I need single precision multiplication intrinsic for C6678

Here is another relevant question.

When I tried double precision intrinsic _complex_mpysp, it has no problem on compiling. But, when I insert _fod(), such as:

temp1 = _fod(A_det_real, A_det_imag);

according to C66x CPU and ISA Reference Guide , it also gives warning: a function declared implicitly warning. How to use _fod()? Thanks.

Robert,

The _cmpysp intrinsic is documented in the C6000 Compiler Users Guide as supported for C6600. Are you sure you are compiling the file with -mv6600 option?

I didn't see the _fod intrinsic documented in this Users Guide. Where did you see it documented?

Hi,

The compiler has the option of -mv6600. This, and other variable definition, _cmpysp() and the link error all can be seen from the below picture.

What is reason for the link error on my CCS v5?

Thanks.

I am not clear about the rts library settings. I also have tried rts6600_elf.lib in the "Properties for matrix_inv0/General", it doesn't work either instead of "<automatic>". What is the relationship between rts6600_elf.lib and -mv6600?

Regards,

_fod, which is from below text of Page 14, "Optimizing Loops on the C66x DSP", is a typo?

I find _ftod in "DSPF_sp_autocor.h", which can pass compiling in my project.

/* Store the result */
_amemd8((void*)&r[i]) = _ftod(_hif(sum2)+_lof(sum2), _hif(sum1)+_lof(sum1));
_amemd8((void*)&r[i+2]) = _ftod(_hif(sum4)+_lof(sum4), _hif(sum3)+_lof(sum3));

Please confirm this for me.

..............

 "Optimizing Loops on the C66x DSP"

3.1.6.2 Form a Register Pair and Extract a Value From a Register Pair
If the complex number is stored in a register pair in the way defined previously,
_hif(src) can be used to get the real part of the complex value, and _lof(src) can be used
to get the imaginary part. This is similar to _hill(src) and _loll(src) intrinsics in the
C64x+.
To form a register pair from 2 floating point numbers, _fod(src1, scr2) is needed. If this
intrinsics is used to form a complex number, _fod(real, imag) must be done. This is
similar to _itoll(srcq, src2) intrinsic in C64x+.
See the C66x CPU and ISA Reference Guide [2] for other intrinsics in this category.

Robert W said:

The compiler has the option of -mv6600. This, and other variable definition, _cmpysp() and the link error all can be seen from the below picture.

The _cmpysp intrinsic takes two __float2_t types as input and returns a __x128_t type. For this to work please include c6x.h to your source file and make sure the return type is __x128_t (this type is documented in section 7.5.6 of the Compiler Users Guide).