Revision 2a406b243730a0a50aa99355786b3ac41340a386 authored by Keno Fischer on 06 July 2023, 21:20:40 UTC, committed by GitHub on 06 July 2023, 21:20:40 UTC
* Add pattern matching for `typeof` into field type tparam This PR allows full elimination of the following, even in ill-typed code. ``` struct TParamTypeofTest{T} x::T @eval TParamTypeofTest(x) = $(Expr(:new, :(TParamTypeofTest{typeof(x)}), :x)) end function tparam_typeof_test_elim(x) TParamTypeofTest(x).x end ``` Before this PR, we would get: ``` julia> code_typed(tparam_typeof_test_elim, Tuple{Any}) 1-element Vector{Any}: CodeInfo( 1 ─ %1 = Main.typeof(x)::DataType │ %2 = Core.apply_type(Main.TParamTypeofTest, %1)::Type{TParamTypeofTest{_A}} where _A │ %new(%2, x)::TParamTypeofTest └── return x ``` Where the `new` is non-eliminable, because the compiler did not know that `x::_A`. Fix this by pattern matching this particular pattern (where the condition is guaranteed, because we computed `_A` by `typeof`). This is not particularly general, but this pattern comes up a lot, so it's surprisingly effective. * add test case for optimizing multiple abstract fields * improve robustness --------- Co-authored-by: Shuhei Kadowaki <aviatesk@gmail.com>
1 parent 46477cc
precompile_utils.c
// f{<:Union{...}}(...) is a common pattern
// and expanding the Union may give a leaf function
static void _compile_all_tvar_union(jl_value_t *methsig)
{
int tvarslen = jl_subtype_env_size(methsig);
jl_value_t *sigbody = methsig;
jl_value_t **roots;
JL_GC_PUSHARGS(roots, 1 + 2 * tvarslen);
jl_value_t **env = roots + 1;
int *idx = (int*)alloca(sizeof(int) * tvarslen);
int i;
for (i = 0; i < tvarslen; i++) {
assert(jl_is_unionall(sigbody));
idx[i] = 0;
env[2 * i] = (jl_value_t*)((jl_unionall_t*)sigbody)->var;
env[2 * i + 1] = jl_bottom_type; // initialize the list with Union{}, since T<:Union{} is always a valid option
sigbody = ((jl_unionall_t*)sigbody)->body;
}
for (i = 0; i < tvarslen; /* incremented by inner loop */) {
jl_value_t **sig = &roots[0];
JL_TRY {
// TODO: wrap in UnionAll for each tvar in env[2*i + 1] ?
// currently doesn't matter much, since jl_compile_hint doesn't work on abstract types
*sig = (jl_value_t*)jl_instantiate_type_with(sigbody, env, tvarslen);
}
JL_CATCH {
goto getnext; // sigh, we found an invalid type signature. should we warn the user?
}
if (!jl_has_concrete_subtype(*sig))
goto getnext; // signature wouldn't be callable / is invalid -- skip it
if (jl_is_concrete_type(*sig)) {
if (jl_compile_hint((jl_tupletype_t *)*sig))
goto getnext; // success
}
getnext:
for (i = 0; i < tvarslen; i++) {
jl_tvar_t *tv = (jl_tvar_t*)env[2 * i];
if (jl_is_uniontype(tv->ub)) {
size_t l = jl_count_union_components(tv->ub);
size_t j = idx[i];
if (j == l) {
env[2 * i + 1] = jl_bottom_type;
idx[i] = 0;
}
else {
jl_value_t *ty = jl_nth_union_component(tv->ub, j);
if (!jl_is_concrete_type(ty))
ty = (jl_value_t*)jl_new_typevar(tv->name, tv->lb, ty);
env[2 * i + 1] = ty;
idx[i] = j + 1;
break;
}
}
else {
env[2 * i + 1] = (jl_value_t*)tv;
}
}
}
JL_GC_POP();
}
// f(::Union{...}, ...) is a common pattern
// and expanding the Union may give a leaf function
static void _compile_all_union(jl_value_t *sig)
{
jl_tupletype_t *sigbody = (jl_tupletype_t*)jl_unwrap_unionall(sig);
size_t count_unions = 0;
size_t i, l = jl_svec_len(sigbody->parameters);
jl_svec_t *p = NULL;
jl_value_t *methsig = NULL;
for (i = 0; i < l; i++) {
jl_value_t *ty = jl_svecref(sigbody->parameters, i);
if (jl_is_uniontype(ty))
++count_unions;
else if (ty == jl_bottom_type)
return; // why does this method exist?
else if (jl_is_datatype(ty) && !jl_has_free_typevars(ty) &&
((!jl_is_kind(ty) && ((jl_datatype_t*)ty)->isconcretetype) ||
((jl_datatype_t*)ty)->name == jl_type_typename))
return; // no amount of union splitting will make this a leaftype signature
}
if (count_unions == 0 || count_unions >= 6) {
_compile_all_tvar_union(sig);
return;
}
int *idx = (int*)alloca(sizeof(int) * count_unions);
for (i = 0; i < count_unions; i++) {
idx[i] = 0;
}
JL_GC_PUSH2(&p, &methsig);
int idx_ctr = 0, incr = 0;
while (!incr) {
p = jl_alloc_svec_uninit(l);
for (i = 0, idx_ctr = 0, incr = 1; i < l; i++) {
jl_value_t *ty = jl_svecref(sigbody->parameters, i);
if (jl_is_uniontype(ty)) {
assert(idx_ctr < count_unions);
size_t l = jl_count_union_components(ty);
size_t j = idx[idx_ctr];
jl_svecset(p, i, jl_nth_union_component(ty, j));
++j;
if (incr) {
if (j == l) {
idx[idx_ctr] = 0;
}
else {
idx[idx_ctr] = j;
incr = 0;
}
}
++idx_ctr;
}
else {
jl_svecset(p, i, ty);
}
}
methsig = jl_apply_tuple_type(p);
methsig = jl_rewrap_unionall(methsig, sig);
_compile_all_tvar_union(methsig);
}
JL_GC_POP();
}
static int compile_all_collect__(jl_typemap_entry_t *ml, void *env)
{
jl_array_t *allmeths = (jl_array_t*)env;
jl_method_t *m = ml->func.method;
if (m->external_mt)
return 1;
if (m->source) {
// method has a non-generated definition; can be compiled generically
jl_array_ptr_1d_push(allmeths, (jl_value_t*)m);
}
return 1;
}
static int compile_all_collect_(jl_methtable_t *mt, void *env)
{
jl_typemap_visitor(jl_atomic_load_relaxed(&mt->defs), compile_all_collect__, env);
return 1;
}
static void jl_compile_all_defs(jl_array_t *mis)
{
jl_array_t *allmeths = jl_alloc_vec_any(0);
JL_GC_PUSH1(&allmeths);
jl_foreach_reachable_mtable(compile_all_collect_, allmeths);
size_t i, l = jl_array_len(allmeths);
for (i = 0; i < l; i++) {
jl_method_t *m = (jl_method_t*)jl_array_ptr_ref(allmeths, i);
if (jl_is_datatype(m->sig) && jl_isa_compileable_sig((jl_tupletype_t*)m->sig, jl_emptysvec, m)) {
// method has a single compilable specialization, e.g. its definition
// signature is concrete. in this case we can just hint it.
jl_compile_hint((jl_tupletype_t*)m->sig);
}
else {
// first try to create leaf signatures from the signature declaration and compile those
_compile_all_union(m->sig);
// finally, compile a fully generic fallback that can work for all arguments
jl_method_instance_t *unspec = jl_get_unspecialized(m);
if (unspec)
jl_array_ptr_1d_push(mis, (jl_value_t*)unspec);
}
}
JL_GC_POP();
}
static int precompile_enq_specialization_(jl_method_instance_t *mi, void *closure)
{
assert(jl_is_method_instance(mi));
jl_code_instance_t *codeinst = jl_atomic_load_relaxed(&mi->cache);
while (codeinst) {
int do_compile = 0;
if (jl_atomic_load_relaxed(&codeinst->invoke) != jl_fptr_const_return) {
jl_value_t *inferred = jl_atomic_load_relaxed(&codeinst->inferred);
if (inferred &&
inferred != jl_nothing &&
jl_ir_flag_inferred(inferred) &&
(jl_ir_inlining_cost(inferred) == UINT16_MAX)) {
do_compile = 1;
}
else if (jl_atomic_load_relaxed(&codeinst->invoke) != NULL || jl_atomic_load_relaxed(&codeinst->precompile)) {
do_compile = 1;
}
}
if (do_compile) {
jl_array_ptr_1d_push((jl_array_t*)closure, (jl_value_t*)mi);
return 1;
}
codeinst = jl_atomic_load_relaxed(&codeinst->next);
}
return 1;
}
static int precompile_enq_all_specializations__(jl_typemap_entry_t *def, void *closure)
{
jl_method_t *m = def->func.method;
if (m->external_mt)
return 1;
if ((m->name == jl_symbol("__init__") || m->ccallable) && jl_is_dispatch_tupletype(m->sig)) {
// ensure `__init__()` and @ccallables get strongly-hinted, specialized, and compiled
jl_method_instance_t *mi = jl_specializations_get_linfo(m, m->sig, jl_emptysvec);
jl_array_ptr_1d_push((jl_array_t*)closure, (jl_value_t*)mi);
}
else {
jl_value_t *specializations = jl_atomic_load_relaxed(&def->func.method->specializations);
if (!jl_is_svec(specializations)) {
precompile_enq_specialization_((jl_method_instance_t*)specializations, closure);
}
else {
size_t i, l = jl_svec_len(specializations);
for (i = 0; i < l; i++) {
jl_value_t *mi = jl_svecref(specializations, i);
if (mi != jl_nothing)
precompile_enq_specialization_((jl_method_instance_t*)mi, closure);
}
}
}
if (m->ccallable)
jl_array_ptr_1d_push((jl_array_t*)closure, (jl_value_t*)m->ccallable);
return 1;
}
static int precompile_enq_all_specializations_(jl_methtable_t *mt, void *env)
{
return jl_typemap_visitor(jl_atomic_load_relaxed(&mt->defs), precompile_enq_all_specializations__, env);
}
static void *jl_precompile_(jl_array_t *m, int external_linkage)
{
jl_array_t *m2 = NULL;
jl_method_instance_t *mi = NULL;
JL_GC_PUSH2(&m2, &mi);
m2 = jl_alloc_vec_any(0);
for (size_t i = 0; i < jl_array_len(m); i++) {
jl_value_t *item = jl_array_ptr_ref(m, i);
if (jl_is_method_instance(item)) {
mi = (jl_method_instance_t*)item;
size_t min_world = 0;
size_t max_world = ~(size_t)0;
if (mi != jl_atomic_load_relaxed(&mi->def.method->unspecialized) && !jl_isa_compileable_sig((jl_tupletype_t*)mi->specTypes, mi->sparam_vals, mi->def.method))
mi = jl_get_specialization1((jl_tupletype_t*)mi->specTypes, jl_atomic_load_acquire(&jl_world_counter), &min_world, &max_world, 0);
if (mi)
jl_array_ptr_1d_push(m2, (jl_value_t*)mi);
}
else {
assert(jl_is_simplevector(item));
assert(jl_svec_len(item) == 2);
jl_array_ptr_1d_push(m2, item);
}
}
void *native_code = jl_create_native(m2, NULL, NULL, 0, 1, external_linkage,
jl_atomic_load_acquire(&jl_world_counter));
JL_GC_POP();
return native_code;
}
static void *jl_precompile(int all)
{
// array of MethodInstances and ccallable aliases to include in the output
jl_array_t *m = jl_alloc_vec_any(0);
JL_GC_PUSH1(&m);
if (all)
jl_compile_all_defs(m);
jl_foreach_reachable_mtable(precompile_enq_all_specializations_, m);
void *native_code = jl_precompile_(m, 0);
JL_GC_POP();
return native_code;
}
static void *jl_precompile_worklist(jl_array_t *worklist, jl_array_t *extext_methods, jl_array_t *new_specializations)
{
if (!worklist)
return NULL;
// this "found" array will contain function
// type signatures that were inferred but haven't been compiled
jl_array_t *m = jl_alloc_vec_any(0);
JL_GC_PUSH1(&m);
size_t i, n = jl_array_len(worklist);
for (i = 0; i < n; i++) {
jl_module_t *mod = (jl_module_t*)jl_array_ptr_ref(worklist, i);
assert(jl_is_module(mod));
foreach_mtable_in_module(mod, precompile_enq_all_specializations_, m);
}
n = jl_array_len(extext_methods);
for (i = 0; i < n; i++) {
jl_method_t *method = (jl_method_t*)jl_array_ptr_ref(extext_methods, i);
assert(jl_is_method(method));
jl_value_t *specializations = jl_atomic_load_relaxed(&method->specializations);
if (!jl_is_svec(specializations)) {
precompile_enq_specialization_((jl_method_instance_t*)specializations, m);
}
else {
size_t j, l = jl_svec_len(specializations);
for (j = 0; j < l; j++) {
jl_value_t *mi = jl_svecref(specializations, j);
if (mi != jl_nothing)
precompile_enq_specialization_((jl_method_instance_t*)mi, m);
}
}
}
n = jl_array_len(new_specializations);
for (i = 0; i < n; i++) {
jl_code_instance_t *ci = (jl_code_instance_t*)jl_array_ptr_ref(new_specializations, i);
precompile_enq_specialization_(ci->def, m);
}
void *native_code = jl_precompile_(m, 1);
JL_GC_POP();
return native_code;
}
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...