Developer Guide

FPGA Optimization Guide for Intel® oneAPI Toolkits

Download PDF
ID 767853
Date 3/31/2023
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A newer version of this document is available.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
FPGA Optimization Guide for Intel® oneAPI Toolkits Introduction To FPGA Design Concepts Analyze Your Design Optimize Your Design FPGA Optimization Flags, Attributes, Pragmas, and Extensions Quick Reference Additional Information Document Revision History for the FPGA Optimization Guide for Intel® oneAPI Toolkits
Introduction To FPGA Design Concepts x
FPGA Architecture Overview Concepts of FPGA Hardware Design Methods of Hardware Design How Source Code Becomes a Custom Hardware Datapath Scheduling Mapping Parallelism Models to FPGA Hardware Memory Types
FPGA Architecture Overview x
Adaptive Logic Module (ALM) Lookup Table (LUT) Register Digital Signal Processing (DSP) Block Random Access Memory (RAM) Blocks
Concepts of FPGA Hardware Design x
Maximum Frequency (fMAX) Latency Pipelining Throughput Datapath Control Path Occupancy
How Source Code Becomes a Custom Hardware Datapath x
Mapping Source Code Instructions to Hardware Mapping Arrays and Their Accesses to Hardware
Scheduling x
Dynamic Scheduling Clustering the Datapath Handshaking Between Clusters
Mapping Parallelism Models to FPGA Hardware x
Data Parallelism Task Parallelism
Data Parallelism x
Executing Independent Operations Simultaneously Pipelining
Memory Types x
Kernel Memory Global Memory
Analyze Your Design x
Analyze the FPGA Early Image Analyze the FPGA Image
Analyze the FPGA Early Image x
Review the FPGA Optimization Report Access HLD FPGA Reports in JSON Format
Review the FPGA Optimization Report x
Loop Analysis Bottlenecks Viewer Area Estimates System Viewer Kernel Memory Viewer Schedule Viewer
Analyze the FPGA Image x
Quartus (Static) Summary Intel® FPGA Dynamic Profiler for DPC++ System-level Profiling Using the Intercept Layer for OpenCL* Applications
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 Runtime Reduce Area Resource Use While Profiling Profiler Analyses of Example SYCL* Design Scenarios Limitations
Obtain Profiling Data During Runtime x
Invoke the Profiler Runtime Wrapper to Obtain Profiling Data Use Intel® VTune™ Profiler
Use Intel® VTune™ Profiler x
Interpret Performance Counter Data
System-level Profiling Using the Intercept Layer for OpenCL* Applications x
Set Up the Intercept Layer for OpenCL* Applications
Optimize Your Design x
Throughput Resource Use
Throughput x
Single Work-item Kernels NDRange Kernels Memory Accesses Pipes Host
Single Work-item Kernels x
Single Work-item Kernel Design Guidelines Loops Single-Cycle Floating-Point Accumulator for Single Work-Item Kernels
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 Shannonization to Improve FMAX/II Optimize Inner Loop Throughput Improve Loop Performance by Caching On-Chip Memory
Single-Cycle Floating-Point Accumulator for Single Work-Item Kernels x
Strategies for Inferring the Accumulator
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
Host x
Multi-Threaded 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
Double Buffering Host Utilizing Kernel Invocation Queue x
Applying Double-Buffering Using the Intercept Layer for OpenCL* Applications
Resource Use x
Data Types and Operations Kernel Variable Accesses
Data Types and Operations x
Optimize Floating-point Operation Avoid Expensive Functions Variable-Precision Integer and Floating-Point Support
Variable-Precision Integer and Floating-Point Support x
Advantages and Limitations of Arbitrary 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 ap_float 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
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) FPGA Extensions
Optimization Flags x
Specify Schedule FMAX Target for Kernels (-Xsclock=<clock target>) Disable Burst-Interleaving of Global Memory (-Xsno-interleaving=<global_memory_type>) 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>)
Optimization Targets x
Minimum Latency Flow
Kernel Attributes x
Specify Schedule FMAX Target for Kernels Specify a Workgroup Size Specify Number of SIMD Work Items Omit Hardware that Generates and Dispatches Kernel IDs Omit Hardware to Support the no_global_work_offset Attribute in parallel_for Kernels Reduce Kernel Area and Latency
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)
FPGA Extensions x
Pipes Extension System of Tasks Extension (task_sequence) device_global Extension (Beta)
Pipes Extension x
Key Properties of a Pipe Accessing Pipes The pipe Class and its Use I/O Pipes I/O Pipes Implementation The I/O Pipe Classes and Their Use Example Code for I/O Pipes Using I/O Pipes to Stream Data Faking I/O Pipes Characteristics of Pipes Restrictions of Pipes Guidelines for Designing Pipes Pipe and Atomic Fence
System of Tasks Extension (task_sequence) x
Task Functions task_sequence Use Cases
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
Document Revision History for the FPGA Optimization Guide for Intel® oneAPI Toolkits x
Notices and Disclaimers
FPGA Optimization Guide for Intel® oneAPI Toolkits
Introduction To FPGA Design Concepts
Analyze Your Design
Optimize Your Design
FPGA Optimization Flags, Attributes, Pragmas, and Extensions
Quick Reference
Additional Information
Document Revision History for the FPGA Optimization Guide for Intel® oneAPI Toolkits

I/O Pipes

An I/O pipe is a unidirectional (source or sink) connection to the hardware that may be connected to input or output features of an FPGA board. These features might include network interfaces, PCIe®, cameras, or other data capture or processing devices or protocols.

A source I/O device provides data that a SYCL kernel can read whereas a sink I/O device accepts data written by a SYCL kernel and sends it to the hardware device. A common use of I/O pipes is to interface to ethernet connections that interface directly with the FPGA. The source reads from the network and the sink writes to the network. This allows a SYCL kernel to process data from the network and resend it back to the network.

For testing purposes, you can use files on the disk as input or output devices, allowing you to debug using emulation or simulation for faster compilation.

I/O Pipes Implementation

To implement I/O pipes, follow these steps:

  1. Consult your board vendor documentation before you implement I/O pipes in your kernel program.
  2. Use a struct with a numeric id to define a kernel_readable_io_pipe or kernel_writable_io_pipe type to declare an I/O pipe to interface with hardware peripherals. These definitions are typically provided by a board vendor.
    • The numeric id value is the 0-origin index into the interfaces in the channels section of the board_spec.xml for the device.
    • The chan_id argument is necessary for simulation and it is the name of the I/O interface listed in the board_spec.xml file as shown in the following: 
       <channels>
   <interface name="board" port="c1"  type="streamsource" width="32" chan_id="c1"/>
   <interface name="board" port="c2"  type="streamsink" width="32" chan_id="c2"/>
 </channels>
    NOTE:

    Only channels marked type streamsource or streamsink are used for indexing.

  3. Implement the interface to hardware I/O pipes or files (emulator or simulator) by mapping the id variable, as shown in the following: 
    // Specialize a pipe type
struct read_io_pipe {
  static constexpr unsigned id = 0;
};

struct write_io_pipe {
  static constexpr unsigned id = 1;
};

// id 0 -> file name or channel name: "c1" for hardware, "0" for emulator, "c1" for simulation.
using read_iopipe = ext::intel::kernel_readable_io_pipe<read_io_pipe, unsigned, 4>;

// id 1 -> file name or channel name: "c2" for hardware, "1" for emulator, "c2" for simulation.
using write_iopipe = ext::intel::kernel_writeable_io_pipe<write_io_pipe, unsigned, 3>;

    where:

    Interface Description
    For Hardware id N is mapped to the channel (not file) in the associated hardware. For example:
    • id 0 is mapped to the chan_id in the first interface defined.
    • id 1 is mapped to the chan_id in the second interface defined, and so on.

    If there is no matching channel, an error is generated.

    For Emulator id N is mapped to a file named N, which means id 0 is file "0". This file is read or written by reading or writing to the associated I/O pipe.
    For Simulator id N is mapped to chan_id names in the board_spec.xml file supplied with the BSP (See channels). For example:
    • id 0 is mapped to the chan_id in the first interface defined.
    • id 1 is mapped to the chan_id in the second interface defined, and so on.

    If there is no matching channel, an error is generated.

The I/O Pipe Classes and Their Use

The I/O pipe APIs exposed by the FPGA implementations is equivalent to the following class declarations: 

template <class name,
          class dataT,
          size_t min_capacity = 0>
class kernel_readable_io_pipe {
  public:
    static dataT read();  // Blocking
    static dataT read(bool &success_code);  // Non-blocking
};

template <class name,
          class dataT,
          size_t min_capacity = 0>
class kernel_writeable_io_pipe {
  public:
    static void write(dataT data);  // Blocking
    static void write(dataT data, bool &success_code);  // Non-blocking
}

The following table describes the template parameters:

Template Parameters

Parameter

Description

name

The type that is the basis of an I/O pipe identification. It may be provided by the device vendor or user defined. The type must contain a static constexpr unsigned expression with name id, which is used to determine the hardware device referenced by the I/O pipe.

dataT

The type of data packet contained within an I/O pipe. This is the data type that is read during a successful pipe read() operation, or written during a successful pipe write() operation. The type must have a standard layout and be trivially copyable.

min_capacity

User-defined minimum number of words (in units of dataT) that an I/O pipe must be able to store without any being read out. The compiler may create an I/O pipe with a larger capacity due to performance considerations.

NOTE:

A data word in this context is the data type that the pipe contains (dataT pipe template argument).

Example Code for I/O Pipes

Here is an example that includes the definitions above: 

// "Built-in pipes" provide interfaces with hardware peripherals
// These definitions are typically provided by a device vendor and
// made available to developers for use.
namespace example_platform {
  template <unsigned ID>
  struct ethernet_pipe_id {
    static constexpr unsigned id = ID;
  };

  using ethernet_read_pipe = kernel_readable_io_pipe<ethernet_pipe_id<0>, int, 0>;
  using ethernet_write_pipe = kernel_writeable_io_pipe<ethernet_pipe_id<1>, int, 0>;
}
NOTE:

For additional information, refer to the FPGA tutorial sample "IO Streaming with IO Pipes" listed in the Intel® oneAPI Samples Browser on Linux* or Windows*, or access the code sample in GitHub.

Using I/O Pipes to Stream Data

In many of the design examples, an FPGA device is treated as an accelerator, as illustrated in the following diagram, where the main computation (and therefore the data to be computed) resides on the host (CPU) and you accelerate some compute-intensive task using kernels on the device. The host moves the data to the device, performs the calculation, and moves the data back:

FPGA Device Used as an Accelerator

However, a key feature of FPGAs is their rich input-output (I/O) capabilities (for example, Ethernet). Taking advantage of these capabilities within the oneAPI programming environment requires a different programming model than the accelerator model as described above. In the model illustrated in the following figure, consider a kernel (or kernels) where some of the kernels are connected to the FPGA's I/O via I/O pipes and the main data flow is through the FPGA device rather than from CPU to FPGA and back:

Data Flow Through the FPGA Device Using I/O Pipes

In the Figure 2, there are four possible directions of data flow:

  • I/O to device
  • Device to I/O
  • Device to host
  • Host to device

The direction and amount of data flowing in the system is application dependent. The main source of data is I/O. Data streams into the device from I/O (I/O to device) and out of the device to I/O (device to I/O). However, the host and device exchange low-bandwidth control signals using a host-to-device and device-to-host connection (or side channel).

I/O pipes have the same interface API as inter-kernel pipes. This abstraction is implemented in the Board Support Package (BSP) and provides a much simpler programming model for accessing the FPGA I/O. See The I/O Pipe Classes and Their Use.

Consider an example scenario where you want the kernel to be able to receive UDP packets through the FPGA's Ethernet interface. Implementing the necessary hardware to process the data coming from the Ethernet pins in oneAPI would be both extremely difficult and inefficient. Moreover, there are already many RTL solutions for doing this. So, instead, you can implement this low-level logic with RTL in the BSP. This example is illustrated in the following figure, where the BSP connects the Ethernet I/O pins to a MAC IP, the MAC IP to a UDP offload engine IP, and finally the UDP offload engine IP to an I/O pipe. This I/O pipe can then be simply connected to the processing kernels. The details of these connections are abstracted from the oneAPI kernel developer. The following figure shows only an input I/O pipe. The process is similar for an output I/O pipe, but the data flows in the opposite direction.

I/O Input Pipe

Faking I/O Pipes

Unfortunately, designs that use these I/O pipes are confined to BSPs that support that specific I/O pipe, which makes prototyping and testing the processing kernels difficult. To address these concerns, a fake I/O pipes library is available. With this library, you can create kernels that behave like I/O pipes without having a BSP that actually supports them. This allows you to start prototyping and testing your processing kernels without a BSP that supports I/O pipes.

There are currently two options for faking IO pipes:

  • Device memory allocations: This option requires no special BSP support, as shown in the following figure:
    Device Memory Allocations

  • Unified Shared Memory (USM) host allocations: This option requires USM support in the BSP as shown in the following figure. See also Unified Shared Memory topic in the open-access book Data Parallel C++: Mastering DPC++ for Programming of Heterogeneous Systems using C++ and SYCL.
    Unified Shared Memory (USM) Host Allocations

Parent topic: Pipes Extension
Level Two Title