Could someone please sense check this assembly code?

[Ernst Blofeld]

It's taken me the best part of a day to make something that agrees with my calculator, which just enforces the fact that assembler and me will never be friends.

Could someone please look at this and tell me how sh_t it is?

All I'm trying to do is multiply a word (that is a 1:15 fixed point number) by a long, then divide by two and remove the digits after the point, taking advantage of 15 + 1 = 16, and saving the long result.

void CalculateDynamicPressure(Aircraft * this_Aircraft) {
    UWORD airDensity = AirDensity_15(this_Aircraft->position3D.up);

    ULONG velocitySquared = this_Aircraft->deltaPosition3D.east * this_Aircraft->deltaPosition3D.east +
                            this_Aircraft->deltaPosition3D.north * this_Aircraft->deltaPosition3D.north +
                            this_Aircraft->deltaPosition3D.up * this_Aircraft->deltaPosition3D.up;

    KPrintF("Dynamic pressure = 1/2 * air density * velocity squared = 1/2 * %ld * %ld", airDensity, velocitySquared);

    ULONG t0, t1;

    asm (
        "       move.l  %[velocitySquared], %[t0]           \n"
        "       move.l  %[t0], %[t1]                        \n"
        "       swap    %[t0]                               \n"

        "       mulu.w  %[airDensity], %[t0]                \n"

        "       mulu.w  %[airDensity], %[t1]                \n"
        "       clr.w   %[t1]                               \n"
        "       swap    %[t1]                               \n"
        
        "       add.l   %[t1], %[t0]                        \n"
        "       move.l  %[t0], %[dynamicPressure]           \n"

        : // outputs
            [dynamicPressure] "=g" (this_Aircraft->dynamicPressure),
            [t0] "=d" (t0),
            [t1] "=d" (t1)

        : // inputs
            [airDensity] "g" (airDensity),
            [velocitySquared] "g" (velocitySquared)
        
        : // clobbers
            "cc"
    );
    
    KPrintF("Dynamic pressure = 1/2 * air density * velocity squared = 1/2 * %ld * %ld = %ld", airDensity, velocitySquared, this_Aircraft->dynamicPressure);
}

[jotd]

where do you divide by 2 ?

[a/b]

You have to mul by 2 to convert from 1:15 after a swap.

  AAAAAAAAAAAAAAAA Aaaaaaaaaaaaaaaa 0000000000000000 = t0
+                  BBBBBBBBBBBBBBBB Bbbbbbbbbbbbbbbb = t1
= CCCCCCCCCCCCCCCC Cccccccccccccccc                  = 15-bit fraction, so 1 bit owed because of early t0 swap

[Ernst Blofeld]

Originally Posted by jotd

where do you divide by 2 ?

By shifting by 16 (swapping) rather than by 15.

Edit: You're right, I've lost a swap.

Edit 2:
Or have I?...

Testing:

air_density     velocity^2          me       excel

      39197         181476      108540      108540
      38673          37889       22358       22358
      39659         271133      164075      164075

[a/b]

Ah, you wanted to additionally divide the result by two. I missed that part, was looking at the asm code only (late hour ><). In that case, that's it. The extra *2 you needed to compensate for a swap is annulated.