Adding linear regression model breaks flow
jackpanderson opened this issue · comments
Describe the bug
As said in the title, adding the linear regression model in the .py config settings breaks the characterization.
Version
1.2.15
To Reproduce
Adding model_name = "linear_regression" to the model config
Logs
LIB: Characterizing...
[openram.characterizer.lib/prepare_tables]: Slews: [0.00125 0.005 0.04 ]
[openram.characterizer.lib/prepare_tables]: Loads: [ 1.7225 6.89 27.56 ]
[openram.characterizer.lib/prepare_tables]: self.load_slews : [(1.7225, 0.00125), (6.89, 0.00125), (27.56, 0.00125), (1.7225, 0.005), (6.89, 0.005), (27.56, 0.005), (1.7225, 0.04), (6.89, 0.04), (27.56, 0.04)]
[openram.characterizer.lib/characterize_corners]: Characterizing corners: [('TT', 1.8, 25)]
[openram.characterizer.lib/characterize_corners]: Corner: ('TT', 1.8, 25)
[openram.characterizer.lib/characterize_corners]: Writing to /home/jande180/CARP/OpenRAM/macros/sram_32_32_sky130_1r1rw/sram_32_32_sky130_1r1rw_TT_1p8V_25C.lib
[openram.characterizer.regression_model/get_lib_values]: Characterizing SRAM using regression models.
Traceback (most recent call last):
File "/home/jande180/CARP/OpenRAM/sram_compiler.py", line 74, in
s.save()
File "/home/jande180/CARP/OpenRAM/compiler/sram.py", line 163, in save
lib(out_dir=OPTS.output_path, sram=self.s, sp_file=sp_file)
File "/home/jande180/CARP/OpenRAM/compiler/characterizer/lib.py", line 46, in init
self.characterize_corners()
File "/home/jande180/CARP/OpenRAM/compiler/characterizer/lib.py", line 163, in characterize_corners
self.characterize()
File "/home/jande180/CARP/OpenRAM/compiler/characterizer/lib.py", line 175, in characterize
self.compute_delay()
File "/home/jande180/CARP/OpenRAM/compiler/characterizer/lib.py", line 646, in compute_delay
char_results = m.get_lib_values(self.load_slews)
File "/home/jande180/CARP/OpenRAM/compiler/characterizer/regression_model.py", line 72, in get_lib_values
models = self.train_models()
File "/home/jande180/CARP/OpenRAM/compiler/characterizer/regression_model.py", line 138, in train_models
self.output_names = get_data_names(data_path)[self.num_inputs:]
File "/home/jande180/CARP/OpenRAM/compiler/characterizer/analytical_util.py", line 22, in get_data_names
with open(file_name, newline='') as csvfile:
FileNotFoundError: [Errno 2] No such file or directory: '/home/jande180/CARP/OpenRAM/technology/sky130/sim_data/sim_data.csv'
make: *** [Makefile:82: sram_32_32_sky130_1r1rw.ok] Error 1
This mode requires you to precharacterize a bunch of memories in that file so that it can use regression to predict models for new memories. We haven't done that for Skywater yet and instead recommend regular characterization. This is from the following MS thesis:
https://www.proquest.com/openview/273a1089a88606160de418f5b75f29e7/1.pdf?pq-origsite=gscholar&cbl=18750&diss=y
The error should definitely be improved though.