def cuda_svd():
# demo_types = [np.float32, np.complex64]
demo_types = [np.float32]
if cula._libcula_toolkit == "premium" and cumisc.get_compute_capability(pycuda.autoinit.device) >= 1.3:
demo_types.extend([np.float64, np.complex128])
for t in demo_types:
print "Testing svd for type " + str(np.dtype(t))
# numpy.float32
a = np.asarray((np.random.rand(1000, 1000) - 0.5) / 10, t)
print a.shape
u_cpu, s_cpu, v_cpu = np.linalg.svd(a)
# gpu array
a_gpu = gpuarray.to_gpu(a)
# call cula rutine
u_gpu, s_gpu, vh_gpu = culinalg.svd(a_gpu)
print u_gpu.get()
print u_cpu
# print s_gpu.get()
# print vh_gpu.get()
a_rec = np.dot(u_gpu.get(), np.dot(np.diag(s_gpu.get()), vh_gpu.get()))
print "Success status: ", np.allclose(a, a_rec, atol=1e-3)
print "Maximum error: ", np.max(np.abs(a - a_rec))
print ""
def test_svd_so_float64(self):
a = np.asarray(np.random.randn(6, 6), np.float64)
a_gpu = gpuarray.to_gpu(a)
u_gpu, s_gpu, vh_gpu = linalg.svd(a_gpu, 's', 'o')
assert np.allclose(a,
np.dot(u_gpu.get(),
np.dot(np.diag(s_gpu.get()), vh_gpu.get())),
atol=atol_float64)
def test_svd_so_complex128(self):
a = np.asarray(np.random.randn(6, 6) + 1j*np.random.randn(6, 6), np.complex128)
a_gpu = gpuarray.to_gpu(a)
u_gpu, s_gpu, vh_gpu = linalg.svd(a_gpu, 's', 'o')
assert np.allclose(a, np.dot(u_gpu.get(),
np.dot(np.diag(s_gpu.get()),
vh_gpu.get())),
atol=atol_float64)
def test_svd_ss_float64(self):
a = np.asarray(np.random.randn(9, 6), np.float64)
a_gpu = gpuarray.to_gpu(a)
u_gpu, s_gpu, vh_gpu = linalg.svd(a_gpu, 's', 's')
assert np.allclose(a, np.dot(u_gpu.get(),
np.dot(np.diag(s_gpu.get()),
vh_gpu.get())),
atol=atol_float64)
def test_svd_ss_complex128(self):
a = np.asarray(
np.random.randn(9, 6) + 1j * np.random.randn(9, 6), np.complex128)
a_gpu = gpuarray.to_gpu(a)
u_gpu, s_gpu, vh_gpu = linalg.svd(a_gpu, 's', 's')
assert np.allclose(a,
np.dot(u_gpu.get(),
np.dot(np.diag(s_gpu.get()), vh_gpu.get())),
atol=atol_float64)
def testForSize(x):
print 'Image Size %dx%d' % (x,x)
x = np.random.rand(x**2, 40).astype(np.float32)
def svdoverwrite(a_gpu, u_gpu, s_gpu, v_gpu, m, n, lda, ldu, ldvt):
data_type = a_gpu.dtype.type
real_type = np.float32
cula_func = cula._libcula.culaDeviceSgesvd
jobu = 'S'
jobvt = 'S'
status = cula_func(jobu, jobvt, m, n, int(a_gpu.gpudata),
lda, int(s_gpu.gpudata), int(u_gpu.gpudata),
ldu, int(v_gpu.gpudata), ldvt)
cula.culaCheckStatus(status)
# Free internal CULA memory:
cula.culaFreeBuffers()
t = time.time()
gpux = gpuarray.to_gpu(x)
pushtime = time.time() - t
print '[Push results]', pushtime
t = time.time()
u_g, s_g, v_g = cla.svd(gpux, 'S', 'S')
gpusvdtime = time.time() - t
print '[GPU results]', gpusvdtime
t = time.time()
u_g = u_g.get()
s_g = s_g.get()
v_g = v_g.get()
fetchtime = time.time() - t
print '[Fetch time]', fetchtime
t = time.time()
u_c, s_c, v_c = la.svd(x, full_matrices=False)
cputime = time.time() - t
print '[CPU time]', cputime
print '[GPU time]', pushtime + gpusvdtime + fetchtime
# Result on desktop Quadro FX1800
print
def test_svd_so_float32(self):
a = np.asarray(np.random.randn(6, 6), np.float32)
a_gpu = gpuarray.to_gpu(a)
u_gpu, s_gpu, vh_gpu = linalg.svd(a_gpu, "s", "o")
assert np.allclose(a, np.dot(u_gpu.get(), np.dot(np.diag(s_gpu.get()), vh_gpu.get())), atol=atol_float32)
def test_svd_ss_complex64(self):
a = np.asarray(np.random.randn(9, 6) + 1j * np.random.randn(9, 6), np.complex64)
a_gpu = gpuarray.to_gpu(a)
u_gpu, s_gpu, vh_gpu = linalg.svd(a_gpu, "s", "s")
assert np.allclose(a, np.dot(u_gpu.get(), np.dot(np.diag(s_gpu.get()), vh_gpu.get())), atol=atol_float32)
import pycuda.autoinit
import pycuda.driver as drv
import pycuda.gpuarray as gpuarray
import numpy as np
import scikits.cuda.linalg as culinalg
import scikits.cuda.misc as cumisc
culinalg.init()
# Double precision is only supported by devices with compute
# capability >= 1.3:
import string
import scikits.cuda.cula as cula
demo_types = [np.float32, np.complex64]
if cula._libcula_toolkit == 'premium' and \
cumisc.get_compute_capability(pycuda.autoinit.device) >= 1.3:
demo_types.extend([np.float64, np.complex128])
for t in demo_types:
print 'Testing svd for type ' + str(np.dtype(t))
a = np.asarray((np.random.rand(50, 50) - 0.5) / 10, t)
a_gpu = gpuarray.to_gpu(a)
u_gpu, s_gpu, vh_gpu = culinalg.svd(a_gpu)
a_rec = np.dot(u_gpu.get(), np.dot(np.diag(s_gpu.get()), vh_gpu.get()))
print 'Success status: ', np.allclose(a, a_rec, atol=1e-3)
print 'Maximum error: ', np.max(np.abs(a - a_rec))
print ''
"""
import pycuda.autoinit
import pycuda.driver as drv
import pycuda.gpuarray as gpuarray
import numpy as np
import scikits.cuda.linalg as culinalg
import scikits.cuda.misc as cumisc
culinalg.init()
# Double precision is only supported by devices with compute
# capability >= 1.3:
import string
import scikits.cuda.cula as cula
demo_types = [np.float32, np.complex64]
if cula._libcula_toolkit == 'premium' and \
cumisc.get_compute_capability(pycuda.autoinit.device) >= 1.3:
demo_types.extend([np.float64, np.complex128])
for t in demo_types:
print 'Testing svd for type ' + str(np.dtype(t))
a = np.asarray((np.random.rand(50, 50)-0.5)/10, t)
a_gpu = gpuarray.to_gpu(a)
u_gpu, s_gpu, vh_gpu = culinalg.svd(a_gpu)
a_rec = np.dot(u_gpu.get(), np.dot(np.diag(s_gpu.get()), vh_gpu.get()))
print 'Success status: ', np.allclose(a, a_rec, atol=1e-3)
print 'Maximum error: ', np.max(np.abs(a-a_rec))
print ''