Strange behavior of MPY instruction (C5505)

Michal Szymanski

Hi, I've written set of c-callable assembler functions. The purpose was to make some fractional arithmetic faster, however now it's mostly curiosity why it's not working. I have functions for adding, subtracting, multiplying etc. All of them look quite similar: set status registers ,copy 1st arg from T0 to AC0, make certain arithmetic and store result in AC0, shift AC0 (Q15), store return result in T0,restore status register to their previous state. ADD and SUB are working fine, but when using MPY instruction result in AC0 is always zero, for any arguments. I'm not talking about return value, I've checked in debuger, that right after MPY instruction AC0 is zero. I have even tried to make multiplying by constant (MPYK #2,AC0), and it's still zero. Source code is below. Status register initialization is same as in DSPLIB mul32 function, but other combinations was also tested. Hope that someone have some assembler experience and can tell what is going on.

Thanks in advance

extern Int16 fmul(Int16,Int16);

ASM:

    .def _fmul
    .cpl_on
    .arms_on
    .c54cm_off

    .text

_fmul:

    PSH    mmap(ST3_55)

PSH mmap(ST2_55)

PSH mmap(ST1_55)

    PSH    mmap(ST0_55)

    AND    #001FFh, mmap(ST0_55)    ;clear all ACOVx,TC1, TC2, C
    OR    #04140h, mmap(ST1_55)    ;set CPL, SXMD, FRCT
    AND    #0F9DFh, mmap(ST1_55)    ;clear M40, SATD, 54CM
    AND #07A00h, mmap(ST2_55)    ;clear ARMS, RDM, CDPLC, AR[0-7]LC
    AND    #0FFDDh, mmap(ST3_55)    ;clear SATA, SMUL

    MOV T0,AC0
    MPY T1,AC0       ; <- thats feral instruction
    SFTA AC0,# -15
    MOV AC0,T0

    POP mmap(ST0_55)

POP mmap(ST1_55)

POP mmap(ST2_55)

POP mmap(ST3_55)
RET

over 14 years ago

0 Mark M over 14 years ago

TI__Mastermind 30120 points

Hi Michal,

We will have to look into this one. With the holidays, it may be until next week.

Thanks,
Mark

0 Michal Szymanski over 14 years ago in reply to Mark M

Expert 1215 points

I might add that I want to use signed numbers (twos complement). For example:

Int16 a,b;
a=32768*(-800)/1000;
b=32768*(500)/1000;

When negative 16-bit value is loaded to 40-bit accumulator AC0 it's filed with leadings ones. However, even when both values are positive, result in AC0 is zero.

0 Michal Szymanski over 14 years ago

Expert 1215 points

Ok, so I've found out what's wrong. 16-bit value should be stored in higher part of accumulator => MOV T0,HI(AC0).

0 Mark M over 14 years ago in reply to Michal Szymanski

TI__Mastermind 30120 points

Nice catch, Michal.

You are correct. The MPY instruction uses the high part of ACx (bits 32-16) as one of its operands.

MPY[R] [ACx,] ACy --> This instruction performs a multiplication in the D-unit MAC. The input
operands of the multiplier are ACx(32−16) and ACy(32−16)

Reference: http://focus.ti.com/lit/ug/swpu067e/swpu067e.pdf#page=527

Is your code working okay now?

Regards,
Mark

0 Michal Szymanski over 14 years ago in reply to Mark M

Expert 1215 points

Thanks Mark, and yes - function is working fine now. I still can't believe I've read that description so many times while the answer was in first line. But it was worthy. Using asm to multiply 2 values is not good idea - it takes more cycles than multiplying in C, but when you have e.g. 128 pairs of values boost is huge.

Processors

Processors forum

Strange behavior of MPY instruction (C5505)