2024-11-01 19:06:07 +00:00
|
|
|
--!strict
|
|
|
|
local function prequire(name) local success, result = pcall(require, name); return if success then result else nil end
|
|
|
|
local bench = script and require(script.Parent.bench_support) or prequire("bench_support") or require("../bench_support")
|
|
|
|
|
|
|
|
function test()
|
|
|
|
|
|
|
|
type Vertex = { p: vector, uv: vector, n: vector, t: vector, b: vector, h: number }
|
|
|
|
|
|
|
|
local grid_size = 100
|
|
|
|
|
|
|
|
local mesh: {
|
|
|
|
vertices: {Vertex},
|
|
|
|
indices: {number},
|
|
|
|
triangle_cone_p: {vector},
|
|
|
|
triangle_cone_n: {vector}
|
|
|
|
} = {
|
|
|
|
vertices = table.create(grid_size * grid_size),
|
|
|
|
indices = table.create((grid_size - 1) * (grid_size - 1) * 6),
|
|
|
|
triangle_cone_p = table.create((grid_size - 1) * (grid_size - 1) * 2),
|
|
|
|
triangle_cone_n = table.create((grid_size - 1) * (grid_size - 1) * 2)
|
|
|
|
}
|
|
|
|
|
|
|
|
function init_vertices()
|
|
|
|
local i = 1
|
|
|
|
for y = 1,grid_size do
|
|
|
|
for x = 1,grid_size do
|
|
|
|
local v: Vertex = {}
|
|
|
|
|
|
|
|
v.p = vector.create(x, y, math.cos(x) + math.sin(y))
|
|
|
|
v.uv = vector.create((x-1)/(grid_size-1), (y-1)/(grid_size-1), 0)
|
|
|
|
v.n = vector.create(0, 0, 0)
|
|
|
|
v.b = vector.create(0, 0, 0)
|
|
|
|
v.t = vector.create(0, 0, 0)
|
|
|
|
v.h = 0
|
|
|
|
|
|
|
|
mesh.vertices[i] = v
|
|
|
|
i += 1
|
|
|
|
end
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
|
|
|
function init_indices()
|
|
|
|
local i = 1
|
|
|
|
for y = 1,grid_size-1 do
|
|
|
|
for x = 1,grid_size-1 do
|
|
|
|
mesh.indices[i] = x + (y-1)*grid_size
|
|
|
|
i += 1
|
|
|
|
mesh.indices[i] = x + y*grid_size
|
|
|
|
i += 1
|
|
|
|
mesh.indices[i] = (x+1) + (y-1)*grid_size
|
|
|
|
i += 1
|
|
|
|
mesh.indices[i] = (x+1) + (y-1)*grid_size
|
|
|
|
i += 1
|
|
|
|
mesh.indices[i] = x + y*grid_size
|
|
|
|
i += 1
|
|
|
|
mesh.indices[i] = (x+1) + y*grid_size
|
|
|
|
i += 1
|
|
|
|
end
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
|
|
|
function calculate_normals()
|
|
|
|
local norm_sum = 0
|
|
|
|
|
|
|
|
for i = 1,#mesh.indices,3 do
|
|
|
|
local a = mesh.vertices[mesh.indices[i]]
|
|
|
|
local b = mesh.vertices[mesh.indices[i + 1]]
|
|
|
|
local c = mesh.vertices[mesh.indices[i + 2]]
|
|
|
|
|
|
|
|
local n = vector.cross(a.p - b.p, a.p - c.p)
|
|
|
|
|
|
|
|
a.n += n
|
|
|
|
b.n += n
|
|
|
|
c.n += n
|
|
|
|
end
|
|
|
|
|
|
|
|
for _,v in ipairs(mesh.vertices) do
|
|
|
|
v.n = vector.normalize(v.n)
|
|
|
|
|
|
|
|
norm_sum += vector.dot(v.n, v.n)
|
|
|
|
end
|
|
|
|
|
|
|
|
return norm_sum
|
|
|
|
end
|
|
|
|
|
|
|
|
function compute_triangle_cones()
|
|
|
|
local mesh_area = 0
|
|
|
|
|
Fix mesh-normal-vector benchmark array access (#1514)
mesh-normal-scalar correctly fills sequential values in the output for
triangle cone function, but mesh-normal-vector accidentally reuses the
loop index, which results in writes to every third index of the array
(1, 4, etc.).
This is both slower (as the table turns into a hash map), and incorrect,
especially as we have a scalar version of the benchmark that does the
right thing.
Note: there's a bunch of inefficiencies in the benchmark code that I
have not fixed (around field access mostly, e.g. writing to `v.n` and
then immediately reading it again). These are not ideal for performance,
but they can be valuable to keep as is because this redundancy is common
in real-world code, and it would be nice to see codegen optimizations
eliminating most of that overhead. This one, however, is a straight up
bug, and sparse arrays should not really be the thing this benchmark
hits.
2024-11-11 20:39:09 +00:00
|
|
|
local pos = 1
|
|
|
|
|
2024-11-01 19:06:07 +00:00
|
|
|
for i = 1,#mesh.indices,3 do
|
|
|
|
local p0 = mesh.vertices[mesh.indices[i]]
|
|
|
|
local p1 = mesh.vertices[mesh.indices[i + 1]]
|
|
|
|
local p2 = mesh.vertices[mesh.indices[i + 2]]
|
|
|
|
|
|
|
|
local p10 = p1.p - p0.p
|
|
|
|
local p20 = p2.p - p0.p
|
|
|
|
|
|
|
|
local normal = vector.cross(p10, p20)
|
|
|
|
|
|
|
|
local area = vector.magnitude(normal)
|
|
|
|
local invarea = (area == 0) and 0 or 1 / area;
|
|
|
|
|
Fix mesh-normal-vector benchmark array access (#1514)
mesh-normal-scalar correctly fills sequential values in the output for
triangle cone function, but mesh-normal-vector accidentally reuses the
loop index, which results in writes to every third index of the array
(1, 4, etc.).
This is both slower (as the table turns into a hash map), and incorrect,
especially as we have a scalar version of the benchmark that does the
right thing.
Note: there's a bunch of inefficiencies in the benchmark code that I
have not fixed (around field access mostly, e.g. writing to `v.n` and
then immediately reading it again). These are not ideal for performance,
but they can be valuable to keep as is because this redundancy is common
in real-world code, and it would be nice to see codegen optimizations
eliminating most of that overhead. This one, however, is a straight up
bug, and sparse arrays should not really be the thing this benchmark
hits.
2024-11-11 20:39:09 +00:00
|
|
|
mesh.triangle_cone_p[pos] = (p0.p + p1.p + p2.p) / 3
|
|
|
|
mesh.triangle_cone_n[pos] = normal * invarea
|
|
|
|
pos += 1
|
2024-11-01 19:06:07 +00:00
|
|
|
|
|
|
|
mesh_area += area
|
|
|
|
end
|
|
|
|
|
|
|
|
return mesh_area
|
|
|
|
end
|
|
|
|
|
|
|
|
function compute_tangent_space()
|
|
|
|
local checksum = 0
|
|
|
|
|
|
|
|
for i = 1,#mesh.indices,3 do
|
|
|
|
local a = mesh.vertices[mesh.indices[i]]
|
|
|
|
local b = mesh.vertices[mesh.indices[i + 1]]
|
|
|
|
local c = mesh.vertices[mesh.indices[i + 2]]
|
|
|
|
|
|
|
|
local vba = b.p - a.p
|
|
|
|
local vca = c.p - a.p
|
|
|
|
|
|
|
|
local uvba = b.uv - a.uv
|
|
|
|
local uvca = c.uv - a.uv
|
|
|
|
|
|
|
|
local r = 1.0 / (uvba.X * uvca.Y - uvca.X * uvba.Y);
|
|
|
|
|
|
|
|
local sdir = (uvca.Y * vba - uvba.Y * vca) * r
|
|
|
|
local tdir = (uvba.X * vca - uvca.X * vba) * r
|
|
|
|
|
|
|
|
a.t += sdir
|
|
|
|
b.t += sdir
|
|
|
|
c.t += sdir
|
|
|
|
|
|
|
|
a.b += tdir
|
|
|
|
b.b += tdir
|
|
|
|
c.b += tdir
|
|
|
|
end
|
|
|
|
|
|
|
|
for _,v in ipairs(mesh.vertices) do
|
|
|
|
local t = v.t
|
|
|
|
|
|
|
|
-- Gram-Schmidt orthogonalize
|
|
|
|
v.t = vector.normalize(t - v.n * vector.dot(v.n, t))
|
|
|
|
|
|
|
|
local ht = vector.dot(vector.cross(v.n, t), v.b)
|
|
|
|
|
|
|
|
v.h = ht < 0 and -1 or 1
|
|
|
|
|
|
|
|
checksum += v.t.X + v.h
|
|
|
|
end
|
|
|
|
|
|
|
|
return checksum
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
|
|
init_vertices()
|
|
|
|
init_indices()
|
|
|
|
calculate_normals()
|
|
|
|
compute_triangle_cones()
|
|
|
|
compute_tangent_space()
|
|
|
|
end
|
|
|
|
|
|
|
|
bench.runCode(test, "mesh-normal-vector")
|