Files
swift-mirror/lib/SILOptimizer/Transforms/AccessEnforcementReleaseSinking.cpp
Michael Gottesman 20c5dff4b5 [builtin] Implement polymorphic builtins for all BUILTIN_BINARY_OPERATIONs.
TLDR: This patch introduces a new kind of builtin, "a polymorphic builtin". One
calls it like any other builtin, e.x.:

```
Builtin.generic_add(x, y)
```

but it has a contract: it must be specialized to a concrete builtin by the time
we hit Lowered SIL. In this commit, I add support for the following generic
operations:

Type           | Op
------------------------
FloatOrVector  |FAdd
FloatOrVector  |FDiv
FloatOrVector  |FMul
FloatOrVector  |FRem
FloatOrVector  |FSub
IntegerOrVector|AShr
IntegerOrVector|Add
IntegerOrVector|And
IntegerOrVector|ExactSDiv
IntegerOrVector|ExactUDiv
IntegerOrVector|LShr
IntegerOrVector|Mul
IntegerOrVector|Or
IntegerOrVector|SDiv
IntegerOrVector|SRem
IntegerOrVector|Shl
IntegerOrVector|Sub
IntegerOrVector|UDiv
IntegerOrVector|Xor
Integer        |URem

NOTE: I only implemented support for the builtins in SIL and in SILGen. I am
going to implement the optimizer parts of this in a separate series of commits.

DISCUSSION
----------

Today there are polymorphic like instructions in LLVM-IR. Yet, at the
swift and SIL level we represent these operations instead as Builtins whose
names are resolved by splatting the builtin into the name. For example, adding
two things in LLVM:

```
  %2 = add i64 %0, %1
  %2 = add <2 x i64> %0, %1
  %2 = add <4 x i64> %0, %1
  %2 = add <8 x i64> %0, %1
```

Each of the add operations are done by the same polymorphic instruction. In
constrast, we splat out these Builtins in swift today, i.e.:

```
let x, y: Builtin.Int32
Builtin.add_Int32(x, y)
let x, y: Builtin.Vec4xInt32
Builtin.add_Vec4xInt32(x, y)
...
```

In SIL, we translate these verbatim and then IRGen just lowers them to the
appropriate polymorphic instruction. Beyond being verbose, these prevent these
Builtins (which need static types) from being used in polymorphic contexts where
we can guarantee that eventually a static type will be provided.

In contrast, the polymorphic builtins introduced in this commit can be passed
any type, with the proviso that the expert user using this feature can guarantee
that before we reach Lowered SIL, the generic_add has been eliminated. This is
enforced by IRGen asserting if passed such a builtin and by the SILVerifier
checking that the underlying builtin is never called once the module is in
Lowered SIL.

In forthcoming commits, I am going to add two optimizations that give the stdlib
tool writer the tools needed to use this builtin:

1. I am going to add an optimization to constant propagation that changes a
"generic_*" op to the type of its argument if the argument is a type that is
valid for the builtin (i.e. integer or vector).

2. I am going to teach the SILCloner how to specialize these as it inlines. This
ensures that when we transparent inline, we specialize the builtin automatically
and can then form SSA at -Onone using predictable memory access operations.

The main implication around these polymorphic builtins are that if an author is
not able to specialize the builtin, they need to ensure that after constant
propagation, the generic builtin has been DCEed. The general rules are that the
-Onone optimizer will constant fold branches with constant integer operands. So
if one can use a bool of some sort to trigger the operation, one can be
guaranteed that the code will not codegen. I am considering putting in some sort
of diagnostic to ensure that the stdlib writer has a good experience (e.x. get
an error instead of crashing the compiler).
2019-09-19 11:42:10 -07:00

238 lines
9.1 KiB
C++

//===--- AccessEnforcementReleaseSinking.cpp - release sinking opt ---===//
//
// This source file is part of the Swift.org open source project
//
// Copyright (c) 2014 - 2018 Apple Inc. and the Swift project authors
// Licensed under Apache License v2.0 with Runtime Library Exception
//
// See https://swift.org/LICENSE.txt for license information
// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
//
//===----------------------------------------------------------------------===//
///
/// This function pass sinks releases out of access scopes.
///
/// General case:
/// begin_access A
/// ...
/// strong_release / release_value / destroy
/// end_access
///
/// The release instruction can be sunk below the end_access instruction,
/// This extends the lifetime of the released value, but, might allow us to
/// Mark the access scope as no nested conflict.
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "access-enforcement-release"
#include "swift/SIL/ApplySite.h"
#include "swift/SIL/DebugUtils.h"
#include "swift/SIL/InstructionUtils.h"
#include "swift/SIL/SILFunction.h"
#include "swift/SILOptimizer/PassManager/Transforms.h"
using namespace swift;
// Returns a bool: If this is a "sinkable" instruction type for this opt
static bool isSinkable(SILInstruction &inst) {
switch (inst.getKind()) {
default:
return false;
case SILInstructionKind::DestroyValueInst:
case SILInstructionKind::ReleaseValueInst:
case SILInstructionKind::ReleaseValueAddrInst:
case SILInstructionKind::StrongReleaseInst:
case SILInstructionKind::UnmanagedReleaseValueInst:
return true;
}
}
// Returns a bool: If this is a "barrier" instruction for this opt
static bool isBarrier(SILInstruction *inst) {
// Calls hide many dangers, from checking reference counts, to beginning
// keypath access, to forcing memory to be live. Checking for these and other
// possible barries at ever call is certainly not worth it.
if (FullApplySite::isa(inst) != FullApplySite())
return true;
// Don't extend lifetime past any sort of uniqueness check.
if (mayCheckRefCount(inst))
return true;
// Don't extend object lifetime past deallocation.
if (isa<DeallocationInst>(inst))
return true;
// Avoid introducing access conflicts.
if (isa<BeginAccessInst>(inst) || isa<BeginUnpairedAccessInst>(inst))
return true;
if (auto *BI = dyn_cast<BuiltinInst>(inst)) {
auto kind = BI->getBuiltinKind();
if (!kind)
return false; // LLVM intrinsics are not barriers.
// Whitelist the safe builtin categories. Builtins should generally be
// treated conservatively, because introducing a new builtin does not
// require updating all passes to be aware of it.
switch (kind.getValue()) {
case BuiltinValueKind::None:
llvm_unreachable("Builtin must has a non-empty kind.");
// Unhandled categories don't generate a case. Instead, they result
// in a build error: enumeration values not handled in switch.
#define BUILTIN(Id, Name, Attrs)
#define BUILTIN_NO_BARRIER(Id) \
case BuiltinValueKind::Id: \
return false;
#define BUILTIN_CAST_OPERATION(Id, Name, Attrs) BUILTIN_NO_BARRIER(Id)
#define BUILTIN_CAST_OR_BITCAST_OPERATION(Id, Name, Attrs) \
BUILTIN_NO_BARRIER(Id)
#define BUILTIN_BINARY_OPERATION(Id, Name, Attrs) BUILTIN_NO_BARRIER(Id)
#define BUILTIN_BINARY_OPERATION_WITH_OVERFLOW(Id, Name, UncheckedID, Attrs, \
Overload) \
BUILTIN_NO_BARRIER(Id)
#define BUILTIN_UNARY_OPERATION(Id, Name, Attrs, Overload) \
BUILTIN_NO_BARRIER(Id)
#define BUILTIN_BINARY_PREDICATE(Id, Name, Attrs, Overload) \
BUILTIN_NO_BARRIER(Id)
#define BUILTIN_SIL_OPERATION(Id, Name, Overload) \
case BuiltinValueKind::Id: \
llvm_unreachable("SIL operation must be lowered to instructions.");
#define BUILTIN_RUNTIME_CALL(Id, Name, Attrs) \
case BuiltinValueKind::Id: \
return true; // A runtime call could be anything.
#define BUILTIN_SANITIZER_OPERATION(Id, Name, Attrs) BUILTIN_NO_BARRIER(Id)
#define BUILTIN_TYPE_CHECKER_OPERATION(Id, Name) BUILTIN_NO_BARRIER(Id)
#define BUILTIN_TYPE_TRAIT_OPERATION(Id, Name) BUILTIN_NO_BARRIER(Id)
#include "swift/AST/Builtins.def"
// Handle BUILTIN_MISC_OPERATIONs individually.
case BuiltinValueKind::Sizeof:
case BuiltinValueKind::Strideof:
case BuiltinValueKind::IsPOD:
case BuiltinValueKind::IsConcrete:
case BuiltinValueKind::IsBitwiseTakable:
case BuiltinValueKind::IsSameMetatype:
case BuiltinValueKind::Alignof:
case BuiltinValueKind::OnFastPath:
case BuiltinValueKind::ExtractElement:
case BuiltinValueKind::InsertElement:
case BuiltinValueKind::StaticReport:
case BuiltinValueKind::AssertConf:
case BuiltinValueKind::StringObjectOr:
case BuiltinValueKind::UToSCheckedTrunc:
case BuiltinValueKind::SToUCheckedTrunc:
case BuiltinValueKind::SToSCheckedTrunc:
case BuiltinValueKind::UToUCheckedTrunc:
case BuiltinValueKind::IntToFPWithOverflow:
case BuiltinValueKind::ZeroInitializer:
case BuiltinValueKind::Once:
case BuiltinValueKind::OnceWithContext:
case BuiltinValueKind::GetObjCTypeEncoding:
case BuiltinValueKind::Swift3ImplicitObjCEntrypoint:
case BuiltinValueKind::WillThrow:
case BuiltinValueKind::CondFailMessage:
case BuiltinValueKind::PoundAssert:
case BuiltinValueKind::GlobalStringTablePointer:
return false;
// Handle some rare builtins that may be sensitive to object lifetime
// or deinit side effects conservatively.
case BuiltinValueKind::AllocRaw:
case BuiltinValueKind::DeallocRaw:
case BuiltinValueKind::Fence:
case BuiltinValueKind::AtomicLoad:
case BuiltinValueKind::AtomicStore:
case BuiltinValueKind::AtomicRMW:
case BuiltinValueKind::Unreachable:
case BuiltinValueKind::CmpXChg:
case BuiltinValueKind::CondUnreachable:
case BuiltinValueKind::DestroyArray:
case BuiltinValueKind::CopyArray:
case BuiltinValueKind::TakeArrayNoAlias:
case BuiltinValueKind::TakeArrayFrontToBack:
case BuiltinValueKind::TakeArrayBackToFront:
case BuiltinValueKind::AssignCopyArrayNoAlias:
case BuiltinValueKind::AssignCopyArrayFrontToBack:
case BuiltinValueKind::AssignCopyArrayBackToFront:
case BuiltinValueKind::AssignTakeArray:
case BuiltinValueKind::UnsafeGuaranteed:
case BuiltinValueKind::UnsafeGuaranteedEnd:
return true;
}
}
return false;
}
// Processes a block bottom-up, keeping a lookout for end_access instructions
// If we encounter a "barrier" we clear out the current end_access
// If we encounter a "release", and we have a current end_access, we sink it
static void processBlock(SILBasicBlock &block) {
EndAccessInst *bottomEndAccessInst = nullptr;
for (auto reverseIt = block.rbegin(); reverseIt != block.rend();
++reverseIt) {
SILInstruction &currIns = *reverseIt;
if (auto *currEAI = dyn_cast<EndAccessInst>(&currIns)) {
if (!bottomEndAccessInst) {
bottomEndAccessInst = currEAI;
}
} else if (isBarrier(&currIns)) {
LLVM_DEBUG(llvm::dbgs() << "Found a barrier " << currIns
<< ", clearing last seen end_access\n");
bottomEndAccessInst = nullptr;
} else if (isSinkable(currIns)) {
LLVM_DEBUG(llvm::dbgs()
<< "Found a sinkable instruction " << currIns << "\n");
if (!bottomEndAccessInst) {
LLVM_DEBUG(
llvm::dbgs()
<< "Cannot be sunk: no open barrier-less end_access found\n");
continue;
}
LLVM_DEBUG(llvm::dbgs() << "Moving sinkable instruction below "
<< *bottomEndAccessInst << "\n");
// We need to avoid iterator invalidation:
// We know this is not the last instruction of the block:
// 1) not a TermInst
// 2) bottomEndAccessInst != nil
assert(reverseIt != block.rbegin() &&
"Did not expect a sinkable instruction at block's end");
// Go back to previous iteration
auto prevIt = reverseIt;
--prevIt;
// Move the instruction after the end_access
currIns.moveAfter(bottomEndAccessInst);
// make reverseIt into a valid iterator again
reverseIt = prevIt;
}
}
}
namespace {
struct AccessEnforcementReleaseSinking : public SILFunctionTransform {
void run() override {
SILFunction *F = getFunction();
if (F->empty())
return;
// FIXME: Support ownership.
if (F->hasOwnership())
return;
LLVM_DEBUG(llvm::dbgs() << "Running AccessEnforcementReleaseSinking on "
<< F->getName() << "\n");
for (SILBasicBlock &currBB : *F) {
processBlock(currBB);
}
}
};
} // namespace
SILTransform *swift::createAccessEnforcementReleaseSinking() {
return new AccessEnforcementReleaseSinking();
}