Skip to content

Instantly share code, notes, and snippets.

@onionhammer
Last active August 29, 2015 13:56
Show Gist options
  • Save onionhammer/8979260 to your computer and use it in GitHub Desktop.
Save onionhammer/8979260 to your computer and use it in GitHub Desktop.
pnoise
# Imports
import math
# Constants
const RAND_MAX = 0x7fff
# Types
type
TVec2 = object
x, y: float
TNoise2DContext = object
rgradients, gradients, origins: array[0.. 255, TVec2]
permutations: array[0.. 255, int]
# Procedures
proc rand: cint {.importc: "rand", header: "<stdlib.h>".}
proc lerp(a, b, v: float): float {.inline, noinit.} =
a * (1 - v) + b * v
proc smooth(v: float): float {.inline, noinit.} =
v * v * (3 - 2 * v)
proc random_gradient: TVec2 {.noinit.} =
let v = rand().float / RAND_MAX * math.PI * 2.0
TVec2( x: math.cos(v), y: math.sin(v) )
proc gradient(orig, grad, p: TVec2): float {.noinit.} =
let sp = TVec2(x: p.x - orig.x, y: p.y - orig.y)
return grad.x * sp.x + grad.y * sp.y
proc get_gradient(ctx: var TNoise2DContext, x, y: int): TVec2 {.noinit.} =
let idx = ctx.permutations[x and 255] + ctx.permutations[y and 255];
return ctx.rgradients[idx and 255]
proc get_gradients(ctx: var TNoise2DContext, x, y: float) =
let
x0f = math.floor(x)
y0f = math.floor(y)
x0 = x0f.int
y0 = y0f.int
x1 = x0 + 1
y1 = y0 + 1
ctx.gradients[0] = get_gradient(ctx, x0, y0)
ctx.gradients[1] = get_gradient(ctx, x1, y0)
ctx.gradients[2] = get_gradient(ctx, x0, y1)
ctx.gradients[3] = get_gradient(ctx, x1, y1)
ctx.origins[0] = TVec2(x: x0f + 0.0, y: y0f + 0.0)
ctx.origins[1] = TVec2(x: x0f + 1.0, y: y0f + 0.0)
ctx.origins[2] = TVec2(x: x0f + 0.0, y: y0f + 1.0)
ctx.origins[3] = TVec2(x: x0f + 1.0, y: y0f + 1.0)
proc noise2d_get(ctx: var TNoise2DContext, x, y: float): float {.noinit.} =
let p = TVec2(x: x, y: y)
get_gradients(ctx, x, y)
let
v0 = gradient(ctx.origins[0], ctx.gradients[0], p)
v1 = gradient(ctx.origins[1], ctx.gradients[1], p)
v2 = gradient(ctx.origins[2], ctx.gradients[2], p)
v3 = gradient(ctx.origins[3], ctx.gradients[3], p)
fx = smooth(x - ctx.origins[0].x)
vx0 = lerp(v0, v1, fx)
vx1 = lerp(v2, v3, fx)
fy = smooth(y - ctx.origins[0].y)
return lerp(vx0, vx1, fy)
proc init_noise2d(ctx: var TNoise2DContext) =
for i in 0.. 255:
ctx.rgradients[i] = random_gradient()
for i in 0.. 255:
let j = rand() mod (i + 1)
ctx.permutations[i] = ctx.permutations[j]
ctx.permutations[j] = i
block main:
math.randomize()
const symbols = [ " ", "░", "▒", "▓", "█", "█" ]
var pixels: array[256 * 256, float]
var n2d = TNoise2DContext()
init_noise2d(n2d)
for i in 0.. 99:
for y in 0.. 255:
for x in 0.. 255:
let v = noise2d_get(n2d, x.float * 0.1, y.float * 0.1) * 0.5 + 0.5
pixels[y * 256 + x] = v
for y in 0.. 255:
for x in 0.. 255:
let idx = int(pixels[y * 256 + x] / 0.2)
stdout.write(symbols[idx])
stdout.write("\L")
--opt:speed
--stackTrace:off
--assertions:off
--threadAnalysis:off
--checks:off
--lineTrace:off
--debugger:off
@onionhammer
Copy link
Author

Rest of benchmark here:
https://github.com/nsf/pnoise

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment