Revert "remove the sound"

This reverts commit debdd36ce0.

dither4.s

@@ -26,6 +26,7 @@ sample_ptrl = $84
 sample_ptrh = $85
 sample_ptr = sample_ptrl
 scanline = $86
+audiotemp = $87
 frame_counter = $89
 
 ;height = 160
@@ -39,6 +40,9 @@ scanline_max = (lines_per_frame - scanline_offset) / 2
 
 .data
 
+.import audio_samples
+.import audio_samples_end
+
 .import frame1_top
 .import frame1_bottom
 .import frame1_palette1_even
@@ -49,11 +53,41 @@ scanline_max = (lines_per_frame - scanline_offset) / 2
 .import frame1_palette3_odd
 .import displaylist
 
+audio_high_byte:
+    .scope
+        .macro byteseq val
+            .repeat 16
+                .byte val | $10
+            .endrep
+        .endmacro
+        byteseq $0
+        byteseq $1
+        byteseq $2
+        byteseq $3
+        byteseq $4
+        byteseq $5
+        byteseq $7
+        byteseq $8
+        byteseq $9
+        byteseq $a
+        byteseq $b
+        byteseq $c
+        byteseq $d
+        byteseq $e
+        byteseq $f
+    .endscope
+
 .code
 
 .export start
 
 .proc start
+    ; Set up the audio sample buffer
+    lda #.lobyte(audio_samples)
+    sta sample_ptrl
+    lda #.hibyte(audio_samples)
+    sta sample_ptrh
+
     ; Disable display DMA
     lda #$00
     sta DMACTL
@@ -95,6 +129,12 @@ wait_loop:
     ldy VCOUNT    ; 4 cycles
     bne wait_loop ; 2 cycles
 
+    .macro audio_prep
+        ; Y is VCOUNT at entry
+        lda (sample_ptr),y ; 5/6 cyc
+        sta audiotemp      ; 3 cyc
+    .endmacro
+
     .macro inner_scanline frame_offset, line_offset
         ; Y should be VCOUNT at entry
         ; it'll fire on unused lines, but harmlessly
@@ -119,6 +159,51 @@ wait_loop:
         sty COLPF2 ; 4
     .endmacro
 
+    .macro audio_play_raw
+        ;ldy VCOUNT ; set on entry
+        lda (sample_ptr),y ; 5/6 cyc
+        sta AUDC1          ; 4 cyc
+    .endmacro
+
+    .macro audio_play_lo
+        lda audiotemp ; 3 cyc
+        and #$0f ; 2 cyc
+        ora #$10 ; 2 cyc
+        sta AUDC1 ; 4 cyc
+    .endmacro
+
+    .macro audio_play_hi ; 12 cycles
+        ldy audiotemp ; 3 cyc
+        lda audio_high_byte,y ; 5 cyc
+        sta AUDC1 ; 4 cyc
+    .endmacro
+
+    .macro audio_inc
+        ; 22 cycles
+        lda sample_ptrl ; 3 cyc
+        clc             ; 2 cyc
+        adc #131        ; 2 cyc
+        sta sample_ptrl ; 3 cyc
+        lda sample_ptrh ; 3 cyc
+        adc #0          ; 2 cyc
+        sta sample_ptrh ; 3 cyc
+        cmp #.hibyte(audio_samples_end) ; 2 cyc
+        bmi audio_cont                  ; 2 cyc
+
+        sta WSYNC
+
+        ; 10 cycles, optional
+        lda #.lobyte(audio_samples)     ; 2
+        sta sample_ptrl                 ; 3
+        lda #.hibyte(audio_samples)     ; 2
+        sta sample_ptrh                 ; 3
+
+        sta WSYNC
+        ldy VCOUNT             ; 4 cycles
+
+        audio_cont:
+    .endmacro
+
     .macro run_frame frame_offset
     .scope
         ; each scanline is 228 color clocks
@@ -130,12 +215,16 @@ wait_loop:
         inner_scanline frame_offset, 0 ; 23-26 cycles before break, 12 cycles after
 
         ldy scanline   ; 3 cycles
+        audio_play_raw ; 8-10 cycles
+
         inner_scanline frame_offset, 128 ; 23-26 cycles before break, 12 cycles after
 
         ; pair cleanup: 6 cycles
         ldy VCOUNT             ; 4 cycles
         bne each_scanline_pair ; 2 cycles
 
+        audio_inc      ; 22-32 cycles
+
         ; frame cleanup: 11 cycles
         lda frame_counter ; 3 cycles
         eor #1            ; 2 cycles