2022-11-20 16:52:33 -05:00

51 lines
1.7 KiB
Python

import warnings
from .basic_utils import basicOperator, deep_copy_dicts, overwrite_dicts
default_params = {
# op related
"op": "MAT_MUL",
"input_idx": None,
"input2_idx": None,
"output_idx": None,
# tensor related
"matA_row": None,
"matA_col": None,
"matB_row": None,
"matB_col": None,
"weight_value": None,
"weight_name": None,
"input_dtype": "int8",
"input2_dtype": "int8",
"output_dtype": "int8",
}
class mat_mul(basicOperator):
def __init__(self, params: dict) -> None:
self.params = deep_copy_dicts(default_params)
overwrite_dicts(self.params, params)
super().__init__()
# handle input/output tensors in HWC format
self._add_input(
self.params["input_idx"], self.params["input_dtype"], self.params["matA_row"], self.params["matA_col"], 1
)
self._add_output(
self.params["output_idx"], self.params["output_dtype"], self.params["matA_row"], self.params["matB_col"], 1
)
if None in default_params:
warnings.warn(f"parameters are not all set for op {self.params['op']}")
def generate_inference_str(self):
params = self.params
if params["input_dtype"] == "float32":
string = f"mat_mul({self._getBufferstrCast(params['input_buf_add'], params['input_buf_add_offset'])},"
string += f"{str(params['matA_row'])},{str(params['matA_col'])},"
string += f"weight_fp{params['parsed_trainable']},{str(params['matB_col'])},"
string += f"{self._getBufferstrCast(params['output_buf_add'], params['output_buf_add_offset'])});\n"
else:
raise NotImplementedError
return string