Shadertoy shader port to Metal slow on iOS - ios

Question

I'm trying to learn Metal shaders so I ported this mountain generation shader I found on Shadertoy to Metal.
https://www.shadertoy.com/view/llsGW7
The port works, but is very slow on iOS. It is reasonably fast on OS X but gets slower when I increase the window size. It is also slow within an OS X playground.
I've done the tutorials on MetalKit.org and read the apple docs on the metal shading language but feel like I'm a bit conceptually lacking of how everything is working under the hood. If there is anything that jumps out in this code that is obviously slowing things down I would be very grateful to learn. I am not sure if the slowdown is due to the shader code itself, or the way it's all set up.
Here's the metal shader:
#include <metal_stdlib>
using namespace metal;
constexpr sampler textureSampler(coord::normalized,
address::repeat,
min_filter::linear,
mag_filter::linear,
mip_filter::linear );
kernel void compute(texture2d<float, access::write> output [[texture(0)]],
texture2d<float, access::sample> input [[texture(1)]],
constant float &timer [[buffer(0)]],
uint2 gid [[thread_position_in_grid]])
{
int width = input.get_width();
int height = input.get_height();
float2 uv = float2(gid) / float2(width, height);
float4 p = float4(uv,1,1)-0.5;
p.y = -p.y;
float4 d = p*0.5;
float4 t;
float4 c;
p.z += timer*200;
d.y-=0.2;
for(float i=1.7;i>0.0;i-=0.002) {
float s=0.5;
t = input.sample(textureSampler,0.3+p.xz*s/3e3) / (s+=s);
// this makes it purple
c = float4(1.0,-0.9,0.8,9.0)+d.x-t*i;
// c = float4(1.0,0.9,0.8,9.0)+d.x-t*i;
if (t.x > p.y*.01+1.3) {
break;
}
p += d;
}
output.write(c, gid);
}
and here is a subclass of MTKView I'm using to render the shader
import Cocoa
import MetalKit
class MetalView: MTKView {
var queue: MTLCommandQueue!
var cps: MTLComputePipelineState!
var timer: Float = 0
var timerBuffer: MTLBuffer!
var shaderName: String!
var texture: MTLTexture!
required public init(coder: NSCoder) {
super.init(coder: coder)
self.framebufferOnly = false
self.preferredFramesPerSecond = 60
registerShaders()
setupTexture()
}
func setupTexture() {
let path = Bundle.main.path(forResource: "texture", ofType: "jpg")
let textureLoader = MTKTextureLoader(device: device!)
texture = try! textureLoader.newTexture(withContentsOf: URL(fileURLWithPath: path!), options: nil)
}
func registerShaders() {
device = MTLCreateSystemDefaultDevice()!
queue = device!.makeCommandQueue()
do {
let library = device!.newDefaultLibrary()!
let kernel = library.makeFunction(name: "compute")!
cps = try device!.makeComputePipelineState(function: kernel)
} catch let e {
Swift.print("\(e)")
}
timerBuffer = device!.makeBuffer(length: MemoryLayout<Float>.size, options: [])
}
override public func draw(_ dirtyRect: CGRect) {
if let drawable = currentDrawable {
let commandBuffer = queue.makeCommandBuffer()
let commandEncoder = commandBuffer.makeComputeCommandEncoder()
commandEncoder.setComputePipelineState(cps)
commandEncoder.setTexture(drawable.texture, at: 0)
commandEncoder.setTexture(texture, at: 1)
commandEncoder.setBuffer(timerBuffer, offset: 0, at: 0)
update()
let threadGroupCount = MTLSizeMake(8, 8, 1)
let threadGroups = MTLSizeMake(drawable.texture.width / threadGroupCount.width, drawable.texture.height / threadGroupCount.height, 1)
commandEncoder.dispatchThreadgroups(threadGroups, threadsPerThreadgroup: threadGroupCount)
commandEncoder.endEncoding()
commandBuffer.present(drawable)
commandBuffer.commit()
}
}
func update() {
timer += Float(1.0 / TimeInterval(self.preferredFramesPerSecond))
let bufferPointer = timerBuffer.contents()
memcpy(bufferPointer, &timer, MemoryLayout<Float>.size)
}
}