Cortex-M4 DSP operations with nRF52

[Jackson Wilson]

The ARM Cortex-M4 supports many assembly-level DSP instructions and we would like to use them (not just the packaged ones in the CMSIS DSP libraries).

For example the set of signed multiply-accumulate halfword operations (SMLABB, for example) is very useful.  

How can we use these valuable M4 assembly instructions, which require several qualifiers (references to variables in the C program)?  Is there a good example of inline C for the nRF52?

Thanks,

Jackson  

[SmallerPond]

Are you asking how to write inline assembly?  Or are you looking for pre-written wrappers for those instructions, like a library?  If the former, it's just something like:

int32_t signed_mult_accum( int16_t m, int16_t n, int32_t a )
{
    int32_t ret;
    __asm__( "SMLABB %%r0, %%r2, %%r1, %%r10;"
                : "=r" (ret)
                : "m" (m), "n" (n), "a" (a) );
    return ret;
}

or something very much like it.  Then just repeat for the other instructions you want.  There aren't really that many of them to do.

(Warning: I'm not an assembly developer and I didn't test that, I'm on a phone. This is from what I remember of assembly.)

**EDIT** And actually, here is a not too bad-looking tutorial: www.codeproject.com/.../Using-Inline-Assembly-in-C-C

[Jackson Wilson]

I was hoping to find some wrappers for these instructions, since the syntax for inline assembly varies across platforms (and I'm not an assembly developer either).

After some digging (this is a good page for ARM GCC inline assembly:  http://www.ethernut.de/en/documents/arm-inline-asm.html), here's the edited function:

int32_t signed_mult_accum_SMLABB( int32_t m, int32_t n, int32_t a )
{
    int32_t ret;
    __asm__( "SMLABB %[rtn], %[w1], %[w2], %[add]"
                : [rtn] "=r" (ret)
                : [w1] "r" (m), [w2] "r" (n), [add] "r" (a)
                :);
    return ret;
}