Intel® FPGA SDK for OpenCL™ Standard Edition: Best Practices Guide

Download PDF
ID 683176
Date 9/24/2018
Public
Document Table of Contents
Document Table of Contents x
1. Introduction to Standard Edition Best Practices Guide 2. Reviewing Your Kernel's report.html File 3. OpenCL Kernel Design Best Practices 4. Profiling Your Kernel to Identify Performance Bottlenecks 5. Strategies for Improving Single Work-Item Kernel Performance 6. Strategies for Improving NDRange Kernel Data Processing Efficiency 7. Strategies for Improving Memory Access Efficiency 8. Strategies for Optimizing FPGA Area Usage A. Additional Information
1. Introduction to Standard Edition Best Practices Guide x
1.1. FPGA Overview 1.2. Pipelines 1.3. Single Work-Item Kernel versus NDRange Kernel 1.4. Multi-Threaded Host Application
2. Reviewing Your Kernel's report.html File x
2.1. High Level Design Report Layout 2.2. Reviewing the Report Summary 2.3. Reviewing Loop Information 2.4. Reviewing Area Information 2.5. Verifying Information on Memory Replication and Stalls 2.6. Optimizing an OpenCL Design Example Based on Information in the HTML Report 2.7. HTML Report: Area Report Messages 2.8. HTML Report: Kernel Design Concepts
2.3. Reviewing Loop Information x
2.3.1. Loop Analysis Report of an OpenCL Design Example 2.3.2. Changing the Memory Access Pattern Example 2.3.3. Reducing the Area Consumed by Nested Loops Using loop_coalesce
2.4. Reviewing Area Information x
2.4.1. Area Analysis by Source 2.4.2. Area Analysis of System
2.5. Verifying Information on Memory Replication and Stalls x
2.5.1. Features of the System Viewer 2.5.2. Features of the Kernel Memory Viewer
2.7. HTML Report: Area Report Messages x
2.7.1. Area Report Message for Board Interface 2.7.2. Area Report Message for Function Overhead 2.7.3. Area Report Message for State 2.7.4. Area Report Message for Feedback 2.7.5. Area Report Message for Constant Memory 2.7.6. Area Report Messages for Private Variable Storage
2.8. HTML Report: Kernel Design Concepts x
2.8.1. Kernels 2.8.2. Global Memory Interconnect 2.8.3. Local Memory 2.8.4. Nested Loops 2.8.5. Loops in a Single Work-Item Kernel 2.8.6. Channels 2.8.7. Load-Store Units
3. OpenCL Kernel Design Best Practices x
3.1. Transferring Data Via Channels or OpenCL Pipes 3.2. Unrolling Loops 3.3. Optimizing Floating-Point Operations 3.4. Allocating Aligned Memory 3.5. Aligning a Struct with or without Padding 3.6. Maintaining Similar Structures for Vector Type Elements 3.7. Avoiding Pointer Aliasing 3.8. Avoid Expensive Functions 3.9. Avoiding Work-Item ID-Dependent Backward Branching
3.1. Transferring Data Via Channels or OpenCL Pipes x
3.1.1. Characteristics of Channels and Pipes 3.1.2. Execution Order for Channels and Pipes 3.1.3. Optimizing Buffer Inference for Channels or Pipes 3.1.4. Best Practices for Channels and Pipes
3.3. Optimizing Floating-Point Operations x
3.3.1. Floating-Point versus Fixed-Point Representations
4. Profiling Your Kernel to Identify Performance Bottlenecks x
4.1. Best Practices 4.2. GUI 4.3. Interpreting the Profiling Information 4.4. Limitations
4.2. GUI x
4.2.1. Source Code Tab 4.2.2. Kernel Execution Tab 4.2.3. Autorun Captures Tab
4.2.1. Source Code Tab x
4.2.1.1. Tool Tip Options
4.3. Interpreting the Profiling Information x
4.3.1. Stall, Occupancy, Bandwidth 4.3.2. Activity 4.3.3. Cache Hit 4.3.4. Profiler Analyses of Example OpenCL Design Scenarios 4.3.5. Autorun Profiler Data
4.3.1. Stall, Occupancy, Bandwidth x
4.3.1.1. Stalling Channels
4.3.4. Profiler Analyses of Example OpenCL Design Scenarios x
4.3.4.1. High Stall Percentage 4.3.4.2. Low Occupancy Percentage 4.3.4.3. Low Bandwidth Efficiency 4.3.4.4. High Stall and High Occupancy Percentages 4.3.4.5. No Stalls, Low Occupancy Percentage, and Low Bandwidth Efficiency 4.3.4.6. No Stalls, High Occupancy Percentage, and Low Bandwidth Efficiency 4.3.4.7. Stalling Channels 4.3.4.8. High Stall and Low Occupancy Percentages
5. Strategies for Improving Single Work-Item Kernel Performance x
5.1. Addressing Single Work-Item Kernel Dependencies Based on Optimization Report Feedback 5.2. Removing Loop-Carried Dependencies Caused by Accesses to Memory Arrays 5.3. Good Design Practices for Single Work-Item Kernel
5.1. Addressing Single Work-Item Kernel Dependencies Based on Optimization Report Feedback x
5.1.1. Removing Loop-Carried Dependency 5.1.2. Relaxing Loop-Carried Dependency 5.1.3. Simplifying Loop-Carried Dependency 5.1.4. Transferring Loop-Carried Dependency to Local Memory 5.1.5. Removing Loop-Carried Dependency by Inferring Shift Registers
6. Strategies for Improving NDRange Kernel Data Processing Efficiency x
6.1. Specifying a Maximum Work-Group Size or a Required Work-Group Size 6.2. Kernel Vectorization 6.3. Multiple Compute Units 6.4. Combination of Compute Unit Replication and Kernel SIMD Vectorization 6.5. Reviewing Kernel Properties and Loop Unroll Status in the HTML Report
6.2. Kernel Vectorization x
6.2.1. Static Memory Coalescing
6.3. Multiple Compute Units x
6.3.1. Compute Unit Replication versus Kernel SIMD Vectorization
7. Strategies for Improving Memory Access Efficiency x
7.1. General Guidelines on Optimizing Memory Accesses 7.2. Optimize Global Memory Accesses 7.3. Performing Kernel Computations Using Constant, Local or Private Memory 7.4. Improving Kernel Performance by Banking the Local Memory 7.5. Optimizing Accesses to Local Memory by Controlling the Memory Replication Factor 7.6. Minimizing the Memory Dependencies for Loop Pipelining
7.2. Optimize Global Memory Accesses x
7.2.1. Contiguous Memory Accesses 7.2.2. Manual Partitioning of Global Memory
7.2.2. Manual Partitioning of Global Memory x
7.2.2.1. Heterogeneous Memory Buffers
7.3. Performing Kernel Computations Using Constant, Local or Private Memory x
7.3.1. Constant Cache Memory 7.3.2. Preloading Data to Local Memory 7.3.3. Storing Variables and Arrays in Private Memory
7.4. Improving Kernel Performance by Banking the Local Memory x
7.4.1. Optimizing the Geometric Configuration of Local Memory Banks Based on Array Index
8. Strategies for Optimizing FPGA Area Usage x
8.1. Compilation Considerations 8.2. Board Variant Selection Considerations 8.3. Memory Access Considerations 8.4. Arithmetic Operation Considerations 8.5. Data Type Selection Considerations
A. Additional Information x
A.1. Document Revision History for the Standard Edition Best Practices Guide
1. Introduction to Standard Edition Best Practices Guide
2. Reviewing Your Kernel's report.html File
3. OpenCL Kernel Design Best Practices
4. Profiling Your Kernel to Identify Performance Bottlenecks
5. Strategies for Improving Single Work-Item Kernel Performance
6. Strategies for Improving NDRange Kernel Data Processing Efficiency
7. Strategies for Improving Memory Access Efficiency
8. Strategies for Optimizing FPGA Area Usage
A. Additional Information

A.1. Document Revision History for the Standard Edition Best Practices Guide

Document Version Intel® Quartus® Prime Version Changes
2018.09.24 18.1
  • Maintenance release
2018.05.04 18.0
  • Removed Pro Edition information.
  • In Preloading Data to Local Memory, added recommendation to create local array elements to be a power of 2 bytes.
  • Removed the topic Resource-Driven Optimization because it described an obsolete optimization behavior.
Table 12.   Best Practices Guide Document Revision History
Date Version Changes
December 2017 2017.12.08
November 2017 2017.11.06
May 2017 2017.05.08
December 2016 2016.12.02 Minor editorial modification.
October 2016 2016.10.31
  • Rebranded the Altera SDK for OpenCL to .
  • Rebranded the Altera Offline Compiler to .
  • In Align a Struct with or without Padding, modified code snippets to correct the placement of attributes with respect to the struct declaration.
  • Added the topic Review Your Kernel's report.html File, with subtopics describing the HTML GUI, the various reports the GUI provides, and a walkthrough on how to leverage the information in the HTML report to optimize an OpenCL design example.
  • Changed Review Your Kernel's Area Report to Identify Inefficiencies in Resource Usage to HTML Report: Area Report Messages, and removed the following subsections:
    • Area Report Messages for Global Memory and Global Memory Interconnect
    • Area Report Messages for Local Memory
    • Area Report Messages for Channels
  • Added the topic HTML Report: Kernel Design Concepts, which includes subtopics on kernels, global memory interconnect, local memory, nested loops, loops in single work-item kernels, and channels.
  • In Interpreting the Profiling Information, reorganized the content and added the following:
    • Additional explanations on stall, occupancy, bandwidth, activity, and cache hit.
    • Suggestions on addressing unsatisfactory Profiler metrics.
  • In Addressing Single Work-Item Kernel Dependencies Based On Optimization Report Feedback, modified the figure Optimization Work Flow of a Single Work-Item Kernel to replace area report with HTML report.
  • Removed the Optimization Report section along with the associated subsections because the information is now part of the HTML report.
  • Changed Review Kernel Properties and Loop Unroll Status in the Optimization Report to Review Kernel Properties and Loop Unroll Status in the HTML Report because the optimization report is now part of the report.html file.
May 2016 2016.05.02
  • Added the topic Removing Loop-Carried Dependencies Caused by Accesses to Memory Arrays to introduce the ivdep pragma.
  • Under Strategies for Improving Memory Access Efficiency, added the following topics to explain how to use the numbanks and bankwidth kernel attributes to configure the geometry of local memory system:
    • Improve Kernel Performance by Banking the Local Memory
    • Optimize the Geometric Configuration of Local Memory Banks Based on Array Index
  • Under Strategies for Improving Memory Access Efficiency, added the topic Optimize Accesses to Local Memory by Controlling the Memory Replication Factor to explain the usage of the singlepump and doublepump kernel attributes.
  • Added information on the area report messages. Refer to the Review Your Kernel's Area Report to Identify Inefficiencies in Resource Usage section for more information.
  • Removed the Kernel-Specific Area Report section because it is replaced by the enhanced area report. Refer to the Review Your Kernel's Area Report to Identify Inefficiencies in Resource Usage section for more information.
  • Updated the subsections under Optimization Report to include the enhanced optimization report messages.
    • Added the Optimization Report Message for Speed-Limiting Constructs
  • Updated the subsections under Addressing Single Work-Item Kernel Dependencies Based on Optimization Report Feedback to include the enhanced optimization report messages.
  • Updated the figure Optimization Work Flow for a Single Work-Item Kernel to include steps on accessing the enhanced area report to review resource usage.
  • Under Strategies for Improving NDRange Kernel Data Processing Efficiency, added the Review Kernel Properties and Loop Unroll Status in the Optimization Report section.
November 2015 2015.11.02
  • Added the topic Multi-Threaded Host Application.
  • Added Caution note regarding memory barrier in Specify a Maximum Work-Group Size or a Required Work-Group Size.
May 2015 15.0.0
  • In Memory Access Considerations, added Caution note regarding performance degradation that might occur when declaring __constant pointer arguments in kernels targeting Cyclone® V devices.
  • In Good Design Practices for Single Work-Item Kernel, removed the Initialize Data Prior to Usage in a Loop section and added a Declare Variables in the Deepest Scope Possible section.
  • Added Removing Loop-Carried Dependency by Inferring Shift Registers. The topic discusses how, in single work-item kernels, inferring double precision floating-point array as a shift register can remove loop-carried dependencies.
  • Added Kernel-Specific Area Reports to show examples of kernel-specific .area files that the Altera Offline Compiler generates during compilation.
  • Renamed Transfer Data Via offline compiler Channels to Transfer Data Via offline compiler Channels or OpenCL Pipes and added the following:
    • More information on how channels can help improve kernel performance.
    • Information on OpenCL pipes.
  • Renamed Data Type Considerations to Data Type Selection Considerations.
December 2014 14.1.0
  • Reorganized the information flow in the Optimization Report Messages section to update report messages and the layout of the optimization report.
  • Included new optimization report messages detailing the reasons for unsuccessful and suboptimal pipelined executions.
  • Added the Optimization Report Messages for Simplified Analysis of a Complex Design subsection under Optimization Report Messages to describe new report message for simplified kernel analysis.
  • Renamed Using Feedback from the Optimization Report to Address Single Work-Item Kernels Dependencies to Addressing Single Work-Item Kernel Dependencies Based on Optimization Report Feedback.
  • Added the Transferring Loop-Carried Dependency to Local Memory subsection under Addressing Single Work-Item Kernel Dependencies Based on Optimization Report Feedback to describe new strategy for resolving loop-carried dependency.
  • Updated the Resource-Driven Optimization and Compilation Considerations sections to reflect the deprecation of the -O3 and --util <N> Altera® Offline Compiler (offline compiler) command options.
  • Consolidated and simplified the Heterogeneous Memory Buffers and Host Application Modifications for Heterogeneous Memory Accesses sections.
  • Added the section Align a Struct and Remove Padding between Struct Fields.
  • Removed the section Ensure 4-Byte Alignment to All Data Structures.
  • Modified the figure Single Work-Item Optimization Work Flow to include emulation and profiling.
June 2014 14.0.0
  • Renamed document as the Best Practices Guide.
  • Reorganized information flow.
  • Renamed Good Design Practices to Good OpenCL Kernel Design Practices.
  • Added channels information in Transfer data via offline compilerL Channels.
  • Added profiler information in Profile Your Kernel to Identify Performance Bottlenecks.
  • Added the section Single Work-Item Kernel Versus NDRange Kernel.
  • Updated Single Work-Item Execution section.
  • Removed Performance Warning Messages section.
  • Renamed Single Work-Item Kernel Programming Considerations to Good Design Practices for Single Work-Item Kernel.
  • Added the section Strategies for Improving Single Work-Item Kernel Performance.
  • Renamed Optimization of Data Processing Efficiency to Strategies for Improving NDRange Kernel Data Processing Efficiency.
  • Removed Resource Sharing section.
  • Renamed Floating-Point Operations to Optimize Floating-Point Operations.
  • Renamed Optimization of Memory Access Efficiency to Strategies for Improving Memory Access Efficiency.
  • Updated Manual Partitioning of Global Memory section.
  • Added the section Strategies for Optimizing FPGA Area Usage.
December 2013 13.1.1
  • Updated the section Specify a Maximum Work-Group Size or a Required Work-Group Size.
  • Added the section Heterogeneous Memory Buffers.
  • Updated the section Single Work-Item Execution.
  • Added the section Performance Warning Messages.
  • Updated the section Single Work-Item Kernel Programming Considerations .
November 2013 13.1.0
  • Reorganized information flow.
  • Updated the section Compilation Flow.
  • Updated the section Pipelines; inserted the figure Example Multistage Pipeline Diagram.
  • Removed the following figures:
    • Instruction Flow through a Five-Stage Pipeline Processor.
    • Vector Addition Kernel Compiled to an FPGA.
    • Effect of Kernel Vectorization on Array Summation.
    • Data Flow Implementation of a Four-Element Accumulation Kernel.
    • Data Flow Implementation of a Four-Element Accumulation Kernel with Loop Unrolled.
    • Complete Loop Unrolling.
    • Unrolling Two Loop Iterations.
    • Memory Master Interconnect.
    • Local Memory Read and Write Ports.
    • Local Memory Configuration.
  • Updated the section Good Design Practices.
  • Removed the following sections:
    • Predicated Execution.
    • Throughput Analysis.
    • Case Studies.
  • Updated and renamed Optimizing Data Processing Efficiency to Optimization of Data Processing Efficiency.
  • Renamed Replicating Compute Units versus Kernel SIMD Vectorization to Compute Unit Replication versus Kernel SIMD Vectorization.
  • Renamed Using num_compute_units and num_simd_work_items Together to Combination of Compute Unit Replication and Kernel SIMD Vectorization.
  • Updated and renamed Memory Streaming to Contiguous Memory Accesses.
  • Updated and renamed Optimizing Memory Access to General Guidelines on Optimizing Memory Accesses.
  • Updated and renamed Optimizing Memory Efficiency to Optimization of Memory Access Efficiency.
  • Inserted the subsection Single Work-Item Execution under Optimization of Memory Access Efficiency.
June 2013 13.0 SP1.0
  • Updated support status of OpenCL kernel source code containing complex exit paths.
  • Updated the figure Effect of Kernel Vectorization on Array Summation to correct the data flow between Store and Global Memory.
  • Updated content for the unroll pragma directive in the section Loop Unrolling.
  • Updated content of the Local Memory section.
  • Updated the figure Local Memories Transferring Data Blocks within Matrices A and B to correct the data transfer pattern in Matrix B.
  • Removed the figure Loop Unrolling with Vectorization.
  • Removed the section Optimizing Local Memory Bandwidth.
May 2013 13.0.1
  • Updated terminology. For example, pipeline is replaced with compute unit; vector lane is replaced with SIMD vector lane.
  • Added the following sections under Good Design Practices:
    • Preprocessor Macros.
    • Floating-Point versus Fixed-Point Representations.
    • Recommended Optimization Methodology.
    • Sequence of Optimization Techniques.
  • Updated code fragments.
  • Updated the figure Data Flow with Multiple Compute Units.
  • Updated the figure Compute Unit Replication versus Kernel SIMD Vectorization.
  • Updated the figure Optimizing Throughput Using Compute Unit Replication and SIMD Vectorization.
  • Updated the figure Memory Streaming.
  • Inserted the figure Local Memories Transferring Data Blocks within Matrices A and B.
  • Reorganized the flow of information. Number of figures, tables, and examples have been updated.
  • Included information on new kernel attributes: max_share_resources and num_share_resources .
May 2013 13.0.0
  • Updated pipeline discussion.
  • Updated case study code examples and results tables.
  • Updated figures.
November 2012 12.1.0 Initial release.
Level Two Title