diff --git a/gpu.h b/gpu.h
index 8b0623e..e1e5a30 100644
--- a/gpu.h
+++ b/gpu.h
@@ -6,12 +6,18 @@
 
 #define SWAPCHAIN_IMG_BUFFER_SIZE 4
 
-typedef float mat4_t[16];
-
 typedef struct {
 	float x, y, z;
 } vec3_t;
 
+typedef struct {
+	float x, y, z, w;
+} vec4_t;
+
+typedef struct {
+	vec4_t x, y, z, w;
+} mat4_t;
+
 typedef struct {
 	vec3_t position;
 	vec3_t colour;
diff --git a/main.c b/main.c
index 37a3c46..5a2ab0b 100644
--- a/main.c
+++ b/main.c
@@ -31,55 +31,53 @@ typedef enum {
 
 static axis_t rot_axis = AXIS_Y;
 
-// clang-format off
 #define ID \
 	{ \
-		1, 0, 0, 0, \
-		0, 1, 0, 0, \
-		0, 0, 1, 0, \
-		0, 0, 0, 1, \
+		{ 1, 0, 0, 0 }, \
+		{ 0, 1, 0, 0 }, \
+		{ 0, 0, 1, 0 }, \
+		{ 0, 0, 0, 1 }, \
 	}
 
 #define SCALE(sf) \
 	{ \
-		sf, 0,  0,  0, \
-		0,  sf, 0,  0, \
-		0,  0,  sf, 0, \
-		0,  0,  0,  1, \
+		{ sf, 0, 0, 0 }, \
+		{ 0, sf, 0, 0 }, \
+		{ 0, 0, sf, 0 }, \
+		{ 0, 0, 0, 1 }, \
 	}
 
 #define TRANSLATE(x, y, z) \
 	{ \
-		1, 0, 0, 0, \
-		0, 1, 0, 0, \
-		0, 0, 1, 0, \
-		x, y, z, 1, \
+		{ 1, 0, 0, 0 }, \
+		{ 0, 1, 0, 0 }, \
+		{ 0, 0, 1, 0 }, \
+		{ x, y, z, 1 }, \
 	}
 
 #define ROT_X(a) \
 	{ \
-		1, 0,       0,        0, \
-		0, cosf(a), sinf(a),  0, \
-		0, -sinf(a), cosf(a), 0, \
-		0, 0,       0,        1, \
+		{ 1, 0, 0, 0 }, \
+		{ 0, cosf(a), sinf(a), 0 }, \
+		{ 0, -sinf(a), cosf(a), 0 }, \
+		{ 0, 0, 0, 1 }, \
 	}
 
 #define ROT_Y(a) \
 	{ \
-		cosf(a), 0, -sinf(a), 0, \
-		0,       1, 0,        0, \
-		sinf(a), 0, cosf(a),  0, \
-		0,       0, 0,        1, \
+		{ cosf(a), 0, -sinf(a), 0 }, \
+		{ 0, 1, 0, 0 }, \
+		{ sinf(a), 0, cosf(a), 0 }, \
+		{ 0, 0, 0, 1 }, \
 	}
 
 #define ROT_Z(a) \
 	{ \
-		cosf(a),  sinf(a), 0, 0, \
-		-sinf(a), cosf(a), 0, 0, \
-		0,        0,       1, 0, \
-		0,        0,       0, 1, \
+		{ cosf(a), sinf(a), 0, 0 }, \
+		{ -sinf(a), cosf(a), 0, 0 }, \
+		{ 0, 0, 1, 0 }, \
+		{ 0, 0, 0, 1 }, \
 	}
-// clang-format on
 
 static void
 key_callback(GLFWwindow *window, int key, int scancode, int action, int mods)
@@ -108,32 +106,29 @@ key_callback(GLFWwindow *window, int key, int scancode, int action, int mods)
 	}
 }
 
-static void mul(mat4_t out, const mat4_t left, const mat4_t right)
+static mat4_t mul(mat4_t m, mat4_t n)
 {
-	for (unsigned y = 0; y < 4; ++y) {
-		for (unsigned x = 0; x < 4; ++x) {
-			out[y + 4 * x] = 0;
-			for (unsigned i = 0; i < 4; ++i)
-				out[y + 4 * x] += left[y + 4 * i] * right[i + 4 * x];
-		}
-	}
+#define ELEM(i, j) \
+	m.x.i *n.j.x + m.y.i *n.j.y + m.z.i *n.j.z + m.w.i *n.j.w
+	return (mat4_t) {
+		{ ELEM(x, x), ELEM(y, x), ELEM(z, x), ELEM(w, x) },
+		{ ELEM(x, y), ELEM(y, y), ELEM(z, y), ELEM(w, y) },
+		{ ELEM(x, z), ELEM(y, z), ELEM(z, z), ELEM(w, z) },
+		{ ELEM(x, w), ELEM(y, w), ELEM(z, w), ELEM(w, w) },
+	};
+#undef MUL_ELEM
 }
 
-static void persp(mat4_t out, float aspect, float fov, float ncp, float fcp)
+static mat4_t persp(float aspect, float fov, float ncp, float fcp)
 {
 	const float r = 2 * ncp * tanf(fov / 2);
 	const float t = r * aspect;
-
-	// clang-format off
-	const mat4_t m = {
-		ncp / t, 0,        0,                        0,
-		0,       -ncp / r, 0,                        0,
-		0,       0,        fcp / (fcp - ncp),        1,
-		0,       0,        -fcp * ncp / (fcp - ncp), 0,
+	return (mat4_t) {
+		{ ncp / t, 0, 0, 0 },
+		{ 0, -ncp / r, 0, 0 },
+		{ 0, 0, fcp / (fcp - ncp), 1 },
+		{ 0, 0, -fcp * ncp / (fcp - ncp), 0 },
 	};
-	// clang-format on
-
-	memcpy(out, m, sizeof(mat4_t));
 }
 
 int main()
@@ -163,8 +158,8 @@ int main()
 	mvp_t mvp = {
 		.model = SCALE(2),
 		.view = TRANSLATE(0, 0, 4.0),
+		.proj = persp((float)W / H, M_PI / 6, 0.1, 10.0),
 	};
-	persp(mvp.proj, (float)W / H, M_PI / 6, 0.1, 10.0);
 
 	const mat4_t rotx = ROT_X(M_PI / 128);
 	const mat4_t roty = ROT_Y(M_PI / 128);
@@ -173,17 +168,15 @@ int main()
 	while (!glfwWindowShouldClose(window)) {
 		glfwPollEvents();
 
-		mat4_t tmp;
-		memcpy(tmp, mvp.model, sizeof(mat4_t));
 		switch (rot_axis) {
 		case AXIS_X:
-			mul(mvp.model, rotx, tmp);
+			mvp.model = mul(rotx, mvp.model);
 			break;
 		case AXIS_Y:
-			mul(mvp.model, roty, tmp);
+			mvp.model = mul(roty, mvp.model);
 			break;
 		case AXIS_Z:
-			mul(mvp.model, rotz, tmp);
+			mvp.model = mul(rotz, mvp.model);
 			break;
 		}