depooling.py

# -*- coding: utf-8 -*-
"""
.. invisible:
     _   _ _____ _     _____ _____
    | | | |  ___| |   |  ___/  ___|
    | | | | |__ | |   | |__ \ `--.
    | | | |  __|| |   |  __| `--. \
    \ \_/ / |___| |___| |___/\__/ /
     \___/\____/\_____|____/\____/

Created on Jul 10, 2014

Depoling unit.

███████████████████████████████████████████████████████████████████████████████

Licensed to the Apache Software Foundation (ASF) under one
or more contributor license agreements.  See the NOTICE file
distributed with this work for additional information
regarding copyright ownership.  The ASF licenses this file
to you under the Apache License, Version 2.0 (the
"License"); you may not use this file except in compliance
with the License.  You may obtain a copy of the License at

  http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing,
software distributed under the License is distributed on an
"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
KIND, either express or implied.  See the License for the
specific language governing permissions and limitations
under the License.

███████████████████████████████████████████████████████████████████████████████
"""


from __future__ import division
import numpy
from zope.interface import implementer

import veles.error as error
from veles.accelerated_units import IOpenCLUnit, ICUDAUnit, INumpyUnit
import veles.znicz.nn_units as nn_units


@implementer(IOpenCLUnit, ICUDAUnit, INumpyUnit)
class Depooling(nn_units.Forward):
    MAPPING = {"depooling"}

    """Depooling unit for *Max* poolings.

    Must be assigned before initialize():
        input
        output_offset
        output_shape_source

    Updates after run():
        output

    Creates within initialize():
        output

    Attributes:
        input: data to depool.
        output: depooled data.
        output_offset: input_offset from the corresponding pooling unit.
        output_shape_source: Array to get output shape from.
        krn_output_clear_: kernel for zeroing the output.
    """
    def __init__(self, workflow, **kwargs):
        super(Depooling, self).__init__(workflow, **kwargs)
        self.output_offset = None
        self.demand("output_offset", "output_shape_source")

    def init_unpickled(self):
        super(Depooling, self).init_unpickled()
        self.sources_["depooling"] = {}
        self.krn_output_clear_ = None

    def initialize(self, device, **kwargs):
        super(Depooling, self).initialize(device, **kwargs)

        if self.output_offset.size != self.input.size:
            raise error.BadFormatError("output_offset.size != input.size")

        if self.output_offset.dtype != numpy.int32:
            raise error.BadFormatError("output_offset.dtype != numpy.int32")

        if self.output:
            assert self.output.shape[1:] == self.output_shape_source.shape[1:]

        if (not self.output or
                self.output.shape[0] != self.output_shape_source.shape[0]):
            self.output.reset(numpy.zeros(self.output_shape_source.shape,
                                          dtype=self.input.dtype))

        self.init_vectors(self.input, self.output_offset, self.output)

    def _gpu_init(self):
        self.build_program(
            {"INPUT_SIZE": self.input.size}, "%s_%s" %
            (self.__class__.__name__,
             "_".join(str(i) for i in self.input.shape)),
            dtype=self.input.dtype)

        self.assign_kernel("feed_layer")
        self.set_args(self.input, self.output_offset, self.output)

    def ocl_init(self):
        self._gpu_init()

        if self.krn_output_clear_ is None:
            self.krn_output_clear_ = self.get_kernel("output_clear")
            self.krn_output_clear_.set_arg(0, self.output.devmem)

    def cuda_init(self):
        self._gpu_init()

        block_size = self.device.suggest_block_size(self._kernel_)
        self._global_size = (
            int(numpy.ceil(self.input.size / block_size)), 1, 1)
        self._local_size = (block_size, 1, 1)

    def ocl_run(self):
        """Do gradient descent.
        """
        self.unmap_vectors(self.input, self.output_offset, self.output)
        self.execute_kernel([self.output.size], None, self.krn_output_clear_)
        self.execute_kernel([self.input.size], None)

    def cuda_run(self):
        self.unmap_vectors(self.input, self.output_offset, self.output)
        self.output.devmem.memset32_async()
        self.execute_kernel(self._global_size, self._local_size)

    def numpy_run(self):
        raise NotImplementedError()

    def generate_data_for_slave(self):
        pass

    def apply_data_from_slave(self):
        pass