Developer Guide

oneAPI GPU Optimization Guide

Download PDF
ID 771772
Date 4/18/2024
Public
Document Table of Contents
Document Table of Contents x
oneAPI GPU Optimization Guide
oneAPI GPU Optimization Guide x
Introduction Getting Started Parallelization Intel® Xe GPU Architecture General-Purpose Computing on GPU Media Graphics Computing on GPU References Terms and Conditions
General-Purpose Computing on GPU x
Execution Model Overview Thread Mapping and GPU Occupancy Kernels Using Libraries for GPU Offload Host/Device Memory, Buffer and USM Unified Shared Memory Allocations Performance Impact of USM and Buffers Avoiding Moving Data Back and Forth between Host and Device Optimizing Data Transfers Avoiding Declaring Buffers in a Loop Buffer Accessor Modes Host/Device Coordination Using Multiple Heterogeneous Devices Compilation OpenMP Offloading Tuning Guide Multi-GPU and Multi-Stack Architecture and Programming Level Zero Performance Profiling and Analysis Configuring GPU Device
Kernels x
Sub-Groups and SIMD Vectorization Removing Conditional Checks Registers and Performance Shared Local Memory Pointer Aliasing and the Restrict Directive Synchronization among Threads in a Kernel Considerations for Selecting Work-Group Size Prefetch Reduction Kernel Launch Executing Multiple Kernels on the Device at the Same Time Submitting Kernels to Multiple Queues Avoiding Redundant Queue Constructions Programming Intel® XMX Using SYCL Joint Matrix Extension Doing I/O in the Kernel
Registers and Performance x
Finding Kernels with Register Spills Small Register Mode vs. Large Register Mode Optimizing Register Spills Porting Code with High Register Pressure to Intel® Max GPUs
Synchronization among Threads in a Kernel x
Atomic Operations Local Barriers vs Global Atomics
Using Libraries for GPU Offload x
Using Performance Libraries Using Standard Library Functions in SYCL Kernels Efficiently Implementing Fourier Correlation Using oneAPI Math Kernel Library (oneMKL) Boost Matrix Multiplication Performance with Intel® Xe Matrix Extensions
Host/Device Coordination x
Asynchronous and Overlapping Data Transfers Between Host and Device
Compilation x
Just-In-Time Compilation Ahead-Of-Time Compilation Specialization Constants Accuracy Versus Performance Tradeoffs in Floating-Point Computations
OpenMP Offloading Tuning Guide x
OpenMP Directives OpenMP Execution Model Terminology Compiling and Running an OpenMP Application Offloading oneMKL Computations onto the GPU Tools for Analyzing Performance of OpenMP Applications OpenMP Offload Best Practices
OpenMP Offload Best Practices x
Using More GPU Resources Minimizing Data Transfers and Memory Allocations Making Better Use of OpenMP Constructs Memory Allocation Fortran Example Clauses: is_device_ptr, use_device_ptr, has_device_addr, use_device_addr Prefetching
Multi-GPU and Multi-Stack Architecture and Programming x
Multi-Stack GPU Architecture Exposing the Device Hierarchy FLAT Mode Programming COMPOSITE Mode Programming Using Intel® oneAPI Math Kernel Library (oneMKL) Using Intel® MPI Library Advanced Topics Terminology
FLAT Mode Programming x
FLAT Mode Example - SYCL FLAT Mode Example - OpenMP
COMPOSITE Mode Programming x
Explicit Scaling Implicit Scaling
Explicit Scaling x
Explicit Scaling - SYCL Explicit Scaling - OpenMP Explicit Scaling Summary
Implicit Scaling x
Introduction Work Scheduling and Memory Distribution STREAM Example Programming Principles
Using Intel® oneAPI Math Kernel Library (oneMKL) x
Scaling Performance with Intel® oneAPI Math Kernel Library(oneMKL) in SYCL Scaling Performance with Intel® oneAPI Math Kernel Library(oneMKL) in OpenMP
Using Intel® MPI Library x
Running MPI Applications on GPUs Intra-Device and Inter-Device Data Transfers for MPI+OpenMP Programs
Level Zero x
Immediate Command Lists
Performance Profiling and Analysis x
Using the Timers Intel® VTuneTM Profiler Intel® Advisor Intel® Intercept Layer for OpenCLTM Applications Performance Tools in Intel® Profiling Tools Interfaces for GPU
Intel® VTuneTM Profiler x
Hardware-assisted Stall Sampling Hardware-assisted Stall Sampling
Intel® Advisor x
Identify Regions to Offload to GPU with Offload Modeling Run a GPU Roofline Analysis Optimize Memory-bound Applications with GPU Roofline
Media Graphics Computing on GPU x
Optimizing Media Pipelines Performance Analysis with Intel® Graphics Performance Analyzers
Optimizing Media Pipelines x
Media Engine Hardware Media API Options for Hardware Acceleration Media Pipeline Parallelism Media Pipeline Inter-operation and Memory Sharing SYCL-Blur Example
oneAPI GPU Optimization Guide

Avoiding Moving Data Back and Forth between Host and Device

Level Two Title