description for defad29
  ddavis-2015

  committed
  defad29

  Copy full SHA for defad29

Commits on Dec 2, 2024

• support for alternate decompression memory.

  ddavis-2015

  committed
  2788d32

  Copy full SHA for 2788d32

Commits on Dec 6, 2024

• initial refactor of compression unit tests

  ddavis-2015

  committed
  b9c62b9

  Copy full SHA for b9c62b9

• finished refactor of compression unit tests

  ddavis-2015

  committed
  81795e7

  Copy full SHA for 81795e7

• place TestConvFloat() back into conv_test_common.cc

  ddavis-2015

  committed
  cc1fb65

  Copy full SHA for cc1fb65

• fix concatenation unit test to match refactored unit test helper framework.

  ddavis-2015

  committed
  9353cda

  Copy full SHA for 9353cda

Commits on Dec 11, 2024

• updated compression and generic benchmark documentation

  ddavis-2015

  committed
  4d07fef

  Copy full SHA for 4d07fef

Commits on Dec 13, 2024

• Add GetOptionalTensorData (four parameter version).

  Show description for 81e548b
  ddavis-2015

  committed
  81e548b

  Copy full SHA for 81e548b

• Update transpose_conv for optional bias tensors when compression is enabled.

  ddavis-2015

  committed
  4c5079b

  Copy full SHA for 4c5079b

• Fixes to depthwise_conv for optional bias tensor when compression is enabled.

  ddavis-2015

  committed
  e788456

  Copy full SHA for e788456

• Fixed scratch size calculation for conv for HiFi targets for scenarios when input, filter and output heights are 1. (tensorflow#3004)

  Show description for 01eb927
  ddavis-2015

  committed
  01eb927

  Copy full SHA for 01eb927

Commits on Dec 14, 2024

• Fixes to CONV for optional bias tensor when compression is enabled.

  ddavis-2015

  committed
  e54866c

  Copy full SHA for e54866c