From 40820d5a566d90b05cf27951b9402e0dcf62b125 Mon Sep 17 00:00:00 2001
From: Phillip Stephens <antidote.crk@gmail.com>
Date: Fri, 24 Feb 2023 21:05:35 -0800
Subject: [PATCH] Finally match salInvertMatrix

Former-commit-id: ec1893fd201b6fd2053aa8edf8ccb43c0a5262ab
---
 libc/math.h                  |   4 +-
 src/musyx/runtime/snd_math.c | 198 +++++------------------------------
 2 files changed, 31 insertions(+), 171 deletions(-)

diff --git a/libc/math.h b/libc/math.h
index 0ad14fe5..1bc6d7f6 100644
--- a/libc/math.h
+++ b/libc/math.h
@@ -160,7 +160,7 @@ static inline int __fpclassifyd(double x) {
 #define isinf(x) (fpclassify(x) == FP_INFINITE)
 #define isfinite(x) ((fpclassify(x) > FP_INFINITE))
 
-inline float sqrtf(float x) {
+static float sqrtf(float x) {
   const double _half = .5;
   const double _three = 3.0;
   volatile float y;
@@ -179,7 +179,7 @@ inline float sqrtf(float x) {
   return x;
 }
 
-static inline double sqrt(double x) {
+static double sqrt(double x) {
   if (x > 0.0) {
     double guess = __frsqrte(x);                    /* returns an approximation to  */
     guess = .5 * guess * (3.0 - guess * guess * x); /* now have 8 sig bits          */
diff --git a/src/musyx/runtime/snd_math.c b/src/musyx/runtime/snd_math.c
index 7ac63567..4d96b9a8 100644
--- a/src/musyx/runtime/snd_math.c
+++ b/src/musyx/runtime/snd_math.c
@@ -1,18 +1,18 @@
 #include "math.h"
 #include "musyx/musyx_priv.h"
 
-void salApplyMatrix(const SND_FMATRIX* a, const SND_FVECTOR* b, SND_FVECTOR* out) {
-  out->x = a->t[0] + ((a->m[0][2] * b->z) + ((a->m[0][0] * b->x) + (a->m[0][1] * b->y)));
-  out->y = a->t[1] + ((a->m[1][2] * b->z) + ((a->m[1][0] * b->x) + (a->m[1][1] * b->y)));
-  out->z = a->t[2] + ((a->m[2][2] * b->z) + ((a->m[2][0] * b->x) + (a->m[2][1] * b->y)));
+void salApplyMatrix(const SND_FMATRIX* mat, const SND_FVECTOR* in, SND_FVECTOR* out) {
+  out->x = mat->m[0][0] * in->x + mat->m[0][1] * in->y + mat->m[0][2] * in->z + mat->t[0];
+  out->y = mat->m[1][0] * in->x + mat->m[1][1] * in->y + mat->m[1][2] * in->z + mat->t[1];
+  out->z = mat->m[2][0] * in->x + mat->m[2][1] * in->y + mat->m[2][2] * in->z + mat->t[2];
 }
 
 float salNormalizeVector(SND_FVECTOR* vec) {
-  float mag = sqrtf(vec->x * vec->x + vec->y * vec->y + vec->z * vec->z);
-  vec->x /= mag;
-  vec->y /= mag;
-  vec->z /= mag;
-  return mag;
+  float l = sqrtf(vec->x * vec->x + vec->y * vec->y + vec->z * vec->z);
+  vec->x /= l;
+  vec->y /= l;
+  vec->z /= l;
+  return l;
 }
 
 void salCrossProduct(SND_FVECTOR* out, const SND_FVECTOR* a, const SND_FVECTOR* b) {
@@ -21,166 +21,26 @@ void salCrossProduct(SND_FVECTOR* out, const SND_FVECTOR* a, const SND_FVECTOR*
   out->z = (a->x * b->y) - (a->y * b->x);
 }
 
-#if NONMATCHING
 void salInvertMatrix(SND_FMATRIX* out, const SND_FMATRIX* in) {
-    float fVar1;
-    float fVar2;
-    float fVar3;
-    float fVar4;
+  float a; // r62
+  float b; // r61
+  float c; // r60
+  float f; // r63
 
-    fVar1 = in->m[1][1] * in->m[2][2] - in->m[2][1] * in->m[1][2];
-    fVar4 = -(in->m[1][0] * in->m[2][2] - in->m[2][0] * in->m[1][2]);
-    fVar2 = in->m[1][0] * in->m[2][1] - in->m[2][0] * in->m[1][1];
-    fVar3 = 1.f / (in->m[0][2] * fVar2 + in->m[0][0] * fVar1 + in->m[0][1] * fVar4);
-    out->m[0][0] = fVar3 * fVar1;
-    out->m[1][0] = fVar3 * fVar4;
-    out->m[2][0] = fVar3 * fVar2;
-    out->m[0][1] = -fVar3 * (in->m[0][1] * in->m[2][2] - in->m[2][1] * in->m[0][2]);
-    out->m[1][1] = fVar3 * (in->m[0][0] * in->m[2][2] - in->m[2][0] * in->m[0][2]);
-    out->m[2][1] = -fVar3 * (in->m[0][0] * in->m[2][1] - in->m[2][0] * in->m[0][1]);
-    out->m[0][2] = fVar3 * (in->m[0][1] * in->m[1][2] - in->m[1][1] * in->m[0][2]);
-    out->m[1][2] = -fVar3 * (in->m[0][0] * in->m[1][2] - in->m[1][0] * in->m[0][2]);
-    out->m[2][2] = fVar3 * (in->m[0][0] * in->m[1][1] - in->m[1][0] * in->m[0][1]);
-    out->t[0] = (-in->t[0] * out->m[0][0] - in->t[1] * out->m[0][1]) - in->t[2] * out->m[0][2];
-    out->t[1] = (-in->t[0] * out->m[1][0] - in->t[1] * out->m[1][1]) - in->t[2] * out->m[1][2];
-    out->t[2] = (-in->t[0] * out->m[2][0] - in->t[1] * out->m[2][1]) - in->t[2] * out->m[2][2];
+  a = in->m[1][1] * in->m[2][2] - in->m[2][1] * in->m[1][2];
+  c = -(in->m[1][0] * in->m[2][2] - in->m[2][0] * in->m[1][2]);
+  f = in->m[1][0] * in->m[2][1] - in->m[2][0] * in->m[1][1];
+  b = 1.f / (in->m[0][0] * a + in->m[0][1] * c + in->m[0][2] * f);
+  out->m[0][0] = b * a;
+  out->m[1][0] = b * c;
+  out->m[2][0] = b * f;
+  out->m[0][1] = -b * (in->m[0][1] * in->m[2][2] - in->m[2][1] * in->m[0][2]);
+  out->m[1][1] = b * (in->m[0][0] * in->m[2][2] - in->m[2][0] * in->m[0][2]);
+  out->m[2][1] = -b * (in->m[0][0] * in->m[2][1] - in->m[2][0] * in->m[0][1]);
+  out->m[0][2] = b * (in->m[0][1] * in->m[1][2] - in->m[1][1] * in->m[0][2]);
+  out->m[1][2] = -b * (in->m[0][0] * in->m[1][2] - in->m[1][0] * in->m[0][2]);
+  out->m[2][2] = b * (in->m[0][0] * in->m[1][1] - in->m[1][0] * in->m[0][1]);
+  out->t[0] = (-in->t[0] * out->m[0][0] - in->t[1] * out->m[0][1]) - in->t[2] * out->m[0][2];
+  out->t[1] = (-in->t[0] * out->m[1][0] - in->t[1] * out->m[1][1]) - in->t[2] * out->m[1][2];
+  out->t[2] = (-in->t[0] * out->m[2][0] - in->t[1] * out->m[2][1]) - in->t[2] * out->m[2][2];
 }
-#else
-/* clang-format off */
-#pragma push
-#pragma optimization_level 0
-#pragma optimizewithasm off
-const float one = 1.f;
-asm void salInvertMatrix(SND_FMATRIX* out, const SND_FMATRIX* in) {
-  nofralloc
-  lfs f5, 0x20(r4)
-  lfs f10, 0xc(r4)
-  lfs f0, 0x14(r4)
-  lfs f6, 0x1c(r4)
-  fmuls f3, f10, f5
-  lfs f9, 0x18(r4)
-  lfs f8, 0x10(r4)
-  fmuls f4, f6, f0
-  fmuls f2, f9, f0
-  lfs f1, 0(r4)
-  fmuls f7, f8, f5
-  lfs f0, 4(r4)
-  fmuls f5, f10, f6
-  fsubs f6, f3, f2
-  fsubs f7, f7, f4
-  lfs f2, 8(r4)
-  fmuls f4, f9, f8
-  lfs f3, one
-  fneg f6, f6
-  fmuls f1, f1, f7
-  fsubs f4, f5, f4
-  fmuls f0, f0, f6
-  fmuls f2, f2, f4
-  fadds f0, f1, f0
-  fadds f0, f2, f0
-  fdivs f1, f3, f0
-  fmuls f0, f1, f7
-  fmuls f3, f1, f6
-  fmuls f2, f1, f4
-  stfs f0, 0(r3)
-  fneg f0, f1
-  stfs f3, 0xc(r3)
-  stfs f2, 0x18(r3)
-  lfs f5, 4(r4)
-  lfs f4, 0x20(r4)
-  lfs f3, 0x1c(r4)
-  lfs f2, 8(r4)
-  fmuls f4, f5, f4
-  fmuls f2, f3, f2
-  fsubs f2, f4, f2
-  fmuls f2, f0, f2
-  stfs f2, 4(r3)
-  lfs f5, 0(r4)
-  lfs f4, 0x20(r4)
-  lfs f3, 0x18(r4)
-  lfs f2, 8(r4)
-  fmuls f4, f5, f4
-  fmuls f2, f3, f2
-  fsubs f2, f4, f2
-  fmuls f2, f1, f2
-  stfs f2, 0x10(r3)
-  lfs f5, 0(r4)
-  lfs f4, 0x1c(r4)
-  lfs f3, 0x18(r4)
-  lfs f2, 4(r4)
-  fmuls f4, f5, f4
-  fmuls f2, f3, f2
-  fsubs f2, f4, f2
-  fmuls f2, f0, f2
-  stfs f2, 0x1c(r3)
-  lfs f5, 4(r4)
-  lfs f4, 0x14(r4)
-  lfs f3, 0x10(r4)
-  lfs f2, 8(r4)
-  fmuls f4, f5, f4
-  fmuls f2, f3, f2
-  fsubs f2, f4, f2
-  fmuls f2, f1, f2
-  stfs f2, 8(r3)
-  lfs f5, 0(r4)
-  lfs f4, 0x14(r4)
-  lfs f3, 0xc(r4)
-  lfs f2, 8(r4)
-  fmuls f4, f5, f4
-  fmuls f2, f3, f2
-  fsubs f2, f4, f2
-  fmuls f0, f0, f2
-  stfs f0, 0x14(r3)
-  lfs f4, 0(r4)
-  lfs f3, 0x10(r4)
-  lfs f2, 0xc(r4)
-  lfs f0, 4(r4)
-  fmuls f3, f4, f3
-  fmuls f0, f2, f0
-  fsubs f0, f3, f0
-  fmuls f0, f1, f0
-  stfs f0, 0x20(r3)
-  lfs f0, 0x24(r4)
-  lfs f4, 0(r3)
-  fneg f5, f0
-  lfs f3, 0x28(r4)
-  lfs f2, 4(r3)
-  lfs f1, 0x2c(r4)
-  lfs f0, 8(r3)
-  fmuls f4, f5, f4
-  fmuls f2, f3, f2
-  fmuls f0, f1, f0
-  fsubs f1, f4, f2
-  fsubs f0, f1, f0
-  stfs f0, 0x24(r3)
-  lfs f0, 0x24(r4)
-  lfs f4, 0xc(r3)
-  fneg f5, f0
-  lfs f3, 0x28(r4)
-  lfs f2, 0x10(r3)
-  lfs f1, 0x2c(r4)
-  lfs f0, 0x14(r3)
-  fmuls f4, f5, f4
-  fmuls f2, f3, f2
-  fmuls f0, f1, f0
-  fsubs f1, f4, f2
-  fsubs f0, f1, f0
-  stfs f0, 0x28(r3)
-  lfs f1, 0x24(r4)
-  lfs f0, 0x18(r3)
-  fneg f4, f1
-  lfs f3, 0x28(r4)
-  lfs f2, 0x1c(r3)
-  lfs f1, 0x2c(r4)
-  fmuls f4, f4, f0
-  lfs f0, 0x20(r3)
-  fmuls f2, f3, f2
-  fmuls f0, f1, f0
-  fsubs f1, f4, f2
-  fsubs f0, f1, f0
-  stfs f0, 0x2c(r3)
-  blr
-}
-#pragma pop
-/* clang-format on */
-#endif