Got an issue with internal stdlib function rand() when setting ALU saturation bit.

Hi,

Can you please provide more details about the issue you are facing when both the rand() function and ALU saturation are used?

Is it possible to share the example code to replicate the issue in our side to assist you better on this?

Regards,
Nandini.C

#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>

#define BUFFER_SIZE 512
float TEST_Buffer[BUFFER_SIZE];

void store_rand_values( float *outBuf,// Output buffer address
uint32_t outSize // Output size
)
{
float *outPtr;
if( outSize == 0 ) return;
outPtr = outBuf;
for(int inDex = 0 ; inDex < outSize ; inDex++ )
{
// store rand values to buffer
*outPtr++ = rand();
}
}

void main(void)
{

// take register browser and set the ALU sat bit on MODE1 register
//plot the buffer you will be able to see the issue.
//After that clear the ALU saturation bit on MODE1 register we will be able to actual rand values.
while(1)
{
store_rand_values(TEST_Buffer,BUFFER_SIZE);
}

}

Hi,

The behaviour you have mentioned seems to be expected.

The rand function uses a Linear Congruential Generator (LCG) to generate pseudo-random numbers. A linear congruential generator is computed as:
X[n+1] = mod ((X[n] * a) + c), m)

The rand function sets a, c, and m from the formula above as follows:
a = 6364136223846793005
c = 1442695040888963407
m = 2^31

When the ALU bit is set, the multiplier operation can result in overflow or underflow, causing the intermediate result to be either 0x7FFFFFFF or 0x80000000. Adding 0x1442695040888963407 to this saturated value reduces the randomness.

For further details, please refer to the “rand.c” file located at the following path. Also please refer the disassembly view for saturated values:
<installation path>\cces\3.0.1\SHARC\lib\src\libc_src

Regards,
Nandini C