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
}
void
vips_vector_asm2( VipsVector *vector,
const char *op, const char *a, const char *b )
{
vector->n_instruction += 1;
#ifdef DEBUG
printf( " %s %s %s\n", op, a, b );
#endif /*DEBUG*/
#ifdef HAVE_ORC
orc_program_append_ds_str( vector->program, op, a, b );
#endif /*HAVE_ORC*/
}
void
vips_vector_asm2( VipsVector *vector,
const char *op, const char *a, const char *b )
{
vector->n_instruction += 1;
#ifdef DEBUG
printf( " %s %s %s\n", op, a, b );
#endif /*DEBUG*/
#ifdef HAVE_ORC
#ifdef DEBUG_TRACE
printf( "orc_program_append_ds_str( %s, \"%s\", \"%s\", \"%s\" );\n",
vector->unique_name, op, a, b );
#endif /*DEBUG_TRACE*/
orc_program_append_ds_str( vector->program, op, a, b );
#endif /*HAVE_ORC*/
}
