本系列文章是對 http://metalkit.org 上面MetalKit內(nèi)容的全面翻譯和學(xué)習(xí).

MetalKit系統(tǒng)文章目錄

有一個讀者聯(lián)系我說我看到一個非常奇怪的現(xiàn)象:當(dāng)運行我們教程的代碼時,MTLLibrary會在幾百次繪制調(diào)用后返回nil.這讓我意識到,我沒有考慮到根據(jù)文檔Metal documentation有些Metal對象是暫時的而有些則不是.謝謝Mike的提醒!

要解決這個問題,我們需要再一次重構(gòu)代碼.這其實是件好事.我們需要把非瞬時的Metal對象(devices, queues, data buffers, textures, states和pipelines)從drawRect(_:)中拿出來,把它們放到視圖加載時只執(zhí)行一次的方法里.命令緩沖器和編碼器是僅有的兩個瞬時對象,設(shè)計出來供一次性使用的,所以我們可以在每次繪制調(diào)用時創(chuàng)建它們.
我們將繼續(xù)從本系列的第5部分 part 5 開始.讓我們創(chuàng)建一個新方法-一個初始化方法-它只在視圖加載時運行一次:

required init(coder: NSCoder) {
    super.init(coder: coder)
    device = MTLCreateSystemDefaultDevice()
    createBuffers()
    registerShaders()
}

下一步,刪除render()方法及在drawRect(_:)中的調(diào)用,因為我們不再需要它了.然后從sendToGPU()移動所有代碼到drawRect(_:)中,并刪除sendToGPU()因為這個也不需要了.這樣我們就從drawRect(_:)中移出了所有非瞬時的對象,只保留了command buffer命令緩沖和encoder編碼器在里面,只有它們是瞬時對象.

override func drawRect(dirtyRect: NSRect) {
    super.drawRect(dirtyRect)
    if let rpd = currentRenderPassDescriptor, drawable = currentDrawable {
        rpd.colorAttachments[0].clearColor = MTLClearColorMake(0.5, 0.5, 0.5, 1.0)
        let command_buffer = device!.newCommandQueue().commandBuffer()
        let command_encoder = command_buffer.renderCommandEncoderWithDescriptor(rpd)
        command_encoder.setRenderPipelineState(rps)
        command_encoder.setVertexBuffer(vertex_buffer, offset: 0, atIndex: 0)
        command_encoder.setVertexBuffer(uniform_buffer, offset: 0, atIndex: 1)
        command_encoder.drawPrimitives(.Triangle, vertexStart: 0, vertexCount: 3, instanceCount: 1)
        command_encoder.endEncoding()
        command_buffer.presentDrawable(drawable)
        command_buffer.commit()
    }
}

最后,我們創(chuàng)建一個新的類命名為MathUtils,并將兩個結(jié)構(gòu)體移動到里面,這樣我們就有了一個干凈的視圖類.

import simd

struct Vertex {
    var position: vector_float4
    var color: vector_float4
}

struct Matrix {
    var m: [Float]
    
    init() {
        m = [1, 0, 0, 0,
            0, 1, 0, 0,
            0, 0, 1, 0,
            0, 0, 0, 1
        ]
    }
    
    func translationMatrix(var matrix: Matrix, _ position: float3) -> Matrix {
        matrix.m[12] = position.x
        matrix.m[13] = position.y
        matrix.m[14] = position.z
        return matrix
    }
    
    func scalingMatrix(var matrix: Matrix, _ scale: Float) -> Matrix {
        matrix.m[0] = scale
        matrix.m[5] = scale
        matrix.m[10] = scale
        matrix.m[15] = 1.0
        return matrix
    }
    
    func rotationMatrix(var matrix: Matrix, _ rot: float3) -> Matrix {
        matrix.m[0] = cos(rot.y) * cos(rot.z)
        matrix.m[4] = cos(rot.z) * sin(rot.x) * sin(rot.y) - cos(rot.x) * sin(rot.z)
        matrix.m[8] = cos(rot.x) * cos(rot.z) * sin(rot.y) + sin(rot.x) * sin(rot.z)
        matrix.m[1] = cos(rot.y) * sin(rot.z)
        matrix.m[5] = cos(rot.x) * cos(rot.z) + sin(rot.x) * sin(rot.y) * sin(rot.z)
        matrix.m[9] = -cos(rot.z) * sin(rot.x) + cos(rot.x) * sin(rot.y) * sin(rot.z)
        matrix.m[2] = -sin(rot.y)
        matrix.m[6] = cos(rot.y) * sin(rot.x)
        matrix.m[10] = cos(rot.x) * cos(rot.y)
        matrix.m[15] = 1.0
        return matrix
    }
    
    func modelMatrix(var matrix: Matrix) -> Matrix {
        matrix = rotationMatrix(matrix, float3(0.0, 0.0, 0.1))
        matrix = scalingMatrix(matrix, 0.25)
        matrix = translationMatrix(matrix, float3(0.0, 0.5, 0.0))
        return matrix
    }
}

運行程序確保你仍能看到壯麗的三角形,就像我們上一部分看到的那樣.
源代碼source code 已發(fā)布在Github上.
下次見!