pt-4052, introduction to amd developer tools, by yaki tebeka and gordon selley

INTRODUCTION TO AMD DEVELOPER TOOLS YAKI TEBEKA

GORDON SELLEY

2 | INTRODUCTION TO AMD DEVELOPER TOOLS | NOVEMBER 2013

APU: ACCELERATED PROCESSING UNIT

!  The APU has arrived and it is a great advance over previous plaKorms

!  Combines scalar processing on CPU, parallel processing on GPU, co-‐processors and high bandwidth access to memory

!  Higher performance with lower power

Dual Core x86 Module

Dual Core x86 Module

L2 Cache L2 Cache

DDR3 Controller

PCIe

UNB Channel

GPU

PCIe PCIe

DisplayPort /HDMI

UVD/ VCE


HSA: HETEROGENEOUS SYSTEM ARCHITECTURE

Processor and soTware design that makes it easy for developers to harness the compuVng power of an APU, for faster and more power-‐efficient computaVon


AMD'S DEVELOPER TOOLS STRATEGY

!  Converge tools and technologies to an “HSA Developer Tools Suite” !  Maintain capabiliVes for specific market needs

CPU Tools

GPU Tools

Need for System Level APU Tool Convergence of Technologies and Tools

GPU PerfStudio

AMD CodeXL


CONVERGE TOOLS AND TECHNOLOGIES

gDEBugger Code Analyst APP Profiler

APP Kernel Analyzer AMD Driver innovaVons


AMD CODE XL

!  Coherent, innovaVve and unified developer tools suite ‒ Debug, Profile, and Analyze applicaVons ‒  System-‐wide “white box” view ‒ AMD CPUs, GPUs and APUs ‒ Does not require source code modificaVons


AMD CODE XL AVAILABILITY

!  Windows® Standalone

!  MicrosoT® Visual Studio® 2010, 2012 Extension

!  Linux® Standalone

CodeXL Demo


GPU DEBUGGING

!  Transforms GPU debugging process to a “white box” model

!  Finds “hard to catch” heterogeneous compuVng related bugs

!  IntuiVve and familiar debugging experience

!  Feature highlights: ‒ OpenCL™ and OpenGL API Level Debugger ‒ OpenCL™ Kernel Source Code Debugger ‒ MulV-‐Watch view ‒ OpenCL Buffers’ Data view


CPU TIME BASED PROFILING

!  Usage ‒  Is used for finding program level/algorithmic/data structure related performance issues

!  Profiling Method ‒ The user defines a Vme interval (default is 1 msec) ‒ On every Vme interval, for each thread, the profiler logs the instrucVon pointer ‒ Call-‐stack logging is opVonal (very useful)


CPU EVENT BASED PROFILING

!  Usage ‒  Is used for locaVng CPU micro-‐architectural related performance issues

!  Profiling Method ‒ The user selects an event (example: cache misses) ‒ The user defines event count (example: 5000) ‒ A counter counts the event occurrences (cache misses) ‒ When the counter reaches the event count, the profiler logs the instrucVon pointer ‒ Call-‐stack logging is opVonal


GPU COMPUTE PROFILING

!  Analyzes and profiles OpenCL™ and MicrosoT® DirectCompute applicaVons

!  API Trace View: Traces OpenCL™ API calls, IdenVfy failed API calls, resource leaks and best pracVces !  Timeline VisualizaNon: Visualize host, kernel and data transfer execuVon in a Vmeline chart

!  Summary Pages: Finds the top booleneck, top ten kernel execuVon and data transfer operaVons

!  Performance Counter View: Collects GPU performance counters for OpenCL™ kernels


STATIC KERNEL ANALYSIS

!  StaVc analysis tool to compile, analyze and disassemble an OpenCL™ kernel for AMD CPUs, APUs and GPUs

!  View AMD Intermediate Language (AMD IL) and hardware disassembly (ISA)

!  Generate offline compilaVon of OpenCL™ kernel binary

!  View kernel compilaVon warning and error messages

!  View compiler staVsVcs and esVmate performance

!  Only require the OpenCL™ kernel source code as an input !  Does not require a GPU in the system


WHAT’S NEW IN CODEXL?

!  Version 1.1 ‒ Windows® 8 support ‒ MicrosoT® Visual Studio® 2012 integraVon ‒  Linux® KDE support ‒  Support AMD’s Sea Island family of GPUs ‒  Single CodeXL Explorer tree

!  Version 1.2 ‒ Windows® 8.1 support ‒ Enhanced CPU Profiling user interface ‒ CPU profiling of Java /.NET applicaVons ‒  Support Kabini APUs


WHAT’S NEW IN CODEXL /2

!  Version 1.3 ‒ Remote GPU Profiling and Remote GPU debugging ‒  IntegraVng Kernel Analyzer ‒ CPU Profiling of Cache Line UVlizaVon ‒ CPU Profiling support for Java inline funcVons ‒ GPU API Level Debugging support for OpenGL 4.3 ‒  Support Kaveri APUs

APP SDK


ACCELERATED PARALLEL PROCESSING (APP) SDK

!  What is AMD APP SDK? ‒ Development plaKorm for heterogeneous compuVng ‒ Helps developers to exploit the highly parallel compute acceleraVon of GPUs ‒ Designed to allow a GPU to work harmoniously with a mulVcore CPU to execute demanding compuVng tasks ‒ Helps enable a beoer balanced plaKorm opVmizaVon for APUs ‒  Supports open industry standards and embraces open source soTware

‒ OpenCL™ ‒  C++ AMP ‒  Aparapi (Java) ‒  Bolt C++ template library ‒ OpenCV ‒ OpenNI

‒ Helps enable development for cross-‐OS and cross-‐plaKorm heterogeneous soluVons ‒ Contains numerous programming samples


WHAT’S NEW IN APP SDK 2.9

!  Unified SDK installer

!  Sample browser

!  New SDK samples ‒ OpenCV and OpenNI ‒ Bolt samples ‒ Concurrent kernels ‒ GCN samples

!  Bolt C++ template library support on Linux®

!  Support for latest AMD hardware (“Kabini” APUs, R9 & R7 series GPUs)

!  Download the latest APP SDK from AMD today from AMD Developer Central (developer.amd.com)

GPU PerfStudio2


!  GPU PerfStudio2 is AMD's performance and debugging tool for graphics applicaVons.

!  Key Features ‒  Integrated Frame Profiler, Frame Debugger, and API Trace with CPU Vming informaVon ‒  Shader Debugger with support for DirectX®10 & 11 HLSL and Assembly code ‒ Client / Server model -‐ GPU PerfStudio 2 Client runs locally or remotely over the network ‒ GPU PerfStudio 2 Server supports 32-‐bit and 64-‐bit applicaVons ‒  Supports DirectX®11, DirectX®10.1 , DirectX®10 and OpenGL 4.2 applicaVons ‒ No special build required for your applicaVon ‒  Small footprint – no installaVon ‒  Free download hop://developer.amd.com

GPU PERFSTUDIO 2 INTRODUCTION


INTRODUCTION TO GPU PERFSTUDIO 2

!  Capture, play back and view the contents of a frame

!  Scrub through draw calls !  Visualize the GPU Vme for each draw call !  View all game resources and state bound

at each draw call !  Inspect the resources at each stage of the

pipeline !  View, edit and debug shader code

FRAME DEBUGGER



!  IdenVfy costly draw calls !  Detect GPU pipeline stage

boolenecks !  InvesVgate performance issues at

the counter level

FRAME PROFILER



!  Edit the live HLSL and GLSL code inside your app while running in the tool

!  Debug the live HLSL and Assembly code inside your app while running in the tool ‒  Step through shader code ‒  Inspect all register values ‒  Insert and run-‐to break points

!  Compare before and aTer edit performance using the Profiler

SHADER DEBUGGER & EDITOR



!  Inspect all API calls (with arguments)

!  CPU Vmeline informaVon for each API call

!  Visualize mulV-‐threaded API usage

!  Supports DirectX® 11 Command Lists and deferred contexts

API TRACE



!  Widely used by internal groups in AMD ‒ AMD Developer Technology Engineers: OpVmize & debug game Vtles in conjuncVon with developers ‒ AMD Driver Performance Team: Improve GPU benchmarks and Vtles at the driver level ‒ AMD Driver Team: Inspect apps that cause driver problems ‒ AMD Game Compute Team: Debug and opVmize game technologies for new GPU hardware -‐ AMD Mecha Demo, Ladybug, Leo demo

!  External users ‒ Graphics developers: Used in the development of DirectX®11 and OpenGL graphics applicaVons

WHO USES IT?


GPU PERFSTUDIO 2

!  Remote usage allows the game to be run full screen. Higher profiling accuracy, useful during final opVmizaVon.

!  Local usage on a single machine (preferred by developers). More suited for use during development.

REMOTE AND LOCAL DEBUG SESSIONS

Server Client

Remote debugging

Server Client

Local debugging server & client on same computer

Server & Client


OPTIMIZING GAME DEVELOPMENT USING AMD'S GPU PERFSTUDIO 2

!  What’s new in GPS2.14? ‒ Hardware counter support for AMD “Hawaii” (R9 290 series) GPU’s ‒  Improved support for mulVthreaded applicaVons ‒ Pipeline specific counters for OpenGL ‒  Support for OpenGL Compute

LATEST VERSION

!  Currently in development ‒  Support for Linux®/OpenGL applicaVons ‒  Support for Mantle applicaVons

‒ There’s opVmizaVon, and then there’s Mantle. Games enabled with Mantle speak the language of GCN architecture to unlock revoluVonary performance and image quality.

GPU PerfStudio2 Demo


GPU PERFSTUDIO 2

!  GS-‐4136 -‐ OpVmizing Game Development using AMD’s GPU PerfStudio 2 Tue 4:00 -‐ 4:45 LL21E

!  Take a look at three usage scenarios from developers ‒ Assassins Creed 3 – Understanding the frame ‒  Far Cry 3 – Automated data mining of game content ‒  Steam Linux® – PorVng Source2 to OpenGL

!  Currently in development ‒  Support for Linux® (OpenGL) applicaVons ‒  Support for Mantle applicaVons

FIND OUT MORE

APU 2013


APU 2013 DEVELOPER TOOLS SESSIONS

!  PT-‐4052 IntroducVon to AMD Developer Tools Tue 10:00 -‐ 10:45

!  PT-‐4055 -‐ OpVmizing Raytracing on GCN with AMD Development Tools Tue 5:00 -‐ 5:45

!  PT-‐4053 -‐ Advanced OpenCL Debugging and Profiling using CodeXL Wed 11:00 -‐ 11:45

!  GS-‐4136 -‐ OpVmizing Game Development using AMD’s GPU PerfStudio 2 Tue 4:00 -‐ 4:45

!  Come and chat with us at the AMD Developer Tools Demo Booth

QuesVons

CONTACT US: [email protected] [email protected] DEVELOPER.AMD.COM


DISCLAIMER & ATTRIBUTION

The informaVon presented in this document is for informaVonal purposes only and may contain technical inaccuracies, omissions and typographical errors.

The informaVon contained herein is subject to change and may be rendered inaccurate for many reasons, including but not limited to product and roadmap changes, component and motherboard version changes, new model and/or product releases, product differences between differing manufacturers, soTware changes, BIOS flashes, firmware upgrades, or the like. AMD assumes no obligaVon to update or otherwise correct or revise this informaVon. However, AMD reserves the right to revise this informaVon and to make changes from Vme to Vme to the content hereof without obligaVon of AMD to noVfy any person of such revisions or changes.

AMD MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY INACCURACIES, ERRORS OR OMISSIONS THAT MAY APPEAR IN THIS INFORMATION.

AMD SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL AMD BE LIABLE TO ANY PERSON FOR ANY DIRECT, INDIRECT, SPECIAL OR OTHER CONSEQUENTIAL DAMAGES ARISING FROM THE USE OF ANY INFORMATION CONTAINED HEREIN, EVEN IF AMD IS EXPRESSLY ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.

ATTRIBUTION

© 2013 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo and combinaVons thereof are trademarks of Advanced Micro Devices, Inc. in the United States and/or other jurisdicVons. OpenCL is a trademark of Apple Inc. and Linux is a trademark of Linus Torvalds. Windows, MicrosoT , Visual Studio and DirectX are trademarks of MicrosoT Corp. Other names are for informaVonal purposes only and may be trademarks of their respecVve owners.

pt-4052, introduction to amd developer tools, by yaki tebeka and gordon selley

Technology

instrucvon

amd developer

opvmizing

converge tools

draw call

development

cache misses

stack logging