[racket-dev] ffi vectors

Okay. I'm kind of glad that I came up with the normal thing.

I'll put it on my infinitely long list of TODO items to come back and
extend the ctypes so that they can specify both aspects of their size.

Jay

On Thu, Sep 30, 2010 at 7:25 PM, Matthew Flatt <mflatt at cs.utah.edu> wrote:
> I guess I misunderstood what you were looking for.
>
> It would be nice to have a compact ctype that adapts like the C array
> type to different contexts. For my FFI tasks, I've gotten by with
> structure types like `_float4'; I manually choose between `_float4' or
> `_float4-pointer' as needed in different contexts (the former for
> `malloc' or a struct member, the latter for a function argument or
> result). That strategy was was particularly awkward for a struct that
> contained an array of 32 bytes, though.
>
> At Thu, 30 Sep 2010 17:19:43 -0600, Jay McCarthy wrote:
>> Yes, but this program:
>>
>> typedef float float4[4];
>> typedef struct {
>>   float4 a;
>>   float b;
>> } astruct;
>>
>> void go(float4 a, astruct b)
>>  {
>>   printf("in go: %d\n", sizeof(a));
>>   printf("in go: %d\n", sizeof(float4));
>>   printf("in go: %d\n", sizeof(b));
>>   printf("in go: %d\n", sizeof(astruct));
>>
>>  }
>>
>>  int main() {
>>   float4 a;
>>   astruct b;
>>   printf("in main: %d\n", sizeof(a));
>>   printf("in main: %d\n", sizeof(float4));
>>   printf("in main: %d\n", sizeof(b));
>>   printf("in main: %d\n", sizeof(astruct));
>>   go(a, b);
>>  }
>>
>> produces
>>
>> in main: 16
>> in main: 16
>> in main: 20
>> in main: 20
>> in go: 8
>> in go: 16
>> in go: 20
>> in go: 20
>>
>> The FFI is not just used for making function calls, it is also used
>> for specify structures and malloc-ing on behalf of the functions you
>> want to call. It is very awkward to make cstructs that are the correct
>> size and malloc the right amount without similar functionality to C in
>> this regard.
>>
>> Jay
>>
>> On Thu, Sep 30, 2010 at 4:51 PM, Matthew Flatt <mflatt at cs.utah.edu> wrote:
>> > When you run this program on a 32-bit machine:
>> >
>> >  #include <stdio.h>
>> >
>> >  void go(float a[4])
>> >  {
>> >   printf("in go: %d\n", sizeof(a));
>> >  }
>> >
>> >  int main() {
>> >   float a[4];
>> >   printf("in main: %d\n", sizeof(a));
>> >   go(a);
>> >  }
>> >
>> > you'll see "in main: 16" and "in go: 4".
>> >
>> > As far as I know, the "4" in " void go(float a[4])" is ignored. It's
>> > really the same as "void go(float a[])" or "void go(float *a)". Unless
>> > the declaration of an array variable is allocating the variable, the
>> > variable is really a pointer, and sizeof() reflects that.
>> >
>> > Along the same lines, a `_cvector' in the FFI always has a pointer
>> > size, because it's always like a pointer.
>> >
>> > A `(_cvector o _float 4)' or `(_f32vector o 4)' is probably
>> > what you want, if the function you'll calling fills in the vector. A
>> > `_float4-pointer' (not `_float4'!) if you allocate it yourself or
>> > `(_pointer o _float4)' is also fine.
>> >
>> >
>> > At Thu, 30 Sep 2010 16:41:29 -0600, Jay McCarthy wrote:
>> >> I'd like to be able to define ctypes like I would make a typedef in C like
>> >>
>> >> typedef float           float4[4];
>> >>
>> >> But it doesn't seem like this works in the FFI. See the program below
>> >> with its awkward work-around:
>> >>
>> >> #lang racket
>> >> (require ffi/unsafe
>> >>          ffi/unsafe/cvector
>> >>          ffi/vector
>> >>          tests/eli-tester)
>> >>
>> >> (test
>> >>  (ctype-sizeof _float) => 4
>> >>
>> >>  (ctype-sizeof _cvector) => 4
>> >>  (ctype-sizeof (_cvector i _float)) => 4
>> >>  (ctype-sizeof (_cvector o _float 4)) => (* 4 4)
>> >>  (ctype-sizeof (_cvector io _float 4)) => (* 4 4)
>> >>
>> >>  (ctype-sizeof _f32vector) => 4
>> >>  (ctype-sizeof (_f32vector i)) => 4
>> >>  (ctype-sizeof (_f32vector o 4)) => (* 4 4)
>> >>  (ctype-sizeof (_f32vector io 4)) => (* 4 4)
>> >>
>> >>  (local [(define-cstruct _float4
>> >>            ([f0 _float]
>> >>             [f1 _float]
>> >>             [f2 _float]
>> >>             [f3 _float]))]
>> >>    (test
>> >>     (ctype-sizeof _float4) => 16)))
>> >>
>> >> Output is:
>> >>
>> >> test: 5/11 test failures:
>> >> unsaved-editor4119:11:1: test failure in (ctype-sizeof (_cvector i _float))
>> >>   expected: 4
>> >>        got: error: expand: unbound identifier in module
>> >> unsaved-editor4119:12:1: test failure in (ctype-sizeof (_cvector o _float
>> 4))
>> >>   expected: 16
>> >>        got: 4
>> >> unsaved-editor4119:13:1: test failure in (ctype-sizeof (_cvector io _float
>> 4))
>> >>   expected: 16
>> >>        got: error: expand: unbound identifier in module
>> >> unsaved-editor4119:17:1: test failure in (ctype-sizeof (_f32vector o 4))
>> >>   expected: 16
>> >>        got: 4
>> >> unsaved-editor4119:18:1: test failure in (ctype-sizeof (_f32vector io 4))
>> >>   expected: 16
>> >>        got: error: expand: unbound identifier in module
>> >>
>> >> Normally I would just go do this, but I don't really understand the
>> >> FFI. If someone can point me appropriately, I'll go do it.
>> >>
>> >> Jay
>> >>
>> >> --
>> >> Jay McCarthy <jay at cs.byu.edu>
>> >> Assistant Professor / Brigham Young University
>> >> http://teammccarthy.org/jay
>> >>
>> >> "The glory of God is Intelligence" - D&C 93
>> >> _________________________________________________
>> >>   For list-related administrative tasks:
>> >>   http://lists.racket-lang.org/listinfo/dev
>> >
>>
>>
>>
>> --
>> Jay McCarthy <jay at cs.byu.edu>
>> Assistant Professor / Brigham Young University
>> http://teammccarthy.org/jay
>>
>> "The glory of God is Intelligence" - D&C 93
>



-- 
Jay McCarthy <jay at cs.byu.edu>
Assistant Professor / Brigham Young University
http://teammccarthy.org/jay

"The glory of God is Intelligence" - D&C 93