OpenCL Overview. Shanghai March Neil Trevett Vice President Mobile Content, NVIDIA President, The Khronos Group

Transcription

1 Copyright Khronos Group, Page 1 OpenCL Overview Shanghai March 2012 Neil Trevett Vice President Mobile Content, NVIDIA President, The Khronos Group

2 Copyright Khronos Group, Page 2 Processor Parallelism CPUs Multiple cores driving performance increases Emerging Intersection GPUs Increasingly general purpose data-parallel computing Multiprocessor programming e.g. OpenMP Heterogeneous Computing Graphics APIs and Shading Languages OpenCL is a programming framework for heterogeneous compute resources

3 Copyright Khronos Group, Page 3 The BIG Idea behind OpenCL OpenCL execution model - Define N-dimensional computation domain - Execute a kernel at each point in computation domain C Derivative to write kernels based on ISO C99 - APIs to discover devices in a system and distribute work to them Targeting many types of device - GPUs, CPUs, DSPs, embedded systems, mobile phones.. Even FPGAs Traditional loops void trad_mul(int n, const float *a, const float *b, float *c) { int i; for (i=0; i<n; i++) c[i] = a[i] * b[i]; } Data Parallel OpenCL kernel void dp_mul(global const float *a, global const float *b, global float *c) { int id = get_global_id(0); c[id] = a[id] * b[id]; } // execute over n work-items

4 Copyright Khronos Khronos 2009 Group, Page 4 OpenCL Heterogeneous Computing Framework for programming diverse parallel computing resources in a system Platform Layer API - Query, select and initialize compute devices Kernel Language Specification - Subset of ISO C99 with language extensions Runtime API - Execute compute kernels gather results OpenCL has Embedded profile - No need for a separate ES spec

5 Copyright Khronos Group, Page 5 OpenCL Working Group Members Diverse industry participation many industry experts - Processor vendors, system OEMs, middleware vendors, application developers - Academia and research labs, FPGA vendors NVIDIA is chair, Apple is specification editor Apple

6 Copyright Khronos Group, Page 6 OpenCL Milestones Six months from proposal to released OpenCL 1.0 specification - Due to a strong initial proposal and a shared commercial incentive Multiple conformant implementations shipping - For CPUs and GPUs on multiple OS 18 month cadence between OpenCL 1.0, OpenCL 1.1 and now OpenCL Backwards compatibility protect software investment OpenCL working group formed Dec08 OpenCL 1.0 conformance tests released Jun10 OpenCL 1.2 Specification and conformance tests released! Jun08 OpenCL 1.0 released May09 OpenCL 1.1 Specification and conformance tests released Nov11

7 Copyright Khronos Group, Page 7 OpenCL 1.2 Announced in December Significant updates - Khronos being responsive to developer requests - Updated OpenCL 1.2 conformance tests available - Multiple implementations underway Backward compatible upgrade to OpenCL OpenCL 1.2 will run any OpenCL 1.0 and OpenCL 1.1 programs - OpenCL 1.2 platform can contain 1.0, 1.1 and 1.2 devices - Maintains embedded profile for mobile and embedded devices

8 Copyright Khronos Group, Page 8 Programming Kernels: OpenCL C Derived from ISO C99 - But without some C99 features such as standard C99 headers, function pointers, recursion, variable length arrays, and bit fields Language Features Added - Work-items and workgroups - Vector types - Synchronization - Address space qualifiers Also includes a large set of built-in functions - Image manipulation - Work-item manipulation, - Math functions, etc.

9 Copyright Khronos Group, Page 9 OpenCL Platform Model One Host + one or more Compute Devices - Each Compute Device is composed of one or more Compute Units - Each Compute Unit is further divided into one or more Processing Elements

10 Copyright Khronos Group, Page 10 OpenCL Execution Model OpenCL application runs on a host which submits work to the compute devices - Context: The environment within which work-items executes includes devices and their memories and command queues - Program: Collection of kernels and other functions (Analogous to a dynamic library) - Kernel: the code for a work item. Basically a C function - Work item: the basic unit of work on an OpenCL device GPU Queue Context CPU Queue Applications queue kernel execution - Executed in-order or out-of-order

11 Copyright Khronos Group, Page 11 Creating an OpenCL Program CPU GPU DSP Context Programs Kernels Memory Objects Command Queue Kernel0 Images Programs Kernel1 Kernel2 Buffers In order & out of order Compile Create data and arguments Send for execution

12 1024 Copyright Khronos Group, Page 12 An N-dimension domain of work-items Kernels executed across a global domain of work-items Work-items grouped into local workgroups Define the best N-dimensioned index space for your algorithm - Global Dimensions: 1024 x 1024 (whole problem space) - Local Dimensions: 128 x 128 (work group executes together) 1024 Synchronization between work-items possible only within workgroups: barriers and memory fences Cannot synchronize outside of a workgroup

13 Enqueue Kernel 1 Enqueue Kernel 2 Enqueue Kernel 1 Enqueue Kernel 2 Copyright Khronos Group, Page 13 Synchronization: Queues & Events Events can be used to synchronize kernel executions between queues Example: 2 queues with 2 devices Kernel 2 starts before the results from Kernel 1 are ready Kernel 2 waits for an event from Kernel 1 and does not start until the results are ready CPU Kernel 2 CPU Kernel 2 GPU Kernel 1 GPU Kernel 1 Time Time

14 OpenCL Memory Model Private Memory Per work-item Local Memory Shared within a workgroup Global/Constant Memory Visible to all workgroups Host Memory On the CPU Private Memory Work-Item Workgroup Compute Device Work-Item Local Memory Private Memory Workgroup Global/Constant Memory Host Memory Private Memory Work-Item Local Memory Private Memory Work-Item Host Memory management is Explicit You must move data from host -> global -> local and back Copyright Khronos Group, Page 14

15 Copyright Khronos Group, Page 15 Partitioning Devices Devices can be partitioned into sub-devices - More control over how computation is assigned to compute units Sub-devices may be used just like a normal device - Create contexts, building programs, further partitioning and creating command-queues Three ways to partition a device - Split into equal-size groups - Provide list of group sizes - Group devices sharing a part of a cache hierarchy Compute Device Compute Device Host Compute Device Compute Unit Compute Unit Compute Unit Sub-device #1 Real-time processing tasks Compute Unit Compute Unit Compute Unit Sub-device #2 Mainline processing tasks

16 Copyright Khronos Group, Page 16 Custom Devices and Built-in Kernels Embedded platforms often contain specialized hardware and firmware - That cannot support OpenCL C Built-in kernels can represent these hardware and firmware capabilities - Such as video encode/decode Hardware can be integrated and controlled from the OpenCL framework - Can enqueue built-in kernels to custom devices alongside OpenCL kernels FPGAs are one example of device that can expose built-in kernels - Latest FPGAs can support full OpenCL C as well OpenCL becomes a powerful coordinating framework for diverse resources - Programmable and non-programmable devices controlled by one run-time Built-in kernels enable control of specialized processors and hardware from OpenCL run-time

17 Copyright Khronos Group, Page 17 Installable Client Driver Analogous to OpenGL ICDs in use for many years - Used to handle multiple OpenGL implementations installed on a system Optional extension - Platform vendor will choose whether to use ICD mechanisms Khronos OpenCL installable client driver loader - Exposes multiple separate vendor installable client drivers (Vendor ICDs) Application can access all vendor implementations - The ICD Loader acts as a de-multiplexor Application ICD Loader ICD Loader enables application to use any of the installed implementations Vendor #1 OpenCL Vendor #2 OpenCL Vendor #3 OpenCL ICD Loader ensures multiple implementations are installed cleanly

18 Copyright Khronos Group, Page 18 Other Major New Features in OpenCL 1.2 Separate compilation and linking of objects - Provides the capabilities and flexibility of traditional compilers - Create a library of OpenCL programs that other programs can link to Enhanced Image Support - Added support for 1D images, 1D & 2D image arrays - OpenGL sharing extension now enables an OpenCL image to be created from an OpenGL 1D texture, 1D and 2D texture arrays DX9 Media Surface Sharing - Efficient sharing between OpenCL and DirectX 9 or DXVA media surfaces DX11 surface sharing - Efficient sharing between OpenCL and DirectX 11 surfaces And many other updates and additions..

19 Copyright Khronos Group, Page 19 OpenCL Desktop Implementations

20 Copyright Khronos Group, Page 20 OpenCL Books Available Now! OpenCL Programming Guide - The Red Book of OpenCL - OpenCL in Action - Heterogeneous Computing with OpenCL - The OpenCL Programming Book -

21 Copyright Khronos Group, Page 21 Spec Translations Japanese OpenCL 1.1 spec translation available today Valued partnership between Khronos and CUTT in Japan Working on OpenCL 1.2 specification translations - Japanese, Korean and Chinese

22 Copyright Khronos Group, Page 22 Khronos OpenCL Resources OpenCL is 100% free for developers - Download drivers from your silicon vendor OpenCL Registry - OpenCL 1.2 Reference Card - PDF version - Online Man pages - OpenCL Developer Forums - Give us your feedback! -

23 Copyright Khronos Group, Page 23 Looking Forward OpenCL-HLM (High Level Model) Exploring high-level programming model, unifying host and device execution environments through language syntax for increased usability and broader optimization opportunities Long-term Core Roadmap Exploring enhanced memory and execution model flexibility to catalyze and expose emerging hardware capabilities WebCL Bring parallel computation to the Web through a JavaScript binding to OpenCL OpenCL-SPIR (Standard Parallel Intermediate Representation) Exploring low-level Intermediate Representation for code obfuscation/security and to provide target back-end for alternative high-level languages