BENCH: sync codspeed-benchmarks with BLAS-benchmarks

benchmark/pybench/benchmarks/bench_blas.py

@@ -1,66 +1,76 @@
 import pytest
 import numpy as np
-from openblas_wrap import (
-    # level 1
-    dnrm2, ddot, daxpy,
-    # level 3
-    dgemm, dsyrk,
-    # lapack
-    dgesv,                   # linalg.solve
-    dgesdd, dgesdd_lwork,    # linalg.svd
-    dsyev, dsyev_lwork,      # linalg.eigh
-)
+import openblas_wrap as ow
+
+dtype_map = {
+    's': np.float32,
+    'd': np.float64,
+    'c': np.complex64,
+    'z': np.complex128,
+    'dz': np.complex128,
+}
+
 
 # ### BLAS level 1 ###
 
 # dnrm2
 
-dnrm2_sizes = [100, 1000]
+dnrm2_sizes = [100, 200, 400, 600, 800, 1000]
 
-def run_dnrm2(n, x, incx):
-    res = dnrm2(x, n, incx=incx)
+def run_dnrm2(n, x, incx, func):
+    res = func(x, n, incx=incx)
     return res
 
 
+@pytest.mark.parametrize('variant', ['d', 'dz'])
 @pytest.mark.parametrize('n', dnrm2_sizes)
-def test_nrm2(benchmark, n):
+def test_nrm2(benchmark, n, variant):
     rndm = np.random.RandomState(1234)
-    x = np.array(rndm.uniform(size=(n,)), dtype=float)
-    result = benchmark(run_dnrm2, n, x, 1)
+    dtyp = dtype_map[variant]
+
+    x = np.array(rndm.uniform(size=(n,)), dtype=dtyp)
+    nrm2 = ow.get_func('nrm2', variant)
+    result = benchmark(run_dnrm2, n, x, 1, nrm2)
 
 
 # ddot
 
-ddot_sizes = [100, 1000]
+ddot_sizes = [100, 200, 400, 600, 800, 1000]
 
-def run_ddot(x, y,):
-    res = ddot(x, y)
+def run_ddot(x, y, func):
+    res = func(x, y)
     return res
 
 
 @pytest.mark.parametrize('n', ddot_sizes)
 def test_dot(benchmark, n):
     rndm = np.random.RandomState(1234)
+
     x = np.array(rndm.uniform(size=(n,)), dtype=float)
     y = np.array(rndm.uniform(size=(n,)), dtype=float)
-    result = benchmark(run_ddot, x, y)
+    dot = ow.get_func('dot', 'd')
+    result = benchmark(run_ddot, x, y, dot)
 
 
 # daxpy
 
-daxpy_sizes = [100, 1000]
+daxpy_sizes = [100, 200, 400, 600, 800, 1000]
 
-def run_daxpy(x, y,):
-    res = daxpy(x, y, a=2.0)
+def run_daxpy(x, y, func):
+    res = func(x, y, a=2.0)
     return res
 
 
+@pytest.mark.parametrize('variant', ['s', 'd', 'c', 'z'])
 @pytest.mark.parametrize('n', daxpy_sizes)
-def test_daxpy(benchmark, n):
+def test_daxpy(benchmark, n, variant):
     rndm = np.random.RandomState(1234)
-    x = np.array(rndm.uniform(size=(n,)), dtype=float)
-    y = np.array(rndm.uniform(size=(n,)), dtype=float)
-    result = benchmark(run_daxpy, x, y)
+    dtyp = dtype_map[variant]
+
+    x = np.array(rndm.uniform(size=(n,)), dtype=dtyp)
+    y = np.array(rndm.uniform(size=(n,)), dtype=dtyp)
+    axpy = ow.get_func('axpy', variant)
+    result = benchmark(run_daxpy, x, y, axpy)
 
 
 
@@ -69,40 +79,46 @@ def test_daxpy(benchmark, n):
 
 # dgemm
 
-gemm_sizes = [100, 1000]
+gemm_sizes = [100, 200, 400, 600, 800, 1000]
 
-def run_gemm(a, b, c):
+def run_gemm(a, b, c, func):
     alpha = 1.0
-    res = dgemm(alpha, a, b, c=c, overwrite_c=True)
+    res = func(alpha, a, b, c=c, overwrite_c=True)
     return res
 
 
+@pytest.mark.parametrize('variant', ['s', 'd', 'c', 'z'])
 @pytest.mark.parametrize('n', gemm_sizes)
-def test_gemm(benchmark, n):
+def test_gemm(benchmark, n, variant):
     rndm = np.random.RandomState(1234)
-    a = np.array(rndm.uniform(size=(n, n)), dtype=float, order='F')
-    b = np.array(rndm.uniform(size=(n, n)), dtype=float, order='F')
-    c = np.empty((n, n), dtype=float, order='F')
-    result = benchmark(run_gemm, a, b, c)
+    dtyp = dtype_map[variant]
+    a = np.array(rndm.uniform(size=(n, n)), dtype=dtyp, order='F')
+    b = np.array(rndm.uniform(size=(n, n)), dtype=dtyp, order='F')
+    c = np.empty((n, n), dtype=dtyp, order='F')
+    gemm = ow.get_func('gemm', variant)
+    result = benchmark(run_gemm, a, b, c, gemm)
     assert result is c
 
 
 # dsyrk
 
-syrk_sizes = [100, 1000]
+syrk_sizes = [100, 200, 400, 600, 800, 1000]
 
 
-def run_syrk(a, c):
-    res = dsyrk(1.0, a, c=c, overwrite_c=True)
+def run_syrk(a, c, func):
+    res = func(1.0, a, c=c, overwrite_c=True)
     return res
 
 
+@pytest.mark.parametrize('variant', ['s', 'd', 'c', 'z'])
 @pytest.mark.parametrize('n', syrk_sizes)
-def test_syrk(benchmark, n):
+def test_syrk(benchmark, n, variant):
     rndm = np.random.RandomState(1234)
-    a = np.array(rndm.uniform(size=(n, n)), dtype=float, order='F')
-    c = np.empty((n, n), dtype=float, order='F')
-    result = benchmark(run_syrk, a, c)
+    dtyp = dtype_map[variant]
+    a = np.array(rndm.uniform(size=(n, n)), dtype=dtyp, order='F')
+    c = np.empty((n, n), dtype=dtyp, order='F')
+    syrk = ow.get_func('syrk', variant)
+    result = benchmark(run_syrk, a, c, syrk)
     assert result is c
 
 
@@ -110,21 +126,25 @@ def test_syrk(benchmark, n):
 
 # linalg.solve
 
-gesv_sizes = [100, 1000]
+gesv_sizes =  [100, 200, 400, 600, 800, 1000]
 
 
-def run_gesv(a, b):
-    res = dgesv(a, b, overwrite_a=True, overwrite_b=True)
+def run_gesv(a, b, func):
+    res = func(a, b, overwrite_a=True, overwrite_b=True)
     return res
 
 
+@pytest.mark.parametrize('variant', ['s', 'd', 'c', 'z'])
 @pytest.mark.parametrize('n', gesv_sizes)
-def test_gesv(benchmark, n):
+def test_gesv(benchmark, n, variant):
     rndm = np.random.RandomState(1234)
-    a = (np.array(rndm.uniform(size=(n, n)), dtype=float, order='F') +
-         np.eye(n, order='F'))
-    b = np.array(rndm.uniform(size=(n, 1)), order='F')
-    lu, piv, x, info = benchmark(run_gesv, a, b)
+    dtyp = dtype_map[variant]
+
+    a = (np.array(rndm.uniform(size=(n, n)), dtype=dtyp, order='F') +
+         np.eye(n, dtype=dtyp, order='F'))
+    b = np.array(rndm.uniform(size=(n, 1)), dtype=dtyp, order='F')
+    gesv = ow.get_func('gesv', variant)
+    lu, piv, x, info = benchmark(run_gesv, a, b, gesv)
     assert lu is a
     assert x is b
     assert info == 0
@@ -135,49 +155,63 @@ def test_gesv(benchmark, n):
 gesdd_sizes = [(100, 5), (1000, 222)]
 
 
-def run_gesdd(a, lwork):
-    res = dgesdd(a, lwork=lwork, full_matrices=False, overwrite_a=False)
+def run_gesdd(a, lwork, func):
+    res = func(a, lwork=lwork, full_matrices=False, overwrite_a=False)
     return res
 
 
+@pytest.mark.parametrize('variant', ['s', 'd'])
 @pytest.mark.parametrize('mn', gesdd_sizes)
-def test_gesdd(benchmark, mn):
+def test_gesdd(benchmark, mn, variant):
     m, n = mn
     rndm = np.random.RandomState(1234)
-    a = np.array(rndm.uniform(size=(m, n)), dtype=float, order='F')
+    dtyp = dtype_map[variant]
+
+    a = np.array(rndm.uniform(size=(m, n)), dtype=dtyp, order='F')
+
+    gesdd_lwork = ow.get_func('gesdd_lwork', variant)
 
-    lwork, info = dgesdd_lwork(m, n)
+    lwork, info = gesdd_lwork(m, n)
     lwork = int(lwork)
     assert info == 0
 
-    u, s, vt, info = benchmark(run_gesdd, a, lwork)
+    gesdd = ow.get_func('gesdd', variant)
+    u, s, vt, info = benchmark(run_gesdd, a, lwork, gesdd)
 
     assert info == 0
-    np.testing.assert_allclose(u @ np.diag(s) @ vt, a, atol=1e-13)
+
+    atol = {'s': 1e-5, 'd': 1e-13}
+
+    np.testing.assert_allclose(u @ np.diag(s) @ vt, a, atol=atol[variant])
 
 
 # linalg.eigh
 
-syev_sizes = [50, 200]
+syev_sizes = [50, 64, 128, 200]
 
 
-def run_syev(a, lwork):
-    res = dsyev(a, lwork=lwork, overwrite_a=True)
+def run_syev(a, lwork, func):
+    res = func(a, lwork=lwork, overwrite_a=True)
     return res
 
 
+@pytest.mark.parametrize('variant', ['s', 'd'])
 @pytest.mark.parametrize('n', syev_sizes)
-def test_syev(benchmark, n):
+def test_syev(benchmark, n, variant):
     rndm = np.random.RandomState(1234)
+    dtyp = dtype_map[variant]
+
     a = rndm.uniform(size=(n, n))
-    a = np.asarray(a + a.T, dtype=float, order='F')
+    a = np.asarray(a + a.T, dtype=dtyp, order='F')
     a_ = a.copy()
 
+    dsyev_lwork = ow.get_func('syev_lwork', variant)
     lwork, info = dsyev_lwork(n)
     lwork = int(lwork)
     assert info == 0
 
-    w, v, info = benchmark(run_syev, a, lwork)
+    syev = ow.get_func('syev', variant)
+    w, v, info = benchmark(run_syev, a, lwork, syev)
 
     assert info == 0
     assert a is v  # overwrite_a=True

benchmark/pybench/openblas_wrap/__init__.py

@@ -6,23 +6,12 @@ from benchmarking.
 __version__ = "0.1"  
 
 
-#from scipy.linalg.blas import (
-from ._flapack import (
-    # level 1
-    dnrm2 as dnrm2,
-    ddot as ddot,
-    daxpy as daxpy,
-    # level 3
-    dgemm as dgemm,
-    dsyrk as dsyrk,
-)
+from . import _flapack
+
+PREFIX = ''
+
+
+def get_func(name, variant):
+    """get_func('gesv', 'c') -> cgesv etc."""
+    return getattr(_flapack, PREFIX + variant + name)
 
-#from scipy.linalg.lapack import (
-from openblas_wrap._flapack import (
-    # linalg.solve
-    dgesv as dgesv,
-    # linalg.svd
-    dgesdd as dgesdd, dgesdd_lwork as dgesdd_lwork,
-    # linalg.eigh
-    dsyev as dsyev, dsyev_lwork as dsyev_lwork
-)