Raster Interrupt Speed Tricks (LowRes NX)

nathanielbabiak 2020-12-04 21:50 (Edited)

Raster Interrupt Speed Tricks - Disclaimer

Caution! Syntactic (within-same-line) speed optimizations are discussed frequently on this console. But, there's a 99% chance algorithmic (not syntactic) improvements will provide an order-of-magnitude speed improvement vs. doing any of these! These are only recommended for use within raster interrupts (not even VBL interrupts require them).

Raster Interrupt Speed Tricks

Non-zero values (int and float) evaluate to true, these are smaller and faster, for example with 12.34 (or any variable replacing 12.34):

Replace IF X<>0 THEN with IF X THEN
Replace IF X<>0 AND Y<>0 THEN with IF X*Y THEN
Replace IF X=0 OR Y=0 THEN with IF X*Y=0 THEN
Replace IF X < 0 OR X>= 12.34 THEN with IF INT(X/12.34) THEN
Replace IF X>= 0 AND X< 12.34 THEN with IF INT(X/12.34)=0 THEN

There's two non-zero evaluation tricks if you're sure X and Y will always have matching signs:

Replace IF X<>0 OR Y<>0 THEN with IF X+Y THEN
Replace IF X=0 AND Y=0 THEN with IF X+Y=0 THEN

There's a few more non-zero evaluation tricks when X and, for example 1234 (or any variable replacing 1234), are integers:

Replace IF (X AND %001)=%001 THEN with IF X AND %001 THEN
Replace IF (X AND %010)=%010 THEN with IF X AND %010 THEN
Replace IF (X AND %100)=%100 THEN with IF X AND %100 THEN
Replace IF X<=-1234 OR X>=1234 THEN with IF X\1234 THEN
Replace IF X> -1234 AND X<1234 THEN with IF X\1234=0 THEN

You can store evaluations per-raster-line in an array to save a single clock cycle:

Replace IF RASTER=64 THEN with IF LOOKUP(RASTER) THEN (after running DIM GLOBAL LOOKUP(127) and LOOKUP(64)=1 elsewhere in your code)

If you're only performing one instruction occasionally per-raster-line, note that some instructions perform a graceful exit when executed with unrealistic arguments (for example COPY SRC, 0 TO DST), so the above tip can be shortened further (depending on the number of nonzero LOOKUP() entries and your specific instruction):

Replace IF LOOKUP(RASTER) THEN COPY SRC, 40 TO DST with COPY SRC, LOOKUP(RASTER) TO DST (after initializing the values of the LOOKUP() array elsewhere in your code)

If you're performing various raster tricks on continuous sequences of RASTER scanlines, don't use either switch-case structure described below (they're too slow), instead change raster subprograms within the raster interrupt itself. Try this:

SUB RASTER1 \\ [first trick, upper screen] \\ IF LOOKUP(RASTER) THEN ON RASTER CALL RASTER2 \\ END SUB
SUB RASTER2 \\ [second trick, middle screen] \\ IF LOOKUP(RASTER) THEN ON RASTER CALL RASTER3 \\ END SUB
SUB RASTER3 \\ [third trick, bottom screen] \\ END SUB
ON VBL CALL RASTER_RESET
SUB RASTER_RESET \\ ON RASTER CALL RASTER1 \\ END SUB

Keywords from C

This switch-case structure requires few tokens, but it's slow:

IF X=0 THEN
  CALL X0
ELSE IF X=1 THEN
  CALL X1
ELSE IF X=2 THEN
  CALL X2
ELSE IF X=3 THEN
  CALL X3
ELSE IF X=4 THEN
  CALL X4
ELSE IF X=5 THEN
  CALL X5
ELSE IF X=6 THEN
  CALL X6
ELSE IF X=7 THEN
  CALL X7
END IF

This switch-case structure is much faster, but requires more tokens:

IF X<4 THEN
  IF X<2 THEN
    IF X<1 THEN CALL X0 ELSE CALL X1
  ELSE
    IF X<3 THEN CALL X2 ELSE CALL X3
  END IF
ELSE
  IF X<6 THEN
    IF X<5 THEN CALL X4 ELSE CALL X5
  ELSE
    IF X<7 THEN CALL X6 ELSE CALL X7
  END IF
END IF

Goto

In small projects, using GOTO can be much easier to read than avoiding it, but it's harder to maintain nonlooped code, so avoid GOTO in large projects. In the vast majority of instances you think you need it, the algorithm isn't coherent. There's one instance where GOTO is common: implementing the C keywords continue[(n)] and break[(n)].

Small Declaration

If an algorithm needs more than 3 new variable names, this is smaller than declaring one variable per-line. It's slower only because of overhead, otherwise the same speed O(n).

DECLARE:
DATA 0, 0, 0, 0, 0, 0, 0, 0
LOAD DECLARE
READ Y0, Y1, Y2, Y3, Y4, Y5, Y6, Y7

Single-command if-then statements are smaller (and sometimes faster too) when written on a single line like this:

IF TEST1 THEN CALL FOO
IF TEST1 THEN CALL FOO ELSE CALL BAR
IF TEST1 THEN CALL FOO ELSE IF TEST2 THEN CALL BAR ELSE CALL ZOT

Loops

This loop is small but slow:

FOR X=0 TO 7
  TEXT 0,X,"-"
NEXT X

This unrolled loop is fast but large:

TEXT 0,0,"-"
TEXT 0,1,"-"
TEXT 0,2,"-"
TEXT 0,3,"-"
TEXT 0,4,"-"
TEXT 0,5,"-"
TEXT 0,6,"-"
TEXT 0,7,"-"

Misc. Stuff

X = ASC( CHR$( X ) ) will convert an 8-bit value from unsigned to signed.
Additional ROM space may be mimicked (with sequential access only) using DATA commands, containing strings of hex-values, use VAL("0X"+VALUE$) to retrieve up to three bytes at a time.

Short-Circuit Evaluation

LowRes NX supports a limited format for short-circuit evaluations. If you have evaluations A, B, and C, you can replace:

IF A AND B AND C THEN CALL ALL_TRUE ELSE CALL SOME_FALSE

With:

IF A THEN IF B THEN IF C THEN CALL ALL_TRUE ELSE CALL SOME_FALSE

How To

Raster Interrupt Speed Tricks

Raster Interrupt Speed Tricks - Disclaimer