Files
banjo-kazooie/src/core2/code_B9770.c
2024-10-16 21:50:24 +02:00

1316 lines
38 KiB
C

#include <ultra64.h>
#include "functions.h"
#include "variables.h"
extern void sfxsource_setSampleRate(u8, s32);
extern bool func_80323240(struct56s *, f32, f32[3]);
extern f32 func_803234FC(struct56s *, f32, f32);
extern f32 func_80323540(struct56s *, f32, f32, f32);
extern f32 func_803237E8(struct56s *);
extern f32 func_80323F74(struct56s *, f32, f32);
extern f32 func_80323FDC(struct56s *, f32, f32, s32 *);
extern f32 func_803240E0(struct56s *, f32, f32, s32 *);
extern void func_8032417C(struct56s *, f32, f32[3], f32[3]);
extern void func_80328FF0(Actor *, f32);
extern void func_8032DC70(s32);
typedef struct {
f32 unk0;
u8 pad4[0x8];
struct{
u32 pad_bit31: 9;
u32 bit22: 11;
u32 pad_bit11: 12;
}unkC;
} Struct_glspline_0;
typedef struct {
u32 bit31: 16;
u32 bit15: 12;
u32 bit3: 3;
u32 bit0: 1;
}Struct_glspline_t0_unk4_common;
typedef struct {
u32 bit31: 1;
u32 bit30: 15;
u32 bit15: 8;
u32 bit7: 4;
u32 bit3: 3;
u32 bit0: 1;
}Struct_glspline_t0_unk4_t7;
typedef struct{
u8 pad0[0x4];
union
{
Struct_glspline_t0_unk4_common common;
Struct_glspline_t0_unk4_t7 t7;
}unk4;
struct {
u32 bit31:24;
u32 pad4_7:8;
}unk8;
struct{
u32 pad_bit31:8;
u32 bit23:2;
u32 bit21:1;
u32 bit20:8;
u32 bit12:12;
u32 bit0:1;
}unkC;
union {
struct {
u32 pad_bit31: 24;
u32 bit7:7;
u32 bit0:1;
} common;
struct {
u32 pad_bit31: 24;
u32 bit7:6;
u32 pad_bit1:1;
u32 bit0:1;
}t7;
u32 word;
} unk10;
}Struct_glspline_t0;
typedef struct{
f32 unk0;
struct{
u32 bit31: 12;
u32 bit19: 2;
u32 bit17: 2;
u32 bit15: 9;
u32 bit6: 3;
u32 bit3: 3;
u32 bit0: 1;
}unk4;
struct{
u32 bit31: 10;
u32 bit21: 11;
u32 bit10: 1;
u32 bit9: 1;
u32 bit8: 1;
u32 pad_bit7: 8;
} unk8;
struct{
// u32 padC_31:8;
// u32 unkC_23:2;
// u32 unkC_21:1;
// u32 unkC_20:8;
u32 bit31:9;
u32 bit22:11;
u32 bit11:11;
u32 bit0 : 1;
}unkC;
struct {
u32 bit31: 12;
u32 pad_bit19: 13;
u32 bit6: 1;
u32 bit5: 1;
u32 bit4: 1;
u32 pad_bit3:3;
}unk10;
}Struct_glspline_t1;
typedef struct{
f32 unk0;
u8 pad0[0x8];
struct{
u32 pad_bit31:31;
u32 bit0:1;
}unkC;
struct {
u32 pad_31:31;
u32 bit0:1;
}unk10;
}Struct_glspline_common;
typedef union{
Struct_glspline_common common;
Struct_glspline_t0 t0;
Struct_glspline_t1 t1;
/* data */
}Union_glspline;
typedef struct{
s32 unk0;
Union_glspline spline[];
}SplineList;
typedef struct {
s16 unk0;
s16 unk2[3];
u16 unk8_15:1;
u16 unk8_14:1;
u16 unk8_13:1;
u16 pad8_12:13;
}Struct_glspline_803411B0;
s32 func_80341BC8(struct56s *arg0, SplineList * arg1);
/* .data */
struct56s **D_80371E70 = NULL;
SplineList **D_80371E74 = NULL;
s32 D_80371E78 = 0;
s32 D_80371E7C = 0;
s32 D_80371E80 = 0;
u8 D_80371E84[] = {
0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0xFF,
0xFF, 0xFF, 0x00, 0xFF, 0x00, 0xFF, 0x00, 0xFF, 0xFF, 0x80, 0x80, 0x80, 0xFF, 0x80, 0x00, 0xFF,
0x00, 0x80, 0x80, 0x80, 0x00, 0x00, 0x80, 0x80, 0xFF, 0x40, 0x00, 0x00, 0x40, 0x80, 0x00, 0x00,
0x40, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
};
/* .bss */
s16 *D_803858A0;
/* .h */
struct56s *func_80342038(s32 indx);
/* .code */
f32 func_80340700(f32 value, f32 min, f32 max) {
return (value < min) ? min
: (max < value) ? max
: value;
}
bool func_80340748(s32 arg0, s32 arg1, s32 arg2, f32 arg3[3], s32 arg4, s32 arg5, s32 arg6){
return FALSE;
}
s32 func_80340760(s32 arg0, s32 *arg1, f32 *arg2, s32 arg3, s32 arg4, f32 *arg5, f32 *arg6) {
SplineList **temp_t0 = D_80371E74;
SplineList *temp_a0;
Union_glspline *temp_v0;
Union_glspline *var_v1;
s32 var_a2;
*arg5 = -9999.0f;
*arg6 = -9999.0f;
if (arg4 != -1) {
temp_a0 = temp_t0[arg4];
var_v1 = (temp_a0 + 1);
temp_v0 = var_v1 + temp_a0->unk0;
for(var_v1 = var_v1; (arg0 != var_v1->t1.unk10.bit31) && (var_v1 < temp_v0); var_v1++){
continue;
}
if (var_v1 < temp_v0) {
if (var_v1->t1.unk4.bit0 == 1) {
if (!var_v1->t1.unkC.bit0 && (var_v1->t1.unk4.bit6 == 7)) {
*arg5 = (f32)var_v1->t1.unkC.bit31;
*arg6 = (f32)var_v1->t1.unk4.bit15;
}
}
*arg1 = arg4;
*arg2 = var_v1->t1.unk0;
func_80323240(func_80342038(arg4), var_v1->t1.unk0, arg3);
return 1;
}
}
for(var_a2 = 0; var_a2 < D_80371E78; var_a2++){
if (var_a2 != arg4) {
temp_a0 = temp_t0[var_a2];
var_v1 = (temp_a0 + 1);
temp_v0 = var_v1 + temp_a0->unk0;
for(var_v1 = var_v1; (arg0 != var_v1->t1.unk10.bit31) && (var_v1 < temp_v0); var_v1++){
continue;
}
if (var_v1 < temp_v0) {
if (var_v1->t1.unk4.bit0 == 1) {
if (!var_v1->t1.unkC.bit0 && (var_v1->t1.unk4.bit6 == 7)) {
*arg5 = (f32)var_v1->t1.unkC.bit31;
*arg6 = (f32)var_v1->t1.unk4.bit15;
}
}
*arg1 = var_a2;
*arg2 = var_v1->t1.unk0;
func_80323240(func_80342038(var_a2), var_v1->t1.unk0, arg3);
return 1;
}
}
}
return 0;
}
f32 glspline_catmull_rom_interpolate(f32 x, s32 knotCount, f32 *knotList) {
s32 start_knot;
s32 tmp_t7;
f32 sp24[3];
f32 *var_a0;
tmp_t7 = knotCount - 3;
start_knot = x = tmp_t7 * func_80340700(x, 0.0f, 1.0f);
start_knot = (start_knot >= (knotCount - 4)) ? (knotCount - 4) : (start_knot);
knotList += start_knot;
x -= start_knot;
sp24[2] = -0.5*knotList[0] + 1.5*knotList[1] + -1.5*knotList[2] + 0.5*knotList[3];
sp24[1] = 1.0*knotList[0] + -2.5*knotList[1] + 2.0*knotList[2] + -0.5*knotList[3];
sp24[0] = -0.5*knotList[0] + 0.0*knotList[1] + 0.5*knotList[2] + 0.0*knotList[3];
return (((((sp24[2] * x) + sp24[1]) * x) + sp24[0]) * x) + (1.0*knotList[1]);
}
void func_80340BE4(f32 x, s32 length, s32 stride, s32 width, f32 *arg4, f32 dst[3]){
s32 padding;
s32 max_interval;
s32 a3;
f32 temp_f2;
f32 f3;
f32 f1;
f32 f2;
f32 *t0;
f32 sp94[3][3];
max_interval = length - 1;
if (length < 4) {
/* not enough points in spline to perform catmull-rom interpolation*/
ml_vec3f_copy(sp94[0], arg4);
ml_vec3f_copy(sp94[1], arg4);
ml_vec3f_copy(sp94[2], &arg4[stride]);
if (length - 1 == 1) {
ml_vec3f_copy(sp94[3], &arg4[stride]);
} else {
ml_vec3f_copy(sp94[3], &arg4[stride*2]);
}
func_80340BE4(x, 4, 3, 3, (f32 *) sp94, dst);
return;
}
temp_f2 = func_80340700(x, 0.0f, 1.0f) * max_interval;
a3 = temp_f2;
if (a3 > (length - 1)) {
a3 = (length - 1);
}
x = temp_f2 - a3;
if (a3 == 0) {
/* desired point in interval[0,1], duplicate P0 to left */
t0 = arg4;
t0 += (stride * 2);
for (; a3 < width; a3++) {
f3 = -0.5*arg4[a3] + 1.5*arg4[a3] + -1.5*arg4[stride + a3] + 0.5*t0[a3];
f1 = 1.0*arg4[a3] + -2.5*arg4[a3] + 2.0*arg4[stride + a3] + -0.5*t0[a3];
if (1);
f2 = -0.5*arg4[a3] + 0.0*arg4[a3] + 0.5*arg4[stride + a3] + 0.0*t0[a3];
dst[a3] = (1.0*arg4[a3]) + (((f3 * x + f1) * x + f2) * x);
}
} else {
a3--;
if (a3 == length - 2) {
arg4 += stride * (length - 1);
for (a3 = 0; a3 < width; a3++) {
/* desired point in interval[n, inf], linearly interpret */
dst[a3] = arg4[a3];
}
} else if (a3 == length - 3) {
/* desired point in interval[n-1, n], duplicate Pn to right */
arg4 += a3 * stride;
t0 = arg4;
t0 += (stride * 2);
for (a3 = 0; a3 < width; a3++) {
f3 = -0.5*arg4[a3] + 1.5*arg4[stride + a3] + -1.5*t0[a3] + 0.5*t0[a3];
f1 = 1.0*arg4[a3] + -2.5*arg4[stride + a3] + 2.0*t0[a3] + -0.5*t0[a3];
f2 = -0.5*arg4[a3] + 0.0*arg4[stride + a3] + 0.5*t0[a3] + 0.0*t0[a3];
dst[a3] = (1.0*arg4[stride + a3]) + (((f3 * x + f1) * x + f2) * x);
}
} else {
/* normal conditions */
arg4 += a3 * stride;
t0 = arg4;
t0 += (stride * 2);
for (a3 = 0; a3 < width; a3++) {
f3 = -0.5*arg4[a3] + 1.5*arg4[stride + a3] + -1.5*t0[a3] + 0.5*arg4[stride*3 + a3];
f1 = 1.0*arg4[a3] + -2.5*arg4[stride + a3] + 2.0*t0[a3] + -0.5*arg4[stride*3 + a3];
f2 = -0.5*arg4[a3] + 0.0*arg4[stride + a3] + 0.5*t0[a3] + 0.0*arg4[stride*3 + a3];
dst[a3] = (1.0*arg4[stride + a3]) + (((f3 * x + f1) * x + f2) * x);
}
}
}
}
void func_80341180(f32 arg0, s32 arg1, s32 arg2, f32 arg3[3], f32 arg4[3]){
func_80340BE4(arg0, arg1, arg2, arg2, arg3, arg4);
}
void func_803411B0(void){
s32 spE4;
s32 spE0;
s32 a0;
Struct_glspline_803411B0 *spD8;
s32 spCC[3];
s32 spC8;
s32 spC4;
s32 padding[3];
struct56s *spB4;
Struct_glspline_803411B0 *spline;
Union_glspline *var_s0_2;
SplineList *spA8;
f32 *spA4;
Union_glspline *var_s1_2;
s32 var_fp;
s32 var_s0;
s32 var_s2;
Union_glspline sp80;
s32 var_s5;
u32 var_s7;
Union_glspline *temp_s3_2;
Union_glspline *temp_v0_16;
s32 tmp;
D_80371E80 = 0;
D_80371E70 = malloc(0);
D_80371E74 = malloc(0);
D_80371E78 = 0;
D_803858A0 = malloc(128 * sizeof(s16));
for (spE0 = 0; spE0 < 128; spE0++) {
D_803858A0[spE0] = 0;
}
spE4 = func_80307E1C() + 1;
if (spE4 <= 1) {
return;
}
spD8 = (Struct_glspline_803411B0 *) malloc(spE4 * sizeof(Struct_glspline_803411B0));
for (spE0 = 0; spE0 < spE4; spE0++) {
(spD8 + spE0)->unk0 = -1;
}
func_80307EA8(0, spCC, &spC8, &spC4);
do {
spE0 = func_80307EA8(1, spCC, &spC8, &spC4);
if (spE0 >= 0) {
(spD8+spE0)->unk0 = spC8;
(spD8+spE0)->unk8_13 = spC4;
(spD8+spE0)->unk2[0] = spCC[0];
(spD8+spE0)->unk2[1] = spCC[1];
(spD8+spE0)->unk2[2] = spCC[2];
}
} while (spE0 >= 0);
for (spE0 = 1; spE0 < spE4; spE0++) {
(spD8+spE0)->unk8_15 = 0;
if ((spD8+spE0)->unk0 >= spE4 || (spD8+spE0)->unk0 == -1) {
(spD8+spE0)->unk0 = 0;
}
}
for (spE0 = 1; spE0 < spE4; spE0++) {
if ((spD8+spE0)->unk0 > 0) {
(spD8+(spD8+spE0)->unk0)->unk8_15 = 1;
}
}
for (spE0 = 1; spE0 < spE4; spE0++) {
if ((spD8+spE0)->unk8_13 == 1
&& ((spD8+spE0)->unk8_15 != 0 || (spD8+spE0)->unk0 <= 0)
&& ((spD8+spE0)->unk8_15 != 1 || (spD8+spE0)->unk0 <= 0)
&& ((spD8+spE0)->unk8_15 != 1 || (spD8+spE0)->unk0 != 0)) {
func_8032DC70(spE0);
(spD8+spE0)->unk0 = -1;
(spD8+spE0)->unk8_13 = 0;
}
}
for (spE0 = 1; spE0 < spE4; spE0++) {
if ((spD8+spE0)->unk0 > 0 && (spD8+spE0)->unk8_15 == 0 && (spD8+spE0)->unk8_13 == 0) {
var_s5 = 1;
var_s7 = 1;
var_s0 = 0;
var_fp = 0;
for (; var_s5 < spE4; var_s5++) {
(spD8+var_s5)->unk8_14 = 0;
}
for (var_s5 = spE0; (spD8+var_s5)->unk0 != 0 && !(spD8+var_s5)->unk8_14; var_s5 = (spD8+var_s5)->unk0) {
(spD8+var_s5)->unk8_14 = 1;
tmp = (spD8+var_s5)->unk8_13 == 0 ? 1 : 0;
var_s7 += tmp;
var_s0 += ((spD8+var_s5)->unk8_13 == 0) ? 0 : 1;
}
tmp = var_s7 * 3;
spB4 = (struct56s *) malloc(8 + tmp * sizeof(f32));
spA4 = (f32 *) spB4;
spB4->unk0 = var_s7;
spB4->unk4 = 0;
spA8 = (SplineList *) malloc(4 + var_s0 * sizeof(Union_glspline));
spA8->unk0 = var_s0;
var_s1_2 = &spA8->spline[0];
a0 = spE0;
var_s2 = spA8->unk0;
for (; var_s2 != 0; a0 = (spD8+a0)->unk0) {
if (0);
if ((spD8+a0)->unk8_13 == 1) {
temp_v0_16 = func_803080C8(a0);
temp_v0_16->t1.unk8.pad_bit7 = D_80371E78;
memcpy(var_s1_2, temp_v0_16, sizeof(Union_glspline));
var_s1_2++;
var_s2--;
}
}
temp_s3_2 = spA8->unk0 - 1 + spA8->spline;
do {
var_s2 = 0;
for (var_s0_2 = &spA8->spline[0]; var_s0_2 < temp_s3_2; var_s0_2++) {
var_s1_2 = var_s0_2 + 1;
if (var_s0_2->common.unk0 > var_s1_2->common.unk0) {
memcpy(&sp80, var_s0_2, sizeof(Union_glspline));
memcpy(var_s0_2, var_s1_2, sizeof(Union_glspline));
memcpy(var_s1_2, &sp80, sizeof(Union_glspline));
var_s2++;
}
}
} while (var_s2);
a0 = var_s7;
for (var_s5 = spE0; a0 != 0; var_s5 = (spD8 + var_s5)->unk0) {
if (!(spD8 + var_s5)->unk8_13) {
spA4[var_fp + 2] = (spD8 + var_s5)->unk2[0];
spA4[var_fp + 3] = (spD8 + var_s5)->unk2[1];
spA4[var_fp + 4] = (spD8 + var_s5)->unk2[2];
var_fp += 3;
a0--;
}
}
func_80341BC8(spB4, spA8);
}
}
free(spD8);
}
//glspline_free
void func_80341A54(void) {
s32 var_s0;
for(var_s0 = 0; var_s0 < 0x40; var_s0++){
if (D_803858A0[var_s0] != 0) {
func_8030DA44(D_803858A0[var_s0]);
}
}
for(var_s0 = 0x40; var_s0 < 0x80; var_s0++){
if (D_803858A0[var_s0] != 0) {
func_8025A7DC(lookup_getCoMusicId(D_803858A0[var_s0]));
}
}
for(var_s0 = 0; var_s0 < D_80371E78; var_s0++){
free(D_80371E70[var_s0]);
free(D_80371E74[var_s0]);
}
free(D_80371E70);
free(D_80371E74);
free(D_803858A0);
D_80371E70 = NULL;
D_80371E74 = NULL;
D_80371E78 = 0;
D_803858A0 = NULL;
}
void func_80341BA0(void){
func_80341A54();
func_803411B0();
}
s32 func_80341BC8(struct56s *arg0, SplineList * arg1) {
s32 temp_t7;
void *temp_v0;
void *temp_v0_2;
D_80371E78++;
D_80371E70 = (struct56s **)realloc(D_80371E70, D_80371E78 * sizeof(struct56s *));
D_80371E70[D_80371E78 - 1] = arg0;
D_80371E74 = (SplineList **)realloc(D_80371E74, D_80371E78 * sizeof(SplineList *));
D_80371E74[D_80371E78 - 1] = arg1;
return D_80371E78 - 1;
}
s32 func_80341C78(s32 arg0[3]) {
struct56s *temp_a0;
s32 var_a3;
s32 var_v1;
f32 *var_a2;
f32 sp4[3];
sp4[0] = (f32) arg0[0];
sp4[1] = (f32) arg0[1];
sp4[2] = (f32) arg0[2];
for(var_v1 = 0; var_v1 < D_80371E78; var_v1++){
temp_a0 = D_80371E70[var_v1];
var_a2 = temp_a0->unk8;
for (var_a3 = 0; var_a3 < temp_a0->unk0; var_a3++) {
if( (sp4[0] == var_a2[0]) && (sp4[1] == var_a2[1]) && (sp4[2] == var_a2[2])) {
return var_v1;
}
var_a2+=3;
}
}
return -1;
}
s32 func_80341D5C(s32 arg0[3], s32 arg1[3])
{
int i;
int j;
struct56s *a0;
f32 (*a2)[3];
f32 spC[3];
f32 sp0[3];
struct56s *new_var;
spC[0] = (f32) arg0[0];
spC[1] = (f32) arg0[1];
spC[2] = (f32) arg0[2];
sp0[0] = (f32) arg1[0];
sp0[1] = (f32) arg1[1];
sp0[2] = (f32) arg1[2];
for (i = 0; i < D_80371E78; i++)
{
a2 = a0 + 1;
new_var = D_80371E70[i];
a0 = new_var;
a2 = a0 + 1;
for (j = 0; j < new_var->unk0; j++)
{
if (((spC[0] == a2[j][0]) && (spC[1] == a2[j][1]) && (spC[2] == a2[j][2])) \
&& ((sp0[0] == a2[j + 1][0]) && (sp0[1] == a2[j + 1][1]) && (sp0[2] == a2[j + 1][2])))
{
return i;
}
}
}
return -1;
}
s32 func_80341EC4(f32 arg0[3]){
s32 sp1C[3];
ml_vec3f_to_vec3w(sp1C, arg0);
return func_80341C78(sp1C);
}
struct56s *func_80341EF0(f32 arg0[3]){
s32 sp1C[3];
ml_vec3f_to_vec3w(sp1C, arg0);
return D_80371E70[func_80341C78(sp1C)];
}
s32 func_80341F2C(s32 arg0){
s32 sp1C[3];
if(!_nodeProp_findPositionFromActorId(arg0, sp1C)){
return -1;
}
return func_80341C78(sp1C);
}
struct56s *func_80341F64(s32 arg0){
s32 sp1C[3];
if(!_nodeProp_findPositionFromActorId(arg0, sp1C)){
return NULL;
}
return D_80371E70[func_80341C78(sp1C)];
}
f32 func_80341FB0(s32 arg0, f32 arg1, s32 arg2, f32 arg3) {
f32 sp1C;
if (arg0 == -1) {
return arg1;
}
sp1C = func_80323F74(D_80371E70[arg0], arg1, arg3);
func_80323240(D_80371E70[arg0], sp1C, arg2);
return sp1C;
}
struct56s *func_80342038(s32 indx){
if(indx == -1)
return NULL;
return D_80371E70[indx];
}
int func_80342064(s32 arg0){
return 0;
}
s32 func_80342070(s32 arg0){
SplineList **temp_v0 = D_80371E74;
SplineList *temp_v1;
if(arg0 == -1)
return 0;
temp_v1 = temp_v0[arg0];
if (temp_v1->unk0 == 0) {
return 0;
}
return temp_v1->spline[0].t1.unk10.bit31;
}
s32 func_803420BC(s32 arg0, s32 arg1, s32 arg2){
SplineList **temp_v0 = D_80371E74;
SplineList *temp_v1;
s32 temp_a1;
s32 var_a0;
if(arg0 == -1)
return 0;
temp_v1 = temp_v0[arg0];
temp_a1 = temp_v1->unk0;
if (temp_a1 == 0) {
return 0;
}
for(var_a0 = 0; (var_a0 < temp_a1) && arg1 != temp_v1->spline[var_a0].t1.unk10.bit31; var_a0++);
if (var_a0 == temp_a1) {
return 0;
}
var_a0 += arg2;
while(var_a0 < 0){ var_a0 += temp_a1; }
while(var_a0 >= temp_a1){ var_a0 -= temp_a1; }
return temp_v1->spline[var_a0].t1.unk10.bit31;
}
s32 func_803421A4(s32 arg0, f32 arg1) {
SplineList **temp_v0 = D_80371E74;
SplineList *temp_v1;
s32 var_a3;
Union_glspline *i_ptr;
Union_glspline *temp_a2;
if (arg0 == -1) {
return 0;
}
temp_v1 = temp_v0[arg0];
if (temp_v1->unk0 == 0) {
return 0;
}
i_ptr = temp_v1->spline;
temp_a2 = i_ptr + temp_v1->unk0;
for(var_a3 = 0; (i_ptr < temp_a2) && (i_ptr->common.unk0 < arg1); i_ptr++){
var_a3++;
}
return var_a3;
}
s32 func_80342244(s32 arg0){
return (arg0 < 180) ? (arg0 + 180) : (arg0 - 180);
}
f32 func_80342260(s32 arg0, f32 arg1, s32 arg2){
return func_803234FC(func_80342038(arg0), arg1, 1.0f) / ((f32)arg2 /4);
}
void func_803422AC(ActorMarker *caller, enum asset_e text_id, s32 arg2){
jiggyscore_total();
}
s32 func_803422D4(Actor *arg0, Union_glspline *arg1, SplineList *arg2){
s32 sp84;
u8 sfxsourceIdx;
f32 sp7C;
s32 sp78;
s32 sp74;
u32 sp70;
s32 sp6C;
s32 sp68;
s32 sp64 = 1;
f32 sp60;
f32 sp54[3];
f32 sp50;
f32 sp4C;
Struct_glspline_t1 *sp48;
Struct_glspline_t1 *temp_v0_6;
sp84 = 0;
if (arg1->common.unkC.bit0){
if(sp64 == D_80371E80) sp84 = sp64;
switch (arg1->t0.unk4.common.bit3) {
case 1:
func_80250530(0, arg1->t0.unk4.common.bit31, 3.0f);
break;
case 2:
D_803858A0[arg1->t0.unk10.common.bit7] = arg1->t0.unk4.common.bit31;
func_8025A6EC(lookup_getCoMusicId(arg1->t0.unk4.common.bit31), arg1->t0.unk4.common.bit15 * 8);
break;
case 3:
sp78 = arg1->t0.unk4.common.bit31;
sp7C = arg1->t0.unk8.bit31 / 8.388608e6f;
sp74 = arg1->t0.unk4.common.bit15 * 8;
sp70 = arg1->t0.unk10.common.bit7;
if (func_8030ED70(lookup_getSfxId(sp78))){
sfxsourceIdx = sfxsource_createSfxsourceAndReturnIndex();
if (sfxsourceIdx == 0){
return sp84;
}
sfxsource_setSfxId(sfxsourceIdx, lookup_getSfxId(sp78));
sfxsource_playSfxAtVolume(sfxsourceIdx, sp7C);
sfxsource_setSampleRate(sfxsourceIdx, sp74);
func_8030E2C4(sfxsourceIdx);
if (D_803858A0[sp70] != 0){
func_8030DA44(D_803858A0[sp70]);
}
D_803858A0[sp70] = sfxsourceIdx;
} else {
func_8030E6A4(lookup_getSfxId(sp78), sp7C, sp74);
}
break;
case 7:
if (((arg1->t0.unk10.word << 0x1e) >> 0x1f) == 1){
arg0->unk124_11 = (arg1->t0.unk4.t7.bit31 == 0) ? (2) : ((sp64 == arg1->t0.unk4.t7.bit31) ? (1) : (0));
arg0->unk1C[0] = (arg1->t0.unk4.t7.bit30 != 0) ? ((f32) (((f32) arg1->t0.unk4.t7.bit30) * 0.25)) : (0.0f);
arg0->unk1C[1] = 0.0f;
}
if (arg1->t0.unkC.bit21 == 1){
func_803110F8(arg1->t0.unkC.bit20, arg1->t0.unk10.t7.bit7, arg1->t0.unk4.t7.bit15, arg1->t0.unk4.t7.bit7, func_803422AC);
}
if ((arg1->t0.unkC.bit23 == 1) || (arg1->t0.unkC.bit23 == 2)){
D_80371E80 = arg1->t0.unkC.bit23 - 1;
}
break;
case 6:
mapSpecificFlags_set(arg1->t0.unk4.common.bit31, 1);
break;
case 4:
if (D_803858A0[arg1->t0.unk10.common.bit7] != 0){
func_8025A7DC(lookup_getCoMusicId(D_803858A0[arg1->t0.unk10.common.bit7]));
D_803858A0[arg1->t0.unk10.common.bit7] = 0;
}
break;
case 5:
if (D_803858A0[arg1->t0.unk10.common.bit7] != 0){
func_8030DA44(D_803858A0[arg1->t0.unk10.common.bit7]);
D_803858A0[arg1->t0.unk10.common.bit7] = 0;
}
break;
}
if (arg1->t0.unkC.bit12 != 0) {
sp64 = 0;
sp68 = func_80340760(arg1->t0.unkC.bit12, &sp6C, &sp60, sp54, arg0->unk44_14, &sp50, &sp4C);
if (sp68 == 0) {
sp64 = func_80340748(arg1->t0.unkC.bit12, &sp6C, &sp60, sp54, arg0->unk44_14, &sp50, &sp4C);
}
sp84 = 2;
arg0->unk44_14 = sp6C;
arg0->unk48 = sp60;
arg0->position[0] = sp54[0];
arg0->position[1] = sp54[1];
arg0->position[2] = sp54[2];
if (sp68 != 0) {
if (sp50 != (-9999.0f)) {
arg0->yaw = sp50;
if (arg0->marker->id == 0x12) {
arg0->yaw = (f32) func_80342244((s32) arg0->yaw);
}
arg0->yaw_ideal = arg0->yaw;
}
if (sp4C != (-9999.0f)){
arg0->pitch = sp4C;
arg0->unk6C = arg0->pitch;
}
if ((sp50 != (-9999.0f)) || (sp4C != (-9999.0f))){
func_802CDAC4(arg0);
}
}
if (sp64);
}
} else {
if (sp64 == D_80371E80){
sp84 = sp64;
}
if (arg1->t1.unk8.bit8){
if (arg1->t1.unk10.bit5){
arg0->unk138_3 = arg1->t1.unkC.bit11 - 0x69;
} else {
arg0->unk54 = ((f32) arg1->t1.unkC.bit11) / 4;
}
}
if (arg1->t1.unk8.bit10){
arg0->unk5C = ((f32) arg1->t1.unk8.bit21) / 4;
arg0->animctrl_asset_id = lookup_getAnimAssetId(arg1->t1.unk8.bit31);
if (arg0->animctrl == 0) {
arg0->animctrl = animctrl_new(0);
animctrl_reset(arg0->animctrl);
}
animctrl_setIndex(arg0->animctrl, arg0->animctrl_asset_id);
animctrl_setDuration(arg0->animctrl, arg0->unk5C);
animctrl_setDirection(arg0->animctrl, 1);
if (arg0->animctrl != 0) {
animctrl_start(arg0->animctrl, "glspline.c", 0x545);
}
}
switch (arg1->t1.unk4.bit6){
case 1:
arg0->unk10_4 = 1;
arg0->unk10_8 = 1;
arg0->unk10_7 = 1;
break;
case 2:
arg0->unk10_4 = 0;
arg0->unk10_8 = 0;
break;
case 3:
arg0->unk10_4 = 0;
arg0->unk10_8 = 1;
break;
case 4:
arg0->unk10_4 = 0;
arg0->unk10_7 = 0;
break;
case 5:
arg0->unk10_4 = 0;
arg0->unk10_7 = 1;
break;
case 6:
arg0->unk10_4 = 0;
arg0->unk10_8 = 0;
arg0->unk10_7 = 0;
break;
case 7:
arg0->unk10_4 = 0;
arg0->unk10_8 = 1;
arg0->unk10_7 = 1;
break;
}
switch (arg1->t1.unk4.bit3){
case 2:
arg0->unk10_6 = 0;
if (arg0->unk4C < 0.0f){
arg0->unk4C = -arg0->unk4C;
} else {
arg0->unk4C = arg0->unk4C;
}
actor_playAnimationOnce(arg0);
break;
case 3:
arg0->unk10_6 = 0;
if (arg0->unk4C < 0.0f) {
arg0->unk4C = arg0->unk4C;
} else {
arg0->unk4C = -arg0->unk4C;
}
actor_playAnimationOnce(arg0);
break;
case 4:
arg0->unk10_6 = 1;
if (arg0->unk4C < 0.0f) {
arg0->unk4C = -arg0->unk4C;
} else {
arg0->unk4C = arg0->unk4C;
}
actor_loopAnimation(arg0);
break;
case 5:
arg0->unk10_6 = 1;
if (arg0->unk4C < 0.0f){
arg0->unk4C = arg0->unk4C;
} else {
arg0->unk4C = -arg0->unk4C;
}
actor_loopAnimation(arg0);
break;
case 6:
arg0->unk10_6 = 2;
break;
}
if (arg1->t1.unk4.bit19 != 0){
arg0->unk10_3 = arg1->t1.unk4.bit19;
}
if (arg0->unk138_30 != arg1->t1.unk10.bit6){
arg0->unk138_30 = arg1->t1.unk10.bit6;
}
if (arg1->t1.unk8.bit9){
if ((!func_802E4A08()) && (arg0->modelCacheIndex == ACTOR_123_MAGIC_CARPET_2)){
arg0->unk4C = func_80323540(func_80342038(arg0->unk44_14), arg0->unk48, 1.0f, 500.0f) / (((f32) arg1->t1.unkC.bit22) / 4);
} else {
arg0->unk4C = func_80342260(arg0->unk44_14, arg0->unk48, arg1->t1.unkC.bit22);
}
}
if (arg1->t1.unk4.bit31 != 0){
temp_v0_6 = arg2->unk0 + &arg2->spline[0];
for (sp48 = &arg2->spline[0]; (sp48->unk10.bit31 != (arg1->t1.unk4.bit31 ^ 0)) && (sp48 < temp_v0_6); sp48++){
;
}
arg0->unkF4_24 = arg1->t0.unk4.common.bit31;
arg0->unk70 = arg1->t1.unk0;
arg0->unk74 = sp48->unk0;
if (arg1->t1.unk4.bit17 & 1)
{
arg0->unk38_13 = arg1->t1.unkC.bit31;
arg0->unk38_21 = sp48->unkC.bit31;
arg0->unk78_31 = arg1->t1.unk4.bit15;
arg0->unk78_22 = sp48->unk4.bit15;
if (arg0->marker->id == 0x12)
{
arg0->unk38_13 = func_80342244(arg0->unk38_13);
arg0->unk38_21 = func_80342244(arg0->unk38_21);
}
arg0->yaw = (arg0->yaw_ideal = (f32) arg0->unk38_13);
arg0->pitch = (arg0->unk6C = (f32) arg0->unk78_31);
}
if (arg1->t1.unk4.bit17 & 2){
arg0->unk140 = arg0->unk4C;
arg0->unk144 = func_803234FC(func_80342038(arg0->unk44_14), sp48->unk0, 1.0f) / (((f32) sp48->unkC.bit22) / 4);
}
}
if (!arg0->unk10_4){
if (arg0->unk10_8){
arg0->yaw_ideal = (f32) arg1->t1.unkC.bit31;
if (arg0->marker->id == 0x12){
arg0->yaw_ideal = (f32) func_80342244((s32) arg0->yaw_ideal);
}
}
if (arg0->unk10_7){
arg0->unk6C = (f32) arg1->t1.unk4.bit15;
}
}
}
return sp84;
}
void func_803430B4(Actor *arg0) {
f32 d_yaw;
if ((arg0->marker->id == 0x12) && (arg0->unk138_30 == 1)) {
d_yaw = arg0->yaw_ideal - arg0->yaw;
while (d_yaw >= 360.0f) { d_yaw -= 360.0f; }
while (d_yaw < 0.0f) { d_yaw += 360.0f; }
if (d_yaw >= 180.0f) {
d_yaw = 360.0f - d_yaw;
}
arg0->unk10_3 = (d_yaw < 1.8) ? 3
: (d_yaw < 3.2) ? 2
: 1;
}
}
void func_803431D0(Actor *arg0, s32 arg1, s32 arg2, s32 arg3) {
f32 sp64[3];
f32 sp58[3];
f32 sp4C[3];
f32 sp40[3];
f32 sp3C;
s32 pad38;
f32 var_f2;
if (arg1 & 0x300) {
func_8032417C(func_80342038(arg0->unk44_14), arg0->unk48, sp64, sp58);
if (arg1 & 1) {
if (arg1 & 0x100) {
arg0->yaw_ideal = sp58[1];
}
if (arg1 & 0x200) {
arg0->unk6C = sp58[0];
}
} else {
if (arg1 & 0x100) {
if (sp58[1] >= 180.0f) {
arg0->yaw_ideal = sp58[1] - 180.0f;
} else {
arg0->yaw_ideal = sp58[1] + 180.0f;
}
}
if (arg1 & 0x200) {
arg0->unk6C = (360.0 - sp58[0]);
}
}
}
if (arg0->unk10_4) {
player_getPosition(sp4C);
func_802BC434(sp40, sp4C, arg0->position);
if (arg0->marker->id != 0x12) {
sp40[1] += 180.0f;
while (sp40[1] >= 360.0f) {
sp40[1] -= 360.0f;
}
sp40[0] = 360.0f - sp40[0];
}
arg0->yaw_ideal = sp40[1];
arg0->unk6C = sp40[0];
} else {
if ((arg0->unk70 <= arg0->unk48) && (arg0->unk48 < arg0->unk74)) {
sp3C = (arg0->unk48 - arg0->unk70) / (arg0->unk74 - arg0->unk70);
if (arg0->unkF4_24 & 1) {
if (arg1 & 0x800) {
var_f2 = mlDiffDegF(arg0->unk38_21, arg0->unk38_13) * sp3C + arg0->unk38_13;
while (var_f2 >= 360.0f) { var_f2 -= 360.0f; }
while (var_f2 < 0.0f) { var_f2 += 360.0f; }
arg0->yaw_ideal = var_f2;
}
if (arg1 & 0x1000) {
var_f2 = mlDiffDegF(arg0->unk78_22, arg0->unk78_31) * sp3C + arg0->unk78_31;
while (var_f2 >= 360.0f) { var_f2 -= 360.0f; }
while (var_f2 < 0.0f) { var_f2 += 360.0f; }
arg0->unk6C = var_f2;
}
}
if (arg0->unkF4_24 & 2) {
arg0->unk4C = ((arg0->unk144 - arg0->unk140) * sp3C) + arg0->unk140;
}
} else if ((arg0->unk70 != 0.0f) || (arg0->unk74 != 0.0f)) {
arg0->unk70 = 0.0f;
arg0->unk74 = 0.0f;
}
}
func_803430B4(arg0);
func_80328FB0(arg0, (f32) arg2);
func_80328FF0(arg0, (f32) arg3);
}
s32 func_80343654(Actor *this){
s32 tmp_v1;
s32 tmp_a0;
tmp_v1 = this->unk10_8 ? 0x800 : 0x100;
tmp_a0 = this->unk10_7 ? 0x1000 : 0x200;
return tmp_a0 + tmp_v1;
}
s32 func_80343694(Actor *actor, s32 indx, s32 begin, s32 end, s32 count, s32 stride) {
SplineList **temp_v0;
SplineList *temp_s5;
Union_glspline *start_ptr;
Union_glspline *end_ptr;
bool var_v1;
f32 sp48;
Union_glspline *i_ptr;
temp_v0 = D_80371E74;
temp_s5 = temp_v0[indx];
var_v1 = 0;
if (temp_s5->unk0 == 0) {
return 0;
}
start_ptr = temp_s5->spline;
i_ptr = temp_s5->spline + ((begin == -9999) ? temp_s5->unk0 - 1 : begin);
end_ptr = temp_s5->spline + ((end == -9999) ? temp_s5->unk0 + count : end);
for(; (i_ptr != end_ptr) && (var_v1 != 1); i_ptr += stride){
if( ((stride >= 1) && (i_ptr->common.unk10.bit0 == 0))
|| ((stride < 0) && (i_ptr->common.unk10.bit0 == 1))
) {
if (D_80371E80 == 1) {
sp48 = i_ptr->common.unk0;
}
var_v1 = func_803422D4(actor, i_ptr, temp_s5);
if( ((actor->unk70 != 0.0f) || (actor->unk74 != 0.0f))
&& (!(actor->unk70 <= actor->unk48) || !(actor->unk48 < actor->unk74))
) {
actor->unk70 = 0.0f;
actor->unk74 = 0.0f;
}
if ((var_v1 == 1) && (actor->modelCacheIndex == 0x66) && (actor->unk70 != actor->unk74)) {
var_v1 = 0;
}
}
}
if ((D_80371E80 == 1) && (var_v1 == 1)) {
actor->unk48 = sp48 + 0.00001;
}
return var_v1;
}
s32 func_803438E0(Actor *actor, s32 arg1, s32 arg2, s32 arg3) {
s32 sp4C;
s32 sp48;
s32 sp44;
f32 sp40;
s32 sp3C;
s32 temp_v0;
sp44 = 0;
sp40 = actor->unk4C;
if (actor->unk44_14 == -1) {
return 0;
}
sp4C = func_803421A4(actor->unk44_14, actor->unk48);
if ((actor->unk54 == 0.0f) && (actor->unk138_3 == 0)) {
if (actor->marker->id == 0xB0) {
actor->unk48 = func_803240E0(D_80371E70[actor->unk44_14], actor->unk48, sp40, &sp44);
} else {
actor->unk48 = func_80323FDC(D_80371E70[actor->unk44_14], actor->unk48, sp40, &sp44);
}
func_80323240(D_80371E70[actor->unk44_14], actor->unk48, actor->position);
} else {
if (actor->unk54 != 0.0f) {
actor->unk54 -= time_getDelta();
if (actor->unk54 < 0.0f) {
actor->unk54 = 0.0f;
}
}
if ((actor->unk138_3 != 0) && func_802501A0(0, actor->unk138_3 + 0x69, 0)) {
func_80250170(0, actor->unk138_3 + 0x69, 0);
actor->unk138_3 = 0;
}
}
sp48 = func_803421A4(actor->unk44_14, actor->unk48);
if (!actor->unk44_1) {
actor->unk44_1 = TRUE;
actor->yaw = actor->yaw_ideal;
actor->pitch = actor->unk6C;
}
if ((sp4C == sp48) && (sp44 == 0)) {
func_803431D0(actor, arg1, arg2, arg3);
} else {
sp3C = func_80343654(actor);
if (sp40 > 0.0f) {
if (sp44 == 0) {
if (func_80343694(actor, actor->unk44_14, sp4C, sp48, 0, 1) == 2) {
D_80371E7C = 1;
}
} else if (func_80343694(actor, actor->unk44_14, sp4C, -0x270F, 0, 1) == 0) {
func_80343694(actor, actor->unk44_14, func_803421A4(actor->unk44_14, func_803237E8(func_80342038(actor->unk44_14))), sp48, 0, 1);
}
} else if (sp40 < 0.0f) {
if (sp44 == 0) {
func_80343694(actor, actor->unk44_14, sp4C - 1, sp48 - 1, 0, -1);
} else if (!func_80343694(actor, actor->unk44_14, sp4C - 1, -1, 0, -1)) {
func_80343694(actor, actor->unk44_14, -0x270F, sp48 - 1, 0, -1);
}
}
temp_v0 = func_80343654(actor);
if ((sp3C == temp_v0) || ((sp3C != temp_v0) && (sp3C != (arg1 & 0x3F00)))) {
func_803431D0(actor, arg1, arg2, arg3);
} else {
func_803431D0(actor, (arg1 & ~0x3F00) | (temp_v0 & 0x3F00), arg2, arg3);
}
}
return 1;
}
int func_80343D50(Actor *this, s32 arg1, s32 arg2, s32 arg3){
s32 s0;
s0 = 0;
do{
D_80371E7C = 0;
s0 += func_803438E0(this, arg1, arg2, arg3);
}while(D_80371E7C);
return s0 ? 1 : 0;
}
void func_80343DEC(Actor *this){
func_80343D50(this, func_80343654(this), 0x19, 0x19);
}
void func_80343E20(s32 arg0, s32 arg1, f32 arg2, s32 arg3) {
u8 temp_v0;
if (func_8030ED70(lookup_getSfxId(arg0)) != 0) {
temp_v0 = sfxsource_createSfxsourceAndReturnIndex();
if (temp_v0 != 0) {
sfxsource_setSfxId(temp_v0, lookup_getSfxId(arg0));
sfxsource_playSfxAtVolume(temp_v0, arg2);
sfxsource_setSampleRate(temp_v0, arg1);
func_8030E2C4(temp_v0);
if (D_803858A0[arg3] != 0) {
func_8030DA44(D_803858A0[arg3]);
}
D_803858A0[arg3] = temp_v0;
}
} else {
func_8030E6A4(lookup_getSfxId(arg0), arg2, arg1);
}
}
struct56s *func_80343F00(s32 indx, f32 arg1[3]){
struct56s *out;
out = func_80342038(indx);
out->unk8[0][0] = arg1[0];
out->unk8[0][1] = arg1[1];
out->unk8[0][2] = arg1[2];
return out;
}
void glspline_defrag(void) {
s32 i;
if (D_80371E70 != 0) {
D_80371E70 = (struct56s **) defrag(D_80371E70);
}
if (D_80371E74 != 0) {
D_80371E74 = (SplineList **) defrag(D_80371E74);
}
if (D_803858A0 != 0) {
D_803858A0 = (s16 *) defrag(D_803858A0);
}
for(i = 0; i < D_80371E78; i++){
D_80371E74[i] = (SplineList *) defrag(D_80371E74[i]);
D_80371E70[i] = (struct56s *) defrag(D_80371E70[i]);
}
}
bool func_80344040(Actor *this){
return func_80323240(D_80371E70[this->unk44_14], this->unk48, this->position);
}