Use scipy to generate random signal

scripts/py_benchmarks.py

@@ -2,6 +2,7 @@
 import pyfftw
 import time
 import csv
+from scipy.stats import unitary_group
 import matplotlib.pyplot as plt
 
 from pybindings import fft
@@ -17,14 +18,16 @@ def main() -> None:
         for n in range(4, 29):
             print(f"n = {n}")
             big_n = 1 << n
-            a_re = np.asarray([float(i) for i in range(big_n)])
-            a_im = np.asarray([float(i) for i in range(big_n)])
+            x = unitary_group.rvs(big_n)
+
+            a_re = x[:, 0].copy().real # np.asarray([float(i) for i in range(big_n)])
+            a_im = x[:, 0].copy().imag # np.asarray([float(i) for i in range(big_n)])
 
             start = time.time()
             fft(a_re, a_im)
             phastft_elapsed = round((time.time() - start) * 10**6)
 
-            a = [complex(i, i) for i in range(big_n)]
+            a = x[:, 0].copy()
 
             start = time.time()
             expected = np.fft.fft(a)
@@ -41,7 +44,7 @@ def main() -> None:
             )
             np.testing.assert_allclose(actual, expected)
 
-            arr = np.asarray([complex(i, i) for i in range(big_n)])
+            arr = x[:, 0].copy()
             a = pyfftw.empty_aligned(big_n, dtype="complex128")
             a[:] = arr
             start = time.time()
@@ -94,7 +97,6 @@ def plot_elapsed_times(data: dict) -> None:
     plt.ylabel("Elapsed Time (microseconds)")
     plt.yscale("log")
     plt.legend()
-
     plt.savefig("py_benchmarks.png", dpi=600)