Runtime polymorphism for unboxed types?

Shea_Levy · February 20, 2015, 2:57pm

Hi all,

I know ATS2 supports function templates, and polymorphic functions for boxed types. Is there any support for runtime polymorphism (with vtables or standalone structs of functions or similar) available in the language?

Related question: Is there support for value templates?

~Shea

Artyom_Shalkhakov · February 21, 2015, 2:46am

I added an example of loop unrolling here:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/ATS-QA-LIST/qa-list-2015-02-20.dats

This is ingenious.

I do have a plan to implement the following feature

fun vec_add{i:int}(…) =
sif i == 0
then …
else …

vec_add{10}(…) can then be expanded at compile-time.

That would be awesome. It would drastically simplify e.g. the
implementation of generic vectors in little math libraries for use in
graphics.> On Friday, February 20, 2015 at 10:29:50 AM UTC-5, Artyom Shalkhakov wrote:

On Friday, February 20, 2015 at 8:58:01 PM UTC+6, Shea Levy wrote:

Hi all,

I know ATS2 supports function templates, and polymorphic functions for
boxed types. Is there any support for runtime polymorphism (with vtables or
standalone structs of functions or similar) available in the language?

HX once implemented Smalltalk-style objects in an ATS prototype, but I
think he abandoned the plan to implement it a long time ago.

Related question: Is there support for value templates?

Are you asking about something like:

fun{i:pos}
vec_add(x: &(@[float][i]), y: &(@[float][i]), o: &(@[float][i])): void =
// do an element-wise addition, putting results into [o]
// i is known at compile time, so we can get it at runtime
// using something like constant<int,i>, similarly to sizeof

I’d like to know the answer as well.

~Shea

H_Zhang · March 7, 2015, 1:41am

I find the example intriguing and I am trying to understand how the
unrolling is actually carried out. Is this paper
http://www.cs.bu.edu/~hwxi/academic/papers/icfp05.pdf still a good
reference on how this part of ATS works? The erasure function defined in
fig 6 seems to simply remove the proof terms. Could you talk about how the
program transformation is achieved or give a reference paper?

Thanks,
HaitaoOn Friday, February 20, 2015 at 12:42:15 PM UTC-8, gmhwxi wrote:

I added an example of loop unrolling here:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/ATS-QA-LIST/qa-list-2015-02-20.dats

I do have a plan to implement the following feature

fun vec_add{i:int}(…) =
sif i == 0
then …
else …

vec_add{10}(…) can then be expanded at compile-time.

On Friday, February 20, 2015 at 10:29:50 AM UTC-5, Artyom Shalkhakov wrote:

On Friday, February 20, 2015 at 8:58:01 PM UTC+6, Shea Levy wrote:

Hi all,

I know ATS2 supports function templates, and polymorphic functions for
boxed types. Is there any support for runtime polymorphism (with vtables or
standalone structs of functions or similar) available in the language?

HX once implemented Smalltalk-style objects in an ATS prototype, but I
think he abandoned the plan to implement it a long time ago.

Related question: Is there support for value templates?

Are you asking about something like:

fun{i:pos}
vec_add(x: &(@[float][i]), y: &(@[float][i]), o: &(@[float][i])): void =
// do an element-wise addition, putting results into [o]
// i is known at compile time, so we can get it at runtime
// using something like constant<int,i>, similarly to sizeof

I’d like to know the answer as well.

~Shea

Artyom_Shalkhakov · March 6, 2015, 3:01am

I tried to implement what is mentioned below. Unfortunately, my plan did
not pan out
well. The simple reason is that the condition following each ‘sif’ is
static but the static
information is essentially all erased at the stage where ‘sif’ needs to be
eliminated.
I could try to eliminate ‘sif’ at the point where such information is
still available but this
means templates making use of ‘sif’ cannot be handled (as template
instantiation happens
at a later stage).

For now, I think we simply need to implement a version of array_foreach
that supports
loop unrolling. For instance, see:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/TYPEVAL/dotprod.dats

Could you post the resulting C code for dotprod.dats? I’d really like to
take a look, even if it’s very long.> On Monday, March 2, 2015 at 12:22:36 PM UTC-5, Shea Levy wrote:

Cool! If we had that feature, would we be able to parameterize function
templates by any sort, and have full use of scase/sif in the
implementations? That would be very useful, I’m currently adapting your
example to implement type lists (to use where in C++ I’d use variadic
templates), but having a proper type list datasort would be much nicer than
passing around proofs that a given list is proper

~Shea

On Feb 20, 2015, at 3:42 PM, gmhwxi gmh...@gmail.com wrote:

I added an example of loop unrolling here:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/ATS-QA-LIST/qa-list-2015-02-20.dats

I do have a plan to implement the following feature

fun vec_add{i:int}(…) =
sif i == 0
then …
else …

vec_add{10}(…) can then be expanded at compile-time.

On Friday, February 20, 2015 at 10:29:50 AM UTC-5, Artyom Shalkhakov wrote:

On Friday, February 20, 2015 at 8:58:01 PM UTC+6, Shea Levy wrote:

Hi all,

I know ATS2 supports function templates, and polymorphic functions for
boxed types. Is there any support for runtime polymorphism (with vtables or
standalone structs of functions or similar) available in the language?

HX once implemented Smalltalk-style objects in an ATS prototype, but I
think he abandoned the plan to implement it a long time ago.

Related question: Is there support for value templates?

Are you asking about something like:

fun{i:pos}
vec_add(x: &(@[float][i]), y: &(@[float][i]), o: &(@[float][i])): void =
// do an element-wise addition, putting results into [o]
// i is known at compile time, so we can get it at runtime
// using something like constant<int,i>, similarly to sizeof

I’d like to know the answer as well.

~Shea

–
You received this message because you are subscribed to the Google Groups
“ats-lang-users” group.
To unsubscribe from this group and stop receiving emails from it, send an
email to ats-lang...@googlegroups.com.
To post to this group, send email to ats-l...@googlegroups.com.
Visit this group at http://groups.google.com/group/ats-lang-users.
To view this discussion on the web visit
https://groups.google.com/d/msgid/ats-lang-users/4e92a56c-bb36-4901-aa58-205fcc67f373%40googlegroups.com
https://groups.google.com/d/msgid/ats-lang-users/4e92a56c-bb36-4901-aa58-205fcc67f373%40googlegroups.com?utm_medium=email&utm_source=footer
.

gmhwxi · March 6, 2015, 12:27am

I tried to implement what is mentioned below. Unfortunately, my plan did
not pan out
well. The simple reason is that the condition following each ‘sif’ is
static but the static
information is essentially all erased at the stage where ‘sif’ needs to be
eliminated.
I could try to eliminate ‘sif’ at the point where such information is still
available but this
means templates making use of ‘sif’ cannot be handled (as template
instantiation happens
at a later stage).

For now, I think we simply need to implement a version of array_foreach
that supports
loop unrolling. For instance, see:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/TYPEVAL/dotprod.datsOn Monday, March 2, 2015 at 12:22:36 PM UTC-5, Shea Levy wrote:

Cool! If we had that feature, would we be able to parameterize function
templates by any sort, and have full use of scase/sif in the
implementations? That would be very useful, I’m currently adapting your
example to implement type lists (to use where in C++ I’d use variadic
templates), but having a proper type list datasort would be much nicer than
passing around proofs that a given list is proper

~Shea

On Feb 20, 2015, at 3:42 PM, gmhwxi <gmh...@gmail.com <javascript:>> wrote:

I added an example of loop unrolling here:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/ATS-QA-LIST/qa-list-2015-02-20.dats

I do have a plan to implement the following feature

fun vec_add{i:int}(…) =
sif i == 0
then …
else …

vec_add{10}(…) can then be expanded at compile-time.

On Friday, February 20, 2015 at 10:29:50 AM UTC-5, Artyom Shalkhakov wrote:

On Friday, February 20, 2015 at 8:58:01 PM UTC+6, Shea Levy wrote:

Hi all,

I know ATS2 supports function templates, and polymorphic functions for
boxed types. Is there any support for runtime polymorphism (with vtables or
standalone structs of functions or similar) available in the language?

HX once implemented Smalltalk-style objects in an ATS prototype, but I
think he abandoned the plan to implement it a long time ago.

Related question: Is there support for value templates?

Are you asking about something like:

fun{i:pos}
vec_add(x: &(@[float][i]), y: &(@[float][i]), o: &(@[float][i])): void =
// do an element-wise addition, putting results into [o]
// i is known at compile time, so we can get it at runtime
// using something like constant<int,i>, similarly to sizeof

I’d like to know the answer as well.

~Shea

–
You received this message because you are subscribed to the Google Groups
“ats-lang-users” group.
To unsubscribe from this group and stop receiving emails from it, send an
email to ats-lang...@googlegroups.com <javascript:>.
To post to this group, send email to ats-l...@googlegroups.com
<javascript:>.
Visit this group at http://groups.google.com/group/ats-lang-users.
To view this discussion on the web visit
https://groups.google.com/d/msgid/ats-lang-users/4e92a56c-bb36-4901-aa58-205fcc67f373%40googlegroups.com
https://groups.google.com/d/msgid/ats-lang-users/4e92a56c-bb36-4901-aa58-205fcc67f373%40googlegroups.com?utm_medium=email&utm_source=footer
.

Shea_Levy · March 2, 2015, 5:22pm

Cool! If we had that feature, would we be able to parameterize function templates by any sort, and have full use of scase/sif in the implementations? That would be very useful, I’m currently adapting your example to implement type lists (to use where in C++ I’d use variadic templates), but having a proper type list datasort would be much nicer than passing around proofs that a given list is proper

~Shea> On Feb 20, 2015, at 3:42 PM, gmhwxi gmh...@gmail.com wrote:

I added an example of loop unrolling here:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/ATS-QA-LIST/qa-list-2015-02-20.dats https://github.com/githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/ATS-QA-LIST/qa-list-2015-02-20.dats

I do have a plan to implement the following feature

fun vec_add{i:int}(…) =
sif i == 0
then …
else …

vec_add{10}(…) can then be expanded at compile-time.

On Friday, February 20, 2015 at 10:29:50 AM UTC-5, Artyom Shalkhakov wrote:
On Friday, February 20, 2015 at 8:58:01 PM UTC+6, Shea Levy wrote:
Hi all,

I know ATS2 supports function templates, and polymorphic functions for boxed types. Is there any support for runtime polymorphism (with vtables or standalone structs of functions or similar) available in the language?

HX once implemented Smalltalk-style objects in an ATS prototype, but I think he abandoned the plan to implement it a long time ago.

Related question: Is there support for value templates?

Are you asking about something like:

fun{i:pos}
vec_add(x: &(@[float][i]), y: &(@[float][i]), o: &(@[float][i])): void =
// do an element-wise addition, putting results into [o]
// i is known at compile time, so we can get it at runtime
// using something like constant<int,i>, similarly to sizeof

I’d like to know the answer as well.

~Shea

–
You received this message because you are subscribed to the Google Groups “ats-lang-users” group.
To unsubscribe from this group and stop receiving emails from it, send an email to ats-lang-user...@googlegroups.com mailto:ats-lang-user...@googlegroups.com.
To post to this group, send email to ats-lan...@googlegroups.com mailto:ats-lan...@googlegroups.com.
Visit this group at http://groups.google.com/group/ats-lang-users http://groups.google.com/group/ats-lang-users.
To view this discussion on the web visit https://groups.google.com/d/msgid/ats-lang-users/4e92a56c-bb36-4901-aa58-205fcc67f373%40googlegroups.com https://groups.google.com/d/msgid/ats-lang-users/4e92a56c-bb36-4901-aa58-205fcc67f373%40googlegroups.com?utm_medium=email&utm_source=footer.

Artyom_Shalkhakov · February 20, 2015, 3:29pm

Hi all,

I know ATS2 supports function templates, and polymorphic functions for
boxed types. Is there any support for runtime polymorphism (with vtables or
standalone structs of functions or similar) available in the language?

HX once implemented Smalltalk-style objects in an ATS prototype, but I
think he abandoned the plan to implement it a long time ago.

Related question: Is there support for value templates?

Are you asking about something like:

fun{i:pos}
vec_add(x: &(@[float][i]), y: &(@[float][i]), o: &(@[float][i])): void =
// do an element-wise addition, putting results into [o]
// i is known at compile time, so we can get it at runtime
// using something like constant<int,i>, similarly to sizeof

I’d like to know the answer as well.

gmhwxi · March 7, 2015, 2:24am

The template system of ATS2 is largely undocumented.

Unlike dependent types and linear types in ATS, I have done very little
theoretical study on templates in ATS. I just don’t feel that I have time
and energy for it

This example is actually quite straightforward.

Say you want to evaluate tally<S(S(Z))>(… | 2)

The compiler essentially generates three functions:

tally_0 for tally
tally_1 for tally<S(Z)>
tally_2 for tally<S(S(Z))>

It goes like this:

First, the compiler introduces tally_2 for tally<S(S(Z))> using the
template tally<S(t)>;
when generating the body of tally_2, the compiler introduces tally_1 for
tally<S(Z)> (as t
is mapped to S(Z) this time); when generating the body of tally_1; the
compiler introduces
tally_0 (as t is mapped to Z this time).On Friday, March 6, 2015 at 8:41:55 PM UTC-5, H Zhang wrote:

I find the example intriguing and I am trying to understand how the
unrolling is actually carried out. Is this paper
http://www.cs.bu.edu/~hwxi/academic/papers/icfp05.pdf still a good
reference on how this part of ATS works? The erasure function defined in
fig 6 seems to simply remove the proof terms. Could you talk about how the
program transformation is achieved or give a reference paper?

Thanks,
Haitao

On Friday, February 20, 2015 at 12:42:15 PM UTC-8, gmhwxi wrote:

I added an example of loop unrolling here:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/ATS-QA-LIST/qa-list-2015-02-20.dats

I do have a plan to implement the following feature

fun vec_add{i:int}(…) =
sif i == 0
then …
else …

vec_add{10}(…) can then be expanded at compile-time.

On Friday, February 20, 2015 at 10:29:50 AM UTC-5, Artyom Shalkhakov wrote:

On Friday, February 20, 2015 at 8:58:01 PM UTC+6, Shea Levy wrote:

Hi all,

I know ATS2 supports function templates, and polymorphic functions for
boxed types. Is there any support for runtime polymorphism (with vtables or
standalone structs of functions or similar) available in the language?

HX once implemented Smalltalk-style objects in an ATS prototype, but I
think he abandoned the plan to implement it a long time ago.

Related question: Is there support for value templates?

Are you asking about something like:

fun{i:pos}
vec_add(x: &(@[float][i]), y: &(@[float][i]), o: &(@[float][i])): void =
// do an element-wise addition, putting results into [o]
// i is known at compile time, so we can get it at runtime
// using something like constant<int,i>, similarly to sizeof

I’d like to know the answer as well.

~Shea

gmhwxi · March 6, 2015, 3:07am

See the attached file.

If you do:

patscc -O2 -S dotprod_dats.c

you can see that gcc can actually does a very thorough job:

dotprod3(…) is already evaluated to 14 at compile-time.On Thu, Mar 5, 2015 at 10:01 PM, Artyom Shalkhakov < artyom.s...@gmail.com> wrote:

On Friday, March 6, 2015 at 6:27:26 AM UTC+6, gmhwxi wrote:

I tried to implement what is mentioned below. Unfortunately, my plan did
not pan out
well. The simple reason is that the condition following each ‘sif’ is
static but the static
information is essentially all erased at the stage where ‘sif’ needs to
be eliminated.
I could try to eliminate ‘sif’ at the point where such information is
still available but this
means templates making use of ‘sif’ cannot be handled (as template
instantiation happens
at a later stage).

For now, I think we simply need to implement a version of array_foreach
that supports
loop unrolling. For instance, see:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/
EXAMPLE/TYPEVAL/dotprod.dats

Could you post the resulting C code for dotprod.dats? I’d really like to
take a look, even if it’s very long.

On Monday, March 2, 2015 at 12:22:36 PM UTC-5, Shea Levy wrote:

Cool! If we had that feature, would we be able to parameterize function
templates by any sort, and have full use of scase/sif in the
implementations? That would be very useful, I’m currently adapting your
example to implement type lists (to use where in C++ I’d use variadic
templates), but having a proper type list datasort would be much nicer than
passing around proofs that a given list is proper

~Shea

On Feb 20, 2015, at 3:42 PM, gmhwxi gmh...@gmail.com wrote:

I added an example of loop unrolling here:

https://github.com/githwxi/ATS-Postiats/blob/master/doc/
EXAMPLE/ATS-QA-LIST/qa-list-2015-02-20.dats

I do have a plan to implement the following feature

fun vec_add{i:int}(…) =
sif i == 0
then …
else …

vec_add{10}(…) can then be expanded at compile-time.

On Friday, February 20, 2015 at 10:29:50 AM UTC-5, Artyom Shalkhakov wrote:

On Friday, February 20, 2015 at 8:58:01 PM UTC+6, Shea Levy wrote:

Hi all,

I know ATS2 supports function templates, and polymorphic functions for
boxed types. Is there any support for runtime polymorphism (with vtables or
standalone structs of functions or similar) available in the language?

HX once implemented Smalltalk-style objects in an ATS prototype, but I
think he abandoned the plan to implement it a long time ago.

Related question: Is there support for value templates?

Are you asking about something like:

fun{i:pos}
vec_add(x: &(@[float][i]), y: &(@[float][i]), o: &(@[float][i])): void =
// do an element-wise addition, putting results into [o]
// i is known at compile time, so we can get it at runtime
// using something like constant<int,i>, similarly to sizeof

I’d like to know the answer as well.

~Shea

–
You received this message because you are subscribed to the Google
Groups “ats-lang-users” group.
To unsubscribe from this group and stop receiving emails from it, send
an email to ats-lang...@googlegroups.com.
To post to this group, send email to ats-l...@googlegroups.com.
Visit this group at http://groups.google.com/group/ats-lang-users.
To view this discussion on the web visit https://groups.google.
com/d/msgid/ats-lang-users/4e92a56c-bb36-4901-aa58-
205fcc67f373%40googlegroups.com
https://groups.google.com/d/msgid/ats-lang-users/4e92a56c-bb36-4901-aa58-205fcc67f373%40googlegroups.com?utm_medium=email&utm_source=footer
.

–
You received this message because you are subscribed to the Google Groups
“ats-lang-users” group.
To unsubscribe from this group and stop receiving emails from it, send an
email to ats-lang-user...@googlegroups.com.
To post to this group, send email to ats-lan...@googlegroups.com.
Visit this group at http://groups.google.com/group/ats-lang-users.
To view this discussion on the web visit
https://groups.google.com/d/msgid/ats-lang-users/7c7a0e69-5bb1-4090-bdd3-57f97adb2070%40googlegroups.com
https://groups.google.com/d/msgid/ats-lang-users/7c7a0e69-5bb1-4090-bdd3-57f97adb2070%40googlegroups.com?utm_medium=email&utm_source=footer
.

dotprod_dats.c (174 KB)

gmhwxi · February 20, 2015, 8:42pm

I added an example of loop unrolling here:

github.com

githwxi/ATS-Postiats/blob/master/doc/EXAMPLE/ATS-QA-LIST/qa-list-2015-02-20.dats

(* ****** ****** *)
//
// A kind of loop unrolling
//
(* ****** ****** *)
//
#include
"share/atspre_staload.hats"
//
(* ****** ****** *)

abstype Z
abstype S(type)

(* ****** ****** *)
//
dataprop
tieq (type, int) =
| TIEQZ(Z, 0)
| {t:type}{n:nat}

This file has been truncated. show original

I do have a plan to implement the following feature

fun vec_add{i:int}(…) =
sif i == 0
then …
else …

vec_add{10}(…) can then be expanded at compile-time.On Friday, February 20, 2015 at 10:29:50 AM UTC-5, Artyom Shalkhakov wrote:

On Friday, February 20, 2015 at 8:58:01 PM UTC+6, Shea Levy wrote:

Hi all,

I know ATS2 supports function templates, and polymorphic functions for
boxed types. Is there any support for runtime polymorphism (with vtables or
standalone structs of functions or similar) available in the language?

HX once implemented Smalltalk-style objects in an ATS prototype, but I
think he abandoned the plan to implement it a long time ago.

Related question: Is there support for value templates?

Are you asking about something like:

fun{i:pos}
vec_add(x: &(@[float][i]), y: &(@[float][i]), o: &(@[float][i])): void =
// do an element-wise addition, putting results into [o]
// i is known at compile time, so we can get it at runtime
// using something like constant<int,i>, similarly to sizeof

I’d like to know the answer as well.

~Shea

Topic		Replies	Views
Templates and boxing/unboxing Google Group	0	271	November 16, 2014
Some monads without higher-order functions Google Group	4	377	January 1, 2015
Formal specification of ATS2 abstract syntax tree (Template ATS2?) Google Group	5	419	November 3, 2014
Varargs functions? Google Group	5	441	September 20, 2014
C++ Expression templates in ATS2 Google Group	8	477	March 15, 2015

Runtime polymorphism for unboxed types?

Related topics