Intel® High Level Synthesis Compiler Pro Edition: Best Practices Guide

Download PDF
ID 683152
Date 12/19/2022
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A newer version of this document is available.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
1. Intel® HLS Compiler Pro Edition Best Practices Guide 2. Best Practices for Coding and Compiling Your Component 3. FPGA Concepts 4. Interface Best Practices 5. Loop Best Practices 6. fMAX Bottleneck Best Practices 7. Memory Architecture Best Practices 8. System of Tasks Best Practices 9. Datatype Best Practices 10. Advanced Troubleshooting A. Intel® HLS Compiler Pro Edition Best Practices Guide Archives B. Document Revision History for Intel® HLS Compiler Pro Edition Best Practices Guide
1. Intel® HLS Compiler Pro Edition Best Practices Guide x
1.1. Pending Deprecation of the Intel® HLS Compiler
3. FPGA Concepts x
3.1. FPGA Architecture Overview 3.2. Concepts of FPGA Hardware Design 3.3. Methods of Hardware Design
3.1. FPGA Architecture Overview x
3.1.1. Adaptive Logic Module (ALM) 3.1.2. Digital Signal Processing (DSP) Block 3.1.3. Random-Access Memory (RAM) Blocks
3.1.1. Adaptive Logic Module (ALM) x
3.1.1.1. Lookup Table (LUT) 3.1.1.2. Register
3.2. Concepts of FPGA Hardware Design x
3.2.1. Maximum Frequency (fMAX) 3.2.2. Latency 3.2.3. Pipelining 3.2.4. Throughput 3.2.5. Datapath 3.2.6. Control Path 3.2.7. Occupancy
3.3. Methods of Hardware Design x
3.3.1. How Source Code Becomes a Custom Hardware Datapath 3.3.2. Scheduling 3.3.3. Mapping Parallelism Models to FPGA Hardware 3.3.4. Memory Types
3.3.1. How Source Code Becomes a Custom Hardware Datapath x
3.3.1.1. Mapping Source Code Instructions to Hardware 3.3.1.2. Mapping Arrays and Their Accesses to Hardware
3.3.2. Scheduling x
3.3.2.1. Dynamic Scheduling 3.3.2.2. Clustering the Datapath 3.3.2.3. Handshaking Between Clusters
3.3.3. Mapping Parallelism Models to FPGA Hardware x
3.3.3.1. Data Parallelism 3.3.3.2. Task Parallelism
3.3.3.1. Data Parallelism x
3.3.3.1.1. Executing Independent Operations Simultaneously 3.3.3.1.2. Pipelining
3.3.3.1.2. Pipelining x
3.3.3.1.2.1. Pipelining Loops Within A Component 3.3.3.1.2.2. Pipelining Across Component Invocations
3.3.4. Memory Types x
3.3.4.1. Component Memory 3.3.4.2. External Memory
4. Interface Best Practices x
4.1. Choose the Right Interface for Your Component 4.2. Control LSUs For Your Variable-Latency MM Host Interfaces 4.3. Avoid Pointer Aliasing
4.1. Choose the Right Interface for Your Component x
4.1.1. Pointer Interfaces 4.1.2. Avalon® Memory Mapped Host Interfaces 4.1.3. Avalon® Memory Mapped Agent Memories 4.1.4. Avalon® Memory Mapped Agent Registers 4.1.5. Avalon® Streaming Interfaces 4.1.6. Pass-by-Value Interface
5. Loop Best Practices x
5.1. Reuse Hardware By Calling It In a Loop 5.2. Parallelize Loops 5.3. Construct Well-Formed Loops 5.4. Minimize Loop-Carried Dependencies 5.5. Avoid Complex Loop-Exit Conditions 5.6. Convert Nested Loops into a Single Loop 5.7. Place if-Statements in the Lowest Possible Scope in a Loop Nest 5.8. Declare Variables in the Deepest Scope Possible 5.9. Raise Loop II to Increase fMAX 5.10. Control Loop Interleaving
5.2. Parallelize Loops x
5.2.1. Pipeline Loops 5.2.2. Unroll Loops 5.2.3. Example: Loop Pipelining and Unrolling
6. fMAX Bottleneck Best Practices x
6.1. Balancing Target fMAX and Target II
7. Memory Architecture Best Practices x
7.1. Example: Overriding a Coalesced Memory Architecture 7.2. Example: Overriding a Banked Memory Architecture 7.3. Merge Memories to Reduce Area 7.4. Example: Specifying Bank-Selection Bits for Local Memory Addresses
7.3. Merge Memories to Reduce Area x
7.3.1. Example: Merging Memories Depth-Wise 7.3.2. Example: Merging Memories Width-Wise
8. System of Tasks Best Practices x
8.1. Executing Multiple Loops in Parallel 8.2. Sharing an Expensive Compute Block 8.3. Implementing a Hierarchical Design 8.4. Balancing Capacity in a System of Tasks
8.4. Balancing Capacity in a System of Tasks x
8.4.1. Enable the Intel® HLS Compiler to Infer Data Path Buffer Capacity Requirements 8.4.2. Explicitly Add Buffer Capacity to Your Design When Needed
9. Datatype Best Practices x
9.1. Avoid Implicit Data Type Conversions 9.2. Avoid Negative Bit Shifts When Using the ac_int Datatype
10. Advanced Troubleshooting x
10.1. Component Fails Only In Simulation 10.2. Component Gets Poor Quality of Results
1. Intel® HLS Compiler Pro Edition Best Practices Guide
2. Best Practices for Coding and Compiling Your Component
3. FPGA Concepts
4. Interface Best Practices
5. Loop Best Practices
6. fMAX Bottleneck Best Practices
7. Memory Architecture Best Practices
8. System of Tasks Best Practices
9. Datatype Best Practices
10. Advanced Troubleshooting
A. Intel® HLS Compiler Pro Edition Best Practices Guide Archives
B. Document Revision History for Intel® HLS Compiler Pro Edition Best Practices Guide

5.2.3. Example: Loop Pipelining and Unrolling

Consider a design where you want to perform a dot-product of every column of a matrix with each other column of a matrix, and store the six results in a different upper-triangular matrix. The rest of the elements of the matrix should be set to zero.



The code might look like the following code example: 
1.	#define ROWS 4
2.	#define COLS 4
3.	
4.	component void dut(...) {
5.		float a_matrix[COLS][ROWS]; // store in column-major format
6.		float r_matrix[ROWS][COLS]; // store in row-major format
7.	
8.		// setup...
9.	
10.		for (int i = 0; i < COLS; i++) {
11.			for (int j = i + 1; j < COLS; j++) {
12.	
13.				float dotProduct = 0;
14.				for (int mRow = 0; mRow < ROWS; mRow++) {
15.					dotProduct += a_matrix[i][mRow] * a_matrix[j][mRow];
16.				}
17.				r_matrix[i][j] = dotProduct;
18.			}
19.		}
20.	 	
21.	 // continue...
22.		
23.	}

You can improve the performance of this component by unrolling the loops that iterate across each entry of a particular column. If the loop operations are independent, then the compiler executes them in parallel.

Floating-point operations typically must be carried out in the same order that they are expressed in your source code to preserve numerical precision. However, you can use the -ffp-reassociate compiler flag to relax the ordering of floating-point operations. With the order of floating-point operations relaxed, the following conditions occur in this loop:
  • The multiplication operations can occur in parallel.
  • The addition operations can be composed into an adder tree instead of an adder chain.
To learn more, review the tutorial: <quartus_installdir>/hls/examples/ tutorials/best_practices/ floating_point_ops
You can improve the throughput by unrolling the j-loop at line 11, but to allow the compiler to unroll the loop, you must ensure that it has constant bounds. You can ensure constant bounds by starting the j-loop at j = 0 instead of j = i + 1. You must also add a predication statement to prevent r_matrix from being assigned with invalid data during iterations 0,1,2,…i of the j-loop. 
01: #define ROWS 4
02: #define COLS 4
03: 
04: component void dut(...) {
05: 	float a_matrix[COLS][ROWS]; // store in column-major format
06: 	float r_matrix[ROWS][COLS]; // store in row-major format
07: 
08: 	// setup...
09: 
10: 	for (int i = 0; i < COLS; i++) {
11: 
12: #pragma unroll
13: 			for (int j = 0; j < COLS; j++) {
14: 				float dotProduct = 0;
15: 
16: #pragma unroll

17: 				for (int mRow = 0; mRow < ROWS; mRow++) {
18: 					dotProduct += a_matrix[i][mRow] * a_matrix[j][mRow];
19: 				}
20: 
21: 				r_matrix[i][j] = (j > i) ? dotProduct : 0; // predication
22: 			}
23: 		}
24:  	}
25:  	
26: 	// continue...
27: 	
28: }

Now the j-loop is fully unrolled. Because they do not have any dependencies, all four iterations run at the same time.

Refer to the resource_sharing_filter tutorial located at <quartus_installdir>/hls/examples/tutorials/best_practices/resource_sharing_filter for more details.

You could continue and also unroll the loop at line 10, but unrolling this loop would result in the area increasing again. By allowing the compiler to pipeline this loop instead of unrolling it, you can avoid increasing the area and pay about only four more clock cycles assuming that the i-loop only has an II of 1. If the II is not 1, the Details pane of the Loops Analysis page in the high-level design report (report.html) gives you tips on how to improve it.

The following factors are factors that can typically affect loop II:
  • loop-carried dependencies

    See the tutorial at <quartus_installdir>/hls/examples/tutorials/best_practices/loop_memory_dependency

  • long critical loop path
  • inner loops with a loop II > 1
Level Two Title