Skip to content

Latest commit

 

History

History
206 lines (156 loc) · 14 KB

README.md

File metadata and controls

206 lines (156 loc) · 14 KB

SNAP setup

Please check ../README.md for dependencies.

Executing the following commands is pre-requisite for the usage of the SNAP framework:

source <xilinx_root>/Vivado/<version>/settings64.sh
export XILINXD_LICENSE_FILE=<pointer to Xilinx license>

In order to handle paths and other environment settings, the SNAP make process is always sourcing the script ${SNAP_ROOT}/snap_env.sh. You may modify that script (or create it if it doesn't already exist) and add your environment settings to that script (see also SNAP environment variables).

Note: The SNAP make process is internally defining a variable ${SNAP_ROOT} which is pointing to SNAP's root directory. Therefore, this variable may be used in the specification of paths for the make process, and we are using this variable in this document in the notation of file names.

SNAP configuration

In order to configure the SNAP framework and to prepare the environment for the SNAP build process, you may call

make snap_config

from the SNAP root directory. Making use of a standalone application configuration tool based on kernel kconfig the features for a specific SNAP framework configuration can be selected. Among the features that get configured via make snap_config are

  • the card type
  • the action type (see also Specifying the action)
  • enablement of the on-card SDRAM
  • enablement of the Xilinx Integrated Logic Analyzer
  • the simulator

If additional path settings are required, the step make snap_config will tell you which variable definitions to add to the script ${SNAP_ROOT}/snap_env.sh (see also SNAP environment variables).

By calling make snap_config again, a previously defined configuration may be modified.

If you want to clean your repository (i.e. remove the generated files) you may do so by calling

make clean

But, that call will keep the configuration settings. If you want to also reset the configuration settings you may call

make clean_config

It is recommended to call make clean_config each time you want to start over with a new configuration.

SNAP environment variables

A side effect of calling make snap_config is the modification (or creation in case of first time execution) of the file ${SNAP_ROOT}/snap_env.sh. The main purpose of this file is the definition of paths that are required during the SNAP build and simulation process. The file gets sourced during SNAP's make process. As a result of the execution of make snap_config a version of ${SNAP_ROOT}/snap_env.sh containing at least five (maybe commented and incomplete) lines will exist:

export TIMING_LABLIMIT=<lower slack limit for timing considered acceptable for bitstream build>
export ACTION_ROOT=<pointer to the directory containing the action sources>
export PSLVER=<8: PSL version for POWER8 | 9: PSL version for POWER9>
export PSL_DCP=<pointer to the Vivado CAPI PSL design checkpoint file (b_route_design.dcp)>
export PSLSE_ROOT=<pointer to the path containing the PSLSE github clone>

If a file ${SNAP_ROOT}/snap_env.sh is already existing when calling make snap_config that file will be taken, and the definition of ACTION_ROOT will be adapted according to the selection of the action type.

In case of a setup for cloud builds (see Cloud Support) the following setup will be modified as well:

DCP_ROOT=<pointer to the directory for design checkpoints required in the Partial Reconfiguration flow>

As already indicated above, the notation ${SNAP_ROOT} may be used when pointing to directories or files below the SNAP root directory. For instance, if during the make snap_config step you select hdl_example as action the file ${SNAP_ROOT}/snap_env.sh will contain the line

export ACTION_ROOT=${SNAP_ROOT}/actions/hdl_example

Note: When calling make snap_config for the first time without a given ${SNAP_ROOT}/snap_env.sh, you need to edit the generated file ${SNAP_ROOT}/snap_env.sh in order to set the correct path names.

Image and model build

Specifying the action

Which action is getting integrated into the SNAP framework is specified in ${SNAP_ROOT}/snap_env.sh via the path ACTION_ROOT. When selecting "HDL Action" or "HLS Action" (for example when creating your own new action), the variable ACTION_ROOT is either set to an empty string or to the value that was selected in the previous run of make snap_config. You may simply modify the file ${SNAP_ROOT}/snap_env.sh manually to let ACTION_ROOT point to the directory containing the action.

As part of the Vivado project configuration step, the make process will call the target hw that is expected to exist in a Makefile contained in the directory that ACTION_ROOT is pointing to (see section Action wrapper). Specific configurations/preparations for the action may be added via this make process step.

CAPI board support and PSL for image build

In order to benefit from the advantage of CAPI (low latency high bandwith access to host memory as well as memory coherency) a specific implementation of the FPGA is required to use the PCIe link and CAPP (Coherent Accelerator Processor Proxy which is located in the POWER processor) properly. The PSL (POWER Service Layer) is providing the necessary infrastructure. The PSL differs whether we have a POWER8 with CAPI 1.0 or a POWER9 with CAPI 2.0.

In any case, to get the latest CAPI board support and PSL download, visit IBM Portal for OpenPOWER at

https://www.ibm.com/systems/power/openpower

From the menu, select "CAPI"->"Coherent Accelerator Processor Interface (CAPI)" or directly click the "CAPI" icon to go to the CAPI section. Then download the appropriate files depending on your target system being POWER8 (CAPI 1.0) or POWER9 (CAPI 2.0).

POWER8

CAPI board support and PSL are provided combined in a Vivado DCP (Design CheckPoint). Please go to the IBM Portal for OpenPOWER and download the required files under "PSL Checkpoint Files for the POWER8 CAPI SNAP Design Kit" according to the selected FPGA card.
For the image build process you need to make sure that in snap_env.sh the environment variable PSL_DCP is defined such that it is pointing to the downloaded Vivado DCP.

POWER9

CAPI BSP (Board Support Package) and PSL got separated. For the CAPI 2.0 BSP, the open source git repository https://github.com/open-power/capi2-bsp is integrated into the SNAP framework as git submodule under hardware/capi2-bsp.

The PSL is integrated into the CAPI BSP as an IP core with encrypted sources. Please go to the IBM Portal for OpenPOWER and download the required files under "PSL IP Source Files for POWER9 CAPI".

Alternatively, the following link will provide direct access:

https://www-355.ibm.com/systems/power/openpower/posting.xhtml?postingId=1BED44BCA884D845852582B70076A89A

If the environment variable PSL9_IP_CORE is defined (e.g. via snap_env.sh), the build process uses it as a pointer to the archived PSL9 IP core. Otherwise, the build process is assuming the archived PSL9 IP core is located in the directory snap/hardware/capi2-bsp/psl.

The make process

If you call make without any targets, then a help message will be printed explaining the different targets supported by the make process.

A simulation model (for the simulator defined in the snap_config step) may be created via the target model:

make model

This will also build the software tools and the PSLSE which are required to run the simulation.

If you want to build an image (a bitstream), you may call make with target image:

make image

Note: You must still build the software tools on the POWER target system.

FPGA bitstream image update

Please see snap/hardware/doc/Bitstream_flashing.md for instructions on how to update the FPGA bitstream image, build factory images and program cards from scratch.

Action wrapper

The definition of ACTION_ROOT in ${SNAP_ROOT}/snap_env.sh specifies the path to the set of actions that shall be included. It has to point to a directory within snap/actions.
The SNAP hardware build process is expecting each action example's root directory to contain a Makefile providing at least the targets clean and hw (see also snap/actions/README.md).

At this point SNAP supports the integration of one action. Multi-action support will follow.

Corresponding to the ports that the SNAP framework provides, each action has to provide ports for the following AXI interfaces:

  • an AXI slave port for MMIO based control
  • an AXI master port for host DMA traffic
  • an optional AXI master port for on-card SDRAM traffic
  • an optional AXI master port for communication with an NVMe host controller Furthermore, HDL actions have to implement the interrupt ports as shown in the HDL example action wrapper.
    For HLS actions, the HLS compiler will automatically generate the necessary interrupt logic.

Note that the ID widths of the AXI interfaces to host memory and to the on-card SDRAM have to be large enough to support the number of actions that shall be instantiated. For the build process this is controlled via the option NUM_OF_ACTIONS which is configurable in the make snap_config step.
Note* The SNAP framework currently does not support more than one action.

The support for actions created with HLS as opposed to HDL actions written directly in VHDL or Verilog differs slightly.

HDL Action

In order to integrate an HDL action into the SNAP framework the directory that ACTION_ROOT is pointing to, needs to contain an entity named action_wrapper which is serving as interface between the action and the SNAP framework.

An example for such an action together with a corresponding wrapper may be found in snap/actions/hdl_example/hw. It can be used for various verification scenarios like copying memory content.

HLS Actions

The top level entity for an HLS action needs to be named hls_action.

You'll also find examples for HLS actions in snap/actions. For each example, the HLS action to be integrated into the SNAP framework is contained in a file hls_<action>.cpp. Before initiating the SNAP framework's make process, you need to create the necessary RTL files (VHDL or Verilog) out of the hls_<action>.cpp file. Each example contains a Makefile that takes care of this. In order to make sure that the resulting top level entity fits into the SNAP framework, the HLS actions top level function needs to be named hls_action.

The variable ACTION_ROOT needs to point to a subdirectory of snap/actions. That directory should contain a Makefile for generating the RTL code from the HLS .cpp sources. As shown in the examples, the Makefile should include the file snap/actions/hls.mk. That way the RTL code for the HLS action will be placed into a subdirectory hw contained in the directory that $ACTION_ROOT points to, and the SNAP framework's make process is able to include it.

Example: By setting up ${SNAP_ROOT}/snap_env.sh such that ACTION_ROOT points to ${SNAP_ROOT}/actions/hls_memcopy the HLS memcopy example contained in snap/actions/hls_memcopy will get integrated into the SNAP framework.

Note: For the integration of HLS actions into the SNAP framework, ACTION_ROOT needs to point to (a subdirectory of) a directory starting with hls (case doesn't matter), or the environment variable $HLS_SUPPORT needs to be defined and to be set to TRUE (this is an option that is handled during the environment setup via make snap_config).

SDRAM Card Memory

The SNAP framework supports the usage of the on-card SDRAM. It is accessible for the actions via the action wrapper through an AXI master interface. In the SNAP configuration step the existence of that interface is controlled via the option 'Enable SDRAM'.

NVMe support

For FPGA cards with NVMe flash attached, the SNAP framework supports the integration of up to two Flash drives. Via the SNAP configuration option 'Enable NVMe' the instantiation of the NVMe host controller together with the corresponding PCIe root complexes and the required AXI interfaces can be configured.

The actions directory contains HDL and HLS based examples on how to use NVMe:

Note the data itself doesn't flow through the NVMe host controller, but transits through the on-card SDRAM. Please see the NVMe main documentation page for more details.

Hardware debug with ILA cores

In order to create an image that allows debugging the design using the Vivado Integrated Logic Analyzer (ILA) you may prepare a .xdc file for adding ILA cores to the design (an example for such a file is located here: ./doc/ila_debug.xdc).
Enabling ILA_DEBUG in the SNAP configuration and defining the environment variable

ILA_SETUP_FILE

in ${SNAP_ROOT}/snap_env.sh such that it points to a .xdc file with ILA core definitions, will configure the preparation of the ILA cores during the image build process. Additionally to the image files itself, the build process will create the required .ltx debug probes file which will be located in the results directory ${SNAP_ROOT}/hardware/build/Images.

Cloud support

TBD...

Simulation

see ./sim/README.md for further instructions about Simulation