void
vips_executor_run( VipsExecutor *executor )
{
#ifdef HAVE_ORC
orc_executor_run( &executor->executor );
#endif /*HAVE_ORC*/
}
static int
sum_square_diff_u8 (uint8_t * s1, uint8_t * s2, int n)
{
#ifndef HAVE_ORC
int sum = 0;
int i;
int x;
for (i = 0; i < n; i++) {
x = s1[i] - s2[i];
sum += x * x;
}
return sum;
#else
static OrcProgram *p = NULL;
OrcExecutor *ex;
int val;
if (p == NULL) {
OrcCompileResult ret;
p = orc_program_new_ass (4, 1, 1);
orc_program_add_temporary (p, 2, "t1");
orc_program_add_temporary (p, 2, "t2");
orc_program_add_temporary (p, 4, "t3");
orc_program_append_ds_str (p, "convubw", "t1", "s1");
orc_program_append_ds_str (p, "convubw", "t2", "s2");
orc_program_append_str (p, "subw", "t1", "t1", "t2");
orc_program_append_str (p, "mullw", "t1", "t1", "t1");
orc_program_append_ds_str (p, "convuwl", "t3", "t1");
orc_program_append_ds_str (p, "accl", "a1", "t3");
ret = orc_program_compile (p);
if (!ORC_COMPILE_RESULT_IS_SUCCESSFUL (ret)) {
GST_ERROR ("Orc compiler failure");
return 0;
}
}
ex = orc_executor_new (p);
orc_executor_set_n (ex, n);
orc_executor_set_array_str (ex, "s1", s1);
orc_executor_set_array_str (ex, "s2", s2);
orc_executor_run (ex);
val = orc_executor_get_accumulator (ex, 0);
orc_executor_free (ex);
return val;
#endif
}
double
orc_test_performance_full (OrcProgram *program, int flags,
const char *target_name)
{
OrcExecutor *ex;
int n;
int m;
OrcArray *dest_exec[4] = { NULL, NULL, NULL, NULL };
OrcArray *dest_emul[4] = { NULL, NULL, NULL, NULL };
OrcArray *src[8] = { NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL };
int i, j;
OrcCompileResult result;
OrcProfile prof;
double ave, std;
OrcTarget *target;
int misalignment;
ORC_DEBUG ("got here");
target = orc_target_get_by_name (target_name);
if (!(flags & ORC_TEST_FLAGS_BACKUP)) {
unsigned int flags;
flags = orc_target_get_default_flags (target);
result = orc_program_compile_full (program, target, flags);
if (!ORC_COMPILE_RESULT_IS_SUCCESSFUL(result)) {
//printf("compile failed\n");
orc_program_reset (program);
return 0;
}
}
if (program->constant_n > 0) {
n = program->constant_n;
} else {
//n = 64 + (orc_random(&rand_context)&0xf);
n = 1000;
}
ex = orc_executor_new (program);
orc_executor_set_n (ex, n);
if (program->is_2d) {
if (program->constant_m > 0) {
m = program->constant_m;
} else {
m = 8 + (orc_random(&rand_context)&0xf);
}
} else {
m = 1;
}
orc_executor_set_m (ex, m);
ORC_DEBUG("size %d %d", ex->n, ex->params[ORC_VAR_A1]);
misalignment = 0;
for(i=0;i<ORC_N_VARIABLES;i++){
if (program->vars[i].name == NULL) continue;
if (program->vars[i].vartype == ORC_VAR_TYPE_SRC) {
src[i-ORC_VAR_S1] = orc_array_new (n, m, program->vars[i].size,
misalignment);
orc_array_set_random (src[i-ORC_VAR_S1], &rand_context);
misalignment++;
} else if (program->vars[i].vartype == ORC_VAR_TYPE_DEST) {
dest_exec[i-ORC_VAR_D1] = orc_array_new (n, m, program->vars[i].size,
misalignment);
orc_array_set_pattern (dest_exec[i], ORC_OOB_VALUE);
dest_emul[i-ORC_VAR_D1] = orc_array_new (n, m, program->vars[i].size,
misalignment);
orc_array_set_pattern (dest_emul[i], ORC_OOB_VALUE);
misalignment++;
} else if (program->vars[i].vartype == ORC_VAR_TYPE_PARAM) {
orc_executor_set_param (ex, i, 2);
}
}
ORC_DEBUG ("running");
orc_profile_init (&prof);
for(i=0;i<10;i++){
orc_executor_set_n (ex, n);
orc_executor_set_m (ex, m);
for(j=0;j<ORC_N_VARIABLES;j++){
if (program->vars[j].vartype == ORC_VAR_TYPE_DEST) {
orc_executor_set_array (ex, j, dest_exec[j-ORC_VAR_D1]->data);
orc_executor_set_stride (ex, j, dest_exec[j-ORC_VAR_D1]->stride);
}
if (program->vars[j].vartype == ORC_VAR_TYPE_SRC) {
orc_executor_set_array (ex, j, src[j-ORC_VAR_S1]->data);
orc_executor_set_stride (ex, j, src[j-ORC_VAR_S1]->stride);
}
}
if (flags & ORC_TEST_FLAGS_BACKUP) {
orc_profile_start (&prof);
orc_executor_run_backup (ex);
orc_profile_stop (&prof);
} else if (flags & ORC_TEST_FLAGS_EMULATE) {
orc_profile_start (&prof);
orc_executor_emulate (ex);
orc_profile_stop (&prof);
} else {
orc_profile_start (&prof);
orc_executor_run (ex);
orc_profile_stop (&prof);
}
}
ORC_DEBUG ("done running");
orc_profile_get_ave_std (&prof, &ave, &std);
for(i=0;i<4;i++){
if (dest_exec[i]) orc_array_free (dest_exec[i]);
if (dest_emul[i]) orc_array_free (dest_emul[i]);
}
for(i=0;i<8;i++){
if (src[i]) orc_array_free (src[i]);
}
orc_executor_free (ex);
orc_program_reset (program);
return ave/(n*m);
}
OrcTestResult
orc_test_compare_output_full (OrcProgram *program, int flags)
{
OrcExecutor *ex;
int n;
int m;
OrcArray *dest_exec[4] = { NULL, NULL, NULL, NULL };
OrcArray *dest_emul[4] = { NULL, NULL, NULL, NULL };
OrcArray *src[8] = { NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL };
int i;
int j;
int k;
int have_dest = FALSE;
OrcCompileResult result;
int have_acc = FALSE;
int acc_exec = 0, acc_emul = 0;
int ret = ORC_TEST_OK;
int bad = 0;
int misalignment;
ORC_DEBUG ("got here");
{
OrcTarget *target;
unsigned int flags;
target = orc_target_get_default ();
flags = orc_target_get_default_flags (target);
result = orc_program_compile_full (program, target, flags);
if (ORC_COMPILE_RESULT_IS_FATAL(result)) {
ret = ORC_TEST_FAILED;
goto out;
}
if (!ORC_COMPILE_RESULT_IS_SUCCESSFUL(result)) {
ret = ORC_TEST_INDETERMINATE;
goto out;
}
}
if (program->constant_n > 0) {
n = program->constant_n;
} else {
n = 64 + (orc_random(&rand_context)&0xf);
}
ex = orc_executor_new (program);
orc_executor_set_n (ex, n);
if (program->is_2d) {
if (program->constant_m > 0) {
m = program->constant_m;
} else {
m = 8 + (orc_random(&rand_context)&0xf);
}
} else {
m = 1;
}
orc_executor_set_m (ex, m);
ORC_DEBUG("size %d %d", ex->n, ex->params[ORC_VAR_A1]);
misalignment = 0;
for(i=0;i<ORC_N_VARIABLES;i++){
if (program->vars[i].name == NULL) continue;
if (program->vars[i].vartype == ORC_VAR_TYPE_SRC) {
src[i-ORC_VAR_S1] = orc_array_new (n, m, program->vars[i].size,
misalignment);
orc_array_set_random (src[i-ORC_VAR_S1], &rand_context);
misalignment++;
} else if (program->vars[i].vartype == ORC_VAR_TYPE_DEST) {
dest_exec[i-ORC_VAR_D1] = orc_array_new (n, m, program->vars[i].size,
misalignment);
orc_array_set_pattern (dest_exec[i], ORC_OOB_VALUE);
dest_emul[i-ORC_VAR_D1] = orc_array_new (n, m, program->vars[i].size,
misalignment);
orc_array_set_pattern (dest_emul[i], ORC_OOB_VALUE);
misalignment++;
} else if (program->vars[i].vartype == ORC_VAR_TYPE_PARAM) {
switch (program->vars[i].param_type) {
case ORC_PARAM_TYPE_INT:
orc_executor_set_param (ex, i, 2);
break;
case ORC_PARAM_TYPE_FLOAT:
orc_executor_set_param_float (ex, i, 2.0);
break;
case ORC_PARAM_TYPE_INT64:
orc_executor_set_param_int64 (ex, i, 2);
break;
case ORC_PARAM_TYPE_DOUBLE:
orc_executor_set_param_double (ex, i, 2.0);
break;
}
}
}
for(i=0;i<ORC_N_VARIABLES;i++){
if (program->vars[i].vartype == ORC_VAR_TYPE_DEST) {
orc_executor_set_array (ex, i, dest_exec[i-ORC_VAR_D1]->data);
orc_executor_set_stride (ex, i, dest_exec[i-ORC_VAR_D1]->stride);
have_dest = TRUE;
}
if (program->vars[i].vartype == ORC_VAR_TYPE_SRC) {
orc_executor_set_array (ex, i, src[i-ORC_VAR_S1]->data);
orc_executor_set_stride (ex, i, src[i-ORC_VAR_S1]->stride);
}
}
ORC_DEBUG ("running");
if (flags & ORC_TEST_FLAGS_BACKUP) {
orc_executor_run_backup (ex);
} else {
orc_executor_run (ex);
}
ORC_DEBUG ("done running");
for(i=0;i<ORC_N_VARIABLES;i++){
if (program->vars[i].vartype == ORC_VAR_TYPE_ACCUMULATOR) {
acc_exec = ex->accumulators[0];
have_acc = TRUE;
}
}
for(i=0;i<ORC_N_VARIABLES;i++){
if (program->vars[i].vartype == ORC_VAR_TYPE_DEST) {
orc_executor_set_array (ex, i, dest_emul[i]->data);
orc_executor_set_stride (ex, i, dest_emul[i]->stride);
}
if (program->vars[i].vartype == ORC_VAR_TYPE_SRC) {
ORC_DEBUG("setting array %p", src[i-ORC_VAR_S1]->data);
orc_executor_set_array (ex, i, src[i-ORC_VAR_S1]->data);
orc_executor_set_stride (ex, i, src[i-ORC_VAR_S1]->stride);
}
}
orc_executor_emulate (ex);
for(i=0;i<ORC_N_VARIABLES;i++){
if (program->vars[i].vartype == ORC_VAR_TYPE_ACCUMULATOR) {
acc_emul = ex->accumulators[0];
}
}
for(k=ORC_VAR_D1;k<ORC_VAR_D1+4;k++){
if (program->vars[k].size > 0) {
if (!orc_array_compare (dest_exec[k-ORC_VAR_D1], dest_emul[k-ORC_VAR_D1], flags)) {
printf("dest array %d bad\n", k);
bad = TRUE;
}
if (!orc_array_check_out_of_bounds (dest_exec[k-ORC_VAR_D1])) {
printf("out of bounds failure\n");
ret = ORC_TEST_FAILED;
}
}
}
if (bad) {
for(j=0;j<m;j++){
for(i=0;i<n;i++){
orc_uint64 a,b;
int l;
int line_bad = 0;
printf("%2d %2d:", i, j);
for(l=ORC_VAR_S1;l<ORC_VAR_S1+8;l++){
if (program->vars[l].size > 0) {
if (flags & ORC_TEST_FLAGS_FLOAT) {
print_array_val_float (src[l-ORC_VAR_S1], i, j);
} else {
print_array_val_hex (src[l-ORC_VAR_S1], i, j);
}
}
}
printf(" ->");
for(l=ORC_VAR_D1;l<ORC_VAR_D1+4;l++){
if (program->vars[l].size > 0) {
if (flags & ORC_TEST_FLAGS_FLOAT) {
a = print_array_val_float (dest_emul[l-ORC_VAR_D1], i, j);
b = print_array_val_float (dest_exec[l-ORC_VAR_D1], i, j);
if (!float_compare (dest_emul[l-ORC_VAR_D1], dest_exec[l-ORC_VAR_D1], i, j) != 0) {
line_bad = TRUE;
}
} else {
a = print_array_val_hex (dest_emul[l-ORC_VAR_D1], i, j);
b = print_array_val_hex (dest_exec[l-ORC_VAR_D1], i, j);
if (a != b) {
line_bad = TRUE;
}
}
}
}
if (line_bad) {
printf(" *");
}
printf("\n");
}
}
ret = ORC_TEST_FAILED;
}
if (have_acc) {
if (acc_emul != acc_exec) {
for(j=0;j<m;j++){
for(i=0;i<n;i++){
printf("%2d %2d:", i, j);
for(k=0;k<ORC_N_VARIABLES;k++){
if (program->vars[k].name == NULL) continue;
if (program->vars[k].vartype == ORC_VAR_TYPE_SRC &&
program->vars[k].size > 0) {
if (flags & ORC_TEST_FLAGS_FLOAT) {
print_array_val_float (src[k-ORC_VAR_S1], i, j);
} else {
print_array_val_signed (src[k-ORC_VAR_S1], i, j);
}
}
}
printf(" -> acc\n");
}
}
printf("acc %d %d\n", acc_emul, acc_exec);
ret = ORC_TEST_FAILED;
}
}
if (ret == ORC_TEST_FAILED) {
printf("%s", orc_program_get_asm_code (program));
}
for(i=0;i<4;i++){
if (dest_exec[i]) orc_array_free (dest_exec[i]);
if (dest_emul[i]) orc_array_free (dest_emul[i]);
}
for(i=0;i<8;i++){
if (src[i]) orc_array_free (src[i]);
}
orc_executor_free (ex);
out:
orc_program_reset (program);
return ret;
}
void
show (OrcProgram *program)
{
OrcCompileResult result;
OrcTarget *target;
const char *target_name;
unsigned int target_flags;
int n, m;
OrcExecutor *ex;
OrcArray *dest[4] = { NULL, NULL, NULL, NULL };
OrcArray *src[8] = { NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL };
int i,j;
OrcRandomContext rand_context = { 0 };
target_name = NULL;
target = orc_target_get_by_name (target_name);
target_flags = orc_target_get_default_flags (target);
result = orc_program_compile_full (program, target, target_flags);
if (!ORC_COMPILE_RESULT_IS_SUCCESSFUL(result)) {
printf("%s: compile failed\n", program->name);
return;
}
printf("%s:\n", program->name);
if (program->constant_n > 0) {
n = program->constant_n;
} else {
n = array_n;
}
ex = orc_executor_new (program);
orc_executor_set_n (ex, n);
if (program->is_2d) {
if (program->constant_m > 0) {
m = program->constant_m;
} else {
m = 2;
}
} else {
m = 1;
}
orc_executor_set_m (ex, m);
for(i=0;i<ORC_N_VARIABLES;i++){
if (program->vars[i].name == NULL) continue;
if (program->vars[i].vartype == ORC_VAR_TYPE_SRC) {
src[i-ORC_VAR_S1] = orc_array_new (n, m, program->vars[i].size, 0);
orc_array_set_random (src[i-ORC_VAR_S1], &rand_context);
} else if (program->vars[i].vartype == ORC_VAR_TYPE_DEST) {
dest[i-ORC_VAR_D1] = orc_array_new (n, m, program->vars[i].size, 0);
orc_array_set_pattern (dest[i], ORC_OOB_VALUE);
} else if (program->vars[i].vartype == ORC_VAR_TYPE_PARAM) {
switch (program->vars[i].param_type) {
case ORC_PARAM_TYPE_INT:
orc_executor_set_param (ex, i, 2);
break;
case ORC_PARAM_TYPE_FLOAT:
orc_executor_set_param_float (ex, i, 2.0);
break;
case ORC_PARAM_TYPE_INT64:
orc_executor_set_param_int64 (ex, i, 2);
break;
case ORC_PARAM_TYPE_DOUBLE:
orc_executor_set_param_double (ex, i, 2.0);
break;
default:
ORC_ASSERT(0);
}
}
}
orc_executor_set_n (ex, n);
orc_executor_set_m (ex, m);
for(j=0;j<ORC_N_VARIABLES;j++){
if (program->vars[j].vartype == ORC_VAR_TYPE_DEST) {
orc_executor_set_array (ex, j, dest[j-ORC_VAR_D1]->data);
orc_executor_set_stride (ex, j, dest[j-ORC_VAR_D1]->stride);
}
if (program->vars[j].vartype == ORC_VAR_TYPE_SRC) {
orc_executor_set_array (ex, j, src[j-ORC_VAR_S1]->data);
orc_executor_set_stride (ex, j, src[j-ORC_VAR_S1]->stride);
}
}
orc_executor_run (ex);
{
int i,j;
for(j=0;j<m;j++){
for(i=0;i<n;i++){
int l;
printf("%2d %2d:", i, j);
for(l=ORC_VAR_S1;l<ORC_VAR_S1+8;l++){
if (program->vars[l].size > 0) {
switch (format) {
case FORMAT_FLOAT:
print_array_val_float (src[l-ORC_VAR_S1], i, j);
break;
case FORMAT_HEX:
print_array_val_hex (src[l-ORC_VAR_S1], i, j);
break;
case FORMAT_SIGNED:
print_array_val_signed (src[l-ORC_VAR_S1], i, j);
break;
case FORMAT_UNSIGNED:
print_array_val_unsigned (src[l-ORC_VAR_S1], i, j);
break;
}
}
}
printf(" ->");
for(l=ORC_VAR_D1;l<ORC_VAR_D1+4;l++){
if (program->vars[l].size > 0) {
switch (format) {
case FORMAT_FLOAT:
print_array_val_float (dest[l-ORC_VAR_D1], i, j);
break;
case FORMAT_HEX:
print_array_val_hex (dest[l-ORC_VAR_D1], i, j);
break;
case FORMAT_SIGNED:
print_array_val_signed (dest[l-ORC_VAR_D1], i, j);
break;
case FORMAT_UNSIGNED:
print_array_val_unsigned (dest[l-ORC_VAR_D1], i, j);
break;
}
}
}
printf("\n");
}
}
}
for(i=0;i<4;i++){
if (dest[i]) orc_array_free (dest[i]);
}
for(i=0;i<8;i++){
if (src[i]) orc_array_free (src[i]);
}
orc_executor_free (ex);
}
