Cortex-A vs Cortex-M. Differents CRC output calculation ?

[mr_max]

Hello,

I have an issue with a software that originally work for an embedded Linux system on ARM-A7 core.That software contain a check sum based on a polynomial CRC-16 modbus. It work very well on ARM-A7 core and even other architectures (x86, amd64). But when I port the code to KL25 ARM-M0+ core, the check sum result give me a wrong output.

My input buffer contain 6 bytes : 0x 01 05 00 02 FF 00 (from MSB to LSB respectively)

The output result should be : 0x FA 2D. I double checked with this link On-line CRC calculation and free library

But with ARM-M0+, I got : 0x 45 92 ?!

So, where is the bug ? From ALU ? Compilation ? Data type or sign ?

Down below is the code I used. I did not write it myself. It is just a copy paste from a website found somewhere on internet. I believe that if I analyse the process of that code and read the link above, I could find a way to solve my problem but... Nah ! I am bit lazy to night.:smileylaugh:

I am very curious to understand why it does that. Anyone have an idea ?

unsigned int CRC(char* buf, int len)
{
  unsigned int crc = 0xFFFF;
  int pos;
  for(pos = 0; pos < len; pos++) {
    crc ^= (unsigned int)buf[pos];          // XOR byte into least sig. byte of crc
    int i;
    for (i = 8; i != 0; i--) {            // Loop over each bit
      if ((crc & 0x0001) != 0) {      // If the LSB is set
        crc >>= 1;                    // Shift right and XOR 0xA001
        crc ^= 0xA001;
      }
      else                            // Else LSB is not set
        crc >>= 1;                    // Just shift right
    }
  }
  // Note, this number has low and high bytes swapped, so use it accordingly (or swap bytes)
  return crc;
}

Max

[carlos_neri]

do you know what's the "int" size on the other compilers used? My first guess is that in an implementation "int" is 16-bit and other uses 32-bit.

Do a quick test: run the code by replacing any "unsigned int" to "uint16_t" and any int (assuming default is signed) as "int16_t" and see the output, then change the same types to "uint32_t" and "int32_t" and check the output.

[mr_max]

Yes I did it Carlos. I replaced all unsigned int to uint16_t. But it only changed the outputs result by another wrong one.

In this link : On-line CRC calculation and free library​, I downloaded the sources and after some modifications, I implemented into my application. Now it work ! . The only bad thing is that CRC calculation cost now more spaces in memory because it use a table of 2*256 bytes...

If someone need a CRC-16 Modbus and do not really care about space memory code, below is the modified source code with its CRC-16 table. Tested with ARM Cortex-M0+ core. Enjoy it !

//CRC-16 (Modbus) table
static short crc_tab16[256] =
{
        0x0000,0xC0C1,0xC181,0x0140,0xC301,0x03C0,0x0280,
        0xC241,0xC601,0x06C0,0x0780,0xC741,0x0500,0xC5C1,
        0xC481,0x0440,0xCC01,0x0CC0,0x0D80,0xCD41,0x0F00,
        0xCFC1,0xCE81,0x0E40,0x0A00,0xCAC1,0xCB81,0x0B40,
        0xC901,0x09C0,0x0880,0xC841,0xD801,0x18C0,0x1980,
        0xD941,0x1B00,0xDBC1,0xDA81,0x1A40,0x1E00,0xDEC1,
        0xDF81,0x1F40,0xDD01,0x1DC0,0x1C80,0xDC41,0x1400,
        0xD4C1,0xD581,0x1540,0xD701,0x17C0,0x1680,0xD641,
        0xD201,0x12C0,0x1380,0xD341,0x1100,0xD1C1,0xD081,
        0x1040,0xF001,0x30C0,0x3180,0xF141,0x3300,0xF3C1,
        0xF281,0x3240,0x3600,0xF6C1,0xF781,0x3740,0xF501,
        0x35C0,0x3480,0xF441,0x3C00,0xFCC1,0xFD81,0x3D40,
        0xFF01,0x3FC0,0x3E80,0xFE41,0xFA01,0x3AC0,0x3B80,
        0xFB41,0x3900,0xF9C1,0xF881,0x3840,0x2800,0xE8C1,
        0xE981,0x2940,0xEB01,0x2BC0,0x2A80,0xEA41,0xEE01,
        0x2EC0,0x2F80,0xEF41,0x2D00,0xEDC1,0xEC81,0x2C40,
        0xE401,0x24C0,0x2580,0xE541,0x2700,0xE7C1,0xE681,
        0x2640,0x2200,0xE2C1,0xE381,0x2340,0xE101,0x21C0,
        0x2080,0xE041,0xA001,0x60C0,0x6180,0xA141,0x6300,
        0xA3C1,0xA281,0x6240,0x6600,0xA6C1,0xA781,0x6740,
        0xA501,0x65C0,0x6480,0xA441,0x6C00,0xACC1,0xAD81,
        0x6D40,0xAF01,0x6FC0,0x6E80,0xAE41,0xAA01,0x6AC0,
        0x6B80,0xAB41,0x6900,0xA9C1,0xA881,0x6840,0x7800,
        0xB8C1,0xB981,0x7940,0xBB01,0x7BC0,0x7A80,0xBA41,
        0xBE01,0x7EC0,0x7F80,0xBF41,0x7D00,0xBDC1,0xBC81,
        0x7C40,0xB401,0x74C0,0x7580,0xB541,0x7700,0xB7C1,
        0xB681,0x7640,0x7200,0xB2C1,0xB381,0x7340,0xB101,
        0x71C0,0x7080,0xB041,0x5000,0x90C1,0x9181,0x5140,
        0x9301,0x53C0,0x5280,0x9241,0x9601,0x56C0,0x5780,
        0x9741,0x5500,0x95C1,0x9481,0x5440,0x9C01,0x5CC0,
        0x5D80,0x9D41,0x5F00,0x9FC1,0x9E81,0x5E40,0x5A00,
        0x9AC1,0x9B81,0x5B40,0x9901,0x59C0,0x5880,0x9841,
        0x8801,0x48C0,0x4980,0x8941,0x4B00,0x8BC1,0x8A81,
        0x4A40,0x4E00,0x8EC1,0x8F81,0x4F40,0x8D01,0x4DC0,
        0x4C80,0x8C41,0x4400,0x84C1,0x8581,0x4540,0x8701,
        0x47C0,0x4680,0x8641,0x8201,0x42C0,0x4380,0x8341,
        0x4100,0x81C1,0x8081,0x4040
};

unsigned short crc16(char* data, int len)
{
    unsigned short crc=0xFFFF;// crc_16_modbus;
    int i;
    for(i=0;i<len;i++)
    {
        crc = update_crc16(crc, data[i] );
    }
    return crc;
}
unsigned short update_crc16( unsigned short crc, char c ) {
  unsigned short tmp, short_c;
  short_c = 0x00ff & (unsigned short) c;
  tmp =  crc       ^ short_c;
  crc = (crc >> 8) ^ crc_tab16[ tmp & 0xff ];
  return crc;
}

int main(void)
{

unsigned short crc;

char txbuf[8] = {0x01, 0x05, 0x00, 0x02, 0xFF, 0x00};  

//Compute CRC

crc = crc16((txbuf,6);

}

Max