term: cache tile attrs in COL_BUF's spare stride bytes (3.5x scroll)

term_write_tilemap's attribute pass ran color_setup - the full Fano
palette walk - for all 360 tiles on every scroll. render_tile already
computes the attr when a tile changes, so store it (COL_BUF + slot*64
+ ACACHE + tcol; the stride's 24 spare bytes were free) and the attr
pass becomes a table read. update_cursor_attr reads the cache too.
term_init now redraws before writing the tilemap so the cache is warm.

Measured entry-to-return with the emulator cycle counter, ANSI test
screen content: term_write_tilemap 723k -> 114k T-cycles (6.3x),
term_scroll 857k -> 248k (3.5x) - a scroll drops from ~12 frames of
guest CPU to ~3.5.

1 files changed, 38 insertions, 28 deletions

diff --git a/src/term.asm b/src/term.asm
index d62310e..aebe7e3 100644
--- a/src/term.asm
+++ b/src/term.asm
@@ -25,6 +25,13 @@ DEF TSTRIDE EQU 64      ; bytes per line-slot in the buffer (power of 2)
 DEF TERM_BUF EQU $D600  ; 18*64 = 1152 B (WRAMX, above the FS caches)
 DEF COL_BUF  EQU $DA80  ; per-cell fg color, parallel to TERM_BUF (18*64 = 1152 B)
 DEF COL_OFF  EQU COL_BUF - TERM_BUF   ; $0480: add to a char addr -> its color addr
+DEF ACACHE   EQU 40     ; attr cache in COL_BUF's spare stride bytes: the tile
+                        ; attribute (palette bits) for [slot][tcol] lives at
+                        ; COL_BUF + slot*64 + ACACHE + tcol (tcol 0..19).
+                        ; render_tile writes it; term_write_tilemap and
+                        ; update_cursor_attr read it - so the tilemap attr
+                        ; pass never re-runs the Fano color math (O(1) lookup
+                        ; per tile instead of color_setup per tile per scroll).
 
 ; terminal state (WRAMX gap $D580-$D5FF)
 DEF wGlyphL   EQU $D580  ; 2
@@ -112,8 +119,8 @@ term_init::
     or e
     jr nz, .clrc
 
-    call term_write_tilemap
-    call term_redraw            ; build all tiles (blank + cursor)
+    call term_redraw            ; build all tiles (also fills the attr cache)
+    call term_write_tilemap     ; ...which the tilemap attr pass reads
 
     ld a, %10010001             ; LCD on, BG on, tile data $8000, map $9800
     ld [rLCDC], a
@@ -149,10 +156,10 @@ term_write_tilemap::
     call .slot20                ; DE = pos, HL = tilemap addr, wTMSlot = slot
     ld b, TCOLS/2
 .a_col
-    ; attribute = palette(from this tile's two cell colors) | bank bit(pos hi).
+    ; attribute = cached palette bits (render_tile keeps the cache fresh)
+    ; | bank bit(pos hi). No color math here - just a table read.
     push hl                     ; save tilemap write ptr
     push de                     ; save pos
-    ; color addr = COL_BUF + slot*64 + tcol*2 ; tcol = (TCOLS/2) - b
     ld a, [wTMSlot]
     ld l, a
     ld h, 0
@@ -162,31 +169,21 @@ term_write_tilemap::
     add hl, hl
     add hl, hl
     add hl, hl                  ; slot*64
+    ld de, COL_BUF + ACACHE
+    add hl, de
     ld a, TCOLS/2
-    sub b
-    add a                       ; tcol*2
+    sub b                       ; tcol
     add l
     ld l, a
     ld a, h
     adc 0
     ld h, a
-    ld de, COL_BUF
-    add hl, de
-    ld a, [hl+]
-    ld d, a                     ; colL
-    ld a, [hl]
-    ld e, a                     ; colR
-    call color_setup            ; -> wPalAttr (preserves B, C)
+    ld a, [hl]                  ; A = cached attr (palette bits)
     pop de                      ; restore pos
     pop hl                      ; restore tilemap write ptr
-    ld a, d                     ; HIGH(pos): 0 or 1 -> VRAM bank bit
-    or a
-    jr z, .a0
-    ld a, [wPalAttr]
+    bit 0, d                    ; HIGH(pos): 0 or 1 -> VRAM bank bit
+    jr z, .aw
     or $08                      ; VRAM bank 1 tile
-    jr .aw
-.a0
-    ld a, [wPalAttr]
 .aw
     ld [hl+], a
     inc de
@@ -549,7 +546,22 @@ render_tile:
     ld e, a                     ; E = colR
     ld a, [hl]
     ld d, a                     ; D = colL
+    push hl                     ; &colL = COL_BUF + slot*64 + tcol*2
     call color_setup
+    ; refresh the attr cache: cache[slot][tcol] = wPalAttr.
+    ; addr = &colL - tcol + ACACHE  (tcol*2 back off, tcol forward)
+    pop hl
+    ld a, [wRT_tcol]
+    ld e, a
+    ld a, ACACHE
+    sub e                       ; ACACHE - tcol (always positive: 21..40)
+    add l
+    ld l, a
+    ld a, h
+    adc 0
+    ld h, a
+    ld a, [wPalAttr]
+    ld [hl], a
     ; cursor mask
     xor a
     ld [wCurMask], a
@@ -638,7 +650,8 @@ update_cursor_attr:
     ld h, a
     ld a, [hl]
     ld [wTMSlot], a
-    ; color addr = COL_BUF + slot*64 + (wCurCol & $FE)
+    ; attr = cache[slot][wCurCol/2] - the cursor tile was just rendered
+    ; (render_cursor_tile falls through to here), so the cache is fresh.
     ld l, a
     ld h, 0
     add hl, hl
@@ -647,20 +660,17 @@ update_cursor_attr:
     add hl, hl
     add hl, hl
     add hl, hl                  ; slot*64
+    ld de, COL_BUF + ACACHE
+    add hl, de
     ld a, [wCurCol]
-    and $FE                     ; left cell of the tile
+    srl a                       ; tile col
     add l
     ld l, a
     ld a, h
     adc 0
     ld h, a
-    ld de, COL_BUF
-    add hl, de
-    ld a, [hl+]
-    ld d, a                     ; colL
     ld a, [hl]
-    ld e, a                     ; colR
-    call color_setup            ; -> wPalAttr (masks recomputed, harmless)
+    ld [wPalAttr], a            ; downstream bank-bit code reads this
     ; bank bit: pos = slot*20 + (wCurCol/2)
     ld a, [wTMSlot]
     ld l, a