Skip to content

Low-level abstraction layer for HICANN-DLS-based chips

License

Notifications You must be signed in to change notification settings

electronicvisions/haldls

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Hardware Abstraction Layer (and STAteful encapsulation) for the HICANN-DLS

The software in this repository has been developed by staff and students of Heidelberg University as part of the research carried out by the Electronic Vision(s) group at the Kirchhoff-Institute for Physics. The research is funded by Heidelberg University, the State of Baden-Württemberg, the European Union Sixth Framework Programme no. 15879 (FACETS), the Seventh Framework Programme under grant agreements no 604102 (HBP), 269921 (BrainScaleS), 243914 (Brain-i-Nets), the Horizon 2020 Framework Programme under grant agreement 720270, 785907, 945539 (HBP) as well as from the Manfred Stärk Foundation.

This repository contains

  • HALDLS -- the data structures for configuring the hardware
  • STADLS -- code for experiment control and experiment encapsulation

How to build

Build- and runtime dependencies

All build- and runtime dependencies are encapsulated in a Singularity Container. If you want to build this project outside the Electronic Vision(s) cluster, please start by downloading the most recent version from here.

For all following steps, we assume that the most recent Singularity container is located at /containers/stable/latest – you are free to choose any other path.

Github-based build

To build this project from public resources, adhere to the following guide:

# 1) Most of the following steps will be executed within a singularity container
#    To keep the steps clutter-free, we start by defining an alias
shopt -s expand_aliases
alias c="singularity exec --app dls-core /containers/stable/latest"

# 2) Add the cross-compiler and toolchain for the embedded processor to your environment
#    If you don't have access to the module, you may build it as noted here:
#    https://github.com/electronicvisions/oppulance
module load ppu-toolchain

# 3) Prepare a fresh workspace and change directory into it
mkdir workspace && cd workspace

# 4) Fetch a current copy of the symwaf2ic build tool
git clone https://github.com/electronicvisions/waf -b symwaf2ic symwaf2ic

# 5) Build symwaf2ic
c make -C symwaf2ic
ln -s symwaf2ic/waf

# 6) Setup your workspace and clone all dependencies (--clone-depth=1 to skip history)
c ./waf setup --repo-db-url=https://github.com/electronicvisions/projects --project=haldls

# 7) Build the project
#    Adjust -j1 to your own needs, beware that high parallelism will lead to high memory consumption!
c ./waf configure
c ./waf build -j1

# 8) Install the project to ./bin and ./lib
c ./waf install

# 9) If you run programs outside waf, you'll need to add ./lib and ./bin to your path specifications
export SINGULARITYENV_PREPEND_PATH=`pwd`/bin:$SINGULARITYENV_PREPEND_PATH
export SINGULARITYENV_LD_LIBRARY_PATH=`pwd`/lib:$SINGULARITYENV_LD_LIBRARY_PATH
export PYTHONPATH=`pwd`/lib:$PYTHONPATH

# 10) You can now run any program, e.g. plain unit tests
c ./bin/stadls_swtest_vx_v3

On the Electronic Vision(s) Cluster

  • Work on the frontend machine, helvetica. You should have received instructions how to connect to it.
  • Follow aforementioned instructions with the following simplifications
    • Replace steps 4) and 5) by module load waf
    • Make sure to run step 7) within a respective slurm allocation: Prefix e.g. srun -p einc -c128 --mem 0 and remove the -j1 option; depending on your shell, you might need to roll out the c-alias.
    • Replace step 9) by module load localdir.

Build from internal sources

If you have access to the internal Gerrit server, you may drop the --repo-db-url-specification in step 6) of aforementioned instructions.

How to add a new container (HICANN-X)

Creating a new container

A container in haldls abstracts the payload to a smallest-possible/feasible read/write operation using fisch register-access containers. To each container, a unique coordinate_type coordinate from halco describes the range of accessible locations of this container unit:

class MyNewContainer
{
public:
    typedef halco::hicann_dls::vx::MyNewContainerOnDLS coordinate_type;
};

Containers can either issue register read/write operations and are then called leaf nodes or use existing containers for higher-order structuring. A leaf node container has to have a is_leaf_node typedef:

typedef std::true_type is_leaf_node;

A non-leaf node container on the other hand provides a has_local_data typedef:

typedef std::false_type has_local_data;

Creating a leaf-node container

A leaf node container provides a compile-time fixed number of register-access coordinates and for writing data payload values via the addresses and encode member functions. A decode member function is used to decode a compile-time fixed number of register values to container data.

If the container only supports register access via one fisch register container type, the function signatures look like:

class MyNewContainer
{
public:
    typedef halco::hicann_dls::vx::MyNewContainerOnDLS coordinate_type;

    constexpr static size_t config_size_in_words = N;

    static std::array<typename fisch::vx::MyNewRegister::coordinate_type, config_size_in_words> addresses();

    std::array<fisch::vx::MyNewRegister, config_size_in_words> encode() const;

    void decode(std::array<fisch::vx::MyNewRegister, config_size_in_words> const& words);
};

If multiple register types are supportes, e.g. fisch::vx::OmnibusOnChipOverJTAG and fisch::vx::Omnibus for omnibus register access on the chip, templated functions provide specializationsfor the different register types:

class MyNewContainer
{
public:
    typedef halco::hicann_dls::vx::MyNewContainerOnDLS coordinate_type;

    constexpr static size_t config_size_in_words = N;

    template <typename AddressT>
    std::array<AddressT, config_size_in_words> addresses() const;

    template <typename WordT>
    std::array<WordT, config_size_in_words> encode() const;

    template <typename WordT>
    void decode(std::array<WordT, config_size_in_words> const& words);
};

// *.cpp

template <>
std::array<typename fisch::vx::SupportedRegisterType::coordinate_type, config_size_in_words> MyNewContainer::addresses() const
{ /* ... */ }

template <>
std::array<typename fisch::vx::AnotherSupportedRegisterType::coordinate_type, config_size_in_words> MyNewContainer::addresses() const
{ /* ... */ }

// ... same for encode and decode ...

In addition to supporting a common number of words for reading and writing configuration, separated numbers can be supplied by providing read_addresses and write_addresses functions replacing to addresses with {read,write}_config_size_in_words number of entries. The associated encode function is to provide write_config_size_in_words number of words and decode is to take read_config_size_in_words number of register words as argument:

class MyNewContainer
{
public:
    typedef halco::hicann_dls::vx::MyNewContainerOnDLS coordinate_type;

    constexpr static size_t read_config_size_in_words = N;
    constexpr static size_t write_config_size_in_words = M;

    static std::array<typename fisch::vx::MyNewRegister::coordinate_type, read_config_size_in_words> read_addresses();
    static std::array<typename fisch::vx::MyNewRegister::coordinate_type, write_config_size_in_words> write_addresses();

    std::array<fisch::vx::MyNewRegister, write_config_size_in_words> encode() const;

    void decode(std::array<fisch::vx::MyNewRegister, read_config_size_in_words> const& words);
};

See DACChannel in include/haldls/vx/spi.h for a single-register-type container implementation, PPUMemoryWord for a container with support for two register types.

For the new leaf node container, a BackendContainerTrait specialization is to be provided:

namespace haldls::vx::detail {

template <>
BackendContainerTrait<MyNewContainer> : public BackendContainerBase<MyNewContainer, fisch::vx::DefaultRegister, fisch::vx::AnotherSupportedRegister>

}

The first register container specified in the list is used as the default type in PlaybackProgramBuilder::{read,write} if no explicit register container to set the transport protocol is given.

Creating a non leaf-node container

A composed non leaf-node container can only be created for containers with (a) common fisch register container(s).

A composed container typically has the following form:

class MyComposedContainer
{
public:
    typedef std::false_type has_local_data;

private:
    // some collection of other haldls containers
    halco::common::typed_array<MyNewContainer, typename MyNewContainer::coordinate_type> m_entries;
};

Opposed to leaf-node containers, a non leaf-node container does not feature addresses, encode and decode functions, as the already implemented features of the member containers are used by applying a visitor pattern. A visitor pattern specifies the order of encoding/decoding operations performed on the member containers. It is to be defined for each non leaf-node container as described in the following:

namespace haldls::vx::detail {

template <>
struct VisitPreorderImpl<MyComposedContainer>
{
    template <typename ContainerT, typename VisitorT>
    static void call(ContainerT& config, typename MyComposedContainer::coordinate_type const& coord, VisitorT&& visitor)
    {
        visitor(coord, config);

        for (auto const entry_coord : halco::common::iter_all<typename MyNewContainer::coordinate_type>()) {
            visit_preorder(config.m_entries[entry_coord], entry_coord, visitor);
        }
    }
};

} // namespace haldls::vx::detail

For a non leaf-node container the BackendContainerTrait is to be specified the same way as for leaf-node containers:

namespace haldls::vx::detail {

template <>
BackendContainerTrait<MyComposedContainer> : public BackendContainerBase<MyComposedContainer, fisch::vx::DefaultCommonRegister, fisch::vx::AnotherSupportedCommonRegister>

}

As for leaf-node containers, the first register container specified in the list is used as the default type in PlaybackProgramBuilder::{read,write} if no explicit register container to set the transport protocol is given.

Integrating a new container for builder and python support.

After having done the above steps the container is to be integrated in order to be supported by the PlaybackProgramBuilder and be available in python. The container is to be added into the list in include/haldls/vx/container.def for support by the builder. For python bindings, the header containing the container declaration is to be included in include/haldls/vx/container.h.

In addition to the above interface, each container is to support serialization by using cereal and an std::ostream operator for printing its content, see JTAGClockScaler for an exemplary implementation using debug printout of the encoded words.

Testing a new container

Software tests of containers are to be written in C++. A complete software test includes

  • General tests
    • Testing all member functions for value access
    • Testing the equality and inequality operators
    • Testing value ranges of ranged types if applicable
  • Testing encoding and decoding for unity if bijective and against fixed expected values
  • Testing serialization coverage with cereal, which includes an identity test for altered members and adding the container in the list of tested types in tests/sw/haldls/vx/test-serialization_cereal.cpp

For e.g. random member changes, test helpers can be found in tests/sw/haldls/common/test-helper.h. See tests/sw/haldls/vx/test-dac_control.cpp for an exemplary test.