BramFactor and co-sim

[sei-jgwohlbier]

Hi, I'm trying to use BramFactor = 1 so that weights are reloadable. However, my co-sim outputs are all zero. In the documentation I see this note: "It is the responsibility of the user to ensure the weights are properly loaded during the operation of the design." This suggests that maybe co-simulation also won't work?

c-sim

Processing input 0
Predictions
-0.269481 -0.328914 0.379265 0.183073 -0.138055 -0.159306 0.554042 -0.191527 -0.270243 0.0851547 -0.739033 0.380615 -0.219884 -0.411622 -0.0200179 -0.529608 -0.205685 -0.0636143 0.44733 0.364676 -0.0438207 0.24263 0.498287 -0.0265807 -0.359685 -0.0841182 -0.346833 0.747636 -0.291152 -0.347639 0.176567 -0.393037 -0.257834 -0.206479 0.306983 0.0293598 -0.103387 0.24123 0.686169 -0.567621 -0.21184 0.0610552 -0.868607 0.63263 -0.305043 -0.39712 -0.000507704 -1.28251 -0.257254 -0.122686 0.356424 0.198907 0.0660207 0.215679 0.495663 -0.368156 -0.429782 0.329403 -0.555253 0.499191 -0.0225185 -0.301449 0.0892304 -0.634397 -0.201487 -0.29578 0.269681 0.528679 0.00368112 0.190562 0.875874 -0.24836 -0.267228 0.000161877 -0.090502 0.757431 -0.417993 -0.54548 0.181006 -0.488625 -0.0957165 -0.039659 0.383149 0.0737789 -0.127552 -0.145742 0.379787 -0.17251 -0.292812 0.0658792 -0.636039 0.428666 -0.259141 -0.359458 -0.254334 -0.708474 -0.567171 -0.0612806 0.307971 0.296532 -0.113456 0.337465 0.479268 -0.253867 -0.104861 0.231379 -0.84189 0.589407 -0.216098 -0.237486 0.0628981 -0.577348 -0.100937 0.00491157 0.29714 0.231292 -0.0583395 0.406773 0.254594 -0.274028 -0.365104 -0.200824 -0.0750172 0.264358 0.0150978 -0.0164963 -0.149388 -0.400248 
Quantized predictions
-0.269504 -0.328939 0.379245 0.18305 -0.138081 -0.159328 0.554017 -0.191551 -0.270266 0.0851316 -0.739056 0.38059 -0.219908 -0.411646 -0.0200424 -0.529629 -0.205719 -0.0636463 0.447301 0.364642 -0.0438576 0.242591 0.498249 -0.0266123 -0.359718 -0.084157 -0.346869 0.747599 -0.291189 -0.347672 0.176528 -0.393071 -0.257867 -0.206515 0.306947 0.0293245 -0.103422 0.241192 0.68613 -0.56766 -0.211876 0.0610228 -0.868643 0.632588 -0.305079 -0.397156 -0.000546455 -1.28254 -0.257289 -0.122716 0.356391 0.198872 0.0659838 0.215644 0.495624 -0.368191 -0.429813 0.329366 -0.555285 0.499153 -0.0225496 -0.301485 0.0891981 -0.634429 -0.20152 -0.295815 0.269645 0.528644 0.00364304 0.190529 0.875838 -0.248392 -0.267261 0.000123978 -0.090538 0.757395 -0.418029 -0.545516 0.180972 -0.488657 -0.0957508 -0.0396948 0.383117 0.0737429 -0.12759 -0.145774 0.379753 -0.17254 -0.292842 0.0658484 -0.636076 0.428632 -0.259176 -0.359487 -0.254368 -0.708508 -0.567205 -0.0613174 0.307939 0.296498 -0.113491 0.337429 0.479233 -0.253901 -0.10489 0.231348 -0.841926 0.589371 -0.216134 -0.237521 0.0628662 -0.577381 -0.100957 0.00489426 0.29712 0.231267 -0.0583649 0.406751 0.254573 -0.274049 -0.365127 -0.200847 -0.0750427 0.264338 0.0150757 -0.0165186 -0.149407 -0.400273 
INFO: Saved inference results to file: tb_data/rtl_cosim_results.log
INFO [HLS SIM]: The maximum depth reached by any hls::stream() instance in the design is 10
INFO: [COSIM 212-323] Starting verilog simulation...
INFO: [COSIM 212-15] Starting XSIM ...

co-sim

INFO: [Common 17-206] Exiting xsim at Mon Jul  7 11:43:17 2025...
INFO: [COSIM 212-316] Starting C post checking ...
Processing input 0
Predictions
-0.269481 -0.328914 0.379265 0.183073 -0.138055 -0.159306 0.554042 -0.191527 -0.270243 0.0851547 -0.739033 0.380615 -0.219884 -0.411622 -0.0200179 -0.529608 -0.205685 -0.0636143 0.44733 0.364676 -0.0438207 0.24263 0.498287 -0.0265807 -0.359685 -0.0841182 -0.346833 0.747636 -0.291152 -0.347639 0.176567 -0.393037 -0.257834 -0.206479 0.306983 0.0293598 -0.103387 0.24123 0.686169 -0.567621 -0.21184 0.0610552 -0.868607 0.63263 -0.305043 -0.39712 -0.000507704 -1.28251 -0.257254 -0.122686 0.356424 0.198907 0.0660207 0.215679 0.495663 -0.368156 -0.429782 0.329403 -0.555253 0.499191 -0.0225185 -0.301449 0.0892304 -0.634397 -0.201487 -0.29578 0.269681 0.528679 0.00368112 0.190562 0.875874 -0.24836 -0.267228 0.000161877 -0.090502 0.757431 -0.417993 -0.54548 0.181006 -0.488625 -0.0957165 -0.039659 0.383149 0.0737789 -0.127552 -0.145742 0.379787 -0.17251 -0.292812 0.0658792 -0.636039 0.428666 -0.259141 -0.359458 -0.254334 -0.708474 -0.567171 -0.0612806 0.307971 0.296532 -0.113456 0.337465 0.479268 -0.253867 -0.104861 0.231379 -0.84189 0.589407 -0.216098 -0.237486 0.0628981 -0.577348 -0.100937 0.00491157 0.29714 0.231292 -0.0583395 0.406773 0.254594 -0.274028 -0.365104 -0.200824 -0.0750172 0.264358 0.0150978 -0.0164963 -0.149388 -0.400248 
Quantized predictions
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 
INFO: Saved inference results to file: tb_data/rtl_cosim_results.log

[sei-jgwohlbier]

I have confirmed that using the default value of BramFactor generates the correct quantized predictions.

[vloncar]

Yes, I think we'd need to update the testbench for the co-sim to work. Or the documentation to say it doesn't work