C-Mera

The C-Mera system is a set of very simple compilers that transform a notation based on S-Expressions (sexp) for C-like languages to the native syntax of that language, e.g. from sexp-C to C, and from sexp-CUDA to CUDA. The semantics of the sexp-based notation is identical to that of the native language, i.e. no inherent abstraction or layering is introduced.

There are a number of different code generators available, all based on C-Mera with a few syntactic extensions.

• cm-c is the default C code generator
• cm-cxx is an extension supporting a subset of C++
• cm-cuda is an extension featuring cuda kernel definition and call syntax
• cm-glsl is an extension of cgen that can generate opengl shader code
• cm-ocl (or cm-opencl) is an extension that can generate opencl code (currently not actively maintained and tested, though)

The recommended way to select a generator is by using the front-end program cm:

$ cm c ...
$ cm c++ ...
$ ...

The code for C-Mera and the C-backend is found in src/{c-mera,c,usr} and is rather comprehensive while the other generators (each in their own subdirectory) are quite concise. Browse the files of the derived generators to see how far the respective language support has grown.

The C-Mera system (which is the collective term for the code in the repository) is provided under the conditions of the GNU GPL version 3 or later, see the file COPYING.

To generate a C source file choose the appropriate generator and simply add the input and output file:

$ cm c input.lisp -o test.c
$ cm c++ input.lisp -o test.cpp

We recommend CCL due to long code-transformation times with SBCL.

The easiest way to configure your Lisp to load C-Mera is by adding it to quicklisp, as follows

$ ln -s <path-to-cmera> ~/quicklisp/local-projects/c-mera

Slime

With this setup it is possible to use Slime for the development process. The relevant C-Mera modules can be loaded by

(asdf:load-system :c-mera)
(asdf:load-system :cmu-c)    ; or :cmu-c++, cmu-cuda, etc. 
(in-package :cmu-c)          ; cl-user equivalent with c-mera environment for c
(cm-reader)                  ; switch to c-mera reader; optional for prototyping
                             ; switch back with (cl-reader)

After that you can enter Lisp expressions that print valid C Code to the REPL.

(simple-print 
  (function main () -> int 
    (return 0)))

Emacs Minor Mode (cm-mode)

To support proper indentation and highlighting of keywords, especially when your forms are not known to a SLIME session, we provide a simple minor mode for Emacs. You can set it up by

$ cp <path-to-cmera>/util/emacs/cm-mode.el <load-path>/cm-mode.el
$ cp <path-to-cmera>/util/emacs/cm.indent ~/.emacs.d/cm.indent

You can then add (require 'cm-mode) to your .emacs file and load it using M-x cm-mode. To load it automatically you can add a mode specification to the top of your file:

; -*- mode: Lisp; eval: (cm-mode 1); -*-

You can extend the indentation and keyword information by having an additional file called cm.indent along your source files, see the provided cm.indent for the layout.

Our cm-mode ist still rather basic and we are open for extensions (e.g. better syntax matching).

With Vim 8 asyc processes spawned Vlime, a project that strives to provide a Slime-like worlflow for Vim. We use is (via a small plugin) to drive indentation of C-Mera code. With Vim set up for Vlime you only have to drop the plugin in the appropriate place:

$ ln -s <path-to-cmera>/util/vim/lisp_cmera.vim ~/.vim/ftplugin/

To get the default behavior (see Emacs integraion) it still has to be told where to look for the cm.indent file. This can be set in your ~/.vimrc

let g:cmera_base_indent_file = '/home/kai/.emacs.d/cm.indent'

Suppose the file wc-l.lisp contains the code of the line counting example shown above. Here is a cmdline session:

$ ls
wc-l.lisp
$ cm c wc-l.lisp
#include <stdio.h>

int main(void)
{
        int c;
        int nl = 0;
        while ((c = getchar()) != EOF) {
                if (c == '\n') 
                        ++nl;
        }
        printf("%d\n", nl);
}
$ cm c wc-l.lisp -o wc-l.c
$ ls
wc-l.c wc-l.lisp
$ gcc -std=c99 wc-l.c -o wc-l

This section describes how some aspects of the system work. We only describe what we believe may be noteworthy for either the seasoned Lisp or the seasoned C programmer. This part will be in motion as we add information that some of our users would have liked to have :) So please get back to us with your experience what might be helpful to mention.

Changes from c-mera-2015

For the old version see its branch. Here we only shortly list the major differences.

• decl and for forms now require the use of = to distinguish the declarator from the initializer. Earlier we had elaborate guesses in place that worked most of the time, but not every time.

• For C++ you can also use (decl ((int v[] { 1 2 3 })) ...) instead of (decl ((int v[] = (clist 1 2 3))) ...). This change is required to be able to distinguish between regular initialization and initializer lists. The differences is easily illustrated by printing the values of the follwing vectors:

  (typedef (instantiate #:std:vector (int)) T)
  (decl ((T vec1 = (T 10 20))
         (T vec2 { 10 20 })))

• You almost never have to use use-variables and use-functions anymore.

Simple Syntax

Conditionals

if statements have exactly two or three subforms. The third subform represents the else part and is optional. Thus, the following example is not correct:

(if (!= a 0)
    (printf "all is safe")
    (return (/ b a)))

You can use progn to group multiple sub-forms

(if (!= a 0)
    (progn
      (printf "all is safe")
      (return (/ b a))))

or, equivalently, when

(when (!= a 0)
   (printf "all is safe")
   (return (/ b a)))

which expands to the previous form using progn, which, in turn, expands to:

if (a != 0) {
    ...
}

In contrast, the first example expands to

if (a != 0) {
    printf(...);
else
    return ...;

We also support cond.

Open Issues

We currently don't have unless.

Loops

A for loop is written with the loop-head grouped:

(for ((int i = 0) (< i n) (+= i 1))
  ...)

Note that C-Mera supports C-like increments and decrements for simple expressions:

(for ((int i = 0) (< i n) ++i)
  ...)

while is straighforward

(while (< a b)
   ...
   ...)

Open Issues

do-while is not implemented at the moment.

Declarations

A set of declarations is introduced with

(decl ((T name [= init])
       ...)
  ...)

or (for C++ based languages)

(decl ((T name [{ init }])
       ...)
  ...)

the initializer is optional and C-Mera collects as many symbols to be part of the type as possible, e.g.

(decl ((const unsigned long int x = 0)) ...)

is correctly identified.

As mentioned above, typenames are not checked.

In declarations (such as decl, in function parameters and (sizeof ...)) the type does not have to be enclosed in parens (and must not be). There are places, however, where for the sake of simplicity type names must be grouped, as e.g. in function return values:

(function foo ((const int *i) ...) -> (unsigned int)
  ...)

As shown in this example C-Mera also supports some C-style decorations, i.e.

(decl ((int *i 0)) ...)
(decl ((int* i 0)) ...)

are both recognized.

Namespace (Lisp vs C-Mera)

Some C-Mera symbols are also defined in Common Lisp. Initially, C-Mera starts out in the cmu-<generator> (user package, depending on the code generator used, e.g. cmu-c) which imports all cl symbols that do not conflict to provide metaprogramming as seamlessly as possible.

Especially with symbols like if etc care has to be taken to use the right one. This can be done by explicitly naming the symbol cl:if, but to define lisp functions or lisp-heavy parts of the meta code it is often more convenient to use the lisp form, such as in the example from our ELS'14 presentation:

(defmacro match (expression &rest clauses)
  `(macrolet 
     ((match-int (expression &rest clauses)
        `(progn 
           (set reg_err 
                (regcomp &reg ,(caar clauses) REG_EXTENDED))
           (if (regexec &reg ,expression 0 0 0)
               (progn ,@(cdar clauses))
               ,(lisp (if (cdr clauses)      
                          `(match-int 
                             ,expression 
                             ,@(cdr clauses))))))))
     (decl ((regex_t reg)
            (int reg_err))
       (match-int ,expression ,@clauses))))

Here we define a recursively expanding macrolet, match-int, that inserts conditional clauses (as in (if (regexec ....)) and also checks to terminate the iteration (with ,(lisp (if ...))).

Codestrings

tbd.