GCC support.

Makefile

@@ -8,10 +8,23 @@ COMPARE ?= 1
 NON_MATCHING ?= 0
 # If ORIG_COMPILER is 1, compile with QEMU_IRIX and the original compiler
 ORIG_COMPILER ?= 0
+# If COMPILER is GCC, compile with GCC instead of IDO.
+COMPILER ?= ido
+# Convert text in source files (.c) to EUC-JP before compilation, but only on GCC.
+# This is required in order to preserve debug print functionality on GCC since it
+# doesn't seem to recognize EUC-JP marked files without doing this beforehand.
+CONV_SOURCE_TEXT ?= 1
+
+# If gcc is used, define the NON_MATCHING flag respectively so the files that
+# are safe to be used can avoid using GLOBAL_ASM which doesn't work with gcc.
+ifeq ($(COMPILER),gcc)
+  NON_MATCHING := 1
+  CPPFLAGS := -DCOMPILER_GCC
+endif
 
 ifeq ($(NON_MATCHING),1)
-  CFLAGS := -DNON_MATCHING
-  CPPFLAGS := -DNON_MATCHING
+  CFLAGS += -DNON_MATCHING
+  CPPFLAGS += -DNON_MATCHING
   COMPARE := 0
 endif
 
@@ -41,8 +54,22 @@ else
   $(error Please install or build mips-linux-gnu)
 endif
 
-CC       := tools/ido_recomp/$(DETECTED_OS)/7.1/cc
-CC_OLD   := tools/ido_recomp/$(DETECTED_OS)/5.3/cc
+# Detect compiler and set variables appropriately.
+ifeq ($(COMPILER),gcc)
+  ifeq ($(CONV_SOURCE_TEXT),1)
+    CC = iconv -f UTF-8 -t EUC-JP $< | mips-linux-gnu-gcc -x c -
+  else
+    CC = mips-linux-gnu-gcc $<
+  endif
+  CC_OLD   := $(CC)
+else 
+ifeq ($(COMPILER),ido)
+  CC       := tools/ido_recomp/$(DETECTED_OS)/7.1/cc
+  CC_OLD   := tools/ido_recomp/$(DETECTED_OS)/5.3/cc
+else
+$(error Unsupported compiler. Please use either ido or gcc as the COMPILER variable.)
+endif
+endif
 
 # if ORIG_COMPILER is 1, check that either QEMU_IRIX is set or qemu-irix package installed
 ifeq ($(ORIG_COMPILER),1)
@@ -77,10 +104,17 @@ ZAPD       := tools/ZAPD/ZAPD.out
 
 OPTFLAGS := -O2
 ASFLAGS := -march=vr4300 -32 -Iinclude
-MIPS_VERSION := -mips2
 
-# we support Microsoft extensions such as anonymous structs, which the compiler does support but warns for their usage. Surpress the warnings with -woff.
-CFLAGS += -G 0 -non_shared -Xfullwarn -Xcpluscomm $(INC) -Wab,-r4300_mul -woff 649,838,712
+ifeq ($(COMPILER),gcc)
+  CFLAGS += -c -G 0 -nostdinc -Iinclude -Isrc -Iassets -Ibuild -I. -DNON_MATCHING=1 -DMODDING=1 -DNORMAL_GAMEPLAY=1 -DAVOID_UB=1 -mno-shared -march=vr4300 -mfix4300 -mabi=32 -mhard-float -mdivide-breaks -fno-stack-protector -fno-common -fno-zero-initialized-in-bss -mno-abicalls -fno-strict-aliasing -fno-inline-functions -fno-inline-small-functions -fno-toplevel-reorder -ffreestanding -fwrapv $(CHECK_WARNINGS) -g -mno-explicit-relocs -mno-split-addresses -funsigned-char
+  MIPS_VERSION := -mips3
+  PIPEIN = 
+else 
+  # we support Microsoft extensions such as anonymous structs, which the compiler does support but warns for their usage. Surpress the warnings with -woff.
+  CFLAGS += -G 0 -non_shared -Xfullwarn -Xcpluscomm $(INC) -Wab,-r4300_mul -woff 649,838,712 
+  MIPS_VERSION := -mips2
+  PIPEIN = $<
+endif
 
 ifeq ($(shell getconf LONG_BIT), 32)
   # Work around memory allocation bug in QEMU
@@ -126,6 +160,7 @@ TEXTURE_FILES_OUT := $(foreach f,$(TEXTURE_FILES_PNG:.png=.inc.c),build/$f) \
 # create build directories
 $(shell mkdir -p build/baserom build/assets/text $(foreach dir,$(SRC_DIRS) $(ASM_DIRS) $(ASSET_BIN_DIRS),build/$(dir)))
 
+ifeq ($(COMPILER),ido)
 build/src/code/fault.o: CFLAGS += -trapuv
 build/src/code/fault.o: OPTFLAGS := -O2 -g3
 build/src/code/fault_drawer.o: CFLAGS += -trapuv
@@ -162,6 +197,7 @@ build/src/code/%.o: CC := python3 tools/asm_processor/build.py $(CC) -- $(AS) $(
 build/src/overlays/%.o: CC := python3 tools/asm_processor/build.py $(CC) -- $(AS) $(ASFLAGS) --
 
 build/assets/%.o: CC := python3 tools/asm_processor/build.py $(CC) -- $(AS) $(ASFLAGS) --
+endif
 
 #### Main Targets ###
 
@@ -231,7 +267,7 @@ build/assets/text/nes_message_data_static.o: build/assets/text/message_data.enc.
 build/assets/text/staff_message_data_static.o: build/assets/text/message_data_staff.enc.h
 
 build/assets/%.o: assets/%.c
-	$(CC) -c $(CFLAGS) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $<
+	$(CC) -c $(CFLAGS) -I $(<D) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $(PIPEIN)
 	$(OBJCOPY) -O binary $@ $@.bin
 
 build/dmadata_table_spec.h: build/$(SPEC)
@@ -241,25 +277,25 @@ build/src/boot/z_std_dma.o: build/dmadata_table_spec.h
 build/src/dmadata/dmadata.o: build/dmadata_table_spec.h
 
 build/src/overlays/%.o: src/overlays/%.c
-	$(CC) -c $(CFLAGS) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $<
+	$(CC) -c $(CFLAGS) -I $(<D) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $(PIPEIN)
 	$(CC_CHECK) $<
-	$(ZAPD) bovl -eh -i $@ -cfg $< --outputpath $(@D)/$(notdir $(@D))_reloc.s
+	$(ZAPD) bovl -eh -i $@ -cfg $< --outputpath $(@D)/$(notdir $(@D))_reloc.s --gcc-compat
 	-test -f $(@D)/$(notdir $(@D))_reloc.s && $(AS) $(ASFLAGS) $(@D)/$(notdir $(@D))_reloc.s -o $(@D)/$(notdir $(@D))_reloc.o
 	@$(OBJDUMP) -d $@ > $(@:.o=.s)
 
 build/src/%.o: src/%.c
-	$(CC) -c $(CFLAGS) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $<
+	$(CC) -c $(CFLAGS) -I $(<D) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $(PIPEIN)
 	$(CC_CHECK) $<
 	@$(OBJDUMP) -d $@ > $(@:.o=.s)
 
 build/src/libultra/libc/ll.o: src/libultra/libc/ll.c
-	$(CC) -c $(CFLAGS) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $<
+	$(CC) -c $(CFLAGS) -I $(<D) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $(PIPEIN)
 	$(CC_CHECK) $<
 	python3 tools/set_o32abi_bit.py $@
 	@$(OBJDUMP) -d $@ > $(@:.o=.s)
 
 build/src/libultra/libc/llcvt.o: src/libultra/libc/llcvt.c
-	$(CC) -c $(CFLAGS) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $<
+	$(CC) -c $(CFLAGS) -I $(<D) $(MIPS_VERSION) $(OPTFLAGS) -o $@ $(PIPEIN)
 	$(CC_CHECK) $<
 	python3 tools/set_o32abi_bit.py $@
 	@$(OBJDUMP) -d $@ > $(@:.o=.s)

asm/llmuldiv_gcc.s

@@ -0,0 +1,103 @@
+.include "macro.inc"
+
+# assembler directives
+.set noat      # allow manual use of $at
+.set noreorder # don't insert nops after branches
+.set gp=64     # allow use of 64-bit general purpose registers
+
+.section .text
+
+.balign 16
+
+/* Copied from Super Mario 64 decomp project */
+
+/* -------------------------------------------------------------------------------------- */
+/* need to asm these functions because lib32gcc-7-dev-mips-cross does not exist so we     */
+/* cannot naturally link a libgcc variant for this target given this architecture and     */
+/* compiler. Until we have a good workaround with a gcc target that doesn't involve       */
+/* assuming a 32-bit to 64-bit change, we have to encode these functions as raw assembly  */
+/* for it to compile.                                                                     */
+/* -------------------------------------------------------------------------------------- */
+
+/* TODO: Is there a non-insane way to fix this hack that doesn't involve the user compiling */
+/* a library themselves?                                                                    */
+glabel __umoddi3
+    sw    $a0, ($sp)
+    sw    $a1, 4($sp)
+    sw    $a2, 8($sp)
+    sw    $a3, 0xc($sp)
+    ld    $t7, 8($sp)
+    ld    $t6, ($sp)
+    ddivu $zero, $t6, $t7
+    bnez  $t7, .L80324144
+     nop
+    break 7
+.L80324144:
+    mfhi  $v0
+    dsll32 $v1, $v0, 0
+    dsra32 $v1, $v1, 0
+    jr    $ra
+     dsra32 $v0, $v0, 0
+
+glabel __udivdi3
+    sw    $a0, ($sp)
+    sw    $a1, 4($sp)
+    sw    $a2, 8($sp)
+    sw    $a3, 0xc($sp)
+    ld    $t7, 8($sp)
+    ld    $t6, ($sp)
+    ddivu $zero, $t6, $t7
+    bnez  $t7, .L80324180
+     nop
+    break 7
+.L80324180:
+    mflo  $v0
+    dsll32 $v1, $v0, 0
+    dsra32 $v1, $v1, 0
+    jr    $ra
+     dsra32 $v0, $v0, 0
+
+glabel __moddi3
+    sw    $a0, ($sp)
+    sw    $a1, 4($sp)
+    sw    $a2, 8($sp)
+    sw    $a3, 0xc($sp)
+    ld    $t7, 8($sp)
+    ld    $t6, ($sp)
+    ddivu $zero, $t6, $t7
+    bnez  $t7, .L803241E8
+     nop
+    break 7
+.L803241E8:
+    mfhi  $v0
+    dsll32 $v1, $v0, 0
+    dsra32 $v1, $v1, 0
+    jr    $ra
+     dsra32 $v0, $v0, 0
+
+glabel __divdi3
+    sw    $a0, ($sp)
+    sw    $a1, 4($sp)
+    sw    $a2, 8($sp)
+    sw    $a3, 0xc($sp)
+    ld    $t7, 8($sp)
+    ld    $t6, ($sp)
+    ddiv  $zero, $t6, $t7
+    nop
+    bnez  $t7, .L80324228
+     nop
+    break 7
+.L80324228:
+    daddiu $at, $zero, -1
+    bne   $t7, $at, .L80324244
+     daddiu $at, $zero, 1
+    dsll32 $at, $at, 0x1f
+    bne   $t6, $at, .L80324244
+     nop
+    break 6
+.L80324244:
+    mflo  $v0
+    dsll32 $v1, $v0, 0
+    dsra32 $v1, $v1, 0
+    jr    $ra
+     dsra32 $v0, $v0, 0

include/functions.h

@@ -4,7 +4,14 @@
 #include "z64.h"
 
 f32 fabsf(f32 f);
+#ifndef __sgi
+#define fabsf __builtin_fabsf
+f32 __floatundisf(u32 c);
+f64 __floatundidf(u32 c); 
+unsigned long __udivdi3(unsigned long a, unsigned long b);
+#else
 #pragma intrinsic(fabsf)
+#endif
 f32 sqrtf(f32 f);
 #pragma intrinsic(sqrtf)
 f64 sqrt(f64 d);

spec

@@ -109,6 +109,10 @@ beginseg
     include "build/src/libultra/os/gethwintrroutine.o"
     include "build/asm/__osSetWatchLo.o"
     include "build/data/rsp_boot.text.o"
+#ifdef COMPILER_GCC
+    include "build/asm/llmuldiv_gcc.o"
+    include "build/src/boot/missing_gcc_functions.o"
+#endif
 endseg
 
 beginseg

src/boot/missing_gcc_functions.c

@@ -0,0 +1,68 @@
+#include "global.h"
+
+// Define functions needed for the GCC build here.
+
+// Self-hosted memcmp.
+int memcmp(void *s1, const void *s2, size_t n) {
+    u8 *m1 = (u8 *)s1;
+    u8 *m2 = (u8 *)s2;
+    int i;
+    for (i = 0; i < n; i++) {
+        if (m1[i] < m2[i]) {
+            return -1;
+        } else if (m1[i] > m2[i]) {
+            return 1;
+        }
+    }
+    return 0;
+}
+
+void *memset(void *str, int c, size_t n) {
+    u8 *m1 = (u8 *)str;
+    int i;
+    for(i = 0; i < n; i++) {
+        m1[i] = c;
+    }
+    return str;
+}
+
+// These functions convert c to an unsigned integer, rounding toward zero. Negative values
+// all become zero.
+u32 __fixunssfdi(f32 a) {
+    if (a < 0.0f)
+        a = 0.0f;
+    return (u32)a;
+}
+
+u32 __fixunsdfdi(f64 a) {
+    if (a < 0.0)
+        a = 0.0;
+    return (u32)a;
+}
+
+// These functions convert c to a signed integer, rounding toward zero.
+s32 __fixsfdi(f32 c) {
+    return (s32)c;
+}
+
+s32 __fixdfdi(f64 c) {
+    return (s32)c;
+}
+
+// These functions convert c, a signed integer, to floating point.
+f32 __floatdisf(s32 c) {
+    return (f32)c;
+}
+
+f64 __floatdidf(s32 c) {
+    return (f64)c;
+}
+
+// These functions convert c, an unsigned integer, to floating point.
+f32 __floatundisf(u32 c) {
+    return (f32)c;    
+}
+
+f64 __floatundidf(u32 c) {
+    return (f64)c;
+}

src/code/audio_heap.c

@@ -1021,7 +1021,9 @@ void* AudioHeap_AllocPermanent(s32 tableType, s32 id, u32 size) {
     gAudioContext.permanentCache[index].size = size;
     //! @bug UB: missing return. "ret" is in v0 at this point, but doing an
     // explicit return uses an additional register.
-    // return ret;
+#ifdef AVOID_UB
+    return ret;
+#endif
 }
 
 void* AudioHeap_AllocSampleCache(u32 size, s32 fontId, void* sampleAddr, s8 medium, s32 cache) {

src/overlays/actors/ovl_En_Tana/z_en_tana.c

@@ -35,6 +35,9 @@ const ActorInit En_Tana_InitVars = {
 static const char* sShelfTypes[] = {
     "木の棚", // "Wooden Shelves"
     "石の棚", // "Stone Shelves"
+#ifdef AVOID_UB
+    "",
+#endif
 };
 
 static const ActorFunc sDrawFuncs[] = {