Embedding Julia in C++ - Determining returned array types

[Kyle Kotowick]

I'm trying to embed Julia code in a C++ app I'm building. I have it all working and can do some basic calls, all good so far.

One thing I'm struggling with, though, is determining what type of array is returned by a Julia evaluation. The only available documentation seems to be this page, which is far from comprehensive and doesn't show how to do this. Let's say I call some abstract Julia code (dynamic, unknown at compile time) like this:

jl_value_t *ret = jl_eval_string(some_code_string);

if(jl_is_array(ret)) {

    

    // we now know that the returned value is an array, but how do we know what the inner type is? (e.g. Float64, Int64, etc.)

    

}

After determining that an array was returned, how would you determine what the inner type of the array is (i.e. the type of the objects it contains)?

And furthermore, if it returns an array of type "Any", would there be any way to tell what the type is of any arbitrary element in that array?

Thanks!

[Isaiah Norton]

On Fri, Oct 14, 2016 at 2:28 PM, Kyle Kotowick <ky...@kotowick.ca> wrote:

After determining that an array was returned, how would you determine what the inner type of the array is (i.e. the type of the objects it contains)?

`jl_array_eltype`

 

And furthermore, if it returns an array of type "Any", would there be any way to tell what the type is of any arbitrary element in that array?

`jl_typeof`, after retrieving the element (which will be boxed)

 

Thanks!

[Bart Janssens]

In addition to this, CxxWrap.jl has some additional convenience classes to work with Julia arrays from C++, see

https://github.com/JuliaInterop/CxxWrap.jl#working-with-arrays

Cheers,

Bart

[Kyle Kotowick]

Awesome, thanks. Could you show how to use it in a minimal code example? Here's what I'm currently trying, but it does not appear to be working:

jl_value_t *ret = jl_eval_string(code_string);
void* array_type = jl_array_eltype(ret);
jl_array_t *ret_array = (jl_array_t*)ret;



if (jl_typeis(array_type, jl_int64_type)) {
 long *data = (long*) jl_array_data(ret_array);
}
else if (jl_typeis(array_type, jl_float64_type)) { double *data = (double*) jl_array_data(ret_array);
}

[Kyle Kotowick]

I apologize for the formatting, that should be:

jl_value_t *ret = jl_eval_string(code_string);
void* array_type = jl_array_eltype(ret);
jl_array_t *ret_array = (jl_array_t*)ret;

if (jl_typeis(array_type, jl_int64_type)) {
  long *data = (long*) jl_array_data(ret_array);
}
else if (jl_typeis(array_type, jl_float64_type)) {
  double *data = (double*) jl_array_data(ret_array);
}

And the issue is that even if it is an Int64 or Float64, neither of those IF statements will return true.

[Isaiah Norton]

The issue here is that `jl_array_eltype` is already returning a type.

`jl_typeis(v, t)` becomes `jl_typeof(v) == t`, so your checks become:

    jl_typeof(array_type) == jl_int64_type

But

    jl_typeof(array_type) -> DataType

Instead, either do the equality check directly:

  array_type == jl_int64_type

Or use the exported API only (jlapi.c):

  strcmp(jl_typename_str(array_type), jl_typename_str(jl_int64_type))

[Yichao Yu]

On Tue, Oct 18, 2016 at 8:57 PM, Isaiah Norton <isaiah...@gmail.com> wrote:

The issue here is that `jl_array_eltype` is already returning a type.

`jl_typeis(v, t)` becomes `jl_typeof(v) == t`, so your checks become:

    jl_typeof(array_type) == jl_int64_type

But

    jl_typeof(array_type) -> DataType

Instead, either do the equality check directly:

  array_type == jl_int64_type

Or use the exported API only (jlapi.c):

  strcmp(jl_typename_str(array_type), jl_typename_str(jl_int64_type))

Just to be clear, this does not guarantee that the two types are equal. Directly comparing the pointer for types is also "exported API".

[Kyle Kotowick]

Aha, that fixed it!

I'm running into one issue though. What do I do with the data when it's a "Float16" type? C++ has no way to represent a 16-bit float, so I'm having difficulty converting it to a regular 32-bit float.

jl_value_t *ret = jl_eval_string(code_string);
jl_array_t *ret_array = (jl_array_t*)ret;

if (jl_array_eltype(ret) == jl_float16_type) {

 /* now I need to convert the 'ret_array' to a C++ array of floats, but how do I do that when they're given in Float16? */

}

[Yichao Yu]

On Wed, Oct 19, 2016 at 12:54 PM, Kyle Kotowick <ky...@kotowick.ca> wrote:

Aha, that fixed it!

I'm running into one issue though. What do I do with the data when it's a "Float16" type? C++ has no way to represent a 16-bit float, so I'm having difficulty converting it to a regular 32-bit float.

Find a compiler / library that supports it? ;-p

I'm only half joking. You can always just keep a pointer to the value or store it in a int16_t as long as you don't need to do any operation on it. If you want to operate on them in C++ natively you would have to find a C/C++ compiler/library (or write the necessary operations yourself) that supports __fp16.

[Stefan Karpinski]

You can pull the significant bits and the exponent values out of a Float16 and then use those to compute the value as a Float32: https://en.wikipedia.org/wiki/Half-precision_floating-point_format

[Isaiah Norton]

e.g. http://half.sourceforge.net/

But if you aren't sure you need to support this, then just throw an error (Float16 is relatively new and AFAICT only useful in specialized applications because of the precision limit, which is why support is rare)

If you are sure you (a) need it but (b) only need to up- and down-convert, see the routines in `base/float.jl`. They should be straightforward to port to C++.

On Wed, Oct 19, 2016 at 1:00 PM, Yichao Yu <yyc...@gmail.com> wrote: