From ce7e051504f7a83ef6d8ce8ecfa174a9f26690a6 Mon Sep 17 00:00:00 2001
From: Brion Vibber <brion@pobox.com>
Date: Sat, 26 Nov 2022 08:48:01 -0800
Subject: [PATCH] WORKS

---
 dither-image.js | 10 +++---
 dither4.s       | 89 +++++++++++++++++++++++++++++++++++++++++--------
 2 files changed, 81 insertions(+), 18 deletions(-)

diff --git a/dither-image.js b/dither-image.js
index 040ed5e..88bbcbb 100644
--- a/dither-image.js
+++ b/dither-image.js
@@ -663,7 +663,7 @@ function genAssembly(width, height, nbits, lines) {
     let palette1 = new Uint8Array(height);
     let palette2 = new Uint8Array(height);
     let palette3 = new Uint8Array(height);
-    let framebuffer = new Uint8Array(stride * height);
+    let bitmap = new Uint8Array(stride * height);
     for (let y = 0; y < height; y++) {
         palette1[y] = lines[y].palette[1];
         palette2[y] = lines[y].palette[2];
@@ -672,7 +672,7 @@ function genAssembly(width, height, nbits, lines) {
             width,
             nbits,
             lines[y].output,
-            framebuffer.subarray(y * stride, (y + 1) * stride)
+            bitmap.subarray(y * stride, (y + 1) * stride)
         );
     }
 
@@ -680,7 +680,7 @@ function genAssembly(width, height, nbits, lines) {
 .export palette1
 .export palette2
 .export palette3
-.export framebuffer
+.export bitmap
 
 palette1:
 ${byte2byte(palette1)}
@@ -691,8 +691,8 @@ ${byte2byte(palette2)}
 palette3:
 ${byte2byte(palette3)}
 
-framebuffer:
-${byte2byte(framebuffer)}
+bitmap:
+${byte2byte(bitmap)}
 `;
 }
 
diff --git a/dither4.s b/dither4.s
index 58f4d41..ac8f2eb 100644
--- a/dither4.s
+++ b/dither4.s
@@ -12,29 +12,41 @@ WSYNC = $D40A
 VCOUNT = $D40B
 NMIEN = $D40E
 
+framebuffer = $a000
+framebuffer2 = $b000
+
+temp1l = $80
+temp1h = $81
+temp1 = temp1l
+temp2l = $82
+temp2h = $83
+temp2 = temp2l
+
+height = 192
+bytes_per_line = 40
+pages_per_frame = 32
+
 .data
 
 .import palette1
 .import palette2
 .import palette3
-.import framebuffer
+.import bitmap
 
 displaylist:
     ; @fixme resolve the line counts
     .repeat 3
         .byte $70 ; 8 blank lines
     .endrep
-    ; Note the framebuffer cannot cross a 4KiB boundary
-    ; so this is split in half
     ; ANTIC mode e (160px 2bpp, 1 scan line per line)
     .byte $4e
     .addr framebuffer
-    .repeat 95
+    .repeat height / 2 - 1
         .byte $0e
     .endrep
     .byte $4e
-    .addr framebuffer + 96*40
-    .repeat 95
+    .addr framebuffer2
+    .repeat height / 2 - 1
         .byte $0e
     .endrep
     .byte $41 ; jump and blank
@@ -46,19 +58,35 @@ displaylist:
 .export start
 
 .proc start
+    ; Copy the bitmap into our framebuffer
+    lda #.lobyte(bitmap)
+    sta temp1l
+    lda #.hibyte(bitmap)
+    sta temp1h
+    lda #.lobyte(framebuffer)
+    sta temp2h
+    lda #.hibyte(framebuffer)
+    sta temp2h
+    jsr copy_half_frame
+
+    ; Second half of bitmap has to be separately aligned
+    lda #.lobyte(framebuffer2)
+    sta temp2h
+    lda #.hibyte(framebuffer2)
+    sta temp2h
+    jsr copy_half_frame
+
+
     ; Disable display DMA
     lda #$00
     sta SDMCTL
+
     ; Set up the display list
     lda #.lobyte(displaylist)
     sta SDLSTL
     lda #.hibyte(displaylist)
     sta SDLSTH
 
-    ; Turn off interrupts except reset button
-    ;lda #$20
-    ;sta NMIEN
-
     ; Manually wait for first scan line
 wait_vblank:
     lda VCOUNT
@@ -97,14 +125,49 @@ each_scanline:
     pla
     tay
     iny
-    cpy #192
+    cpy height
     bne each_scanline
 
     jmp wait_start    
 
 .endproc
 
-; Loop through scan lines updating the palette
-.proc update_palette
+; temp1 contains source address
+; temp2 contains dest address
+; clobbers a/x/y
+; incremepts temp1 and temp2 to the end of their regions
+.proc copy_half_frame
+    ldx #0
+
+copy_loop_lines:
+    ldy #00
+
+copy_loop_bytes:
+    lda (temp1),y
+    sta (temp2),y
+    iny
+    cpy #bytes_per_line
+    bne copy_loop_bytes
+
+    clc
+    lda temp1l
+    adc #bytes_per_line
+    sta temp1l
+    lda temp1h
+    adc #00
+    sta temp1h
+
+    clc
+    lda temp2l
+    adc #bytes_per_line
+    sta temp2l
+    lda temp2h
+    adc #00
+    sta temp2h
+
+    inx
+    cpx #(height / 2)
+    bne copy_loop_lines
+
     rts
 .endproc