cpp: export operators for isl::val, isl::aff, and isl::pw_aff

[tobiasgrosser]

Add operator overloads for isl::val, isl::aff, and isl::pw_aff. We currently only overload basic arithmetic functions and comparisons between values for which a direct mathematical correspondance exists and consequently confusion is unlikely. Here an example use for isl::val:

isl::val half(ctx, "1/2");
isl::val one(ctx, "1");
isl::val neg_one(ctx, "-1");
isl::val two(ctx, "2");
isl::val four(ctx, "4");

assert(-one == neg_one);
assert(one + one == two);
assert(two - one == one);
assert(two * two == four);
assert(one / two == half);
assert(one < two);
assert(one <= two);
assert(two > one );
assert(two >= one );
assert(one == one);
assert(one != two);

Signed-off-by: Tobias Grosser tobias@grosser.es

[tobiasgrosser]

@nicolasvasilache, @ftynse, @chandangreddy: Here a first set of patches for operator overloading. A close review would be very much appreciated.

[ftynse]

Have you considered an unordered_map instead?

[tobiasgrosser]

Sven prefers to not use C++11 in the generator yet (but the generated code is C++11). Hence, I use now a std::map, instead.

[ftynse]

But uniform initialization is C++11...

…

On 03/08/17 23:42, Tobias Grosser wrote: ***@***.**** commented on this pull request. ------------------------------------------------------------------------ In interface/cpp.cc <#10 (comment)>: > @@ -553,13 +553,42 @@ void cpp_generator::print_methods_impl(ostream &os, const isl_class &clazz) } } +/* Map isl function names to C++ operators. + */ +static const char *op_map[][2] = { Sven prefers to not use C++11 in the generator yet (but the generated code is C++11). Hence, I use now a std::map, instead. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#10 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABcTa14kA4OEGLlqNyULqvijfp2NNmxkks5sUj6rgaJpZM4OqZYd>.

[tobiasgrosser]

You are right. This means I will need to move back to scanning the array. I think that's fine.

[ftynse]

Sorry about the fuss.

Everything else looks fine. There are some randomly placed spaces in tests, though.

Any ideas on how overloaded operators can be exposed, e.g. isl_val_add_ui ?

[tobiasgrosser]

@ftynse : Thanks for the review. Please be as picky as possible. Finding such spaces and such is really imporant. Sven cares about this and any such issue faster to fix in a review here, than on the mailing list. Also, the hope is that these reviews take some of the review load from Sven, such that it is easier for him to have us upstream some of these patches. So again, be as picky as possible.

[tobiasgrosser]

@ftynse : I leaft overloaded operators out on purpose. Doing so is likely requiring some discussions, which I wanted to avoid at this stage. The code snippets I gave to Nicolas should have some examples how to expose them (but not added to the generator yet). We can likely expose them just the way they are today by adding __isl_export, __isl_overload. However, I would like to also allow symmetric operations 42 + val, which requires some more codegen changes. I suggest to leave this to a future change.

[tobiasgrosser]

Proposed for inclusion on the isl mailing list.

[aisoard]

I see you didn't add isl_val_mod as the % operator. Is that on purpose?

[tobiasgrosser]

% does not count to the basic arithmetic operations for me. In fact, I mostly wanted to get the standard affine expressions with addition and multiplication to work. I expect that after this is in, other trivial extensions can be easily added. In general I aim to minimize the actual extensions when adding new functionality. % seems one of the first to add afterwards.

[aisoard]

Poor little %. Although you are right that it does not fit well yet, because the / is not an integer division here...