import os
import numpy as np
np.random.seed(23)
def gen_golden_data(param):
src_type = param.src_type
dst_type = param.dst_type
flag = param.flag
row = param.row
col = param.col
valid_row = param.valid_row
valid_col = param.valid_col
value_max = 100
value_min = -100
input_arr = np.random.uniform(low=value_min, high=value_max, size=(row, col)).astype(src_type)
temp_arr = input_arr.astype(dst_type)
scale_arr = np.random.uniform(low=value_min, high=value_max, size=(row, 1)).astype(dst_type)
offset_arr = np.random.uniform(low=value_min, high=value_max, size=(row, 1)).astype(dst_type)
output_arr = np.zeros((row, col), dtype=dst_type)
if flag:
for i in range(valid_row):
offset_arr[i, :] = 0
for i in range(valid_row):
for j in range(valid_col):
output_arr[i, j] = (temp_arr[i, j] - offset_arr[i, 0]) * scale_arr[i, 0]
input_arr.tofile('input.bin')
scale_arr.tofile('scale.bin')
offset_arr.tofile('offset.bin')
output_arr.tofile('golden.bin')
class TDequantParams:
def __init__(self, name, src_type, dst_type, flag, row, col, valid_row, valid_col):
self.name = name
self.src_type = src_type
self.dst_type = dst_type
self.flag = flag
self.row = row
self.col = col
self.valid_row = valid_row
self.valid_col = valid_col
if __name__ == "__main__":
case_params_list = [
TDequantParams("TDEQUANTTest.case1", np.int16, np.float32, True, 64, 64, 64, 64),
TDequantParams("TDEQUANTTest.case2", np.int8, np.float32, True, 64, 64, 64, 64)
]
for _, case in enumerate(case_params_list):
if not os.path.exists(case.name):
os.makedirs(case.name)
original_dir = os.getcwd()
os.chdir(case.name)
gen_golden_data(case)
os.chdir(original_dir)