etc/sm64coopdx
0
0
Fork 0
mirror of https://github.com/coop-deluxe/sm64coopdx.git synced 2024-10-18 11:18:30 +00:00
Code Issues Projects Releases Packages Wiki Activity

Fix the branch bloat found in the atan2s and atan2_lookup functions (#375)

Browse source 
* Fix branch bloat in atan2s and atan2_lookup

* Fix unecessary inline on lookup
This commit is contained in:
John S 2024-10-18 02:47:00 -04:00 committed by GitHub
parent 479cc9d79d
commit b712d4b6a3
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 23 additions and 45 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

68
src/engine/math_util.c
View file

@ -872,58 +872,36 @@ f32 approach_f32(f32 current, f32 target, f32 inc, f32 dec) {
* the resulting angle is in range [0, 0x2000] (1/8 of a circle). * the resulting angle is in range [0, 0x2000] (1/8 of a circle).
*/ */
static u16 atan2_lookup(f32 y, f32 x) { static u16 atan2_lookup(f32 y, f32 x) {
u16 ret; s16 idx = (s16)(y / x * 1024.0f + 0.5f);
idx = (idx >= 0 && idx < 0x401) ? idx : 0;
if (x == 0) { return gArctanTable[idx];
ret = gArctanTable[0];
} else {
s32 index = (s32)(y / x * 1024 + 0.5f);
if (index >= 0x401 || index < 0) { index = 0; }
ret = gArctanTable[index];
}
return ret;
} }
/** /**
* Compute the angle from (0, 0) to (x, y) as a s16. Given that terrain is in * Compute the angle from (0, 0) to (x, y) as a s16. Given that terrain is in
* the xz-plane, this is commonly called with (z, x) to get a yaw angle. * the xz-plane, this is commonly called with (z, x) to get a yaw angle.
*/ */
s16 atan2s(f32 y, f32 x) { inline s16 atan2s(f32 y, f32 x) {
u16 ret; // Extract sign bits: 1 if negative, 0 otherwise
u8 signx = (x < 0.0f);
u8 signy = (y < 0.0f);
if (x >= 0) { // Take absolute values
if (y >= 0) { f32 absx = absx(x);
if (y >= x) { f32 absy = absx(y);
ret = atan2_lookup(x, y);
} else { // Compute the angle in the first octant
ret = 0x4000 - atan2_lookup(y, x); u16 angle = atan2_lookup(min(absx, absy), max(absy, absx));
}
} else { // Create an index based on the signs and swap status
y = -y; u8 idx = ((absy > absx) << 2) | (signx << 1) | signy;
if (y < x) {
ret = 0x4000 + atan2_lookup(y, x); // Combined lookup tables for offsets and sign multipliers
} else { static const s16 offsets[] = {0x4000, 0x4000, 0xC000, 0xC000, 0x0000, 0x8000, 0x0000, 0x8000};
ret = 0x8000 - atan2_lookup(x, y); static const s8 signs[] = {-1, 1, 1, -1, 1, -1, -1, 1};
}
} // Ensure the result fits into 16 bits via an explicit cast on angle
} else { return (offsets[idx] + (signs[idx] * (s16)angle)) & 0xFFFF;
x = -x;
if (y < 0) {
y = -y;
if (y >= x) {
ret = 0x8000 + atan2_lookup(x, y);
} else {
ret = 0xC000 - atan2_lookup(y, x);
}
} else {
if (y < x) {
ret = 0xC000 + atan2_lookup(y, x);
} else {
ret = -atan2_lookup(x, y);
}
}
}
return ret;
} }
/** /**