Code

MLPerf Tiny developments#503

Code

Open

jmduarte wants to merge 117 commits intofastmachinelearning:mainfastmachinelearning/hls4ml:mainfrom hls4ml-finn-mlperftiny:fifo_depth_opthls4ml-finn-mlperftiny/hls4ml:fifo_depth_optCopy head branch name to clipboard

+3,165-39

Commits

Commits on Mar 8, 2021

• Added QConv2DBatchnorm support.

  nghielme

  committed
  89c3d85

  Copy full SHA for 89c3d85

• Added QConv2DBatchnorm support.

  nghielme

  committed
  0ae678d

  Copy full SHA for 0ae678d

• Added support for QConv2DBatchnorm.

  nghielme

  committed
  0846766

  Copy full SHA for 0846766

Commits on Mar 9, 2021

• Fixed problems emerged from PR discussion.

  nghielme

  committed
  dade7de

  Copy full SHA for dade7de

• Fixed problems emerged from PR discussion.

  nghielme

  committed
  ae5afbf

  Copy full SHA for ae5afbf

• Fixed problems emerged from PR discussion.

  nghielme

  committed
  02f92e2

  Copy full SHA for 02f92e2

• Fixed problems emerged from PR discussion.

  nghielme

  committed
  b829039

  Copy full SHA for b829039

Commits on Mar 10, 2021

• Some improvements + testing if weights are provided already folded.

  nghielme

  committed
  20d9a62

  Copy full SHA for 20d9a62

Commits on Mar 12, 2021

• Added support for QConv2DBatchnorm with use_bias=False.

  nghielme

  committed
  6e6afab

  Copy full SHA for 6e6afab

Commits on Mar 19, 2021

• Removing the check on batch size. "" -> ''.

  nghielme

  committed
  7649a80

  Copy full SHA for 7649a80

Commits on Mar 23, 2021

• Last updates before merging PR.

  nghielme

  committed
  a1d86b8

  Copy full SHA for a1d86b8

Commits on Jun 24, 2021

• Merge branch 'fastmachinelearning:master' into master

  nghielme

  authored
  5e988c2

  Copy full SHA for 5e988c2

Commits on Aug 6, 2021

• HLS implementation of Concatenate for io_stream

  vloncar

  committed
  aa246b8

  Copy full SHA for aa246b8

• Fix concatenate3d_2 for io_parallel

  vloncar

  committed
  bee6c05

  Copy full SHA for bee6c05

Commits on Aug 9, 2021

• Merge branch 'concat_io_stream' of https://github.com/vloncar/hls4ml into vivado-accel-concat

  nghielme

  committed
  a0b4a00

  Copy full SHA for a0b4a00

• Merge remote-tracking branch 'origin/pynq-pr' into vivado-accel-concat

  nghielme

  committed
  33d1e91

  Copy full SHA for 33d1e91

Commits on Aug 10, 2021

• Support overriding stream depth in config

  vloncar

  committed
  810e2d5

  Copy full SHA for 810e2d5

Commits on Sep 2, 2021

• Added a first version of build_prj.tcl that is able to generate the .vcd file. Next step -> add the parser, find the max for each fifo and set the new fifo depth

  nghielme

  committed
  59e3766

  Copy full SHA for 59e3766

• Merge remote-tracking branch 'vlad/pynq_enet' into fifo_depth_opt

  nghielme

  committed
  b72260c

  Copy full SHA for b72260c

• Fixed the assignment of the unassigned inp variable

  nghielme

  committed
  306e381

  Copy full SHA for 306e381

Commits on Sep 24, 2021

• Bump version

  thesps

  committed
  da2c46a

  Copy full SHA for da2c46a

Commits on Sep 27, 2021

• First version of the optimization flow automated. Tested with ENet model.

  nghielme

  committed
  707dba5

  Copy full SHA for 707dba5

• Fixing minor issues

  nghielme

  committed
  ee4733b

  Copy full SHA for ee4733b

• Fixing minor issues

  nghielme

  committed
  93cca6c

  Copy full SHA for 93cca6c

Commits on Sep 28, 2021

• Fixing minor issues

  nghielme

  committed
  14fa32d

  Copy full SHA for 14fa32d

Commits on Sep 30, 2021

• fix conv1d io_parallel resource (fastmachinelearning#403)

  Show description for fe8519f
  

  jmitrevs

  and

  thesps

  committed
  fe8519f

  Copy full SHA for fe8519f

Commits on Oct 1, 2021

• Fixing minor issues

  nghielme

  committed
  038c4b6

  Copy full SHA for 038c4b6

• Handling fifos not implemented in brams

  nghielme

  committed
  0f915bf

  Copy full SHA for 0f915bf

• Modified name to cppname for InplaceVariable

  nghielme

  committed
  c76ea08

  Copy full SHA for c76ea08

• Added exit in build_prj.tcl file

  nghielme

  committed
  4614bfe

  Copy full SHA for 4614bfe

• name -> cppname

  nghielme

  committed
  c136ff8

  Copy full SHA for c136ff8

• Small fix

  nghielme

  committed
  3d48d0d

  Copy full SHA for 3d48d0d

Commits on Oct 7, 2021

• Parallelise pytests in CI

  thesps

  committed
  fd8a618

  Copy full SHA for fd8a618

• Fix GlobalPooling1D Layers (fastmachinelearning#399)

  Show description for f5acaca
  

  3 peoplecommitted
  f5acaca

  Copy full SHA for f5acaca

• Set appropriate data type for quantized_relu activations

  thesps

  committed
  231fdcb

  Copy full SHA for 231fdcb

• Display unsigned types properly in profiling

  thesps

  committed
  b361265

  Copy full SHA for b361265

Commits on Oct 18, 2021

• Some fixes

  nghielme

  committed
  c0143a1

  Copy full SHA for c0143a1

• Some fixes

  nghielme

  committed
  0ea8d88

  Copy full SHA for 0ea8d88

Commits on Oct 19, 2021

• Adding one dependency

  nghielme

  committed
  7ed6fcc

  Copy full SHA for 7ed6fcc

Commits on Oct 20, 2021

• Fixing bugs in the optimize_fifos_depth function

  nghielme

  committed
  92bf513

  Copy full SHA for 92bf513

Commits on Oct 21, 2021

• Fixing bugs in optimize_fifos_depth function

  nghielme

  committed
  888713b

  Copy full SHA for 888713b

Commits on Nov 8, 2021

• Testing without log_wave -r /

  nghielme

  committed
  ffc9845

  Copy full SHA for ffc9845

• - Testing with all build parameters true by default

  Show description for 0cfb79c
  nghielme

  committed
  0cfb79c

  Copy full SHA for 0cfb79c

• - Testing with all build parameters true by default

  Show description for be7dd00
  nghielme

  committed
  be7dd00

  Copy full SHA for be7dd00

Commits on Nov 11, 2021

• fix batched multiple inputs

  

  jmduarte

  authored and

  thesps

  committed
  f10e0c3

  Copy full SHA for f10e0c3

• Fixed 'qkeras_mnist_dense' example build problem fastmachinelearning#423

  

  siorpaes

  authored and

  thesps

  committed
  d1d5eba

  Copy full SHA for d1d5eba

• Update for pyyaml 6.0 (fastmachinelearning#435)

  Show description for 997c07b
  thesps

  committed
  997c07b

  Copy full SHA for 997c07b

• axi_stream_driver update (fastmachinelearning#420)

  Show description for d0ff8ba
  

  nghielme

  authored and

  thesps

  committed
  d0ff8ba

  Copy full SHA for d0ff8ba

• Reshape fixes: don't repack stream for flatten; remove final reshape (fastmachinelearning#443)

  Show description for ba3902f
  

  jmduarte

  and

  thesps

  committed
  ba3902f

  Copy full SHA for ba3902f

• Reorder loops in im2col_2d_cl given resource strategy issue. Reenable relevant test. Use 5000 MNIST samples rather than full dataset for faster testing

  thesps

  committed
  86e4397

  Copy full SHA for 86e4397

• Support applying Softmax over multidimensional tensors (fastmachinelearning#384)

  Show description for 36cd38c
  

  2 peopleauthored and

  thesps

  committed
  36cd38c

  Copy full SHA for 36cd38c

• Disable some unsupported layers

  thesps

  committed
  9f5c249

  Copy full SHA for 9f5c249

Commits on Nov 26, 2021

• Start integrating Vivado Accelerator AXI-m backend

  

  Giuseppe Di Guglielmo

  committed
  ab076fc

  Copy full SHA for ab076fc

• build_prj.tcl fixed for vivadoaccelerator - there will be problems for vivado backend

  nghielme

  committed
  fa4bd2b

  Copy full SHA for fa4bd2b

• update example-models

  nghielme

  committed
  a1384cd

  Copy full SHA for a1384cd

Commits on Nov 28, 2021

• Setup AXI-m backend for PYNQ-Z1 and PYNQ-Z2

  

  Giuseppe Di Guglielmo

  committed
  8f64675

  Copy full SHA for 8f64675

• Add C drivers to PYNQ-Z1

  

  Giuseppe Di Guglielmo

  committed
  1a64d4b

  Copy full SHA for 1a64d4b

Commits on Nov 29, 2021

• Move data.h generator into hls4ml

  

  Giuseppe Di Guglielmo

  committed
  c1f4670

  Copy full SHA for c1f4670

Commits on Dec 1, 2021

• Merge remote-tracking branch 'hls4ml/master' into fifo_depth_opt

  Show description for ca26680
  nghielme

  committed
  ca26680

  Copy full SHA for ca26680

Commits on Dec 2, 2021

• Copy directory even when the directory exists

  

  Giuseppe Di Guglielmo

  committed
  aefe0c9

  Copy full SHA for aefe0c9

Commits on Dec 11, 2021

• String compare with '==' rather than 'is'

  

  Giuseppe Di Guglielmo

  committed
  77ff677

  Copy full SHA for 77ff677

Commits on Jan 3, 2022

• Add validation besides verification to the software application (Pynq-Z1)

  

  Giuseppe Di Guglielmo

  committed
  a9aeb33

  Copy full SHA for a9aeb33

Commits on Jan 5, 2022

• Set stack/heap size in the linker script

  

  Giuseppe Di Guglielmo

  committed
  d146994

  Copy full SHA for d146994

• ATTENTION: force 1 the size of the scratchpads in the AXI-M wrapper (this may break co-simulation, but uses far less resources)

  

  Giuseppe Di Guglielmo

  committed
  08ee073

  Copy full SHA for 08ee073

Commits on Jan 7, 2022

• Set software for ultra96-v2

  

  Giuseppe Di Guglielmo

  committed
  cd27c4c

  Copy full SHA for cd27c4c

Commits on Jan 11, 2022

• init commit of QDenseBN

  julesmuhizi

  committed
  feef1f9

  Copy full SHA for feef1f9

Commits on Jan 16, 2022

• Added changes needed to carry out the relu_merge optimization

  anmeza

  committed
  97b14ea

  Copy full SHA for 97b14ea

Commits on Jan 18, 2022

• clean up code and currently testing if checking the merged relu flag in the dense_resource() functions has worse resource reductions than the previous code duplication version

  oliviaweng

  committed
  c499f47

  Copy full SHA for c499f47

Commits on Jan 19, 2022

• remove comments and extra brace

  oliviaweng

  committed
  04a382e

  Copy full SHA for 04a382e

• revert back to verbose template because it achieves maximum resource savings

  oliviaweng

  committed
  7164673

  Copy full SHA for 7164673

Commits on Jan 24, 2022

• Merge pull request #1 from anmeza/relu_fuse

  Show description for b0aa186
  jmduarte

  authored
  b0aa186

  Copy full SHA for b0aa186

Commits on Jan 25, 2022

• Merge branch 'fifo_depth_opt' into gdg/axi-m

  jmduarte

  authored
  c513c91

  Copy full SHA for c513c91

• add back cppname

  jmduarte

  committed
  1ee5708

  Copy full SHA for 1ee5708

• added weight transponse and qkeras quantization support

  julesmuhizi

  committed
  4c7321a

  Copy full SHA for 4c7321a

• patch-2 - added weight transponse and qkeras quantization support

  julesmuhizi

  committed
  afd8249

  Copy full SHA for afd8249

• add c drivers for pynq-z2

  jmduarte

  committed
  023321d

  Copy full SHA for 023321d

Commits on Jan 26, 2022

• remove uncessary _add_supported_quantized_objects call

  julesmuhizi

  committed
  2e069ae

  Copy full SHA for 2e069ae

Commits on Feb 9, 2022

• Add scripts and application for ARTY-A7 with VivadoAccelerator backend

  

  Giuseppe Di Guglielmo

  committed
  15eb287

  Copy full SHA for 15eb287

• Cleanup timers code for ARTY-A7

  

  Giuseppe Di Guglielmo

  committed
  cb7fc44

  Copy full SHA for cb7fc44

• Merge pull request #5 from GiuseppeDiGuglielmo/gdg/axi-m

  Show description for d1dab8d
  jmduarte

  authored
  d1dab8d

  Copy full SHA for d1dab8d

Commits on Feb 14, 2022

• Merge pull request #2 from hls4ml-finn-mlperftiny/gdg/axi-m

  Show description for fd4df8e
  jmduarte

  authored
  fd4df8e

  Copy full SHA for fd4df8e

Commits on Feb 16, 2022

• Integration of features for EEMBC power analysis.

  Show description for 361628d
  

  Giuseppe Di Guglielmo

  committed
  361628d

  Copy full SHA for 361628d

Commits on Feb 18, 2022

• Merge pull request #7 from hls4ml-finn-mlperftiny/fifo_depth_opt

  Show description for 17fc6f8
  julesmuhizi

  authored
  17fc6f8

  Copy full SHA for 17fc6f8

Commits on Feb 21, 2022

• try out jules's fix to relu_merge optimizer pass to work with dense-relu layers

  oliviaweng

  committed
  a497085

  Copy full SHA for a497085

• have dense layer check merged_relu flag to see whether or not to set merged_relu flag in HLS firmware code

  oliviaweng

  committed
  c9daa06

  Copy full SHA for c9daa06

• fix how merged_relu flag is set for Dense and Conv2D in hls_layers.py

  oliviaweng

  committed
  949bafd

  Copy full SHA for 949bafd

• debug print

  oliviaweng

  committed
  8e3df9e

  Copy full SHA for 8e3df9e

• add merged_relu attribute to Layer class

  oliviaweng

  committed
  f298517

  Copy full SHA for f298517

• remove debug prints and comments and DenseBatchnorm from relu_merge supported layers list since it's not supported yet

  oliviaweng

  committed
  f6bf06a

  Copy full SHA for f6bf06a

Commits on Feb 22, 2022

• Delete relu_merge_old.py

  jmduarte

  authored
  9c08a24

  Copy full SHA for 9c08a24

• Update core.py

  jmduarte

  authored
  77a2db1

  Copy full SHA for 77a2db1

• Update core.py

  jmduarte

  authored
  8b494a2

  Copy full SHA for 8b494a2

• apply patches

  jmduarte

  committed
  a18a204

  Copy full SHA for a18a204

• remove unnecessary patches

  jmduarte

  committed
  3590aa4

  Copy full SHA for 3590aa4

• Merge pull request #8 from anmeza/relu_fuse

  Show description for 4173797
  jmduarte

  authored
  4173797

  Copy full SHA for 4173797

• Merge pull request #10 from hls4ml-finn-mlperftiny/div_by_256_patch

  Show description for b2f499c
  jmduarte

  authored
  b2f499c

  Copy full SHA for b2f499c

• Merge pull request #4 from hls4ml-finn-mlperftiny/qdense_batchnorm

  Show description for ad311fc
  jmduarte

  authored
  ad311fc

  Copy full SHA for ad311fc

• small fixes

  jmduarte

  committed
  532aff8

  Copy full SHA for 532aff8

• Merge pull request #6 from hls4ml-finn-mlperftiny/fifo_depth_opt_dev

  Show description for 541acac
  jmduarte

  authored
  541acac

  Copy full SHA for 541acac

• just pass hls_model

  jmduarte

  committed
  e48ae63

  Copy full SHA for e48ae63

• no need to pass keras model

  jmduarte

  committed
  c18064c

  Copy full SHA for c18064c

Commits on Feb 24, 2022

• correct scope for arty

  jmduarte

  committed
  d8b435d

  Copy full SHA for d8b435d

Commits on Feb 28, 2022

• weird hack for now

  jmduarte

  committed
  8c51558

  Copy full SHA for 8c51558

Commits on Mar 1, 2022

• remove prints

  jmduarte

  authored
  5f3b468

  Copy full SHA for 5f3b468

• Merge pull request #11 from hls4ml-finn-mlperftiny/arty_fifo

  Show description for 6c0d205
  jmduarte

  authored
  6c0d205

  Copy full SHA for 6c0d205

Commits on Mar 2, 2022

• Support overriding stream depth in config

  

  vloncar

  authored and

  jmduarte

  committed
  e808a05

  Copy full SHA for e808a05

• bug fixes

  jmduarte

  committed
  e58a986

  Copy full SHA for e58a986

• get correct type for data header

  jmduarte

  committed
  6bbf88c

  Copy full SHA for 6bbf88c

• smaller MCU

  jmduarte

  committed
  7f209e2

  Copy full SHA for 7f209e2

• hardcode name which works for current models 🤷

  jmduarte

  committed
  c424ac1

  Copy full SHA for c424ac1

• generalize with regex

  jmduarte

  committed
  6e6730d

  Copy full SHA for 6e6730d

Commits on Mar 3, 2022

• make eembc_power configurable

  jmduarte

  committed
  fb3def1

  Copy full SHA for fb3def1

Commits on Mar 4, 2022

• Use SystemVerilog define to control EEMBC setup

  

  Giuseppe Di Guglielmo

  committed
  ed3ffc7

  Copy full SHA for ed3ffc7

• Update axi_master_design.tcl

  Show description for 7452d5a
  GiuseppeDiGuglielmo

  authored
  7452d5a

  Copy full SHA for 7452d5a

• Update design_1_wrapper.v

  GiuseppeDiGuglielmo

  authored
  f4613f0

  Copy full SHA for f4613f0

Commits on Mar 9, 2022

• Fix but w/ clock connection that stopped the QSPI boot

  

  Giuseppe Di Guglielmo

  committed
  9a411f0

  Copy full SHA for 9a411f0

• Fix bug w/ clock connection that stopped the QSPI boot

  

  Giuseppe Di Guglielmo

  committed
  5c0ad77

  Copy full SHA for 5c0ad77