Rearrange the segments a bit

* put TABLES in the low memory, before the bank switch window
* reserve bank switch window
* put rest of the code after that and before the framebuffer

so TABLES lives just before $4000
and MAIN lives in $8000-$bfff

could split some more code and/or data into low mem

and/or move the tables not used in extended memory mode into
the bank switch window so they take no address space on XE
or expanded memory machines

.gitignore

@@ -1,3 +1,4 @@
 *.o
 *.xex
+tables.s
 .DS_Store

Makefile

@@ -2,13 +2,21 @@
 
 all : mandel.xex
 
-%.xex : %.o
-	ld65 -C atari-asm-xex.cfg -o $@ $<
+mandel.xex : mandel.o mandel-core.o tables.o atari-xex.cfg
+	ld65 -C ./atari-xex.cfg --mapfile mandel.map -o $@ mandel.o mandel-core.o tables.o atari.lib
+
+mandel.s : mandel.c mandel.h
+	cc65 -o $@ mandel.c
 
 %.o : %.s
 	ca65 -o $@ $<
 
+tables.s : tables.js
+	node tables.js > tables.s
+
 clean :
+	rm -f tables.s
+	rm -f mandel.s
 	rm -f *.o
 	rm -f *.xex
-
+	rm -f mandel.map

atari-asm-xex.cfg

@@ -0,0 +1,28 @@
+FEATURES {
+    STARTADDRESS: default = $2E00;
+}
+SYMBOLS {
+    __STARTADDRESS__: type = export, value = %S;
+}
+MEMORY {
+    ZP:      file = "", define = yes, start = $0082, size = $007E;
+    MAIN:    file = %O, define = yes, start = %S,    size = $4000 - %S;
+    # Keep $4000-7fff clear for expanded RAM access window
+    TABLES:  file = %O, define = yes, start = $8000, size = $a000 - $8000;
+    # Keep $a000-$bfff clear for BASIC cartridge
+}
+FILES {
+    %O: format = atari;
+}
+FORMATS {
+    atari: runad = start;
+}
+SEGMENTS {
+    ZEROPAGE: load = ZP,      type = zp,  optional = yes;
+    EXTZP:    load = ZP,      type = zp,  optional = yes; # to enable modules to be able to link to C and assembler programs
+    CODE:     load = MAIN,    type = rw,                  define = yes;
+    RODATA:   load = MAIN,    type = ro   optional = yes;
+    DATA:     load = MAIN,    type = rw   optional = yes;
+    BSS:      load = MAIN,    type = bss, optional = yes, define = yes;
+    TABLES:   load = TABLES,  type = ro,  optional = yes, align = 256;
+}

atari-xex.cfg

@@ -0,0 +1,69 @@
+# Sample linker configuration for C programs using the Atari binary file support.
+# Use with: cl65 -tatari -Catari-xex.cfg prog.c -o prog.xex
+FEATURES {
+    STARTADDRESS: default = $8000;
+}
+SYMBOLS {
+    __SYSTEM_CHECK__:    type = import;  # force inclusion of "system check" load chunk
+    __STACKSIZE__:       type = weak, value = $0800; # 2k stack
+    __STARTADDRESS__:    type = export, value = %S;
+    __RESERVED_MEMORY__: type = weak, value = $0000;
+    __SYSCHKHDR__:       type = export, value = 0; # Disable system check header
+    __SYSCHKTRL__:       type = export, value = 0; # Disable system check trailer
+    __TABLESEG_START__:    type = weak, value = $2E00 + $0300;
+    __TABLESEG_SIZE__:     type = weak, value = 6 * $100;
+    __BANKSY_START__:  type = weak, value = $4000;
+    __BANKSY_SIZE__:   type = weak, value = $4000;
+    __FRAMEBUFFER_START__: type = weak, value = $A000;
+}
+MEMORY {
+# Note -- $80 and $81 (LOMEM) appear to be reserved in ZP.
+    ZP:         file = "", define = yes, start = $0082, size = $007E;
+# "system check" load chunk
+    SYSCHKCHNK: file = %O,               start = $2E00, size = $0300;
+# Note $a000-$bfff is against the BASIC cartridge, may require booting with OPTION.
+    TABLES:     file = %O, define = yes, start = __TABLESEG_START__, size = __TABLESEG_SIZE__;
+# We reserve $4000-7fff for the bank-switch window.
+# In theory we could keep data and code here that we only use on 48k/64k systems.
+    BANKSWITCH: file = "", define = yes, start = __BANKSY_START__, size = __BANKSY_SIZE__;
+# "main program" load chunk
+    MAIN:       file = %O, define = yes, start = %S, size = __FRAMEBUFFER_START__ - __STACKSIZE__ - __RESERVED_MEMORY__ - %S;
+}
+FILES {
+    %O: format = atari;
+}
+FORMATS {
+    atari: runad = start,
+           initad = SYSCHKCHNK: __SYSTEM_CHECK__;
+}
+SEGMENTS {
+    ZEROPAGE:  load = ZP,         type = zp;
+    EXTZP:     load = ZP,         type = zp,                optional = yes;
+    SYSCHK:    load = SYSCHKCHNK, type = rw,  define = yes, optional = yes;
+    TABLES:    load = TABLES,     type = ro,  optional = yes, align = 256;
+    BANKSWICH: load = BANKSWITCH, type = ro,  optional = yes;
+    STARTUP:   load = MAIN,       type = ro,  define = yes;
+    LOWBSS:    load = MAIN,       type = rw,                optional = yes;  # not zero initialized
+    LOWCODE:   load = MAIN,       type = ro,  define = yes, optional = yes;
+    ONCE:      load = MAIN,       type = ro,                optional = yes;
+    CODE:      load = MAIN,       type = ro,  define = yes;
+    RODATA:    load = MAIN,       type = ro;
+    DATA:      load = MAIN,       type = rw;
+    INIT:      load = MAIN,       type = rw,                optional = yes;
+    BSS:       load = MAIN,       type = bss, define = yes;
+}
+FEATURES {
+    CONDES: type    = constructor,
+            label   = __CONSTRUCTOR_TABLE__,
+            count   = __CONSTRUCTOR_COUNT__,
+            segment = ONCE;
+    CONDES: type    = destructor,
+            label   = __DESTRUCTOR_TABLE__,
+            count   = __DESTRUCTOR_COUNT__,
+            segment = RODATA;
+    CONDES: type    = interruptor,
+            label   = __INTERRUPTOR_TABLE__,
+            count   = __INTERRUPTOR_COUNT__,
+            segment = RODATA,
+            import  = __CALLIRQ__;
+}

mandel.c

@@ -0,0 +1,15 @@
+/**
+ * The UI and I/O wrapper for the Mandelbrot runner, in C.
+ *
+ * For the moment *all* logic is in mandel-core.s, I'm just
+ * trying to get this to run within a cc65 environment.
+ * Eventually just the inner loop fun will live in there.
+ */
+
+#include <stdlib.h>
+#include <stdio.h>
+#include "mandel.h"
+
+void main(void) {
+    mandel_start();
+}

mandel.h

@@ -0,0 +1,4 @@
+#include <inttypes.h>
+
+// From mandel-core.s:
+extern void mandel_start(void);

readme.md

@@ -14,32 +14,37 @@ Non-goals:
 
 Enjoy! I'll probably work on this off and on for the next few weeks until I've got it producing fractals.
 
--- brooke, january 2023 - february 2024
+-- brooke, january 2023 - december 2024
 
 ## Current state
 
-Basic rendering is functional, but no interactive behavior (zoom/pan) or benchmarking is done yet.
+Basic rendering is functional, with interactive zoom/pan (+/-/arrows) and 6 preset viewports via the number keys.
 
-The 16-bit signed integer multiplication works; it 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.
+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.
 
-The main loop is a basic add-and-shift, using 16-bit adds which requires flipping the sign of negative inputs (otherwise you'd have to add all those sign-extension bits). Runs in 470-780 cycles depending on input.
+* 16-bit multiplies are decomposed into 4 8-bit unsigned multiplies and some addition
+* an optimized case for squares uses a table of 8-bit squares to reduce the number of 8-bit multiplication sub-ops
+* 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.
 
 The pixels are run in a progressive layout to get the basic shape on screen faster.
 
-## Next steps
+There is a running counter of ms/px using the vertical blank interrupts as a timer, used to track our progress. :D
 
-Add a running counter of ms/px using the vertical blank interrupts as a timer. This'll show how further work improves it!
+There's a check for cycles in (zx,zy) output when in the 'lake'; if values repeat, they cannot escape. This is a big time saver in fractint.
 
-Check for cycles in (zx,zy) output when in the 'lake'; if values repeat, they cannot escape. This is a big time saver in fractint.
-
-I may be able to do a faster multiply using tables of squares for 8-bit component multiplication.
+There's some cute color cycling.
 
 ## Deps and build instructions
 
 I'm using `ca65` as a macro assembler, and have a Unix-style `Makefile` for building. Should work fairly easily on Linux and Mac. Might work on "raw" Windows but I use WSL for that.
 
 Currently produces a `.xex` executable, which can be booted up in common Atari emulators and some i/o devices.
+
+## Todo
+
+See ideas in `todo.md`.

tables.js

@@ -0,0 +1,50 @@
+function db(func) {
+    let lines = [];
+    for (let i = 0; i < 256; i += 16) {
+        let items = [];
+        for (let j = 0; j < 16; j++) {
+            let x = i + j;
+            items.push(func(x));
+        }
+        lines.push('    .byte ' + items.join(', '));
+    }
+    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_lobyte256
+.export mul_hibyte256
+.export mul_hibyte512
+.export sqr_lobyte
+.export sqr_hibyte
+
+; (i * i + 1) / 2 for the multiplier
+.align 256
+mul_lobyte256:
+${db((i) => squares[i] & 0xff)}
+
+.align 256
+mul_hibyte256:
+${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)}
+
+`);

testme.js

@@ -0,0 +1,41 @@
+// ax = (a + x)2/2 - a2/2 - x2/2 
+
+function half_square(x) {
+    return Math.round(x * x / 2) & 0xffff >>> 0;
+}
+
+function mul8(a, b) {
+    let result = half_square(a + b) & 0xffff;
+    result = (result - half_square(a)) & 0xffff;
+    result = (result - half_square(b)) & 0xffff;
+    result = (result + (b & a & 1)) & 0xffff;
+    return result >>> 0;
+}
+
+function mul16(a, b) {
+    let ah = (a & 0xff00) >>> 8;
+    let al = (a & 0x00ff) >>> 0;
+    let bh = (b & 0xff00) >>> 8;
+    let bl = (b & 0x00ff) >>> 0;
+    let result = (mul8(al, bl) & 0xffff) >>> 0;
+    result = ((result + (mul8(ah, bl) << 8)) & 0x00ffffff) >>> 0;
+    result = ((result + (mul8(al, bh) << 8)) & 0x01ffffff) >>> 0;
+    result = ((result + (mul8(ah, bh) << 16)) & 0xffffffff) >>> 0;
+    return result;
+}
+
+let max = 65536;
+//let max = 256;
+//let max = 128;
+//let max = 8;
+
+for (let a = 0; a < max; a++) {
+    for (let b = 0; b < max; b++) {
+        let expected = Math.imul(a, b) >>> 0;
+        //let actual = mul8(a, b);
+        let actual = mul16(a, b);
+        if (expected !== actual) {
+            console.log(`wrong! ${a} * ${b} expected ${expected} got ${actual}`);
+        }
+    }
+}

todo.md

@@ -0,0 +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
+
+* maybe clean up the load/layout of the big mul table
+
+* consider alternate lookup tables in the top 16KB under ROM
+
+* y-axis mirror optimization
+
+* extract viewport for display & re-input via keyboard
+
+* fujinet screenshot/viewport uploader