Erroneous array reading inside a thread - c

Question

I Have a multi-threaded C program where I have 4 threads doing some arithmetic computations using some global arrays. here is example of the code.
__m256 *array_1;
__m256 *array_2;
__m256 *array_3;
#define ALIGNMENT 32
#define SIMD_STEP 8
void Init_arrays()
{
int i;
posix_memalign((void**) &array_1, ALIGNMENT, 32*sizeof(__m256));
posix_memalign((void**) &array_2, ALIGNMENT, 4 *sizeof(__m256));
posix_memalign((void**) &array_3, ALIGNMENT, 2 *sizeof(__m256));
for(i=0;i < 256; i+= SIMD_STEP)
{
// Filling array for the 1st stage
}
for(i=0;i < 64; i+= SIMD_STEP)
{
// Filling array for the 2nd stage
}
for(i=0;i < 16; i+= SIMD_STEP)
{
// Filling array for the 3rd stage
}
}
void *routine(void *thread_info)
{
int n;
unsigned t_start,t_stop;
unsigned ind1, ind2, ind3;
float *arr_in , *arr_out;
struct thread_data *mydata;
mydata = (struct thread_data*) thread_info;
t_start = mydata->start;
t_stop = mydata->stop;
arr_in = mydata->input;
arr_out = mydata->output;
for (n = t_start; n < t_stop; n += 8)
{
ind1 = 256 + n;
ind2 = 512 + n;
vec_a = _mm256_load_ps((float *) (&arr_in[n ]) );
vec_b = _mm256_load_ps((float *) (&arr_in[ind1]) );
vec_c = _mm256_load_ps((float *) (&arr_in[ind2]) );
T_fac1 = array_1[n];
T_fac2 = array_2[n];
T_fac3 = array_3[n];
// print data 'printf()'
// further computations
_mm256_store_ps((float *) (&arr_out[n ]), (vec_a) );
_mm256_store_ps((float *) (&arr_out[ind1]), (vec_b) );
_mm256_store_ps((float *) (&arr_out[ind2]), (vec_c) );
}
pthread_exit(NULL);
}
void foo(float* in,float* out)
{
unsigned t,i=0;
for(t=0;t<256;t+=64)
{
thread_data_array[i].start = t;
thread_data_array[i].stop = t+QUARTER;
thread_data_array[i].input = in;
thread_data_array[i].output = out;
pthread_create(&threads[i],NULL,routine,(void*)&thread_data_array[i]);
i++;
}
for(i=0; i<NUM_THREADS; i++)
{
int rc = pthread_join(threads[i], NULL);
if (rc)
{
fprintf(stderr, "failed to join thread #%u - %s\n",i, strerror(rc));
}
}
}
int main()
{
float *data1;
float *data2;
posix_memalign((void**)&data1, 32, 1024 * sizeof(float));
posix_memalign((void**)&data2, 32, 1024 * sizeof(float));
Load_inputs(reals,imags);//load data into the two arrays
Init_arrays();
// print data 'printf()'
foo(data1,data2);
return EXIT_SUCCESS;
}
For some reason reading from the array_1 for example doesnt work as it should be inside the thread, and I did not know the reason behind it. Here is display of the array_1 as it should be
Display from the main Display from the thread
RE = 1.000000 IM = -0.000000 RE = 1.000000 IM = -0.000000
RE = 0.999981 IM = -0.006136 RE = 0.399624 IM = 0.671559
RE = 0.999925 IM = -0.012272 RE = 0.416430 IM = 0.634393
RE = 0.999831 IM = -0.018407 RE = 0.433094 IM = 0.595699
RE = 0.999699 IM = -0.024541 RE = 0.449612 IM = 0.555570
RE = 0.999529 IM = -0.030675 RE = 0.465977 IM = 0.514103
RE = 0.999322 IM = -0.036807 RE = 0.482184 IM = 0.471397
RE = 0.999078 IM = -0.042938 RE = 0.498228 IM = 0.427555
RE = 0.998795 IM = -0.049068 // the same
RE = 0.998476 IM = -0.055195 // the same
RE = 0.998118 IM = -0.061321 // the same
RE = 0.997723 IM = -0.067444 // the same
RE = 0.997290 IM = -0.073565 // the same
RE = 0.996820 IM = -0.079682 // the same
RE = 0.996313 IM = -0.085797 // the same
RE = 0.995767 IM = -0.091909 // the same
Could somebody know what is the reason behind this erroneous results?

Answer 1

Given
__m256 *array_1;
__m256 *array_2;
__m256 *array_3;
#define ALIGNMENT 32
#define SIMD_STEP 8
void Init_arrays()
{
int i;
posix_memalign((void**) &array_1, ALIGNMENT, 32*sizeof(__m256));
posix_memalign((void**) &array_2, ALIGNMENT, 4 *sizeof(__m256));
posix_memalign((void**) &array_3, ALIGNMENT, 2 *sizeof(__m256));
.
.
.
This loop references array elements that are way out of range:
for (n = t_start; n < t_stop; n += 8)
{
.
.
.
T_fac1 = array_1[n];
T_fac2 = array_2[n];
T_fac3 = array_3[n];
array_3 has all of two members: 2 *sizeof(__m256) yet the index is incremented by eight?