236 lines
5.3 KiB
V
236 lines
5.3 KiB
V
import gg.m4
|
|
|
|
pub fn test_m4() {
|
|
unsafe {
|
|
// Test Mat4
|
|
mut a := m4.Mat4{ e: [
|
|
f32(0), 1, 2, 3,
|
|
4, 5, 6, 7,
|
|
8, 9, 10, 11,
|
|
12, 13, 14, 15,
|
|
]!
|
|
}
|
|
mut b := m4.Mat4{}
|
|
mut c := m4.Mat4{}
|
|
|
|
// equal test
|
|
assert a.e == [
|
|
f32(0), 1, 2, 3,
|
|
4, 5, 6, 7,
|
|
8, 9, 10, 11,
|
|
12, 13, 14, 15,
|
|
]!
|
|
|
|
// copy test
|
|
b.copy(a)
|
|
assert a.e == b.e
|
|
|
|
// test: transpose, scale
|
|
assert b.transpose().mul_scalar(2.0).mul_scalar(0.5).transpose().e == a.e
|
|
assert b.sum_all() == 120.0
|
|
|
|
// test rows/columns set/get
|
|
for i in 0 .. 4 {
|
|
b = m4.zero_m4()
|
|
b.set_row(i, m4.Vec4{ e: [f32(1.0), 2, 3, 4]! })
|
|
assert b.get_f(0, i) == 1.0
|
|
assert b.get_f(1, i) == 2.0
|
|
assert b.get_f(2, i) == 3.0
|
|
assert b.get_f(3, i) == 4.0
|
|
// println(b)
|
|
c = m4.zero_m4()
|
|
c.set_col(i, m4.Vec4{ e: [f32(1.0), 2, 3, 4]! })
|
|
assert c.get_f(i, 0) == 1.0
|
|
assert c.get_f(i, 1) == 2.0
|
|
assert c.get_f(i, 2) == 3.0
|
|
assert c.get_f(i, 3) == 4.0
|
|
// println(c)
|
|
}
|
|
}
|
|
}
|
|
|
|
fn test_swap_col_row() {
|
|
unsafe {
|
|
// swap_col / swap_row
|
|
b := m4.Mat4{ e: [
|
|
f32(1), 2, 3, 4,
|
|
5, 6, 7, 8,
|
|
9, 10, 11, 12,
|
|
13, 14, 15, 16,
|
|
]!
|
|
}
|
|
b.swap_col(0, 2)
|
|
assert b.e == [
|
|
f32(3), 2, 1, 4,
|
|
7, 6, 5, 8,
|
|
11, 10, 9, 12,
|
|
15, 14, 13, 16,
|
|
]!
|
|
b = m4.Mat4{ e: [
|
|
f32(1), 2, 3, 4,
|
|
5, 6, 7, 8,
|
|
9, 10, 11, 12,
|
|
13, 14, 15, 16,
|
|
]!
|
|
}
|
|
b.swap_row(0, 2)
|
|
assert b.e == [
|
|
f32(9), 10, 11, 12,
|
|
5, 6, 7, 8,
|
|
1, 2, 3, 4,
|
|
13, 14, 15, 16,
|
|
]!
|
|
}
|
|
}
|
|
|
|
fn test_sum_sub() {
|
|
unsafe {
|
|
// test sum/sub
|
|
b := m4.unit_m4()
|
|
c := m4.unit_m4()
|
|
assert m4.sub(m4.add(b, c), b).e == m4.unit_m4().e
|
|
assert (b + c - b).e == m4.unit_m4().e
|
|
}
|
|
}
|
|
|
|
fn test_transpose() {
|
|
unsafe {
|
|
b := m4.Mat4{ e: [
|
|
f32(0), 1, 2, 3,
|
|
4, 5, 6, 7,
|
|
8, 9, 10, 11,
|
|
12, 13, 14, 15,
|
|
]!
|
|
}
|
|
assert b.transpose().transpose().e == b.e
|
|
}
|
|
}
|
|
|
|
fn test_multiplication() {
|
|
unsafe {
|
|
b := m4.Mat4{ e: [
|
|
f32(1), 0, 0, 0,
|
|
0, 2, 0, 0,
|
|
0, 0, 3, 0,
|
|
0, 0, 0, 4,
|
|
]!
|
|
}
|
|
c := m4.Mat4{ e: [
|
|
f32(1), 2, 3, 4,
|
|
5, 6, 7, 8,
|
|
9, 10, 11, 12,
|
|
13, 14, 15, 16,
|
|
]!
|
|
}
|
|
|
|
assert (c * c).e == [
|
|
f32(90),100,110,120,
|
|
202,228,254,280,
|
|
314,356,398,440,
|
|
426,484,542,600,
|
|
]!
|
|
|
|
assert m4.mul(c, c).e == [
|
|
f32(90),100,110,120,
|
|
202,228,254,280,
|
|
314,356,398,440,
|
|
426,484,542,600,
|
|
]!
|
|
|
|
assert m4.mul(b, c).e == [
|
|
f32(1), 2, 3, 4,
|
|
10, 12, 14, 16,
|
|
27, 30, 33, 36,
|
|
52, 56, 60, 64,
|
|
]!
|
|
|
|
assert (b * c).e == [
|
|
f32(1), 2, 3, 4,
|
|
10, 12, 14, 16,
|
|
27, 30, 33, 36,
|
|
52, 56, 60, 64,
|
|
]!
|
|
|
|
assert m4.det(b) == 24
|
|
}
|
|
}
|
|
|
|
fn test_det() {
|
|
unsafe {
|
|
b := m4.Mat4{ e: [
|
|
f32(5), 6, 6, 8,
|
|
2, 2, 2, 8,
|
|
6, 6, 2, 8,
|
|
2, 3, 6, 7,
|
|
]!
|
|
}
|
|
assert m4.det(b) == -8
|
|
|
|
c := m4.Mat4{ e: [
|
|
f32(1), 8, 2, 3,
|
|
8, 2, 3, 1,
|
|
2, 3, 3, 2,
|
|
3, 1, 2, 4,
|
|
]!
|
|
}
|
|
// println("*** INVERSE ****")
|
|
// println(m4.mul(b.inverse(),b))
|
|
// println(m4.clean_small(m4.mul(c.inverse(),c)))
|
|
// println("****************")
|
|
assert m4.mul(b.inverse(), b).e == m4.unit_m4().e
|
|
assert m4.mul(c.inverse(), c).is_equal(m4.unit_m4())
|
|
}
|
|
}
|
|
|
|
fn test_vec4() {
|
|
// Test Vec4
|
|
// println("*** Vector4 ****")
|
|
assert m4.vec3(1,2,3) == m4.Vec4{[f32(1), 2, 3, 1]!}
|
|
mut v := m4.Vec4{[f32(1), 2, 3, 4]!}
|
|
assert v * v.inv() == 4
|
|
assert v.mul_scalar(1.0 / v.mod()).mod() == 1
|
|
assert v + m4.Vec4{ e: [f32(5), 6, 7, 8]! } == m4.Vec4{ e: [f32(6), 8, 10, 12]! }
|
|
assert v - m4.Vec4{ e: [f32(1), 2, 3, 4]! } == m4.Vec4{ e: [f32(0), 0, 0, 0]! }
|
|
assert v.mul_vec4(m4.Vec4{ e: [f32(2), 2, 2, 2]! }) == m4.Vec4{ e: [f32(2), 4, 6, 8]! }
|
|
assert f32_abs(v.normalize().mod() - 1) < m4.precision
|
|
v = m4.Vec4{[f32(1), 2, 3, 0]!}
|
|
assert f32_abs(v.normalize3().mod3() - 1) < m4.precision
|
|
assert f32_abs(v.normalize3().mod() - 1) < m4.precision
|
|
// cross product
|
|
// x y z
|
|
// 1 2 3 ==> -3 6 -3 0
|
|
// 4 5 6
|
|
// println(m4.Vec4{[f32(1),2,3,2]!} % m4.Vec4{[f32(4),5,6,2]!})
|
|
assert m4.Vec4{[f32(1), 2, 3, 0]!} % m4.Vec4{[f32(4), 5, 6, 0]!} == m4.Vec4{[ f32(-3), 6, -3, 0, ]!}
|
|
assert m4.Vec4{[f32(1), 2, 3, 13]!} % m4.Vec4{[f32(4), 5, 6, 11]!} == m4.Vec4{[ f32(-3), 6, -3, 0, ]!}
|
|
// matrix * vector
|
|
a := m4.Mat4{ e: [
|
|
f32(1),2,3,4
|
|
5,6,7,8
|
|
9,10,11,12
|
|
13,14,15,16
|
|
]!
|
|
}
|
|
assert m4.mul_vec(a, m4.Vec4{[f32(1), 2, 3, 4]!}) == m4.Vec4{[ f32(30), 70, 110,150, ]!}
|
|
// Rotation
|
|
// println("*** Rotation ****")
|
|
rotx := m4.rotate(m4.rad(-90), m4.Vec4{ e: [f32(1.0), 0, 0, 0]! }).clean()
|
|
roty := m4.rotate(m4.rad(-90), m4.Vec4{ e: [f32(0), 1.0, 0, 0]! }).clean()
|
|
rotz := m4.rotate(m4.rad(-90), m4.Vec4{ e: [f32(0), 0, 1, 0]! }).clean()
|
|
// println( rotx )
|
|
// println( roty )
|
|
// println( rotz )
|
|
// println( m4.mul_vec(rotx, m4.Vec4{e:[f32(0),0,1,0]!}).clean())
|
|
assert m4.mul_vec(roty, m4.Vec4{ e: [f32(1.0), 0.0, 0, 0]! }).clean() == m4.Vec4{ e: [f32(0), 0.0, -1, 0]! }
|
|
assert m4.mul_vec(rotz, m4.Vec4{ e: [f32(1.0), 0.0, 0, 0]! }).clean() == m4.Vec4{ e: [f32(0), 1, 0, 0]! }
|
|
assert m4.mul_vec(rotx, m4.Vec4{ e: [f32(0), 0, 1, 0]! }).clean() == m4.Vec4{ e: [f32(0), -1, 0, 0]! }
|
|
// println("****************")
|
|
}
|
|
|
|
fn test_proj() {
|
|
ort := m4.ortho(0,300,0,200,0,0)
|
|
assert m4.mul_vec(ort, m4.Vec4{[ f32(150), 100, 0, 1]!}) == m4.Vec4{[ f32(0), 0, 0, 1]!}
|
|
assert m4.mul_vec(ort, m4.Vec4{[ f32(0), 0, 0, 1]!}) == m4.Vec4{[ f32(-1), -1, 0, 1]!}
|
|
assert m4.mul_vec(ort, m4.Vec4{[ f32(300), 200, 0, 1]!}) == m4.Vec4{[ f32(1), 1, 0, 1]!}
|
|
}
|