xblocks_devel

This library contains scripting functions for building fixed-point DSP for Xilinx FPGAs using the XSG (Xilinx System Generator) tool, associated functional implementations of those blocks using MATLAB's fixed-point libraries, and testing infrastructure for verification that the blocks work.

Configuring MATLAB & Simulink

Add the following lines (or similar) to your MATLAB startup.m file:

homedir = getenv('HOME');
xblocks_lib_path = 'lib/xblocks_devel/xblocks_library';
addpath(fullfile(homedir, xblocks_lib_path))
addpath(fullfile(homedir, xblocks_lib_path, 'Testing'))
addpath(fullfile(homedir, xblocks_lib_path, 'Testing/util'))
addpath(fullfile(homedir, xblocks_lib_path, 'matlab_data_xfer'))
addpath(fullfile(homedir, xblocks_lib_path, 'dsp48e_util_lib'))
addpath(fullfile(homedir, xblocks_lib_path, 'fixed_point_dsp'))
addpath(fullfile(homedir, xblocks_lib_path, 'scripting'))
addpath(fullfile(homedir, xblocks_lib_path, 'automation'))
addpath(fullfile(homedir, xblocks_lib_path, 'gen_test_signals'))

where xblocks_lib_path should change depending on where you've checked out the repository. An example file appears in code xblocks_library/startup.m

To speed block drawing and avoid pointless scolling, Simulink provides a way to customize the ordering of the blocks and various other features of the library browser. To make all the relevant libraries appear in a useful order,

ln -s $HOME/lib/xblocks_library/sl_customization.m $HOME/Documents/MATLAB/sl_customization.m

Creating new blocks

All blocks in this library are created with the xBlocks scripting functions (see https://casper.berkeley.edu/wiki/XBlock_Scripting) for an overview of how these scripting functions can be used to draw blocks. This has certain advantages and disadvantages over the standard simulink drawing methods. Primarily, the tradeoff is that drawing takes a bit longer than the standard CASPER library methods of block-drawing, as there is some redundancy, but after this cost is paid, your block becomes library version independent. In addition, because the library blocks are now fully text-based, they can actually be version controlled. The value of this tradeoff is debatable and ultimately comes down to personal preference.

To aid in the creation of new blocks, the functions disp_params.m and get_mask_config.m have been added to code xblocks_library/scripting/. These functions display print the current parameters of a block (in the case of the second function, just the Mask parameters). These can be useful because while scripting blocks, it can be quite annoying to determine the parameter names that need to be set as there is no master-list of parameter names.

Block generator functions

The xBlock interface, while purely text, can be a bit clunky. To combat this, as well as improve the compactness of the block definitions, this library includes a set of functions that we are calling 'block generator functions'. The idea here is that

Testing infrastructure

Testbenches for this library are handled primarily through MATLAB functional validation. Simulation data is generated by MATLAB, transferred to the Simulink model for simulation, and the results are dumped back to MATLAB for validation. The (poorly named) simulink_xsg_bridge library contains a set of blocks that transfer MATLAB variables, both real and complex, to their respective Xilinx-compatible data types, as well as a set of blocks to transfer data from Simulink to MATLAB data types.

All test cases are stored in subdirectories of

addpath(fullfile(homedir, xblocks_lib_path, 'Testing'))

New testbenches should be created by:

1. code mkdir testing/blktype/subblkname; cd testing/blktype/subblkname where 'blktype' is a categorization for the block (e.g. fft, vacc, binary_signal, etc.) and subblkname is the name of the actual block being tested, i.e. the init script would be named _init_xblock.m

2. Call the function create_tb.m, which should be on the path if startup.m was properly set. This will create 2 files, a testbench script file and a simulnk model file.

3. From the Simulink library browser, pull appropriate blocks from the 'MATLAB data xfer' library, which has been configured to be the browser name for the simulink_xsg_bridge library file if you followed the step for the sl_customization.m file above.