Release merge for v0.10.1 #1123

auphelia · 2024-07-03T15:24:29Z

This release contains

Fix for FIFO sizing, see issue StreamingFIFO count width mismatch between Python and RTL #998
Independent input and threshold precision and resource efficient per tensor quantization with RTL Thresholding
Select specific DSP48 (version 1 or version 2) depending on target device for RTL MVU
Extend resource estimation to take fpga part as argument and improved post synthesis resource reporting (Enable analysis report of post synthesis resource utilization in json #965)
Fix for data packing in rtlsim (Fix fast data packing for rtlsim #1080)
cpp helper functions for numpy to hls::vector conversion (Helper function for reading numpy files into hls::vector #1047)
Smaller bugfixes, like calling the correct .onnx file in Brevitas export tutorial (Update 1_brevitas_network_import_via_QONNX.ipynb #1053)

…ore default transformation steps tested

… input shape

Note: This was already implemented, it was just not called from the step_synthesize_bitfile.

According to the last synthesis I ran using Vitis/Vivado 2022.2, the DSP48 utilization seems to be reported in row 10 of the XML table.

Signed-off-by: aziz bahri <[email protected]>

…test Extension to end2end mobilenet test

[Threshold RTL] padd threshold steps based on activation bitwidth

…and count in FIFO template

Signed-off-by: Jonas Kühle <[email protected]>

Bug fix for FIFO count range mismatch between Python and RTL

Helper function for reading numpy files into hls::vector

Update xrt download command in Dockerfile

… params

Enable optional xrt installation from local deb

Fix fast data packing for rtlsim

Install setupstools explicitly in Dockerfile

Expose DSP variant for RTL MVU

[Deps] Update finn-experimental commit hash

…idth_converter_hls-duplicate-stream

…erter_hls-duplicate-stream Fix redeclaration of streams in StreamingDataWidthConverter_hls

Optimizing resources for per tensor quantization

mmrahorovic and others added 30 commits March 29, 2024 12:53

[test]: extension to end2end mobilenet test; use RTL layers and few m…

2d0e174

…ore default transformation steps tested

[test]: extended with testing functional correctnes models

25f75b5

[test]: disabled checking probabilities (fails test) and minor fix to…

a99619a

… input shape

[Data packing] Disable unwanted sign extension in fast mode

1a11858

Enable analysis report of post synthesis resource in json

de1f024

Note: This was already implemented, it was just not called from the step_synthesize_bitfile.

Fix table row of the DSP48 post synthesis resource utilization report

b435655

According to the last synthesis I ran using Vitis/Vivado 2022.2, the DSP48 utilization seems to be reported in row 10 of the XML table.

[Analysis] try to infer restype-to-index dynamically from report

79a9015

Address linting issues running pre-commit

5f1f80a

DSP48 should simply be reported as DSP

f2aa440

[Tests] Adjust liveness thresholds in mobilenet end2end rtlsim tests

b602500

[Tests] Cleanup end2end mobilenet test

0ad7cfb

[Tests] Fix typo and save model in mobilenet ipgen test

66902e4

[Tests] Remove copy ip to separate directory

12da374

[Tests] Removing double ip generation

201867a

[Threshold RTL] padd threshold steps based on activation bitwidth

4132c75

Signed-off-by: aziz bahri <[email protected]>

[BNN Test] remove comment

b02f72e

Signed-off-by: aziz bahri <[email protected]>

Merge pull request #1022 from mmrahorovic/refactor/end2end_mobilenet_…

2fffa76

…test Extension to end2end mobilenet test

Merge pull request #1035 from Xilinx/bugfix/threshold_weight_padding

e188b4c

[Threshold RTL] padd threshold steps based on activation bitwidth

[FIFO] Add additional count width parameter to set range of maxcount …

1b63517

…and count in FIFO template

[FIFO] Adjust count width and make param local in verilog files

3bf03b6

[FIFO] Delete obsolete comma in template code

50c4e6f

added helper function for reading from numpy files into hls::vector

1acf207

Signed-off-by: Jonas Kühle <[email protected]>

disabled debug output

e5f603a

Signed-off-by: Jonas Kühle <[email protected]>

updated copyright header

7635460

Signed-off-by: Jonas Kühle <[email protected]>

[Deps] Update finn-experimental commit hash

fe6852a

Merge remote-tracking branch 'upstream/dev' into feature/vector

7f38048

Merge pull request #1044 from Xilinx/bugfix/FIFO_countrange

6f193cb

Bug fix for FIFO count range mismatch between Python and RTL

[Test] Simplify access to last element of shape

b1ecd4e

[npy2vectorstream] Clean up file

1dd18ec

Merge pull request #1047 from vopade/feature/vector

10bb1ba

Helper function for reading numpy files into hls::vector

auphelia and others added 29 commits May 31, 2024 15:38

[GHA] Add debug flag for GHA

8d6543c

[Deps] Introduce env var to skip xrt download

356528f

Merge pull request #1092 from Xilinx/hotfix/quicktest_gha

507d72e

Update xrt download command in Dockerfile

Merge branch 'dev' into bigfix/mvu4bit

bc5ce2d

[RTL Thresh] Enable narrow and per tensor mode with runtime writeable…

79c46bb

… params

[Tests] Change target board for subset of mvau tests

304337b

[Docker] Enable optional xrt installation from local deb

1c46131

[GHA] Add path for skipping xrt download

3f428a0

[Docker] Workaround to allow for optional COPY command

048557f

Merge pull request #1098 from Xilinx/fix/xrt_download

9f0fad5

Enable optional xrt installation from local deb

Merge branch 'dev' into bigfix/mvu4bit

599cf3a

Merge pull request #1080 from fpjentzsch/fix/packing_mask

4a2e429

Fix fast data packing for rtlsim

[Tests] Add res type and depth triggers

abb6daf

[Deps] Move setuptools installation to Dockerfile

b9894c0

Merge pull request #1102 from Xilinx/fix/setuptools_version

3d597ac

Install setupstools explicitly in Dockerfile

[Tests] Fix bnn pynq to use default hw variants

aacdaee

Merge branch 'dev' into bigfix/mvu4bit

10704f7

Merge pull request #1091 from Xilinx/bigfix/mvu4bit

a4bbb59

Expose DSP variant for RTL MVU

[Deps] Update finn-experimental commit hash

0ef0ca4

Merge pull request #1104 from Xilinx/update/finn_experimental

34f73d5

[Deps] Update finn-experimental commit hash

[RTL thresh] Duplicate per tensor thresholds per PE

29e2256

[StreamingDataWidthConverter_hls] Remove duplicate intermediate stream

a325a5f

[StreamingDataWidthConverter_hls] Remove the other duplicate stream

7b80e1a

Merge remote-tracking branch 'upstream/dev' into fix/streaming_data_w…

6fd8b1a

…idth_converter_hls-duplicate-stream

[Tests] Update dwc testing to test rtl and hls variant

961e48b

Merge pull request #1107 from iksnagreb/fix/streaming_data_width_conv…

25c1c99

…erter_hls-duplicate-stream Fix redeclaration of streams in StreamingDataWidthConverter_hls

Merge branch 'dev' into feature/per_tensor_rtl_thresh

bc8dc04

Merge pull request #1110 from Xilinx/feature/per_tensor_rtl_thresh

9cd8bab

Optimizing resources for per tensor quantization

[AUTHORS] Update list

f649cda

auphelia merged commit 9d29968 into main Jul 8, 2024
4 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Release merge for v0.10.1 #1123

Release merge for v0.10.1 #1123

auphelia commented Jul 3, 2024 •

edited

Loading

Release merge for v0.10.1 #1123

Release merge for v0.10.1 #1123

Conversation

auphelia commented Jul 3, 2024 • edited Loading

auphelia commented Jul 3, 2024 •

edited

Loading