unity-tech-cn
/
ScriptableRenderPipeline
29
0
派生 4
代码 合并请求 版本发布
浏览代码

Added async compute to fptl light list build. Removed shader that was not in use and causing compile errors on ps4

/main
rthompson-unity 7 年前
当前提交
7e1b650c
共有 3 个文件被更改，包括 36 次插入 和 258 次删除
  1. 138
      TestbedPipelines/Fptl/FptlLighting.cs
  2. 10
      TestbedPipelines/Fptl/Internal-DeferredComputeShading.compute.meta
  3. 146
      TestbedPipelines/Fptl/Internal-DeferredComputeShading.compute

138
TestbedPipelines/Fptl/FptlLighting.cs
查看文件


public bool enableDrawLightBoundsDebug = false;
public bool enableDrawTileDebug = false;
public bool enableReflectionProbeDebug = false;
public bool enableComputeLightEvaluation = false;
const bool k_UseAsyncCompute = true; // should not use on mobile
const bool k_UseAsyncCompute = true;// should not use on mobile
const int k_Log2NumClusters = 6; // accepted range is from 0 to 6. NumClusters is 1<<g_iLog2NumClusters
const float k_ClustLogBase = 1.02f; // each slice 2% bigger than the previous

//cmd.SetRenderTarget(new RenderTargetIdentifier(kGBufferEmission), new RenderTargetIdentifier(kGBufferZ));
//cmd.Blit (kGBufferNormal, (RenderTexture)null); // debug: display normals
if (enableComputeLightEvaluation) //TODO: temporary workaround for "All kernels must use same constant buffer layouts"
{
var w = camera.pixelWidth;
var h = camera.pixelHeight;
var numTilesX = (w + 7) / 8;
var numTilesY = (h + 7) / 8;
string kernelName = "ShadeDeferred" + (bUseClusteredForDeferred ? "_Clustered" : "_Fptl") + (enableDrawTileDebug ? "_Debug" : "");
int kernel = deferredComputeShader.FindKernel(kernelName);
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_CameraDepthTexture", new RenderTargetIdentifier(s_CameraDepthTexture));
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_CameraGBufferTexture0", new RenderTargetIdentifier(s_GBufferAlbedo));
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_CameraGBufferTexture1", new RenderTargetIdentifier(s_GBufferSpecRough));
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_CameraGBufferTexture2", new RenderTargetIdentifier(s_GBufferNormal));
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_CameraGBufferTexture3", new RenderTargetIdentifier(s_GBufferEmission));
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_spotCookieTextures", m_CookieTexArray.GetTexCache());
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_pointCookieTextures", m_CubeCookieTexArray.GetTexCache());
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_reflCubeTextures", m_CubeReflTexArray.GetTexCache());
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_reflRootCubeTexture", ReflectionProbe.defaultTexture);
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "g_tShadowBuffer", new RenderTargetIdentifier(m_shadowBufferID));
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "unity_NHxRoughness", m_NHxRoughnessTexture);
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "_LightTextureB0", m_LightAttentuationTexture);
cmd.SetComputeBufferParam(deferredComputeShader, kernel, "g_vLightListGlobal", bUseClusteredForDeferred ? s_PerVoxelLightLists : s_LightList);
cmd.SetComputeBufferParam(deferredComputeShader, kernel, "g_vLightData", s_LightDataBuffer);
cmd.SetComputeBufferParam(deferredComputeShader, kernel, "g_dirLightData", s_DirLightList);
var defdecode = ReflectionProbe.defaultTextureHDRDecodeValues;
cmd.SetComputeFloatParam(deferredComputeShader, "_reflRootHdrDecodeMult", defdecode.x);
cmd.SetComputeFloatParam(deferredComputeShader, "_reflRootHdrDecodeExp", defdecode.y);
cmd.SetComputeFloatParam(deferredComputeShader, "g_fClustScale", m_ClustScale);
cmd.SetComputeFloatParam(deferredComputeShader, "g_fClustBase", k_ClustLogBase);
cmd.SetComputeFloatParam(deferredComputeShader, "g_fNearPlane", camera.nearClipPlane);
cmd.SetComputeFloatParam(deferredComputeShader, "g_fFarPlane", camera.farClipPlane);
cmd.SetComputeIntParam(deferredComputeShader, "g_iLog2NumClusters", k_Log2NumClusters);
cmd.SetComputeIntParam(deferredComputeShader, "g_isLogBaseBufferEnabled", k_UseDepthBuffer ? 1 : 0);
cmd.SetComputeIntParam(deferredComputeShader, "g_isOpaquesOnlyEnabled", 0);
//
var proj = camera.projectionMatrix;
var temp = new Matrix4x4();
temp.SetRow(0, new Vector4(1.0f, 0.0f, 0.0f, 0.0f));
temp.SetRow(1, new Vector4(0.0f, 1.0f, 0.0f, 0.0f));
temp.SetRow(2, new Vector4(0.0f, 0.0f, 0.5f, 0.5f));
temp.SetRow(3, new Vector4(0.0f, 0.0f, 0.0f, 1.0f));
var projh = temp * proj;
var invProjh = projh.inverse;
temp.SetRow(0, new Vector4(0.5f * w, 0.0f, 0.0f, 0.5f * w));
temp.SetRow(1, new Vector4(0.0f, 0.5f * h, 0.0f, 0.5f * h));
temp.SetRow(2, new Vector4(0.0f, 0.0f, 0.5f, 0.5f));
temp.SetRow(3, new Vector4(0.0f, 0.0f, 0.0f, 1.0f));
var projscr = temp * proj;
var invProjscr = projscr.inverse;
cmd.SetComputeIntParam(deferredComputeShader, "g_iNrVisibLights", numLights);
cmd.SetComputeMatrixParam(deferredComputeShader, "g_mScrProjection", projscr);
cmd.SetComputeMatrixParam(deferredComputeShader, "g_mInvScrProjection", invProjscr);
cmd.SetComputeMatrixParam(deferredComputeShader, "g_mViewToWorld", camera.cameraToWorldMatrix);
if (bUseClusteredForDeferred)
{
cmd.SetComputeBufferParam(deferredComputeShader, kernel, "g_vLayeredOffsetsBuffer", s_PerVoxelOffset);
if (k_UseDepthBuffer)
{
cmd.SetComputeBufferParam(deferredComputeShader, kernel, "g_logBaseBuffer", s_PerTileLogBaseTweak);
}
}
cmd.SetComputeIntParam(deferredComputeShader, "g_widthRT", w);
cmd.SetComputeIntParam(deferredComputeShader, "g_heightRT", h);
cmd.SetComputeIntParam(deferredComputeShader, "g_nNumDirLights", numDirLights);
cmd.SetComputeBufferParam(deferredComputeShader, kernel, "g_dirLightData", s_DirLightList);
cmd.SetComputeTextureParam(deferredComputeShader, kernel, "uavOutput", new RenderTargetIdentifier(s_CameraTarget));
cmd.SetComputeMatrixArrayParam(deferredComputeShader, "g_matWorldToShadow", m_MatWorldToShadow);
cmd.SetComputeVectorArrayParam(deferredComputeShader, "g_vDirShadowSplitSpheres", m_DirShadowSplitSpheres);
cmd.SetComputeVectorParam(deferredComputeShader, "g_vShadow3x3PCFTerms0", m_Shadow3X3PCFTerms[0]);
cmd.SetComputeVectorParam(deferredComputeShader, "g_vShadow3x3PCFTerms1", m_Shadow3X3PCFTerms[1]);
cmd.SetComputeVectorParam(deferredComputeShader, "g_vShadow3x3PCFTerms2", m_Shadow3X3PCFTerms[2]);
cmd.SetComputeVectorParam(deferredComputeShader, "g_vShadow3x3PCFTerms3", m_Shadow3X3PCFTerms[3]);
cmd.DispatchCompute(deferredComputeShader, kernel, numTilesX, numTilesY, 1);
}
else
{
cmd.Blit(BuiltinRenderTextureType.CameraTarget, s_CameraTarget, m_DeferredMaterial, 0);
cmd.Blit(BuiltinRenderTextureType.CameraTarget, s_CameraTarget, m_DeferredReflectionMaterial, 0);
}
cmd.Blit(BuiltinRenderTextureType.CameraTarget, s_CameraTarget, m_DeferredMaterial, 0);
cmd.Blit(BuiltinRenderTextureType.CameraTarget, s_CameraTarget, m_DeferredReflectionMaterial, 0);
// Set the intermediate target for compositing (skybox, etc)
cmd.SetRenderTarget(new RenderTargetIdentifier(s_CameraTarget), new RenderTargetIdentifier(s_CameraDepthTexture));

for (int i = 0; i < lcnt; ++i)
{
VisibleLight vl = inputs.visibleLights[i];
if (vl.light.shadows != LightShadows.None && vl.light.GetComponent<AdditionalShadowData>().shadowDimmer > 0.0f)
AdditionalShadowData asd = vl.light.GetComponent<AdditionalShadowData>();
if (vl.light.shadows != LightShadows.None && asd != null && asd.shadowDimmer > 0.0f)
m_ShadowRequests.Add(i);
}
// pass this list to a routine that assigns shadows based on some heuristic

temp.SetRow(3, new Vector4(0.0f, 0.0f, 0.0f, 1.0f));
var projscr = temp * proj;
var invProjscr = projscr.inverse;
BuildPerTileLightLists(camera, loop, numLights, projscr, invProjscr);
CommandBuffer cmdPreShadows = CommandBufferPool.Get();
GPUFence preShadowsFence = cmdPreShadows.CreateGPUFence();
loop.ExecuteCommandBuffer(cmdPreShadows);
CommandBufferPool.Release(cmdPreShadows);
GPUFence postLightListFence = BuildPerTileLightLists(camera, loop, numLights, projscr, invProjscr, preShadowsFence);
//BuildPerTileLightLists(camera, loop, numLights, projscr, invProjscr);
CommandBuffer cmdShadow = CommandBufferPool.Get();
m_ShadowMgr.RenderShadows( m_FrameId, loop, cmdShadow, cullResults, cullResults.visibleLights );

// Push all global params
var numDirLights = UpdateDirectionalLights(camera, cullResults.visibleLights, m_ShadowIndices);
PushGlobalParams(camera, loop, CameraToWorld(camera), projscr, invProjscr, numDirLights);
PushGlobalParams(camera, loop, CameraToWorld(camera), projscr, invProjscr, numDirLights, postLightListFence);
// do deferred lighting
DoTiledDeferredLighting(camera, loop, numLights, numDirLights);

cmd.DispatchCompute(buildPerVoxelLightListShader, s_GenListPerVoxelKernel, nrTilesClustX, nrTilesClustY, 1);
}
void BuildPerTileLightLists(Camera camera, ScriptableRenderContext loop, int numLights, Matrix4x4 projscr, Matrix4x4 invProjscr)
GPUFence BuildPerTileLightLists(Camera camera, ScriptableRenderContext loop, int numLights, Matrix4x4 projscr, Matrix4x4 invProjscr, GPUFence startFence)
{
var w = camera.pixelWidth;
var h = camera.pixelHeight;

var numBigTilesY = (h + 63) / 64;
var cmd = CommandBufferPool.Get("Build light list" );
cmd.WaitOnGPUFence(startFence);
bool isOrthographic = camera.orthographic;

VoxelLightListGeneration(cmd, camera, numLights, projscr, invProjscr);
}
loop.ExecuteCommandBuffer(cmd);
GPUFence completeFence = cmd.CreateGPUFence();
if (k_UseAsyncCompute)
{
loop.ExecuteCommandBufferAsync(cmd, ComputeQueueType.Default);
}
else
{
loop.ExecuteCommandBuffer(cmd);
}
return completeFence;
void PushGlobalParams(Camera camera, ScriptableRenderContext loop, Matrix4x4 viewToWorld, Matrix4x4 scrProj, Matrix4x4 incScrProj, int numDirLights)
void PushGlobalParams(Camera camera, ScriptableRenderContext loop, Matrix4x4 viewToWorld, Matrix4x4 scrProj, Matrix4x4 incScrProj, int numDirLights, GPUFence startFence)
cmd.WaitOnGPUFence(startFence);
bool isOrthographic = camera.orthographic;
cmd.SetGlobalFloat("g_isOrthographic", (float) (isOrthographic ? 1 : 0));

10
TestbedPipelines/Fptl/Internal-DeferredComputeShading.compute.meta
查看文件


fileFormatVersion: 2
guid: 6994e4f8a80d87b428cca7242cb32a0e
timeCreated: 1503477807
licenseType: Pro
ComputeShaderImporter:
externalObjects: {}
currentAPIMask: 4
userData:
assetBundleName:
assetBundleVariant:

146
TestbedPipelines/Fptl/Internal-DeferredComputeShading.compute
查看文件


#pragma kernel ShadeDeferred_Fptl SHADE_DEFERRED_ENTRY=ShadeDeferred_Fptl USE_FPTL_LIGHTLIST=1 ENABLE_DEBUG=0
#pragma kernel ShadeDeferred_Clustered SHADE_DEFERRED_ENTRY=ShadeDeferred_Clustered USE_CLUSTERED_LIGHTLIST=1 ENABLE_DEBUG=0 //TODO: disabled clustered permutations so far as it leads to the error "All kernels must use same constant buffer layouts"
#pragma kernel ShadeDeferred_Fptl_Debug SHADE_DEFERRED_ENTRY=ShadeDeferred_Fptl_Debug USE_FPTL_LIGHTLIST=1 ENABLE_DEBUG=1
#pragma kernel ShadeDeferred_Clustered_Debug SHADE_DEFERRED_ENTRY=ShadeDeferred_Clustered_Debug USE_CLUSTERED_LIGHTLIST=1 ENABLE_DEBUG=1
#define TILE_SIZE 8
// Hacks to get the header to compile in compute
#define SHADER_TARGET 50
#define UNITY_PBS_USE_BRDF1
#define fixed4 float4
#include "UnityLightingCommon.cginc"
#undef fixed4
float3 EvalMaterial(UnityLight light, UnityIndirect ind);
float3 EvalIndirectSpecular(UnityLight light, UnityIndirect ind);
// uses the optimized single layered light list for opaques only
#ifdef USE_FPTL_LIGHTLIST
#define OPAQUES_ONLY
#endif
#include "TiledLightingTemplate.hlsl"
#include "TiledReflectionTemplate.hlsl"
UNITY_DECLARE_TEX2D_FLOAT(_CameraDepthTexture);
Texture2D _CameraGBufferTexture0;
Texture2D _CameraGBufferTexture1;
Texture2D _CameraGBufferTexture2;
Texture2D _CameraGBufferTexture3;
RWTexture2D<float4> uavOutput : register(u0);
struct v2f {
float4 vertex : SV_POSITION;
float2 texcoord : TEXCOORD0;
};
v2f vert(float4 vertex : POSITION, float2 texcoord : TEXCOORD0)
{
v2f o;
o.vertex = UnityObjectToClipPos(vertex);
o.texcoord = texcoord.xy;
return o;
}
struct StandardData
{
float3 specularColor;
float3 diffuseColor;
float3 normalWorld;
float smoothness;
float occlusion;
float3 emission;
};
struct LocalDataBRDF
{
StandardData gbuf;
// extras
float oneMinusReflectivity;
float3 Vworld;
};
static LocalDataBRDF g_localParams;
StandardData UnityStandardDataFromGbufferAux(float4 gbuffer0, float4 gbuffer1, float4 gbuffer2, float4 gbuffer3)
{
StandardData data;
data.normalWorld = normalize(2 * gbuffer2.xyz - 1);
data.smoothness = gbuffer1.a;
data.diffuseColor = gbuffer0.xyz; data.specularColor = gbuffer1.xyz;
data.occlusion = gbuffer0.a;
data.emission = gbuffer3.xyz;
return data;
}
half3 BRDF3_Direct2(half3 diffColor, half3 specColor, half rlPow4, half smoothness)
{
half LUT_RANGE = 16.0; // must match range in NHxRoughness() function in GeneratedTextures.cpp
// Lookup texture to save instructions
half specular = tex2Dlod(unity_NHxRoughness, half4(rlPow4, SmoothnessToPerceptualRoughness(smoothness), 0, 0)).UNITY_ATTEN_CHANNEL * LUT_RANGE;
#if defined(_SPECULARHIGHLIGHTS_OFF)
specular = 0.0;
#endif
return diffColor + specular * specColor;
}
float3 EvalMaterial(UnityLight light, UnityIndirect ind)
{
StandardData data = g_localParams.gbuf;
return UNITY_BRDF_PBS(data.diffuseColor, data.specularColor, g_localParams.oneMinusReflectivity, data.smoothness, data.normalWorld, g_localParams.Vworld, light, ind);
}
float3 EvalIndirectSpecular(UnityLight light, UnityIndirect ind)
{
StandardData data = g_localParams.gbuf;
return data.occlusion * UNITY_BRDF_PBS(0, data.specularColor, g_localParams.oneMinusReflectivity, data.smoothness, data.normalWorld, g_localParams.Vworld, light, ind).rgb;
}
[numthreads(TILE_SIZE, TILE_SIZE, 1)]
void SHADE_DEFERRED_ENTRY(uint2 dispatchThreadId : SV_DispatchThreadID, uint2 groupId : SV_GroupID)
{
uint2 pixCoord = dispatchThreadId;
float zbufDpth = FetchDepth(_CameraDepthTexture, pixCoord.xy).x;
float linDepth = GetLinearDepth(zbufDpth);
float3 vP = GetViewPosFromLinDepth(pixCoord, linDepth);
float3 vPw = mul(g_mViewToWorld, float4(vP, 1)).xyz;
float3 Vworld = normalize(mul((float3x3) g_mViewToWorld, -vP).xyz); //unity_CameraToWorld
float4 gbuffer0 = _CameraGBufferTexture0.Load(uint3(pixCoord.xy, 0));
float4 gbuffer1 = _CameraGBufferTexture1.Load(uint3(pixCoord.xy, 0));
float4 gbuffer2 = _CameraGBufferTexture2.Load(uint3(pixCoord.xy, 0));
float4 gbuffer3 = _CameraGBufferTexture3.Load(uint3(pixCoord.xy, 0));
StandardData data = UnityStandardDataFromGbufferAux(gbuffer0, gbuffer1, gbuffer2, gbuffer3);
g_localParams.gbuf = data;
g_localParams.oneMinusReflectivity = 1.0 - SpecularStrength(data.specularColor.rgb);
g_localParams.Vworld = Vworld;
uint2 tileCoord = groupId >> 1;
uint numLightsProcessed = 0;
float3 c = data.emission + ExecuteLightList(numLightsProcessed, tileCoord, vP, vPw, Vworld);
uint numReflectionsProcessed = 0;
c += ExecuteReflectionList(numReflectionsProcessed, tileCoord, vP, data.normalWorld, Vworld, data.smoothness);
#if ENABLE_DEBUG
c = OverlayHeatMap(pixCoord & 15, numLightsProcessed, c);
#endif
uavOutput[pixCoord] = float4(c, 1.0);
}
撰写 预览
正在加载...
取消
保存