Handwritten asm build prep, add handwritten asm for the code segment (#1807)

* Convert entry.s to IDO, use objcopy instead of ipl3.s and audio/lib/stack.s, add handwritten asm for the code segment

* Fix disasm, review comments

* Other suggestions

12 files changed, 611 insertions, 96 deletions

diff --git a/Makefile b/Makefile
index 51e1dfd8a..9db358570 100644
--- a/Makefile
+++ b/Makefile
@@ -116,9 +116,11 @@ endif
 # Detect compiler and set variables appropriately.
 ifeq ($(COMPILER),gcc)
   CC       := $(MIPS_BINUTILS_PREFIX)gcc
+  CCAS     := $(CC) -x assembler-with-cpp
 else ifeq ($(COMPILER),ido)
   CC       := tools/ido_recomp/$(DETECTED_OS)/7.1/cc
   CC_OLD   := tools/ido_recomp/$(DETECTED_OS)/5.3/cc
+  CCAS     := $(CC_OLD)
 else
 $(error Unsupported compiler. Please use either ido or gcc as the COMPILER variable.)
 endif
@@ -133,6 +135,7 @@ ifeq ($(ORIG_COMPILER),1)
   endif
   CC        = $(QEMU_IRIX) -L tools/ido7.1_compiler tools/ido7.1_compiler/usr/bin/cc
   CC_OLD    = $(QEMU_IRIX) -L tools/ido5.3_compiler tools/ido5.3_compiler/usr/bin/cc
+  CCAS      = $(CC_OLD)
 endif
 
 AS      := $(MIPS_BINUTILS_PREFIX)as
@@ -140,6 +143,7 @@ LD      := $(MIPS_BINUTILS_PREFIX)ld
 NM      := $(MIPS_BINUTILS_PREFIX)nm
 OBJCOPY := $(MIPS_BINUTILS_PREFIX)objcopy
 OBJDUMP := $(MIPS_BINUTILS_PREFIX)objdump
+STRIP   := $(MIPS_BINUTILS_PREFIX)strip
 
 IINC := -Iinclude -Iinclude/libc -Isrc -I$(BUILD_DIR) -I. -I$(EXTRACTED_DIR)
 
@@ -213,7 +217,8 @@ ifeq ($(COMPILER),gcc)
   CFLAGS           += -nostdinc -fno-PIC -fno-common -ffreestanding -fbuiltin -fno-builtin-sinf -fno-builtin-cosf -funsigned-char
 
   WARNINGS         := $(CC_CHECK_WARNINGS)
-  ASFLAGS          := -march=vr4300 -32 -G0
+  ASFLAGS          := -march=vr4300 -32 -G0 -no-pad-sections
+  CCASFLAGS        := $(GBI_DEFINES) -G 0 -nostdinc -march=vr4300 -mfix4300 -mabi=32 -mno-abicalls -fno-PIC -fno-common -Wa,-no-pad-sections
   COMMON_DEFINES   := $(GBI_DEFINES)
   AS_DEFINES       := $(COMMON_DEFINES) -DMIPSEB -D_LANGUAGE_ASSEMBLY -D_ULTRA64
   C_DEFINES        := $(COMMON_DEFINES) -D_LANGUAGE_C
@@ -225,7 +230,8 @@ else
   CFLAGS           += -G 0 -non_shared -Xcpluscomm -nostdinc -Wab,-r4300_mul
 
   WARNINGS         := -fullwarn -verbose -woff 624,649,838,712,516,513,596,564,594,807
-  ASFLAGS          := -march=vr4300 -32 -G0
+  ASFLAGS          := -march=vr4300 -32 -G0 -no-pad-sections
+  CCASFLAGS        := $(GBI_DEFINES) -G 0 -non_shared -Xcpluscomm -nostdinc -Wab,-r4300_mul $(WARNINGS) -o32
   COMMON_DEFINES   := -D_MIPS_SZLONG=32 $(GBI_DEFINES)
   AS_DEFINES       := $(COMMON_DEFINES) -DMIPSEB -D_LANGUAGE_ASSEMBLY -D_ULTRA64
   C_DEFINES        := $(COMMON_DEFINES) -D_LANGUAGE_C
@@ -234,6 +240,7 @@ else
   OPTFLAGS         := -O2 -g3
   MIPS_VERSION     := -mips2
 endif
+ASOPTFLAGS := -O1
 
 # Use relocations and abi fpr names in the dump
 OBJDUMP_FLAGS := --disassemble --reloc --disassemble-zeroes -Mreg-names=32
@@ -380,7 +387,8 @@ O_FILES        := $(foreach f,$(S_FILES:.s=.o),$(BUILD_DIR)/$f) \
                   $(foreach f,$(ASSET_C_FILES_EXTRACTED:.c=.o),$(f:$(EXTRACTED_DIR)/%=$(BUILD_DIR)/%)) \
                   $(foreach f,$(ASSET_C_FILES_COMMITTED:.c=.o),$(BUILD_DIR)/$f) \
                   $(foreach f,$(BASEROM_FILES),$(BUILD_DIR)/$f.o) \
-                  $(foreach f,$(ARCHIVES_O),$(BUILD_DIR)/$f)
+                  $(foreach f,$(ARCHIVES_O),$(BUILD_DIR)/$f) \
+                  $(BUILD_DIR)/src/makerom/ipl3.o $(BUILD_DIR)/src/audio/lib/stack.o
 
 SHIFTJIS_C_FILES := src/libultra/voice/voicecheckword.c src/audio/voice_external.c src/code/z_message.c src/code/z_message_nes.c
 SHIFTJIS_O_FILES := $(foreach f,$(SHIFTJIS_C_FILES:.c=.o),$(BUILD_DIR)/$f)
@@ -426,7 +434,11 @@ endif
 
 $(BUILD_DIR)/src/audio/lib/seqplayer.o: C_DEFINES += -DMML_VERSION=MML_VERSION_MM
 
+# Command to patch certain object files after they are built
+POSTPROCESS_OBJ := @:
+
 ifeq ($(COMPILER),ido)
+
 # directory flags
 $(BUILD_DIR)/src/libultra/os/%.o: OPTFLAGS := -O1
 $(BUILD_DIR)/src/libultra/voice/%.o: OPTFLAGS := -O2
@@ -441,6 +453,8 @@ $(BUILD_DIR)/src/boot/libc/%.o: OPTFLAGS := -O2
 $(BUILD_DIR)/src/boot/libm/%.o: OPTFLAGS := -O2
 $(BUILD_DIR)/src/boot/libc64/%.o: OPTFLAGS := -O2
 
+$(BUILD_DIR)/src/code/%.o: ASOPTFLAGS := -O2
+
 $(BUILD_DIR)/src/audio/%.o: OPTFLAGS := -O2
 
 $(BUILD_DIR)/assets/%.o: OPTFLAGS := -O1
@@ -448,8 +462,14 @@ $(BUILD_DIR)/assets/%.o: OPTFLAGS := -O1
 # file flags
 $(BUILD_DIR)/src/libultra/libc/ll.o: OPTFLAGS := -O1
 $(BUILD_DIR)/src/libultra/libc/ll.o: MIPS_VERSION := -mips3 -32
+$(BUILD_DIR)/src/libultra/libc/ll.o: POSTPROCESS_OBJ := $(PYTHON) tools/set_o32abi_bit.py
+
 $(BUILD_DIR)/src/libultra/libc/llcvt.o: OPTFLAGS := -O1
 $(BUILD_DIR)/src/libultra/libc/llcvt.o: MIPS_VERSION := -mips3 -32
+$(BUILD_DIR)/src/libultra/libc/llcvt.o: POSTPROCESS_OBJ := $(PYTHON) tools/set_o32abi_bit.py
+
+$(BUILD_DIR)/src/libultra/os/exceptasm.o: MIPS_VERSION := -mips3 -32
+$(BUILD_DIR)/src/libultra/os/exceptasm.o: POSTPROCESS_OBJ := $(PYTHON) tools/set_o32abi_bit.py
 
 $(BUILD_DIR)/src/boot/fault.o: CFLAGS += -trapuv
 $(BUILD_DIR)/src/boot/fault_drawer.o: CFLAGS += -trapuv
@@ -613,8 +633,44 @@ $(BUILD_DIR)/dmadata/dmadata_table_spec.h $(BUILD_DIR)/dmadata/compress_ranges.t
 $(BUILD_DIR)/src/boot/z_std_dma.o: $(BUILD_DIR)/dmadata/dmadata_table_spec.h
 $(BUILD_DIR)/src/dmadata/dmadata.o: $(BUILD_DIR)/dmadata/dmadata_table_spec.h
 
-$(BUILD_DIR)/%.o: %.s
-	$(CPP) $(CPPFLAGS) $(IINC) $< | $(AS) $(ASFLAGS) $(IINC) $(ENDIAN) -o $@
+$(BUILD_DIR)/asm/%.o: asm/%.s
+	$(AS) $(ASFLAGS) $(IINC) $(ENDIAN) $< -o $@
+
+$(BUILD_DIR)/data/%.o: data/%.s
+	$(AS) $(ASFLAGS) $(IINC) $(ENDIAN) $< -o $@
+
+$(BUILD_DIR)/rsp/%.o: rsp/%.s
+	$(AS) $(ASFLAGS) $(IINC) $(ENDIAN) $< -o $@
+
+# Assemble the ROM header with GNU AS always
+$(BUILD_DIR)/src/makerom/rom_header.o: src/makerom/rom_header.s
+ifeq ($(COMPILER),ido)
+	$(CPP) $(CPPFLAGS) $(MIPS_BUILTIN_DEFS) $(IINC) $< | $(AS) $(ASFLAGS) $(IINC) $(ENDIAN) -o $@
+else
+	$(CCAS) -c $(CCASFLAGS) $(IINC) $(MIPS_VERSION) $(ASOPTFLAGS) -o $@ $<
+endif
+	$(OBJDUMP_CMD)
+
+$(BUILD_DIR)/src/makerom/ipl3.o: $(EXTRACTED_DIR)/incbin/ipl3
+	$(OBJCOPY) -I binary -O elf32-big --rename-section .data=.text $< $@
+
+$(BUILD_DIR)/src/audio/lib/stack.o: $(EXTRACTED_DIR)/incbin/aspMainStack
+	$(OBJCOPY) -I binary -O elf32-big --add-symbol aspMainStack=.data:0,global $< $@
+
+$(BUILD_DIR)/src/%.o: src/%.s
+ifeq ($(COMPILER),ido)
+	$(CCAS) -c $(CCASFLAGS) $(IINC) $(MIPS_VERSION) $(ASOPTFLAGS) -o $(@:.o=.tmp.o) $<
+# IDO generates bad symbol tables, fix the symbol table with strip..
+	$(STRIP) $(@:.o=.tmp.o) -N dummy-symbol-name
+# but strip doesn't know about file-relative offsets in .mdebug and doesn't relocate them, ld will
+# segfault unless .mdebug is removed
+	$(OBJCOPY) --remove-section .mdebug $(@:.o=.tmp.o) $@
+else
+	$(CCAS) -c $(CCASFLAGS) $(IINC) $(MIPS_VERSION) $(ASOPTFLAGS) -o $@ $<
+endif
+	$(POSTPROCESS_OBJ) $@
+	$(OBJDUMP_CMD)
+	$(RM_MDEBUG)
 
 $(BUILD_DIR)/assets/text/%.o: assets/text/%.c
 ifneq ($(COMPILER),gcc)
@@ -681,23 +737,10 @@ $(SHIFTJIS_O_FILES): $(BUILD_DIR)/src/%.o: src/%.c
 	$(OBJDUMP_CMD)
 	$(RM_MDEBUG)
 
-$(BUILD_DIR)/src/libultra/libc/ll.o: src/libultra/libc/ll.c
-	$(CC_CHECK_COMP) $(CC_CHECK_FLAGS) $(IINC) $(CC_CHECK_WARNINGS) $(C_DEFINES) $(MIPS_BUILTIN_DEFS) $<
-	$(CC) -c $(CFLAGS) $(IINC) $(WARNINGS) $(C_DEFINES) $(MIPS_VERSION) $(ENDIAN) $(OPTFLAGS) -o $@ $<
-	$(PYTHON) tools/set_o32abi_bit.py $@
-	$(OBJDUMP_CMD)
-	$(RM_MDEBUG)
-
-$(BUILD_DIR)/src/libultra/libc/llcvt.o: src/libultra/libc/llcvt.c
-	$(CC_CHECK_COMP) $(CC_CHECK_FLAGS) $(IINC) $(CC_CHECK_WARNINGS) $(C_DEFINES) $(MIPS_BUILTIN_DEFS) $<
-	$(CC) -c $(CFLAGS) $(IINC) $(WARNINGS) $(C_DEFINES) $(MIPS_VERSION) $(ENDIAN) $(OPTFLAGS) -o $@ $<
-	$(PYTHON) tools/set_o32abi_bit.py $@
-	$(OBJDUMP_CMD)
-	$(RM_MDEBUG)
-
 $(BUILD_DIR)/%.o: %.c
 	$(CC_CHECK_COMP) $(CC_CHECK_FLAGS) $(IINC) $(CC_CHECK_WARNINGS) $(C_DEFINES) $(MIPS_BUILTIN_DEFS) $<
 	$(CC) -c $(CFLAGS) $(IINC) $(WARNINGS) $(C_DEFINES) $(MIPS_VERSION) $(ENDIAN) $(OPTFLAGS) -o $@ $<
+	$(POSTPROCESS_OBJ) $@
 	$(OBJDUMP_CMD)
 	$(RM_MDEBUG)
 
diff --git a/include/PR/asm.h b/include/PR/asm.h
index d5058f5c1..379c1a875 100644
--- a/include/PR/asm.h
+++ b/include/PR/asm.h
@@ -6,55 +6,106 @@
 #define _MIPS_ISA_MIPS2 2
 #define _MIPS_ISA_MIPS3 3
 #define _MIPS_ISA_MIPS4 4
+
+#define _MIPS_SIM_ABI32     1   /* MIPS MSIG calling convention */
+#define _MIPS_SIM_NABI32    2   /* MIPS new 32-bit abi. NABI32 is 64bit calling convention but 32bit type sizes */
+#define _MIPS_SIM_ABI64     3   /* MIPS 64 calling convention */
 #endif
 
 #ifndef _LANGUAGE_C
 
+#ifdef __GNUC__
+#define TYPE(x, t) .type x, @t
+#define SIZE(x)    .size x, . - x
+#else
+#define TYPE(x, t)
+#define SIZE(x)
+#endif
+
 #define LEAF(x)                 \
-    .balign 4                  ;\
+    .align 2                   ;\
     .globl x                   ;\
-    .type x, @function         ;\
+    TYPE(x, function)          ;\
+    .ent x, 0                  ;\
     x:                         ;\
-        .ent x, 0              ;\
-        .frame $sp, 0, $ra
+        .frame sp, 0, ra
 
 #define XLEAF(x)                \
-    .balign 4                  ;\
+    .align 2                   ;\
     .globl x                   ;\
-    .type x, @function         ;\
-    x:                         ;\
-        .aent x, 0
+    TYPE(x, function)          ;\
+    .aent x, 0                 ;\
+    x:
 
 #define NESTED(x, fsize, ra)    \
     .globl x                   ;\
+    .ent x, 0                  ;\
     x:                         ;\
-        .ent x, 0              ;\
-        .frame $sp, fsize, ra
+        .frame sp, fsize, ra
 
 #define XNESTED(x)              \
     .globl x                   ;\
-    x:                         ;\
-        .aent x, 0
+    .aent x, 0                 ;\
+    x:
 
 #define END(x)                  \
-    .size x, . - x             ;\
+    SIZE(x)                    ;\
     .end x
 
-#define IMPORT(x, size)         \
-    .extern x, size
-
 #define EXPORT(x)               \
     .globl x                   ;\
     x:
 
+#ifdef __sgi
+#define IMPORT(sym, size)       \
+    .extern sym, size
+#else
+#define IMPORT(sym, size)
+#endif
+
 #define DATA(x)                 \
-    .balign 4                  ;\
+    .align 2                   ;\
     .globl x                   ;\
-    .type x, @object           ;\
+    TYPE(x, object)            ;\
     x:
 
 #define ENDDATA(x)              \
-    .size x, . - x
+    SIZE(x)
+
+#define MFC0(dst, src) \
+    .set noreorder; mfc0 dst, src; .set reorder
+#define MTC0(dst, src) \
+    .set noreorder; mtc0 dst, src; .set reorder
+
+#define CACHE(op, base)  \
+    .set noreorder; cache op, base; .set reorder
+
+#define CFC1(dst, src) \
+    .set noreorder; cfc1 dst, src; .set reorder
+#define CTC1(src, dst) \
+    .set noreorder; ctc1 src, dst; .set reorder
+
+#define NOP \
+    .set noreorder; nop; .set reorder
+
+#define TLBWI \
+    .set noreorder; tlbwi; .set reorder
+
+#define TLBR \
+    .set noreorder; tlbr; .set reorder
+
+#define TLBP \
+    .set noreorder; tlbp; .set reorder
+
+#ifndef __GNUC__
+#define ABS(x, y)   \
+    .globl  x;      \
+    x = y
+#else
+#define ABS(x, y)   \
+    .globl x;       \
+    .set x, y
+#endif
 
 #endif
 
@@ -62,11 +113,11 @@
  *  Stack Alignment
  */
 #if   (_MIPS_SIM == _ABIO32)
-#define NARGSAVE 4      // space for 4 args must be allocated
+#define NARGSAVE 4      /* space for 4 args must be allocated */
 #define ALSZ    (8-1)
 #define ALMASK ~(8-1)
 #elif (_MIPS_SIM == _ABIN32 || _MIPS_SIM == _ABI64)
-#define NARGSAVE 0      // no caller responsibilities
+#define NARGSAVE 0      /* no caller responsibilities */
 #define ALSZ    (16-1)
 #define ALMASK ~(16-1)
 #endif
diff --git a/include/PR/regdef.h b/include/PR/regdef.h
new file mode 100644
index 000000000..03438c8bb
--- /dev/null
+++ b/include/PR/regdef.h
@@ -0,0 +1,164 @@
+#ifndef PR_REGDEF_H
+#define PR_REGDEF_H
+
+#ifdef __GNUC__
+#define _MIPS_SIM_ABI32 _ABIO32
+#define _MIPS_SIM_NABI32 _ABIN32
+#define _MIPS_SIM_ABI64 _ABI64
+#endif
+
+#if (_MIPS_SIM == _MIPS_SIM_ABI32)
+#define zero    $0
+#define AT      $at
+#define v0      $2
+#define v1      $3
+#define a0      $4
+#define a1      $5
+#define a2      $6
+#define a3      $7
+#define t0      $8
+#define t1      $9
+#define t2      $10
+#define t3      $11
+#define t4      $12
+#define ta0     $12
+#define t5      $13
+#define ta1     $13
+#define t6      $14
+#define ta2     $14
+#define t7      $15
+#define ta3     $15
+#define s0      $16
+#define s1      $17
+#define s2      $18
+#define s3      $19
+#define s4      $20
+#define s5      $21
+#define s6      $22
+#define s7      $23
+#define t8      $24
+#define t9      $25
+#define jp      $25
+#define k0      $26
+#define k1      $27
+#define gp      $28
+#define sp      $29
+#define fp      $30
+#define s8      $30
+#define ra      $31
+#endif
+
+#if (_MIPS_SIM == _MIPS_SIM_ABI64)
+#define zero    $0
+#define AT      $at
+#define v0      $2
+#define v1      $3
+#define a0      $4
+#define a1      $5
+#define a2      $6
+#define a3      $7
+#define a4      $8
+#define ta0     $8
+#define a5      $9
+#define ta1     $9
+#define a6      $10
+#define ta2     $10
+#define a7      $11
+#define ta3     $11
+#define t0      $12
+#define t1      $13
+#define t2      $14
+#define t3      $15
+#define s0      $16
+#define s1      $17
+#define s2      $18
+#define s3      $19
+#define s4      $20
+#define s5      $21
+#define s6      $22
+#define s7      $23
+#define t8      $24
+#define t9      $25
+#define jp      $25
+#define k0      $26
+#define k1      $27
+#define gp      $28
+#define sp      $29
+#define fp      $30
+#define s8      $30
+#define ra      $31
+#endif
+
+#if (_MIPS_SIM == _MIPS_SIM_ABI32)
+#define fv0     $f0
+#define fv0f    $f1
+#define fv1     $f2
+#define fv1f    $f3
+#define fa0     $f12
+#define fa0f    $f13
+#define fa1     $f14
+#define fa1f    $f15
+#define ft0     $f4
+#define ft0f    $f5
+#define ft1     $f6
+#define ft1f    $f7
+#define ft2     $f8
+#define ft2f    $f9
+#define ft3     $f10
+#define ft3f    $f11
+#define ft4     $f16
+#define ft4f    $f17
+#define ft5     $f18
+#define ft5f    $f19
+#define fs0     $f20
+#define fs0f    $f21
+#define fs1     $f22
+#define fs1f    $f23
+#define fs2     $f24
+#define fs2f    $f25
+#define fs3     $f26
+#define fs3f    $f27
+#define fs4     $f28
+#define fs4f    $f29
+#define fs5     $f30
+#define fs5f    $f31
+#endif
+
+#if (_MIPS_SIM == _MIPS_SIM_ABI64)
+#define fv0     $f0
+#define fv1     $f2
+#define fa0     $f12
+#define fa1     $f13
+#define fa2     $f14
+#define fa3     $f15
+#define fa4     $f16
+#define fa5     $f17
+#define fa6     $f18
+#define fa7     $f19
+#define ft0     $f4
+#define ft1     $f5
+#define ft2     $f6
+#define ft3     $f7
+#define ft4     $f8
+#define ft5     $f9
+#define ft6     $f10
+#define ft7     $f11
+#define ft8     $f20
+#define ft9     $f21
+#define ft10    $f22
+#define ft11    $f23
+#define ft12    $f1
+#define ft13    $f3
+#define fs0     $f24
+#define fs1     $f25
+#define fs2     $f26
+#define fs3     $f27
+#define fs4     $f28
+#define fs5     $f29
+#define fs6     $f30
+#define fs7     $f31
+#endif
+
+#define fcr31   $31
+
+#endif
diff --git a/include/kanread.h b/include/kanread.h
index 0ca7ddf9d..5fdc31f6a 100644
--- a/include/kanread.h
+++ b/include/kanread.h
@@ -3,24 +3,6 @@
 
 #include "PR/ultratypes.h"
 
-/**
- * s32 Kanji_OffsetFromShiftJIS(s32 sjis);
- *
- * Returns the offset of the glyph texture data in the file `kanji` corresponding
- * to a given 2-byte Shift-JIS codepoint. No range validity check is carried out.
- *
- * A nice Shift-JIS codepoint table: https://uic.io/en/charset/show/shift_jis/
- * The file `kanji` contains the 'Level 1' kanji (0x889F-0x9872), and a reworked
- * version of the non-kanji section that includes extra English and Hylian glyphs.
- *
- * @note This function assumes that its argument is a valid Shift-JIS codepoint;
- * there is no range protection at all.
- *
- * @param sjis Shift-JIS Codepoint of glyph.
- * @return s32 offset into `kanji` file.
- *
- * @remark Original name: "LeoGetKadr"
- */
 s32 Kanji_OffsetFromShiftJIS(s32 sjis);
 
 #endif
diff --git a/spec/spec b/spec/spec
index 79fd43a88..40e532aa6 100644
--- a/spec/spec
+++ b/spec/spec
@@ -744,7 +744,7 @@ beginseg
     include "$(BUILD_DIR)/src/code/c_keyframe.o"
     include "$(BUILD_DIR)/src/code/sys_slowly.o"
     include "$(BUILD_DIR)/src/code/sys_flashrom.o"
-    include "$(BUILD_DIR)/asm/code/kanread.text.o" // handwritten
+    include "$(BUILD_DIR)/src/code/kanread.o"
     include "$(BUILD_DIR)/src/code/osFlash.o"
     pad_text
     pad_text
@@ -760,9 +760,8 @@ beginseg
     include "$(BUILD_DIR)/src/audio/lib/effects.o"
     include "$(BUILD_DIR)/src/audio/lib/seqplayer.o"
     include "$(BUILD_DIR)/src/audio/lib/stack.o"
-    include "$(BUILD_DIR)/asm/code/code_8019AE40.text.o" // handwritten
-    pad_text
-    include "$(BUILD_DIR)/asm/code/code_8019AEC0.text.o" // handwritten
+    include "$(BUILD_DIR)/src/audio/code_8019AE40.o"
+    include "$(BUILD_DIR)/src/audio/code_8019AEC0.o"
     include "$(BUILD_DIR)/src/audio/code_8019AF00.o"
     include "$(BUILD_DIR)/src/audio/voice_external.o"
     include "$(BUILD_DIR)/src/audio/voice_internal.o"
diff --git a/src/audio/code_8019AE40.s b/src/audio/code_8019AE40.s
new file mode 100644
index 000000000..f9c6bb2d5
--- /dev/null
+++ b/src/audio/code_8019AE40.s
@@ -0,0 +1,39 @@
+#include "PR/asm.h"
+#include "PR/regdef.h"
+
+.text
+
+/* void func_8019AE40(UNK* a0, UNK* a1, u16 a2, s32 a3) */
+LEAF(func_8019AE40)
+    andi    a2, a2, 0xFFFF
+    srl     a2, a2, 1
+    lw      t0, 0x28(a0)
+    lw      t1, 0x2C(a0)
+    lw      t2, 0xC(a1)
+    bnez    a3, 1f
+    sll     t2, t2, 1
+    add     t0, t0, t2
+    add     t1, t1, t2
+1:
+    lw      t2, 0x4(a1)
+    lw      t3, 0x8(a1)
+    lbu     t4, 0x4(a0)
+    sll     a3, a3, 1
+    add     t2, t2, a3
+    add     t3, t3, a3
+    sll     t4, t4, 1
+2:
+.set noreorder
+    lh      t5, (t2)
+    lh      t6, (t3)
+    add     t2, t2, t4
+    add     t3, t3, t4
+    addi    a2, a2, -1
+    sh      t5, (t0)
+    addi    t0, t0, 2
+    sh      t6, (t1)
+.set reorder
+    addi    t1, t1, 2
+    bgtz    a2, 2b
+    jr      ra
+END(func_8019AE40)
diff --git a/src/audio/code_8019AEC0.s b/src/audio/code_8019AEC0.s
new file mode 100644
index 000000000..817879895
--- /dev/null
+++ b/src/audio/code_8019AEC0.s
@@ -0,0 +1,26 @@
+#include "PR/asm.h"
+#include "PR/regdef.h"
+
+.text
+.align 5
+
+/* Copy 0x20 bytes at a1 to a0, with word alignment */
+/* void func_8019AEC0(u32 dst[8], u32 src[8]) */
+LEAF(func_8019AEC0)
+    addi    t7, a1, 0x20
+1:
+.set noreorder
+    lw      t0, 0x0(a1)
+    lw      t1, 0x4(a1)
+    lw      t2, 0x8(a1)
+    lw      t3, 0xC(a1)
+    addi    a1, a1, 0x10
+    sw      t0, 0x0(a0)
+    sw      t1, 0x4(a0)
+    sw      t2, 0x8(a0)
+    sw      t3, 0xC(a0)
+.set reorder
+    addi    a0, a0, 0x10
+    bne     a1, t7, 1b
+    jr      ra
+END(func_8019AEC0)
diff --git a/src/audio/lib/stack.s b/src/audio/lib/stack.s
deleted file mode 100644
index 831b1b159..000000000
--- a/src/audio/lib/stack.s
+++ /dev/null
@@ -1,6 +0,0 @@
-.include "macro.inc"
-
-.section .data
-
-dlabel aspMainStack
-    .incbin "incbin/aspMainStack"
diff --git a/src/code/kanread.s b/src/code/kanread.s
new file mode 100644
index 000000000..13851760e
--- /dev/null
+++ b/src/code/kanread.s
@@ -0,0 +1,205 @@
+#include "PR/asm.h"
+#include "PR/regdef.h"
+
+.text
+
+/**
+ * s32 Kanji_OffsetFromShiftJIS(s32 sjis);
+ *
+ * Returns the offset of the glyph texture data in the file `kanji` corresponding
+ * to a given 2-byte Shift-JIS codepoint. No range validity check is carried out.
+ *
+ * A nice Shift-JIS codepoint table: https://uic.io/en/charset/show/shift_jis/
+ * The file `kanji` contains the 'Level 1' kanji (0x889F-0x9872), and a reworked
+ * version of the non-kanji section that includes extra English and Hylian glyphs.
+ *
+ * @note This function assumes that its argument is a valid Shift-JIS codepoint;
+ * there is no range protection at all.
+ *
+ * @param sjis Shift-JIS Codepoint of glyph.
+ * @return s32 offset into `kanji` file.
+ *
+ * @remark Almost identical to "LeoGetKadr" from libleo.
+ */
+LEAF(Kanji_OffsetFromShiftJIS)
+    /* Characters with codepoints >= 0x8800 are kanji. Arrangement is regular, */
+    /* so convert index directly. */
+    li      a2, 0xBC
+    blt     a0, 0x8800, .nonkanji
+     /* 0xBC is number of glyphs in one block in the `kanji` file: */
+     /* 0x100 possible codepoints with the same byte1 */
+     /* - 0x40 unused at beginning */
+     /* - 1 unused at 0x7F */
+     /* - 3 unused at 0xFD, 0xFE, 0xFF */
+    /* Get byte1 and adjust so starts at 0 */
+    srl     a1, a0, 8
+    addi    a1, a1, -0x88
+    /* Get byte2 and adjust so starts at 0 */
+    andi    a3, a0, 0xFF
+    addi    a3, a3, -0x40
+    mul     a2, a2, a1
+    blt     a3, 0x40, .kanji_lower_halfblock
+
+    /* 0x__7F is always empty and elided in the file, so decrement if larger */
+    addi    a3, a3, -1
+.kanji_lower_halfblock:
+    mflo    a2  /* Unncessary mflo: the pseudo-op `mul` already performs mflo */
+    addi    a3, a3, 0x30A
+    add     a3, a3, a2
+    /* returns (0x30A + (adjusted byte2) + (adjusted byte1) * 0xBC) * FONT_CHAR_TEX_SIZE */
+    sll     v0, a3, 7
+    jr      ra
+
+/* Non-kanji are arranged with irregular gaps, use the lookup table. */
+.nonkanji:
+    /* Get byte1 and adjust so starts at 0 */
+    srl     a1, a0, 8
+    addi    a1, a1, -0x81
+    /* Get byte2 and adjust so starts at 0 */
+    andi    a3, a0, 0xFF
+    addi    a3, a3, -0x40
+    mul     a2, a2, a1
+    blt     a3, 0x40, .nonkanji_lower_halfblock
+
+    /* 0x__7F is always empty and elided in the file, so decrement if larger */
+    addi    a3, a3, -1
+.nonkanji_lower_halfblock:
+    mflo    a2  /* Unncessary mflo: the pseudo-op `mul` already performs mflo */
+    add     a3, a3, a2
+    sll     a3, a3, 1
+    la      a2, sNonKanjiIndices
+    add     a3, a3, a2
+    lh      a2, (a3)
+    /* returns sNonKanjiIndices[(adjusted byte2) + (adjusted byte1) * 0xBC] * FONT_CHAR_TEX_SIZE */
+    sll     v0, a2, 7
+    jr      ra
+END(Kanji_OffsetFromShiftJIS)
+
+/**
+ * Nearly equivalent C code (Equivalent for all valid input, will behave differently on overflow due to the use of `add` in the original):
+ *
+ * extern u16 sNonKanjiIndices[];
+ *
+ * s32 Kanji_OffsetFromShiftJIS(s32 sjis) {
+ *     u32 byte1 = (u32)sjis >> 8;
+ *     u32 byte2 = sjis & 0xFF;
+ *
+ *     byte2 -= 0x40;
+ *     if (byte2 >= 0x40) {
+ *         byte2--;
+ *     }
+ *
+ *     if (sjis >= 0x8800) {
+ *         byte1 -= 0x88;
+ *         return (0x30A + byte2 + byte1 * 0xBC) * FONT_CHAR_TEX_SIZE;
+ *     } else {
+ *         byte1 -= 0x81;
+ *         return sNonKanjiIndices[byte2 + byte1 * 0xBC] * FONT_CHAR_TEX_SIZE;
+ *     }
+ * }
+ */
+
+/**
+ * Index table, formatting follows conventional Shift-JIS character table layout,
+ * with space for the missing 0x__7F,0x__FD,0x__FE,0x__FF entries, and split
+ * into blocks by high byte.
+ */
+DATA(sNonKanjiIndices)
+/*                 0x___0  0x___1  0x___2  0x___3  0x___4  0x___5  0x___6  0x___7  0x___8  0x___9  0x___A  0x___B  0x___C  0x___D  0x___E  0x___F */
+/* 0x814_ */ .half 0x0000, 0x0001, 0x0002, 0x0003, 0x0004, 0x0005, 0x0006, 0x0007, 0x0008, 0x0009, 0x000A, 0x000B, 0x000C, 0x000D, 0x000E, 0x000F
+/* 0x815_ */ .half 0x0010, 0x0011, 0x0012, 0x0013, 0x0014, 0x0015, 0x0016, 0x0017, 0x0018, 0x0019, 0x001A, 0x001B, 0x001C, 0x001D, 0x001E, 0x001F
+/* 0x816_ */ .half 0x0020, 0x0021, 0x0022, 0x0023, 0x0024, 0x0025, 0x0026, 0x0027, 0x0028, 0x0029, 0x002A, 0x002B, 0x002C, 0x002D, 0x002E, 0x002F
+/* 0x817_ */ .half 0x0030, 0x0031, 0x0032, 0x0033, 0x0034, 0x0035, 0x0036, 0x0037, 0x0038, 0x0039, 0x003A, 0x003B, 0x003C, 0x003D, 0x003E
+/* 0x818_ */ .half 0x003F, 0x0040, 0x0041, 0x0042, 0x0043, 0x0044, 0x0045, 0x0046, 0x0047, 0x0048, 0x0049, 0x004A, 0x004B, 0x004C, 0x004D, 0x004E
+/* 0x819_ */ .half 0x004F, 0x0050, 0x0051, 0x0052, 0x0053, 0x0054, 0x0055, 0x0056, 0x0057, 0x0058, 0x0059, 0x005A, 0x005B, 0x005C, 0x005D, 0x005E
+/* 0x81A_ */ .half 0x005F, 0x0060, 0x0061, 0x0062, 0x0063, 0x0064, 0x0065, 0x0066, 0x0067, 0x0068, 0x0069, 0x006A, 0x006B, 0x0000, 0x0000, 0x0000
+/* 0x81B_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x006C, 0x006D, 0x006E, 0x006F, 0x0070, 0x0071, 0x0072, 0x0073
+/* 0x81C_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0074, 0x0075, 0x0076, 0x0077, 0x0078, 0x0079, 0x007A, 0x0000
+/* 0x81D_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x007B, 0x007C, 0x007D, 0x007E, 0x007F
+/* 0x81E_ */ .half 0x0080, 0x0081, 0x0082, 0x0083, 0x0084, 0x0085, 0x0086, 0x0087, 0x0088, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x81F_ */ .half 0x0089, 0x008A, 0x008B, 0x008C, 0x008D, 0x008E, 0x008F, 0x0090, 0x0000, 0x0000, 0x0000, 0x0000, 0x0091
+
+/*                 0x___0  0x___1  0x___2  0x___3  0x___4  0x___5  0x___6  0x___7  0x___8  0x___9  0x___A  0x___B  0x___C  0x___D  0x___E  0x___F */
+/* 0x824_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0092
+/* 0x825_ */ .half 0x0093, 0x0094, 0x0095, 0x0096, 0x0097, 0x0098, 0x0099, 0x009A, 0x009B, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x826_ */ .half 0x009C, 0x009D, 0x009E, 0x009F, 0x00A0, 0x00A1, 0x00A2, 0x00A3, 0x00A4, 0x00A5, 0x00A6, 0x00A7, 0x00A8, 0x00A9, 0x00AA, 0x00AB
+/* 0x827_ */ .half 0x00AC, 0x00AD, 0x00AE, 0x00AF, 0x00B0, 0x00B1, 0x00B2, 0x00B3, 0x00B4, 0x00B5, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x828_ */ .half 0x0000, 0x00B6, 0x00B7, 0x00B8, 0x00B9, 0x00BA, 0x00BB, 0x00BC, 0x00BD, 0x00BE, 0x00BF, 0x00C0, 0x00C1, 0x00C2, 0x00C3, 0x00C4
+/* 0x829_ */ .half 0x00C5, 0x00C6, 0x00C7, 0x00C8, 0x00C9, 0x00CA, 0x00CB, 0x00CC, 0x00CD, 0x00CE, 0x00CF, 0x0000, 0x0000, 0x0000, 0x0000, 0x00D0
+/* 0x82A_ */ .half 0x00D1, 0x00D2, 0x00D3, 0x00D4, 0x00D5, 0x00D6, 0x00D7, 0x00D8, 0x00D9, 0x00DA, 0x00DB, 0x00DC, 0x00DD, 0x00DE, 0x00DF, 0x00E0
+/* 0x82B_ */ .half 0x00E1, 0x00E2, 0x00E3, 0x00E4, 0x00E5, 0x00E6, 0x00E7, 0x00E8, 0x00E9, 0x00EA, 0x00EB, 0x00EC, 0x00ED, 0x00EE, 0x00EF, 0x00F0
+/* 0x82C_ */ .half 0x00F1, 0x00F2, 0x00F3, 0x00F4, 0x00F5, 0x00F6, 0x00F7, 0x00F8, 0x00F9, 0x00FA, 0x00FB, 0x00FC, 0x00FD, 0x00FE, 0x00FF, 0x0100
+/* 0x82D_ */ .half 0x0101, 0x0102, 0x0103, 0x0104, 0x0105, 0x0106, 0x0107, 0x0108, 0x0109, 0x010A, 0x010B, 0x010C, 0x010D, 0x010E, 0x010F, 0x0110
+/* 0x82E_ */ .half 0x0111, 0x0112, 0x0113, 0x0114, 0x0115, 0x0116, 0x0117, 0x0118, 0x0119, 0x011A, 0x011B, 0x011C, 0x011D, 0x011E, 0x011F, 0x0120
+/* 0x82F_ */ .half 0x0121, 0x0122, 0x0123, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+
+/*                 0x___0  0x___1  0x___2  0x___3  0x___4  0x___5  0x___6  0x___7  0x___8  0x___9  0x___A  0x___B  0x___C  0x___D  0x___E  0x___F */
+/* 0x834_ */ .half 0x0124, 0x0125, 0x0126, 0x0127, 0x0128, 0x0129, 0x012A, 0x012B, 0x012C, 0x012D, 0x012E, 0x012F, 0x0130, 0x0131, 0x0132, 0x0133
+/* 0x835_ */ .half 0x0134, 0x0135, 0x0136, 0x0137, 0x0138, 0x0139, 0x013A, 0x013B, 0x013C, 0x013D, 0x013E, 0x013F, 0x0140, 0x0141, 0x0142, 0x0143
+/* 0x836_ */ .half 0x0144, 0x0145, 0x0146, 0x0147, 0x0148, 0x0149, 0x014A, 0x014B, 0x014C, 0x014D, 0x014E, 0x014F, 0x0150, 0x0151, 0x0152, 0x0153
+/* 0x837_ */ .half 0x0154, 0x0155, 0x0156, 0x0157, 0x0158, 0x0159, 0x015A, 0x015B, 0x015C, 0x015D, 0x015E, 0x015F, 0x0160, 0x0161, 0x0162
+/* 0x838_ */ .half 0x0163, 0x0164, 0x0165, 0x0166, 0x0167, 0x0168, 0x0169, 0x016A, 0x016B, 0x016C, 0x016D, 0x016E, 0x016F, 0x0170, 0x0171, 0x0172
+/* 0x839_ */ .half 0x0173, 0x0174, 0x0175, 0x0176, 0x0177, 0x0178, 0x0179, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x017A
+/* 0x83A_ */ .half 0x017B, 0x017C, 0x017D, 0x017E, 0x017F, 0x0180, 0x0181, 0x0182, 0x0183, 0x0184, 0x0185, 0x0186, 0x0187, 0x0188, 0x0189, 0x018A
+/* 0x83B_ */ .half 0x018B, 0x018C, 0x018D, 0x018E, 0x018F, 0x0190, 0x0191, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0192
+/* 0x83C_ */ .half 0x0193, 0x0194, 0x0195, 0x0196, 0x0197, 0x0198, 0x0199, 0x019A, 0x019B, 0x019C, 0x019D, 0x019E, 0x019F, 0x01A0, 0x01A1, 0x01A2
+/* 0x83D_ */ .half 0x01A3, 0x01A4, 0x01A5, 0x01A6, 0x01A7, 0x01A8, 0x01A9, 0x01AA, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x83E_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x83F_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+
+/*                 0x___0  0x___1  0x___2  0x___3  0x___4  0x___5  0x___6  0x___7  0x___8  0x___9  0x___A  0x___B  0x___C  0x___D  0x___E  0x___F */
+/* 0x844_ */ .half 0x01AB, 0x01AC, 0x01AD, 0x01AE, 0x01AF, 0x01B0, 0x01B1, 0x01B2, 0x01B3, 0x01B4, 0x01B5, 0x01B6, 0x01B7, 0x01B8, 0x01B9, 0x01BA
+/* 0x845_ */ .half 0x01BB, 0x01BC, 0x01BD, 0x01BE, 0x01BF, 0x01C0, 0x01C1, 0x01C2, 0x01C3, 0x01C4, 0x01C5, 0x01C6, 0x01C7, 0x01C8, 0x01C9, 0x01CA
+/* 0x846_ */ .half 0x01CB, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x847_ */ .half 0x01CC, 0x01CD, 0x01CE, 0x01CF, 0x01D0, 0x01D1, 0x01D2, 0x01D3, 0x01D4, 0x01D5, 0x01D6, 0x01D7, 0x01D8, 0x01D9, 0x01DA
+/* 0x848_ */ .half 0x01DB, 0x01DC, 0x01DD, 0x01DE, 0x01DF, 0x01E0, 0x01E1, 0x01E2, 0x01E3, 0x01E4, 0x01E5, 0x01E6, 0x01E7, 0x01E8, 0x01E9, 0x01EA
+/* 0x849_ */ .half 0x01EB, 0x01EC, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x01ED
+/* 0x84A_ */ .half 0x01EE, 0x01EF, 0x01F0, 0x01F1, 0x01F2, 0x01F3, 0x01F4, 0x01F5, 0x01F6, 0x01F7, 0x01F8, 0x01F9, 0x01FA, 0x01FB, 0x01FC, 0x01FD
+/* 0x84B_ */ .half 0x01FE, 0x01FF, 0x0200, 0x0201, 0x0202, 0x0203, 0x0204, 0x0205, 0x0206, 0x0207, 0x0208, 0x0209, 0x020A, 0x020B, 0x020C, 0x020D
+/* 0x84C_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x84D_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x84E_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x84F_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+
+/*                 0x___0  0x___1  0x___2  0x___3  0x___4  0x___5  0x___6  0x___7  0x___8  0x___9  0x___A  0x___B  0x___C  0x___D  0x___E  0x___F */
+/* 0x854_ */ .half 0x020E, 0x020F, 0x0210, 0x0211, 0x0212, 0x0213, 0x0214, 0x0215, 0x0216, 0x0217, 0x0218, 0x0219, 0x021A, 0x021B, 0x021C, 0x021D
+/* 0x855_ */ .half 0x021E, 0x021F, 0x0220, 0x0221, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0222, 0x0223
+/* 0x856_ */ .half 0x0224, 0x0225, 0x0226, 0x0227, 0x0228, 0x0229, 0x022A, 0x022B, 0x022C, 0x022D, 0x022E, 0x022F, 0x0230, 0x0231, 0x0232, 0x0233
+/* 0x857_ */ .half 0x0234, 0x0235, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0236, 0x0237, 0x0238
+/* 0x858_ */ .half 0x0239, 0x023A, 0x023B, 0x023C, 0x023D, 0x023E, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x859_ */ .half 0x0000, 0x023F, 0x0240, 0x0241, 0x0242, 0x0243, 0x0244, 0x0245, 0x0246, 0x0247, 0x0248, 0x0000, 0x0000, 0x0000, 0x0000, 0x0249
+/* 0x85A_ */ .half 0x024A, 0x024B, 0x024C, 0x024D, 0x024E, 0x024F, 0x0250, 0x0251, 0x0252, 0x0253, 0x0254, 0x0255, 0x0256, 0x0257, 0x0000, 0x0000
+/* 0x85B_ */ .half 0x0000, 0x0000, 0x0000, 0x0258, 0x0259, 0x025A, 0x025B, 0x025C, 0x025D, 0x025E, 0x025F, 0x0260, 0x0261, 0x0262, 0x0263, 0x0264
+/* 0x85C_ */ .half 0x0265, 0x0266, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x85D_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0267, 0x0268, 0x0269, 0x026A, 0x026B
+/* 0x85E_ */ .half 0x026C, 0x026D, 0x026E, 0x026F, 0x0270, 0x0271, 0x0272, 0x0273, 0x0274, 0x0275, 0x0276, 0x0277, 0x0278, 0x0279, 0x027A, 0x027B
+/* 0x85F_ */ .half 0x027C, 0x027D, 0x027E, 0x027F, 0x0280, 0x0281, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+
+/*                 0x___0  0x___1  0x___2  0x___3  0x___4  0x___5  0x___6  0x___7  0x___8  0x___9  0x___A  0x___B  0x___C  0x___D  0x___E  0x___F */
+/* 0x864_ */ .half 0x0282, 0x0283, 0x0284, 0x0285, 0x0286, 0x0287, 0x0288, 0x0289, 0x028A, 0x028B, 0x028C, 0x028D, 0x028E, 0x028F, 0x0290, 0x0291
+/* 0x865_ */ .half 0x0292, 0x0293, 0x0294, 0x0295, 0x0296, 0x0297, 0x0298, 0x0299, 0x029A, 0x029B, 0x029C, 0x029D, 0x029E, 0x029F, 0x0000, 0x0000
+/* 0x866_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x867_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x868_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x869_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x02A0, 0x02A1, 0x02A2, 0x02A3, 0x02A4
+/* 0x86A_ */ .half 0x02A5, 0x02A6, 0x02A7, 0x02A8, 0x02A9, 0x02AA, 0x02AB, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x86B_ */ .half 0x0000, 0x0000, 0x0000, 0x02AC, 0x02AD, 0x02AE, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x86C_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x02AF, 0x02B0, 0x02B1, 0x02B2, 0x02B3, 0x02B4, 0x02B5, 0x02B6, 0x02B7
+/* 0x86D_ */ .half 0x02B8, 0x02B9, 0x02BA, 0x02BB, 0x02BC, 0x02BD, 0x02BE, 0x02BF, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x86E_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x86F_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+
+/*                 0x___0  0x___1  0x___2  0x___3  0x___4  0x___5  0x___6  0x___7  0x___8  0x___9  0x___A  0x___B  0x___C  0x___D  0x___E  0x___F */
+/* 0x874_ */ .half 0x02C0, 0x02C1, 0x02C2, 0x02C3, 0x02C4, 0x02C5, 0x02C6, 0x02C7, 0x02C8, 0x02C9, 0x02CA, 0x02CB, 0x02CC, 0x02CD, 0x02CE, 0x02CF
+/* 0x875_ */ .half 0x02D0, 0x02D1, 0x02D2, 0x02D3, 0x02D4, 0x02D5, 0x02D6, 0x02D7, 0x02D8, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x876_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x877_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x878_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x879_ */ .half 0x0000, 0x02D9, 0x02DA, 0x02DB, 0x02DC, 0x02DD, 0x02DE, 0x02DF, 0x02E0, 0x02E1, 0x02E2, 0x02E3, 0x02E4, 0x02E5, 0x02E6, 0x02E7
+/* 0x87A_ */ .half 0x02E8, 0x02E9, 0x02EA, 0x02EB, 0x02EC, 0x02ED, 0x02EE, 0x02EF, 0x02F0, 0x02F1, 0x02F2, 0x02F3, 0x02F4, 0x02F5, 0x02F6, 0x02F7
+/* 0x87B_ */ .half 0x02F8, 0x02F9, 0x02FA, 0x02FB, 0x02FC, 0x02FD, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x02FE, 0x02FF, 0x0300
+/* 0x87C_ */ .half 0x0301, 0x0302, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x87D_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x87E_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0303, 0x0304, 0x0305, 0x0306, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000
+/* 0x87F_ */ .half 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0307, 0x0308, 0x0309
+ENDDATA(sNonKanjiIndices)
diff --git a/src/makerom/entry.s b/src/makerom/entry.s
index e28bc8b50..513e5358a 100644
--- a/src/makerom/entry.s
+++ b/src/makerom/entry.s
@@ -1,34 +1,49 @@
 #include "PR/asm.h"
+#include "PR/regdef.h"
 #include "boot.h"
 
-.set noreorder
+.text
 
-.section .text
+#define PAD_TO 0x60
+#define LA(reg, sym) la reg, sym
 
-.balign 16
-
-LEAF(entrypoint)
-    // Clear boot segment .bss
-    la      $t0, _bootSegmentBssStart
-#ifndef AVOID_UB
-    // UB: li only loads the lower 16 bits of _bootSegmentBssSize when it may be larger than this,
-    // so not all of bss may be cleared if it is too large
-    li      $t1, _bootSegmentBssSize
+#if !defined(AVOID_UB)
+/* Old assembler workarounds: The makerom tool in the N64 SDK was given the bss segment size as a const
+ * literal, and since this literal was < 0x10000 it was loaded in one instruction. We don't have access
+ * to the bss segment size until we link everything so we cannot do the same thing. Instead we must load
+ * only the lower 16 bits of the bss size for matching.
+ * When AVOID_UB is enabled, don't do this and instead load the full symbol value, otherwise not all of
+ * bss may be cleared. */
+# if defined(__sgi)
+#  define LOAD_BSS_SIZE(reg) li reg, %half(_bootSegmentBssSize)
+# else
+#  define LOAD_BSS_SIZE(reg) ori reg, zero, %lo(_bootSegmentBssSize)
+# endif
 #else
-    la      $t1, _bootSegmentBssSize
+# define LOAD_BSS_SIZE(reg) LA(reg, _bootSegmentBssSize)
 #endif
+
+LEAF(entrypoint)
+    /* Clear boot segment .bss */
+    LA(     t0, _bootSegmentBssStart)
+    LOAD_BSS_SIZE(t1)
 .clear_bss:
-    addi    $t1, $t1, -8
-    sw      $zero, 0($t0)
-    sw      $zero, 4($t0)
-    bnez    $t1, .clear_bss
-     addi   $t0, $t0, 8
-    // Set up stack and enter program code
-    lui     $t2, %hi(bootproc)
-    lui     $sp, %hi(sBootStack + BOOT_STACK_SIZE)
-    addiu   $t2, %lo(bootproc)
-    jr      $t2
-     addiu  $sp, %lo(sBootStack + BOOT_STACK_SIZE)
+    sw      zero, 0(t0)
+    sw      zero, 4(t0)
+    addi    t0, t0, 8
+    addi    t1, t1, -8
+    bnez    t1, .clear_bss
+
+    /* Set up stack and enter program code */
+    LA(     sp, sBootStack + BOOT_STACK_SIZE)
+    LA(     t2, bootproc)
+    jr      t2
 END(entrypoint)
 
-.fill 0x60 - (. - entrypoint)
+#ifdef __sgi
+/* IDO can't take absolute differences of symbols so the size of the above is hardcoded */
+.space PAD_TO - 0x34
+#else
+/* Pad to a total size taking into account the size of the above */
+.space PAD_TO - (. - entrypoint)
+#endif
diff --git a/src/makerom/ipl3.s b/src/makerom/ipl3.s
deleted file mode 100644
index c2dc845b5..000000000
--- a/src/makerom/ipl3.s
+++ /dev/null
@@ -1,3 +0,0 @@
-.section .text
-
-.incbin "incbin/ipl3"
diff --git a/tools/disasm/disasm.py b/tools/disasm/disasm.py
index 79d41bd96..dcdde2f8d 100755
--- a/tools/disasm/disasm.py
+++ b/tools/disasm/disasm.py
@@ -122,7 +122,7 @@ def discard_decomped_files(files_spec, include_files):
                             with open(root_path / last_line, "r") as f2:
                                 if "GLOBAL_ASM" in f2.read():
                                     include = True
-                        else:
+                        elif last_line.rsplit(".", 1)[0] not in ("src/makerom/ipl3"):
                             assert os.path.exists(root_path / last_line.replace(".c", ".s"))
 
             include |= force_include