Decompile mt_trig.cpp

config/RSBE01_02/splits.txt

@@ -147,6 +147,10 @@ sora/mt/mt_prng.cpp:
 	.sbss       start:0x805A00B8 end:0x805A00C0
 	.sdata2     start:0x805A1748 end:0x805A176C
 
+sora/mt/mt_trig.cpp:
+	.text       start:0x8003FD5C end:0x8003FFC4
+	.sdata2     start:0x805A1770 end:0x805A179C
+
 sora/mt/mt_prng_log.cpp:
 	.text       start:0x8004392C end:0x80043B18
 	.ctors      start:0x80406514 end:0x80406518

config/RSBE01_02/symbols.txt

@@ -31850,7 +31850,7 @@ lbl_805A1780 = .sdata2:0x805A1780; // type:object size:0x4 align:4 data:float
 lbl_805A1784 = .sdata2:0x805A1784; // type:object size:0x4 align:4 data:float
 lbl_805A1788 = .sdata2:0x805A1788; // type:object size:0x4 align:4 data:float
 lbl_805A1790 = .sdata2:0x805A1790; // type:object size:0x8 align:8 data:double
-lbl_805A1798 = .sdata2:0x805A1798; // type:object size:0x8 align:4 data:float
+lbl_805A1798 = .sdata2:0x805A1798; // type:object size:0x4 align:4 data:float
 lbl_805A17A0 = .sdata2:0x805A17A0; // type:object size:0x8 align:4 data:float
 lbl_805A17A8 = .sdata2:0x805A17A8; // type:object size:0x4 align:4 data:float
 lbl_805A17AC = .sdata2:0x805A17AC; // type:object size:0x4 align:4 data:float

configure.py

@@ -198,7 +198,7 @@
     "-maxerrors 1",
     "-nosyspath",
     "-RTTI off",
-    "-fp_contract on",
+    "-fp_contract off",
     "-str reuse",
     "-enc SJIS",
     "-i include",
@@ -320,6 +320,7 @@ def MatchingFor(*versions):
             Object(Matching, "sora/gf/gf_resource_loader.cpp"),
             Object(NonMatching, "sora/mt/mt_vector_old.cpp"),
             Object(Matching, "sora/mt/mt_prng.cpp", extra_cflags=["-RTTI off"]),
+            Object(NonMatching, "sora/mt/mt_trig.cpp"),
             Object(Matching, "sora/mt/mt_prng_log.cpp"),
             Object(Matching, "sora/st/module.cpp"),
             Object(Matching, "sora/ut/ut_nw.cpp"),

src/sora/mt/mt_trig.cpp

@@ -0,0 +1,64 @@
+#include <mt/mt_trig.h>
+
+// pi/2
+#define HALF_PI 1.5707964f
+// 2^24 / (2*pi)
+#define RAD_TO_FIXED_PT 2670176.8f
+// (2*pi) / 2^24
+#define FIXED_PT_TO_RAD 3.7450704e-7f
+
+// Quarter-turn and reduce range to [0, 2^24)
+static inline s32 reduceRange(s32 fixedPtAngle) {
+    return (fixedPtAngle + 0x400000) & 0x00FFFFFF;
+}
+
+// Fit to range [-2^22, 2^22), i.e. [-pi/2, pi/2)
+static inline s32 fitToRange(s32 theta) {
+    if (theta > 0x800000)
+        theta = 0x1000000 - theta;
+    theta -= 0x400000;
+
+    return theta;
+}
+
+// 9th order Taylor polynomial approximation of sin(x)
+static inline float taylor(float x) {
+    const float C1 = 1.0f;            // 1
+    const float C2 = -0.16666657f;    // -1/3!
+    const float C3 = 0.0083330255f;   // 1/5!
+    const float C4 = -0.00019807414f; // -1/7!
+    const float C5 = 0.000002601887f; // 1/9!
+    return x * (C1 + ((x * x) * (C2 + ((x * x) * (C3 + ((x * x) * (C4 + (C5 * (x * x)))))))));
+}
+
+float mtSinf(float angle) {
+    return taylor(FIXED_PT_TO_RAD * fitToRange(reduceRange((RAD_TO_FIXED_PT * angle))));
+}
+
+float mtCosf(float angle) {
+    return mtSinf(angle + HALF_PI);
+}
+
+// NONMATCHING regswaps - should simply interleave calls to mtCosf and mtSinf
+void mtSinCosf(float rad, float* sinOut, float* cosOut) {
+    s32 s0 = RAD_TO_FIXED_PT * rad;
+    s32 c0 = RAD_TO_FIXED_PT * (rad + HALF_PI);
+
+    s32 s1 = reduceRange(s0);
+    s32 c1 = reduceRange(c0);
+
+    s32 c2 = fitToRange(s1);
+    s32 s2 = fitToRange(c1);
+
+    float s3 = s2;
+    float c3 = c2;
+
+    float c4 = FIXED_PT_TO_RAD * c3;
+    float s4 = FIXED_PT_TO_RAD * s3;
+
+    float s5 = taylor(s4);
+    float c5 = taylor(c4);
+
+    *sinOut = s5;
+    *cosOut = c5;
+}