Revision c22889e76cb9b7fd8a4710d9bf53e827aaa907e4 authored by Shuhei Kadowaki on 24 August 2021, 04:21:29 UTC, committed by Shuhei Kadowaki on 26 October 2021, 14:45:33 UTC
Currently our constant-prop' heuristics work in the following way:
1. `const_prop_entry_heuristic`
2. `const_prop_argument_heuristic` & `const_prop_rettype_heuristic`
3. `force_const_prop` custom heuristic & `!const_prop_function_heuristic`
4. `MethodInstance` specialization and `const_prop_methodinstance_heuristic`
This PR changes it so that the step 1. now works like:
1. `force_const_prop` custom heuristic & `const_prop_entry_heuristic`
and the steps 2., 3. and 4. don't change
This change particularly allows us to more forcibly constant-propagate
for `getproperty` and `setproperty!`, and inline them more, e.g.:
```julia
mutable struct Foo
val
_::Int
end
function setter(xs)
for x in xs
x.val = nothing # `setproperty!` can be inlined with this PR
end
end
```
It might be useful because now we can intervene into the constant-prop'
heuristic in a more reliable way with the `aggressive_constprop` interface.
I did the simple benchmark below, and it looks like this change doesn't
cause the latency problem for this particular example:
```zsh
~/julia master aviatesk@amdci2 6s
❯ ./usr/bin/julia -e '@time using Plots; @time plot(rand(10,3))'
3.708500 seconds (7.28 M allocations: 506.128 MiB, 3.45% gc time, 1.13% compilation time)
2.817794 seconds (3.45 M allocations: 195.127 MiB, 7.84% gc time, 53.76% compilation time)
~/julia avi/forceconstantprop aviatesk@amdci2 6s
❯ ./usr/bin/julia -e '@time using Plots; @time plot(rand(10,3))'
3.622109 seconds (7.02 M allocations: 481.710 MiB, 4.19% gc time, 1.17% compilation time)
2.863419 seconds (3.44 M allocations: 194.210 MiB, 8.02% gc time, 53.53% compilation time)
```
1 parent da71d29
opaque_closure.c
#include "julia.h"
#include "julia_internal.h"
JL_DLLEXPORT jl_value_t *jl_invoke_opaque_closure(jl_opaque_closure_t *oc, jl_value_t **args, size_t nargs)
{
jl_value_t *ret = NULL;
JL_GC_PUSH1(&ret);
jl_task_t *ct = jl_current_task;
size_t last_age = ct->world_age;
ct->world_age = oc->world;
ret = jl_interpret_opaque_closure(oc, args, nargs);
jl_typeassert(ret, jl_tparam1(jl_typeof(oc)));
ct->world_age = last_age;
JL_GC_POP();
return ret;
}
jl_opaque_closure_t *jl_new_opaque_closure(jl_tupletype_t *argt, jl_value_t *isva,
jl_value_t *rt_lb, jl_value_t *rt_ub, jl_value_t *source, jl_value_t **env, size_t nenv)
{
if (!jl_is_tuple_type((jl_value_t*)argt)) {
jl_error("OpaqueClosure argument tuple must be a tuple type");
}
JL_TYPECHK(new_opaque_closure, bool, isva);
JL_TYPECHK(new_opaque_closure, type, rt_lb);
JL_TYPECHK(new_opaque_closure, type, rt_ub);
JL_TYPECHK(new_opaque_closure, method, source);
jl_task_t *ct = jl_current_task;
jl_value_t *oc_type JL_ALWAYS_LEAFTYPE;
oc_type = jl_apply_type2((jl_value_t*)jl_opaque_closure_type, (jl_value_t*)argt, rt_ub);
JL_GC_PROMISE_ROOTED(oc_type);
jl_value_t *captures = NULL;
JL_GC_PUSH1(&captures);
captures = jl_f_tuple(NULL, env, nenv);
jl_opaque_closure_t *oc = (jl_opaque_closure_t*)jl_gc_alloc(ct->ptls, sizeof(jl_opaque_closure_t), oc_type);
JL_GC_POP();
oc->source = (jl_method_t*)source;
oc->isva = jl_unbox_bool(isva);
oc->invoke = (jl_fptr_args_t)jl_invoke_opaque_closure;
oc->specptr = NULL;
oc->captures = captures;
oc->world = jl_world_counter;
return oc;
}
JL_CALLABLE(jl_new_opaque_closure_jlcall)
{
if (nargs < 5)
jl_error("new_opaque_closure: Not enough arguments");
return (jl_value_t*)jl_new_opaque_closure((jl_tupletype_t*)args[0],
args[1], args[2], args[3], args[4], &args[5], nargs-5);
}
// check whether the specified number of arguments is compatible with the
// specified number of parameters of the tuple type
STATIC_INLINE int jl_tupletype_length_compat(jl_value_t *v, size_t nargs) JL_NOTSAFEPOINT
{
v = jl_unwrap_unionall(v);
assert(jl_is_tuple_type(v));
size_t nparams = jl_nparams(v);
if (nparams == 0)
return nargs == 0;
jl_value_t *va = jl_tparam(v,nparams-1);
if (jl_is_vararg(va)) {
jl_value_t *len = jl_unwrap_vararg_num(va);
if (len &&jl_is_long(len))
return nargs == nparams - 1 + jl_unbox_long(len);
return nargs >= nparams - 1;
}
return nparams == nargs;
}
JL_CALLABLE(jl_f_opaque_closure_call)
{
jl_opaque_closure_t* oc = (jl_opaque_closure_t*)F;
jl_value_t *argt = jl_tparam0(jl_typeof(oc));
if (!jl_tupletype_length_compat(argt, nargs))
jl_error("Incorrect argument count for OpaqueClosure");
argt = jl_unwrap_unionall(argt);
assert(jl_is_datatype(argt));
jl_svec_t *types = jl_get_fieldtypes((jl_datatype_t*)argt);
size_t ntypes = jl_svec_len(types);
for (int i = 0; i < nargs; ++i) {
jl_value_t *typ = i >= ntypes ? jl_svecref(types, ntypes-1) : jl_svecref(types, i);
if (jl_is_vararg(typ))
typ = jl_unwrap_vararg(typ);
jl_typeassert(args[i], typ);
}
return oc->invoke(F, args, nargs);
}
Computing file changes ...
