Need Help to round number

[sandruzzo]

I'm using this shitty way to round number:

lsr.w #4,d1
lsl.w #4,d1

Is there a fast way beside table?

[a/b]

and.w #-1<<4,d1    ; %1111111111110000

[sandruzzo]

@a/b

Soucery! How does it work?

[hooverphonique]

your shifting exercise does nothing but clear the lower 4 bits ;-)

edit: well, it also sets some status flags which you aren't using anyway, so...

[sandruzzo]

Originally Posted by hooverphonique

your shifting exercise does nothing but clear the lower 4 bits ;-)

edit: well, it also sets some status flags which you aren't using anyway, so...

Ok, I was focused on math operation not on bits...

[Samurai_Crow]

Except that it should be the bitwise inverse of #-1.

[mcgeezer]

errrr...

and.w #$fff0

that's the sourcery rounding....

[StingRay]

Yeah, a/b's solution is completely correct!

[lmimmfn]

Originally Posted by mcgeezer

errrr...

and.w #$fff0

that's the sourcery rounding....

This makes the most sense to me and is easily readable

[a/b]

The idea behind -1 is that it works with any size, it's automagically expanded to $ff, $ffff, or $ffffffff by assembler.
If you want to be strict, you could go with ~(1<<N-1), where N is the number of bits you want to clear, and get the same size-independent expression but -1 is just so simple and I personally don't find it confusing. No big deal either way...

[mcgeezer]

Originally Posted by a/b

The idea behind -1 is that it works with any size, it's automagically expanded to $ff, $ffff, or $ffffffff by assembler.
If you want to be strict, you could go with ~(1<<N-1), where N is the number of bits you want to clear, and get the same size-independent expression but -1 is just so simple and I personally don't find it confusing. No big deal either way...

Personally I find the solution over complicated, why complicate a simple boolean AND with assembler directives unnecessarily?

I use the bit shift stuff alot in VASM.

Consider the BLTSIZE register:

move.w #(32<<6)+(32/16),BLTSIZE

....makes perfect sense to me because you are telling the assembler you want to blit a 32x32 block while assigning the correct bits in BLTSIZE.

Using the OR directive is equally useful.

BIT15: equ $8000
BIT14: equ $4000

or.w #(BIT15|BIT14),d0

will set both bits 15 and 14 in d0, great for dealing with DMA stuff... like this.

move.w	#(DMAF_SETCLR!DMAF_MASTER!DMAF_COPPER!DMAF_RASTER!DMAF_BLITTER),DMACON(a5)

[a/b]

Originally Posted by mcgeezer

Personally I find the solution over complicated, why complicate a simple boolean AND with assembler directives unnecessarily?
...

Because when you decide that 4 bits is not enough and you have to go change your code it's easier to change a single EQU from 4 to say 5, and the rest is automagically calculated, than figure out which instructions contain related hardcoded stuff, be it $fff0, $000f, etc.
And when I do hardcode something related to a 'global parameter', the reason typically being optimizations, I add an assert (simple macro) that stop the assember and I don't have to deal with the code exploding during run-time.