Merge branch 'config-optimizations' into 'main'

doc: Guide for optimizing RAM and flash using config options See merge request app-frameworks/esp-matter!596
2026-04-27 19:13:13 +00:00 · 2024-02-27 00:59:08 +08:00
parent d4cbddc4d3 d01fc5d03b
commit 16caf828e9
3 changed files with 201 additions and 2 deletions
@@ -1,4 +1,4 @@
-6. API Reference
+7. API Reference
 ================

 .. toctree::
@@ -22,5 +22,6 @@ Table of Contents
   3. Matter Certification <certification>
   4. Production Considerations <production>
   5. Security Considerations <security>
-   6. API Reference <api-reference/index>
+   6. RAM and Flash Optimizations <optimizations>
+   7. API Reference <api-reference/index>
   A1 Appendix FAQs <faq>
@@ -0,0 +1,198 @@
+6. Configuration options to optimize RAM and Flash
+==================================================
+
+6.1 Overview
+------------
+
+There are several configuration options available to optimize Flash and RAM storage.
+The following list highlights key options that significantly increase the free DRAM, heap, and reduce the flash
+footprint.
+
+For more optimizations, we've also listed the reference links to esp-idf's optimization guide.
+
+6.2 Configurations
+------------------
+
+6.2.1 Test Environment setup
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+All numbers mentioned below are collected in the following environment:
+
+- esp-idf `v5.1.2`_
+- esp-matter `cb3bc9d`_
+- Example: `light`_
+- SoC: ESP32-C3
+
+.. note::
+
+    - These numbers may vary slightly in a different environment.
+    - All numbers are in bytes
+    - As we are using BLE only for commissioning, BLE memory is freed post commissioning,
+      hence there is an increase in the free heap post commissioning. (``CONFIG_USE_BLE_ONLY_FOR_COMMISSIONING=y``)
+    - After building an example, some DRAM will be utilized, and the remaining DRAM will be
+      allocated as heap. Therefore, a direct increase in the free DRAM will reflect as an increase in free heap.
+
+
+6.2.2 Default Configuration
+~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+We have used the default light example here, and below listed are the static and dynamic sizes.
+
+.. csv-table:: Static memory stats
+  :header: "", "Size", "Decreased by"
+
+  Used D/IRAM,169042,--
+  Used Flash,1370786,--
+
+.. csv-table:: Dynamic memory stats
+  :header: "", "Free Heap", "Increased by"
+
+  On Bootup, 64824, --
+  Post Commissioning, 126808, --
+
+
+6.2.3 Disable the chip-shell
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Console shell is helpful when developing/debugging the application, but may not be necessary in
+production. Disabling the shell can save space. Disable the below configuration option.
+
+::
+
+    CONFIG_ENABLE_CHIP_SHELL=n
+
+.. csv-table:: Static memory stats
+  :header: "", "Size", "Decreased by"
+
+  Used D/IRAM,165480,3562
+  Used Flash,1311926,58860
+
+.. csv-table:: Dynamic memory stats
+  :header: "", "Free Heap", "Increased by"
+
+  On Bootup, 76344, 11520
+  Post Commissioning, 137696, 10888
+
+
+6.2.4 Adjust the dynamic endpoint count
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The default dynamic endpoint count is 16, which may be excessive for a normal application creating only 2 endpoints.
+eg: light, only has two endpoints, one for root endpoint and one for actual light.
+Adjusting this to a lower value, corresponding to the actual number of endpoints the application will create, can save DRAM.
+
+Here, we have set the dynamic endpoint count to 4. Increase in the DRAM per endpoint is ~275 bytes.
+
+::
+
+    CONFIG_ESP_MATTER_MAX_DYNAMIC_ENDPOINT_COUNT=4
+
+.. csv-table:: Static memory stats
+  :header: "", "Size", "Decreased by"
+
+  Used D/IRAM,162136,3344
+  Used Flash,1311914,12
+
+.. csv-table:: Dynamic memory stats
+  :header: "", "Free Heap", "Increased by"
+
+  On Bootup, 79688, 3344
+  Post Commissioning, 141204, 3508
+
+
+6.2.5 Use the newlib nano formatting
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+This optimization saves approximately 25-50K of flash, depending on the target. In our case, it results in a flash
+reduction of 61.5 KB.
+
+Additionally, it lowers the high watermark of task stack for functions that call printf() or other string formatting
+functions. Fore more details please take a look at esp-idf's `newlib nano formatting guide`_.
+
+::
+
+    CONFIG_NEWLIB_NANO_FORMAT=y
+
+.. csv-table:: Static memory stats
+  :header: "", "Size", "Decreased by"
+
+  Used D/IRAM,162136,0
+  Used Flash,1281354,30560
+
+.. csv-table:: Dynamic memory stats
+  :header: "", "Free Heap", "Increased by"
+
+  On Bootup, 82748, 3060
+  Post Commissioning, 143956, 2752
+
+
+6.2.6 Few BLE Optimizations
+~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Since most devices will primarily operate as BLE peripherals and typically won't need more than one connection
+(especially if it's just a Matter app), we can optimize by reducing the maximum allowed connections, thereby
+saving DRAM. Additionally, given the peripheral nature of these devices, we can disable the central and
+observer roles, for further optimization.
+
+Below are the configuration options that can be set to achieve these optimizations.
+
+::
+
+    CONFIG_NIMBLE_MAX_CONNECTIONS=1
+    CONFIG_BTDM_CTRL_BLE_MAX_CONN=1
+    CONFIG_BT_NIMBLE_ROLE_CENTRAL=n
+    CONFIG_BT_NIMBLE_ROLE_OBSERVER=n
+
+.. csv-table:: Static memory stats
+  :header: "", "Size", "Decreased by"
+
+  Used D/IRAM,161682,454
+  Used Flash,1275860,5494
+
+.. csv-table:: Dynamic memory stats
+  :header: "", "Free Heap", "Increased by"
+
+  On Bootup, 83220, 472
+  Post Commissioning, 143804, -152
+
+
+6.2.7 Configuring logging event buffer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Matter events serve as a historical record, stored in chronological order in the logging event buffer.
+By reducing the buffer size we can potentially save the DRAM. However, it's important to note that this reduction
+could lead to the omission of events.
+
+For instance, reducing the critical log buffer from 4K to 1K could save 3K DRAM, but it comes with the trade-off of
+potentially missing critical events.
+
+::
+
+    CONFIG_EVENT_LOGGING_CRIT_BUFFER_SIZE=1024
+
+.. csv-table:: Static memory stats
+  :header: "", "Size", "Decreased by"
+
+  Used D/IRAM,158610,3072
+  Used Flash,1275860,0
+
+6.3 References for futher optimizations
+---------------------------------------
+
+- `RAM optimization`_
+- `Binary size optimization`_
+- `Speed Optimization`_
+- `ESP32 Memory Analysis — Case Study`_
+- `Optimizing IRAM`_ can provide additional heap area but at the cost of execution speed. Relocating frequently-called
+  functions from IRAM to flash may result in increased execution time
+
+
+.. _`v5.1.2`: https://github.com/espressif/esp-idf/tree/v5.1.2
+.. _`cb3bc9d`: https://github.com/espressif/esp-matter/tree/cb3bc9d
+.. _`light`: https://github.com/espressif/esp-matter/tree/cb3bc9d/examples/light
+.. _`newlib nano formatting guide`: https://docs.espressif.com/projects/esp-idf/en/latest/esp32c3/api-guides/performance/size.html#newlib-nano-formatting
+.. _`RAM optimization`: https://docs.espressif.com/projects/esp-idf/en/latest/esp32c3/api-guides/performance/ram-usage.html
+.. _`Binary size optimization`: https://docs.espressif.com/projects/esp-idf/en/latest/esp32c3/api-guides/performance/size.html
+.. _`Speed Optimization`: https://docs.espressif.com/projects/esp-idf/en/latest/esp32c3/api-guides/performance/speed.html
+.. _`ESP32 Memory Analysis — Case Study`: https://blog.espressif.com/esp32-memory-analysis-case-study-eacc75fe5431
+.. _`Optimizing IRAM`: https://docs.espressif.com/projects/esp-idf/en/latest/esp32c3/api-guides/performance/ram-usage.html#optimizing-iram-usage