def init_module():
global context, context_wrapper
if context_wrapper is not None:
return
log_sys_info()
device_id, device = select_device()
context = device.make_context(flags=driver.ctx_flags.SCHED_YIELD
| driver.ctx_flags.MAP_HOST)
debug("testing with context=%s", context)
debug("api version=%s", context.get_api_version())
free, total = driver.mem_get_info()
debug("using device %s", device_info(device))
debug("memory: free=%sMB, total=%sMB", int(free / 1024 / 1024),
int(total / 1024 / 1024))
context_wrapper = CudaContextWrapper(context)
#generate kernel sources:
for rgb_format, yuv_formats in COLORSPACES_MAP.items():
m = gen_rgb_to_yuv_kernels(rgb_format, yuv_formats)
KERNELS_MAP.update(m)
_kernel_names_ = sorted(set([x[0] for x in KERNELS_MAP.values()]))
log.info("%s csc_nvcuda kernels: %s", len(_kernel_names_),
", ".join(_kernel_names_))
#now, pre-compile the kernels:
for src_format, dst_format in KERNELS_MAP.keys():
get_CUDA_kernel(device_id, src_format, dst_format)
context.pop()
def init_module():
global context, context_wrapper
if context_wrapper is not None:
return
log_sys_info()
device_id, device = select_device()
context = device.make_context(flags=driver.ctx_flags.SCHED_YIELD | driver.ctx_flags.MAP_HOST)
debug("testing with context=%s", context)
debug("api version=%s", context.get_api_version())
free, total = driver.mem_get_info()
debug("using device %s", device_info(device))
debug("memory: free=%sMB, total=%sMB", int(free/1024/1024), int(total/1024/1024))
context_wrapper = CudaContextWrapper(context)
#generate kernel sources:
for rgb_format, yuv_formats in COLORSPACES_MAP.items():
m = gen_rgb_to_yuv_kernels(rgb_format, yuv_formats)
KERNELS_MAP.update(m)
_kernel_names_ = sorted(set([x[0] for x in KERNELS_MAP.values()]))
log.info("%s csc_nvcuda kernels: %s", len(_kernel_names_), ", ".join(_kernel_names_))
#now, pre-compile the kernels:
for src_format, dst_format in KERNELS_MAP.keys():
get_CUDA_kernel(device_id, src_format, dst_format)
context.pop()
def gen_all_kernels():
"""
Generates the source code for all the kernels.
Returns a dictionary:
* key: (src_format, dst_format)
* value: (function_name, kernel_src)
"""
kernels = {}
for rgb_format, yuv_formats in COLORSPACES_MAP.items():
m = gen_rgb_to_yuv_kernels(rgb_format, yuv_formats)
kernels.update(m)
_kernel_names_ = sorted(set([x[0] for x in kernels.values()]))
log.info("%s csc_nvcuda kernels: %s", len(_kernel_names_), ", ".join(_kernel_names_))
return kernels
def gen_all_kernels():
"""
Generates the source code for all the kernels.
Returns a dictionary:
* key: (src_format, dst_format)
* value: (function_name, kernel_src)
"""
kernels = {}
for rgb_format, yuv_formats in COLORSPACES_MAP.items():
m = gen_rgb_to_yuv_kernels(rgb_format, yuv_formats)
kernels.update(m)
_kernel_names_ = sorted(set([x[0] for x in kernels.values()]))
log.info("%s csc_nvcuda kernels: %s", len(_kernel_names_),
", ".join(_kernel_names_))
return kernels
