Ported libc from pokeruby

Makefile

@@ -59,8 +59,7 @@ asm/libmks4agb.o \
 asm/libagbbackup.o \
 asm/librtc.o \
 asm/librfu.o \
-asm/libagbsyscall.o \
+asm/libagbsyscall.o
-asm/libc.o
 
 DATA_ASM_OBJS := data/data2.o data/anim_mon_front_pics.o \
 data/graphics.o data/unknown_serial_data.o data/multiboot_berry_glitch_fix.o \
@@ -102,6 +101,9 @@ include graphics_file_rules.mk
 %.lz: % ; $(GFX) $< $@
 %.rl: % ; $(GFX) $< $@
 
+src/libc.o: CC1 := tools/agbcc/bin/old_agbcc
+src/libc.o: CFLAGS := -O2
+
 src/siirtc.o: CFLAGS := -mthumb-interwork
 
 src/agb_flash.o: CFLAGS := -O -mthumb-interwork

asm/libc.s

@@ -1,213 +0,0 @@
-	.include "asm/macros.s"
-
-	.syntax unified
-
-	.text
-
-	thumb_func_start memcpy
-@ void *memcpy(void *dest, void *src, int size)
-memcpy: @ 82E93D4
-	push {r4,r5,lr}
-	adds r5, r0, 0
-	adds r4, r5, 0
-	adds r3, r1, 0
-	cmp r2, 0xF
-	bls _082E9414
-	adds r0, r3, 0
-	orrs r0, r5
-	movs r1, 0x3
-	ands r0, r1
-	cmp r0, 0
-	bne _082E9414
-	adds r1, r5, 0
-_082E93EE:
-	ldm r3!, {r0}
-	stm r1!, {r0}
-	ldm r3!, {r0}
-	stm r1!, {r0}
-	ldm r3!, {r0}
-	stm r1!, {r0}
-	ldm r3!, {r0}
-	stm r1!, {r0}
-	subs r2, 0x10
-	cmp r2, 0xF
-	bhi _082E93EE
-	cmp r2, 0x3
-	bls _082E9412
-_082E9408:
-	ldm r3!, {r0}
-	stm r1!, {r0}
-	subs r2, 0x4
-	cmp r2, 0x3
-	bhi _082E9408
-_082E9412:
-	adds r4, r1, 0
-_082E9414:
-	subs r2, 0x1
-	movs r0, 0x1
-	negs r0, r0
-	cmp r2, r0
-	beq _082E942E
-	adds r1, r0, 0
-_082E9420:
-	ldrb r0, [r3]
-	strb r0, [r4]
-	adds r3, 0x1
-	adds r4, 0x1
-	subs r2, 0x1
-	cmp r2, r1
-	bne _082E9420
-_082E942E:
-	adds r0, r5, 0
-	pop {r4,r5,pc}
-	thumb_func_end memcpy
-
-	thumb_func_start memset
-@ void *memset(void *dest, char c, int size)
-memset: @ 82E9434
-	push {r4,r5,lr}
-	adds r5, r0, 0
-	adds r4, r1, 0
-	adds r3, r5, 0
-	cmp r2, 0x3
-	bls _082E947A
-	movs r0, 0x3
-	ands r0, r5
-	cmp r0, 0
-	bne _082E947A
-	adds r1, r5, 0
-	movs r0, 0xFF
-	ands r4, r0
-	lsls r3, r4, 8
-	orrs r3, r4
-	lsls r0, r3, 16
-	orrs r3, r0
-	cmp r2, 0xF
-	bls _082E946E
-_082E945A:
-	stm r1!, {r3}
-	stm r1!, {r3}
-	stm r1!, {r3}
-	stm r1!, {r3}
-	subs r2, 0x10
-	cmp r2, 0xF
-	bhi _082E945A
-	b _082E946E
-_082E946A:
-	stm r1!, {r3}
-	subs r2, 0x4
-_082E946E:
-	cmp r2, 0x3
-	bhi _082E946A
-	adds r3, r1, 0
-	b _082E947A
-_082E9476:
-	strb r4, [r3]
-	adds r3, 0x1
-_082E947A:
-	adds r0, r2, 0
-	subs r2, 0x1
-	cmp r0, 0
-	bne _082E9476
-	adds r0, r5, 0
-	pop {r4,r5,pc}
-	thumb_func_end memset
-
-	thumb_func_start strcmp
-@ int strcmp(char *s1, char *s2)
-strcmp: @ 82E9488
-	push {r4,r5,lr}
-	adds r2, r0, 0
-	adds r3, r1, 0
-	orrs r0, r3
-	movs r1, 0x3
-	ands r0, r1
-	cmp r0, 0
-	bne _082E94CE
-	ldr r1, [r2]
-	ldr r0, [r3]
-	cmp r1, r0
-	bne _082E94CE
-	ldr r5, _082E94B4
-	ldr r4, _082E94B8
-_082E94A4:
-	ldr r1, [r2]
-	adds r0, r1, r5
-	bics r0, r1
-	ands r0, r4
-	cmp r0, 0
-	beq _082E94BC
-	movs r0, 0
-	b _082E94E0
-	.align 2, 0
-_082E94B4: .4byte 0xfefefeff
-_082E94B8: .4byte 0x80808080
-_082E94BC:
-	adds r2, 0x4
-	adds r3, 0x4
-	ldr r1, [r2]
-	ldr r0, [r3]
-	cmp r1, r0
-	beq _082E94A4
-	b _082E94CE
-_082E94CA:
-	adds r2, 0x1
-	adds r3, 0x1
-_082E94CE:
-	ldrb r0, [r2]
-	cmp r0, 0
-	beq _082E94DA
-	ldrb r1, [r3]
-	cmp r0, r1
-	beq _082E94CA
-_082E94DA:
-	ldrb r2, [r2]
-	ldrb r3, [r3]
-	subs r0, r2, r3
-_082E94E0:
-	pop {r4,r5,pc}
-	thumb_func_end strcmp
-
-	thumb_func_start strcat
-strcat: @ 82E94E4
-	push {r4-r6,lr}
-	adds r6, r0, 0
-	adds r3, r6, 0
-	adds r2, r1, 0
-	adds r0, r2, 0
-	orrs r0, r6
-	movs r1, 0x3
-	ands r0, r1
-	cmp r0, 0
-	bne _082E951C
-	ldr r1, [r2]
-	ldr r5, _082E9504
-	adds r0, r1, r5
-	bics r0, r1
-	ldr r4, _082E9508
-	b _082E9516
-	.align 2, 0
-_082E9504: .4byte 0xfefefeff
-_082E9508: .4byte 0x80808080
-_082E950C:
-	ldm r2!, {r0}
-	stm r3!, {r0}
-	ldr r1, [r2]
-	adds r0, r1, r5
-	bics r0, r1
-_082E9516:
-	ands r0, r4
-	cmp r0, 0
-	beq _082E950C
-_082E951C:
-	ldrb r0, [r2]
-	strb r0, [r3]
-	lsls r0, 24
-	adds r2, 0x1
-	adds r3, 0x1
-	cmp r0, 0
-	bne _082E951C
-	adds r0, r6, 0
-	pop {r4-r6,pc}
-
-	.align 2, 0 @ Don't pad with nop.

asm/rom_80093CC.s

@@ -2523,7 +2523,7 @@ task00_link_test: @ 800A850
 	sub sp, 0x20
 	ldr r1, =gUnknown_082ED1E4
 	mov r0, sp
-	bl strcat
+	bl strcpy
 	mov r0, sp
 	movs r1, 0x5
 	movs r2, 0x2

ld_script.txt

@@ -86,7 +86,7 @@ SECTIONS {
         tools/agbcc/lib/libgcc.a:fp-bit.o(.text);
         tools/agbcc/lib/libgcc.a:_lshrdi3.o(.text);
         tools/agbcc/lib/libgcc.a:_negdi2.o(.text);
-        asm/libc.o(.text);
+        src/libc.o(.text);
     } =0
 
     .rodata :

src/libc.c

@@ -0,0 +1,173 @@
+#include "global.h"
+#include <stddef.h>
+
+#define LBLOCKSIZE (sizeof(long))
+
+// Nonzero if (long)X contains a NULL byte.
+#define CONTAINSNULL(X) (((X) - 0x01010101) & ~(X) & 0x80808080)
+
+// Nonzero if X is not aligned on a "long" boundary.
+#define UNALIGNED(X) ((long)X & (LBLOCKSIZE - 1))
+
+void *memcpy(void *dst0, const void *src0, size_t len0)
+{
+	char *dst = dst0;
+	const char *src = src0;
+	long *aligned_dst;
+	const long *aligned_src;
+	unsigned int len = len0;
+	
+	// If the size is small, or either src or dst is unaligned,
+	// then go to the byte copy loop. This should be rare.
+	if(len >= 16 && !(UNALIGNED(src) | UNALIGNED(dst)))
+	{
+		aligned_dst = (long *)dst;
+		aligned_src = (long *)src;
+		
+		// Copy 4X long words at a time if possible.
+		while(len >= 16)
+		{
+			*aligned_dst++ = *aligned_src++;
+			*aligned_dst++ = *aligned_src++;
+			*aligned_dst++ = *aligned_src++;
+			*aligned_dst++ = *aligned_src++;
+			len -= 16;
+		}
+		
+		// Copy one long word at a time if possible
+		while(len >= 4)
+		{
+			*aligned_dst++ = *aligned_src++;
+			len -= 4;
+		}
+		
+		dst = (char *)aligned_dst;
+		src = (char *)aligned_src;
+	}
+	
+	// Pick up any remaining bytes with a byte copier.
+	while(len--)
+		*dst++ = *src++;
+	
+	return dst0;
+}
+
+void *memset(void *m, int c, size_t n)
+{
+	char *s = (char *)m;
+	int count, i;
+	unsigned long buffer;
+	unsigned long *aligned_addr;
+	unsigned char *unaligned_addr;
+	
+	// If the size is small or m is unaligned,
+	// then go to the byte copy loop. This should be rare.
+	if(n >= LBLOCKSIZE && !UNALIGNED(m))
+	{
+		// We know that n is large and m is word-aligned.
+		aligned_addr = (unsigned long *)m;
+		
+		// Store C into each char sized location in buffer so that
+		// we can set large blocks quickly.
+		c &= 0xFF;
+		if(LBLOCKSIZE == 4)
+		{
+			buffer = (c << 8) | c;
+			buffer |= (buffer << 16);
+		}
+		else
+		{
+			buffer = 0;
+			for(i = 0; i < LBLOCKSIZE; i++)
+				buffer = (buffer << 8) | c;
+		}
+		
+		while(n >= LBLOCKSIZE * 4)
+		{
+			*aligned_addr++ = buffer;
+			*aligned_addr++ = buffer;
+			*aligned_addr++ = buffer;
+			*aligned_addr++ = buffer;
+			n -= LBLOCKSIZE * 4;
+		}
+		while(n >= LBLOCKSIZE)
+		{
+			*aligned_addr++ = buffer;
+			n -= LBLOCKSIZE;
+		}
+		
+		s = (char *)aligned_addr;
+	}
+	
+	// Pick up the remainder with a bytewise loop.
+	while(n--)
+		*s++ = (char)c;
+	
+	return m;
+}
+
+int strcmp(const char *s1, const char *s2)
+{
+	unsigned long *a1;
+	unsigned long *a2;
+	
+	// If s1 or s2 are unaligned, then skip this and compare bytes.
+	if(!(UNALIGNED(s1) | UNALIGNED(s2)))
+	{
+		// Compare them a word at a time.
+		a1 = (unsigned long *)s1;
+		a2 = (unsigned long *)s2;
+		while(*a1 == *a2)
+		{
+			// If *a1 == *a2, and we find a null in *a1,
+			// then the strings must be equal, so return zero.
+			if(CONTAINSNULL(*a1))
+				return 0;
+			
+			a1++;
+			a2++;
+		}
+		
+		s1 = (char *)a1;
+		s2 = (char *)a2;
+	}
+	
+	// Check the remaining few bytes.
+	while(*s1 != '\0' && *s1 == *s2)
+	{
+		s1++;
+		s2++;
+	}
+	
+	return (*(unsigned char *) s1) - (*(unsigned char *) s2);
+}
+
+char* strcpy(char *dst0, const char *src0)
+{
+  char *dst = dst0;
+  const char *src = src0;
+	unsigned long *a1;
+	const unsigned long *a2;
+	
+  /* If SRC or DEST is unaligned, then copy bytes.  */
+	if(!(UNALIGNED(src) | UNALIGNED(dst)))
+	{
+		/* SRC and DEST are both "long int" aligned, try to do "long int"
+         sized copies.  */
+		a1 = (unsigned long *)dst;
+		a2 = (unsigned long *)src;
+		while(!CONTAINSNULL(*a2))
+		{
+      *a1++ = *a2++;
+		}
+		
+		dst = (char *)a1;
+		src = (char *)a2;
+	}
+	
+	// Copy the remaining few bytes.
+	while(*dst++ = *src++);
+	
+	return dst0;
+}
+