[GAP Forum] New (in development) package : gapcpp

[Christopher Jefferson]

On 01/04/2015 12:32, "Sandeep Murthy" <s.murthy at mykolab.com> wrote:

>Hi
>
>This could be useful for optimising sorting and search, since GAP is very
>slow.
>
>Can it deal with lists of GAP objects, like groups and tuples?

At the moment no, it only deals with objects which have an "obvious" C++
version. One could write a C++ wrapper for these types, but if it simply
called back to the methods on the original GAP objects, it wouldn't
particularlly be faster.

This will get much easier in HPC-GAP, as by design HPC-GAP makes it much
easier to have GAP and C/C++ objects interact, due to better memory
manager.


>
>Sandeep Murthy
>s.murthy at mykolab.com
>
>> On 1 Apr 2015, at 12:20, Christopher Jefferson <caj21 at st-andrews.ac.uk>
>>wrote:
>> 
>> Ever wish you could just write C and C++ code in GAP? Well now you can,
>> with the gapcpp package from: https://github.com/ChrisJefferson/gapcpp
>> 
>> After running './configure' in the packages directory and loading it,
>>you
>> will find one method, 'CompileMethod', which accepts C++ (or C) code,
>>the
>> name of your function and it's number of arguments. The function is then
>> automatically linked into GAP, and GAP types turned into C++ types (and
>> vice versa).
>> 
>> Here's an easy example:
>> 
>> gap> fun := CompileMethod("int f(int X) { return X; }", "f", 1);;
>> gap> fun(3);
>> 3
>> 
>> That's fairly boring however. Let's consider something more interesting!
>> 
>> Perhaps we want to sort a list of integers by their last digit:
>> 
>> gap> SortBy([1..1000000], x -> x mod 10);;
>> gap> time;
>> 548
>> 
>> Hmm.. Lets give that 'mod' function some C++ power!
>> 
>> gap> m := CompileMethod("int f(int X) { return X % 10; }", "f", 1);;
>> gap> SortBy([1..1000000], m);;
>> gap> time
>> 487
>> 
>> Hmm, a little speedup, but not as much as we would like. Let's push the
>> whole thing into C++, using new 'triple quotes'  to make the code easier
>> to write (in the master branch of GAP, not package specific!)
>> 
>> gap> x := CompileMethod("""
>> bool comp(int i, int j)
>> { return i%10 < j%10; }
>> std::vector<int> sb(std::vector<int> i)
>> { std::sort(i.begin(), i.end(),comp); return i; }""", "sb", 1);
>> gap> x([1..1000000]);;
>> gap> time;
>> 48
>> 
>> Wow, 10x speedup!
>> 
>> Of course, C++ lets us do new things. Perhaps we really wish we had a
>> stable sort? No problem!
>> 
>> gap> x := CompileMethod("""
>> bool comp(int i, int j)
>> { return i%10 < j%10; }
>> std::vector<int> sb(std::vector<int> i)
>> { std::stable_sort(i.begin(), i.end(),comp); return i; }""", "sb", 1);
>> gap> x([1..1000000]);;
>> gap> time;
>> 124
>> 
>> A little slower, but with the bonus of being stable!
>> 
>> Let's try some more things:
>> 
>> gap> PartialSums([1..50000]);;
>> gap> time;
>> 26021
>> 
>> gap> x := CompileMethod("""
>> #include <numeric>
>> std::vector<int> sums(std::vector<int> v)
>> {
>>  std::vector<int> out(v.size());
>>  // r means 'reverse', we do this to agree with GAP's order
>>  std::partial_sum(v.rbegin(), v.rend(), out.rbegin());
>>  return out;
>> }""", "sums", 1);;
>> gap> PartialSums([1..50000]);;
>> gap> time;
>> 1
>> 
>> The following C++ types are currently supported (recursively). More
>>types
>> can be added on request!
>> 
>> std::vector
>> std::list
>> std::deque
>> std::pair
>> std::string
>> int
>> Bool
>> vec1 (a custom vector type which is 1-indexed, for each GAP<->C++
>> integration)
>> 
>> optional (a way of marking missing values, so std::vector<optional<int>
>>>
>> supports lists with missing values, unlike std::vector<int>)
>> 
>> If you understand GAP, You can also use Obj and all the normal GAP
>>method
>> to access objects. There is also initial support for gap records (see
>>the
>> tests for examples).
>> 
>> 
>> I am interested in any requests for improvements, or usage of this
>>package
>> (or wished usages of this package).
>> 
>> 
>> Chris
>> 
>> 
>> _______________________________________________
>> Forum mailing list
>> Forum at mail.gap-system.org
>> http://mail.gap-system.org/mailman/listinfo/forum
>