Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

avoid inferring when compilation signature differs from call site sig #46581

Merged
merged 1 commit into from
Sep 2, 2022
Merged

avoid inferring when compilation signature differs from call site sig #46581

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 22 additions & 16 deletions base/compiler/abstractinterpretation.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,7 +125,7 @@ function abstract_call_gf_by_type(interp::AbstractInterpreter, @nospecialize(f),
for sig_n in splitsigs
result = abstract_call_method(interp, method, sig_n, svec(), multiple_matches, sv)
(; rt, edge, effects) = result
edge !== nothing && push!(edges, edge)
edge === nothing || push!(edges, edge)
this_argtypes = isa(matches, MethodMatches) ? argtypes : matches.applicable_argtypes[i]
this_arginfo = ArgInfo(fargs, this_argtypes)
const_call_result = abstract_call_method_with_const_args(interp, result,
Expand All @@ -148,25 +148,11 @@ function abstract_call_gf_by_type(interp::AbstractInterpreter, @nospecialize(f),
this_conditional = ignorelimited(this_rt)
this_rt = widenwrappedconditional(this_rt)
else
if infer_compilation_signature(interp)
# Also infer the compilation signature for this method, so it's available
# to the compiler in case it ends up needing it (which is likely).
csig = get_compileable_sig(method, sig, match.sparams)
if csig !== nothing && csig !== sig
# The result of this inference is not directly used, so temporarily empty
# the use set for the current SSA value.
saved_uses = sv.ssavalue_uses[sv.currpc]
sv.ssavalue_uses[sv.currpc] = empty_bitset
abstract_call_method(interp, method, csig, match.sparams, multiple_matches, sv)
sv.ssavalue_uses[sv.currpc] = saved_uses
end
end

result = abstract_call_method(interp, method, sig, match.sparams, multiple_matches, sv)
(; rt, edge, effects) = result
this_conditional = ignorelimited(rt)
this_rt = widenwrappedconditional(rt)
edge !== nothing && push!(edges, edge)
edge === nothing || push!(edges, edge)
# try constant propagation with argtypes for this match
# this is in preparation for inlining, or improving the return result
this_argtypes = isa(matches, MethodMatches) ? argtypes : matches.applicable_argtypes[i]
Expand Down Expand Up @@ -225,6 +211,26 @@ function abstract_call_gf_by_type(interp::AbstractInterpreter, @nospecialize(f),

rettype = from_interprocedural!(rettype, sv, arginfo, conditionals)

# Also considering inferring the compilation signature for this method, so
# it is available to the compiler in case it ends up needing it.
if infer_compilation_signature(interp) && 1 == seen == napplicable && rettype !== Any && rettype !== Union{} && !is_removable_if_unused(all_effects)
match = applicable[1]::MethodMatch
method = match.method
sig = match.spec_types
mi = specialize_method(match; preexisting=true)
if mi !== nothing && !const_prop_methodinstance_heuristic(interp, match, mi::MethodInstance, arginfo, sv)
csig = get_compileable_sig(method, sig, match.sparams)
if csig !== nothing && csig !== sig
Comment on lines +216 to +223
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I may want to pack these condition checks into a separate function, e.g.

    if infer_compilation_signature(interp)
        compilation_sig = maybe_compilation_signature(applicable, seen, rettype, all_effects, arginfo, sv)
        if compilation_sig !== nothing
            (; method, csig, sparams) = compilation_sig
            ...
        end
    end

But maybe_compilation_signature's argument list is so long depending on many local variables of abstract_call_gf_by_type, so I'm fine with the current code if you prefer it.

# The result of this inference is not directly used, so temporarily empty
# the use set for the current SSA value.
saved_uses = sv.ssavalue_uses[sv.currpc]
sv.ssavalue_uses[sv.currpc] = empty_bitset
abstract_call_method(interp, method, csig, match.sparams, multiple_matches, sv)
aviatesk marked this conversation as resolved.
Show resolved Hide resolved
sv.ssavalue_uses[sv.currpc] = saved_uses
end
end
end

if call_result_unused(sv) && !(rettype === Bottom)
add_remark!(interp, sv, "Call result type was widened because the return value is unused")
# We're mainly only here because the optimizer might want this code,
Expand Down