def test_cublasSgemv(self):
a = np.random.rand(2, 3).astype(np.float32)
x = np.random.rand(3, 1).astype(np.float32)
a_gpu = gpuarray.to_gpu(a.T.copy())
x_gpu = gpuarray.to_gpu(x)
y_gpu = gpuarray.empty((2, 1), np.float32)
alpha = np.float32(1.0)
beta = np.float32(0.0)
cublas.cublasSgemv('n', 2, 3, alpha, a_gpu.gpudata, 2, x_gpu.gpudata,
1, beta, y_gpu.gpudata, 1)
assert np.allclose(y_gpu.get(), np.dot(a, x))
def test_cublasSgemv(self):
a = np.random.rand(2, 3).astype(np.float32)
x = np.random.rand(3, 1).astype(np.float32)
a_gpu = gpuarray.to_gpu(a.T.copy())
x_gpu = gpuarray.to_gpu(x)
y_gpu = gpuarray.empty((2, 1), np.float32)
alpha = np.float32(1.0)
beta = np.float32(0.0)
cublas.cublasSgemv(self.cublas_handle, 'n', 2, 3, alpha, a_gpu.gpudata,
2, x_gpu.gpudata, 1, beta, y_gpu.gpudata, 1)
assert np.allclose(y_gpu.get(), np.dot(a, x))
def to_unit_variance(H):
''' Scales H so that column has a variance of 1. '''
from scikits.cuda.misc import _global_cublas_handle as cublas_handle
ones = gpu.empty((H.shape[0], 1), np.float32, allocator=_mempool.allocate)
ones.fill(1.0)
Hsq = gpu.empty(H.shape, np.float32, allocator=_mempool.allocate)
mean = gpu.empty((1, H.shape[1]), np.float32, allocator=_mempool.allocate)
cublasSgemv(cublas_handle, "n", H.shape[1], H.shape[0], 1.0 / H.shape[0],
H.gpudata, H.shape[1], ones.gpudata, 1, 0.0, mean.gpudata, 1)
_unitvariance_step1_kernel(H, mean, Hsq, H.shape[1])
cublasSgemv(cublas_handle, "n", Hsq.shape[1], H.shape[0], 1.0 / H.shape[0],
Hsq.gpudata, H.shape[1], ones.gpudata, 1, 0.0, mean.gpudata, 1)
_unitvariance_step2_kernel(mean, H.shape[1])
_unitvariance_step3_kernel(H, mean, H.shape[1])
return H
def to_unit_variance(H):
''' Scales H so that column has a variance of 1. '''
from scikits.cuda.misc import _global_cublas_handle as cublas_handle
ones = gpu.empty((H.shape[0], 1), np.float32, allocator=_mempool.allocate)
ones.fill(1.0)
Hsq = gpu.empty(H.shape, np.float32, allocator=_mempool.allocate)
mean = gpu.empty((1, H.shape[1]), np.float32, allocator=_mempool.allocate)
cublasSgemv(cublas_handle, "n", H.shape[1], H.shape[0],
1.0/H.shape[0], H.gpudata, H.shape[1], ones.gpudata,
1, 0.0, mean.gpudata, 1)
_unitvariance_step1_kernel(H, mean, Hsq, H.shape[1])
cublasSgemv(cublas_handle, "n", Hsq.shape[1], H.shape[0],
1.0/H.shape[0], Hsq.gpudata, H.shape[1], ones.gpudata,
1, 0.0, mean.gpudata, 1)
_unitvariance_step2_kernel(mean, H.shape[1])
_unitvariance_step3_kernel(H, mean, H.shape[1])
return H
