def do_register_gpu_opt(f):
from theano.sandbox.cuda import cuda_available, register_opt
from theano.gof.opt import local_optimizer
f = local_optimizer([base_op_class])(f)
if cuda_available:
f = register_opt()(f)
return f # just ignore
def do_register_gpu_opt(f):
from theano.sandbox.cuda import cuda_available, register_opt
from theano.gof.opt import local_optimizer
f = local_optimizer([base_op_class])(f)
if cuda_available:
f = register_opt()(f)
return f # just ignore
def f(local_opt):
name = (kwargs and kwargs.pop("name")) or local_opt.__name__
if isinstance(local_opt, DB):
opt = local_opt
else:
opt = local_optimizer(tracks)(local_opt)
gpu_optimizer2.register(name, opt, "fast_run", "gpuarray", *tags)
return local_opt
mode=mode)
return unpad_dims(ret_padded, img, 2, nd)
pool_db.register(
"local_gpua_pool_dnn_alternative",
op_lifter([Pool])(local_gpua_pool_dnn_alternative),
"gpuarray",
"fast_compile",
"fast_run",
"cudnn",
position=0,
)
pool_db2.register(
"local_gpua_pool_dnn_alternative",
local_optimizer([Pool])(local_gpua_pool_dnn_alternative),
"gpuarray",
"fast_compile",
"fast_run",
"cudnn",
position=0,
)
def local_gpua_pool_dnn_grad_stride(op, ctx_name, inputs, outputs):
if not dnn_available(ctx_name):
return
if not op.ignore_border:
return
inp, out, out_grad, ws, stride, pad = inputs
nd = op.ndim