def compile_args():
"""
This args will be received by compile_str() in the preargs paramter.
They will also be included in the "hard" part of the key module.
"""
flags = [flag for flag in config.nvcc.flags.split(" ") if flag]
if config.nvcc.fastmath:
flags.append("-use_fast_math")
cuda_ndarray_cuh_hash = hash_from_file(os.path.join(os.path.split(__file__)[0], "cuda_ndarray.cuh"))
flags.append("-DCUDA_NDARRAY_CUH=" + cuda_ndarray_cuh_hash)
# NumPy 1.7 Deprecate the old API. I updated most of the places
# to use the new API, but not everywhere. When finished, enable
# the following macro to assert that we don't bring new code
# that use the old API.
flags.append("-DNPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION")
# numpy 1.7 deprecated the following macro but the didn't
# existed in the past
numpy_ver = [int(n) for n in numpy.__version__.split(".")[:2]]
if bool(numpy_ver < [1, 7]):
flags.append("-DNPY_ARRAY_ENSURECOPY=NPY_ENSURECOPY")
flags.append("-DNPY_ARRAY_ALIGNED=NPY_ALIGNED")
flags.append("-DNPY_ARRAY_WRITEABLE=NPY_WRITEABLE")
flags.append("-DNPY_ARRAY_UPDATE_ALL=NPY_UPDATE_ALL")
flags.append("-DNPY_ARRAY_C_CONTIGUOUS=NPY_C_CONTIGUOUS")
flags.append("-DNPY_ARRAY_F_CONTIGUOUS=NPY_F_CONTIGUOUS")
# If the user didn't specify architecture flags add them
if not any(["-arch=sm_" in f for f in flags]):
# We compile cuda_ndarray.cu during import.
# We should not add device properties at that time.
# As the device is not selected yet!
# TODO: re-compile cuda_ndarray when we bind to a GPU?
import theano.sandbox.cuda
if hasattr(theano.sandbox, "cuda"):
n = theano.sandbox.cuda.use.device_number
if n is None:
_logger.warn(
"We try to get compilation arguments for CUDA"
" code, but the GPU device is not initialized."
" This is probably caused by an Op that work on"
" the GPU that don't inherit from GpuOp."
" We Initialize the GPU now."
)
theano.sandbox.cuda.use(
"gpu",
force=True,
default_to_move_computation_to_gpu=False,
move_shared_float32_to_gpu=False,
enable_cuda=False,
)
n = theano.sandbox.cuda.use.device_number
p = theano.sandbox.cuda.device_properties(n)
flags.append("-arch=sm_" + str(p["major"]) + str(p["minor"]))
return flags
def compile_args():
"""
This args will be received by compile_str() in the preargs paramter.
They will also be included in the "hard" part of the key module.
"""
flags = [flag for flag in config.nvcc.flags.split(' ') if flag]
if config.nvcc.fastmath:
flags.append('-use_fast_math')
cuda_ndarray_cuh_hash = hash_from_file(
os.path.join(os.path.split(__file__)[0], 'cuda_ndarray.cuh'))
flags.append('-DCUDA_NDARRAY_CUH=' + cuda_ndarray_cuh_hash)
# NumPy 1.7 Deprecate the old API. I updated most of the places
# to use the new API, but not everywhere. When finished, enable
# the following macro to assert that we don't bring new code
# that use the old API.
flags.append("-DNPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION")
# numpy 1.7 deprecated the following macro but the didn't
# existed in the past
numpy_ver = [int(n) for n in numpy.__version__.split('.')[:2]]
if bool(numpy_ver < [1, 7]):
flags.append("-DNPY_ARRAY_ENSURECOPY=NPY_ENSURECOPY")
flags.append("-DNPY_ARRAY_ALIGNED=NPY_ALIGNED")
flags.append("-DNPY_ARRAY_WRITEABLE=NPY_WRITEABLE")
flags.append("-DNPY_ARRAY_UPDATE_ALL=NPY_UPDATE_ALL")
flags.append("-DNPY_ARRAY_C_CONTIGUOUS=NPY_C_CONTIGUOUS")
flags.append("-DNPY_ARRAY_F_CONTIGUOUS=NPY_F_CONTIGUOUS")
# If the user didn't specify architecture flags add them
if not any(['-arch=sm_' in f for f in flags]):
# We compile cuda_ndarray.cu during import.
# We should not add device properties at that time.
# As the device is not selected yet!
# TODO: re-compile cuda_ndarray when we bind to a GPU?
import theano.sandbox.cuda
if hasattr(theano.sandbox, 'cuda'):
n = theano.sandbox.cuda.use.device_number
if n is None:
_logger.warn(
"We try to get compilation arguments for CUDA"
" code, but the GPU device is not initialized."
" This is probably caused by an Op that work on"
" the GPU that don't inherit from GpuOp."
" We Initialize the GPU now.")
theano.sandbox.cuda.use(
"gpu",
force=True,
default_to_move_computation_to_gpu=False,
move_shared_float32_to_gpu=False,
enable_cuda=False)
n = theano.sandbox.cuda.use.device_number
p = theano.sandbox.cuda.device_properties(n)
flags.append('-arch=sm_' + str(p['major']) +
str(p['minor']))
return flags
def compile_args():
"""
This args will be received by compile_str() in the preargs paramter.
They will also be included in the "hard" part of the key module.
"""
flags = [flag for flag in config.nvcc.flags.split(' ') if flag]
if config.nvcc.fastmath:
flags.append('-use_fast_math')
cuda_ndarray_cuh_hash = hash_from_file(
os.path.join(os.path.split(__file__)[0], 'cuda_ndarray.cuh'))
flags.append('-DCUDA_NDARRAY_CUH=' + cuda_ndarray_cuh_hash)
# NumPy 1.7 Deprecate the old API.
# The following macro asserts that we don't bring new code
# that use the old API.
flags.append("-DNPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION")
# If the user didn't specify architecture flags add them
if not any(['-arch=sm_' in f for f in flags]):
# We compile cuda_ndarray.cu during import.
# We should not add device properties at that time.
# As the device is not selected yet!
# TODO: re-compile cuda_ndarray when we bind to a GPU?
import theano.sandbox.cuda
if hasattr(theano.sandbox, 'cuda'):
n = theano.sandbox.cuda.use.device_number
if n is None:
_logger.warn(
"We try to get compilation arguments for CUDA"
" code, but the GPU device is not initialized."
" This is probably caused by an Op that work on"
" the GPU that don't inherit from GpuOp."
" We Initialize the GPU now.")
theano.sandbox.cuda.use(
"gpu",
force=True,
default_to_move_computation_to_gpu=False,
move_shared_float32_to_gpu=False,
enable_cuda=False)
n = theano.sandbox.cuda.use.device_number
p = theano.sandbox.cuda.device_properties(n)
flags.append('-arch=sm_' + str(p['major']) +
str(p['minor']))
return flags