| |
Rudi
Account closed
Registered: May 2010
Posts: 125 |
Fast way to rotate a char?
Im not talking about rol or ror, but swap bits so that they are rotated 90 degrees:
Example:
a char (and the bits can be random):
10110010 byte 1..
11010110 byte 2.. etc..
00111001
01010110
11011010
10110101
00110011
10110100 after "rotation" (rows and columns are swapped):
11001101
01011000
10100111
11111111
00101000
01010101
11011010
00100110 is it possible to use lookup tables for this or would that lookup table be too big?
or other lookuptable for getting and setting bits?
-Rudi |
|
... 105 posts hidden. Click here to view all posts....
|
| |
Oswald
Registered: Apr 2002
Posts: 5086 |
Quote: Oswald: Why 3 times and not 8?
coz you rotate in 2 buffers a,b:
a b
a b
a b
but if shifting is bigger than 3 you can just shift bits to opposite dir and then swap a,b (on screen). 3 is just a wild guess maybe 4 is the max, it was a long time ago. |
| |
Rudi
Account closed
Registered: May 2010
Posts: 125 |
There is some interesting cyclic dynamic going on when looking at a bit-swapping method for 8x8 bits:
these figures shows that the outer layer is rotated 8 times first, then the next layer is rotated 6 times and so forth:xxxxxxxx ........ ........ ........
x......x .xxxxxx. ........ ........
x......x .x....x. ..xxxx.. ........
x......x .x....x. ..x..x.. ...xx...
x......x .x....x. ..x..x.. ...xx...
x......x .x....x. ..xxxx.. ........
x......x .xxxxxx. ........ ........
xxxxxxxx ........ ........ ........ in this example one start with c0r0 and swap that with c7r0, then c1r0->c7r1, c2r0->c7r2 etc. so it goes around like a circle would. when that process is finished, the next inner layer does the same thing. until all four layers are done. and the result is a 90 deg rotation. its just one other way of looking at it. but still, a lookup-table for this? its difficult for me to see how. (im just giving out a few ideas for trying out other approach maybe). |
| |
ChristopherJam
Registered: Aug 2004
Posts: 1408 |
Problem with bitshifter is it only deals with one bit at a time.
c2p exchanges four bitpairs in only 20-30 cycles. The expensive part is moving bits to matching positions within the two bytes they're being exchanged between... |
| |
Digger
Registered: Mar 2005
Posts: 427 |
Nice read. But what fx do you have in mind Rudi? |
| |
Rudi
Account closed
Registered: May 2010
Posts: 125 |
Quote: Nice read. But what fx do you have in mind Rudi?
My initial thinking was to make a sine-scroller, where the sinus-transform is done in chunky and then transformed using c2p. Of course the c2p-routine has to be fast enough for this. It would also bring out some new ideas for other vertical effects (that are first done with horisontal tricks, and then transformed using c2p). Unfortunately there seem to be no _really_ fast c2p routine.
Im still figuring out how that 4x4, 2x2 plus 1x1 rotator works. or that is; looking at every detail of the bit swaps.
Edit: hires-sinescroller. |
| |
Frantic
Registered: Mar 2003
Posts: 1647 |
It is not really what you specified in your first post, but just a thought that may be worth considering: Would it be okay to represent the original char in some different way? If so, there might (possibly) be ways to represent that char that would make it easier to "rotate" it to +90˚ angle although at the cost that you may then also need to do some operations on the data to be able to "read out" the normal 0˚ angle.
Just thinking aloud here. Trying to rethink the problem rather than the solution, so to speak. Not sure if it would be allowed. |
| |
Rudi
Account closed
Registered: May 2010
Posts: 125 |
Frantic? Sorry mate, but my first post is what i need. the figures dont lie :) |
| |
Copyfault
Registered: Dec 2001
Posts: 475 |
Interesting Thread :)
Sticking to the figures of the initial post the goal is to have a char mirror* routine with a cycle count as low as possbile for 8x8-Chars HiRes with arbitrary bit patterns [*mirror in the sense of mirroring along the diagonal line from bitpos (0,0) to (7,7)].
Maybe it's possible to weaken the requirement of _arbitrary_ bitpatterns... I'm thinking about using a charset with chars $00-$7f for the "given" bitpatterns and $80-$ff for the corresponding "mirrored" patterns. This way, mirroring would only be a question of setting the msb of the corresponding bytes in the screenram.
Guess in order to have a chance to make this approach work you'd need several charsets spread along the screen - and this still requires that appropriate slices of the screen will have at most 128 different bitpatterns.
Another downside: expanding this approach to 90° rotations will lower the number of different bitpatterns even further to 64.
But maybe this thought is of any help; it just came up while reading so I wanted to share it. |
| |
ChristopherJam
Registered: Aug 2004
Posts: 1408 |
For things like sine scrollers, there may be value in doing a partial rotation; only takes 240 cycles per char to reshape eight bytes into
; 00224466
; 00224466
; 00224466
; 00224466
; 11335577
; 11335577
; 11335577
; 11335577
(digit is index of source byte)
using something like
ldx s+0
ldy s+1 ; 8 cycles priming the xy line cache. Need to do this on row 0 and row 4
lda t00,x
ora t01,y ; 8
ldx s+4
ldy s+6 ; 8
ora t02,x
ora t03,y ; 8
sta dst+0 ; 4 28 cycles for this block; one of these per dest byte
lda t12,x
ora t13,y
ldx s+0
ldy s+2
ora t10,x
ora t11,y
sta dst+1
etc |
| |
Krill
Registered: Apr 2002
Posts: 2969 |
Rudi: So you want to rotate chars by arbitrary angles, not just swap X and Y (90°)? What's the problem with Frantic's suggestion of pre-calculating another representation of the original characters to optimise run-time rendering of the rotated characters, thus trading memory for speed? |
| Previous - 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 - Next |
