diff --git a/include/PR/gu.h b/include/PR/gu.h
index e005382df..30be352cf 100644
--- a/include/PR/gu.h
+++ b/include/PR/gu.h
@@ -185,8 +185,8 @@ extern int guRandom(void);
  */
 extern float sinf(float angle);
 extern float cosf(float angle);
-extern signed short sins (unsigned short angle);
-extern signed short coss (unsigned short angle);
+extern signed int sins (unsigned int angle);
+extern signed int coss (unsigned int angle);
 
 /*
  *  Dump routines for low-level display lists
diff --git a/ld/libfiles.ntsc-final.inc b/ld/libfiles.ntsc-final.inc
index 398326538..4aa65e67a 100644
--- a/ld/libfiles.ntsc-final.inc
+++ b/ld/libfiles.ntsc-final.inc
@@ -157,7 +157,6 @@
 	build/ROMID/lib/ultra/io/devmgr.o (section); \
 	build/ROMID/lib/ultra/os/thread.o (section); \
 	build/ROMID/lib/ultra/os/destroythread.o (section); \
-	build/ROMID/lib/ultra/gu/mtxutil.o (section); \
 	build/ROMID/lib/ultra/libc/string.o (section); \
 	build/ROMID/lib/ultra/libc/ldiv.o (section); \
 	build/ROMID/lib/ultra/io/siacs.o (section); \
diff --git a/src/include/constants.h b/src/include/constants.h
index 8ef4a2be0..18911f814 100644
--- a/src/include/constants.h
+++ b/src/include/constants.h
@@ -13,8 +13,6 @@
 #define false 0
 #define true  1
 
-#define osSyncPrintf
-
 #define S32_MAX  2147483647
 #define U32_MAX  4294967295
 #define MINFLOAT ((float)-3.40282346638528860e+38)
diff --git a/src/lib/mtx.c b/src/lib/mtx.c
index 4f14d9501..f2977f65e 100644
--- a/src/lib/mtx.c
+++ b/src/lib/mtx.c
@@ -456,30 +456,6 @@ void mtx4Align(f32 mtx[4][4], f32 angle, f32 x, f32 y, f32 z)
 	guAlignF(mtx, angle, x, y, z);
 }
 
-void mtx4LoadRotationFrom(f32 src[4][4], f32 dst[4][4])
-{
-	dst[0][0] = src[0][0];
-	dst[0][1] = src[1][0];
-	dst[0][2] = src[2][0];
-
-	dst[1][0] = src[0][1];
-	dst[1][1] = src[1][1];
-	dst[1][2] = src[2][1];
-
-	dst[2][0] = src[0][2];
-	dst[2][1] = src[1][2];
-	dst[2][2] = src[2][2];
-
-	dst[3][0] = 0;
-	dst[3][1] = 0;
-	dst[3][2] = 0;
-
-	dst[0][3] = 0;
-	dst[1][3] = 0;
-	dst[2][3] = 0;
-	dst[3][3] = 1;
-}
-
 void mtx000170e4(f32 src[4][4], f32 dst[4][4])
 {
 	f32 tmp = (src[0][0] * src[0][0] + src[1][0] * src[1][0] + src[2][0] * src[2][0]);
diff --git a/src/lib/mtxasm.s b/src/lib/mtxasm.s
index ea239dc4f..6539e48bb 100644
--- a/src/lib/mtxasm.s
+++ b/src/lib/mtxasm.s
@@ -450,13 +450,21 @@ glabel mtx00015f88
 	jr     $ra
  	nop
 
+glabel guMtxF2L
+	lui    $at, 0x4780
+	mtc1   $at, $f0
+	j      _mtxF2L
+	mtc1   $at, $f2
+
 glabel mtx00016054
 	lui    $t2, %hi(var8005ef10)
 	addiu  $t2, $t2, %lo(var8005ef10)
-	addiu  $t0, $a0, 0x40
-	lui    $t1, 0xffff
 	lwc1   $f0, 0x0($t2)
 	lwc1   $f2, 0x4($t2)
+
+glabel _mtxF2L
+	lui    $t1, 0xffff
+	addiu  $t0, $a0, 0x40
 .L0001606c:
 	lwc1   $f4, 0x0($a0)
 	lwc1   $f6, 0x4($a0)
@@ -495,3 +503,53 @@ glabel mtx00016054
 	addiu  $a1, $a1, 0x8
 	jr     $ra
  	nop
+
+glabel mtx4LoadRotationFrom
+	lw     $t0, 0x00($a0)
+	lw     $t1, 0x10($a0)
+	lw     $t2, 0x20($a0)
+	lw     $t3, 0x04($a0)
+	lw     $t4, 0x14($a0)
+	lw     $t5, 0x24($a0)
+	lw     $t6, 0x08($a0)
+	lw     $t7, 0x18($a0)
+	lw     $t8, 0x28($a0)
+	sw     $t0, 0x00($a1)
+	sw     $t1, 0x04($a1)
+	sw     $t2, 0x08($a1)
+	sw     $t3, 0x10($a1)
+	sw     $t4, 0x14($a1)
+	sw     $t5, 0x18($a1)
+	sw     $t6, 0x20($a1)
+	sw     $t7, 0x24($a1)
+	sw     $t8, 0x28($a1)
+	sw     $zero, 0x30($a1)
+	sw     $zero, 0x34($a1)
+	sw     $zero, 0x38($a1)
+	sw     $zero, 0x0c($a1)
+	sw     $zero, 0x1c($a1)
+	sw     $zero, 0x2c($a1)
+	lui    $at, 0x3f80
+	jr     $ra
+	sw     $at, 0x3c($a1)
+
+glabel guScaleF
+	lui    $at, 0x3f80
+	sw     $zero, 0x4($a0)
+	sw     $zero, 0x8($a0)
+	sw     $zero, 0xc($a0)
+	sw     $zero, 0x10($a0)
+	sw     $zero, 0x18($a0)
+	sw     $zero, 0x1c($a0)
+	sw     $zero, 0x20($a0)
+	sw     $zero, 0x24($a0)
+	sw     $zero, 0x2c($a0)
+	sw     $zero, 0x30($a0)
+	sw     $zero, 0x34($a0)
+	sw     $zero, 0x38($a0)
+	sw     $a1, 0x0($a0)
+	sw     $a2, 0x14($a0)
+	sw     $a3, 0x28($a0)
+	jr     $ra
+	sw     $at, 0x3c($a0)
+
diff --git a/src/lib/ultra/gu/align.c b/src/lib/ultra/gu/align.c
index 1dde87db4..c0d758a09 100644
--- a/src/lib/ultra/gu/align.c
+++ b/src/lib/ultra/gu/align.c
@@ -1,5 +1,6 @@
 #include "guint.h"
 #include "constants.h"
+#include "lib/mtx.h"
 
 void guAlignF(float mf[4][4], float a, float x, float y, float z)
 {
@@ -13,8 +14,6 @@ void guAlignF(float mf[4][4], float a, float x, float y, float z)
 	c = cosf(a);
 	h = sqrtf(x * x + z * z);
 
-	guMtxIdentF(mf);
-
 	if (h != 0) {
 		hinv = 1 / h;
 
@@ -37,5 +36,7 @@ void guAlignF(float mf[4][4], float a, float x, float y, float z)
 		mf[1][3] = 0;
 		mf[2][3] = 0;
 		mf[3][3] = 1;
+	} else {
+		mtx4LoadIdentity((Mtxf *) mf);
 	}
 }
diff --git a/src/lib/ultra/gu/coss.c b/src/lib/ultra/gu/coss.c
index f6c805f1c..f58b8b214 100644
--- a/src/lib/ultra/gu/coss.c
+++ b/src/lib/ultra/gu/coss.c
@@ -1,6 +1,6 @@
 #include "guint.h"
 
-signed short coss(unsigned short x)
+signed int coss(unsigned int x)
 {
-	return sins((unsigned short) (x + 0x4000));
+	return sins((unsigned int) (x + 0x4000));
 }
diff --git a/src/lib/ultra/gu/frustum.c b/src/lib/ultra/gu/frustum.c
index 6857939f3..74b4d3083 100644
--- a/src/lib/ultra/gu/frustum.c
+++ b/src/lib/ultra/gu/frustum.c
@@ -4,20 +4,23 @@ void guFrustumF(float mf[4][4], float l, float r, float b, float t, float n, flo
 {
 	int i, j;
 
-	guMtxIdentF(mf);
+	mf[0][0] = 2 * n / (r - l) * scale;
+	mf[0][1] = 0;
+	mf[0][2] = 0;
+	mf[0][3] = 0;
 
-	mf[0][0] = 2 * n / (r - l);
-	mf[1][1] = 2 * n / (t - b);
-	mf[2][0] = (r + l) / (r - l);
-	mf[2][1] = (t + b) / (t - b);
-	mf[2][2] = -(f + n) / (f - n);
-	mf[2][3] = -1;
-	mf[3][2] = -2 * f * n / (f - n);
+	mf[1][0] = 0;
+	mf[1][1] = 2 * n / (t - b) * scale;
+	mf[1][2] = 0;
+	mf[1][3] = 0;
+
+	mf[2][0] = (r + l) / (r - l) * scale;
+	mf[2][1] = (t + b) / (t - b) * scale;
+	mf[2][2] = -(f + n) / (f - n) * scale;
+	mf[2][3] = -scale;
+
+	mf[3][0] = 0;
+	mf[3][1] = 0;
+	mf[3][2] = -2 * f * n / (f - n) * scale;
 	mf[3][3] = 0;
-
-	for (i = 0; i < 4; i++) {
-		for (j = 0; j < 4; j++) {
-			mf[i][j] *= scale;
-		}
-	}
 }
diff --git a/src/lib/ultra/gu/lookatref.c b/src/lib/ultra/gu/lookatref.c
index 82125181d..80a6b047b 100644
--- a/src/lib/ultra/gu/lookatref.c
+++ b/src/lib/ultra/gu/lookatref.c
@@ -8,8 +8,6 @@ static void guLookAtReflectF(float mf[4][4], LookAt *l,
 {
 	float len, xLook, yLook, zLook, xRight, yRight, zRight;
 
-	guMtxIdentF(mf);
-
 	xLook = xAt - xEye;
 	yLook = yAt - yEye;
 	zLook = zAt - zEye;
diff --git a/src/lib/ultra/gu/mtxutil.c b/src/lib/ultra/gu/mtxutil.c
deleted file mode 100644
index 131e7ada5..000000000
--- a/src/lib/ultra/gu/mtxutil.c
+++ /dev/null
@@ -1,32 +0,0 @@
-#include <ultra64.h>
-
-void guMtxF2L(f32 mf[4][4], Mtx *m)
-{
-	int	i, j;
-	int	e1, e2;
-	int	*ai, *af;
-
-	ai = (int *) &m->m[0][0];
-	af = (int *) &m->m[2][0];
-
-	for (i = 0; i < 4; i++) {
-		for (j = 0; j < 2; j++) {
-			e1 = FTOFIX32(mf[i][j * 2]);
-			e2 = FTOFIX32(mf[i][j * 2 + 1]);
-
-			*(ai++) = (e1 & 0xffff0000) | ((e2 >> 16) & 0xffff);
-			*(af++) = ((e1 << 16) & 0xffff0000) | (e2 & 0xffff);
-		}
-	}
-}
-
-void guMtxIdentF(float mf[4][4])
-{
-	int	i, j;
-
-	for (i = 0; i < 4; i++) {
-		for (j = 0; j < 4; j++) {
-			mf[i][j] = i == j ? 1 : 0;
-		}
-	}
-}
diff --git a/src/lib/ultra/gu/perspective.c b/src/lib/ultra/gu/perspective.c
index 9b4035fca..9ad495176 100644
--- a/src/lib/ultra/gu/perspective.c
+++ b/src/lib/ultra/gu/perspective.c
@@ -5,23 +5,28 @@ void guPerspectiveF(float mf[4][4], u16 *perspNorm, float fovy, float aspect, fl
 	float cot;
 	int	i, j;
 
-	guMtxIdentF(mf);
-
 	fovy *= 3.1415926f / 180.0f;
 	cot = cosf(fovy * 0.5f) / sinf(fovy * 0.5f);
 
-	mf[0][0] = cot / aspect;
-	mf[1][1] = cot;
-	mf[2][2] = (near + far) / (near - far);
-	mf[2][3] = -1;
-	mf[3][2] = (2.0f * near * far) / (near - far);
-	mf[3][3] = 0;
+	mf[0][0] = cot / aspect * scale;
+	mf[0][1] = 0;
+	mf[0][2] = 0;
+	mf[0][3] = 0;
 
-	for (i = 0; i < 4; i++) {
-		for (j = 0; j < 4; j++) {
-			mf[i][j] *= scale;
-		}
-	}
+	mf[1][0] = 0;
+	mf[1][1] = cot * scale;
+	mf[1][2] = 0;
+	mf[1][3] = 0;
+
+	mf[2][0] = 0;
+	mf[2][1] = 0;
+	mf[2][2] = (near + far) / (near - far) * scale;
+	mf[2][3] = -scale;
+
+	mf[3][0] = 0;
+	mf[3][1] = 0;
+	mf[3][2] = (2.0f * near * far) / (near - far) * scale;
+	mf[3][3] = 0;
 
 	if (perspNorm != (u16 *) NULL) {
 		if (near + far <= 2.0f) {
diff --git a/src/lib/ultra/gu/rotate.c b/src/lib/ultra/gu/rotate.c
index 46d48a16b..68cccdbfd 100644
--- a/src/lib/ultra/gu/rotate.c
+++ b/src/lib/ultra/gu/rotate.c
@@ -16,8 +16,6 @@ void guRotateF(float mf[4][4], float a, float x, float y, float z)
 	bc = y * z * t;
 	ca = z * x * t;
 
-	guMtxIdentF(mf);
-
 	t = x * x;
 	mf[0][0] = t + cosine * (1 - t);
 	mf[2][1] = bc - x * sine;
@@ -32,4 +30,13 @@ void guRotateF(float mf[4][4], float a, float x, float y, float z)
 	mf[2][2] = t + cosine * (1 - t);
 	mf[1][0] = ab - z * sine;
 	mf[0][1] = ab + z * sine;
+
+	mf[0][3] = 0;
+	mf[1][3] = 0;
+	mf[2][3] = 0;
+
+	mf[3][0] = 0;
+	mf[3][1] = 0;
+	mf[3][2] = 0;
+	mf[3][3] = 1;
 }
diff --git a/src/lib/ultra/gu/scale.c b/src/lib/ultra/gu/scale.c
index fdb838d97..5ac3fcb26 100644
--- a/src/lib/ultra/gu/scale.c
+++ b/src/lib/ultra/gu/scale.c
@@ -1,15 +1,5 @@
 #include <ultra64.h>
 
-void guScaleF(float mf[4][4], float x, float y, float z)
-{
-	guMtxIdentF(mf);
-
-	mf[0][0] = x;
-	mf[1][1] = y;
-	mf[2][2] = z;
-	mf[3][3] = 1;
-}
-
 void guScale(Mtx *m, float x, float y, float z)
 {
 	f32 mf[4][4];
diff --git a/src/lib/ultra/gu/sins.c b/src/lib/ultra/gu/sins.c
index be1c8e229..8e00a4e7e 100644
--- a/src/lib/ultra/gu/sins.c
+++ b/src/lib/ultra/gu/sins.c
@@ -106,19 +106,19 @@ s16 sintable[] = {
 	0x7ffe, 0x7ffe, 0x7ffe, 0x7fff,
 };
 
-signed short sins(unsigned short x)
+signed int sins(unsigned int x)
 {
-	signed short val;
+	signed int val;
 
-	x >>= 4;
-
-	if (x & 0x400) {
-		val = sintable[0x3ff - (x & 0x3ff)];
-	} else {
-		val = sintable[x & 0x3ff];
-	}
+	x >>= 3;
 
 	if (x & 0x800) {
+		val = sintable[(0x7ff - (x & 0x7ff)) >> 1];
+	} else {
+		val = sintable[(x & 0x7fe) >> 1];
+	}
+
+	if (x & 0x1000) {
 		return -val;
 	} else {
 		return val;