GTC 2012: NVIDIA OpenGL in 2012

  • 3,732 views
Uploaded on

Presented at the GPU Technology Conference 2012 in San Jose, California. …

Presented at the GPU Technology Conference 2012 in San Jose, California.

Monday, May 14, 2012.

Attend this session to get the most out of OpenGL on NVIDIA Quadro and GeForce GPUs. Topics covered include the latest advances available for Cg 3.1, the OpenGL Shading Language (GLSL); programmable tessellation; improved support for Direct3D conventions; integration with Direct3D and CUDA resources; bindless graphics; and more. When you utilize the latest OpenGL innovations from NVIDIA in your graphics applications, you benefit from NVIDIA's leadership driving OpenGL as a cross-platform, open industry standard.

  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
No Downloads

Views

Total Views
3,732
On Slideshare
0
From Embeds
0
Number of Embeds
1

Actions

Shares
Downloads
141
Comments
0
Likes
2

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide
  • One way to view the OpenGL offerings from NVIDIA is as a tool to enable awesome visual applications to be developed through a comprehensive stack of software solutions. We offer: Cg: A shading language and effects system to develop rendering effects and techniques, and deploy them on various platforms and APIs, including OpenGL. Scenix: OpenGL based professional scene graph used in many markets today. Automotive styling, visualization, simulation, broadcast graphics, and more. Applications can quickly add features such as stereo, SDI, 30-bit color, scene distribution and interactive ray tracing Optix: Is an interactive ray-tracing engine built on top of CUDA and OpenGL. Hybrid rendering, mixing of traditional graphics rendering and ray tracing, are also enabled. OptiX integrates with SceniX. With Optix, you accelerate an existing renderer, or build a new one yourself. Complex: Maintains interactivity for large scenes as they exceed the limits of a single GPU, allowing massive data sets to be explored, using multiple GPUs in a system. The CompleX engine can be adopted by any product using OpenGL, and can be enabled immediately by an application using SceniX Parallel Nsight is an advanced debugger and analyzer fully integrated with Visual Studio. It enables you to better see what is going on with your OpenGL, CUDA, Direct3D, DirectCompute or OpenCL application.
  • This is a close-up taken out of the Fermi Rocket Sled demo. This shows how hybrid rendering works. With traditional OpenGL rendering you generate your image, for example using SceniX, and then enhance it by using OptiX to add reflection and shadows. One of our experts on this, Tristan, is available if you have more questions on how you can do this in your application. Tristan, raise your hand please.
  • OpenGL is the only Cross Platform 3D API. Every major Operating System provides a version or flavor of OpenGL. Windows, Mac OS, iOS, Linux and Android.
  • After Mark’s deep-dive, I’ll pull you back up into the higher level view of where OpenGL fits in as the 3D graphics API.
  • OpenGL is a desktop API, as you’ve seen by now. OpenGL ES is the sister API for mobile and embedded devices. By keeping both APIs closely aligned, content can flow from there up into OpenGL enabled platforms and back down to OpenGL ES enabled platforms. WebGL was announced last year. It provides JavaScript bindings to OpenGL ES and provides plug-in less 3D graphics in a web browser. WebGL gives you access to the GPU in the system inside of a browser. Beta implementations of WebGL are already available from Mozilla, Google and Opera. For NVIDIA this means we will support and enhance OpenGL and WebGL on GeForce and OpenGL ES and WebGL on Tegra for the mobile market.
  • Finally, Siggraph is where we introduce 3D Vision Pro as well. Again, come to the booth for more information. Now on to OpenGL!
  • OpenGL Interoperability means the ability to share data among different APIs. You can, for example, map a D3D surface as a texture or renderbuffer in OpenGL directly. Or map an OpenGL buffer object as a CUDA memory object. It is also possible to transfer data between two OpenGL contexts in a multi-GPU situation with maximum efficiency.
  • Here is an example of interop with CUDA. The application makes OpenGL and CUDA calls. For example, process a video stream in CUDA, then transfer it over to OpenGL using interop for more processing and display. This works on a single card as well as multi-card systems. Performance between Quadro cards will be higher than between GeForce cards.
  • This slide is almost the same as the previous one, except it shows interop between two OpenGL contexts, each talking to a separate graphics card. The application is using GPU affinity to direct rendering to a particular GPU. Again, you need interop to transfer the data from one card to the other. For example, to composite the final frame of a rendering split over many GPUs. Note that both GPU affinity and fast GL to GL interop with NV_copy_image are Quadro only solutions.
  • NV_path_rendering provides a new third pipeline—in addition to the vertex and pixel pipelines—for rendering pixels
  • Before Kepler, in order to make a texture available for the GPU to reference it had to be assigned a “slot” in a fixed-size binding table. The number of slots in that table ultimately limits how many unique textures a shader can read from at run time. On Kepler, no such additional setup is necessary - shader can reference textures in memory directly and there is no need to go through the binding tables anymore. This effectively eliminates any limits on the number of unique textures it can use to render a scene.
  • Bindless textures reduce CPU work and provide more efficient access for the GPU
  • In advanced rendering apps such as raytracing it is impossible to know in advance which textures a given ray may hit – thus it is impossible to pre-“bind” them. Bindless model solves this problem by allowing shader reference textures directly

Transcript

  • 1. NVIDIA OpenGL in 2012
  • 2. Talk S0023 - NVIDIA OpenGL for 2012 Mark Kilgard (Principal Software Engineer, NVIDIA) Attend this session to get the most out of OpenGL on NVIDIA Quadro and GeForce GPUs. Topics covered include the latest advances available for Cg 3.1, the OpenGL Shading Language (GLSL); programmable tessellation; improved support for Direct3D conventions; integration with Direct3D and CUDA resources; bindless graphics; and more. When you utilize the latest OpenGL innovations from NVIDIA in your graphics applications, you benefit from NVIDIAs leadership driving OpenGL as a cross-platform, open industry standard. Topic Areas: Computer Graphics; Development Tools & Libraries; Visualization; Audio, Image and Video Processing Level: Intermediate Day: Monday, 05/14 Time: 9:00 am - 10:20 pm Location: Room A3
  • 3. Mark Kilgard• Principal System Software Engineer – OpenGL driver and API evolution – Cg (“C for graphics”) shading language – GPU-accelerated path rendering• OpenGL Utility Toolkit (GLUT) implementer• Author of OpenGL for the X Window System• Co-author of Cg Tutorial
  • 4. Outline• OpenGL’s importance to NVIDIA• OpenGL API improvements & new features – OpenGL 4.2 – Direct3D interoperability – GPU-accelerated path rendering – Kepler Improvements • Bindless Textures• Linux improvements & new features• Cg 3.1 update
  • 5. NVIDIA’s OpenGL Leverage Cg GeForce Parallel Nsight Tegra Quadro OptiX
  • 6. Example of Hybrid Rendering with OptiX OpenGL (Rasterization) OptiX (Ray tracing)
  • 7. Nsight Provides OpenGL Profiling Configure Application Trace Settings
  • 8. Parallel Nsight Provides OpenGL Profiling Magnified trace options shows specific OpenGL (and Cg) tracing options
  • 9. Parallel Nsight Provides OpenGL Profiling
  • 10. Parallel Nsight Provides OpenGL Profiling Trace of mix of OpenGL and CUDA shows glFinish & OpenGL draw calls
  • 11. Only Cross Platform 3D API
  • 12. OpenGL 3D Graphics API• cross-platform• most functional• peak performance• open standard• inter-operable• well specified & documented• 20 years of compatibility
  • 13. OpenGL Spawns Closely Related Standards Congratulations: WebGL officially approved, February 2012 “The web is now 3D enabled”
  • 14. OpenGL 4 – DirectX 11 Superset Buffer and Event Interop• Interop with a complete compute solution – OpenGL is for graphics – CUDA / OpenCL is for compute• Shaders can be saved to and loaded from binary blobs – Ability to query a binary shader, and save it for reuse later• Flow of content between desktop and mobile – All of OpenGL ES 2.0 capabilities available on desktop – EGL on Desktop in the works – WebGL bridging desktop and mobile• Cross platform – Mac, Windows, Linux, Android, Solaris, FreeBSD – Result of being an open standard
  • 15. Increasing Functionality of OpenGL Tessellation and Compute Features 4.X Geometry Shaders 3.X Vertex and Fragment Shaders 2.X Fixed Function 1.X
  • 16. Classic OpenGL State Machine• From 1991-2007 * vertex & fragment processing got programmable 2001 & 2003 [source: GL 1.0 specification, 1992]
  • 17. Complicated from inception
  • 18. OpenGL 3.x Conceptual Processing Flow (pre-2010) uniform/ primitive topology, parameters transformed Legend vertex data Geometric primitive buffer objects Vertex Vertex assembly & vertices assembly primitive processing batch transformed processing programmable pixels operations type, vertex fragments vertex data attributes point, line, and polygon fixed-function filtered texels geometry operations Transform texture fragments buffer data vertex transform feedback fetches buffer objects feedback pixels in framebuffer object textures buffer objects stenciling, depth testing, primitive batch type, vertex blending, accumulation vertex indices, texture vertex attributes texture fetches buffer buffer data, objects Texture Fragment Raster unmap Framebuffer mapping fragment processing operations buffer texture Command Buffer fetches parser store pixel map buffer, pack get buffer buffer data objects image and bitmap texture fragments pixel image pixel image or unpack Pixel specification texture image buffer specification packing objects OpenGL 3.2 pixels to pack image rectangles, bitmaps Image Pixel Pixel primitive unpacking unpacked processing pixels processing copy pixels, copy texture image
  • 19. Patch Control point Patch tessellation Patch evaluation assembly & processing transformed transformed generation transformed processing control points processing transformed patch patch, bivariatetessellation patch control patch domain points patch topology, evaluated patch vertextexturefetches primitive topology, transformed Geometric primitive Legend Vertex Vertex vertex data patch data assembly & assembly primitive processing programmable batch transformed processing vertices operations type, vertex pixels vertex data attributes point, line, fragments and polygon fixed-function geometry filtered texels Transform fragments operations vertex texture transform feedback fetches buffer data buffer objects feedback pixels in framebuffer object textures buffer primitive batch type, objects stenciling, depth testing, vertex vertex indices, texture blending, accumulation vertex attributes texture fetches buffer buffer data, objects Texture Fragment Raster unmap Framebuffer mapping fragment processing operations buffer pixel texture Command Buffer pack fetches parser store buffer map buffer, get buffer objects texture image and bitmap data pixel fragments image pixel image or unpack Pixel specification texture image buffer packing specification objects pixels to pack OpenGL 4.2 image rectangles, Image Pixel Pixel bitmaps primitive copy pixels, unpacking unpacked processing pixels processing copy texture image
  • 20. Accelerating OpenGL Innovation NOW! 2004 2005 2006 2007 2008 2009 2010 2011 2012 DirectX 10.0 DirectX 10.1 DirectX 11 DirectX 9.0c It’s OpenGL 3.1 OpenGL 3.3 + cooking!OpenGL 2.0 OpenGL 2.1 OpenGL 3.0 OpenGL 3.2 OpenGL 4.0 OpenGL 4.1• OpenGL increased pace of innovation - Expect 8 new spec versions in four years - Actual implementations following specifications closely OpenGL 4.2• OpenGL 4.2 is a superset of DirectX 11 functionality - While retaining backwards compatibility
  • 21. What’s new in OpenGL 4.2?• OpenGL 4.2 standardized August 2011 – Immediately shipped by NVIDIA – All Kepler and Fermi GPUs support 4.2• Standardizes: – Compressed RGBA texture – Shader atomic counters – Immutable texture images – Further Direct3Disms – OpenGL Shading Language (GLSL) 4.20
  • 22. OpenGL 4.2• Official compressed RGBA texture formats – GL_ARB_texture_compression_bptc – S3TC/DXTC was widely implemented as EXT, but was never an ARB standard – Now, there’s standard RGBA compressed format• Pixel store modes for compressed block texture specification – ARB_compressed_texture_pixel_storage – New pixel store modes for dealing with sub-rectangle loads of blocked compressed texture formats
  • 23. Details on BPTC• Four new compressed formats – GL_COMPRESSED_RGBA_BPTC_UNORM • RGBA format, lower signal-to-noise than S3TC • [0,1] component range for low-dynamic range (LDR) images – GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM • sRGB version of GL_COMPRESSED_RGBA_BPTC_UNORM – GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT • Compressed high dynamic range (HDR) format – GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT • HDR format for magnitude (unsigned) component values• Each has fixed compression ratio with 4x4 blocks – Much more intricate encoding than S3TC format (see specification)
  • 24. Easy Adoption of BPTC Compression• If you are loading uncompressed texture data – Just use BPTC internal texture format with glTexImage2D – Driver will compress the image for you• And you can read back the compressed image – Then subsequent loads can use glTexCompressedTexImage2D to be faster – However, BPTC compression is hard & takes time to do well so driver’s compression is not optimal • Consider using nvidia-texture-tools Google code project’s compressor• Same approach works for DXTC/S3TC
  • 25. OpenGL 4.2 Compressed Texture Format Table Fixed lossy Block CompressionToken name Size ratio RangeGL_COMPRESSED_RGBA_BPTC_UNORM 4x4 6:1 RGB LDRGL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM 4x4 6:1 sRGB LDRGL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT 4x4 3:1 for RGB Unsigned 4:1 for RGBA HDRGL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT 4x4 3:1 for RGB Signed 4:1 for RGBA HDR
  • 26. OpenGL 4.2 Compressed TexturePixel Storage• New glPixelStorei state settings – Unpack state • GL_UNPACK_COMPRESSED_BLOCK_WIDTH • GL_UNPACK_COMPRESSED_BLOCK_HEIGHT • GL_UNPACK_COMPRESSED_BLOCK_DEPTH • GL_UNPACK_COMPRESSED_BLOCK_SIZE – Pack state, similar just starting GL_PACK_ ...• Allows sub rectangle update for glCompressedTexImage2D, glCompressedTexSubImage2D, etc.
  • 27. OpenGL 4.2 Atomic Counters• Shaders seek to write/read from buffers need a way to get a unique memory location across alls shader instances – Prior to atomic counters, no way to coordinate such reads or writes• New: atomic counters – Allow GLSL shaders to write at unique offset within a buffer object – Also allow GLSL shader to read at unique offset within a buffer object – Counter value stored in a buffer • Updated via glBuferSubData, glMapBuffer, etc.• GLSL declaration of opaque object – layout ( binding = 2, offset = 0 ) uniform atomic_uint variable;• GLSL usage – uint prior_value = atomicCounterIncrement(atomic_uint counter);
  • 28. OpenGL 4.2 Atomic Counters• More GLSL usage of atomic counters – uint updated_value = atomicCounterDecrement(atomic_uint counter); – uint current_value = atomicCounter(atomic_uint counter);• Using – #extension GL_ARB_shader_atomic_counters : enable• Expected usage – Atomic values are offsets for loads and store operations – Using GLSL imageLoad & imageStore built-in functions – Also part of OpenGL 4.2
  • 29. In OpenGL 4.2, Shaders allowed Side-effects• Very exciting development – Prior to 4.2, standard GLSL shaders had their explicit outputs and that was it• In OpenGL 4.2 – Loads and stores to images and buffers allowed from shaders – Use atomic operations or atomic counters to synchronize those updates – You need to manage any required ordering though
  • 30. OpenGL 4.2 Immutable Texture Storage• OpenGL allows flexible layout of mipmap levels with a texture object• New commands gives a frozen mipmap layout – glTexStorage1D, glTexStorage2D, glTexStorage3D • Free of selector – glTextureStorage1DEXT, glTextureStorage2DEXT, glTextureStorage3DEXT • Explicit texture parameter• Allow driver to assume layout of texture immutable – Still can update texels, just not layout
  • 31. Direct3Dism Concept• Allows your 3D content to be API agnostic – OpenGL supports both OpenGL & Direct3D conventions, so support either style OpenGLYour OpenGL Your API OpenGL application OpenGL content OpenGL driver content authored application + Direct3D to OpenGL conventions conventions 3D API hardware Direct3D conventions interface interface GPU supported by OpenGL tooYour Direct3D Your Direct3D API Direct3D application Direct3D content Direct3D driver content authored application conventions to Direct3D conventions
  • 32. Further Direct3Disms in OpenGL 4.2• Direct3Disms = semantic compatibility with Direct3D behaviors – Not just same functionality as Direct3D—but same functionality with Direct3D-style semantics• Alignment constraints for mapped buffers – ARB_buffer_alignment – Makes sure mapped buffers are aligned for SIMD CPU instruction access – Really just a query for minimum alignment • Expect alignment to be no smaller than 64 bytes
  • 33. OpenGL 4.2 Base Instance• Extension on instanced rendering – For drawing multiple instances of geometry • Designed to be compatible with Direct3D – See ARB_instanced_arrays• Allows the specification of – base vertex – base instance – for each instanced draw• Generalizes instanced rendering
  • 34. OpenGL 4.2 GLSL Features• Conservative Depth – Permits depth/stencil tests before shading – Re-declares gl_FragDepth to restrict depth changes• Allows restrictions – New depth can only increase depth • layout (depth_greater) out float gl_FragDepth; – New depth can only decrease depth • layout (depth_less) out float gl_FragDepth;• To use – #extension GL_ARB_conservative_depth : enable
  • 35. OpenGL 4.2 GLSL Features• Miscellaneous GLSL changes – Allow line-continuation character using ‘’ – Use UTF8 for character set, for comments – Implicit conversions of return values – const keyword allowed on variables with functions – No strict order of qualifiers – Adds layout qualifier binding for uniform blocks – C-style aggregate initializers – Allow .length method to return number of components or columns in vectors and matrices – Allow swizzles on scalars – New built-in constants for gl_MinProgramTexelOffset and Max• All minor improvements for consistency with C/C++ and minimize aggravation
  • 36. OpenGL-relatedLinux Improvements Support for X Resize, Rotate, and Reflect Extension  Also known as RandR  Version 1.2 and 1.3 OpenGL enables, by default, “Sync to Vertical Blank”  Locks your glXSwapBuffers to the monitor refresh rates  Matches Windows default now  Previously disabled by default
  • 37. OpenGL-relatedLinux Improvements• Expose additional full-scene antialiasing (FSAA) modes – 16x multisample FSAA on all GeForce GPUs • 2x2 supersampling of 4x multisampling – Ultra high-quality FSAA modes for Quadro GPUs • 32x multisample FSAA – 2x2 supersampling of 8x multisampling • 64x multisample FSAA – 4x4 supersampling of 4x multisampling• Coverage sample FSAA on GeForce 8 series and better – 4 color/depth samples + 12 depth samples
  • 38. Multisampling FSAA Patterns aliased 2x multisampling 4x multisampling 8x multisampling1 sample/pixel 2 samples/pixel 4 samples/pixel 8 samples/pixel64 bits/pixel 128 bits/pixel 256 bits/pixel 512 bits/pixel Assume: 32-bit RGBA + 24-bit Z + 8-bit Stencil = 64 bits/sample
  • 39. Supersampling FSAA Patterns2x2 supersampling 2x2 supersampling 4x4 supersamplingof 4x multisampling of 8x multisampling of 16x multisampling16 samples/pixel 32 samples/pixel 64 samples/pixel 1024 bits/pixel 2048 bits/pixel 4096 bits/pixel Quadro GPUs Assume: 32-bit RGBA + 24-bit Z + 8-bit Stencil = 64 bits/sample
  • 40. NVIDIA X Server Settings for LinuxControl Panel
  • 41. GLX Protocol• Network transparent OpenGL – Run OpenGL app on one machine, display the X and 3D on a different machine 3D app X server GLX OpenGL Server GLX Client OpenGL network connection
  • 42. OpenGL-relatedLinux Improvements Official GLX Protocol support for OpenGL extensions• GL_ARB_half_float_pixel • GL_EXT_point_parameters• GL_ARB_transpose_matrix • GL_EXT_stencil_two_side• GL_EXT_blend_equation_separate • GL_NV_copy_image• GL_EXT_depth_bounds_test • GL_NV_half_float• GL_EXT_framebuffer_blit • GL_NV_occlusion_query• GL_EXT_framebuffer_multisample • GL_NV_point_sprite• GL_EXT_packed_depth_stencil • GL_NV_register_combiners2
  • 43. OpenGL-relatedLinux Improvements Tentative GLX Protocol support for OpenGL extensions• GL_ARB_map_buffer_range • GL_EXT_vertex_attrib_64bit• GL_ARB_shader_subroutine • GL_NV_conditional_render• GL_ARB_stencil_two_side • GL_NV_framebuffer_multisample• GL_EXT_texture_integer _coverage• GL_EXT_transform_feedback2 • GL_NV_texture_barrier • GL_NV_transform_feedback2
  • 44. Synchronizing X11-based OpenGL Streams• New extension – GL_EXT_x11_sync_object• Bridges the X Synchronization Extension with OpenGL 3.2 “sync” objects (ARB_sync)• Introduces new OpenGL command – GLintptr sync_handle; – GLsync glImportSyncEXT (GLenum external_sync_type, GLintptr external_sync, GLbitfield flags); • external_sync_type must be GL_SYNC_X11_FENCE_EXT • flags must be zero
  • 45. Other Linux Updates• GL_CLAMP behaves in conformant way now – Long-standing work around for original Quake 3• Enabled 10-bit per component X desktop support – GeForce 8 and better GPUs• Support for 3D Vision Pro stereo now
  • 46. What is 3D Vision Pro?• For Professionals• All of 3D Vision support, plus • Radio frequency (RF) glasses, Bidirectional • Query compass, accelerometer, battery • Many RF channels – no collision • Up to ~120 feet • No line of sight needed to emitter • NVAPI to control
  • 47. OpenGL Inter-GPU Communication PCI Express• NV_copy_image extension GPU1 GPU0 destRC srcRC – Quadro-only Compute/ Compute/ DMA DMA Render Render – (Asynchronous) copy Engine Engine Engine Engine between GPU’s – Does not require GPU GPU binding textures or Memory Memory destTex srcTex state changes – App handles wglCopyImageSubDataNV(srcRC, srcTex, GL_TEXTURE_2D,0, 0, 0, 0, destRC, destTex, synchronization GL_TEXTURE_2D, 0, 0, 0, 0, width, height, 1);
  • 48. OpenGL Interoperability Multi-GPU
  • 49. DirectX / OpenGL Interoperability Application • DirectX interop is a GL API extension (WGL_NVX_dx_interop) • Premise: create resources in DirectX, get access to them within OpenGL OpenGL DirectX driver driver • Read/write support for DirectX 9 textures, render targets and vertex buffers • Implicit synchronization is done Texture / Vertex buffer object between Direct3D and OpenGL data • Soon: DirectX 10/11 support GPU Display
  • 50. DirectX / OpenGL Interoperability// create Direct3D device and resources the regular waydirect3D->CreateDevice(..., &dxDevice);dxDevice->CreateRenderTarget(width, height, D3DFMT_A8R8G8B8, D3DMULTISAMPLE_4_SAMPLES, 0, FALSE, &dxColorBuffer, NULL);dxDevice->CreateDepthStencilSurface(width, height, D3DFMT_D24S8, D3DMULTISAMPLE_4_SAMPLES, 0, FALSE, &dxDepthBuffer, NULL);
  • 51. DirectX / OpenGL Interoperability// Register DirectX device for OpenGL interopHANDLE gl_handleD3D = wglDXOpenDeviceNVX(dxDevice);// Register DirectX render targets as GL texture objectsGLuint names[2];HANDLE handles[2];handles[0] = wglDXRegisterObjectNVX(gl_handleD3D, dxColorBuffer, names[0], GL_TEXTURE_2D_MULTISAMPLE, WGL_ACCESS_READ_WRITE_NVX);handles[1] = wglDXRegisterObjectNVX(gl_handleD3D, dxDepthBuffer, names[0], GL_TEXTURE_2D_MULTISAMPLE, WGL_ACCESS_READ_WRITE_NVX);// Now textures can be used as normal OpenGL textures
  • 52. DirectX / OpenGL Interoperability// Rendering example: DirectX and OpenGL rendering to the// same render targetdirect3d_render_pass(); // D3D renders to the render targets as usual// Lock the render targets for GL accesswglDXLockObjectsNVX(handleD3D, 2, handles);opengl_render_pass(); // OpenGL renders using the textures as render // targets (e.g., attached to an FBO)wglDXUnlockObjectsNVX(handleD3D, 2, handles);direct3d_swap_buffers(); // Direct3D presents the results on the screen
  • 53. CUDA Interoperability example Application • Interop APIs are extensions to OpenGL and CUDA • Multi-card interop 2x faster on Quadro / Tesla. Transfer between cards, minimal OpenGL CUDA CPU involvement driver driver • GeForce requires CPU copy Quadro fast Tesla or or Interop GeForce GeForce Display
  • 54. Multi-GPU OpenGL Interoperability Application • Interop using NV_copy_image OpenGL extension GPU affinity • Transfer directly between cards, minimal CPU involvement OpenGL OpenGL driver driver • Quadro only Quadro fast Quadro Off-screen Interop buffer Display
  • 55. What is path rendering?• A rendering approach – Resolution-independent two-dimensional graphics – Occlusion & transparency depend on rendering order • So called “Painter’s Algorithm” – Basic primitive is a path to be filled or stroked • Path is a sequence of path commands • Commands are – moveto, lineto, curveto, arcto, closepath, etc.• Standards – Content: PostScript, PDF, TrueType fonts, Flash, Scalable Vector Graphics (SVG), HTML5 Canvas, Silverlight, Office drawings – APIs: Apple Quartz 2D, Khronos OpenVG, Microsoft Direct2D, Cairo, Skia, Qt::QPainter, Anti-grain Graphics,
  • 56. What is NV_path_rendering?• OpenGL extension to GPU-accelerate path rendering• Uses “stencil, then cover” (StC) approach – Create a path object – Step 1: “Stencil” the path object into the stencil buffer • GPU provides fast stenciling of filled or stroked paths – Step 2: “Cover” the path object and stencil test against its coverage stenciled by the prior step • Application can configure arbitrary shading during the step – More details later• Supports the union of functionality of all major path rendering standards – Includes all stroking embellishments – Includes first-class text and font support – Allows this functionality to mix with traditional 3D and programmable shading
  • 57. Pixel pipeline Vertex pipeline Path pipeline Application Path specificationPixel assembly Vertex assembly Transform path (unpack) Vertex operations transform feedback Primitive assemblyPixel operations Primitive operations Fill/Stroke Covering Pixel pack Rasterization read Texture Fragment operations back memory Fill/Stroke Application Raster operations Stenciling Framebuffer Display
  • 58. Talk on Tuesday• “GPU-Accelerated Path Rendering” – Tuesday, May 15 – 14:00-14:50, Room A3• Teaser scene
  • 59. Bindless Graphics• Problem: Binding to different objects (textures, buffers) takes a lot of validation time in driver – And applications are limited to a small palette of bound buffers and textures – Approach of OpenGL, but also Direct3D• Solution: Exposes GPU virtual addresses – Let shaders and vertex puller access buffer and texture memory by its virtual address!
  • 60. Prior to Bindless Graphics• Traditional OpenGL – GPU memory reads are “indirected” through bindings • Limited number of texture units and vertex array attributes – glBindTexture—for texture images and buffers – glBindBuffer—for vertex arrays
  • 61. Buffer-centric Evolution• Data moves onto GPU, away from CPU – Apps on CPUs just too slow at moving data otherwise Array Element Buffer glBegin, glDrawElements, etc. Object (VeBO) Texture Buffer Object (TexBO)Vertex Array Buffer Object Vertex Puller (VaBO) texel data Transform Feedback Buffer (XBO) Vertex Shading Pixel Unpack Buffer (PuBO) vertex data Texturing Geometry Shading glDrawPixels, glTexImage2D, etc. Parameter Buffer Object (PaBO) glReadPixels, Pixel Pack Buffer etc. Fragment Pixel (PpBO) Uniform Buffer Shading Pipeline pixel data Object (UBO)parameter data Framebuffer
  • 62. Kepler – Bindless Textures • Enormous increase in the number of unique textures available to shaders at run-time • More different materials and richer texture detail in a scene texture #0Shader code texture #1 Shader code texture #2 … texture #127 … Pre-Kepler texture binding model Kepler bindless textures over 1 million unique textures
  • 63. Kepler – Bindless TexturesPre-Kepler texture binding model Kepler bindless texturesCPU CPU Load texture A Load textures A, B, C Load texture B Draw() Load texture C Bind texture A to slot I GPU Bind texture B to slot J Read from texture A Draw() Read from texture B Read from texture C GPU Read from texture at slot I Read from texture at slot JCPU Bind texture C to slot K Bindless model reduces CPU Draw() overhead and improves GPU access GPU efficiency Read from texture at slot K
  • 64. Bindless Textures • Apropos for ray-tracing and advanced rendering where textures cannot be “bound” in advanceShader code
  • 65. Bindless performance benefit Numbers obtained with a directed test
  • 66. More Information on Bindless Texture• Kepler has new NV_bindless_texture extension – Texture companion to • NV_vertex_buffer_unified_memory for bindless vertex arrays • NV_shader_buffer_load for bindless shader buffer reads• API specification publically available – http://developer.download.nvidia.com/opengl/specs/GL_NV_bindless_texture.txt
  • 67. API Usage to Initialize Bindless Texture• Make a conventional OpenGL texture object – With a 32-bit GLuint name• Query a 64-bit texture handle from 32-bit texture name – GLuint64 glGetTextureHandleNV(GLuint);• Make handle resident in context’s GPU address space – void glMakeTextureHandleResidentNV(GLuint64);
  • 68. Writing GLSL for Bindless Textures• Request GLSL to understand bindless textures – #version 400 // or later – #extension GL_NV_bindless_texture : require• Declare a sampler in the normal way – in sampler2D bindless_texture;• Alternatively, access bindless samplers in big array: – uniform Samplers { sampler2D lotsOfSamplers[256]; } – Exciting: 256 samplers exceeds the available texture units!
  • 69. Update Sampler Uniforms withBindless Texture Handle• Get a location for a sampler or image uniform – GLint loc = glGetUniformLocation(program, “bindless_texture”); – GLint loc_array = glGetUniformLocation(program, “lotsOfSamplers”);• Then set sampler to the bindless texture handle – glProgramUniformHandleui64NV(program, location, 1, &bindless_handle);
  • 70. Seam-free Cube Map Edges• Cube maps have edges along each face – Traditionally texture mapping hardware simply clamps to these seam edges• Results in “seam” artifacts – Particularly when level-of-detail bias is large • Meaning very blurry levels • But seams appear sharply seam• Use glEnable( GL_TEXTURE_CUBE_MAP_SEA MLESS) to mitigate these artifacts on context-wide basis – Applies to all cube maps
  • 71. Seamless Cube Maps: Before and After • Before: with edge seams • After: without seams
  • 72. Kepler ProvidesPer-texture Seamless Cube Map Support• Kepler GPUs support – GL_AMD_seamless_cubemap_per_texture – Provides per-texture object parameters • glTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_CUBE_MAP_SEAMLESS_ARB, GL_TRUE); – Not sure if you want a mix of seamless and seamed cube maps… • …but supported anyway
  • 73. Cg Toolkit 3.1 Update• Cg 3.0 = big upgrade – Introduced Shader Model 5.0 support for OpenGL • New gp5vp, gp5gp, and gp5fp profiles correspond to Shader Model 5.0 vertex, geometry, and fragment profiles – Compiles to assembly text for the NV_gpu_program5 assembly-level shading language – Also programmable tessellation profiles • gp5tcp = NV_gpu_program5 tessellation control program • gp5tep = NV_gpu_program5 tessellation evaluation program – Also DirectX 11 profiles• Cg 3.1 refines 3.0 functionality – Bug fixes, constant buffer support, better GLSL support
  • 74. Cg 3.1 OpenGL Shader Model 5.0 Profiles indices primitive rectangular & indices primitive topology & triangluar topology & in-band vertex patch in-band vertex attributes index to vertex attributes index to vertex attribute attribute vertex puller puller vertex attributes attributes control point vertex vertex (vertex) shader attributes vertex shader gp4vp shader gp5vp gp5vp control assembled points tessellation assembled primitive primitive level-of-detail control (hull) parameters shader geometry geometry gp4gp shader gp5tcs gp5gp shader tessellation patch control primitive stream primitive stream generator points &programmable clipping, parameters clipping, setup, & setup, & rasterization rasterizationfixed-function fragments gp5tes tessellation fragments fine primitive fragment topology evaluation gp4fp fragment (u, v) coordinates (domain) shader gp5fp shader shaderCg 2.x support Cg 3.x adds programmable tessellation
  • 75. Cg 3.1 profiles OpenGL OpenGL Shader Shader Model Shader Model OpenGL Shading domain 4.0 5.0 Language (GLSL) DirectX 10 DirectX 11 Vertex (or control point) gp4vp gp5vp glslv vs_4_0 vs_5_0 Tessellation control (hull) n/a gp5tcp n/a n/a hs_5_0 Tessellation n/a n/a evaluation gp5tep n/a ds_5_0 (domain) Geometry gp5gp gp5gp glslg gs_4_0 gs_5_0 Fragment gp4fp gp5fp glslf ps_4_0 ps_5_0
  • 76. Cg 3.1 Update• Various improvements & bug fixes – Constant buffer fixes – GLSL translation for different GLSL versions• Support for CENTROID, FLAT, and NOPERSPECTIVE fragment program interpolants• Deprecated functionality – Removed Direct3D 8 support – Minimum Mac OS X version: 10.5 (Leopard)
  • 77. Cg Off-line CompilerCompiles GLSL to Assembly• GLSL is “opaque” about the assembly generated – Hard to know the quality of the code generated with the driver – Cg Toolkit 3.0 is a useful development tool even for GLSL programmers• Cg Toolkit’s cgc command-line compiler actually accepts both the Cg (cgc’s default) and GLSL (with –oglsl flag) languages – Uses the same compiler technology used to compile GLSL within the driver• Example command line for compiling earlier tessellation control and evaluation GLSL shaders – cgc -profile gp5tcp -oglsl -po InputPatchSize=3 -po PATCH_3 filename.glsl (assumes 3 input control points and 3 outputs points) – cgc -profile gp5tep -oglsl -po PATCH_3 filename.glsl (assumes 3 input control points)
  • 78. Questions?
  • 79. Other OpenGL-related Sessions at GTC• S0024: GPU-Accelerated Path Rendering – Tuesday, 14:00, Room A3• S0049: Using the GPU Direct for Video API – Tuesday, 15:00, Room J8• S0356: Optimized Texture Transfers – Tuesday, 16:00, Room J2• S0267A: Mixing Graphics and Compute with Multiple GPUs – Tuesday, 17:00, Room J2• S0353: Programming Multi-GPUs for Scalable Rendering – Wednesday, 9:00, Room A1• S0267B - Mixing Graphics and Compute with Multiple GPUs – Thursday, 17:30, Room L