I'm trying to render a pixel aligned 2D quad using Metal, but can't seem to get it right.
My vertex buffer's contents (as logged from the CPU side, on creation) are:
Vertex(position: float4(0.0, 0.0, 0.5, 1.0), textureCoordinate: float2(0.0, 0.0))
Vertex(position: float4(0.0, 64.0, 0.5, 1.0), textureCoordinate: float2(0.0, 1.0))
Vertex(position: float4(64.0, 0.0, 0.5, 1.0), textureCoordinate: float2(1.0, 0.0))
Vertex(position: float4(64.0, 64.0, 0.5, 1.0), textureCoordinate: float2(1.0, 1.0))
The index buffer to draw two triangles contains the following indices:
0, 1, 2, 2, 1, 3
The texture I'm using is:
...but I get something like this:
When capturing the frame and inspecting the vertex buffer, I get this:
Clearly, position and texture coordinates are mixed up.
This is the code I use to create the geometry:
import Metal
import simd
struct Vertex {
var position = float4(x: 0, y: 0, z: 0, w: 1)
var textureCoordinate = float2(x: 0, y: 0)
}
class Quad {
let vertexBuffer: MTLBuffer
let indexBuffer: MTLBuffer
let indexCount: Int
let indexType: MTLIndexType
let primitiveType: MTLPrimitiveType
init(sideLength: Float, device: MTLDevice) {
self.primitiveType = .triangle
var vertexData = [Vertex]()
var topLeft = Vertex()
topLeft.position.x = 0
topLeft.position.y = 0
topLeft.position.z = 0.5
topLeft.textureCoordinate.x = 0
topLeft.textureCoordinate.y = 0
vertexData.append(topLeft)
var bottomLeft = Vertex()
bottomLeft.position.x = 0
bottomLeft.position.y = sideLength
bottomLeft.position.z = 0.5
bottomLeft.textureCoordinate.x = 0
bottomLeft.textureCoordinate.y = 1
vertexData.append(bottomLeft)
var topRight = Vertex()
topRight.position.x = sideLength
topRight.position.y = 0
topRight.position.z = 0.5
topRight.textureCoordinate.x = 1
topRight.textureCoordinate.y = 0
vertexData.append(topRight)
var bottomRight = Vertex()
bottomRight.position.x = sideLength
bottomRight.position.y = sideLength
bottomRight.position.z = 0.5
bottomRight.textureCoordinate.x = 1
bottomRight.textureCoordinate.y = 1
vertexData.append(bottomRight)
for vertex in vertexData {
Swift.print(vertex) // logs the structs posted above
}
let vertexBufferSize = vertexData.count * MemoryLayout<Vertex>.stride
self.vertexBuffer = device.makeBuffer(bytes: vertexData, length: vertexBufferSize, options: [])
var indexData = [UInt32]()
// First triangle: Top left, bottom left, top right (CCW)
indexData.append(0)
indexData.append(1)
indexData.append(2)
// Second triangle: top right, bottom left, bottom right (CCW)
indexData.append(2)
indexData.append(1)
indexData.append(3)
for index in indexData {
Swift.print(index) // logs the integers posted before
}
self.indexType = .uint32
self.indexCount = indexData.count
let indexBufferSize = indexData.count * MemoryLayout<UInt32>.stride
self.indexBuffer = device.makeBuffer(bytes: indexData, length: indexBufferSize, options: [])
}
}
...the Shaders:
#include <metal_stdlib>
using namespace metal;
struct Constants {
float4x4 modelViewProjectionMatrix;
float4 tintColor;
};
struct VertexIn {
packed_float4 position [[ attribute(0) ]];
packed_float2 texCoords [[ attribute(1) ]];
};
struct VertexOut {
float4 position [[position]];
float2 texCoords;
};
vertex VertexOut sprite_vertex_transform(device VertexIn *vertices [[buffer(0)]],
constant Constants &uniforms [[buffer(1)]],
uint vertexId [[vertex_id]]) {
float4 modelPosition = vertices[vertexId].position;
VertexOut out;
out.position = uniforms.modelViewProjectionMatrix * modelPosition;
out.texCoords = vertices[vertexId].texCoords;
return out;
}
fragment half4 sprite_fragment_textured(
VertexOut fragmentIn [[stage_in]],
texture2d<float, access::sample> tex2d [[texture(0)]],
sampler sampler2d [[sampler(0)]]){
half4 surfaceColor = half4(tex2d.sample(sampler2d, fragmentIn.texCoords).rgba);
return surfaceColor;
}
...and this is the code to render it (indexed):
renderEncoder.setVertexBuffer(quad.vertexBuffer, offset: 0, at: 0)
renderEncoder.setVertexBytes(&constants, length: MemoryLayout<Constants>.stride, at: 1)
renderEncoder.setFragmentTexture(texture, at: 0)
renderEncoder.setFragmentSamplerState(sampler, at: 0)
// quad is an instance of the class above, with sideLength == 64
renderEncoder.drawIndexedPrimitives(
type: quad.primitiveType,
indexCount: quad.indexCount,
indexType: quad.indexType,
indexBuffer: quad.indexBuffer,
indexBufferOffset: 0)
Clearly, the data isn't being copied to the vertex buffer appropriately and I'm getting the strides wrong somewhere, but I can't quite figure out where.
The whole project is on GitHub.
It seems this part was causing the trouble:
struct VertexIn {
packed_float4 position [[ attribute(0) ]];
packed_float2 texCoords [[ attribute(1) ]];
};
From the debugger console, on the CPU side:
(lldb) print MemoryLayout<Vertex>.stride
(Int) $R0 = 32
(lldb) print MemoryLayout<Vertex>.size
(Int) $R1 = 24
(lldb)
Removing the packed_
prefix solved it:
struct VertexIn {
float4 position [[ attribute(0) ]];
float2 texCoords [[ attribute(1) ]];
};
But I guess now my buffer is not optimal in terms of space usage. I'll have to work out how to get the packed_
version to work...
I really need a refresher on struct alignment and packing...
Note: Originally, I adapted the code from a sample project by Apple where the vertex struct had position
and normal
, both of type float4
. Cleary changing the struct "net" size from 8*sizeof(float)
to 6*sizeof(float)
is what caused the problem.