add benchmarking script for testing FFT speeds

examples/xafs/fft_benchmark.py

@@ -0,0 +1,92 @@
+from time import time
+import numpy as np
+fft_numpy =  np.fft.fft
+
+fft_mkln = fft_mkls = fft_ws = fft_wp = fft_wn = None
+
+try:
+    import mkl_fft
+    from mkl_fft.interfaces.numpy_fft import fft as fft_mkln
+    from mkl_fft.interfaces.scipy_fft import fft as fft_mkls
+except ImportError:
+    print("mkl not available")
+
+try:
+    from pyfftw.interfaces.scipy_fft import fft as fft_ws
+    from pyfftw.interfaces.scipy_fftpack import fft as fft_wp
+    from pyfftw.interfaces.numpy_fft import fft as fft_wn
+except:
+    print("pyfftw not available")
+
+from scipy.fftpack import fft as fft_pack
+from scipy.fft import fft as fft_scipy
+
+from larch.io import read_ascii
+from larch.xafs import pre_edge, autobk, xftf, xftf_prep
+
+
+METHODS = {}
+for key, val in {'numpy': fft_numpy, 'scipy': fft_scipy,
+                 'fftpack': fft_pack,
+                 'mkl_numpy': fft_mkln, 'mkl_scipy': fft_mkls,
+                 'fftw_numpy': fft_wn, 'fftw_scipy': fft_ws,
+                 'fftw_pack': fft_wp}.items():
+
+    if callable(val):
+        METHODS[key] = val
+
+fname = '../xafsdata/feo_rt1.xdi'
+dat = read_ascii(fname, labels='energy mu i0')
+
+pre_edge(dat)
+autobk(dat, rbkg=0.9, kweight=2)
+
+xftf(dat, kmin=2, kmax=13, dk=3, window='hanning', kweight=2)
+
+# plot_chir(dat, win=2)
+
+
+xk = dat.k
+
+xchi, win = xftf_prep(dat.k, dat.chi, kmin=2, kmax=13, dk=3, dk2=3,
+                      kstep=0.05, window='hanning', kweight=2)
+
+nfft = 2048
+kstep = 0.05
+rstep = np.pi /(nfft*kstep)
+
+cchi = np.zeros(nfft, dtype='complex128')
+cchi[0:len(xchi)] = xchi*win
+r    = rstep * np.arange(nfft//2)
+
+rscale = kstep/np.sqrt(np.pi)
+results = {}
+for key in METHODS:
+    results[key] = None
+
+out = {}
+for key in METHODS:
+    out[key] = []
+
+#warm up
+for name, meth in METHODS.items():
+    for i in range(5):
+        t0 = time()
+        chir = rscale * meth(cchi)
+        results[name] = chir.real
+
+for name, meth in METHODS.items():
+    for i in range(50):
+        t0 = time()
+        for j in range(25):
+            chir = rscale * meth(cchi)
+        out[name].append(1e4*(time() -t0))
+
+
+k,m,s,x = 'Method', 'Mean', 'Std', 'Max'
+print(f"{k:10s} :  {m:8s}   {s:8s} {x:8s}")
+for k, v in out.items():
+    vn = np.array(v)
+    print(f"{k:10s} :  {vn.mean():8.3f}   {vn.std():8.3f} {vn.max():8.3f}")
+
+print("done")