It was exciting to see two new AVR® microcontroller (MCU) ports in FreeRTOS™ version 10.3.1, or rather six new ports. These ports not only cover the megaAVR® 0-series of MCUs and the brand-new AVR Dx devices, but also the three main compilers for AVR MCUs: MPLAB XC8 compiler, AVR-GCC, and IAR Embedded Workbench® for AVR. Why is this great news? Well, it’s because the ATmega323 and ATmega128 MCUs that are supported in earlier versions of FreeRTOS are older devices. While these devices are still available products, AVR MCUs have evolved significantly - and who doesn’t want to use the latest generation of a product?
With this support for new devices and a new compiler, I could not help wondering how much memory the standard FreeRTOS demo examples consume on one of these new MCUs and how much the choice of compiler actually influences the code size. To keep this post concise, I’ll focus on just the ports for the AVR Dx MCU families.
AVR Dx: Is this a tinyAVR®, megaAVR or AVR XMEGA® MCU?
The most well-known AVR MCU is perhaps the ATmega328, which is the MCU used on the Arduino® UNO kit. The AVR Dx is the newest generation of AVR MCUs and it is similar to the latest tinyAVR 1-series, like the ATtiny817, which again is similar to the AVR XMEGA devices. So, it is a bit more sophisticated than the older megaAVR MCUs. This difference is most clearly seen in the peripherals and how they are better organized in the memory map.
There are currently two families of AVR Dx MCUs. The AVR DA family features the latest version of the Peripheral Touch Controller (PTC) module used to interface capacitive sensors for use in touch-enabled user interfaces. The AVR DB family has built-in op amps and multi-voltage I/O, which means that selected pins can be run from a separate voltage domain. This eliminates the need for expensive level converters.
AVR Dx MCUs offer a significant improvement over past AVR MCUs. The core logic runs from an internal regulated supply, which means that the maximum speed of the core is not limited by the external supply voltage. It can now run at 24 MHz, regardless of the supply voltage. In other words, you have more horsepower available, even in low-voltage applications, with a lower active mode power consumption.
The IDEs and Compilers
The newest compiler to support AVR MCUs is the MPLAB XC8 compiler, which is a variant of the GCC compiler. It is integrated into the MPLAB X IDE and was added to the v7.0.2542 release of Microchip Studio. The MPLAB XC8 compiler is available in a free version, which offers a subset of optimization options. The commercial version, called MPLAB XC8 PRO, offers all the optimizations you’d expect from a professional grade compiler. While IAR EWAVR and Microchip Studio are available for use with Windows® OS only, MPLAB X and the MPLAB XC8 compiler is available for Linux® and macOS®, as well as Windows systems. By the way, MPLAB X IDE supports the AVR-GCC compiler as well.
The AVR-GCC variant of the GNU GCC compiler was maintained by the user community for many years, and the original WIN-AVR version still exists even though it is not actively maintained. Later Atmel decided to make AVR-GCC part of the Atmel Studio IDE and made AVR-GCC available to AVR MCU users as an Atmel Studio plug-in. The AVR-GCC compiler and Arduino boards contributed significantly to the popularity of AVR MCUs in the Maker community. In November 2020, Microchip rebranded Atmel Studio 7 to Microchip Studio for AVR and SAM Devices; I’ll refer to it as Microchip Studio. There are no functional changes, but it is worth mentioning a very nice added feature: support for the MPLAB XC8 compiler in Microchip Studio. This means that users who are familiar with Atmel Studio and AVR-GCC can stay in this well-known IDE while taking advantage of the MPLAB XC8 compiler’s benefits.
IAR Embedded Workbench for AVR (IAR EWAVR for short) was the first C compiler for use with AVR MCUs. The inventors of the AVR core worked closely with IAR when the instruction set was defined, which resulted in a better instruction set that was optimized for C code. IAR EWAVR combines an IDE, a compiler and a simulator. Microchip’s debugging and programming tools can also be used in IAR EWAVR.
The compiler versions used are:
- AR EWAVR (v7.30)
- AVR-GCC (build v18.104.22.1688)
- MPLAB XC8 Compiler (v2.30)
Finding the Right Port and Device
Looking at the supported devices on the FreeRTOS website, I realized that you need to know your semiconductor company history. Although AVR MCUs are Microchip products, they were originally released by Atmel, which was acquired by Microchip in 2016. This is why AVR devices are listed under Atmel and not under Microchip on the FreeRTOS website. You will also find a comprehensive list of Microchip’s Arm® core-based MCUs here too.
Navigating deeper into the FreeRTOS web structure I found the port page for the AVR Dx, which supports both the AVR DA and the AVR DB devices. The demos that come with the port use the AVR128DA48 device and the corresponding AVR128DA48 Curiosity Nano Evaluation Board. This board was used to confirm that the code is running, which it was for all ports.
Download (or Clone with Git)
I downloaded the standard distribution from the FreeRTOS downloads page. It is a zip file that contains all ports and demos, along with the kernel and FreeRTOS libraries. You can also clone it from the FreeRTOS GitHub project, if you prefer to have a local Git branch of the code.
Download the zip file, unpack it and navigate to the Demo folder. If you like working with MCU, this is like a candy store – there’s so much to choose from!
Before going into details with compilers, it is worth mentioning the Demo examples.
The Demo ExamplesThe AVR Dx ports come with three demos. They are integrated into on single code project for each compiler port. The three demos are:
You use the mainSELECTED_APPLICATION define to select the demo at compile time. In the main.c file you define mainSELECTED_APPLICATION to 0, 1 or 2 to select the demo you want before hitting the build button.
MPLAB X IDE (v5.40) and MPLAB XC8 Compiler (v2.31)
From the menu, Open project and navigate to the AVR_Dx_MPLAB project file. A project-file is in the MPLAB X IDE a folder with the extension “.X”, in this case “AVR_Dx_MPLAB.X”.
The Project Properties describe how the compiler is configured. The configuration confirms that the project is configured to use the MPLAB XC8 v2.31 compiler and that the selected device is the AVR128DA48. The port page indicated that the MPLAB XC8 v2.20 compiler was used for the development and test of the port, but using v2.31 (the latest version at the writing of this article) did not cause any difficulties.
So, what optimization level is enabled? The optimization options are located under the XC8 Global Options -> XC8 Compiler in the Options category drop-down.
All optimization checkboxes are checked, except the "Debug" options, which you would primarily want to check if you do debugging with an in-system debugger like the MPLAB PICkit™ 4, MPLAB ICD 4 or Atmel-ICE programmers/debuggers, but that is a totally different story. The Optimization level is set to "s", which is the highest level of size optimization. This optimization level is accessible for the PRO version of the MPLAB XC8 Pro compiler. The free version of the MPLAB XC8 compiler supports optimization level -O1 and -O2. Compiling with -Os causes warnings that some of the optimization settings used for -Os are ignored as they require the MPLAB XC8 PRO compiler, but it compiles without errors.
In addition, the -flto switch is specified in the Additional Options text box. I consulted with the MPLAB XC8 compiler user guide to see what the switch controls. The -flto switch turns on the "standard link-time optimizer," which is a feature that is found in the PRO version of the MPLAB XC8 compiler. As a side note, the -flto option is not documented for the AVR-GCC compiler, so this is a difference between the MPLAB XC8 compiler and AVR-GCC.
Since Microchip has an online archive of all past versions of the MPLAB XC compiler with corresponding user guides, search engines may not serve you the latest version of the user guide; I recommend that you obtain the version you need directly from the MPLAB XC Compiler product pages.
Code Size Results
Since not everyone has a license for the MPLAB XC8 PRO compiler, I have provided the results of the code size both for the free compiler and the PRO compiler. The compiler does not produce any code size information in the Output window, but the Dashboard window shows a nice graphical and plain-text representation.
The code size results for the three examples, so the size of the demo applications:
|MPLAB XC8 PRO Compiler (-Os)||MPLAB XC8 Free Compiler (-Os)|
|Demo||Flash Size||SRAM Size||Flash Size||SRAM Size|
As you can see, the MPLAB XC8 PRO compiler clearly does a better job in producing a smaller program code than the Free version, while the data memory usage is virtually the same for the two settings.
Microchip Studio (v7.0.2542) and AVR-GCC (v22.214.171.1248)
Now it’s time to open the Microchip Studio project. In Microchip Studio select Open Project, navigate to the FreeRTOS/Demo/AVR_Dx_Atmel_Studio folder and open the RTOSDemo file.
Open the compiler configuration by right-clicking on the project in the Solutions Explorer (navigation) and selecting Properties. From there, select the AVR/GNU C Compiler -> Optimization item. All check-boxes are checked and optimization is -Os.
In the Debugging option the debug level is set to -g2, which is the default level.
For the Linker -> Optimization, the "Garbage collection unused section" option is checked. The GUI seems to include most common GCC options, but also allows you to specify additional options manually.
Microchip Studio and AVR-GCC generates a couple of warnings, which is also the case for the other compilers. Some warnings are concerning, like doing a longer shift than the width of the variable can support, but this is outside the scope of this blog post. Microchip Studio shows the code size in the compiler output window:
A warning that states that the code size information may be inaccurate if code is placed in other segments than the standard .text segment used for program code. Using the avr-size tool to look into the ELF output file, I concluded that the code mainly uses the .text segment, so good enough for this comparison as the size error is very small.
|Demo||Flash Size||SRAM Size|
IAR Embedded Workbench for AVR
In IAR EWAVR, the FreeRTOS port project is opened using the workspace file. The optimization settings are found by right-clicking the project and selecting Options. The GUI is very straightforward to use since the options are limited and it seems difficult miss anything crucial. Select the level of optimization and whether you want the code to be size or speed optimized, and then check all the checkboxes.
Porthardware.h includes FreeRTOSConfig.h, and for some reason the compiler was not able to locate that file by itself, so I had to specify the path for the code to compile. This is an easy fix, but I had expected the code to compile out of the box.
Hit Rebuild all in the Project menu…
|Demo:||Flash Size||SRAM Size|
Since I am Not a Compiler Expert…
We cannot all be compiler experts, so I reached out to Microchip’s and IAR’s and compiler experts and asked if the optimization settings used for the ports could be improved further. These are optimization settings that many users may not be familiar with, so I have included what I learned as inspiration to those who need to shrink their code to a minimum.
Expert Advice from Microchip
Microchip’s experts suggested adding the switches listed below to the MPLAB XC8 compiler’s Global Options:
This significantly improved of the code size, reducing it by 7.9% to 10.6%:
|Demo||Flash Size||SRAM Size||Improvement|
Expert Advice from IAR
IAR’s experts proposed adding the following to the command line option:
This results in a code size reduction of around 200 bytes for all three demo examples, which represents a 1.5% to 3.1% improvement:
|IAR EWAVR (Expert)|
|Demo:||Flash Size||SRAM Size||Improvement|
The quest was to find out how well suited the AVR Dx family is for running FreeRTOS in terms of memory consumption. So, let have a look at the code size results out of the box, without the expert feedback from IAR and Microchip:
|AVR-GCC||MPLAB XC8 PRO Compiler||MPLAB XC8 Free Compiler||IAR EWAVR|
|Demo:||Flash Size||SRAM Size||Flash Size||SRAM Size||Flash Size||SRAM Size||Flash Size||SRAM Size|