RenderPass

설명

Object encapsulating the duration of a single renderpass that contains one or more subpasses.

The RenderPass object provides a new way to switch rendertargets in the context of a Scriptable Rendering Pipeline. As opposed to the SetRenderTargets function, the RenderPass object specifies a clear beginning and an end for the rendering, alongside explicit load/store actions on the rendering surfaces.

The RenderPass object also allows running multiple subpasses within the same renderpass, where the pixel shaders have a read access to the current pixel value within the renderpass. This allows for efficient implementation of various rendering methods on tile-based GPUs, such as deferred rendering.

RenderPasses are natively implemented on Metal (iOS) and Vulkan, but the API is fully functional on all rendering backends via emulation (using legacy SetRenderTargets calls and reading the current pixel values via texel fetches).

A quick example on how to use the RenderPass API within the Scriptable Render Pipeline to implement deferred rendering:

The RenderPass mechanism has the following limitations: - All attachments must have the same resolution and MSAA sample count - The rendering results of previous subpasses are only available within the same screen-space pixel coordinate via the UNITY_READ_FRAMEBUFFER_INPUT(x) macro in the shader; the attachments cannot be bound as textures or otherwise accessed until the renderpass has ended - iOS Metal does not allow reading from the Z-Buffer, so an additional render target is needed to work around that - The maximum amount of attachments allowed per RenderPass is currently 8 + depth, but note that various GPUs may have stricter limits.

using UnityEngine;
using UnityEngine.Experimental.Rendering;

public class DeferredRenderer
{
    private RenderPassAttachment m_Albedo;
    private RenderPassAttachment m_SpecRough;
    private RenderPassAttachment m_Normal;
    private RenderPassAttachment m_Emission;
    private RenderPassAttachment m_Depth;

    public DeferredRenderer()
    {
        // Create the attachment objects. If these attachments are not specifically bound to any RenderTexture using the BindSurface calls,
        // these are treated as temporary surfaces that are discarded at the end of the renderpass
        // NOTE: DO NOT ALLOCATE NEW RENDERPASSATTACHMENTS EVERY FRAME!
        // These objects only get garbage collected when a scene is unloaded, so you'll leak a lot of objects.
        m_Albedo = new RenderPassAttachment(RenderTextureFormat.ARGB32);
        m_SpecRough = new RenderPassAttachment(RenderTextureFormat.ARGB32);
        m_Normal = new RenderPassAttachment(RenderTextureFormat.ARGB2101010);
        m_Emission = new RenderPassAttachment(RenderTextureFormat.ARGBHalf);
        m_Depth = new RenderPassAttachment(RenderTextureFormat.Depth);

        // At the beginning of the render pass, clear the emission buffer to all black, and the depth buffer to 1.0f
        m_Emission.Clear(new Color(0.0f, 0.0f, 0.0f, 0.0f), 1.0f, 0);
        m_Depth.Clear(new Color(), 1.0f, 0);
    }

    public void ExecuteRenderLoop(Camera camera, CullResults cullResults, ScriptableRenderContext loop)
    {
        // Bind the albedo surface to the current camera target, so the final pass will render the scene to the screen backbuffer
        // The second argument specifies whether the existing contents of the surface need to be loaded as the initial values;
        // in our case we do not need that because we'll be clearing the attachment anyway. This saves a lot of memory
        // bandwidth on tiled GPUs.
        // The third argument specifies whether the rendering results need to be written out to memory at the end of
        // the renderpass. We need this as we'll be generating the final image there.
        // We could do this in the constructor already, but the camera target may change on the fly, esp. in the editor
        albedo.BindSurface(BuiltinRenderTextureType.CameraTarget, false, true);

        // All other attachments are transient surfaces that are not stored anywhere. If the renderer allows,
        // those surfaces do not even have a memory allocated for the pixel values, saving RAM usage.

        // Start the renderpass using the given scriptable rendercontext, resolution, samplecount, array of attachments that will be used within the renderpass and the depth surface
        using (RenderPass rp = new RenderPass(scriptableRenderContext, camera.pixelWidth, camera.pixelHeight, 1, new[] { albedo, specRough, normal, emission }, depth))
        {
            // Start the first subpass, GBuffer creation: render to albedo, specRough, normal and emission, no need to read any input attachments
            using (new RenderPass.SubPass(rp, new[] { albedo, specRough, normal, emission }, null))
            {
                // Render the deferred G-Buffer
                RenderGBuffer(cullResults, camera, loop);
            }
            // Second subpass, lighting: Render to the emission buffer, read from albedo, specRough, normal and depth.
            // The last parameter indicates whether the depth buffer can be bound as read-only.
            // Note that some renderers (notably iOS Metal) won't allow reading from the depth buffer while it's bound as Z-buffer,
            // so those renderers should write the Z into an additional FP32 render target manually in the pixel shader and read from it instead
            using (new RenderPass.SubPass(rp, new[] { emission }, new[] { albedo, specRough, normal, depth }, true))
            {
                PushGlobalShadowParams(loop);

                RenderLighting(camera, cullResults, loop);

                loop.DrawSkybox(camera);
            }
            // Third subpass, tonemapping: Render to albedo (which is bound to the camera target), read from emission.
            using (new RenderPass.SubPass(rp, new[] { albedo }, new[] { emission }, true))
            {
                // present frame buffer.
                FinalPass(loop);
            }
        }
    }
}

변수

생성자

Public 함수