No need for an oscilloscope! High-precision time measurement using ARM-Cortex M DWT CYCCNT - A practical guide to minimized steps

What is DWT CYCCNT? A high-precision hardware counter for Cortex-M

In embedded development, measuring processing time in microseconds is directly linked to the stability of real-time control. This is because in real-time control, even a delay of just a few tens of microseconds can lead to performance degradation. Traditionally, the most common method of measurement was to use GPIO output and an oscilloscope. However, this requires wiring, securing GPIO, and checking the circuit diagram, which is time-consuming.
The CYCCNT (Cycle Counter) in the DWT (Data Watchpoint and Trace) that comes standard with the Cortex-M3/M4/M7/M23/M33 is an effective solution. CYCCNT is a 32-bit counter that increments with each CPU clock cycle and can Access directly from the software.

DWT CYCCNT operation image

The biggest advantage of CYCCNT is that it has almost zero measurement overhead. There is no I/O delay like GPIO toggle, and it automatically counts every clock. Even short processes of several tens of clocks can be accurately measured.

• Not available on Cortex-M0/M0+. Be sure to check the product manual to see if DWT is installed.

Why is an oscilloscope not needed? Why is DWT CYCCNT so convenient?

With DWT, time measurement can be completed simply by reading the hardware counter inside the CPU directly. No external wiring, oscilloscope, or logic analyzer is required. High-precision measurements are possible using only software.
You may be thinking, "Isn't the built-in timer or SysTick enough for measuring time?" However, in practice, they can be difficult to use due to the following limitations:

• The control method for built-in timers varies depending on the product, making them less portable.
• SysTick is often used exclusively by RTOS (e.g., FreeRTOS)
• Changing timer settings may affect system stability

On the other hand, DWT CYCCNT is:

• No external equipment required (software only)
• High accuracy in units of CPU clock
• Cortex-M common address for high portability

Easy to use even for beginners, no external tools are required, and it's accurate, making it the perfect first step in measuring time.

Minimum steps to enable DWT CYCCNT

The DWT registers are located in the PPB (Private Peripheral Bus) area, which is common to all Cortex-M devices. The PPB is a fixed address area for debugging, and can be Access at the same address even across different products, making code portability easy. You can use the registers by directly defining their addresses.
This article explains how to use DWT CYCCNT by directly writing to registers. Its advantage is that it can be easily implemented by copy and paste, and explains how to implement it using basic address manipulation. In environments where the CMSIS* header is provided, predefined register information can also be used.

• CMSIS (Cortex Microcontroller Software Interface Standard): A standard software interface for Cortex-M microcontrollers defined by ARM.

To enable DWT CYCCNT, follow these steps:

1. Set DEMCR.TRCENA (bit 24) to 1 to enable DWT/ITM

2. (Cortex-M7 only) Write the unlock key to DWT_LAR

3. Clear DWT_CYCCNT to 0

4. DWT_CTRL.CYCCNTENA (set bit 0 to 1 to start counting)

That's all it takes for CYCCNT to start working.

Since it is a fixed address, it can Access directly even without CMSIS.

Super simple measurement block that measures only the difference (copy and paste OK)

Below is the minimum configuration measurement template.
Only initialization and the start and end of measurement are described, making it a minimal configuration that is easy to understand.
The start and end points of the measurement are clearly indicated in the code with comments.

// レジスタ定義
  #define DEMCR        (*(volatile unsigned long*)0xE000EDFCUL)
  #define DWT_CTRL     (*(volatile unsigned long*)0xE0001000UL)
  #define DWT_CYCCNT   (*(volatile unsigned long*)0xE0001004UL)
  #define DWT_LAR      (*(volatile unsigned long*)0xE0001FB0UL)
  
  #define DEMCR_TRCENA    (1UL << 24)
  #define DWT_CYCCNTENA   (1UL << 0)
  
  int main(void) {
    volatile unsigned int sum = 0;
    int i=0;
  
    unsigned long start = 0;
    unsigned long cycles = 0;
    double time_us = 0;
  
    //　DWTの有効化（起動後に1回だけ必要）
    DEMCR |= DEMCR_TRCENA;          // DWT使用許可
    //DWT_LAR = 0xC5ACCE55UL;         // ロック解除（ARM Cortex-M7使用時のみ実行）
    DWT_CYCCNT = 0UL;               // カウンター初期化
    DWT_CTRL |= DWT_CYCCNTENA;      // カウント開始
  
    // 計測
    start = DWT_CYCCNT;  // 計測開始 --->
  
  　　　　　　// 時間計測する処理（今回は例として、ループする処理）
    for (i = 0; i < 1000; i++) {
        sum += i;
    }
  
    cycles = DWT_CYCCNT - start; // <--- 計測終了
  
    time_us = (double)cycles * 1000000 / 48000000.0; // CPU周波数に応じて変更してください（例 48MHz)
  
    // printfを利用できる環境の場合は下記をコメントアウトしてください。
    //printf("cycles=%lu, time=%.3f us\n", cycles, time_us);
  
    while(1){}
  
      return 0;
  }

Execution example (Cortex-M4F 48MHz)

cycles=17017, time=354.521 us

Advantages of the differential measurement method

• Does not include function call overhead
• Easy to copy and paste as a template
• Just write the process you want to measure on the spot

This is a convenient and easy method for beginners to learn first.

Practical points to improve accuracy

To improve the accuracy of differential measurements, keep the following points in mind.

Problems and solutions

Summary: The fastest way to "measure accurately" without any external equipment

Time measurement using DWT CYCCNT does not require "GPIO + oscilloscope" and enables highly accurate processing time measurement without any external equipment.
It counts in CPU clock units, achieving 1-clock accuracy. What's more, it only requires a few lines of initialization and one measurement block, making it easy to incorporate into existing code.

To summarise the points:

• Easy setup with no external equipment required
• High-precision measurement in units of one CPU clock
• By adding simple code, DWT CYCCNT can be measured.

First, let's use DWT CYCCNT to easily measure the processing time.

Related Info

• Cortex-M7 Beginner's Guide: From Comparison with Cortex-M4 to ECU Applications

<Representative NXP microcontrollers equipped with Cortex-M>

• S32K1 General-Purpose Automotive Microcontroller
• S32K3 General-Purpose Automotive Microcontroller
• S32M2 Integrated Solution for 12V Motor Control