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

Jump to bottom

correct codegen bugs introduced by allocation-hoisting-PR #45476

Merged
merged 17 commits into from
Jul 21, 2022
Merged

correct codegen bugs introduced by allocation-hoisting-PR #45476

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
4a99a31
Address reviews for older PR
gbaraldi May 27, 2022
a3cb97b
Merge branch 'master' into fix-heaphoist
gbaraldi May 27, 2022
508f292
Fix whitespace
gbaraldi May 27, 2022
ce8ff75
Correct name mangling
gbaraldi May 31, 2022
30f68bc
Implement different mangling approach
gbaraldi May 31, 2022
e366acb
Remove unused variable and small refactor
gbaraldi Jun 6, 2022
76671d1
Whitespace
gbaraldi Jun 6, 2022
ac65255
change is_promotable to field_promotable
gbaraldi Jun 9, 2022
d645429
Merge branch 'master' into fix-heaphoist
gbaraldi Jun 22, 2022
4cac1d0
Add explicit check instead of assertion
gbaraldi Jun 22, 2022
7558018
Fix the merge error ;(
gbaraldi Jun 22, 2022
e4257cb
Merge branch 'master' into fix-heaphoist
gbaraldi Jun 30, 2022
2e8363b
Explicetely not promote unions
gbaraldi Jul 2, 2022
9fcb436
Merge branch 'master' into fix-heaphoist
gbaraldi Jul 4, 2022
088b36c
Add a test for the issue found
gbaraldi Jul 6, 2022
f8f7028
Fix whitespace
gbaraldi Jul 7, 2022
d653492
Merge branch 'master' into fix-heaphoist
gbaraldi Jul 21, 2022
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
63 changes: 48 additions & 15 deletions src/cgutils.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1005,7 +1005,7 @@ static LoadInst *emit_nthptr_recast(jl_codectx_t &ctx, Value *v, ssize_t n, MDNo
emit_bitcast(ctx, vptr, PointerType::get(type, 0)))));
}

static Value *boxed(jl_codectx_t &ctx, const jl_cgval_t &v);
static Value *boxed(jl_codectx_t &ctx, const jl_cgval_t &v, bool is_promotable=false);
Copy link
Sponsor Member

Choose a reason for hiding this comment

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

So promotion is now always forbidden implicitly?

Copy link
Member Author

Choose a reason for hiding this comment

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

Yes, the regression we were seeing was from something getting promoted when we called boxed on a separate part fo the of the compilation and then hoist something that shouldn't be hoisted.
In the case of the regression we were hosting a string allocation from inside an error to the main path of the function which then impeded an optimization.
From what I understand we should only hoist the allocation when we have nested structs and it's fields are promotable which we only check inside emit_new_struct

static Value *emit_typeof(jl_codectx_t &ctx, Value *v, bool maybenull);

static jl_cgval_t emit_typeof(jl_codectx_t &ctx, const jl_cgval_t &p, bool maybenull)
Expand Down Expand Up @@ -3207,6 +3207,44 @@ static Value *box_union(jl_codectx_t &ctx, const jl_cgval_t &vinfo, const SmallB
return box_merge;
}

static Function *mangleIntrinsic(IntrinsicInst *call) //mangling based on replaceIntrinsicUseWith
{
Intrinsic::ID ID = call->getIntrinsicID();
auto nargs = call->arg_size();
SmallVector<Type*, 8> argTys(nargs);
auto oldfType = call->getFunctionType();
for (unsigned i = 0; i < oldfType->getNumParams(); i++) {
auto argi = call->getArgOperand(i);
argTys[i] = argi->getType();
}

auto newfType = FunctionType::get(
oldfType->getReturnType(),
makeArrayRef(argTys).slice(0, oldfType->getNumParams()),
oldfType->isVarArg());

// Accumulate an array of overloaded types for the given intrinsic
// and compute the new name mangling schema
SmallVector<Type*, 4> overloadTys;
{
SmallVector<Intrinsic::IITDescriptor, 8> Table;
getIntrinsicInfoTableEntries(ID, Table);
ArrayRef<Intrinsic::IITDescriptor> TableRef = Table;
auto res = Intrinsic::matchIntrinsicSignature(newfType, TableRef, overloadTys);
assert(res == Intrinsic::MatchIntrinsicTypes_Match);
(void)res;
bool matchvararg = !Intrinsic::matchIntrinsicVarArg(newfType->isVarArg(), TableRef);
assert(matchvararg);
(void)matchvararg;
}
auto newF = Intrinsic::getDeclaration(call->getModule(), ID, overloadTys);
assert(newF->getFunctionType() == newfType);
newF->setCallingConv(call->getCallingConv());
return newF;
}


//Used for allocation hoisting in *boxed
static void recursively_adjust_ptr_type(llvm::Value *Val, unsigned FromAS, unsigned ToAS)
{
for (auto *User : Val->users()) {
Expand All @@ -3216,13 +3254,8 @@ static void recursively_adjust_ptr_type(llvm::Value *Val, unsigned FromAS, unsig
recursively_adjust_ptr_type(Inst, FromAS, ToAS);
}
else if (isa<IntrinsicInst>(User)) {
IntrinsicInst *II = cast<IntrinsicInst>(User);
SmallVector<Type*, 3> ArgTys;
Intrinsic::getIntrinsicSignature(II->getCalledFunction(), ArgTys);
assert(ArgTys.size() <= II->arg_size());
for (size_t i = 0; i < ArgTys.size(); ++i)
ArgTys[i] = II->getArgOperand(i)->getType();
II->setCalledFunction(Intrinsic::getDeclaration(II->getModule(), II->getIntrinsicID(), ArgTys));
IntrinsicInst *call = cast<IntrinsicInst>(User);
call->setCalledFunction(mangleIntrinsic(call));
}
#ifndef JL_LLVM_OPAQUE_POINTERS
else if (isa<BitCastInst>(User)) {
Expand All @@ -3238,7 +3271,7 @@ static void recursively_adjust_ptr_type(llvm::Value *Val, unsigned FromAS, unsig
// dynamically-typed value is required (e.g. argument to unknown function).
// if it's already a pointer it's left alone.
// Returns ctx.types().T_prjlvalue
static Value *boxed(jl_codectx_t &ctx, const jl_cgval_t &vinfo)
static Value *boxed(jl_codectx_t &ctx, const jl_cgval_t &vinfo, bool is_promotable)
{
jl_value_t *jt = vinfo.typ;
if (jt == jl_bottom_type || jt == NULL)
Expand Down Expand Up @@ -3268,7 +3301,7 @@ static Value *boxed(jl_codectx_t &ctx, const jl_cgval_t &vinfo)
box = _boxed_special(ctx, vinfo, t);
if (!box) {
bool do_promote = vinfo.promotion_point;
if (do_promote) {
if (do_promote && is_promotable) {
auto IP = ctx.builder.saveIP();
ctx.builder.SetInsertPoint(vinfo.promotion_point);
box = emit_allocobj(ctx, jl_datatype_size(jt), literal_pointer_val(ctx, (jl_value_t*)jt));
Expand All @@ -3282,7 +3315,7 @@ static Value *boxed(jl_codectx_t &ctx, const jl_cgval_t &vinfo)
recursively_adjust_ptr_type(originalAlloca, 0, AddressSpace::Derived);
originalAlloca->replaceAllUsesWith(decayed);
// end illegal IR
cast<Instruction>(vinfo.V)->eraseFromParent();
originalAlloca->eraseFromParent();
ctx.builder.restoreIP(IP);
} else {
box = emit_allocobj(ctx, jl_datatype_size(jt), literal_pointer_val(ctx, (jl_value_t*)jt));
Expand Down Expand Up @@ -3651,7 +3684,7 @@ static jl_cgval_t emit_new_struct(jl_codectx_t &ctx, jl_value_t *ty, size_t narg
if (fval_info.typ == jl_bottom_type)
return jl_cgval_t();
// TODO: Use (post-)domination instead.
bool field_promotable = !init_as_value && fval_info.promotion_ssa != -1 &&
bool field_promotable = !jl_is_uniontype(jtype) && !init_as_value && fval_info.promotion_ssa != -1 &&
Copy link
Sponsor Member

Choose a reason for hiding this comment

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

It would be valid though if jl_field_isptr

Copy link
Member Author

Choose a reason for hiding this comment

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

The issue is that unions currently don't get their promotion point correctly, I'm not exactly what's needed to fix that, so for now, not allowing unions at all seems fine.

fval_info.promotion_point && fval_info.promotion_point->getParent() == ctx.builder.GetInsertBlock();
if (field_promotable) {
savedIP = ctx.builder.saveIP();
Expand Down Expand Up @@ -3684,20 +3717,20 @@ static jl_cgval_t emit_new_struct(jl_codectx_t &ctx, jl_value_t *ty, size_t narg
promotion_point = inst;
promotion_ssa = fval_info.promotion_ssa;
}
} else if (!promotion_point) {
}
else if (!promotion_point) {
promotion_point = inst;
}
}
Value *fval = NULL;
if (jl_field_isptr(sty, i)) {
fval = boxed(ctx, fval_info);
fval = boxed(ctx, fval_info, field_promotable);
if (!init_as_value)
cast<StoreInst>(tbaa_decorate(ctx.tbaa().tbaa_stack,
ctx.builder.CreateAlignedStore(fval, dest, Align(jl_field_align(sty, i)))))
->setOrdering(AtomicOrdering::Unordered);
}
else if (jl_is_uniontype(jtype)) {
assert(!field_promotable);
// compute tindex from rhs
jl_cgval_t rhs_union = convert_julia_type(ctx, fval_info, jtype);
if (rhs_union.typ == jl_bottom_type)
Expand Down
4 changes: 2 additions & 2 deletions src/codegen.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5001,7 +5001,7 @@ static jl_cgval_t emit_expr(jl_codectx_t &ctx, jl_value_t *expr, ssize_t ssaidx_
expr_t = (jl_value_t*)jl_any_type;
else {
expr_t = jl_is_long(ctx.source->ssavaluetypes) ? (jl_value_t*)jl_any_type : jl_array_ptr_ref(ctx.source->ssavaluetypes, ssaidx_0based);
is_promotable = ctx.ssavalue_usecount[ssaidx_0based] == 1;
is_promotable = ctx.ssavalue_usecount.at(ssaidx_0based) == 1;
}
jl_cgval_t res = emit_call(ctx, ex, expr_t, is_promotable);
// some intrinsics (e.g. typeassert) can return a wider type
Expand Down Expand Up @@ -5119,7 +5119,7 @@ static jl_cgval_t emit_expr(jl_codectx_t &ctx, jl_value_t *expr, ssize_t ssaidx_
else if (head == jl_new_sym) {
bool is_promotable = false;
if (ssaidx_0based >= 0) {
is_promotable = ctx.ssavalue_usecount[ssaidx_0based] == 1;
is_promotable = ctx.ssavalue_usecount.at(ssaidx_0based) == 1;
}
assert(nargs > 0);
jl_cgval_t *argv = (jl_cgval_t*)alloca(sizeof(jl_cgval_t) * nargs);
Expand Down
7 changes: 7 additions & 0 deletions test/compiler/codegen.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -747,3 +747,10 @@ f_donotdelete_input(x) = Base.donotdelete(x+1)
f_donotdelete_const() = Base.donotdelete(1+1)
@test occursin("call void (...) @jl_f_donotdelete(i64", get_llvm(f_donotdelete_input, Tuple{Int64}, true, false, false))
@test occursin("call void (...) @jl_f_donotdelete()", get_llvm(f_donotdelete_const, Tuple{}, true, false, false))

# Test 45476 fixes
struct MaybeTuple45476
val::Union{Nothing, Tuple{Float32}}
end

@test MaybeTuple45476((0,)).val[1] == 0f0