[erlang-questions] Accessing A Binary as an Array of Floats/Doubles Within a NIF

[Tony Rogvall]

> On 20 jun 2016, at 22:24, Sverker Eriksson <sverker.eriksson@REDACTED> wrote:
> 
> Well, reference counted binaries have 8-byte aligned data.
> Heap binaries only half of the time on 32-bit :-)

Did  not think of that. I guess that it should be fixed ?
I can see traces of an align flag... Maybe resurrect that flag?
Or better to have something like unit syntax for alignment?

<<F:64/float-native-align:64>>

But alignment just for powers of two.

/Tony

> 
> Both driver_alloc_binary() and enif_alloc_binary() guarantee
> 8-byte alignment of the returned payload data.
> 
> /Sverker
> 
> 
> On 06/20/2016 10:07 PM, Tony Rogvall wrote:
>> I just found the answer in binary.c.
>> The answer is that all binary data is always 8 byte aligned.
>> And that is nice :-)
>> 
>> /Tony
>> 
>>> On 20 jun 2016, at 21:59, Tony Rogvall <tony@REDACTED> wrote:
>>> 
>>>> On 20 jun 2016, at 21:50, Sverker Eriksson <sverker.eriksson@REDACTED> wrote:
>>>> 
>>>> 
>>>> If you use <<1.0/float-native>>
>>>> then it's just type cast to a 'double' pointer:
>>>> 
>>>> double* my_array;
>>>> 
>>>> my_array = (double*) Bin.data;
>>>> 
>>>> printf("%lf", my_array[0]);
>>>> 
>>>> 
>>>> Default byte order is 'float-big', which will not work on little endian machines (x86).
>>>> 
>>>> Depending on architecture you also might have to ensure 8 byte alignment.
>>>> x86 will accept unaligned floats but the performance may suffer.
>>>> 
>>> Is there an easy way to ensure 8 byte alignment in Erlang, when constructing binaries?
>>> On a 64-bit machine/arch my guess is that Binary data is 8 byte aligned but
>>> what about a on 32-bit machine/arch?
>>> 
>>> /Tony
>>> 
>>>> /Sverker, Erlang/OTP
>>>> 
>>>> 
>>>> On 06/20/2016 09:04 PM, John Duffy wrote:
>>>>> Hi
>>>>> I'm using FreeBSD 10.3 & Erlang 18.3 and I'm trying to get to grips with passing a binary of floats to a NIF for further processing as as array of C doubles.
>>>>> My simple NIF module compiles and I can access my "my_nif_1" function from the erlang shell. I can access the 'size' structure member but I can't figure out how to access the 'data' structure member as a C array. I understand that 'data' is a block of memory of UINT bytes but I can't figure out how to "cast" this to a C array within the NIF.
>>>>> Any help would be mud appreciated.
>>>>> Kind regards
>>>>> John
>>>>> #include "erl_nif.h"
>>>>> static ERL_NIF_TERM my_nif_1(ErlNifEnv * env, int argc, const ERL_NIF_TERM argv[])
>>>>> {
>>>>>    ErlNifBinary Bin;
>>>>>    if (!enif_inspect_binary(env, argv[0], &Bin)) {
>>>>>        return enif_make_badarg(env);
>>>>>    }
>>>>>    // So far, so good...
>>>>>    // From the erl shell...
>>>>>    // my_nifs:my_nif_1(<<1.0/float>>). returns a Bin.size of 8
>>>>>    // my_nifs:my_nif_1(<<1.0/float,2.0/float>>. returns a Bin.size of 16
>>>>>    //
>>>>>    // How do I access Bin.data as an array of doubles?
>>>>>    //
>>>>>    return enif_make_int(env, Bin.size); // For testing... this works.
>>>>> }
>>>>> static ErlNifFunc nif_funcs[] = {
>>>>>    {"my_nif_1", 1, my_nif_1}
>>>>> };
>>>>> ERL_NIF_INIT(my_nifs, nif_funcs, NULL, NULL, NULL, NULL);
>>>>> 
>>>>> 
>>>>> 
>>>>> 
>>>>> _______________________________________________
>>>>> erlang-questions mailing list
>>>>> 
>>>>> erlang-questions@REDACTED
>>>>> http://erlang.org/mailman/listinfo/erlang-questions
>>>> _______________________________________________
>>>> erlang-questions mailing list
>>>> erlang-questions@REDACTED
>>>> http://erlang.org/mailman/listinfo/erlang-questions
>>> _______________________________________________
>>> erlang-questions mailing list
>>> erlang-questions@REDACTED
>>> http://erlang.org/mailman/listinfo/erlang-questions
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20160621/882018a8/attachment.bin>