ld.exe:linker_script.ld:138 cannot move location counter backwards | ArduinoFFT Error

Question

Board: Arduino Nano 33 BLE Sense

I'm pulling PCM data from an onboard microphone and running it through the arduinoFFT library (version 1.5.6) to obtain a spectrogram. Since the arduinoFFT library requires the data be an array of doubles, I've converted the data from short to double, and started the FFT process. The FFT.Windowing() function worked fine, but I get the error at the FFT.Compute() function call (line 77). The error goes away when I remove the function call. The error:

Found 25 compatible libraries
Scanning dependencies...
Dependency Graph
|-- <arduinoFFT> 1.5.6
|-- <PDM> 1.0
Building in release mode
Linking .pio\build\nano33ble\firmware.elf
c:/users/benlu/.platformio/packages/toolchain-gccarmnoneeabi/bin/../lib/gcc/arm-none- 
eabi/8.2.1/../../../../arm-none-eabi/bin/ld.exe:linker_script.ld:138 cannot move location counter 
backwards (from 000000002004a470 t
collect2.exe: error: ld returned 1 exit status
*** [.pio\build\nano33ble\firmware.elf] Error 1

And the code:

#include "Mic.h"
#include <arduinoFFT.h>
// FFT
arduinoFFT FFT = arduinoFFT(); /* Create FFT object /
/
These values can be changed in order to evaluate the functions
/
const uint16_t samples = 16128; //This value MUST ALWAYS be a power of 2
const double signalFrequency = 16000;
const double samplingFrequency = 16000;
const uint8_t amplitude = 100;
/
These are the input and output vectors
Input vectors receive computed results from FFT
*/
double vImag[samples] = {0};
// buffer to read samples into, each sample is 16-bits
short* sample_buffer;
// number of samples read, read from Mic class on loop
volatile int samples_read;
// The length of the buffer to use. Currently set to 4 seconds.
int buffer_size = 16000;
double double_buffer[16384] = {};
// Instantiate
Mic mic;
// DO NOT PRINT INSIDE THIS FUNCTION - Causes it to hang.
void onPDMdata() {
  mic.sample();
}
void setup() {
  Serial.begin(115200);
  while (!Serial);
  PDM.onReceive(onPDMdata);
  PDM.setBufferSize(16128);
Serial.println("3...");
  delay(1000); 
  Serial.println("2...");
  delay(1000);
  Serial.println("1...");
  delay(1000);

  Serial.println("Listening!");
mic.begin();
}
void loop() {
// wait for samples to be read
  samples_read = mic.samples_read();
// If desired buffer size is reached.
  if (samples_read >= buffer_size) {
    // Retrieves audio buffer
    sample_buffer = mic.sample_buffer();
  // print samples to the serial monitor or plotter
  for (int i = 0; i &lt; samples_read; i++) {
    double_buffer[i] = (double)(sample_buffer[i]); 
    Serial.println(double_buffer[i]);   
  }

Serial.println(&quot;---------- SAMPLE BREAK ----------&quot;);

FFT.Windowing(double_buffer, samples, FFT_WIN_TYP_HAMMING, FFT_FORWARD);
FFT.Compute(double_buffer, vImag, samples, FFT_FORWARD); /* Compute FFT &lt;--------------------- THIS LINE */
// FFT.ComplexToMagnitude(double_buffer, vImag, samples); /* Compute magnitudes */

// delay for data collection only, comment when not collecting
delay(1000);
Serial.println(&quot;Listening!&quot;);

// clear the read count and resume recording
mic.clear_buffer();

}
}

Where can I start to troubleshoot this issue? I've not had to deal with linker errors before, so any pointers would be greatly appreciated. Thanks in advanced!

timemage · Accepted Answer · 2021-02-20T16:49:29.447

Short Version

The short answer is that you've exhausted RAM.

These two arrays alone:

const uint16_t samples = 16128;
double vImag[samples] = {0};
double double_buffer[16384] = {};

...are occupying (16128 + 16384) * 8 == 260096 bytes of memory. There's only 262144 bytes of RAM in entirety.

The way the linker script is set up at least 1024 bytes are reserved for the stack, and there is no doubt at least another kByte in use somewhere. In the Mbed OS and Arduino core files if nothing else.

As for why it fails only when you call Compute, that comes down to the compiler optimizer discovering that one (perhaps both) of your arrays are not actually used and so not entering into the mix when the linker is run.

Details

In theory, you should have gotten a somewhat nicer and easier to understand message from an ASSERT in the linker script:

ASSERT(. <= (ORIGIN(RAM) + LENGTH(RAM) - 1024), "heap region overflowed into stack");

So, you should have seen an error "heap region overflowed into stack" on your screen instead, which is maybe not the easiest thing for an Arduino user to decipher, but heap and stack are terms likely to be recognizable as memory-related sooner or later.

However, the linker executable shipped with the current version 1.3.2 of the Arduino core for Mbed ignores ASSERT statements completely; I have tested this (and tested my understanding of my test using an amd64 linker). Presumably, there's a build option for the linker or a command-line or configuration file option that controls that, but I haven't found it yet. In any case, it ignores a helpful ASSERT and instead proceeds to fail on line 138 mentioned in the error message:

 . = ORIGIN(RAM) + LENGTH(RAM) - 1024;

You can see it involves the same expression, just without a useful message if something goes wrong. The "location counter" in the message is a linker related term that roughly means "where the next thing will be placed in memory" as it cobbles together your final executable from separately compiled chunks. You've used so much RAM in your program that the location counter has been pushed past where the linker script wants to place the __HeapLimit symbol, 1K from the end of the MCU's RAM. The location counter is supposed to increase monotonically. So, evaluating this expression results in an erroneous attempt to "move location counter backwards".

ld.exe:linker_script.ld:138 cannot move location counter backwards | ArduinoFFT Error

1 Answers1

Short Version

Details