From f6d4c2f2ee940cadff5c5792c789ba86b9df7d4d Mon Sep 17 00:00:00 2001
From: John Tsiombikas <nuclear@member.fsf.org>
Date: Tue, 13 Apr 2021 13:09:44 +0300
Subject: [PATCH] 2vbl tunnel! 30fps

---
 .gitignore    |    1 +
 Makefile      |    2 +-
 src/gamescr.c |   92 ++++++++++++++++++++++++++++++++-------------------------
 src/util.c    |    8 ++---
 4 files changed, 58 insertions(+), 45 deletions(-)

diff --git a/.gitignore b/.gitignore
index bf8108b..226cf4e 100644
--- a/.gitignore
+++ b/.gitignore
@@ -16,3 +16,4 @@ data
 *.tar.gz
 *.zip
 pushdata
+link.map
diff --git a/Makefile b/Makefile
index a83188f..301519d 100644
--- a/Makefile
+++ b/Makefile
@@ -41,7 +41,7 @@ $(bin): $(elf)
 	gbafix -r0 $(bin)
 
 $(elf): data/snd.h $(obj) $(libs)
-	$(CC) -o $(elf) $(obj) -specs=gba.specs $(LDFLAGS)
+	$(CC) -o $(elf) $(obj) -specs=gba.specs -Wl,-Map,link.map $(LDFLAGS)
 
 -include $(dep)
 
diff --git a/src/gamescr.c b/src/gamescr.c
index 1b11c25..70991e8 100644
--- a/src/gamescr.c
+++ b/src/gamescr.c
@@ -7,14 +7,17 @@
 #include "intr.h"
 #include "debug.h"
 
+static void draw_tunnel(void);
+
+static int nframes, backbuf;
+static uint16_t *vram[] = { (uint16_t*)VRAM_LFB_FB0_ADDR, (uint16_t*)VRAM_LFB_FB1_ADDR };
+static unsigned char *tex;
+
 void gamescr(void)
 {
-	int i, j, tx, ty, angle, depth, nframes, backbuf, zoffs;
-	static uint16_t *vram[] = { (uint16_t*)VRAM_LFB_FB0_ADDR, (uint16_t*)VRAM_LFB_FB1_ADDR };
-	uint16_t *cdst, *top, *bot;
+	int i;
+	uint16_t *cdst;
 	unsigned char *csrc;
-	uint32_t tun, *tunptr;
-	unsigned char *tex;
 
 	REG_DISPCNT = 4 | DISPCNT_BG2 | DISPCNT_FB1;
 
@@ -37,41 +40,7 @@ void gamescr(void)
 	for(;;) {
 		backbuf = ++nframes & 1;
 
-		zoffs = nframes << 1;
-
-		top = vram[backbuf];
-		bot = vram[backbuf] + 159 * 240 / 2;
-		tunptr = tunmap;
-		for(i=0; i<80; i++) {
-			top++;
-			bot++;
-			tunptr++;
-			for(j=1; j<240/2; j++) {
-				uint16_t pptop, ppbot;
-
-				tun = *tunptr++;
-
-				angle = tun & 0xff;
-				depth = (tun >> 8) & 0xff;
-				tx = ((angle >> 1) + zoffs) & 0x1f;
-				ty = ((depth >> 1) + zoffs) & 0x1f;
-				pptop = tex[(ty << 5) + tx];
-				tx = ((angle >> 1) - zoffs) & 0x1f;
-				ppbot = tex[(ty << 5) + tx];
-
-				angle = (tun >> 16) & 0xff;
-				depth = (tun >> 24) & 0xff;
-				tx = ((angle >> 1) + zoffs) & 0x1f;
-				ty = ((depth >> 1) + zoffs) & 0x1f;
-				pptop |= (uint16_t)tex[(ty << 5) + tx] << 8;
-				tx = ((angle >> 1) - zoffs) & 0x1f;
-				ppbot |= (uint16_t)tex[(ty << 5) + tx] << 8;
-
-				*top++ = pptop;
-				*bot++ = ppbot;
-			}
-			bot -= 240;
-		}
+		draw_tunnel();
 
 		vblperf_end();
 		wait_vblank();
@@ -79,3 +48,46 @@ void gamescr(void)
 		vblperf_begin();
 	}
 }
+
+__attribute__((noinline, target("arm"), section(".iwram")))
+static void draw_tunnel(void)
+{
+	int i, j, tx, ty, angle, depth, zoffs;
+	uint16_t pptop, ppbot;
+	uint16_t *top, *bot;
+	uint32_t tun, *tunptr;
+
+	zoffs = nframes;
+
+	top = vram[backbuf];
+	bot = vram[backbuf] + 159 * 240 / 2;
+	tunptr = tunmap;
+	for(i=0; i<80; i++) {
+		top++;
+		bot++;
+		tunptr++;
+		for(j=1; j<240/2; j++) {
+			tun = *tunptr++;
+
+			angle = tun & 0xff;
+			depth = (tun >> 8) & 0xff;
+			tx = ((angle >> 1) + zoffs) & 0x1f;
+			ty = ((depth >> 1) + zoffs) & 0x1f;
+			pptop = tex[(ty << 5) + tx];
+			tx = ((angle >> 1) - zoffs) & 0x1f;
+			ppbot = tex[(ty << 5) + tx];
+
+			angle = (tun >> 16) & 0xff;
+			depth = (tun >> 24) & 0xff;
+			tx = ((angle >> 1) + zoffs) & 0x1f;
+			ty = ((depth >> 1) + zoffs) & 0x1f;
+			pptop |= (uint16_t)tex[(ty << 5) + tx] << 8;
+			tx = ((angle >> 1) - zoffs) & 0x1f;
+			ppbot |= (uint16_t)tex[(ty << 5) + tx] << 8;
+
+			*top++ = pptop;
+			*bot++ = ppbot;
+		}
+		bot -= 240;
+	}
+}
diff --git a/src/util.c b/src/util.c
index 292f67c..9bba42f 100644
--- a/src/util.c
+++ b/src/util.c
@@ -1,13 +1,13 @@
 #include "util.h"
 #include "debug.h"
 
-extern char __data_end__;
-static char *top = &__data_end__;
+extern char __iheap_start;
+static char *top = &__iheap_start;
 
 int iwram_brk(void *addr)
 {
-	if((char*)addr < &__data_end__) {
-		addr = &__data_end__;
+	if((char*)addr < &__iheap_start) {
+		addr = &__iheap_start;
 	}
 	if(addr > get_sp()) {
 		/*return -1;*/
-- 
1.7.10.4