From 7f2f61f06215fbd6d2f3c6d5f3ae0317c0137a35 Mon Sep 17 00:00:00 2001
From: Forest Belton <forest@homolo.gy>
Date: Tue, 13 Jul 2021 00:24:31 -0400
Subject: [PATCH] Optimize scroll-up/down routines

In order to calculate the VRAM destination for these routines, we
need to compute the row pointer into map RAM:

```
HL = _SCRN0 + 32 * (SCY/8 - 2)
```

The obvious approach of shifting `SCY/8 - 2` right 5 times won't work
since overflow is possible. The first implementation worked by adding
`SCY/8-2` to `_SCRN0` 32 times in a loop.

Now, `SCY/8 - 2` is loaded into a 16-bit register and a 16-bit shift
macro is expanded 5 times before being added to HL.

The first approach took 802 * 4 cycles. This has been reduced to 25 * 4
cycles.
---
 Makefile     |  2 +-
 inc/util.inc | 15 +++++++++++++++
 src/map.s    | 37 ++++++++++++-------------------------
 3 files changed, 28 insertions(+), 26 deletions(-)

diff --git a/Makefile b/Makefile
index 3530a25..3a3ea7f 100644
--- a/Makefile
+++ b/Makefile
@@ -20,7 +20,7 @@ OFILES := $(SFILES:%.s=%.o) $(MAP_S:%.s=%.o)
 is.gb is.gb.sym: $(OFILES)
 	@echo "[LINK]     is.gb"
 	@rgblink -o $@ -n $@.sym $(OFILES)
-	@rgbfix -v $@
+	@rgbfix -v -p 0 $@
 
 $(OFILES): $(MAP_INC)
 $(OFILES): $(SPRITE_2BPP)
diff --git a/inc/util.inc b/inc/util.inc
index 4eebde5..40ecf05 100644
--- a/inc/util.inc
+++ b/inc/util.inc
@@ -49,3 +49,18 @@ MACRO LD16
     ld HIGH(\1), HIGH(\2)
     ld LOW(\1), LOW(\2)
 ENDM
+
+MACRO SLA16
+    sla LOW(\1)
+    rl HIGH(\1)
+ENDM
+
+; Performs SLA on a register N times
+; \1 8-bit register
+; \2 N, a value 1-8
+MACRO slan
+    ASSERT 1 <= \2 && \2 <= 8
+    REPT \2
+        sla \1
+    ENDR
+ENDM
diff --git a/src/map.s b/src/map.s
index fbf655a..2545302 100644
--- a/src/map.s
+++ b/src/map.s
@@ -115,20 +115,12 @@ Map_Scroll::
     ld c, a
 
     ; HL = _SCRN0 + 32 * (SCY/8 - 2)
-.find_vram_ptr:
     ld a, [rSCY]
     sub 16
     srl a
     srl a
     srl a
-    ld d, a
-    ld e, 32
-    ld hl, _SCRN0
-.0:
-    ld a, d
-    ADD16 hl
-    dec e
-    jr nz, .0
+    call get_row_ptr
 
     call enqueue_row_write
 
@@ -151,8 +143,7 @@ Map_Scroll::
     srl a
     srl a
     dec a
-    ld e, a
-    call compute_vram_ptr
+    call get_row_ptr
 
     ; B = CAMERA_X - SCX/8
     ld a, [rSCX]
@@ -297,22 +288,18 @@ Map_Update::
     ret
 
 ; Computes the offset into map RAM
-; @param e The map RAM y-coordinate
+; @param a The map RAM y-coordinate
 ; @return hl Pointer into map RAM
-compute_vram_ptr:
-    push bc
+get_row_ptr:
+    push de
+    ld d, 0
+    ld e, a
+    REPT 5
+        SLA16 de
+    ENDR
     ld hl, _SCRN0
-    ld b, 0
-    ld c, SCRN_VX_B
-    ld a, e
-.loop:
-    or a
-    jr z, .done
-    add hl, bc
-    dec a
-    jr .loop
-.done:
-    pop bc
+    add hl, de
+    pop de
     ret
 
 ; Write a row of map data into row buffer