Memory Access Latency Benchmark

Table of Contents

• Supported Combinations
• Introduction
• Performance statistics
  • AM275x
• R5F
• C7
• Steps to Run the Example
  • Building Memory access latency application
  • Running the Memory access latency application
  • Sample output for Memory access latency example

Supported Combinations

Parameter	Value
CPU + OS	r5fss0-0 nortos
Toolchain	ti-arm-clang
Boards	am275x-evm
Example folder	examples/benchmarks/memory_latency_benchmark

Introduction

• This demo provides a rough measurement of the time taken for a Read or Write instruction to be executed at different memory regions (OCRAM, TCM).
• The demo uses simple assembly load and store commands to read the data at a specific memory location in different regions.
• The example does the following:
  1. Initializes the drivers and board
  2. Calls the Read functions at TCM, OCRAM, Non self TCM (peripheral address) for 32-bit and 64-bit access.
  3. The Read 32-bit functions consist of 32 commands for reading from 32-bit (4-byte) register.
  4. The Read 64-bit functions consist of 32 commands for reading from two 32-bit (4-byte) registers.
  5. Average cycles per byte is calculated in example.

Performance statistics

AM275x

R5F

Operation	Latency(ns)/byte
Read (32-bit access) - TCM	0.796875
Read (32-bit access) - OCRAM	9.234375
Read (32-bit access) - Non self TCM	35.320312
Read (64-bit access) - TCM	0.300781
Read (64-bit access) - OCRAM	4.617188
Read (64-bit access) - Non self TCM	17.660156

C7

Operation	Latency(ns)/byte
Read (32-bit access) - OCRAM	2.726562
Read (32-bit access) - L2RAM	2.437500
Read (64-bit access) - OCRAM	1.160156
Read (64-bit access) - L2RAM	1.152344

Steps to Run the Example

Building Memory access latency application

• When using CCS projects to build, import the CCS project for the required combination and build it using the CCS project menu (see Using SDK with CCS Projects).
• When using makefiles to build, note the required combination and build using make command (see Using SDK with Makefiles)

Running the Memory access latency application

Flash the application binary to the device, follow the steps mentioned here (see Flash an example application).

Sample output for Memory access latency example

BENCHMARK START - ARM R5F - Memory Access latency
 
[32-BIT READ] Self TCM Access Average Latency Per Byte: 0.796875 ns
[32-BIT READ] MSRAM Access Average Latency Per Byte: 9.234375 ns
[32-BIT READ] Non-Self TCM Access Average Latency Per Byte: 35.320312 ns
 
[64-BIT READ] Self TCM Access Average Latency Per Byte: 0.300781 ns
[64-BIT READ] MSRAM Access Average Latency Per Byte: 4.617188 ns
[64-BIT READ] Non-Self TCM Access Average Latency Per Byte: 17.660156 ns
 
BENCHMARK END
 
BENCHMARK START - C7x - Memory Access latency
 
[32-BIT READ] L2RAM Access Average Latency Per Byte: 2.437500 ns
[32-BIT READ] MSRAM Access Average Latency Per Byte: 2.726562 ns
 
[64-BIT READ] L2RAM Access Average Latency Per Byte: 1.152344 ns
[64-BIT READ] MSRAM Access Average Latency Per Byte: 1.160156 ns
 
BENCHMARK END