Developer Guide

Intel® oneAPI DPC++/C++ Compiler Handbook for FPGAs

Download PDF
ID 785441
Date 6/24/2024
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents
Document Table of Contents x
Intel oneAPI DPC++/C++ Compiler Handbook for FPGAs Overview Introduction To FPGA Design Concepts Intel oneAPI FPGA Development Getting Started with the Intel oneAPI DPC++/C++ Compiler for Intel FPGA Development Defining a Kernel for FPGAs Debugging and Verifying Your Design Analyzing Your Design Optimizing Your Kernel Optimizing Your Host Application Integrating Your Kernel into DSP Builder for Intel FPGAs Integrating Your RTL IP Core Into a System RTL IP Core Kernel Interfaces Loops Pipes Data Types and Arithmetic Operations Parallelism Memories and Memory Operations Libraries Additional FPGA Acceleration Flow Considerations FPGA Optimization Flags, Attributes, Pragmas, and Extensions Quick Reference Additional Information Document Revision History for the Intel oneAPI DPC++/C++ Compiler Handbook for Intel FPGAs Notices and Disclaimers
Introduction To FPGA Design Concepts x
FPGA Architecture Overview Concepts of FPGA Hardware Design How Source Code Becomes a Custom Hardware Datapath
How Source Code Becomes a Custom Hardware Datapath x
Mapping Source Code Instructions to Hardware Scheduling Mapping Parallelism Models to FPGA Hardware Memory Types
Intel oneAPI FPGA Development x
FPGA Flow Terminology Intel oneAPI FPGA Development Flow Types of SYCL* FPGA Compilation FPGA Compilation Flags FPGA Workflows in IDEs
Types of SYCL* FPGA Compilation x
Separating Device and Host Code Compilation
Getting Started with the Intel oneAPI DPC++/C++ Compiler for Intel FPGA Development x
Installing the Intel oneAPI FPGA Development Environment FPGA Development for Intel oneAPI Toolkits with Visual Studio* Code
Installing the Intel oneAPI FPGA Development Environment x
Simulation Prerequisites Installing the Questa*-Intel FPGA Edition Software Set Up the Simulation Environment
FPGA Development for Intel oneAPI Toolkits with Visual Studio* Code x
Set the Environment Variables and Launch Visual Studio* Code Create an FPGA Visual Studio* Code Project Enable Code Completion in a Visual Studio* Code Project Configure Running and Debugging in a Visual Studio* Code Project Debugging Your Kernel in Visual Studio* Code with a Native Debugger Generate and View the FPGA Optimization Report Build and Run the FPGA Hardware Image
Defining a Kernel for FPGAs x
Suggested Kernel Coding Styles Single Work-Item Kernels
Single Work-Item Kernels x
Single Work-item Kernel Design Guidelines Single-Cycle Floating-Point Accumulator for Single Work-Item Kernels
Single-Cycle Floating-Point Accumulator for Single Work-Item Kernels x
Strategies for Inferring the Accumulator
Debugging and Verifying Your Design x
Device Selectors for FPGA printf Command Restrictions Emulate and Debug Your Design Evaluate Your Kernel Through Simulation
Emulate and Debug Your Design x
Emulator Environment Variables Compile and Emulate Your Design Limitations of the Emulator Troubleshooting Discrepancies in Hardware and Emulator Results Emulator Known Issues
Evaluate Your Kernel Through Simulation x
Debug During Verification Compile a Kernel for Simulation Simulate Your Kernel Viewing Simulation Waveforms Troubleshooting Simulator Issues
Analyzing Your Design x
Review the FPGA Optimization Report Quartus (Static) Summary Intel® FPGA Dynamic Profiler for DPC++
Review the FPGA Optimization Report x
Loop Analysis Bottlenecks Viewer Area Estimates System Viewer Kernel Memory Viewer Schedule Viewer Access FPGA Optimization Reports in JSON Format
Quartus (Static) Summary x
Timing Failures
Intel® FPGA Dynamic Profiler for DPC++ x
Measure Kernel Performance Instrument the Kernel Pipeline with Performance Counters (-Xsprofile) Obtain Profiling Data During Run Time Reduce Area Resource Use While Profiling Profiler Analyses of Example SYCL* Design Scenarios Limitations
Obtain Profiling Data During Run Time x
Invoke the Profiler Runtime Wrapper to Obtain Profiling Data Use Intel® VTune™ Profiler
Use Intel® VTune™ Profiler x
Interpret Performance Counter Data
Optimizing Your Host Application x
Throughput Resource Use System-level Profiling Using the Intercept Layer for OpenCL™ Applications Multithreaded Host Application Utilizing Hardware Kernel Invocation Queue Double Buffering Host Utilizing Kernel Invocation Queue N-Way Buffering to Overlap Kernel Execution Prepinning Memory Simple Host-Device Streaming Buffered Host-Device Streaming
System-level Profiling Using the Intercept Layer for OpenCL™ Applications x
Set Up the Intercept Layer for OpenCL™ Applications
Double Buffering Host Utilizing Kernel Invocation Queue x
Analyzing Buffering Using the Intercept Layer for OpenCL™ Applications
Integrating Your RTL IP Core Into a System x
Synthesize Your RTL IP Core with Quartus Prime Software Encrypting RTL IP Cores Add an RTL IP Core into a Platform Designer System Add an RTL IP Core into a Quartus Prime Project
Encrypting RTL IP Cores x
Encrypting RTL IP Cores Without Licensing Encrypting RTL IP Core With Licensing
RTL IP Core Kernel Interfaces x
Kernel Argument Interfaces Memory-Mapped (MM) Agent Kernel Invocation Interface Ready/Valid Handshaking Kernel Invocation Interface Memory-Mapped Host Interfaces Structs in RTL IP Core Interfaces IP core Reset Behavior
Memory-Mapped Host Interfaces x
Memory-Mapped Host Interfaces Using Unified Shared Memory Pointers and the annotated_arg Class Memory-Mapped Host Interfaces Using Accessors
Memory-Mapped Host Interfaces Using Unified Shared Memory Pointers and the annotated_arg Class x
Buffer Locations in Memory-Mapped Host Interfaces The annotated_arg Template Class
Loops x
Refactor the Loop-Carried Data Dependency Relax Loop-Carried Dependency Transfer Loop-Carried Dependency to Local Memory Minimize the Memory Dependencies for Loop Pipelining Unroll Loops Fuse Loops to Reduce Overhead and Improve Performance Optimize Loops With Loop Speculation Remove Loop Bottlenecks Improve fMAX/II with Shannonization Optimize Inner Loop Throughput Improve Loop Performance by Caching Data in On-Chip Memory
Pipes x
Host Pipes Pipes Extension
Host Pipes x
Host Pipe Declaration Host Pipe API Host Pipes RTL Interfaces
Pipes Extension x
Key Properties of a Pipe Accessing Pipes The pipe Class and its Use I/O Pipes Characteristics of Pipes Restrictions of Pipes Guidelines for Designing Pipes The atomic_fence Function and Pipes
I/O Pipes x
Emulate Applications with a Pipe That Reads or Writes to an I/O Pipe
Data Types and Arithmetic Operations x
Optimize Floating-point Operation Minimize the Use of Expensive Functions Variable-Precision Data Type Support
Variable-Precision Data Type Support x
Advantages and Limitations of Variable Precision Data Types Declare and Use the AC Data Types
Declare and Use the AC Data Types x
Declare the ac_int Data Type Declare the ac_fixed Data Type Declare the ac_complex Data Type Declare the ap_float Data Type
Declare the ap_float Data Type x
Conversion Rules for the ap_float Data Type Operations with Explicit Precision Controls Comparison Operators Additional ap_float Functions Additional Data Types Provided by the ap_float.hpp Header File Quality of Results and the ap_float Data Type
Parallelism x
Pipelined Kernels NDRange Kernels Asynchronous Parallelism Within Kernels (task_sequence)
Pipelined Kernels x
Stable Arguments
Asynchronous Parallelism Within Kernels (task_sequence) x
Task Functions task_sequence Use Cases
Memories and Memory Operations x
The device_global Extension (Beta) The annotated_ptr Template Class (Beta) Memory Accesses Kernel Variable Accesses
Memory Accesses x
Load-Store Units Global Memory Accesses Optimization Perform Kernel Computations Using Local or Private Memory Local and Private Memory Accesses Optimization Annotating Unified Shared Memory Pointers Zero-Copy Memory Access Additional Recommendations
Load-Store Units x
Load-Store Unit Styles Load-Store Unit Modifiers Load-Store Unit Controls
Global Memory Accesses Optimization x
Global Memory Bandwidth Use Calculation Manual Partition of Global Memory Partitioning Buffers Across Different Memory Types (Heterogeneous Memory) Partitioning Buffers Across Memory Channels of the Same Memory Type Ignoring Dependencies Between Accessor Arguments Contiguous Memory Accesses Static Memory Coalescing
Libraries x
Use SYCL Shared Library With Third-Party Applications Use of RTL Libraries for FPGA Object Manifest File Syntax of an RTL Library Restrictions and Limitations in RTL Support Agilex 7 and Stratix 10 Design-Specific Reset Requirements for Stall-Free and Stallable RTL Libraries Stall-Free RTL
Additional FPGA Acceleration Flow Considerations x
FPGA-CPU Interaction FPGA Boards and Board Support Packages (BSPs) Targeting Multiple Identical FPGA Devices Targeting Multiple Platforms Split Kernel into Multiple FPGA Images (Linux only)
FPGA Boards and Board Support Packages (BSPs) x
The FPGA Board Utility Commands Managing an FPGA Board Advanced Board Management The OpenCL™ Installable Client Driver (ICD)
Advanced Board Management x
Extracting the FPGA Device Image (.aocx) File from a Multiarchitecture Binary File
FPGA Optimization Flags, Attributes, Pragmas, and Extensions x
Optimization Flags Optimization Targets Kernel Variables Kernel Attributes Memory Attributes Loop Directives Floating-Point Pragmas Latency Controls (Beta)
Optimization Flags x
Specify Schedule fMAX Target for Kernels (-Xsclock=<clock target>) Create a 2xclock Interface (-Xsuse-2xclock) Disable Burst-Interleaving of Global Memory (-Xsno-interleaving) Force Ring Interconnect for Global Memory (-Xsglobal-ring) Force a Single Store Ring to Reduce Area (-Xsforce-single-store-ring) Force Fewer Read Data Reorder Units to Reduce Area (-Xsnum-reorder) Disable Hardware Kernel Invocation Queue (-Xsno-hardware-kernel-invocation-queue) Modify the Handshaking Protocol Between Clusters (-Xshyper-optimized-handshaking) Disable Automatic Fusion of Loops (-Xsdisable-auto-loop-fusion) Fuse Adjacent Loops With Unequal Trip Counts (-Xsenable-unequal-tc-fusion) Pipeline Loops in Non-task Kernels (-Xsauto-pipeline) Control Semantics of Floating-Point Operations (-fp-model=<value>) Modify the Rounding Mode of Floating-point Operations (-Xsrounding=<rounding_type>) Global Control of Exit FIFO Latency of Stall-free Clusters (-Xssfc-exit-fifo-type=<value>) Enable the Read-Only Cache for Read-Only Accessors (-Xsread-only-cache-size=<N>) Control Hardware Implementation of the Supported Data Types and Math Operations (-Xsdsp-mode=<option>) Generate Register Map Wrapper (-Xsregister-map-wrapper-type) Allow Wide Memory Initialization (-Xsallow-wide-device-globals)
Optimization Targets x
Minimum Latency Flow Minimum Area Flow Balanced Throughput-Area Trade-Offs Flow
Kernel Attributes x
Specify Schedule fMAX Target for Kernels (scheduler_target_fmax_mhz) Specify a Workgroup Size (max_work_group_size/reqd_work_group_size) Specify Number of SIMD Work Items (num_simd_work_items) Omit Hardware that Generates and Dispatches Kernel IDs (max_global_work_dim) Omit Hardware that Supports Global Work Offsets (no_global_work_offset) Reduce Kernel Area and Latency (use_stall_enable_clusters)
Loop Directives x
disable_loop_pipelining Attribute initiation_interval Attribute ivdep Attribute loop_coalesce Attribute max_concurrency Attribute max_interleaving Attribute speculated_iterations Attribute unroll Pragma Loop Fuse Functions and nofusion Attribute max_reinvocation_delay Attribute (Beta)
Quick Reference x
Algorithmic C Data Types Floating Point Pragmas FPGA Accessor Properties FPGA Extensions FPGA Kernel Attributes FPGA Local Memory Function Latency Control Properties (Beta) FPGA LSU Controls FPGA Loop Directives FPGA Memory Attributes FPGA Optimization Flags Pipe API
Intel oneAPI DPC++/C++ Compiler Handbook for FPGAs Overview
Introduction To FPGA Design Concepts
Intel oneAPI FPGA Development
Getting Started with the Intel oneAPI DPC++/C++ Compiler for Intel FPGA Development
Defining a Kernel for FPGAs
Debugging and Verifying Your Design
Analyzing Your Design
Optimizing Your Kernel
Optimizing Your Host Application
Integrating Your Kernel into DSP Builder for Intel FPGAs
Integrating Your RTL IP Core Into a System
RTL IP Core Kernel Interfaces
Loops
Pipes
Data Types and Arithmetic Operations
Parallelism
Memories and Memory Operations
Libraries
Additional FPGA Acceleration Flow Considerations
FPGA Optimization Flags, Attributes, Pragmas, and Extensions
Quick Reference
Additional Information
Document Revision History for the Intel oneAPI DPC++/C++ Compiler Handbook for Intel FPGAs
Notices and Disclaimers

Simple Host-Device Streaming

You can take advantage of SYCL Unified Shared Memory host allocation and zero-copy host memory to implement a streaming host-device design with low latency and high throughput.

TIP:
If your BSP supports host pipes, use them instead of simple host-device streaming. For information about host pipes, refer to Host Pipes.

TIP:

Before you begin learning how to stream data, review Pipes and Zero-Copy Memory Access concepts.

NOTE:
SYCL USM host allocations are only supported by some BSPs. Check with your BSP vendor to see if they support SYCL USM host allocations.

The OFS Intel® oneAPI Accelerator Support Package (ASP) supports USM. For details, refer https://github.com/OFS/oneapi-asp/releases.

To understand the concept, consider the following design example:

Offload Processing

Typical SYCL designs perform offload processing. All of the input data is prepared by the CPU, and then transferred to an FPGA device. Kernels are started on the device to process the data. When the kernels finish, the CPU copies the output data from the FPGA back to its memory. Memory synchronization is achieved on the host after the device kernel signals its completion.

Offload processing achieves excellent throughput when the memory transfers and kernel computation are performed on large data sets, as the CPU's kernel management overhead is minimized. You can conceal data transfer overhead using double buffering or N-way buffering to maximize kernel throughput. However, a significant shortcoming of this design pattern is latency. The coarse-grain synchronization of waiting for the entire set of data to be processed results in a latency that is equal to the processing time of the entire data.

Host-device Streaming Processing

The method for achieving lower latency between the host and device is to break data set into smaller chunks and, instead of enqueueing a single long-running kernel, launch a set of shorter-running kernels. Together, these shorter-running kernels process the data in smaller batches. As memory synchronization occurs upon kernel completion, this strategy makes the output data available to the CPU in a more granular way. This is illustrated in the following figure, where the orange lines illustrate the time when the first set of data is available in the host:

Host-device Streaming Processing

In the streaming version, the first piece of data is available in the host earlier than in the offload version. To understand how much earlier, consider that you have total_size elements of data to process and you break the computation into chunks chunks of size chunk_size=total_size/chunks (as is the case in Figure 1). Then, in an ideal situation, the streaming design achieves a latency that is chunks times better than the offload version.

Setting the chunk_size

You might wonder if you can set the chunk_size to 1 (for example, chunks=total_size) to minimize the latency. In the figure above, you may notice small gaps between the kernels in the streaming design (for example, between K0 and K1). This is caused by the overhead of launching kernels and detecting kernel completion on the host. These gaps increase the total processing time and therefore, decrease the throughput of the design (that is, when compared to the offload design, it takes more time to process the same amount of data). If these gaps are negligible, then the throughput is negligibly affected.

In the streaming design, the choice of the chunk_size is thus a tradeoff between latency (a smaller chunk size results in a smaller latency) and throughput (a smaller chunk size increases the relevance of the inter-kernel latency).

Lower Bounds on Latency

Lowering the chunk_size can reduce latency, sometimes at the expense of throughput. However, even if you are not concerned with throughput, there still exists a lower-bound on the latency of a kernel. Review the following figure:

Lower Bounds on Latency

In the Figure 2:

  • tlaunch is the time for a kernel launch signal to go from the host to the device.
  • tkernel is the time for the kernel to execute on the device.
  • tfinish is the time for the finished signal to go from the device to the host.

Even if you set chunk_size to 0 (for example, launch a kernel that does nothing) and therefore tkernel ~= 0, the latency is still tlaunch + tfinish. In other words, the lower bound on kernel latency is the time needed for the start signal to go from the host to the device and for the finished signal to go from the device to the host.

In the Setting the chunk_size section above, you learned how gaps between kernel invocations can degrade throughput. In the Figure 2, there appears to be a minimum tlaunch + tfinish gap between kernel invocations. This is illustrated as the Naive Relaunch timeline in the following figure:

Naive Relaunch Timeline

Fortunately, this overhead is circumvented by an automatic runtime kernel launch scheme that buffers kernel arguments on the device before the previous kernel finishes. This enables kernels to queue on the device and to begin execution without waiting for the previous kernel's finished to propagate back to the host. You can refer to this as Fast Kernel Relaunch, and it is also illustrated in the Figure 3. Fast kernel relaunch reduces the gap between kernel invocations and allows you to achieve lower latency while maintaining throughput.

Multiple Kernel Pipeline

More complicated FPGA designs often instantiate multiple kernels connected by SYCL pipes (for examples, see FPGA Reference Designs). Suppose you have a kernel system of N kernels connected by pipes, as shown in the following figure:
Multiple Kernels Connected by Pipes

With the goal of achieving lower latency, you use the Host-device Streaming Processing technique to launch multiple invocations of your N kernels to process chunks of data. This gives you a timeline as shown in the following figure:

Multiple Invocation of Kernels to Process Chunks of Data

Notice the gaps between the start times of the N kernels for a single chunk. This is the tlaunch time discussed in the Host-device Streaming Processing. However, the multi-kernel design introduces a potential new lower bound on the latency for a single chunk because processing a single chunk of data requires launching N kernels, which takes N x tlaunch. If N (the number of kernels in your system) is sufficiently large, this limits your achievable latency.

For designs with N > 2, Intel recommends a different approach. The idea is to enqueue your system of N kernels once, and to introduce Producer (P) and Consumer (C) kernels to handle the production and consumption of data from and to the host, respectively. This method is illustrated in the following figure:

Enqueuing a System of Kernels

The Producer streams data from the host and presents it to the kernel system through a SYCL pipe. The output of the kernel system is consumed by the Consumer and written back into host memory. To achieve low latency, you still process the data in chunks, but instead of having to enqueue N kernels for each chunk, you must only enqueue a single Producer and Consumer kernel per chunk. This enables you to reduce the lower bound on the latency to MAX(2 x tlaunch, tlaunch + tfinish). Observe that this lower bound does not depend on the number of kernels in your system (N).

Use this method only when N > 2. FPGA area is sacrificed to implement the Producer, Consumer, and their pipes to achieve lower overall processing latency.


Limitations

Fundamentally, the ability to stream data between the host and device is built around SYCL USM host allocations. The underlying problem is how to efficiently synchronize between the host and device to signal that some data is ready to be processed, or has been processed. In other words, how does the host signal to the device that some data is ready to be processed? Conversely, how does the device signal to the host that some data is done being processed?

One method to achieve this signaling is to use the start of a kernel to signal to the device that data is ready to be processed, and the end of a kernel to signal to the host that data has been processed. This is the approach taken in the examples discussed above. However, this method has two notable drawbacks. First, the latency to start and end kernels is high. To maintain high throughput, you must size the chunk_size sufficiently large to hide the inter-kernel latency, resulting in a latency increase. Second, the programming model to achieve this performance is non-trivial as you must intelligently manage the SYCL device queue.

NOTE:

For additional information, refer to the FPGA tutorial sample "Host-Device Streaming using USM" on GitHub.

Parent topic: Optimizing Your Host Application
Level Two Title