int main(void)
{
vbx_timestamp_t time_start, time_stop;
double scalar_time, vbx_time, vbx_time_masked;
int i, j, k, l, m, n;
int errors = 0;
vbx_test_init();
vbx_mxp_print_params();
pixel *input, *scalar_input, *vbx_input, *vbx_input_masked;
uint16_t *scalar_short;
input = (pixel *)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
scalar_input = (pixel *)vbx_remap_cached(input, IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
scalar_short = (uint16_t *)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(uint16_t));
vbx_input = (pixel *)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
vbx_input_masked = (pixel *)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
#if UNIT
unsigned char *vbx_img8;
unsigned short *img, *vbx_img;
unsigned int *iImg, *vbx_iImg;
unsigned int *iiImg, *vbx_iiImg;
img = (unsigned short*)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned short));
vbx_img = (unsigned short*)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned short));
vbx_img8 = (unsigned char*)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned char));
iImg = (unsigned int*)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned int));
vbx_iImg = (unsigned int*)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned int));
iiImg = (unsigned int*)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned int));
vbx_iiImg = (unsigned int*)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned int));
#endif//UNIT
printf("Resolution = %dx%d\n", IMAGE_WIDTH, IMAGE_HEIGHT);
printf("Initializing data\n");
vbx_timestamp_start();
for(l = 0; l < 1; l++){
char *src;
char *sdst;
char *vdst;
char *mdst;
if(l == 0){
load_lenna(input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_lenna(vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_lenna(vbx_input_masked, IMAGE_WIDTH, IMAGE_HEIGHT);
printf("\nLenna\n");
src = "lenna";
sdst = "s_lenna";
vdst = "v_lenna";
mdst = "m_lenna";
}else if(l == 1){
load_ms(input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_ms(vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_ms(vbx_input_masked, IMAGE_WIDTH, IMAGE_HEIGHT);
printf("\nMicrosoft\n");
src = "ms";
sdst = "s_ms";
vdst = "v_ms";
mdst = "m_ms";
}else if(l == 2){
load_blank(input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_blank(vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_blank(vbx_input_masked, IMAGE_WIDTH, IMAGE_HEIGHT);
printf("\nblank\n");
src = "blank";
sdst = "s_blank";
vdst = "v_blank";
mdst = "m_blank";
}
#if UNIT
int window = 20;
int log=0;
while(((window/3)>>log) >= 2) log++;
errors += compare_scalar_rgb2luma_to_vbw_rgb2luma16(img, vbx_img, vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH, MAX_PRINT_ERRORS);
vbw_rgb2luma8(vbx_img8, vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH);
int s;
#if LUT_CI
#if DOUBLE_LUT
printf("Testing double lut\n");
printf("Assign lbp double lut\n");
assign_lbp_lut_ci2();
int prev = errors;
printf("Cascade check\n");
/* errors += cascade_check_2w(face_lbp, face_lbp_max_stage, 256); */
/* errors += cascade_check_2h(face_lbp, face_lbp_max_stage, 256); */
errors += cascade_check_2b(face_lbp, face_lbp_max_stage, 256);
if (errors) {
printf("errors %d\n", errors-prev);
}
#else
assign_lbp_lut_ci();
printf("Testing cascade\n");
int prev = errors;
printf("lut check\n");
#if 0
#if 0
errors += lut_check(256, 0, 0, 0);
if (errors) {
printf("errors %d\n", errors-prev);
}
#elif 1
int print_errors = 0;
vbx_mxp_t *this_mxp = VBX_GET_THIS_MXP();
int vci_lanes = this_mxp->vcustom0_lanes;
int num_features = cascade_max_feature();
int input_length = 10;
int lut_length = num_features*vci_lanes;
int lut_iterations = 15;
#if 1
lut_length = input_length = 128;
lut_iterations = 13;
print_errors = 0;
errors += lut_check2(input_length, lut_length, lut_iterations, print_errors);
if (errors) {
printf("errors %d\n", errors-prev);
}
#elif 1
input_length = 64;
lut_length = input_length;
lut_iterations = 13;
print_errors = 1;
errors += lut_check2(input_length, lut_length, lut_iterations, print_errors);
if (errors) {
printf("errors %d\n", errors-prev);
}
#else
for(s = 2; s < 100; s=s+10){
errors += lut_check2(s, lut_length, lut_iterations, print_errors);
if (errors - prev > 0) {
printf("%d\terrors %d\n", s, errors-prev);
} else {
printf("%d\n", s);
}
prev = errors;
}
#endif
#else
for(s = 0; s < 2000; s=s+100){
errors += lut_check(s, 0, 0, 0);
if (errors - prev > 0) {
printf("%d\terrors %d\n", s, errors-prev);
} else {
printf("%d\n", s);
}
prev = errors;
}
#endif
#elif 1
#else
printf("check cascade\n");
prev = errors;
errors += cascade_check(face_lbp, face_lbp_max_stage, 256);
if (errors) {
printf("errors %d\n", errors-prev);
}
printf("Testing LBP LUT CI\n");
prev = errors;
for(s = 0; s < face_lbp_max_stage; s++){
errors += compare_vbx_lut_to_vbx_lut_ci(s, MAX_PRINT_ERRORS);
}
if (errors) {
printf("errors %d\n", errors-prev);
prev = errors;
}
#endif
#endif
#endif
#if 0
printf("Printing grey scale img\n");
printf("grey = [");
for (j = 0; j < IMAGE_HEIGHT; j++) {
printf("[");
for (i = 0; i < IMAGE_WIDTH; i++) {
printf("%d, ", vbx_img8[j*IMAGE_WIDTH+i]);
}
printf("],\n");
}
printf("]\n");
#endif
#if LBP_CI
printf("Testing LBP Pattern CI\n");
errors += compare_LBPRestrictedCI_to_test_scalar_patterns(vbx_img, vbx_img8, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
#endif
#if BLIP
printf("Testing BLIP\n");
for(s = 1; s < 10; s++){
errors += compare_scalar_BLIP2_to_vector_BLIP(img, vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS, s);
}
#endif
#if 0
errors += compare_LBPRestrictedSums_to_test_scalar_sums_byte(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
errors += compare_LBPRestrictedSums2_to_test_scalar_sums_half(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
errors += compare_ScalarLBPRestrictedSums_to_test_scalar_sums_half(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
errors += compare_ScalarLBPRestrictedPatterns_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
errors += compare_LBPRestrictedPatterns2_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
errors += compare_LBPRestricted_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
/* overflow issues -- using bytes changes lbp pattern */
errors += compare_LBPRestrictedPatterns_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
/* requires SKIP_INTEGRALS 0 */
errors += compare_gen_integrals_to_vector_get_img(img, iImg, iiImg, vbx_img, vbx_iImg, vbx_iiImg, vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
/* redundant test, compare to test_scalar_patterns instead */
errors += compare_ScalarLBPRestrictedPatterns_to_SATBinaryPattern(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
errors += compare_SATBinaryPattern_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
errors += compare_LBPPassStage_to_restricted(vbx_img, log, face_lbp[0], window, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
#endif
#else // UNIT
#if PRINT
print_python_pixel(scalar_input, src, IMAGE_WIDTH, IMAGE_HEIGHT);
#endif
time_start = vbx_timestamp();
scalar_rgb2luma(scalar_short, input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH);
scalar_face_detect_luma(scalar_short, input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH, sdst);
time_stop = vbx_timestamp();
scalar_time = vbx_print_scalar_time(time_start, time_stop);
#if PRINT
print_python_pixel(scalar_input, sdst, IMAGE_WIDTH, IMAGE_HEIGHT);
#endif
printf("\nVector");
time_start = vbx_timestamp();
vector_face_detect((pixel *)vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH, 0, vdst);
time_stop = vbx_timestamp();
vbx_time = vbx_print_vector_time(time_start, time_stop, scalar_time);
#if PRINT
print_python_pixel(vbx_input, vdst, IMAGE_WIDTH, IMAGE_HEIGHT);
#endif
printf("\nVector Masked");
time_start = vbx_timestamp();
vector_face_detect((pixel *)vbx_input_masked, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH, 1, mdst);
time_stop = vbx_timestamp();
vbx_time_masked = vbx_print_vector_time(time_start, time_stop, scalar_time);
#if PRINT
print_python_pixel(vbx_input_masked, mdst, IMAGE_WIDTH, IMAGE_HEIGHT);
#endif
/* errors += match_array_pixel(input, vbx_input, "vector", IMAGE_WIDTH, IMAGE_HEIGHT, 0, MAX_PRINT_ERRORS, 0); */
/* errors += match_array_pixel(input, vbx_input_masked, "masked", IMAGE_WIDTH, IMAGE_HEIGHT, 0, MAX_PRINT_ERRORS, 0); */
errors += match_array_pixel(vbx_input, vbx_input_masked, "masked", IMAGE_WIDTH, IMAGE_HEIGHT, 0, MAX_PRINT_ERRORS, 0);
#endif // UNIT
}
VBX_TEST_END(errors);
return errors;
}
int VBX_T(vbw_vec_reverse_test_mm)()
{
unsigned int aN[] = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 12, 15, 16, 17, 20, 25, 31, 32, 33, 35, 40, 48, 60, 61, 62, 63, 64, 64, 65,
66, 67, 68, 70, 80, 90, 99, 100, 101, 110, 128, 128, 144, 144, 160, 160, 176, 176, 192, 192, 224, 224,
256, 256, 288, 288, 320, 320, 352, 352, 384, 384, 400, 450, 512, 550, 600, 650, 700, 768, 768, 900,
900, 1023, 1024, 1200, 1400, 1600, 1800, 2048, 2048, 2100, 2200, 2300, 2400, 2500, 2600, 2700, 2800,
2900, 3000, 3100, 3200, 3300, 3400, 3500, 3600, 3700, 3800, 3900, 4000, 4096, 4096, 4100, 4200, 4300,
4400, 4500, 4600, 4700, 4800, 4900, 5000, 6000, 7000, 8000, 8192, 8192, 9000, 10000, 11000, 12000,
13000, 14000, 15000, 16000, 16384, 16384, 20000, 25000, 30000, 32767, 32768, 32768, 35000, 40000,
45000, 50000, 55000, 60000, 65000, 65535, 65536, 65536, 65537, 100000, 128000, 256000, 333333, 528374,
528374 };
int retval;
unsigned int N;
unsigned int NBYTES;
unsigned int NREPS = 100;
unsigned int NREPSFORLARGE = 10;
unsigned int i,j,k;
vbx_timestamp_t start=0,finish=0;
for( i=0; i<sizeof(aN)/4; i++ ) {
N = aN[i];
//printf( "testing with vector size %d\n", N );
if(N > 10000) NREPS = NREPSFORLARGE;
NBYTES = N*sizeof(vbx_mm_t);
vbx_mm_t *src = (vbx_mm_t *) vbx_shared_malloc( NBYTES );
vbx_mm_t *dst = (vbx_mm_t *) vbx_shared_malloc( NBYTES );
//printf("bytes alloc: %d\n", NBYTES );
if( !src ) VBX_EXIT(-1);
if( !dst ) VBX_EXIT(-1);
for ( j=0; j<N; j++ ) {
dst[j] = -1; // Fill the destination with -1
src[j] = j; // Fill the source with enumerated values
}
// VBX_T(vbw_vec_reverse_ext)( dst, src, N );
/** measure performance of function call **/
start = vbx_timestamp();
for(k=0; __builtin_expect(k<NREPS,1); k++ ) {
retval = VBX_T(vbw_vec_reverse_ext)( dst, src, N );
}
finish = vbx_timestamp();
printf( "length %d (%s):\tvbware mm f():\t%llu", N, VBX_EXPAND_AND_QUOTE(BYTEHALFWORD), (unsigned long long) vbx_mxp_cycles((finish-start)/NREPS) );
#if VERIFY_VBWARE_ALGORITHM
VBX_T(verify_vector)( src, dst, N );
#else
printf(" [VERIFY OFF]");
#endif
printf("\treturn value: %X", retval);
/** measure performance of scalar **/
vbx_mm_t *A = vbx_remap_cached( src, N*sizeof(vbx_mm_t) ); // Use cached pointers for better performance
vbx_mm_t *B = vbx_remap_cached( dst, N*sizeof(vbx_mm_t) );
start = vbx_timestamp();
for(k=0; k<NREPS; k++ ) {
unsigned int m;
for(m=0; m<N; m++) {
B[N-1-m]=A[m];
}
vbx_dcache_flush( A, N*sizeof(vbx_mm_t) ); // Make sure to read from main memory
vbx_dcache_flush( B, N*sizeof(vbx_mm_t) ); // Make sure writes are committed to memory
}
finish = vbx_timestamp();
printf( "\tscalar (cache friendly):\t%llu", (unsigned long long) vbx_mxp_cycles((finish-start)/NREPS) );
#if VERIFY_SIMPLE_ALGORITHM
VBX_T(verify_vector)( src, dst, N );
#else
printf(" [VERIFY OFF]");
#endif
printf("\tcycles\n");
vbx_shared_free(src);
vbx_shared_free(dst);
}
printf("All tests passed successfully.\n");
return 0;
}
int main(void)
{
vbx_timestamp_t time_start, time_stop;
double scalar_time, vbx_time, vbx_time_masked;
int i, j, k, l, m, n;
int errors = 0;
vbx_test_init();
vbx_mxp_print_params();
pixel *input, *scalar_input, *vbx_input, *vbx_input_masked;
uint16_t *scalar_short;
input = (pixel *)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
scalar_input = (pixel *)vbx_remap_cached(input, IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
scalar_short = (uint16_t *)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(uint16_t));
vbx_input = (pixel *)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
vbx_input_masked = (pixel *)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
#if UNIT
unsigned short *img, *vbx_img;
unsigned int *iImg, *vbx_iImg;
unsigned int *iiImg, *vbx_iiImg;
img = (unsigned short*)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned short));
vbx_img = (unsigned short*)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned short));
iImg = (unsigned int*)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned int));
vbx_iImg = (unsigned int*)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned int));
iiImg = (unsigned int*)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned int));
vbx_iiImg = (unsigned int*)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned int));
#endif//UNIT
printf("Resolution = %dx%d\n", IMAGE_WIDTH, IMAGE_HEIGHT);
printf("Initializing data\n");
vbx_timestamp_start();
for(l = 0; l < 1; l++){
char *src;
char *sdst;
char *vdst;
char *mdst;
if(l == 0){
load_lenna(input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_lenna(vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_lenna(vbx_input_masked, IMAGE_WIDTH, IMAGE_HEIGHT);
printf("\nLenna\n");
src = "lenna";
sdst = "s_lenna";
vdst = "v_lenna";
mdst = "m_lenna";
}else if(l == 1){
load_ms(input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_ms(vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_ms(vbx_input_masked, IMAGE_WIDTH, IMAGE_HEIGHT);
printf("\nMicrosoft\n");
src = "ms";
sdst = "s_ms";
vdst = "v_ms";
mdst = "m_ms";
}else if(l == 2){
load_blank(input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_blank(vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT);
load_blank(vbx_input_masked, IMAGE_WIDTH, IMAGE_HEIGHT);
printf("\nblank\n");
src = "blank";
sdst = "s_blank";
vdst = "v_blank";
mdst = "m_blank";
}
#if UNIT
int window = 20;
int log=0;
while(((window/3)>>log) >= 2) log++;
#if LUT_CI
/* errors += compare_vbx_lut_to_vbx_lut_ci(1024, MAX_PRINT_ERRORS); */
#endif
#if LBP_CI
errors += compare_vbx_lbp_ci_to_scalar_patterns(vbx_img, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
#endif
errors += compare_scalar_rgb2luma_to_vbw_rgb2luma16(img, vbx_img, vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH, MAX_PRINT_ERRORS);
/* errors += compare_LBPRestrictedSums_to_test_scalar_sums_byte(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS); */
/* errors += compare_LBPRestrictedSums2_to_test_scalar_sums_half(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS); */
/* errors += compare_ScalarLBPRestrictedSums_to_test_scalar_sums_half(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS); */
/* errors += compare_ScalarLBPRestrictedPatterns_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS); */
/* errors += compare_LBPRestrictedPatterns2_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS); */
errors += compare_LBPRestricted_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
#if 0
/* overflow issues -- using bytes changes lbp pattern */
errors += compare_LBPRestrictedPatterns_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
/* requires SKIP_INTEGRALS 0 */
errors += compare_gen_integrals_to_vector_get_img(img, iImg, iiImg, vbx_img, vbx_iImg, vbx_iiImg, vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
/* currently last values have errors if the scaled images size is not an integer, width * f/ (f+1) */
errors += compare_scalar_BLIP2_to_vector_BLIP(img, vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
/* redundant test, compare to test_scalar_patterns instead */
errors += compare_ScalarLBPRestrictedPatterns_to_SATBinaryPattern(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
errors += compare_SATBinaryPattern_to_test_scalar_patterns(vbx_img, log, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
#endif
errors += compare_LBPPassStage_to_restricted(vbx_img, log, face_lbp[0], window, IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS);
#else // UNIT
#if PRINT
print_python_pixel(scalar_input, src, IMAGE_WIDTH, IMAGE_HEIGHT);
#endif
time_start = vbx_timestamp();
scalar_rgb2luma(scalar_short, input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH);
scalar_face_detect_luma(scalar_short, input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH, sdst);
time_stop = vbx_timestamp();
scalar_time = vbx_print_scalar_time(time_start, time_stop);
#if PRINT
print_python_pixel(scalar_input, sdst, IMAGE_WIDTH, IMAGE_HEIGHT);
#endif
printf("\nVector");
time_start = vbx_timestamp();
vector_face_detect((pixel *)vbx_input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH, 0, vdst);
time_stop = vbx_timestamp();
vbx_time = vbx_print_vector_time(time_start, time_stop, scalar_time);
#if PRINT
print_python_pixel(vbx_input, vdst, IMAGE_WIDTH, IMAGE_HEIGHT);
#endif
printf("\nVector Masked");
time_start = vbx_timestamp();
vector_face_detect((pixel *)vbx_input_masked, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_WIDTH, 1, mdst);
time_stop = vbx_timestamp();
vbx_time_masked = vbx_print_vector_time(time_start, time_stop, scalar_time);
#if PRINT
print_python_pixel(vbx_input_masked, mdst, IMAGE_WIDTH, IMAGE_HEIGHT);
#endif
/* errors += match_array_pixel(input, vbx_input, "vector", IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS, 0); */
/* errors += match_array_pixel(input, vbx_input_masked, "masked", IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS, 0); */
errors += match_array_pixel(vbx_input, vbx_input_masked, "masked", IMAGE_WIDTH, IMAGE_HEIGHT, MAX_PRINT_ERRORS, 0);
#endif // UNIT
}
VBX_TEST_END(errors);
return errors;
}
int vbw_vec_reverse_ext( vbx_mm_t *dst, vbx_mm_t *src, const unsigned int N )
{
typedef vbx_mm_t vbx_sp_t;
const int VBW_ROT16= sizeof(vbx_sp_t) <=sizeof(vbx_half_t);
const int VBW_ROT8= sizeof(vbx_sp_t)== sizeof(vbx_byte_t);
const int VBW_RSHIFT_T_TO_W= (sizeof(vbx_sp_t)==sizeof(vbx_word_t)? 0:
sizeof(vbx_sp_t)==sizeof(vbx_half_t)? 1:/*byte_sized*/2);
const int VBW_LSHIFT_W_TO_T= VBW_RSHIFT_T_TO_W;
// Catch when N is very small
if( N<4 ) {
unsigned int i = 0;
while(i<N) {
dst[N-i-1]=src[i];
i++;
}
return VBW_SUCCESS;
}
vbx_mxp_t *this_mxp = VBX_GET_THIS_MXP();
unsigned int SP_WIDTH_B = this_mxp->scratchpad_alignment_bytes;
unsigned int FREE_BYTES = vbx_sp_getfree();
// Catch when N is small enough that cached scalar does a better job
if( N <= MM_CACHED_SCALAR_THRESHOLD || FREE_BYTES < SP_WIDTH_B*5 ){
unsigned int i;
vbx_mm_t *A = (vbx_mm_t*)vbx_remap_cached(src,N*sizeof(vbx_mm_t));
vbx_mm_t *B = (vbx_mm_t*)vbx_remap_cached(dst,N*sizeof(vbx_mm_t));
for( i=0; i<N; i++ ) {
B[N-i-1]=A[i];
}
vbx_dcache_flush(B,N*sizeof(vbx_mm_t));
return VBW_SUCCESS;
}
unsigned int NUM_LANES = this_mxp->vector_lanes;
unsigned int tile_size_b = VBX_PAD_DN(((FREE_BYTES-SP_WIDTH_B)/2),SP_WIDTH_B);
unsigned int tile_size_w = tile_size_b/4;
unsigned int tile_size_t = tile_size_w << VBW_LSHIFT_W_TO_T;
unsigned int num_tiles = N / tile_size_t;
unsigned int rows_per_tile = tile_size_b / SP_WIDTH_B;
unsigned int tile_part_t = N - num_tiles * tile_size_t;
unsigned int threshold_w = NUM_LANES >= 32 ? VL1_THRESHOLD_V32_UP :
NUM_LANES == 16 ? VL1_THRESHOLD_V16 :
NUM_LANES == 8 ? VL1_THRESHOLD_V8 : UINT_MAX;
if(tile_part_t){
vbx_sp_push();
vbx_sp_t *v_0 = (vbx_sp_t *)vbx_sp_malloc(tile_part_t*sizeof(vbx_sp_t));
vbx_sp_t *v_1 = (vbx_sp_t *)vbx_sp_malloc(tile_part_t*sizeof(vbx_sp_t));
#if !VBX_SKIP_ALL_CHECKS
if( !v_0 || !v_1) {
VBX_PRINTF("vbx_sp_malloc failed when it was predetermined to have enough space.");
VBX_EXIT(-1);
}
#endif
vbx_dma_to_vector(v_0, src+N-tile_part_t, tile_part_t*sizeof(vbx_mm_t));
vbw_vec_reverse(v_1, v_0, tile_part_t);
vbx_dma_to_host(dst, v_1, tile_part_t*sizeof(vbx_sp_t));
dst += tile_part_t;
vbx_sp_pop();
}
if(!num_tiles) {
return VBW_SUCCESS;
}
vbx_sp_push();
vbx_word_t *v_mask = (vbx_word_t *)vbx_sp_malloc(SP_WIDTH_B);
vbx_word_t *v_scratch[2] = { (vbx_word_t *)vbx_sp_malloc(tile_size_b), (vbx_word_t *)vbx_sp_malloc(tile_size_b) };
vbx_word_t *result;
#if !VBX_SKIP_ALL_CHECKS
if( !v_scratch[0] || !v_scratch[1] || !v_mask ) {
VBX_PRINTF("vbx_sp_malloc failed when it was predetermined to have enough space.");
VBX_EXIT(-1);
}
#endif
src += (num_tiles - 1) * tile_size_t;
if( tile_size_w <= threshold_w) {
while( num_tiles ) {
vbx_dma_to_vector( v_scratch[0], src, tile_size_b );
if(VBW_ROT16){
vec_rev_rot16_w(v_scratch[1], v_scratch[0], tile_size_w);
}else{
vec_rev_w(v_scratch[1], v_scratch[0], tile_size_w);
}
if( VBW_ROT8){
vec_rot8_h( v_scratch[1], v_scratch[1], tile_size_w*2 );
}
vbx_dma_to_host( dst, v_scratch[1], tile_size_b );
dst += tile_size_t;
src -= tile_size_t;
num_tiles--;
}
} else {
while( num_tiles ) {
vbx_dma_to_vector( v_scratch[0], src, tile_size_b );
result = vec_rev_merge_w( v_scratch[1], v_scratch[0], tile_size_w, v_scratch[0], v_mask, SP_WIDTH_B,
rows_per_tile, VBW_ROT16 );
if(VBW_ROT8){
vec_rot8_h( result, result, tile_size_w*2 );
}
vbx_dma_to_host( dst, result, tile_size_b );
dst += tile_size_t;
src -= tile_size_t;
num_tiles--;
}
}
vbx_sp_pop();
return VBW_SUCCESS;
}
int main(void)
{
pixel *input;
pixel *scalar_input;
#if USE_LUMA
unsigned char *vbx_luma;
#endif
unsigned short *scalar_luma;
pixel *vbx_output;
pixel *scalar_output;
vbx_timestamp_t time_start, time_stop;
double scalar_time, vbx_time;
int x, y;
int errors = 0;
vbx_test_init();
vbx_mxp_print_params();
input = (pixel *)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
scalar_input = (pixel *)vbx_remap_cached(input, IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
#if USE_LUMA
vbx_luma = (unsigned char *)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned char));
#endif
scalar_luma = (unsigned short *)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(unsigned short));
vbx_output = (pixel *)vbx_shared_malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
scalar_output = (pixel *)malloc(IMAGE_WIDTH*IMAGE_HEIGHT*sizeof(pixel));
printf("\nInitializing data\n");
printf("Resolution = %dx%d\n", IMAGE_WIDTH, IMAGE_HEIGHT);
init_matrix(input, IMAGE_WIDTH, IMAGE_HEIGHT);
printf("Starting Sobel 3x3 edge-detection test\n");
#if USE_LUMA
scalar_rgb2luma(scalar_luma, scalar_input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_PITCH);
#endif
vbx_timestamp_start();
time_start = vbx_timestamp();
#if !USE_LUMA
scalar_rgb2luma(scalar_luma, scalar_input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_PITCH);
#endif
scalar_sobel_argb32_3x3(scalar_output, scalar_luma, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_PITCH, RENORM_AMOUNT);
time_stop = vbx_timestamp();
scalar_time = vbx_print_scalar_time(time_start, time_stop);
#if USE_LUMA
vbw_rgb2luma8(vbx_luma, (unsigned *)input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_PITCH);
#endif
vbx_timestamp_start();
time_start = vbx_timestamp();
#if USE_LUMA
vbw_sobel_luma8_3x3((unsigned *)vbx_output, vbx_luma, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_PITCH, RENORM_AMOUNT);
#else
vbw_sobel_argb32_3x3((unsigned *)vbx_output, (unsigned *)input, IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_PITCH, RENORM_AMOUNT);
#endif
time_stop = vbx_timestamp();
vbx_time = vbx_print_vector_time(time_start, time_stop, scalar_time);
for (y = 0; y < IMAGE_HEIGHT; y++) {
for (x = 0; x < IMAGE_WIDTH; x++) {
#if USE_LUMA
if (scalar_luma[y*IMAGE_WIDTH+x] != vbx_luma[y*IMAGE_WIDTH+x]) {
if (errors < MAX_PRINT_ERRORS) {
printf("Y Error at %d, %d: Expected = %02X, got = %02X\n",
y, x, scalar_luma[y*IMAGE_WIDTH+x], vbx_luma[y*IMAGE_WIDTH+x]);
}
errors++;
}
#endif
if (scalar_output[y*IMAGE_WIDTH+x].r != vbx_output[y*IMAGE_WIDTH+x].r) {
if (errors < MAX_PRINT_ERRORS) {
printf("R Error at %d, %d: Expected = %02X, got = %02X\n",
y, x, scalar_output[y*IMAGE_WIDTH+x].r, vbx_output[y*IMAGE_WIDTH+x].r);
}
errors++;
}
if (scalar_output[y*IMAGE_WIDTH+x].g != vbx_output[y*IMAGE_WIDTH+x].g) {
if (errors < MAX_PRINT_ERRORS) {
printf("G Error at %d, %d: Expected = %02X, got = %02X\n",
y, x, scalar_output[y*IMAGE_WIDTH+x].g, vbx_output[y*IMAGE_WIDTH+x].g);
}
errors++;
}
if (scalar_output[y*IMAGE_WIDTH+x].b != vbx_output[y*IMAGE_WIDTH+x].b) {
if (errors < MAX_PRINT_ERRORS) {
printf("B Error at %d, %d: Expected = %02X, got = %02X\n",
y, x, scalar_output[y*IMAGE_WIDTH+x].b, vbx_output[y*IMAGE_WIDTH+x].b);
}
errors++;
}
}
}
VBX_TEST_END(errors);
return errors;
}
