CBG CtoV: C to Verilog Compiler


CBG CtoV - C to Gates HLS Compiler

The CBG CtoV compiler compiles C code to synthesisable Verilog RTL. It was started in 1995 and developed at the University of Cambridge Computer Laboratory between 1995 and 1999. The tool was licensed to Tenison EDA which was itself purchased by ARC and is now owned by Synopsys. The compiler was used for FPGA designs and for ASIC testbenches. It spurred further synthesis research work using the H2 tool and also the Kiwi HLS tools.

Uses

  • Perform rapid prototyping and performance analysis in C
  • Generate synthesisable Verilog from the C
  • Rapidly repartition a design with different numbers of RAM blocks and processors.
  • Use multiple threads to describe the design in a natural way.

Short Paper

Here is a two page summary PDF.

Summary

This CtoV compiler converts almost arbitrary C/C++ code to a hardware netlist. It operates by building the datapath for a custom VLIW-like processor and a sequencer to control the datapath that may or may not be microprogrammed. The datapath may consist of any number of RAMs and ALUs and also holding registers for data that cannot be transferred to/from RAM at the current time owing to RAM port bandwidth constraints (structural hazards). There are many possible datapaths that will serve to execute a given program, trading time for space. The compiler will generate a datapath using an internal heuristic that places each array present in the source code in a different RAM and converts all other state-bearing variables to flip-flops. However, the user can override the default action to control the datapath shape: he may specify the location of every variable stating whether it should be placed in a RAM or not, and if so, which RAM. The number of RAMs used and the style and number of ports on each RAM can also be fully specified by the user. This allows complete flexibility within a design space that encompasses a pair of opposite points: one point where all state is held in flip-flops and another point that places all programming-model state in a single, single-ported RAM.

The compiler partitions the behaviour of the input program into macrocycles. A macrocycle corresponds to a number of successive steps in the input program by one thread. The number of steps implemented by a macrocycle is controlled by a combination of three things: a size heuristic, decidability of name aliases and user 'barrier' statements inserted in the source code. Each macrocycle may take a number of clock cycles to execute on the hardware, depending on the availability of ALU's and RAM ports to compute, source and sink the data. Either run-time or compile-time arbitration can be used to resolve such structural hazards. Again, a default heuristic is provided: the compiler performs compile-time arbitration for competition between events caused by a common thread and run-time arbitration for inter-thread competition. Un-decidability causes termination of the macrocycle generation when array subscript comparison and/or loop exit conditions cannot be calculated at compile time.

Features

CtoV is a package that enables hardware to be generated from C. It is based on the TT CY compiler core. For performance analysis and fast system simulation, the C can also be run as software by linking with a threads library. We provide this and a set of standard Tenos C libraries for compiling to hardware.

Compiling C and C++ to hardware is a broad subject that ranges from simple processing that is almost like macro-expansion to complex processing involving a great deal of compile-time manipulation and automatic mapping. CtoV sits at the complex processing end of this spectrum, and, in our view (2000), the industry will take some years before this sort of tool is in widespread use.

CtoV can handle a very large subset of C, but not arbitrary recursive functions. With the current version of CtoV, it is unrealistic to take and use a section of C not written specifically for hardware compilation. However, variables can be updated by more than one thread and fork, join and mutexes are provided, which is a major step forward compared with Verilog.

Miniature Tutorial Example From The Manual

AND gate

The first example is a section of C which makes a single AND gate. It is important for practical hardware design that an engineer can instruct the tools to produce gate-level features where required.

    void and2gate(u1 *y, u1 a, u1 b)
    {
       *y = a && b;
    }
In the example, the AND output is via a call by reference parameter instead of a returned value. An alternative would be as follows
    u1 and2gate(u1 a, u1 b)
    {
       return a & b;
    }
The switch to logical AND (&&) from binary AND (&) makes no difference in this example. The output from compiling the first version of this simple gate is as follows. The second version does not make sense to compile as a top level routine since our compiler discards the value returned from the top level routine.

Documentation

The CY compiler is an advanced behavioural elaborator that provides the internal processing for CtoV. The front end elaborates the block-structured imperative source language into a set of control and data units which can be implemented largely in parallel.

The backend performs logic minimisation, folding and allocation of processing units, such as adders and multipliers to implement the required data flow graph. Unlike typical DSP compilers, CtoV does not artificially distinguish between control and data and thereby allows arbitrary data-dependence in the algorithms.

Example Runs

These example runs mostly use the old netlist format. The MT mode format allocates variables representing nets on the heap instead of on the stack. An extension to CtoV also supports most SystemC constructs, but this has not been tested in depth.

  • A simple example CtoV run Serial Transmitter Example.
  • A second example CtoV run Bubble Sorter Example(s).
  • Further examples: Illux suite of tiny examples.

    Future

    It is intended to subsume the functionality of the CtoV tool inside a higher-level synthesis tool that can generate complete hardware/software embedded systems from formal specifications: HPRLS.
  •   total used in directory 272 available 2300872832
      drwxrwxr-x  2 djg11 djg11  4096 Aug 29 07:08 .
      drwxr-xr-x 54 djg11 djg11 24576 Aug 22 14:00 ..
      -rw-rw-r--  1 djg11 djg11  1840 Dec  3  1999 background.gif
      -rw-rw-r--  1 djg11 djg11  1871 Apr  9  1999 ctov.back.gif
      -rw-rw-r--  1 djg11 djg11 23745 Feb  3  2003 ctovbubble.html
      -rw-r--r--  1 djg11 djg11 26173 Feb 28  2003 ctov-manual.html
      -rw-rw-r--  1 djg11 djg11  7428 Aug 29 07:08 ctovpage.html
      -rw-rw-r--  1 djg11 djg11 38229 Jul 13  2007 ctov-short1.pdf
      -rw-rw-r--  1 djg11 djg11 38230 Jul 13  2007 ctov-short.pdf
      -rw-r--r--  1 djg11 djg11 20660 Apr 10  2000 ctov-user-manual.html
      -rw-r--r--  1 djg11 djg11  4301 Feb  3  2003 illux.txt
      lrwxrwxrwx  1 djg11 djg11    13 Nov  3  2004 index.html -> ctovpage.html
      -rw-r--r--  1 djg11 djg11 24847 Feb 16  2007 serialw.html
    

    DJ Greaves, 2002.