Switch from tonccpy to ASM memcpy

hb/dldi/Makefile

@@ -26,11 +26,12 @@ INCLUDES	:=	include ../common/include
 #---------------------------------------------------------------------------------
 # options for code generation
 #---------------------------------------------------------------------------------
-ARCH	:=	-mthumb -mthumb-interwork -march=armv4t
+ARCH	:=	-mthumb -march=armv4t
 
 COMMON	:=	-g -Wall -O2\
 			-mcpu=arm7tdmi -mtune=arm7tdmi -fomit-frame-pointer\
 			-ffast-math \
+			-flto \
 			$(ARCH)
 
 # ADD -DDEBUG here for logging/debug
@@ -39,7 +40,7 @@ COMMON	+=	$(INCLUDE) -DARM9 -fPIC
 CFLAGS	:=	$(COMMON) -std=gnu99
 CXXFLAGS	:= $(CFLAGS) -fno-rtti -fno-exceptions
 
-ASFLAGS	:=	-g $(ARCH) $(INCLUDE)
+ASFLAGS	:=	-g $(ARCH) -flto $(INCLUDE)
 LDFLAGS	=	-nostartfiles -nostdlib -T ../dldi.ld -g $(ARCH) -Wl,-Map,$(TARGET).map
 
 LIBS	:= -lnds9

hb/dldi/include/aeabi.h

@@ -0,0 +1,116 @@
+// SPDX-License-Identifier: Zlib
+// SPDX-FileNotice: Modified from the original version by the BlocksDS project.
+//
+// Copyright (C) 2021-2023 agbabi contributors
+
+#ifndef AEABI_H__
+#define AEABI_H__
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stddef.h>
+
+/**
+ * Alias of __aeabi_memcpy4
+ * @param dest Destination address
+ * @param src Source address
+ * @param n Number of bytes to copy
+ */
+void __aeabi_memcpy8(void* __restrict__ dest, const void* __restrict__ src, size_t n) __attribute__((nonnull(1, 2)));
+
+/**
+ * Copies n bytes from src to dest (forward)
+ * Assumes dest and src are 4-byte aligned
+ * @param dest Destination address
+ * @param src Source address
+ * @param n Number of bytes to copy
+ */
+void __aeabi_memcpy4(void* __restrict__ dest, const void* __restrict__ src, size_t n) __attribute__((nonnull(1, 2)));
+
+/**
+ * Copies n bytes from src to dest (forward)
+ * @param dest Destination address
+ * @param src Source address
+ * @param n Number of bytes to copy
+ */
+void __aeabi_memcpy(void* __restrict__ dest, const void* __restrict__ src, size_t n) __attribute__((nonnull(1, 2)));
+
+/**
+ * Alias of __aeabi_memmove4
+ * @param dest Destination address
+ * @param src Source address
+ * @param n Number of bytes to copy
+ */
+void __aeabi_memmove8(void* dest, const void* src, size_t n) __attribute__((nonnull(1, 2)));
+
+/**
+ * Safely copies n bytes of src to dest
+ * Assumes dest and src are 4-byte aligned
+ * @param dest Destination address
+ * @param src Source address
+ * @param n Number of bytes to copy
+ */
+void __aeabi_memmove4(void* dest, const void* src, size_t n) __attribute__((nonnull(1, 2)));
+
+/**
+ * Safely copies n bytes of src to dest
+ * @param dest Destination address
+ * @param src Source address
+ * @param n Number of bytes to copy
+ */
+void __aeabi_memmove(void* dest, const void* src, size_t n) __attribute__((nonnull(1, 2)));
+
+/**
+ * Alias of __aeabi_memset4
+ * @param dest Destination address
+ * @param n Number of bytes to set
+ * @param c Value to set
+ */
+void __aeabi_memset8(void* dest, size_t n, int c) __attribute__((nonnull(1)));
+
+/**
+ * Set n bytes of dest to (c & 0xff)
+ * Assumes dest is 4-byte aligned
+ * @param dest Destination address
+ * @param n Number of bytes to set
+ * @param c Value to set
+ */
+void __aeabi_memset4(void* dest, size_t n, int c) __attribute__((nonnull(1)));
+
+/**
+ * Set n bytes of dest to (c & 0xff)
+ * @param dest Destination address
+ * @param n Number of bytes to set
+ * @param c Value to set
+ */
+void __aeabi_memset(void* dest, size_t n, int c) __attribute__((nonnull(1)));
+
+/**
+ * Alias of __aeabi_memclr4
+ * @param dest Destination address
+ * @param n Number of bytes to clear
+ */
+void __aeabi_memclr8(void* dest, size_t n) __attribute__((nonnull(1)));
+
+/**
+ * Clears n bytes of dest to 0
+ * Assumes dest is 4-byte aligned
+ * @param dest Destination address
+ * @param n Number of bytes to clear
+ */
+void __aeabi_memclr4(void* dest, size_t n) __attribute__((nonnull(1)));
+
+/**
+ * Clears n bytes of dest to 0
+ * @param dest Destination address
+ * @param n Number of bytes to clear
+ */
+void __aeabi_memclr(void* dest, size_t n) __attribute__((nonnull(1)));
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // AEABI_H__

retail/bootloader/include/asminc.h → hb/dldi/include/asminc.h

@@ -2,8 +2,8 @@
 #define _ASMINC_H_
 
 #if !__ASSEMBLER__
-	#error This header file is only for use in assembly files!
-#endif	// !__ASSEMBLER__
+# error "This header file is only for use in assembly files!"
+#endif
 
 
 .macro BEGIN_ASM_FUNC name section=text
@@ -14,6 +14,13 @@
 \name:
 .endm
 
+.macro BEGIN_ASM_FUNC_NO_SECTION name
+    .global \name
+    .type \name, %function
+    .align 2
+\name:
+.endm
+
 #define ICACHE_SIZE	0x2000
 #define DCACHE_SIZE	0x1000
 #define CACHE_LINE_SIZE	32

hb/dldi/include/macros.inc

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: Zlib
+// SPDX-FileNotice: Modified from the original version by the BlocksDS project.
+//
+// Copyright (C) 2021-2023 agbabi contributors
+//
+// ARM assembly support macros
+
+@ Shift and test upper two bits, clobbering \reg
+@ Use mi for first bit, cs for second bit
+.macro joaobapt_test_lsl reg shift = #0
+    movs    \reg, \reg, lsl \shift
+.endm
+
+@ Test lowest two bits, clobbering \reg
+@ Use mi for low bit, cs for high bit
+.macro joaobapt_test reg
+    joaobapt_test_lsl \reg, #31
+.endm
+
+@ Test lowest two bits of \src, result stored in \dst
+@ Use mi for low bit, cs for high bit
+.macro joaobapt_test_into dst, src
+    movs    \dst, \src, lsl #31
+.endm
+
+@ Branches depending on lowest two bits, clobbering \reg
+@ b_mi = low bit case, b_cs = high bit case
+.macro joaobapt_switch reg, b_mi, b_cs
+    joaobapt_test \reg
+    bmi     \b_mi
+    bcs     \b_cs
+.endm
+
+@ Branches depending on alignment of \a and \b, clobbering \scratch
+@ b_byte = off-by-byte case, b_half = off-by-half case
+.macro align_switch a, b, scratch, b_byte, b_half
+    eor     \scratch, \a, \b
+    joaobapt_switch \scratch, \b_byte, \b_half
+.endm

hb/dldi/source/abi_memcpy.s

@@ -0,0 +1,126 @@
+// SPDX-License-Identifier: Zlib
+// SPDX-FileNotice: Modified from the original version by the BlocksDS project.
+//
+// Copyright (C) 2021-2023 agbabi contributors
+//
+// ABI:
+//    __aeabi_memcpy, __aeabi_memcpy4, __aeabi_memcpy8
+// Standard:
+//    memcpy
+// Support:
+//    __ndsabi_memcpy2, __ndsabi_memcpy1
+
+#include "asminc.h"
+
+#include "macros.inc"
+
+    .syntax unified
+
+    .arm
+
+
+
+BEGIN_ASM_FUNC __aeabi_memcpy
+
+    @ >6-bytes is roughly the threshold when byte-by-byte copy is slower
+    cmp     r2, #6
+    ble     __ndsabi_memcpy1
+
+    align_switch r0, r1, r3, __ndsabi_memcpy1, .Lcopy_halves
+
+    @ Check if r0 (or r1) needs word aligning
+    rsbs    r3, r0, #4
+    joaobapt_test r3
+
+    @ Copy byte head to align
+    ldrbmi  r3, [r1], #1
+    strbmi  r3, [r0], #1
+    submi   r2, r2, #1
+    @ r0, r1 are now half aligned
+
+    @ Copy half head to align
+    ldrhcs  r3, [r1], #2
+    strhcs  r3, [r0], #2
+    subcs   r2, r2, #2
+    @ r0, r1 are now word aligned
+
+
+BEGIN_ASM_FUNC_NO_SECTION __aeabi_memcpy8
+BEGIN_ASM_FUNC_NO_SECTION __aeabi_memcpy4
+
+    cmp     r2, #32
+    blt     .Lcopy_words
+
+    @ Word aligned, 32-byte copy
+    push    {r4-r10}
+.Lloop_32:
+    subs    r2, r2, #32
+    ldmiage r1!, {r3-r10}
+    stmiage r0!, {r3-r10}
+    bgt     .Lloop_32
+    pop     {r4-r10}
+    bxeq    lr
+
+    @ < 32 bytes remaining to be copied
+    add     r2, r2, #32
+
+.Lcopy_words:
+    cmp     r2, #4
+    blt     .Lcopy_halves
+.Lloop_4:
+    subs    r2, r2, #4
+    ldrge   r3, [r1], #4
+    strge   r3, [r0], #4
+    bgt     .Lloop_4
+    bxeq    lr
+
+    @ Copy byte & half tail
+    @ This test still works when r2 is negative
+    joaobapt_test r2
+    @ Copy half
+    ldrhcs  r3, [r1], #2
+    strhcs  r3, [r0], #2
+    @ Copy byte
+    ldrbmi  r3, [r1]
+    strbmi  r3, [r0]
+    bx      lr
+
+.Lcopy_halves:
+    @ Copy byte head to align
+    tst     r0, #1
+    ldrbne  r3, [r1], #1
+    strbne  r3, [r0], #1
+    subne   r2, r2, #1
+    @ r0, r1 are now half aligned
+
+
+BEGIN_ASM_FUNC_NO_SECTION __ndsabi_memcpy2
+
+    subs    r2, r2, #2
+    ldrhge  r3, [r1], #2
+    strhge  r3, [r0], #2
+    bgt     __ndsabi_memcpy2
+    bxeq    lr
+
+    @ Copy byte tail
+    adds    r2, r2, #2
+    ldrbne  r3, [r1]
+    strbne  r3, [r0]
+    bx      lr
+
+
+BEGIN_ASM_FUNC_NO_SECTION __ndsabi_memcpy1
+
+    subs    r2, r2, #1
+    ldrbge  r3, [r1], #1
+    strbge  r3, [r0], #1
+    bgt     __ndsabi_memcpy1
+    bx      lr
+
+
+BEGIN_ASM_FUNC memcpy
+
+    push    {r0, lr}
+    bl      __aeabi_memcpy
+    pop     {r0, lr}
+    bx      lr