github/DeepFaceLive
1
0
Fork 0
mirror of https://github.com/iperov/DeepFaceLive synced 2025-08-19 04:59:28 -07:00
Code Issues Projects Releases Packages Wiki Activity Actions

update xlib.avecl

Browse source
This commit is contained in:
iperov 2021-10-22 19:31:35 +04:00
commit acbe4957e6
3 changed files with 16 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

4
xlib/avecl/_internal/Tensor.py
View file

@ -82,9 +82,9 @@ class Tensor:
self.get_buffer().set(value)
return self
def np(self):
def np(self, out=None):
"""Returns numpy value of a Tensor"""
return self.get_buffer().np(self.shape, self.dtype)
return self.get_buffer().np(self.shape, self.dtype, out=out)
### OPERATORS

11
xlib/avecl/_internal/backend/Buffer.py
View file

@ -85,20 +85,21 @@ class Buffer:
CL.clWaitForEvents(1, ( CL.cl_event * 1 )(ev) )
CL.clReleaseEvent(ev)
def np(self, shape : Iterable, dtype : np.dtype):
def np(self, shape : Iterable, dtype : np.dtype, out=None):
"""
Returns data of buffer as np.ndarray with specified shape and dtype
"""
out_np_value = np.empty (shape, dtype)
if out is None:
out = np.empty (shape, dtype)
if out_np_value.nbytes != self._size:
if out.nbytes != self._size:
raise ValueError(f'Unable to represent Buffer with size {self._size} as shape {shape} with dtype {dtype}')
clr = CL.clEnqueueReadBuffer(self._device._get_ctx_q(), self.get_cl_mem(), True, 0, self._size, out_np_value.ctypes.data, 0, None, None)
clr = CL.clEnqueueReadBuffer(self._device._get_ctx_q(), self.get_cl_mem(), True, 0, self._size, out.ctypes.data, 0, None, None)
if clr != CL.CLERROR.SUCCESS:
raise Exception(f'clEnqueueReadBuffer error: {clr}')
return out_np_value
return out
def __str__(self):
return f'Buffer [{self._size} bytes][{f"{self._cl_mem.value}" if self._cl_mem is not None else "unallocated"}] on {str(self._device)}'

9
xlib/avecl/_internal/backend/Device.py
View file

@ -86,7 +86,6 @@ class Device:
"""
compiled_krn, prog = self._cached_kernels.get(key, (None, None) )
if compiled_krn is None:
clr = CL.CLRESULT()
prog = CL.clCreateProgramWithSource(self._get_ctx(), 1, CL.c_char_p(kernel_text.encode()), None, clr )
@ -343,8 +342,6 @@ N of cacheddata: {len(self._cached_data)}
wait(False) wait execution to complete
"""
ckernel = self._compile_kernel(kernel, kernel.get_kernel_text())
if global_shape is None:
global_shape = kernel.get_global_shape()
if global_shape is None:
@ -376,8 +373,8 @@ N of cacheddata: {len(self._cached_data)}
for i,v in enumerate(local_shape):
global_shape_offsets_ar[i] = v
krn_args = []
for i, arg in enumerate(args):
if isinstance(arg, Buffer):
arg = arg.get_cl_mem()
else:
@ -385,7 +382,11 @@ N of cacheddata: {len(self._cached_data)}
if cl_type is None:
raise ValueError(f'Cannot convert type {arg.__class__} to OpenCL type.')
arg = cl_type(arg)
krn_args.append(arg)
ckernel = self._compile_kernel(kernel, kernel.get_kernel_text())
for i, arg in enumerate(krn_args):
clr = CL.clSetKernelArg(ckernel, i, CL.sizeof(arg), CL.byref(arg))
if clr != CL.CLERROR.SUCCESS:
raise Exception(f'clSetKernelArg error: {clr}')