Top level diagram

Future enhancements:

• Replace nXOR kernel with fix-point unsigned arithmetic dot-product
• Update convolution mechanism. It's not necessary to catch the whole window at the same time. Instead, window can be processed column-by-colum in a FIR manner. This will make implementation of convolution DSP48-friendly

Simulation & Verification

Cocotb simulation was used to generate input stimulus and assert output values

Requirements

• verilator
• cocotb
• numpy

Makefile contains all the required configurations to run simulation Run make and cocotb will run simulation via VPI interface to verilated binaries See the end of this file for example log

Future enhancements:

• Break common test to Testbench and various test-cases
• Chance input-output interface to packed-streaming
• Cover various combinations of N & K, only several are covered manually

Window Buffer 2D

There are 2 versions of window buffer mechanim presented now

• Use version 1 in case of relatively small N,K,BDW (initial case)
• use version 2 for large image processing (better for extended versions)

Future enhancements:

• Utilize BRAM for data storage
• Implement "classic" 2D convolution Kernel

Version 1. 2D Grid

Benefits

• Design has constant logic complexity of the critical path (multiplexor 2 to 1)

Drawback

• Because of the mesh (or 2D grid) structure, where each element has two inputs, this design is not suitable for SRL or BRAM, resource efficient memory storage

Version 2. SRL-Friendly

Benefits

• Feasable for SRL implementation, (N-K)*K registers can be fitted to SRL or BRAM (in the future)

Drawback

• Data-input (1 to K) demultiplexor and output multiplexors (K to 1) may affect critical path in case of large K (> 32)

Click me to see images

Window Mechanism validation

Window mechanism validation with binary values may be obscured, so to simplify window-logic validation:

• Extend bit-width parameter BWD to 8
• Use incrementing counter value to represent indexes as values

See waveform example with N_IMAGE=5, BWD=8, K_KERNEL=3

Image values

[[ 0  1  2  3  4]
 [ 5  6  7  8  9]
 [10 11 12 13 14]
 [15 16 17 18 19]
 [20 21 22 23 24]]

Click me to see Window Buffer V1 workflow

Click me to see Cocotb Run Log

(base) kvinogradov@blackbox:~/work/fpga/conv$ make
rm -f results.xml
make -f Makefile results.xml
make[1]: Entering directory '/home/kvinogradov/work/fpga/conv'
rm -f results.xml
MODULE=test_conv2d_binary TESTCASE= TOPLEVEL=conv2d_binary TOPLEVEL_LANG=verilog \
         sim_build/Vtop 
     -.--ns INFO     gpi                                ..mbed/gpi_embed.cpp:76   in set_program_name_in_venv        Did not detect Python virtual environment. Using system-wide Python interpreter
     -.--ns INFO     gpi                                ../gpi/GpiCommon.cpp:101  in gpi_print_registered_impl       VPI registered
     0.00ns INFO     cocotb                             Running on Verilator version 5.020 2024-01-01
     0.00ns INFO     cocotb                             Running tests with cocotb v1.8.1 from /home/kvinogradov/anaconda3/lib/python3.11/site-packages/cocotb
     0.00ns INFO     cocotb                             Seeding Python random module with 1706469030
     0.00ns INFO     cocotb.regression                  Found test test_conv2d_binary.basic_test
     0.00ns INFO     cocotb.regression                  running basic_test (1/1)
     0.00ns INFO     cocotb                             Image values
     0.00ns INFO     cocotb                             [[1 0 0 0 1 1 1 1]
                                                         [0 0 1 1 0 0 1 0]
                                                         [1 1 0 0 0 1 0 1]
                                                         [1 1 0 0 0 0 1 0]
                                                         [1 0 0 0 0 0 0 0]
                                                         [0 0 0 0 1 1 1 0]
                                                         [1 1 1 1 1 1 1 1]
                                                         [0 1 0 0 0 1 0 1]]
     0.00ns INFO     cocotb                             Kernel values
     0.00ns INFO     cocotb                             [[1 1 0]
                                                         [1 1 0]
                                                         [1 1 1]]
     0.00ns INFO     cocotb                             Expected dot-product
     0.00ns INFO     cocotb                             [[4. 3. 3. 4. 3. 6.]
                                                         [5. 4. 4. 2. 3. 4.]
                                                         [7. 4. 2. 1. 2. 3.]
                                                         [5. 3. 3. 4. 4. 5.]
                                                         [6. 5. 4. 5. 6. 7.]
                                                         [4. 4. 2. 4. 5. 7.]]
   765.00ns INFO     cocotb.regression                  basic_test passed
   765.00ns INFO     cocotb.regression                  ***************************************************************************************
                                                        ** TEST                           STATUS  SIM TIME (ns)  REAL TIME (s)  RATIO (ns/s) **
                                                        ***************************************************************************************
                                                        ** test_conv2d_binary.basic_test   PASS         765.00           0.01      77421.26  **
                                                        ***************************************************************************************
                                                        ** TESTS=1 PASS=1 FAIL=0 SKIP=0                 765.00           0.16       4658.56  **
                                                        ***************************************************************************************