int main(void) {
float a0 = 1.01f;
float a1 = 1.02f;
float a2 = 1.03f;
float a3 = 1.04f;
float b0 = 13.33f;
float b1 = 13.34f;
float b2 = 13.35f;
float b3 = 13.36f;
float res0 = test_scalar(a0, b0);
float res1 = test_scalar(a1, b1);
float res2 = test_scalar(a2, b2);
float res3 = test_scalar(a3, b3);
__m128 av = _mm_set_ps(a0, a1, a2, a3);
__m128 bv = _mm_set_ps(b0, b1, b2, b3);
// fake use to prevent deletion of target function
__m128 resv = test_to_be_generated(av, bv);
printf("res (scalar): %f %f %f %f\n", res0, res1, res2, res3);
printf("res (packetized): %f %f %f %f\n", ((float*)&resv)[0], ((float*)&resv)[1], ((float*)&resv)[2], ((float*)&resv)[3]);
return 0;
}
int main(void)
{
vbx_test_init();
vbx_mxp_t *this_mxp = VBX_GET_THIS_MXP();
const int VBX_SCRATCHPAD_SIZE = this_mxp->scratchpad_size;
const int required_vectors = 4;
int N = VBX_SCRATCHPAD_SIZE / sizeof(vbx_mm_t) / required_vectors;
int PRINT_LENGTH = min( N, MAX_PRINT_LENGTH );
double scalar_time, vector_time;
int errors=0;
vbx_mxp_print_params();
printf( "\nAdd test...\n" );
printf( "Vector length: %d\n", N );
vbx_mm_t *scalar_in1 = malloc( N*sizeof(vbx_mm_t) );
vbx_mm_t *scalar_in2 = malloc( N*sizeof(vbx_mm_t) );
vbx_mm_t *scalar_out = malloc( N*sizeof(vbx_mm_t) );
vbx_mm_t *vector_in1 = vbx_shared_malloc( N*sizeof(vbx_mm_t) );
vbx_mm_t *vector_in2 = vbx_shared_malloc( N*sizeof(vbx_mm_t) );
vbx_mm_t *vector_out = vbx_shared_malloc( N*sizeof(vbx_mm_t) );
// vbx_mm_t *vector_out = vector_in2 - 5;
vbx_sp_t *v_in1 = vbx_sp_malloc( N*sizeof(vbx_sp_t) );
vbx_sp_t *v_in2 = vbx_sp_malloc( N*sizeof(vbx_sp_t) );
vbx_sp_t *v_out = vbx_sp_malloc( N*sizeof(vbx_sp_t) );
// vbx_sp_t *v_out = v_in2-5;
VBX_T(test_zero_array)( scalar_out, N );
VBX_T(test_zero_array)( vector_out, N );
VBX_T(test_init_array)( scalar_in1, N, 1 );
VBX_T(test_copy_array)( vector_in1, scalar_in1, N );
VBX_T(test_init_array)( scalar_in2, N, 1 );
VBX_T(test_copy_array)( vector_in2, scalar_in2, N );
VBX_T(test_print_array)( scalar_in1, PRINT_LENGTH );
VBX_T(test_print_array)( scalar_in2, PRINT_LENGTH );
scalar_time = test_scalar( scalar_out, scalar_in1, scalar_in2, N );
VBX_T(test_print_array)( scalar_out, PRINT_LENGTH);
vbx_dma_to_vector( v_in1, (void *)vector_in1, N*sizeof(vbx_sp_t) );
vbx_dma_to_vector( v_in2, (void *)vector_in1, N*sizeof(vbx_sp_t) );
vector_time = test_vector( v_out, v_in1, v_in2, N, scalar_time );
vbx_dma_to_host( (void *)vector_out, v_out, N*sizeof(vbx_sp_t) );
vbx_sync();
VBX_T(test_print_array)( vector_out, PRINT_LENGTH );
errors += VBX_T(test_verify_array)( scalar_out, vector_out, N );
VBX_TEST_END(errors);
return 0;
}
int main(void)
{
double scalar_time, vector_time;
int errors=0;
vbx_test_init();
vbx_mxp_print_params();
printf("\nVector FIR test...\n");
vbx_mm_t *scalar_sample = malloc( (SAMP_SIZE+NTAPS)*sizeof(vbx_mm_t) );
vbx_mm_t *scalar_coeffs = malloc( NTAPS*sizeof(vbx_mm_t) );
vbx_mm_t *scalar_out = malloc( SAMP_SIZE*sizeof(vbx_mm_t) );
vbx_mm_t *sample = vbx_shared_malloc( (SAMP_SIZE+NTAPS)*sizeof(vbx_mm_t) );
vbx_mm_t *coeffs = vbx_shared_malloc( NTAPS*sizeof(vbx_mm_t) );
vbx_mm_t *vector_out = vbx_shared_malloc( SAMP_SIZE*sizeof(vbx_mm_t) );
VBX_T(test_zero_array)( scalar_out, SAMP_SIZE );
VBX_T(test_zero_array)( vector_out, SAMP_SIZE );
VBX_T(test_init_array)( scalar_sample, SAMP_SIZE, 0xff );
VBX_T(test_copy_array)( sample, scalar_sample, SAMP_SIZE );
VBX_T(test_init_array)( scalar_coeffs, NTAPS, 1 );
VBX_T(test_copy_array)( coeffs, scalar_coeffs, NTAPS );
VBX_T(test_zero_array)( scalar_sample+SAMP_SIZE, NTAPS );
VBX_T(test_zero_array)( sample+SAMP_SIZE, NTAPS );
printf("\nSamples:\n");
VBX_T(test_print_array)( scalar_sample, min(SAMP_SIZE,MAX_PRINT_LENGTH) );
printf("\nCoefficients:\n");
VBX_T(test_print_array)( scalar_coeffs, min(NTAPS,MAX_PRINT_LENGTH) );
scalar_time = test_scalar( scalar_out, scalar_sample, scalar_coeffs);
VBX_T(test_print_array)( scalar_out, min(SAMP_SIZE,MAX_PRINT_LENGTH) );
#ifdef USE_TRANSPOSE
vector_time = test_vector_transpose( vector_out, sample, coeffs, scalar_time );
VBX_T(test_print_array)( vector_out, min(SAMP_SIZE,MAX_PRINT_LENGTH) );
errors += VBX_T(test_verify_array)( scalar_out, vector_out, SAMP_SIZE-NTAPS );
#endif //USE_TRANSPOSE
#ifdef USE_1D
vector_time = test_vector_1d( vector_out, sample, coeffs, scalar_time );
VBX_T(test_print_array)( vector_out, min(SAMP_SIZE,MAX_PRINT_LENGTH) );
errors += VBX_T(test_verify_array)( scalar_out, vector_out, SAMP_SIZE-NTAPS );
#endif //USE_1D
#ifdef USE_2D
vector_time = test_vector_2d( vector_out, sample, coeffs, scalar_time );
VBX_T(test_print_array)( vector_out, min(SAMP_SIZE,MAX_PRINT_LENGTH) );
errors += VBX_T(test_verify_array)( scalar_out, vector_out, SAMP_SIZE-NTAPS );
#endif //USE_2D
VBX_TEST_END(errors);
return 0;
}
int main(int argc, char **argv)
{
_Bool my_bool = true;
_Bool my_bool_array[NUM_ELEMS];
int i;
test_scalar(&my_bool);
for(i = 0; i < NUM_ELEMS; i+=2)
my_bool_array[i] = true;
for(i = 1; i < NUM_ELEMS; i+=2)
my_bool_array[i] = false;
test_array(my_bool_array, NUM_ELEMS);
return 0;
}
void test( const std::string & label ,
const size_t elem_count ,
const size_t iter_count )
{
KokkosArray::Impl::Timer timer ;
double seconds_scalar ;
double seconds_multi ;
double seconds_array1 ;
double seconds_array4 ;
double seconds_array16 ;
{ // Loop 16 times:
Explicit::TestHexGrad<double,float,Device> test_scalar( elem_count );
timer.reset();
for ( size_t i = 0 ; i < iter_count * 16 ; ++i ) {
test_scalar.apply();
}
Device::fence();
seconds_scalar = timer.seconds() / ( 16 * iter_count * elem_count );
}
{ // 16 x elements
Explicit::TestHexGrad<double,float,Device> test_multiple( elem_count * 16 );
timer.reset();
for ( size_t i = 0 ; i < iter_count ; ++i ) {
test_multiple.apply();
}
Device::fence();
seconds_multi = timer.seconds() / ( 16 * iter_count * elem_count );
}
{ // 16 x elements with Array<1>
typedef KokkosArray::Array<double,1> coord_scalar_type ;
typedef KokkosArray::Array<float,1> grad_scalar_type ;
Explicit::TestHexGrad<coord_scalar_type,grad_scalar_type,Device>
test_array( elem_count * 16 );
timer.reset();
for ( size_t i = 0 ; i < iter_count ; ++i ) {
test_array.apply();
}
Device::fence();
seconds_array1 = timer.seconds() / ( 16 * iter_count * elem_count );
}
{ // 4 x elements with Array<4>
typedef KokkosArray::Array<double,4> coord_scalar_type ;
typedef KokkosArray::Array<float,4> grad_scalar_type ;
Explicit::TestHexGrad<coord_scalar_type,grad_scalar_type,Device>
test_array( elem_count * 4 );
timer.reset();
for ( size_t i = 0 ; i < iter_count ; ++i ) {
test_array.apply();
}
Device::fence();
seconds_array4 = timer.seconds() / ( 16 * iter_count * elem_count );
}
{ // 1 x elements with Array<16>
typedef KokkosArray::Array<double,16> coord_scalar_type ;
typedef KokkosArray::Array<float,16> grad_scalar_type ;
Explicit::TestHexGrad<coord_scalar_type,grad_scalar_type,Device> test_array( elem_count );
timer.reset();
for ( size_t i = 0 ; i < iter_count ; ++i ) {
test_array.apply();
}
Device::fence();
seconds_array16 = timer.seconds() / ( 16 * iter_count * elem_count );
}
std::cout << label
<< " scalar( " << seconds_scalar
<< " ) multi( " << seconds_multi << " )"
<< " ) array1( " << seconds_array1 << " )"
<< " ) array4( " << seconds_array4 << " )"
<< " ) array16( " << seconds_array16 << " )"
<< std::endl ;
}
int main(void)
{
vbx_test_init();
#if 0
vbx_mxp_t *this_mxp = VBX_GET_THIS_MXP();
const int VBX_SCRATCHPAD_SIZE = this_mxp->scratchpad_size;
int N = VBX_SCRATCHPAD_SIZE/sizeof(vbx_mm_t)/8;
#endif
int TEST_LENGTH = TEST_ROWS*TEST_COLS;
int NTAP_LENGTH = NTAP_ROWS*NTAP_COLS;
int PRINT_COLS = min( TEST_COLS, MAX_PRINT_LENGTH );
int PRINT_ROWS = min( TEST_ROWS, MAX_PRINT_LENGTH );
double scalar_time, vector_time;
int errors=0;
vbx_mxp_print_params();
printf( "\nMatrix FIR test...\n" );
printf( "Matrix dimensions: %d,%d\n", TEST_ROWS, TEST_COLS );
vbx_mm_t *scalar_in = malloc( TEST_LENGTH*sizeof(vbx_mm_t) );
vbx_mm_t *vector_in = vbx_shared_malloc( TEST_LENGTH*sizeof(vbx_mm_t) );
int32_t *scalar_filt = malloc( NTAP_LENGTH*sizeof(int32_t) );
int32_t *vector_filt = vbx_shared_malloc( NTAP_LENGTH*sizeof(int32_t) );
vbx_mm_t *scalar_out = malloc( TEST_LENGTH*sizeof(vbx_mm_t) );
vbx_mm_t *vector_out = vbx_shared_malloc( TEST_LENGTH*sizeof(vbx_mm_t) );
VBX_T(test_zero_array)( scalar_out, TEST_LENGTH );
VBX_T(test_zero_array)( vector_out, TEST_LENGTH );
VBX_T(test_init_array)( scalar_in, TEST_LENGTH, 1 );
VBX_T(test_copy_array)( vector_in, scalar_in, TEST_LENGTH );
test_init_array_word( scalar_filt, NTAP_LENGTH, 1 );
test_copy_array_word( vector_filt, scalar_filt, NTAP_LENGTH );
VBX_T(test_print_matrix)( scalar_in, PRINT_ROWS, PRINT_COLS, TEST_COLS );
test_print_matrix_word( scalar_filt, NTAP_ROWS, NTAP_COLS, NTAP_COLS );
scalar_time = test_scalar( scalar_out, scalar_in, scalar_filt,
TEST_ROWS, TEST_COLS, NTAP_ROWS, NTAP_COLS);
VBX_T(test_print_matrix)( scalar_out, PRINT_COLS, PRINT_ROWS, TEST_COLS );
vector_time = test_vector( vector_out, vector_in, vector_filt,
TEST_ROWS, TEST_COLS, NTAP_ROWS, NTAP_COLS, scalar_time );
VBX_T(test_print_matrix)( vector_out, PRINT_COLS, PRINT_ROWS, TEST_COLS );
int i;
for(i=0; i<TEST_ROWS-NTAP_ROWS; i++){
errors += VBX_T(test_verify_array)( scalar_out+i*TEST_COLS, vector_out+i*TEST_COLS, TEST_COLS-NTAP_COLS );
}
VBX_TEST_END(errors);
return 0;
}
int main(void)
{
vbx_test_init();
typedef vbx_word_t vbx_mm_t;
vbx_mxp_t *this_mxp = VBX_GET_THIS_MXP();
const int VBX_SCRATCHPAD_SIZE = this_mxp->scratchpad_size;
int N = VBX_SCRATCHPAD_SIZE / sizeof(vbx_mm_t );
N = 20;
int M = 20;
int PRINT_LENGTH = N<MAX_PRINT_LENGTH ? N : MAX_PRINT_LENGTH ;
// int PRINT_ROWS = PRINT_LENGTH;
int PRINT_ROWS = M<MAX_PRINT_LENGTH ? N : MAX_PRINT_LENGTH;
int PRINT_COLS = PRINT_LENGTH;
double scalar_time, vector_time,vector2_time;
int errors=0;
vbx_mxp_print_params();
printf( "\nMatrix multiply test...\n" );
printf( "Matrix dimensions: %d,%d\n", N, M );
vbx_mm_t *scalar_in1 = (vbx_mm_t*)malloc( M*N*sizeof(vbx_mm_t ) );
vbx_mm_t *scalar_in2 = (vbx_mm_t*)malloc( M*N*sizeof(vbx_mm_t ) );
vbx_mm_t *scalar_out = (vbx_mm_t*)malloc( N*N*sizeof(vbx_mm_t ) );
vbx_mm_t *vector_in1 = (vbx_mm_t*)vbx_shared_malloc( M*N*sizeof(vbx_mm_t ) );
vbx_mm_t *vector_in2 = (vbx_mm_t*)vbx_shared_malloc( M*N*sizeof(vbx_mm_t ) );
vbx_mm_t *vector_out = (vbx_mm_t*)vbx_shared_malloc( N*N*sizeof(vbx_mm_t ) );
if ( scalar_in1 == NULL ||
scalar_in2 == NULL ||
scalar_out == NULL ||
vector_in1 == NULL ||
vector_in2 == NULL ||
vector_out == NULL ){
printf("Malloc failed\n");
VBX_TEST_END(1);
return 0;
}
test_zero_array_word(scalar_out, N*N );
test_zero_array_word(vector_out, N*N );
test_init_array_word( scalar_in1, M*N, 1 );
test_copy_array_word( vector_in1, scalar_in1, M*N );
test_init_array_word( scalar_in2, M*N, 999 );
//scalar_mtx_xp_MN_word( vector_in2, scalar_in2, N, N );
test_copy_array_word( vector_in2, scalar_in2, M*N );
test_print_matrix_word( scalar_in1, PRINT_COLS, PRINT_ROWS, M );
test_print_matrix_word( scalar_in2, PRINT_ROWS, PRINT_COLS, N );
//change print sizes for outputs
PRINT_ROWS=PRINT_COLS=N<PRINT_LENGTH?N:PRINT_LENGTH;
scalar_time = test_scalar( scalar_out, scalar_in1, N, M, scalar_in2, M, N);
test_print_matrix_word( scalar_out, PRINT_COLS, PRINT_ROWS, N );
vector_time = test_vector( vector_out, vector_in1, N, M, vector_in2, M, N, scalar_time );
test_print_matrix_word( vector_out, PRINT_COLS, PRINT_ROWS, N );
errors += test_verify_array_word( scalar_out, vector_out, N*N);
vector2_time = test_vector_trans( vector_out, vector_in1, N, M, vector_in2, M, N, scalar_time );
test_print_matrix_word( vector_out, PRINT_COLS, PRINT_ROWS, N );
errors += test_verify_array_word( scalar_out, vector_out, N*N);
vector2_time = test_vector_sp( vector_out, vector_in1, N, M, vector_in2, M, N, scalar_time );
test_print_matrix_word( vector_out, PRINT_COLS, PRINT_ROWS, N );
errors += test_verify_array_word( scalar_out, vector_out, N*N);
vbx_shared_free(vector_out);
vbx_shared_free(vector_in2);
vbx_shared_free(vector_in1);
free(scalar_out);
free(scalar_in2);
free(scalar_in1);
//errors += orig_test();
VBX_TEST_END(errors);
return 0;
}
int main(void)
{
vbx_test_init();
vbx_mxp_t *this_mxp = VBX_GET_THIS_MXP();
const int VBX_SCRATCHPAD_SIZE = this_mxp->scratchpad_size;
const int required_vectors = 4;
int N = VBX_PAD_DN(VBX_SCRATCHPAD_SIZE / sizeof(vbx_mm_t) / required_vectors, this_mxp->scratchpad_alignment_bytes);
int PRINT_LENGTH = min( N, MAX_PRINT_LENGTH );
double scalar_time, vector_time;
int errors=0;
vbx_mxp_print_params();
printf( "\nVector copy test...\n" );
printf( "Vector length: %d\n", N );
vbx_mm_t *scalar_in = malloc( N*sizeof(vbx_mm_t) );
vbx_mm_t *scalar_out = malloc( N*sizeof(vbx_mm_t) );
vbx_mm_t *vector_in = vbx_shared_malloc( N*sizeof(vbx_mm_t) );
vbx_mm_t *vector_out = vbx_shared_malloc( N*sizeof(vbx_mm_t) );
vbx_sp_t *v_out = vbx_sp_malloc( N*sizeof(vbx_sp_t) );
vbx_sp_t *v_in = vbx_sp_malloc( N*sizeof(vbx_sp_t) );
VBX_T(test_zero_array)( scalar_in, N );
VBX_T(test_zero_array)( vector_in, N );
VBX_T(test_init_array)( scalar_in, N, 1 );
VBX_T(test_copy_array)( vector_in, scalar_in, N );
scalar_time = test_scalar( scalar_out, scalar_in, N );
VBX_T(test_print_array)( scalar_out, PRINT_LENGTH );
vbx_dma_to_vector( v_in, vector_in, N*sizeof(vbx_sp_t) );
vector_time = test_vector( v_out, v_in, N, scalar_time );
vbx_dma_to_host(vector_out, v_out, N*sizeof(vbx_sp_t) );
vbx_sync();
VBX_T(test_print_array)( vector_out, PRINT_LENGTH );
errors += VBX_T(test_verify_array)( scalar_out, vector_out, N );
vbx_sp_free();
#if TEST_DEEP_SP
errors += deep_vector_copy_test();
#endif
#if DEBUG_MAKE_SP_FULL
vbx_sp_malloc(vbx_sp_getfree());
#endif
#if TEST_DEEP_MM
errors += deep_vector_copy_ext_test();
#endif
VBX_TEST_END(errors);
return 0;
}
