Developer Guide

FPGA Optimization Guide for Intel® oneAPI Toolkits

Download PDF
ID 767853
Date 7/13/2023
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A newer version of this document is available.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 Example 1: Multiply Chain Example 2: Accumulator 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
Pipes x
Host Pipes
Host Pipes x
Host Pipe Declaration Host Pipe API Host Pipes RTL Interfaces
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>) Create a 2xclock Interface (-Xsuse-2xclock) Disable Burst-Interleaving of Global Memory (-Xsno-interleaving=<global_memory_name>) 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)
Optimization Targets x
Minimum Latency Flow Maximum Throughput Without Area Optimization Heuristics 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 Asynchronous Parallelism Within Kernels (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 Characteristics of Pipes Restrictions of Pipes Guidelines for Designing Pipes Pipe and Atomic Fence
Asynchronous Parallelism Within Kernels (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

Relax Loop-Carried Dependency

Example 1: Multiply Chain

Based on the feedback from the optimization report, you can relax a loop-carried dependency by allowing the compiler to infer a shift register and increase the dependence distance. Increase the dependence distance by increasing the number of loop iterations that occur between the generation of a loop-carried value and its use.

Consider the following code example: 

constexpr int N = 128;
 queue.submit([&](handler &cgh) {
   accessor A(A_buf, cgh, read_only);
   accessor result(result_buf, cgh, write_only);
   cgh.single_task<class unoptimized>([=]() {
     float mul = 1.0f;
     for (int i = 0; i < N; i++)
       mul *= A[i];
 
     result[0] = mul;
   });
 });

The optimization report shows that the Intel® oneAPI DPC++/C++ Compiler infers pipelined execution for the loop successfully. However, the loop-carried dependency on the variable mul causes loop iterations to launch every six cycles. In this case, the floating-point multiplication operation on line 8 (that is, mul *= A[i]) contributes the largest delay to the computation of the variable mul.

To relax the loop-carried data dependency, instead of using a single variable to store the multiplication results, operate on M copies of the variable, and use one copy every M iterations:

  1. Declare multiple copies of the variable mul (for example, in an array called mul_copies).
  2. Initialize all copies of mul_copies.
  3. Use the last copy in the array in the multiplication operation.
  4. Perform a shift operation to pass the last value of the array back to the beginning of the shift register.
  5. Sum up all copies of mul_copies to mul and write the final value to the result.

The following code illustrates the restructured kernel: 

constexpr int N = 128;
 constexpr int M = 8;
 queue.submit([&](handler &cgh) {
   accessor A(A_buf, cgh, read_only);
   accessor result(result_buf, cgh, write_only);
   cgh.single_task<class optimized>([=]() {
     float mul = 1.0f;
 
     // Step 1: Declare multiple copies of variable mul
     float mul_copies[M];
 
     // Step 2: Initialize all copies
     for (int i = 0; i < M; i++)
       mul_copies[i] = 1.0f;
 
     for (int i = 0; i < N; i++) {
       // Step 3: Perform multiplication on the last copy
       float cur = mul_copies[M-1] * A[i];
 
       // Step 4a: Shift copies
       #pragma unroll 
       for (int j = M-1; j > 0; j--)
         mul_copies[j] = mul_copies[j-1];
 
       // Step 4b: Insert updated copy at the beginning
       mul_copies[0] = cur;
     }
 
     // Step 5: Perform reduction on copies
     #pragma unroll 
     for (int i = 0; i < M; i++)
       mul *= mul_copies[i];
 
     result[0] = mul;
   });
 });

Example 2: Accumulator

Similar to Example 1, this example also applies the same technique and demonstrates how to write single work-item kernels that carry out double precision floating-point operations efficiently by inferring a shift register.

Consider the following example: 

queue.submit([&](handler &cgh) { 
  accessor arr(arr_buf, cgh, read_only); 
  accessor result(result_buf, cgh, write_only); 
  accessor N(N_buf, cgh, read_only); 
  cgh.single_task<class unoptimized>([=]() { 
    double temp_sum = 0; 
    for (int i = 0; i < *N; ++i) 
      temp_sum += arr[i]; 
    result[0] = temp_sum; 
  }); 
});

The kernel unoptimized is an accumulator that sums the elements of a double precision floating-point array arr[i]. For each loop iteration, the Intel® oneAPI DPC++/C++ Compiler takes 10 cycles to compute the result of the addition and then stores it in the variable temp_sum. Each loop iteration requires the value of temp_sum from the previous loop iteration, which creates a data dependency on temp_sum. To relax the data dependency, infer the array arr[i] as a shift register.

The following is the restructured kernel optimized: 

//Shift register size must be statically determinable 
constexpr int II_CYCLES = 12; 
queue.submit([&](handler &cgh) { 
  accessor arr(arr_buf, cgh, read_only); 
  accessor result(result_buf, cgh, write_only); 
  accessor N(N_buf, cgh, read_only); 
  cgh.single_task<class optimized>([=]() { 
    //Create shift register with II_CYCLE+1 elements 
    double shift_reg[II_CYCLES+1]; 
 
    //Initialize all elements of the register to 0 
    for (int i = 0; i < II_CYCLES + 1; i++) { 
      shift_reg[i] = 0; 
    } 
     
    //Iterate through every element of input array 
    for(int i = 0; i < N; ++i){ 
      //Load ith element into end of shift register 
      //if N > II_CYCLE, add to shift_reg[0] to preserve values 
      shift_reg[II_CYCLES] = shift_reg[0] + arr[i]; 
   
      #pragma unroll 
      //Shift every element of shift register 
      for(int j = 0; j < II_CYCLES; ++j) 
      { 
        shift_reg[j] = shift_reg[j + 1]; 
      }  
    } 
  
    //Sum every element of shift register 
    double temp_sum = 0; 
   
    #pragma unroll  
    for(int i = 0; i < II_CYCLES; ++i) 
    { 
      temp_sum += shift_reg[i]; 
    } 
  
    result[0] = temp_sum; 
  }); 
});
Parent topic: Loops
Level Two Title