ESP32 Firmware image size too high (IDFGH-5663)

[ashigupta]

Environment

• Development Kit: ESP32-DevKitM-1
• Kit version : -
• Module or chip used: ESP32-SOLO-1 (ESP32-Mini-1)
• IDF version : v4.2.1
• Build System: Make|CMake|idf.py
• Compiler version : xtensa-esp32-elf-gcc (crosstool-NG esp-2020r3) 8.4.0
• Operating System: Windows
• (Windows only) environment type: ESP Command Prompt|
• Using an IDE?: Yes (Eclipse)
• Power Supply: USB

Problem Description

Firmware image size is too high after build even for 'hello world' example code

//Detailed problem description goes here.
I am developing my application on ESP32 and using Wifi Station mode, IOT, BLE(Bluedroid). I realized that my firmware size is going beyond 1.6MB.

I compiled 'Hello World' example code and found that firmware image is **
sdkconfig - Copy.txt
** which is too huge just for hello world program. Below is the output of 'idf.py size' command
Total sizes:
DRAM .data size: 8376 bytes
DRAM .bss size: 2104 bytes
Used static DRAM: 10480 bytes ( 170256 available, 5.8% used)
Used static IRAM: 40689 bytes ( 90383 available, 31.0% used)
Flash code: 77795 bytes
Flash rodata: 24448 bytes
Total image size:~ 153412 bytes (.bin may be padded larger

I read espressif document on below link that explains how we can reduce image size.
https://docs.espressif.com/projects/esp-idf/en/latest/esp32/api-guides/performance/size.html

Following this i am able to reduce the firmware image size to 120KB which is still huge for just hello world program. Below is the output of idf.py size command
Total sizes:
DRAM .data size: 5528 bytes
DRAM .bss size: 1856 bytes
Used static DRAM: 7384 bytes ( 173352 available, 4.1% used)
Used static IRAM: 32363 bytes ( 98709 available, 24.7% used)
Flash code: 72207 bytes
Flash rodata: 11044 bytes
Total image size:~ 122998 bytes (.bin may be padded larger)

I have attached my modified sdkconfig file for hello world with which i am able to reduce size to 120KB.

I want to understand how i can reduce the firmware image size more.

I know Bluedroid consume more memory as compare to nimble but is there any way to reduce size when i am using Wifi, BLE both. I will be using OTA firmware update and with this huge image size upgrade size will be too high and i also have to alot high memory size OTA partition.
I have gone through many blogs, videos but still there is no great help.

Expected Behavior

I expect bin image size should be small for just a hello world program.

Actual Behavior

Bin image size is too high.

Steps to reproduce

1. Build 'Hello world' example code using idf.py build command
2. Check image size using idf.py size

[negativekelvin]

#7007

[ashigupta]

#7007

Hi,

I have already followed and able to reduce around 300KB in my application that uses Wifi station mode + BLE ( Bluedroid). But still the memory footprint is above 1.3MB.

Below is memory consumption component wise in my application:
#####################################################################
Per-archive contributions to ELF file:
Archive File DRAM .data & .bss & other IRAM D/IRAM Flash code & rodata Total
libbt.a 163 26171 0 316 0 214659 61218 302527
libnet80211.a 1038 6181 0 11732 0 98433 21536 138920
libAzure_IOT.a 408 0 0 0 0 66157 39998 106563
libbtdm_app.a 763 2405 0 27634 0 67198 4551 102551
libcrypto_matixssl.a 96 1294 0 0 0 84464 13393 99247
libc.a 4 4 0 0 0 80915 4638 85561
liblwip.a 29 3823 0 0 0 70243 4238 78333
libpp.a 1262 3785 0 20829 0 34943 4485 65304
libwpa_supplicant.a 12 844 0 0 0 42607 3364 46827
libmbedcrypto.a 64 29 0 30 0 40598 3853 44574
libphy.a 1621 675 0 7837 0 32443 0 42576
libCode.a 49 7332 0 0 0 20547 10962 38890
libLib.a 2444 4320 0 0 0 2890 28023 37677
libfreertos.a 2084 736 0 12243 0 0 1406 16469
libESP32_BLE.a 48 92 0 0 0 2361 13679 16180
libIOT.a 0 4 0 0 0 7333 3935 11272
libAzure_ltk_adapters.a 36 26 0 0 0 6605 4600 11267
libcoexist.a 2030 28 0 3704 0 4632 325 10719
libnvs_flash.a 0 24 0 0 0 10487 80 10591
libDCI.a 4 57 0 0 0 9532 283 9876
libsoc.a 155 4 0 6574 0 1639 1273 9645
libspi_flash.a 623 288 0 4826 0 968 0 6705
libgcc.a 8 12 0 193 0 5555 872 6640
libm.a 4 0 0 0 0 5111 69 5184
libvfs.a 308 48 0 0 0 4492 159 5007
libesp_wifi.a 492 76 0 496 0 3477 205 4746
libesp32.a 16 21 0 1938 0 2023 544 4542
libesp_system.a 1682 17 0 2188 0 250 68 4205
libstdc++.a 8 20 0 0 0 2577 1084 3689
libesp_netif.a 12 25 0 0 0 3366 129 3532
libDebugManager.a 48 50 0 0 0 1954 1128 3180
libdriver.a 80 55 0 156 0 2248 284 2823
libheap.a 12 8 0 1811 0 933 35 2799
libnewlib.a 190 272 0 953 0 1117 119 2651
libesp_event.a 4 13 0 0 0 2411 68 2496
librtc.a 0 4 0 2235 0 0 0 2239
libBLE.a 0 320 0 0 0 1589 74 1983
libesp_timer.a 16 20 0 954 0 795 19 1804
libWiFi.a 0 18 0 0 0 1297 428 1743
libOS_FreeRTOS.a 0 73 0 0 0 1485 44 1602
libbootloader_support.a 0 0 0 886 0 544 36 1466
liblog.a 8 268 0 250 0 526 25 1077
libesp_ringbuf.a 0 0 0 709 0 0 183 892
libcore.a 0 29 0 0 0 617 228 874
libpthread.a 16 12 0 186 0 586 0 800
libefuse.a 36 4 0 0 0 676 60 776
libesp_common.a 8 70 0 41 0 582 43 744
libapp_update.a 1 12 0 154 0 97 260 524
libGeneric.a 0 4 0 0 0 434 70 508
libhal.a 0 0 0 443 0 0 32 475
libxtensa.a 0 0 0 416 0 0 58 474
libtcpip_adapter.a 0 17 0 0 0 376 49 442
libcxx.a 8 16 0 0 0 336 0 360
libESP32D0WD.a 0 4 0 0 0 223 64 291
libTimeConversion.a 0 0 0 0 0 221 0 221
libNV.a 0 16 0 0 0 200 0 216
libMiscCode.a 0 4 0 0 0 59 0 63
(exe) 0 0 0 3 0 3 12 18
libesp_eth.a 0 0 0 0 0 0 0 0
libmesh.a 0 0 0 0 0 0 0 0
libsoc_esp32.a 0 0 0 0 0 0 0 0
#####################################################################

I have tried the same sdkconfig changes in 'Hello_World' program and memory footprint for the same reduced to 120KB but still 120KB is too high for just a simple 'Hello_World' program.

Also the link you have attached below is mentioned:

"The default ESP-IDF configuration is also targeted for maximum functionality and easy debug-ability, rather than code size. One of the simplest size improvements is to change the compiler optimization level to Size. There are a number of other possible actions listed in the upcoming Performance Guide, will link to this as soon as it's available."

Is the performance guide available?

[negativekelvin]

It is the same guide you linked to. If you switch to newlib nano printf you can save some more space. Idf.py size-files will show you exactly what is using the space. Remember it is hello world running on top of a full rtos. Optimizing the size of hello world is not going to be that useful in helping you optimize a large app. They sell 16mb modules for a reason.

[ashigupta]

It is the same guide you linked to. If you switch to newlib nano printf you can save some more space. Idf.py size-files will show you exactly what is using the space. Remember it is hello world running on top of a full rtos. Optimizing the size of hello world is not going to be that useful in helping you optimize a large app. They sell 16mb modules for a reason.

Thanks. I am able to further reduce memory footprint by enabling nano printf in hello world. Its now ~80KB.
Yes hello world is running on top of full rtos and this will also contribute in in size. I have attached idf.py file-size details.

What are the options to reduce memory footprint if BLE(Bluedroid) is enabled? I have already disabled debug logs but is there any other options that can help ?
Hello World - File size.txt

[igrr]

Can you consider using NimBLE stack instead of Bluedroid? It has lower code size.
https://docs.espressif.com/projects/esp-idf/en/latest/esp32/api-reference/bluetooth/nimble/index.html

[ashigupta]

Can you consider using NimBLE stack instead of Bluedroid? It has lower code size.
https://docs.espressif.com/projects/esp-idf/en/latest/esp32/api-reference/bluetooth/nimble/index.html

Hi,

Is there any document available that shows memory consumption / footprint differences when using Bluedroid and nimble.
It will be very helpful for understanding before starting any work using nimble.

[rahult-github]

Is there any document available that shows memory consumption / footprint differences when using Bluedroid and nimble.

Hi @ashigupta , As of today, there doesn't exist a document for this. The stack configuration for bluedroid / nimble is different.

However, if you take an existing example in ESP-SDK which works for both bluedroid and nimble ( say wifi_prov_mgr ) and compile it for the respective stacks, then we should the size with nimble stack to be around 240KB lesser than the same when compiled with bluedroid ( for default configuration )

Again, please note, this is a general estimate as configuration change can change the final numbers.

[rahult-github]

Hi @ashigupta , did the above nimble stack suggestion help ?

[rahult-github]

Closing this issue. Please feel free to reopen in case of any queries.