static void
orc_multiply_and_acc_biref_Xxn_s16_u8 (int16_t * d1, int d1_stride,
const int16_t * s1, int s1_stride, const uint8_t * s2, int s2_stride,
const uint8_t * s3, int s3_stride, int p1, int p2, int n, int m)
{
OrcExecutor _ex, *ex = &_ex;
OrcProgram *p;
void (*func) (OrcExecutor *);
p = motion_funcs[n >> 1].block_accumulate_biref;
ex->program = p;
ex->n = n;
ORC_EXECUTOR_M (ex) = m;
ex->arrays[ORC_VAR_D1] = d1;
ex->params[ORC_VAR_D1] = d1_stride;
ex->arrays[ORC_VAR_S1] = (void *) s1;
ex->params[ORC_VAR_S1] = s1_stride;
ex->arrays[ORC_VAR_S2] = (void *) s2;
ex->params[ORC_VAR_S2] = s2_stride;
ex->arrays[ORC_VAR_S3] = (void *) s3;
ex->params[ORC_VAR_S3] = s3_stride;
ex->params[ORC_VAR_P1] = p1;
ex->params[ORC_VAR_P2] = p2;
func = p->code_exec;
func (ex);
}
void
orc_blend_u8 (guint8 * d1, int d1_stride, const guint8 * s1, int s1_stride,
int p1, int n, int m)
{
OrcExecutor _ex, *ex = &_ex;
static int p_inited = 0;
static OrcProgram *p = 0;
void (*func) (OrcExecutor *);
if (!p_inited) {
orc_once_mutex_lock ();
if (!p_inited) {
OrcCompileResult result;
p = orc_program_new ();
orc_program_set_2d (p);
orc_program_set_name (p, "orc_blend_u8");
orc_program_set_backup_function (p, _backup_orc_blend_u8);
orc_program_add_destination (p, 1, "d1");
orc_program_add_source (p, 1, "s1");
orc_program_add_constant (p, 1, 8, "c1");
orc_program_add_parameter (p, 2, "p1");
orc_program_add_temporary (p, 2, "t1");
orc_program_add_temporary (p, 2, "t2");
orc_program_append (p, "convubw", ORC_VAR_T1, ORC_VAR_D1, ORC_VAR_D1);
orc_program_append (p, "convubw", ORC_VAR_T2, ORC_VAR_S1, ORC_VAR_D1);
orc_program_append (p, "subw", ORC_VAR_T2, ORC_VAR_T2, ORC_VAR_T1);
orc_program_append (p, "mullw", ORC_VAR_T2, ORC_VAR_T2, ORC_VAR_P1);
orc_program_append (p, "shlw", ORC_VAR_T1, ORC_VAR_T1, ORC_VAR_C1);
orc_program_append (p, "addw", ORC_VAR_T2, ORC_VAR_T1, ORC_VAR_T2);
orc_program_append (p, "shruw", ORC_VAR_T2, ORC_VAR_T2, ORC_VAR_C1);
orc_program_append (p, "convsuswb", ORC_VAR_D1, ORC_VAR_T2, ORC_VAR_D1);
result = orc_program_compile (p);
}
p_inited = TRUE;
orc_once_mutex_unlock ();
}
ex->program = p;
ex->n = n;
ORC_EXECUTOR_M (ex) = m;
ex->arrays[ORC_VAR_D1] = d1;
ex->params[ORC_VAR_D1] = d1_stride;
ex->arrays[ORC_VAR_S1] = (void *) s1;
ex->params[ORC_VAR_S1] = s1_stride;
ex->params[ORC_VAR_P1] = p1;
func = p->code_exec;
func (ex);
}
void
orc_executor_emulate (OrcExecutor *ex)
{
int i;
int j;
int k;
int m, m_index;
OrcCode *code;
OrcInstruction *insn;
OrcStaticOpcode *opcode;
OrcOpcodeExecutor *opcode_ex;
void *tmpspace[ORC_N_COMPILER_VARIABLES] = { 0 };
if (ex->program) {
code = ex->program->orccode;
} else {
code = (OrcCode *)ex->arrays[ORC_VAR_A2];
}
ex->accumulators[0] = 0;
ex->accumulators[1] = 0;
ex->accumulators[2] = 0;
ex->accumulators[3] = 0;
ORC_DEBUG("emulating");
memset (&opcode_ex, 0, sizeof(opcode_ex));
if (code == NULL) {
ORC_ERROR("attempt to run program that failed to compile");
ORC_ASSERT(0);
}
if (code->is_2d) {
m = ORC_EXECUTOR_M(ex);
} else {
m = 1;
}
for(i=0;i<ORC_N_COMPILER_VARIABLES;i++){
OrcCodeVariable *var = code->vars + i;
if (var->size) {
tmpspace[i] = malloc(ORC_MAX_VAR_SIZE * CHUNK_SIZE);
}
}
opcode_ex = malloc(sizeof(OrcOpcodeExecutor)*code->n_insns);
for(j=0;j<code->n_insns;j++){
insn = code->insns + j;
opcode = insn->opcode;
opcode_ex[j].emulateN = opcode->emulateN;
opcode_ex[j].shift = 0;
if (insn->flags & ORC_INSTRUCTION_FLAG_X2) {
opcode_ex[j].shift = 1;
} else if (insn->flags & ORC_INSTRUCTION_FLAG_X4) {
opcode_ex[j].shift = 2;
}
for(k=0;k<ORC_STATIC_OPCODE_N_SRC;k++) {
OrcCodeVariable *var = code->vars + insn->src_args[k];
if (opcode->src_size[k] == 0) continue;
if (var->vartype == ORC_VAR_TYPE_CONST) {
opcode_ex[j].src_ptrs[k] = tmpspace[insn->src_args[k]];
/* FIXME hack */
load_constant (tmpspace[insn->src_args[k]], 8,
var->value.i);
} else if (var->vartype == ORC_VAR_TYPE_PARAM) {
opcode_ex[j].src_ptrs[k] = tmpspace[insn->src_args[k]];
/* FIXME hack */
load_constant (tmpspace[insn->src_args[k]], 8,
(orc_uint64)(orc_uint32)ex->params[insn->src_args[k]] |
(((orc_uint64)(orc_uint32)ex->params[insn->src_args[k] +
(ORC_VAR_T1 - ORC_VAR_P1)])<<32));
} else if (var->vartype == ORC_VAR_TYPE_TEMP) {
opcode_ex[j].src_ptrs[k] = tmpspace[insn->src_args[k]];
} else if (var->vartype == ORC_VAR_TYPE_SRC) {
if (ORC_PTR_TO_INT(ex->arrays[insn->src_args[k]]) & (var->size - 1)) {
ORC_ERROR("Unaligned array for src%d, program %s",
(insn->src_args[k]-ORC_VAR_S1), ex->program->name);
}
opcode_ex[j].src_ptrs[k] = ex->arrays[insn->src_args[k]];
} else if (var->vartype == ORC_VAR_TYPE_DEST) {
if (ORC_PTR_TO_INT(ex->arrays[insn->src_args[k]]) & (var->size - 1)) {
ORC_ERROR("Unaligned array for dest%d, program %s",
(insn->src_args[k]-ORC_VAR_D1), ex->program->name);
}
opcode_ex[j].src_ptrs[k] = ex->arrays[insn->src_args[k]];
}
}
for(k=0;k<ORC_STATIC_OPCODE_N_DEST;k++) {
OrcCodeVariable *var = code->vars + insn->dest_args[k];
if (opcode->dest_size[k] == 0) continue;
if (var->vartype == ORC_VAR_TYPE_TEMP) {
ORC_DEBUG("dest vartype tmp %d", insn->dest_args[k]);
opcode_ex[j].dest_ptrs[k] = tmpspace[insn->dest_args[k]];
} else if (var->vartype == ORC_VAR_TYPE_ACCUMULATOR) {
opcode_ex[j].dest_ptrs[k] =
&ex->accumulators[insn->dest_args[k] - ORC_VAR_A1];
} else if (var->vartype == ORC_VAR_TYPE_DEST) {
if (ORC_PTR_TO_INT(ex->arrays[insn->dest_args[k]]) & (var->size - 1)) {
ORC_ERROR("Unaligned array for dest%d, program %s",
(insn->dest_args[k]-ORC_VAR_D1), ex->program->name);
}
opcode_ex[j].dest_ptrs[k] = ex->arrays[insn->dest_args[k]];
}
}
ORC_DEBUG("opcode %s %p %p %p", opcode->name,
opcode_ex[j].dest_ptrs[0], opcode_ex[j].src_ptrs[0],
opcode_ex[j].src_ptrs[1]);
}
ORC_DEBUG("src ptr %p stride %d", ex->arrays[ORC_VAR_S1], ex->params[ORC_VAR_S1]);
for(m_index=0;m_index<m;m_index++){
ORC_DEBUG("m_index %d m %d", m_index, m);
for(j=0;j<code->n_insns;j++){
insn = code->insns + j;
opcode = insn->opcode;
for(k=0;k<ORC_STATIC_OPCODE_N_SRC;k++) {
OrcCodeVariable *var = code->vars + insn->src_args[k];
if (opcode->src_size[k] == 0) continue;
if (var->vartype == ORC_VAR_TYPE_SRC) {
opcode_ex[j].src_ptrs[k] =
ORC_PTR_OFFSET(ex->arrays[insn->src_args[k]],
ex->params[insn->src_args[k]]*m_index);
} else if (var->vartype == ORC_VAR_TYPE_DEST) {
opcode_ex[j].src_ptrs[k] =
ORC_PTR_OFFSET(ex->arrays[insn->src_args[k]],
ex->params[insn->src_args[k]]*m_index);
}
}
for(k=0;k<ORC_STATIC_OPCODE_N_DEST;k++) {
OrcCodeVariable *var = code->vars + insn->dest_args[k];
if (opcode->dest_size[k] == 0) continue;
if (var->vartype == ORC_VAR_TYPE_DEST) {
opcode_ex[j].dest_ptrs[k] =
ORC_PTR_OFFSET(ex->arrays[insn->dest_args[k]],
ex->params[insn->dest_args[k]]*m_index);
}
}
}
for(i=0;i<ex->n;i+=CHUNK_SIZE){
for(j=0;j<code->n_insns;j++){
if (ex->n - i >= CHUNK_SIZE) {
opcode_ex[j].emulateN (opcode_ex + j, i, CHUNK_SIZE << opcode_ex[j].shift);
} else {
opcode_ex[j].emulateN (opcode_ex + j, i, (ex->n - i) << opcode_ex[j].shift);
}
}
}
}
free (opcode_ex);
for(i=0;i<ORC_N_COMPILER_VARIABLES;i++){
if (tmpspace[i]) free (tmpspace[i]);
}
}
void
orc_executor_set_m (OrcExecutor *ex, int m)
{
ORC_EXECUTOR_M(ex) = m;
}