update some timings

Instead of relying solely on the JMP thunks added to
imul16_func and sqr16_func, three call sites within the
mandelbrot iteration function are patched directly to
jsr to the XE versions, saving like 15 cycles per iter

Ok so it's not a lot, but every seconds counts. ;)

with XE code disabled:
1539 us/iter
5m13s

with old XE code:
1417 us/iter
4m48s

with new XE code:
1406 us/iter
4m45s

Makefile

@@ -3,7 +3,7 @@
 all : mandel.xex
 
 mandel.xex : mandel.o tables.o atari-asm-xex.cfg
-	ld65 -C ./atari-asm-xex.cfg -o $@ mandel.o tables.o
+	ld65 -C ./atari-asm-xex.cfg --mapfile mandel.map -o $@ mandel.o tables.o
 
 %.o : %.s
 	ca65 -o $@ $<
@@ -15,4 +15,6 @@ clean :
 	rm -f tables.s
 	rm -f *.o
 	rm -f *.xex
+	rm -f mandel.map
+
 

readme.md

@@ -18,7 +18,7 @@ Enjoy! I'll probably work on this off and on for the next few weeks until I've g
 
 ## Current state
 
-Basic rendering is functional, with interactive zoom/pan (+/-/arrows) and 4 preset viewports via the number keys.
+Basic rendering is functional, with interactive zoom/pan (+/-/arrows) and 6 preset viewports via the number keys.
 
 The 16-bit signed integer multiplication takes two 16-bit inputs and emits one 32-bit output in the zero page, using the Atari OS ROM's floating point registers as workspaces. Inputs are clobbered.
 
@@ -27,7 +27,7 @@ The 16-bit signed integer multiplication takes two 16-bit inputs and emits one 3
 * when expanded RAM is available as on 130XE, a 64KB 8-bit multiplication table accelerates the remaining multiplications
 * without expanded RAM, a table of half-squares is used to implement the algorithm from https://everything2.com/title/Fast+6502+multiplication
 
-The mandelbrot calculations are done using 4.12-precision fixed point numbers. It may be possible to squish this down to 3.13.
+The mandelbrot calculations are done using 3.13-precision fixed point numbers with 6.26-precision intermediates.
 
 Iterations are capped at 255.
 
@@ -47,4 +47,4 @@ Currently produces a `.xex` executable, which can be booted up in common Atari e
 
 ## Todo
 
-See ideas in `todo.md`.
+See ideas in `todo.md`.

tables.js

@@ -11,19 +11,40 @@ function db(func) {
     return lines.join('\n');
 }
 
+let squares = [];
+for (let i = 0; i < 512; i++) {
+    squares.push(Math.trunc((i * i + 1) / 2));
+}
+
 console.log(
 `.segment "TABLES"
 
-.export mul_lobyte
+.export mul_lobyte256
-.export mul_hibyte
+.export mul_hibyte256
+.export mul_hibyte512
+.export sqr_lobyte
+.export sqr_hibyte
 
-; (i * i) / 2 for the multiplier
+; (i * i + 1) / 2 for the multiplier
 .align 256
-mul_lobyte:
+mul_lobyte256:
-${db((i) => ((i * i) >> 1) & 0xff)}
+${db((i) => squares[i] & 0xff)}
 
 .align 256
-mul_hibyte:
+mul_hibyte256:
-${db((i) => ((i * i) >> 9) & 0xff)}
+${db((i) => (squares[i] >> 8) & 0xff)}
+
+.align 256
+mul_hibyte512:
+${db((i) => (squares[i + 256] >> 8) & 0xff)}
+
+; (i * i) for the plain squares
+.align 256
+sqr_lobyte:
+${db((i) => (i * i) & 0xff)}
+
+.align 256
+sqr_hibyte:
+${db((i) => ((i * i) >> 8) & 0xff)}
 
 `);

todo.md

@@ -1,15 +1,17 @@
 things to try:
 
+* fix status bar to show elapsed time, per-iter time, per-pixel iter count
+
+* 'turbo' mode disabling graphics in full or part
+
 * patch the entire expanded-ram imul8xe on top of imul8 to avoid the 3-cycle thunk penalty :D
 
-* try 3.13 fixed point instead of 4.12 for more precision
+* maybe clean up the load/layout of the big mul table
-  * can we get away without the extra bit?
+
+* consider alternate lookup tables in the top 16KB under ROM
 
 * y-axis mirror optimization
 
-* 'wide pixels' 2x and 4x for a fuller initial image in the tiered rendering
-  * maybe redo tiering to just 4x4, 2x2, 1x1?
-
 * extract viewport for display & re-input via keyboard
 
 * fujinet screenshot/viewport uploader