Casting between two structs in C (CUDA)

The CUDA library effectively defines its own complex number data type, cuComplex, as follows:

  struct __builtin_align__(8) float2
  {
    float x, y;
  };
  typedef float2 cuFloatComplex;
  typedef cuFloatComplex cuComplex;

I am already using my own complex data type:

  typedef struct  { float re; float im; } complex;

and I wish to be able to cast between the two. Here's an example of what I'm currently trying:

  cuComplex cu_data;
  ...
  complex c_data;
  c_data = (complex)cu_data;

but I'm getting the (nvcc) error:

  'no suitable user-defined conversion from "cuComplex" to "complex" exists'.

So how do I do this? Equivalently: is there a way to do a typedef, but to simply rename the members of that struct? (ie rename 'x' to 're', 'y' to 'im')?

Thanks!
LVL 25
InteractiveMindAsked:
Who is Participating?
I wear a lot of hats...

"The solutions and answers provided on Experts Exchange have been extremely helpful to me over the last few years. I wear a lot of hats - Developer, Database Administrator, Help Desk, etc., so I know a lot of things but not a lot about one thing. Experts Exchange gives me answers from people who do know a lot about one thing, in a easy to use platform." -Todd S.

jkrCommented:
You could provide your own operator, e.g.
  struct /*__builtin_align__(8)*/ float2
  {
    float x, y;
  };
  typedef float2 cuFloatComplex;
  typedef cuFloatComplex cuComplex;

 typedef struct  { float re; float im; operator cuComplex();} complex;

 complex::operator cuComplex() {

   cuComplex out;

   out.x = re;
   out.y = im;

   return out;
 }

  int main () {

  cuComplex out;
  complex in;

  out = (cuComplex) in;

  return 0;
  }

Open in new window

0
jkrCommented:
Ooops, seems that I missed the 'C' part - well, in that case, make that a function:
  struct /*__builtin_align__(8)*/ float2
  {
    float x, y;
  };
  typedef float2 cuFloatComplex;
  typedef cuFloatComplex cuComplex;

 typedef struct  { float re; float im; operator cuComplex();} complex;

 cuComplex to_cuComplex(complex* in) {

   cuComplex out;

   out.x = in->re;
   out.y = in->im;

   return out;
 }

  int main () {

  cuComplex out;
  complex in;

  out = to_cuComplex(&in);

  return 0;
  }

Open in new window

0

Experts Exchange Solution brought to you by

Your issues matter to us.

Facing a tech roadblock? Get the help and guidance you need from experienced professionals who care. Ask your question anytime, anywhere, with no hassle.

Start your 7-day free trial
sridhardCommented:

Hi,

Try using pointers . This will assign/point the correponding values. Note sure whether this helps

  cuComplex cu_data = (cuComplex  *) malloc(sizeof(cuComplex ));
  ...
  complex *c_data;
  c_data = (complex*)cu_data;

You can delete any one pointer later .
0
Become a Microsoft Certified Solutions Expert

This course teaches how to install and configure Windows Server 2012 R2.  It is the first step on your path to becoming a Microsoft Certified Solutions Expert (MCSE).

InteractiveMindAuthor Commented:
Thank you, both. I think I should mention that I am in fact casting pointers/arrays of such types in my code (but I get the same problem).

@jkr, would your method work for casting pointers? (I can't test it at the moment). If so, my code needs to be as fast as possible, so given that I'm going to need to do [probably] millions of such casts per second, would it be quicker to just do a typedef:

   typedef cuComplex complex;

and then a global search and replace of '.re' with '.x' etc? (I'm reluctant to do so)

Also, can you think of any reason why what I'm already doing wouldn't work? Is it the __builtin_align__(8) that's causing problems?
0
jkrCommented:
>>@jkr, would your method work for casting pointers?

If you use the indirecton appropriately - yes.
0
Infinity08Commented:
>> given that I'm going to need to do [probably] millions of such casts per second, would it be quicker to just do a typedef:

Yes. Because no conversion would be needed then. Performing no conversion is faster than performing a conversion ;)


>> Also, can you think of any reason why what I'm already doing wouldn't work?

The error says it :

>>   'no suitable user-defined conversion from "cuComplex" to "complex" exists'.

the compiler doesn't know how to perform the conversion from cuComplex to complex, because you didn't tell it how to do that (see jkr's post for that).


>> I should mention that I am in fact casting pointers/arrays of such types in my code (but I get the same problem).

Casting pointers wouldn't give you the same problem. It would give you a different problem in C++ : the type safety system in C++ wouldn't allow a cast from cuComplex* to complex*, because the two types are not related - ie. the cast is not safe.
If you shut the compiler up, and force the cast, then the __builtin_align__(8) can indeed cause a problem. If the packing of the complex type is not the same as that of the cuComplex type, then something will go wrong badly.
0
ikeworkCommented:
If you use the same alignment for your struct (8 bytes) then the cast shouldn't be a problem.
0
ikeworkCommented:
The safest strategy is to use only cuda's struct in the first place, also in your code, so there is no cast needed anymore.
0
ambienceCommented:
>> so given that I'm going to need to do [probably] millions of such casts per second, would it be quicker to just do a typedef:

My vote -> YES
0
ikeworkCommented:
@ambience, did you consider the different alignments as previous posts said?
0
ikeworkCommented:
@ambience, sorry I misunderstood your post, I thought you were suggesting just cast anyway, nevermind :)
0
soscpdCommented:
Hi

Your answer is a function ponter. There (in the function) you read/write cuComplex/complex structs anyway you like.

Rafael
0
InteractiveMindAuthor Commented:
Thanks
0
It's more than this solution.Get answers and train to solve all your tech problems - anytime, anywhere.Try it for free Edge Out The Competitionfor your dream job with proven skills and certifications.Get started today Stand Outas the employee with proven skills.Start learning today for free Move Your Career Forwardwith certification training in the latest technologies.Start your trial today
C

From novice to tech pro — start learning today.