pasm
Version:
Piston X86-64 Assembler
126 lines (96 loc) • 5.16 kB
Plain Text
This directory contains source code for NPGMP, a browser plug-in that
gives scripts access to GMP, the GNU multiple precision arithmetic
library. The code status is alpha to early beta.
NPGMP supports most GMP functions. For a list, see gmp-entries.h.
The code uses NPAPI, which all major browsers *except* Internet
Explorer support. I have tested it in Firefox 11 and 12 and Chrome
19.0.1084.30.
The plug-in can crash on certain input, such as requesting the square
root of a negative number. This is hard to fix.
The per-call time overhead is quite high, unless you configure your
browser to run the plug-in in-process. But if you do, the plug-in can
crash the browser.
This is free software and comes with NO WARRANTY as explained in
../LICENSE.
INSTALLATION
* GNU/Linux or a similar operating system, or some porting experience.
* npapi.h and accompanying files: http://code.google.com/p/npapi-sdk/
* the GMP development package: http://gmplib.org/devel/
* a C compiler: http://gcc.gnu.org/
* Make: http://www.gnu.org/software/make/
To install, run the following commands. If successful, restart your
browser and open one of the example pages, gmp-example.html or
pi.html. The former simply pops up an alert containing the result of
a calculation.
make
mkdir -p ~/.mozilla/plugins/
cp npgmp.so ~/.mozilla/plugins/
USAGE
Obtain a top-level scriptable object through the browser's Document
Object Model (DOM) as in gmp-example.html, where it is the variable
called "gmplib". This documentation refers to the top-level object as
gmplib or omits it, for example using "mpz.set_ui" to mean
gmplib.mpz.set_ui.
The NPGMP plug-in exposes three main types of objects:
* data types (mpz, mpq, mpf, randstate)
* functions (over 200 as documented in the GMP manual)
* constants (e.g., gmplib.gmp_version)
Each data type is simply a function that returns a fresh value of the
corresponding type. A call to one of these functions serves the
purpose of a variable declaration and default initialization in C.
For example,
var my_mpz_t = gmplib.mpz(); // Create a multi-precision integer,
var my_mpq_t = gmplib.mpq(); // rational number,
var my_mpf_t = gmplib.mpf(); // floating-point number,
var rs = gmplib.randstate(); // and random state (gmp_randstate_t).
Library functions are grouped by prefix; functions whose names begin
with "mpz_" are properties of gmplib.mpz, named by removing the
prefix. Likewise for prefixes "mpq_" "mpf_" and "gmp_". Examples:
mpf_get_prec = gmplib.mpf.get_prec
gmp_urandomb_ui = gmplib.gmp.urandomb_ui
_mpz_realloc = gmplib.mpz._realloc // oddball case
Refer to the GMP manual or http://gmplib.org/manual/ for function
documentation. Where a function accepts or returns a value of type
double, NPGMP expects or returns a JavaScript number. For integral
types like "unsigned long" and "mp_bitcnt_t" a number or decimal
string may be used.
Functions that logically return two values, such as mpz_get_d_2exp,
do so in C by returning the first value and storing the second in the
position pointed to by the first argument. NPGMP lacks the output
pointer argument (so, for example, mpz.get_d_2exp takes only one
argument, of type mpz) and returns both values in an array-like
object, like this:
var values = gmplib.mpz.get_d_2exp(a);
var d = values[0], exp = values[1];
Functions that logically return a string work differently due to C's
lack of a string type. NPGMP does not provide wrappers for
mpz_get_str and mpq_get_str, but mpz and mpq objects support a
toString method accepting a base, similarly to JavaScript numbers.
The mpf_get_str function is more complex, and NPGMP does provide a
wrapper, mpf.get_str. This function accepts three arguments
corresponding to the last three in C: base, n_digits, and the mpf
object. It returns the fraction (string) and exponent in an
array-like object as described above.
Extra functions not found in the C library include the type
predicates:
* mpz.is_mpz
* mpq.is_mpq
* mpf.is_mpf
* gmp.randstate.is_randstate
The above functions all accept one argument of arbitrary type and
return true if it is of the type implied by the function's name, or
otherwise false.
NPGMP does not support the following GMP features:
* mpz_inits, mpz_clears, and other multiple init/clear functions;
* binary input and output via mpz_import and mpz_export;
* stream input and output via mpz_out_str, etc.;
* gmp_randinit, a variadic function described as obsolete;
* functions relating to C types like mpz_fits_sint_p; however, we do
support the ulong and slong versions;
* the mpz_array_init function;
* the low-level mpn family of functions;
* formatted input and output via gmp_printf, etc.;
* the C++ interface;
* the MINT (Berkeley MP compatibility) interface;
* custom memory allocation via mp_set_memory_functions.
Errors? Questions? Ideas? Email me: John.Tobey.com