Presentation is loading. Please wait.

Presentation is loading. Please wait.

OIT and Indirect Illumination using DX11 Linked Lists Holger GruenAMD ISV Relations Nicolas ThibierozAMD ISV Relations.

Similar presentations


Presentation on theme: "OIT and Indirect Illumination using DX11 Linked Lists Holger GruenAMD ISV Relations Nicolas ThibierozAMD ISV Relations."— Presentation transcript:

1

2 OIT and Indirect Illumination using DX11 Linked Lists Holger GruenAMD ISV Relations Nicolas ThibierozAMD ISV Relations

3 Agenda Introduction Linked List Rendering Order Independent Transparency Indirect Illumination Q&A

4 Introduction Direct3D 11 HW opens the door to many new rendering algorithms In particular per pixel linked lists allow for a number of new techniques OIT, Indirect Shadows, Ray Tracing of dynamic scenes, REYES surface dicing, custom AA, Irregular Z-buffering, custom blending, Advanced Depth of Field, etc. This talk will walk you through: A DX11 implementation of per-pixel linked list and two effects that utilize this techique OIT Indirect Illumination

5 Per-Pixel Linked Lists with Direct3D 11 Element Link Element Link Element Link Element Link Nicolas Thibieroz European ISV Relations AMD

6 Why Linked Lists? Data structure useful for programming Very hard to implement efficiently with previous real-time graphics APIs DX11 allows efficient creation and parsing of linked lists Per-pixel linked lists A collection of linked lists enumerating all pixels belonging to the same screen position Element Link Element Link Element Link Element Link

7 Two-step process 1) Linked List Creation Store incoming fragments into linked lists 2) Rendering from Linked List Linked List traversal and processing of stored fragments

8 Creating Per-Pixel Linked Lists

9 PS5.0 and UAVs Uses a Pixel Shader 5.0 to store fragments into linked lists Not a Compute Shader 5.0! Uses atomic operations Two UAV buffers required - Fragment & Link buffer - Start Offset buffer UAV = Unordered Access View

10 Fragment & Link Buffer The Fragment & Link buffer contains data and link for all fragments to store Must be large enough to store all fragments Created with Counter support D3D11_BUFFER_UAV_FLAG_COUNTER flag in UAV view Declaration: struct FragmentAndLinkBuffer_STRUCT { FragmentData_STRUCT FragmentData;// Fragment data uint uNext;// Link to next fragment }; RWStructuredBuffer FLBuffer;

11 Start Offset Buffer The Start Offset buffer contains the offset of the last fragment written at every pixel location Screen-sized: (width * height * sizeof(UINT32) ) Initialized to magic value (e.g. -1) Magic value indicates no more fragments are stored (i.e. end of the list) Declaration: RWByteAddressBuffer StartOffsetBuffer;

12 Linked List Creation (1) No color Render Target bound! No rendering yet, just storing in L.L. Depth buffer bound if needed OIT will need it in a few slides UAVs bounds as input/output: StartOffsetBuffer (R/W) FragmentAndLinkBuffer (W)

13 Viewport Start Offset Buffer Fragment and Link Buffer Linked List Creation (2a) Fragment and Link Buffer Counter = 0

14 Start Offset Buffer Fragment and Link Buffer Linked List Creation (2b) Fragment and Link Buffer Fragment and Link BufferCounter = 0 1 Viewport

15 0 Start Offset Buffer Fragment and Link Buffer Linked List Creation (2c) Fragment and Link Buffer Fragment and Link BufferCounter = 1 23 Viewport

16 Start Offset Buffer 12 Linked List Creation (2d) Fragment and Link Buffer 0 Fragment and Link BufferCounter = 3 4 0 5 Viewport

17 Linked List Creation - Code float PS_StoreFragments(PS_INPUT input) : SV_Target { // Calculate fragment data (color, depth, etc.) FragmentData_STRUCT FragmentData = ComputeFragment(); // Retrieve current pixel count and increase counter uint uPixelCount = FLBuffer.IncrementCounter(); // Exchange offsets in StartOffsetBuffer uint vPos = uint(input.vPos); uint uStartOffsetAddress= 4 * ( (SCREEN_WIDTH*vPos.y) + vPos.x ); uint uOldStartOffset; StartOffsetBuffer.InterlockedExchange( uStartOffsetAddress, uPixelCount, uOldStartOffset); // Add new fragment entry in Fragment & Link Buffer FragmentAndLinkBuffer_STRUCT Element; Element.FragmentData = FragmentData; Element.uNext = uOldStartOffset; FLBuffer[uPixelCount] = Element; }

18 Traversing Per-Pixel Linked Lists

19 Rendering Pixels (1) Start Offset Buffer and Fragment & Link Buffer now bound as SRV Buffer StartOffsetBufferSRV; StructuredBuffer FLBufferSRV; Render a fullscreen quad For each pixel, parse the linked list and retrieve fragments for this screen position Process list of fragments as required Depends on algorithm e.g. sorting, finding maximum, etc. SRV = Shader Resource View

20 Render Target Start Offset Buffer Rendering from Linked List Fragment and Link Buffer 4 12 33 00

21 Rendering Pixels (2) float4 PS_RenderFragments(PS_INPUT input) : SV_Target { // Calculate UINT-aligned start offset buffer address uint vPos = uint(input.vPos); uint uStartOffsetAddress = SCREEN_WIDTH*vPos.y + vPos.x; // Fetch offset of first fragment for current pixel uint uOffset = StartOffsetBufferSRV.Load(uStartOffsetAddress); // Parse linked list for all fragments at this position float4 FinalColor=float4(0,0,0,0); while (uOffset!=0xFFFFFFFF)// 0xFFFFFFFF is magic value { // Retrieve pixel at current offset Element=FLBufferSRV[uOffset]; // Process pixel as required ProcessPixel(Element, FinalColor); // Retrieve next offset uOffset = Element.uNext; } return (FinalColor); }

22 Order-Independent Transparency via Per- Pixel Linked Lists Nicolas Thibieroz European ISV Relations AMD

23 Description Straight application of the linked list algorithm Stores transparent fragments into PPLL Rendering phase sorts pixels in a back-to-front order and blends them manually in a pixel shader Blend mode can be unique per-pixel! Special case for MSAA support

24 Linked List Structure Optimize performance by reducing amount of data to write to/read from UAV E.g. uint instead of float4 for color Example data structure for OIT: struct FragmentAndLinkBuffer_STRUCT { uint uPixelColor;// Packed pixel color uint uDepth;// Pixel depth uint uNext;// Address of next link }; May also get away with packed color and depth into the same uint! (if same alpha) 16 bits color (565) + 16 bits depth Performance/memory/quality trade-off

25 Visible Fragments Only! Use [earlydepthstencil] in front of Linked List creation pixel shader This ensures only transparent fragments that pass the depth test are stored i.e. Visible fragments! Allows performance savings and rendering correctness! [earlydepthstencil] float PS_StoreFragments(PS_INPUT input) : SV_Target {... }

26 Sorting Pixels Sorting in place requires R/W access to Linked List Sparse memory accesses = slow! Better way is to copy all pixels into array of temp registers Then do the sorting Temp array declaration means a hard limit on number of pixel per screen coordinates Required trade-off for performance

27 Sorting and Blending Blend fragments back to front in PS Blending algorithm up to app Example: SRCALPHA-INVSRCALPHA Or unique per pixel! (stored in fragment data) Background passed as input texture Actual HW blending mode disabled 0.87 0 0.98 0.95 12 0.93 34 0.950.930.870.98 Temp Array Background color PS color Render Target

28 Storing Pixels for Sorting (...) static uint2 SortedPixels[MAX_SORTED_PIXELS]; // Parse linked list for all pixels at this position // and store them into temp array for later sorting int nNumPixels=0; while (uOffset!=0xFFFFFFFF) { // Retrieve pixel at current offset Element=FLBufferSRV[uOffset]; // Copy pixel data into temp array SortedPixels[nNumPixels++]= uint2(Element.uPixelColor, Element.uDepth); // Retrieve next offset [flatten]uOffset = (nNumPixels>=MAX_SORTED_PIXELS) ? 0xFFFFFFFF : Element.uNext; } // Sort pixels in-place SortPixelsInPlace(SortedPixels, nNumPixels); (...)

29 Pixel Blending in PS (...) // Retrieve current color from background texture float4 vCurrentColor=BackgroundTexture.Load(int3(vPos.xy, 0)); // Rendering pixels using SRCALPHA-INVSRCALPHA blending for (int k=0; k { "@context": "http://schema.org", "@type": "ImageObject", "contentUrl": "http://images.slideplayer.com/5/1476904/slides/slide_29.jpg", "name": "Pixel Blending in PS (...) // Retrieve current color from background texture float4 vCurrentColor=BackgroundTexture.Load(int3(vPos.xy, 0)); // Rendering pixels using SRCALPHA-INVSRCALPHA blending for (int k=0; k

30 OIT via Per-Pixel Linked Lists with MSAA Support

31 Sample Coverage Storing individual samples into Linked Lists requires a huge amount of memory... and performance will suffer! Solution is to store transparent pixels into PPLL as before But including sample coverage too! Requires as many bits as MSAA mode Declare SV_COVERAGE in PS structure struct PS_INPUT { float3 vNormal : NORMAL; float2 vTex : TEXCOORD; float4 vPos : SV_POSITION; uint uCoverage : SV_COVERAGE; }

32 Linked List Structure Almost unchanged from previously Depth is now packed into 24 bits 8 Bits are used to store coverage struct FragmentAndLinkBuffer_STRUCT { uint uPixelColor;// Packed pixel color uint uDepthAndCoverage;// Depth + coverage uint uNext;// Address of next link };

33 Pixel Center Sample Sample Coverage Example Third sample is covered uCoverage = 0x04 (0100 in binary) Element.uDepthAndCoverage = ( In.vPos.z*(2^24-1) << 8 ) | In.uCoverage;

34 Rendering Samples (1) Rendering phase needs to be able to write individual samples Thus PS is run at sample frequency Can be done by declaring SV_SAMPLEINDEX in input structure Parse linked list and store pixels into temp array for later sorting Similar to non-MSAA case Difference is to only store sample if coverage matches sample index being rasterized

35 static uint2 SortedPixels[MAX_SORTED_PIXELS]; // Parse linked list for all pixels at this position // and store them into temp array for later sorting int nNumPixels=0; while (uOffset!=0xFFFFFFFF) { // Retrieve pixel at current offset Element=FLBufferSRV[uOffset]; // Retrieve pixel coverage from linked list element uint uCoverage=UnpackCoverage(Element.uDepthAndCoverage); if ( uCoverage & (1<=MAX_SORTED_PIXELS) ? 0xFFFFFFFF : Element.uNext; } Rendering Samples (2)

36 OIT Linked List Demo DEMO

37 Holger Gruen European ISV Relations AMD Direct3D 11 Indirect Illumination

38 Indirect Illumination Introduction 1 Real-time Indirect illumination is an active research topic Numerous approaches exist Reflective Shadow Maps (RSM) [ Dachsbacher/Stammiger05] Splatting Indirect Illumination [Dachsbacher/Stammiger2006] Multi-Res Splatting of Illumination [Wyman2009] Light propagation volumes [Kapalanyan2009] Approximating Dynamic Global Illumination in Image Space [Ritschel2009] Only a few support indirect shadows Imperfect Shadow Maps [Ritschel/Grosch2008] Micro-Rendering for Scalable, Parallel Final Gathering(SSDO) [Ritschel2010] Cascaded light propagation volumes for real-time indirect illumination [Kapalanyan/Dachsbacher2010 ] Most approaches somehow extend to multi- bounce lighting

39 Indirect Illumination Introduction 2 This section will cover An efficient and simple DX9-compliant RSM based implementation for smooth one bounce indirect illumination Indirect shadows are ignored here A Direct3D 11 technique that traces rays to compute indirect shadows Part of this technique could generally be used for ray-tracing dynamic scenes

40 Indirect Illumination w/o Indirect Shadows 1. Draw scene g-buffer 2. Draw Reflective Shadowmap (RSM) 1. RSM shows the part of the scene that recieves direct light from the light source 3. Draw Indirect Light buffer at ¼ res 1. RSM texels are used as light sources on g- buffer pixels for indirect lighting 4. Upsample Indirect Light (IL) 5. Draw final image adding IL

41 Step 1 G-Buffer needs to allow reconstruction of World/Camera space position World/Camera space normal Color/ Albedo DXGI_FORMAT_R32G32B32A32_FLOAT positions may be required for precise ray queries for indirect shadows

42 Step 2 RSM needs to allow reconstruction of World space position World space normal Color/ Albedo Only draw emitters of indirect light DXGI_FORMAT_R32G32B32A32_FLOAT position may be required for ray precise queries for indirect shadows

43 Step 3 Render a ¼ res IL as a deferred op Transform g-buffer pix to RSM space ->Light Space->project to RSM texel space Use a kernel of RSM texels as light sources RSM texels also called Virtual Point Light(VPL) Kernel size depends on Desired speed Desired look of the effect RSM resolution

44 Computing IL at a G-buf Pixel 1 Sum up contribution of all VPLs in the kernel

45 Computing IL at a G-buf Pixel 2 g-buffer pixel RSM texel/VPL This term is very similar to terms used in radiosity form factor computations

46 Computing IL at a G-buf Pixel 3 stx : sub RSM texel x position [0.0, 1.0[ sty : sub RSM texel y position [0.0, 1.0[ A naive solution for smooth IL needs to consider four VPL kernels with centers at t0, t1, t2 and t3.

47 Computing IL at a g-buf pixel 4 IndirectLight = (1.0f-sty) * ((1.0f-stx) * + stx * ) + (0.0f+sty) * ((1.0f-stx) * + stx * ) Evaluation of 4 big VPL kernels is slow stx : sub texel x position [0.0, 1.0[ sty : sub texel y position [0.0, 1.0[ VPL kernel at t0 VPL kernel at t1 VPL kernel at t2 VPL kernel at t3

48 Computing IL at a g-buf pixel 5 SmoothIndirectLight = (1.0f-sty)*(((1.0f-stx)*(B0+B3)+stx*(B2+B5))+B1)+ (0.0f+sty)*(((1.0f-stx)*(B6+B3)+stx*(B8+B5))+B7)+B4 stx : sub RSM texel x position of g-buf pix [0.0, 1.0[ sty : sub RSM texel y position of g-buf pix [0.0, 1.0[ This trick is probably known to some of you already. See backup for a detailed explanation !

49 Indirect Light Buffer

50 Step 4 Indirect Light buffer is ¼ res Perform a bilateral upsampling step See Peter-Pike Sloan, Naga K. Govindaraju, Derek Nowrouzezahrai, John Snyder. "Image-Based Proxy Accumulation for Real-Time Soft Global Illumination". Pacific Graphics 2007 Result is a full resolution IL

51 Step 5 Combine Direct Illumination Indirect Illumination Shadows (not mentioned )

52 Combined Image ~280 FPS on a HD5970 @ 1280x1024 for a 15x15 VPL kernel Deffered IL pass + bilateral upsampling costs ~2.5 ms DEMO

53 How to add Indirect Shadows 1. Use a CS and the linked lists technique Insert blockers of IL into 3D grid of lists – lets use triangles for now see backup for alternative data structure 2. Look at a kernel of VPLs again 3. Only accumulate light of VPLs that are occluded by blocker tris Trace rays through 3d grid to detect occluded VPLs Render low res buffer only 4. Subtract blocked indirect light from IL buffer Subtract a blurred version of low res version Blur is combined bilateral blurring/upsampling

54 Insert tris into 3D grid of triangle lists Scene Rasterize dynamic blockers to 3D grid using a CS and atomics

55 Insert tris into 3D grid of triangle lists World space 3D grid of triangle lists around IL blockers laid out in a UAV (0,1,0 ) eol = End of list (0xffffffff) Scene Rasterize dynamic blockers to 3D grid using a CS and atomics

56 3D Grid Demo

57 Indirect Light Buffer Emitter of green light Blocker of green light Expected indirect shadow

58 Blocked Indirect Light

59 Indirect Light Buffer

60 Subtracting Blocked IL

61 Final Image ~300 million rays per second for Indirect Shadows ~70 FPS on a HD5970 @ 1280x1024 DEMO Ray casting costs ~9 ms

62 Future directions Speed up IL rendering Render IL at even lower res Look into multi-res RSMs Speed up ray-tracing Per pixel array of lists for depth buckets Other data structures Raytrace other primitive types Splats, fuzzy ellipsoids etc. Proxy geometry or bounding volumes of blockers Get rid of Interlocked*() ops Just mark grid cells as occupied Lower quality but could work on earlier hardware Need to solve self-occlusion issues

63 Q&A Holger Gruen holger.gruen@AMD.comholger.gruen@AMD.com Nicolas Thibieroz nicolas.thibieroz@AMD.comnicolas.thibieroz@AMD.com Credits for the basic idea of how to implement PPLL under Direct3D 11 go to Jakub Klarowicz (Techland), Holger Gruen and Nicolas Thibieroz (AMD )


Download ppt "OIT and Indirect Illumination using DX11 Linked Lists Holger GruenAMD ISV Relations Nicolas ThibierozAMD ISV Relations."

Similar presentations


Ads by Google