def common_impl(a, out, dpnp_func, print_debug):
if a.size == 0:
raise ValueError("Passed Empty array")
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(a.itemsize, sycl_queue)
axes, axes_ndim = 0, 0
initial = 0
where = 0
dpnp_func(out_usm, a_usm, a.shapeptr, a.ndim, axes, axes_ndim, initial,
where)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([a.size, out.size])
if print_debug:
print("dpnp implementation")
def common_impl_multinomial(n, pvals, res, dpnp_func, print_debug):
sycl_queue = dpctl_functions.get_current_queue()
res_usm = dpctl_functions.malloc_shared(res.size * res.itemsize,
sycl_queue)
pvals_usm = dpctl_functions.malloc_shared(pvals.size * pvals.itemsize,
sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, pvals_usm, pvals.ctypes,
pvals.size * pvals.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpnp_func(res_usm, n, pvals_usm, pvals.size, res.size)
event = dpctl_functions.queue_memcpy(sycl_queue, res.ctypes, res_usm,
res.size * res.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(res_usm, sycl_queue)
dpctl_functions.free_with_queue(pvals_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([res.size])
if print_debug:
print("dpnp implementation")
def dpnp_impl(a):
if a.size == 0:
raise ValueError("Passed Empty array")
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out = np.empty(1, dtype=res_dtype)
out_usm = dpctl_functions.malloc_shared(out.itemsize, sycl_queue)
dpnp_func(a_usm, out_usm, a.size)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([a.size, out.size])
if PRINT_DEBUG:
print("dpnp implementation")
return out[0]
def dpnp_impl(a):
n = a.shape[-1]
if a.shape[-2] != n:
raise ValueError("Input array must be square.")
out = a.copy()
if n == 0:
return out
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize,
sycl_queue)
dpnp_func(a_usm, out_usm, a.shapeptr)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([out.size, a.size])
if PRINT_DEBUG:
print("dpnp implementation")
return out
def dpnp_impl(a, offset=0):
if a.size == 0:
raise ValueError("Passed Empty array")
n = min(a.shape[0], a.shape[1])
res_shape = np.zeros(a.ndim - 1, dtype=np.int64)
if a.ndim > 2:
for i in range(a.ndim - 2):
res_shape[i] = a.shape[i + 2]
if (n + offset) > a.shape[1]:
res_shape[-1] = a.shape[1] - offset
elif (n + offset) > a.shape[0]:
res_shape[-1] = a.shape[0]
else:
res_shape[-1] = n + offset
shape = tuplizer(res_shape)
out = np.empty(shape, dtype=a.dtype)
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(
sycl_queue, a_usm, a.ctypes, a.size * a.itemsize
)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(
out.size * out.itemsize, sycl_queue
)
dpnp_func(
a_usm,
a.size * a.itemsize,
out_usm,
offset,
a.shapeptr,
out.shapeptr,
out.ndim,
)
event = dpctl_functions.queue_memcpy(
sycl_queue, out.ctypes, out_usm, out.size * out.itemsize
)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([a.size, out.size])
if PRINT_DEBUG:
print("dpnp implementation")
return out
def common_shape_impl(a, out, dpnp_func, PRINT_DEBUG):
if a.size == 0:
raise ValueError("Passed Empty array")
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize,
sycl_queue)
dpnp_func(a_usm, out_usm, a.shapeptr, a.ndim)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([a.size, out.size])
if PRINT_DEBUG:
print("dpnp implementation")
def dpnp_impl(a):
if a.size == 0:
return True
out = np.empty(1, dtype=np.bool_)
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize,
sycl_queue)
dpnp_func(a_usm, out_usm, a.size)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([a.size, out.size])
if PRINT_DEBUG:
print("dpnp implementation")
# TODO: sometimes all() returns ndarray
return out[0]
def dpnp_impl(a):
if a.size == 0:
raise ValueError("Passed Empty array")
sycl_queue = dpctl_functions.get_current_queue()
""" We have to pass a array in double precision to DpNp """
if copy_input_to_double:
a_copy_in_double = a.astype(np.float64)
else:
a_copy_in_double = a
a_usm = dpctl_functions.malloc_shared(
a_copy_in_double.size * a_copy_in_double.itemsize, sycl_queue
)
event = dpctl_functions.queue_memcpy(
sycl_queue,
a_usm,
a_copy_in_double.ctypes,
a_copy_in_double.size * a_copy_in_double.itemsize,
)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
if a.ndim == 2:
rows = a.shape[0]
cols = a.shape[1]
out = np.empty((rows, rows), dtype=res_dtype)
elif a.ndim == 1:
rows = 1
cols = a.shape[0]
out = np.empty(rows, dtype=res_dtype)
out_usm = dpctl_functions.malloc_shared(
out.size * out.itemsize, sycl_queue
)
dpnp_func(a_usm, out_usm, rows, cols)
event = dpctl_functions.queue_memcpy(
sycl_queue, out.ctypes, out_usm, out.size * out.itemsize
)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([a_copy_in_double.size, a.size, out.size])
if PRINT_DEBUG:
print("dpnp implementation")
if a.ndim == 2:
return out
elif a.ndim == 1:
return out[0]
def dpnp_impl(a):
n = a.shape[-1]
if a.shape[-2] != n:
msg = "Last 2 dimensions of the array must be square."
raise ValueError(msg)
dpnp_ext._check_finite_matrix(a)
wr = np.empty(n, dtype=res_dtype)
vr = np.empty((n, n), dtype=res_dtype)
if n == 0:
return (wr, vr)
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
wr_usm = dpctl_functions.malloc_shared(wr.size * wr.itemsize,
sycl_queue)
vr_usm = dpctl_functions.malloc_shared(vr.size * vr.itemsize,
sycl_queue)
dpnp_eig(a_usm, wr_usm, vr_usm, n)
event = dpctl_functions.queue_memcpy(sycl_queue, wr.ctypes, wr_usm,
wr.size * wr.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
event = dpctl_functions.queue_memcpy(sycl_queue, vr.ctypes, vr_usm,
vr.size * vr.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(wr_usm, sycl_queue)
dpctl_functions.free_with_queue(vr_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([wr.size, vr.size])
if PRINT_DEBUG:
print("dpnp implementation")
return (wr, vr)
def dpnp_impl(a, kth):
if a.size == 0:
raise ValueError("Passed Empty array")
kth_ = kth if kth >= 0 else (a.ndim + kth)
arr2 = numba_dppy.dpnp.copy(a)
out = np.empty(a.shape, dtype=a.dtype)
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
arr2_usm = dpctl_functions.malloc_shared(arr2.size * arr2.itemsize,
sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, arr2_usm, arr2.ctypes,
arr2.size * arr2.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize,
sycl_queue)
dpnp_func(a_usm, arr2_usm, out_usm, kth_, a.shapeptr, a.ndim)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(arr2_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([a.size, arr2.size, out.size])
if PRINT_DEBUG:
print("dpnp implementation")
return out
def dpnp_impl(a):
n = a.shape[-1]
if a.shape[-2] != n:
raise ValueError("Input array must be square.")
dpnp_ext._check_finite_matrix(a)
if a.ndim == 2:
out = np.empty((1, ), dtype=a.dtype)
out[0] = -4
else:
out = np.empty(a.shape[:-2], dtype=a.dtype)
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize,
sycl_queue)
dpnp_func(a_usm, out_usm, a.shapeptr, a.ndim)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([out.size, a.size])
if PRINT_DEBUG:
print("dpnp implementation")
if a.ndim == 2:
return out[0]
else:
return out
def dpnp_impl(M, tol=None, hermitian=False):
if tol is not None:
raise ValueError(
"tol is not supported for np.linalg.matrix_rank(M)")
if hermitian:
raise ValueError(
"hermitian is not supported for np.linalg.matrix_rank(M)")
if M.ndim > 2:
raise ValueError(
"np.linalg.matrix_rank(M) is only supported on 1 or 2-d arrays"
)
out = np.empty(1, dtype=M.dtype)
sycl_queue = dpctl_functions.get_current_queue()
M_usm = dpctl_functions.malloc_shared(M.size * M.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, M_usm, M.ctypes,
M.size * M.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize,
sycl_queue)
dpnp_func(M_usm, out_usm, M.shapeptr, M.ndim)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(M_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([out.size, M.size])
if PRINT_DEBUG:
print("dpnp implementation")
return out[0]
def dpnp_impl(a, repeats):
if a.size == 0:
raise ValueError("Passed Empty array")
if a.ndim >= 2:
raise ValueError("Not supported in dpnp")
new_size = a.size * repeats
out = np.zeros(new_size, dtype=a.dtype)
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize,
sycl_queue)
dpnp_func(a_usm, out_usm, repeats, a.size)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([a.size, out.size])
if PRINT_DEBUG:
print("dpnp implementation")
return out
def common_impl_multivariate_normal(mean, cov, size, check_valid, tol, res,
dpnp_func, print_debug):
sycl_queue = dpctl_functions.get_current_queue()
res_usm = dpctl_functions.malloc_shared(res.size * res.itemsize,
sycl_queue)
mean_usm = dpctl_functions.malloc_shared(mean.size * mean.itemsize,
sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, mean_usm, mean.ctypes,
mean.size * mean.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
cov_usm = dpctl_functions.malloc_shared(cov.size * cov.itemsize,
sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, cov_usm, cov.ctypes,
cov.size * cov.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpnp_func(res_usm, mean.size, mean_usm, mean.size, cov_usm, cov.size,
res.size)
event = dpctl_functions.queue_memcpy(sycl_queue, res.ctypes, res_usm,
res.size * res.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(res_usm, sycl_queue)
dpctl_functions.free_with_queue(mean_usm, sycl_queue)
dpctl_functions.free_with_queue(cov_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([res.size])
if print_debug:
print("dpnp implementation")
def common_impl(low, high, res, dpnp_func, print_debug):
check_range(low, high)
sycl_queue = dpctl_functions.get_current_queue()
res_usm = dpctl_functions.malloc_shared(res.size * res.itemsize,
sycl_queue)
dpnp_func(res_usm, low, high, res.size)
event = dpctl_functions.queue_memcpy(sycl_queue, res.ctypes, res_usm,
res.size * res.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(res_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([res.size])
if print_debug:
print("dpnp implementation")
def common_impl_hypergeometric(ngood, nbad, nsample, res, dpnp_func,
print_debug):
sycl_queue = dpctl_functions.get_current_queue()
res_usm = dpctl_functions.malloc_shared(res.size * res.itemsize,
sycl_queue)
dpnp_func(res_usm, ngood, nbad, nsample, res.size)
event = dpctl_functions.queue_memcpy(sycl_queue, res.ctypes, res_usm,
res.size * res.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(res_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([res.size])
if print_debug:
print("dpnp implementation")
def common_impl_1_arg(arg1, res, dpnp_func, print_debug):
sycl_queue = dpctl_functions.get_current_queue()
res_usm = dpctl_functions.malloc_shared(res.size * res.itemsize,
sycl_queue)
try:
dpnp_func(res_usm, arg1, res.size)
except Exception:
raise ValueError("Device not supported")
event = dpctl_functions.queue_memcpy(sycl_queue, res.ctypes, res_usm,
res.size * res.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(res_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([res.size])
if print_debug:
print("dpnp implementation")
def common_dot_impl(dpnp_func, a, b, out, m, print_debug):
sycl_queue = dpctl_functions.get_current_queue()
a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes,
a.size * a.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
b_usm = dpctl_functions.malloc_shared(b.size * b.itemsize, sycl_queue)
event = dpctl_functions.queue_memcpy(sycl_queue, b_usm, b.ctypes,
b.size * b.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize,
sycl_queue)
strides = np.array(1)
result_out = out_usm
result_size = out.size
result_ndim = out.ndim
result_shape = out.shapeptr
result_strides = strides.ctypes
input1_in = a_usm
input1_size = a.size
input1_ndim = a.ndim
input1_shape = a.shapeptr
input1_strides = strides.ctypes
input2_in = b_usm
input2_size = b.size
input2_ndim = b.ndim
input2_shape = b.shapeptr
input2_strides = strides.ctypes
dpnp_func(
result_out,
result_size,
result_ndim,
result_shape,
result_strides,
input1_in,
input1_size,
input1_ndim,
input1_shape,
input1_strides,
input2_in,
input2_size,
input2_ndim,
input2_shape,
input2_strides,
)
event = dpctl_functions.queue_memcpy(sycl_queue, out.ctypes, out_usm,
out.size * out.itemsize)
dpctl_functions.event_wait(event)
dpctl_functions.event_delete(event)
dpctl_functions.free_with_queue(a_usm, sycl_queue)
dpctl_functions.free_with_queue(b_usm, sycl_queue)
dpctl_functions.free_with_queue(out_usm, sycl_queue)
dpnp_ext._dummy_liveness_func([a.size, b.size, out.size])
if print_debug:
print("dpnp implementation")
