blob: efe4d6b000499f319eed8f7d5144e59d6299a97a [file] [log] [blame]
/*
* Copyright (C) 2017 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef ART_COMPILER_OPTIMIZING_NODES_VECTOR_H_
#define ART_COMPILER_OPTIMIZING_NODES_VECTOR_H_
// This #include should never be used by compilation, because this header file (nodes_vector.h)
// is included in the header file nodes.h itself. However it gives editing tools better context.
#include "nodes.h"
namespace art {
// Memory alignment, represented as an offset relative to a base, where 0 <= offset < base,
// and base is a power of two. For example, the value Alignment(16, 0) means memory is
// perfectly aligned at a 16-byte boundary, whereas the value Alignment(16, 4) means
// memory is always exactly 4 bytes above such a boundary.
class Alignment {
public:
Alignment(size_t base, size_t offset) : base_(base), offset_(offset) {
DCHECK_LT(offset, base);
DCHECK(IsPowerOfTwo(base));
}
// Returns true if memory is at least aligned at the given boundary.
// Assumes requested base is power of two.
bool IsAlignedAt(size_t base) const {
DCHECK_NE(0u, base);
DCHECK(IsPowerOfTwo(base));
return ((offset_ | base_) & (base - 1u)) == 0;
}
size_t Base() const { return base_; }
size_t Offset() const { return offset_; }
std::string ToString() const {
return "ALIGN(" + std::to_string(base_) + "," + std::to_string(offset_) + ")";
}
bool operator==(const Alignment& other) const {
return base_ == other.base_ && offset_ == other.offset_;
}
private:
size_t base_;
size_t offset_;
};
//
// Definitions of abstract vector operations in HIR.
//
// Abstraction of a vector operation, i.e., an operation that performs
// GetVectorLength() x GetPackedType() operations simultaneously.
class HVecOperation : public HVariableInputSizeInstruction {
public:
// A SIMD operation looks like a FPU location.
// TODO: we could introduce SIMD types in HIR.
static constexpr DataType::Type kSIMDType = DataType::Type::kFloat64;
HVecOperation(InstructionKind kind,
ArenaAllocator* allocator,
DataType::Type packed_type,
SideEffects side_effects,
size_t number_of_inputs,
size_t vector_length,
uint32_t dex_pc)
: HVariableInputSizeInstruction(kind,
kSIMDType,
side_effects,
dex_pc,
allocator,
number_of_inputs,
kArenaAllocVectorNode),
vector_length_(vector_length) {
SetPackedField<PackedTypeField>(packed_type);
DCHECK_LT(1u, vector_length);
}
// Returns the number of elements packed in a vector.
size_t GetVectorLength() const {
return vector_length_;
}
// Returns the number of bytes in a full vector.
size_t GetVectorNumberOfBytes() const {
return vector_length_ * DataType::Size(GetPackedType());
}
// Returns the true component type packed in a vector.
DataType::Type GetPackedType() const {
return GetPackedField<PackedTypeField>();
}
// Assumes vector nodes cannot be moved by default. Each concrete implementation
// that can be moved should override this method and return true.
//
// Note: similar approach is used for instruction scheduling (if it is turned on for the target):
// by default HScheduler::IsSchedulable returns false for a particular HVecOperation.
// HScheduler${ARCH}::IsSchedulable can be overridden to return true for an instruction (see
// scheduler_arm64.h for example) if it is safe to schedule it; in this case one *must* also
// look at/update HScheduler${ARCH}::IsSchedulingBarrier for this instruction.
//
// Note: For newly introduced vector instructions HScheduler${ARCH}::IsSchedulingBarrier must be
// altered to return true if the instruction might reside outside the SIMD loop body since SIMD
// registers are not kept alive across vector loop boundaries (yet).
bool CanBeMoved() const override { return false; }
// Tests if all data of a vector node (vector length and packed type) is equal.
// Each concrete implementation that adds more fields should test equality of
// those fields in its own method *and* call all super methods.
bool InstructionDataEquals(const HInstruction* other) const override {
DCHECK(other->IsVecOperation());
const HVecOperation* o = other->AsVecOperation();
return GetVectorLength() == o->GetVectorLength() && GetPackedType() == o->GetPackedType();
}
// Maps an integral type to the same-size signed type and leaves other types alone.
static DataType::Type ToSignedType(DataType::Type type) {
switch (type) {
case DataType::Type::kBool: // 1-byte storage unit
case DataType::Type::kUint8:
return DataType::Type::kInt8;
case DataType::Type::kUint16:
return DataType::Type::kInt16;
default:
DCHECK(type != DataType::Type::kVoid && type != DataType::Type::kReference) << type;
return type;
}
}
// Maps an integral type to the same-size unsigned type and leaves other types alone.
static DataType::Type ToUnsignedType(DataType::Type type) {
switch (type) {
case DataType::Type::kBool: // 1-byte storage unit
case DataType::Type::kInt8:
return DataType::Type::kUint8;
case DataType::Type::kInt16:
return DataType::Type::kUint16;
default:
DCHECK(type != DataType::Type::kVoid && type != DataType::Type::kReference) << type;
return type;
}
}
// Maps an integral type to the same-size (un)signed type. Leaves other types alone.
static DataType::Type ToProperType(DataType::Type type, bool is_unsigned) {
return is_unsigned ? ToUnsignedType(type) : ToSignedType(type);
}
// Helper method to determine if an instruction returns a SIMD value.
// TODO: This method is needed until we introduce SIMD as proper type.
static bool ReturnsSIMDValue(HInstruction* instruction) {
if (instruction->IsVecOperation()) {
return !instruction->IsVecExtractScalar(); // only scalar returning vec op
} else if (instruction->IsPhi()) {
// Vectorizer only uses Phis in reductions, so checking for a 2-way phi
// with a direct vector operand as second argument suffices.
return
instruction->GetType() == kSIMDType &&
instruction->InputCount() == 2 &&
instruction->InputAt(1)->IsVecOperation();
}
return false;
}
DECLARE_ABSTRACT_INSTRUCTION(VecOperation);
protected:
// Additional packed bits.
static constexpr size_t kFieldPackedType = HInstruction::kNumberOfGenericPackedBits;
static constexpr size_t kFieldPackedTypeSize =
MinimumBitsToStore(static_cast<size_t>(DataType::Type::kLast));
static constexpr size_t kNumberOfVectorOpPackedBits = kFieldPackedType + kFieldPackedTypeSize;
static_assert(kNumberOfVectorOpPackedBits <= kMaxNumberOfPackedBits, "Too many packed fields.");
using PackedTypeField = BitField<DataType::Type, kFieldPackedType, kFieldPackedTypeSize>;
DEFAULT_COPY_CONSTRUCTOR(VecOperation);
private:
const size_t vector_length_;
};
// Abstraction of a unary vector operation.
class HVecUnaryOperation : public HVecOperation {
public:
HVecUnaryOperation(InstructionKind kind,
ArenaAllocator* allocator,
HInstruction* input,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecOperation(kind,
allocator,
packed_type,
SideEffects::None(),
/* number_of_inputs= */ 1,
vector_length,
dex_pc) {
SetRawInputAt(0, input);
}
HInstruction* GetInput() const { return InputAt(0); }
DECLARE_ABSTRACT_INSTRUCTION(VecUnaryOperation);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecUnaryOperation);
};
// Abstraction of a binary vector operation.
class HVecBinaryOperation : public HVecOperation {
public:
HVecBinaryOperation(InstructionKind kind,
ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecOperation(kind,
allocator,
packed_type,
SideEffects::None(),
/* number_of_inputs= */ 2,
vector_length,
dex_pc) {
SetRawInputAt(0, left);
SetRawInputAt(1, right);
}
HInstruction* GetLeft() const { return InputAt(0); }
HInstruction* GetRight() const { return InputAt(1); }
DECLARE_ABSTRACT_INSTRUCTION(VecBinaryOperation);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecBinaryOperation);
};
// Abstraction of a vector operation that references memory, with an alignment.
// The Android runtime guarantees elements have at least natural alignment.
class HVecMemoryOperation : public HVecOperation {
public:
HVecMemoryOperation(InstructionKind kind,
ArenaAllocator* allocator,
DataType::Type packed_type,
SideEffects side_effects,
size_t number_of_inputs,
size_t vector_length,
uint32_t dex_pc)
: HVecOperation(kind,
allocator,
packed_type,
side_effects,
number_of_inputs,
vector_length,
dex_pc),
alignment_(DataType::Size(packed_type), 0) {
DCHECK_GE(number_of_inputs, 2u);
}
void SetAlignment(Alignment alignment) { alignment_ = alignment; }
Alignment GetAlignment() const { return alignment_; }
HInstruction* GetArray() const { return InputAt(0); }
HInstruction* GetIndex() const { return InputAt(1); }
bool InstructionDataEquals(const HInstruction* other) const override {
DCHECK(other->IsVecMemoryOperation());
const HVecMemoryOperation* o = other->AsVecMemoryOperation();
return HVecOperation::InstructionDataEquals(o) && GetAlignment() == o->GetAlignment();
}
DECLARE_ABSTRACT_INSTRUCTION(VecMemoryOperation);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecMemoryOperation);
private:
Alignment alignment_;
};
// Packed type consistency checker ("same vector length" integral types may mix freely).
// Tests relaxed type consistency in which packed same-size integral types can co-exist,
// but other type mixes are an error.
inline static bool HasConsistentPackedTypes(HInstruction* input, DataType::Type type) {
if (input->IsPhi()) {
return input->GetType() == HVecOperation::kSIMDType; // carries SIMD
}
DCHECK(input->IsVecOperation());
DataType::Type input_type = input->AsVecOperation()->GetPackedType();
DCHECK_EQ(HVecOperation::ToUnsignedType(input_type) == HVecOperation::ToUnsignedType(type),
HVecOperation::ToSignedType(input_type) == HVecOperation::ToSignedType(type));
return HVecOperation::ToSignedType(input_type) == HVecOperation::ToSignedType(type);
}
//
// Definitions of concrete unary vector operations in HIR.
//
// Replicates the given scalar into a vector,
// viz. replicate(x) = [ x, .. , x ].
class HVecReplicateScalar final : public HVecUnaryOperation {
public:
HVecReplicateScalar(ArenaAllocator* allocator,
HInstruction* scalar,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecUnaryOperation(
kVecReplicateScalar, allocator, scalar, packed_type, vector_length, dex_pc) {
DCHECK(!ReturnsSIMDValue(scalar));
}
// A replicate needs to stay in place, since SIMD registers are not
// kept alive across vector loop boundaries (yet).
bool CanBeMoved() const override { return false; }
DECLARE_INSTRUCTION(VecReplicateScalar);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecReplicateScalar);
};
// Extracts a particular scalar from the given vector,
// viz. extract[ x1, .. , xn ] = x_i.
//
// TODO: for now only i == 1 case supported.
class HVecExtractScalar final : public HVecUnaryOperation {
public:
HVecExtractScalar(ArenaAllocator* allocator,
HInstruction* input,
DataType::Type packed_type,
size_t vector_length,
size_t index,
uint32_t dex_pc)
: HVecUnaryOperation(
kVecExtractScalar, allocator, input, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(input, packed_type));
DCHECK_LT(index, vector_length);
DCHECK_EQ(index, 0u);
// Yields a single component in the vector.
// Overrides the kSIMDType set by the VecOperation constructor.
SetPackedField<TypeField>(packed_type);
}
// An extract needs to stay in place, since SIMD registers are not
// kept alive across vector loop boundaries (yet).
bool CanBeMoved() const override { return false; }
DECLARE_INSTRUCTION(VecExtractScalar);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecExtractScalar);
};
// Reduces the given vector into the first element as sum/min/max,
// viz. sum-reduce[ x1, .. , xn ] = [ y, ---- ], where y = sum xi
// and the "-" denotes "don't care" (implementation dependent).
class HVecReduce final : public HVecUnaryOperation {
public:
enum ReductionKind {
kSum = 1,
kMin = 2,
kMax = 3
};
HVecReduce(ArenaAllocator* allocator,
HInstruction* input,
DataType::Type packed_type,
size_t vector_length,
ReductionKind reduction_kind,
uint32_t dex_pc)
: HVecUnaryOperation(kVecReduce, allocator, input, packed_type, vector_length, dex_pc),
reduction_kind_(reduction_kind) {
DCHECK(HasConsistentPackedTypes(input, packed_type));
}
ReductionKind GetReductionKind() const { return reduction_kind_; }
bool CanBeMoved() const override { return true; }
bool InstructionDataEquals(const HInstruction* other) const override {
DCHECK(other->IsVecReduce());
const HVecReduce* o = other->AsVecReduce();
return HVecOperation::InstructionDataEquals(o) && GetReductionKind() == o->GetReductionKind();
}
DECLARE_INSTRUCTION(VecReduce);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecReduce);
private:
const ReductionKind reduction_kind_;
};
// Converts every component in the vector,
// viz. cnv[ x1, .. , xn ] = [ cnv(x1), .. , cnv(xn) ].
class HVecCnv final : public HVecUnaryOperation {
public:
HVecCnv(ArenaAllocator* allocator,
HInstruction* input,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecUnaryOperation(kVecCnv, allocator, input, packed_type, vector_length, dex_pc) {
DCHECK(input->IsVecOperation());
DCHECK_NE(GetInputType(), GetResultType()); // actual convert
}
DataType::Type GetInputType() const { return InputAt(0)->AsVecOperation()->GetPackedType(); }
DataType::Type GetResultType() const { return GetPackedType(); }
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecCnv);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecCnv);
};
// Negates every component in the vector,
// viz. neg[ x1, .. , xn ] = [ -x1, .. , -xn ].
class HVecNeg final : public HVecUnaryOperation {
public:
HVecNeg(ArenaAllocator* allocator,
HInstruction* input,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecUnaryOperation(kVecNeg, allocator, input, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(input, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecNeg);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecNeg);
};
// Takes absolute value of every component in the vector,
// viz. abs[ x1, .. , xn ] = [ |x1|, .. , |xn| ]
// for signed operand x.
class HVecAbs final : public HVecUnaryOperation {
public:
HVecAbs(ArenaAllocator* allocator,
HInstruction* input,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecUnaryOperation(kVecAbs, allocator, input, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(input, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecAbs);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecAbs);
};
// Bitwise- or boolean-nots every component in the vector,
// viz. not[ x1, .. , xn ] = [ ~x1, .. , ~xn ], or
// not[ x1, .. , xn ] = [ !x1, .. , !xn ] for boolean.
class HVecNot final : public HVecUnaryOperation {
public:
HVecNot(ArenaAllocator* allocator,
HInstruction* input,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecUnaryOperation(kVecNot, allocator, input, packed_type, vector_length, dex_pc) {
DCHECK(input->IsVecOperation());
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecNot);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecNot);
};
//
// Definitions of concrete binary vector operations in HIR.
//
// Adds every component in the two vectors,
// viz. [ x1, .. , xn ] + [ y1, .. , yn ] = [ x1 + y1, .. , xn + yn ].
class HVecAdd final : public HVecBinaryOperation {
public:
HVecAdd(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecAdd, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
DCHECK(HasConsistentPackedTypes(right, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecAdd);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecAdd);
};
// Adds every component in the two vectors using saturation arithmetic,
// viz. [ x1, .. , xn ] + [ y1, .. , yn ] = [ x1 +_sat y1, .. , xn +_sat yn ]
// for either both signed or both unsigned operands x, y (reflected in packed_type).
class HVecSaturationAdd final : public HVecBinaryOperation {
public:
HVecSaturationAdd(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(
kVecSaturationAdd, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
DCHECK(HasConsistentPackedTypes(right, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecSaturationAdd);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecSaturationAdd);
};
// Performs halving add on every component in the two vectors, viz.
// rounded [ x1, .. , xn ] hradd [ y1, .. , yn ] = [ (x1 + y1 + 1) >> 1, .. , (xn + yn + 1) >> 1 ]
// truncated [ x1, .. , xn ] hadd [ y1, .. , yn ] = [ (x1 + y1) >> 1, .. , (xn + yn ) >> 1 ]
// for either both signed or both unsigned operands x, y (reflected in packed_type).
class HVecHalvingAdd final : public HVecBinaryOperation {
public:
HVecHalvingAdd(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
bool is_rounded,
uint32_t dex_pc)
: HVecBinaryOperation(
kVecHalvingAdd, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
DCHECK(HasConsistentPackedTypes(right, packed_type));
SetPackedFlag<kFieldHAddIsRounded>(is_rounded);
}
bool IsRounded() const { return GetPackedFlag<kFieldHAddIsRounded>(); }
bool CanBeMoved() const override { return true; }
bool InstructionDataEquals(const HInstruction* other) const override {
DCHECK(other->IsVecHalvingAdd());
const HVecHalvingAdd* o = other->AsVecHalvingAdd();
return HVecOperation::InstructionDataEquals(o) && IsRounded() == o->IsRounded();
}
DECLARE_INSTRUCTION(VecHalvingAdd);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecHalvingAdd);
private:
// Additional packed bits.
static constexpr size_t kFieldHAddIsRounded = HVecOperation::kNumberOfVectorOpPackedBits;
static constexpr size_t kNumberOfHAddPackedBits = kFieldHAddIsRounded + 1;
static_assert(kNumberOfHAddPackedBits <= kMaxNumberOfPackedBits, "Too many packed fields.");
};
// Subtracts every component in the two vectors,
// viz. [ x1, .. , xn ] - [ y1, .. , yn ] = [ x1 - y1, .. , xn - yn ].
class HVecSub final : public HVecBinaryOperation {
public:
HVecSub(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecSub, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
DCHECK(HasConsistentPackedTypes(right, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecSub);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecSub);
};
// Subtracts every component in the two vectors using saturation arithmetic,
// viz. [ x1, .. , xn ] + [ y1, .. , yn ] = [ x1 -_sat y1, .. , xn -_sat yn ]
// for either both signed or both unsigned operands x, y (reflected in packed_type).
class HVecSaturationSub final : public HVecBinaryOperation {
public:
HVecSaturationSub(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(
kVecSaturationSub, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
DCHECK(HasConsistentPackedTypes(right, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecSaturationSub);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecSaturationSub);
};
// Multiplies every component in the two vectors,
// viz. [ x1, .. , xn ] * [ y1, .. , yn ] = [ x1 * y1, .. , xn * yn ].
class HVecMul final : public HVecBinaryOperation {
public:
HVecMul(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecMul, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
DCHECK(HasConsistentPackedTypes(right, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecMul);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecMul);
};
// Divides every component in the two vectors,
// viz. [ x1, .. , xn ] / [ y1, .. , yn ] = [ x1 / y1, .. , xn / yn ].
class HVecDiv final : public HVecBinaryOperation {
public:
HVecDiv(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecDiv, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
DCHECK(HasConsistentPackedTypes(right, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecDiv);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecDiv);
};
// Takes minimum of every component in the two vectors,
// viz. MIN( [ x1, .. , xn ] , [ y1, .. , yn ]) = [ min(x1, y1), .. , min(xn, yn) ]
// for either both signed or both unsigned operands x, y (reflected in packed_type).
class HVecMin final : public HVecBinaryOperation {
public:
HVecMin(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecMin, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
DCHECK(HasConsistentPackedTypes(right, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecMin);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecMin);
};
// Takes maximum of every component in the two vectors,
// viz. MAX( [ x1, .. , xn ] , [ y1, .. , yn ]) = [ max(x1, y1), .. , max(xn, yn) ]
// for either both signed or both unsigned operands x, y (reflected in packed_type).
class HVecMax final : public HVecBinaryOperation {
public:
HVecMax(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecMax, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
DCHECK(HasConsistentPackedTypes(right, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecMax);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecMax);
};
// Bitwise-ands every component in the two vectors,
// viz. [ x1, .. , xn ] & [ y1, .. , yn ] = [ x1 & y1, .. , xn & yn ].
class HVecAnd final : public HVecBinaryOperation {
public:
HVecAnd(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecAnd, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(left->IsVecOperation() && right->IsVecOperation());
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecAnd);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecAnd);
};
// Bitwise-and-nots every component in the two vectors,
// viz. [ x1, .. , xn ] and-not [ y1, .. , yn ] = [ ~x1 & y1, .. , ~xn & yn ].
class HVecAndNot final : public HVecBinaryOperation {
public:
HVecAndNot(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(
kVecAndNot, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(left->IsVecOperation() && right->IsVecOperation());
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecAndNot);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecAndNot);
};
// Bitwise-ors every component in the two vectors,
// viz. [ x1, .. , xn ] | [ y1, .. , yn ] = [ x1 | y1, .. , xn | yn ].
class HVecOr final : public HVecBinaryOperation {
public:
HVecOr(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecOr, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(left->IsVecOperation() && right->IsVecOperation());
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecOr);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecOr);
};
// Bitwise-xors every component in the two vectors,
// viz. [ x1, .. , xn ] ^ [ y1, .. , yn ] = [ x1 ^ y1, .. , xn ^ yn ].
class HVecXor final : public HVecBinaryOperation {
public:
HVecXor(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecXor, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(left->IsVecOperation() && right->IsVecOperation());
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecXor);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecXor);
};
// Logically shifts every component in the vector left by the given distance,
// viz. [ x1, .. , xn ] << d = [ x1 << d, .. , xn << d ].
class HVecShl final : public HVecBinaryOperation {
public:
HVecShl(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecShl, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecShl);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecShl);
};
// Arithmetically shifts every component in the vector right by the given distance,
// viz. [ x1, .. , xn ] >> d = [ x1 >> d, .. , xn >> d ].
class HVecShr final : public HVecBinaryOperation {
public:
HVecShr(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecShr, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecShr);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecShr);
};
// Logically shifts every component in the vector right by the given distance,
// viz. [ x1, .. , xn ] >>> d = [ x1 >>> d, .. , xn >>> d ].
class HVecUShr final : public HVecBinaryOperation {
public:
HVecUShr(ArenaAllocator* allocator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecBinaryOperation(kVecUShr, allocator, left, right, packed_type, vector_length, dex_pc) {
DCHECK(HasConsistentPackedTypes(left, packed_type));
}
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecUShr);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecUShr);
};
//
// Definitions of concrete miscellaneous vector operations in HIR.
//
// Assigns the given scalar elements to a vector,
// viz. set( array(x1, .. , xn) ) = [ x1, .. , xn ] if n == m,
// set( array(x1, .. , xm) ) = [ x1, .. , xm, 0, .. , 0 ] if m < n.
class HVecSetScalars final : public HVecOperation {
public:
HVecSetScalars(ArenaAllocator* allocator,
HInstruction* scalars[],
DataType::Type packed_type,
size_t vector_length,
size_t number_of_scalars,
uint32_t dex_pc)
: HVecOperation(kVecSetScalars,
allocator,
packed_type,
SideEffects::None(),
number_of_scalars,
vector_length,
dex_pc) {
for (size_t i = 0; i < number_of_scalars; i++) {
DCHECK(!ReturnsSIMDValue(scalars[i]));
SetRawInputAt(0, scalars[i]);
}
}
// Setting scalars needs to stay in place, since SIMD registers are not
// kept alive across vector loop boundaries (yet).
bool CanBeMoved() const override { return false; }
DECLARE_INSTRUCTION(VecSetScalars);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecSetScalars);
};
// Multiplies every component in the two vectors, adds the result vector to the accumulator vector,
// viz. [ a1, .. , an ] + [ x1, .. , xn ] * [ y1, .. , yn ] = [ a1 + x1 * y1, .. , an + xn * yn ].
// For floating point types, Java rounding behavior must be preserved; the products are rounded to
// the proper precision before being added. "Fused" multiply-add operations available on several
// architectures are not usable since they would violate Java language rules.
class HVecMultiplyAccumulate final : public HVecOperation {
public:
HVecMultiplyAccumulate(ArenaAllocator* allocator,
InstructionKind op,
HInstruction* accumulator,
HInstruction* mul_left,
HInstruction* mul_right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecOperation(kVecMultiplyAccumulate,
allocator,
packed_type,
SideEffects::None(),
/* number_of_inputs= */ 3,
vector_length,
dex_pc),
op_kind_(op) {
DCHECK(op == InstructionKind::kAdd || op == InstructionKind::kSub);
DCHECK(HasConsistentPackedTypes(accumulator, packed_type));
DCHECK(HasConsistentPackedTypes(mul_left, packed_type));
DCHECK(HasConsistentPackedTypes(mul_right, packed_type));
// Remove the following if we add an architecture that supports floating point multiply-add
// with Java-compatible rounding.
DCHECK(DataType::IsIntegralType(packed_type));
SetRawInputAt(0, accumulator);
SetRawInputAt(1, mul_left);
SetRawInputAt(2, mul_right);
}
bool CanBeMoved() const override { return true; }
bool InstructionDataEquals(const HInstruction* other) const override {
DCHECK(other->IsVecMultiplyAccumulate());
const HVecMultiplyAccumulate* o = other->AsVecMultiplyAccumulate();
return HVecOperation::InstructionDataEquals(o) && GetOpKind() == o->GetOpKind();
}
InstructionKind GetOpKind() const { return op_kind_; }
DECLARE_INSTRUCTION(VecMultiplyAccumulate);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecMultiplyAccumulate);
private:
// Indicates if this is a MADD or MSUB.
const InstructionKind op_kind_;
};
// Takes the absolute difference of two vectors, and adds the results to
// same-precision or wider-precision components in the accumulator,
// viz. SAD([ a1, .. , am ], [ x1, .. , xn ], [ y1, .. , yn ]) =
// [ a1 + sum abs(xi-yi), .. , am + sum abs(xj-yj) ],
// for m <= n, non-overlapping sums, and signed operands x, y.
class HVecSADAccumulate final : public HVecOperation {
public:
HVecSADAccumulate(ArenaAllocator* allocator,
HInstruction* accumulator,
HInstruction* sad_left,
HInstruction* sad_right,
DataType::Type packed_type,
size_t vector_length,
uint32_t dex_pc)
: HVecOperation(kVecSADAccumulate,
allocator,
packed_type,
SideEffects::None(),
/* number_of_inputs= */ 3,
vector_length,
dex_pc) {
DCHECK(HasConsistentPackedTypes(accumulator, packed_type));
DCHECK(sad_left->IsVecOperation());
DCHECK(sad_right->IsVecOperation());
DCHECK_EQ(ToSignedType(sad_left->AsVecOperation()->GetPackedType()),
ToSignedType(sad_right->AsVecOperation()->GetPackedType()));
SetRawInputAt(0, accumulator);
SetRawInputAt(1, sad_left);
SetRawInputAt(2, sad_right);
}
DECLARE_INSTRUCTION(VecSADAccumulate);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecSADAccumulate);
};
// Performs dot product of two vectors and adds the result to wider precision components in
// the accumulator.
//
// viz. DOT_PRODUCT([ a1, .. , am], [ x1, .. , xn ], [ y1, .. , yn ]) =
// [ a1 + sum(xi * yi), .. , am + sum(xj * yj) ],
// for m <= n, non-overlapping sums,
// for either both signed or both unsigned operands x, y.
//
// Notes:
// - packed type reflects the type of sum reduction, not the type of the operands.
// - IsZeroExtending() is used to determine the kind of signed/zero extension to be
// performed for the operands.
//
// TODO: Support types other than kInt32 for packed type.
class HVecDotProd final : public HVecOperation {
public:
HVecDotProd(ArenaAllocator* allocator,
HInstruction* accumulator,
HInstruction* left,
HInstruction* right,
DataType::Type packed_type,
bool is_zero_extending,
size_t vector_length,
uint32_t dex_pc)
: HVecOperation(kVecDotProd,
allocator,
packed_type,
SideEffects::None(),
/* number_of_inputs= */ 3,
vector_length,
dex_pc) {
DCHECK(HasConsistentPackedTypes(accumulator, packed_type));
DCHECK(DataType::IsIntegralType(packed_type));
DCHECK(left->IsVecOperation());
DCHECK(right->IsVecOperation());
DCHECK_EQ(ToSignedType(left->AsVecOperation()->GetPackedType()),
ToSignedType(right->AsVecOperation()->GetPackedType()));
SetRawInputAt(0, accumulator);
SetRawInputAt(1, left);
SetRawInputAt(2, right);
SetPackedFlag<kFieldHDotProdIsZeroExtending>(is_zero_extending);
}
bool IsZeroExtending() const { return GetPackedFlag<kFieldHDotProdIsZeroExtending>(); }
bool CanBeMoved() const override { return true; }
DECLARE_INSTRUCTION(VecDotProd);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecDotProd);
private:
// Additional packed bits.
static constexpr size_t kFieldHDotProdIsZeroExtending =
HVecOperation::kNumberOfVectorOpPackedBits;
static constexpr size_t kNumberOfHDotProdPackedBits = kFieldHDotProdIsZeroExtending + 1;
static_assert(kNumberOfHDotProdPackedBits <= kMaxNumberOfPackedBits, "Too many packed fields.");
};
// Loads a vector from memory, viz. load(mem, 1)
// yield the vector [ mem(1), .. , mem(n) ].
class HVecLoad final : public HVecMemoryOperation {
public:
HVecLoad(ArenaAllocator* allocator,
HInstruction* base,
HInstruction* index,
DataType::Type packed_type,
SideEffects side_effects,
size_t vector_length,
bool is_string_char_at,
uint32_t dex_pc)
: HVecMemoryOperation(kVecLoad,
allocator,
packed_type,
side_effects,
/* number_of_inputs= */ 2,
vector_length,
dex_pc) {
SetRawInputAt(0, base);
SetRawInputAt(1, index);
SetPackedFlag<kFieldIsStringCharAt>(is_string_char_at);
}
bool IsStringCharAt() const { return GetPackedFlag<kFieldIsStringCharAt>(); }
bool CanBeMoved() const override { return true; }
bool InstructionDataEquals(const HInstruction* other) const override {
DCHECK(other->IsVecLoad());
const HVecLoad* o = other->AsVecLoad();
return HVecMemoryOperation::InstructionDataEquals(o) && IsStringCharAt() == o->IsStringCharAt();
}
DECLARE_INSTRUCTION(VecLoad);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecLoad);
private:
// Additional packed bits.
static constexpr size_t kFieldIsStringCharAt = HVecOperation::kNumberOfVectorOpPackedBits;
static constexpr size_t kNumberOfVecLoadPackedBits = kFieldIsStringCharAt + 1;
static_assert(kNumberOfVecLoadPackedBits <= kMaxNumberOfPackedBits, "Too many packed fields.");
};
// Stores a vector to memory, viz. store(m, 1, [x1, .. , xn] )
// sets mem(1) = x1, .. , mem(n) = xn.
class HVecStore final : public HVecMemoryOperation {
public:
HVecStore(ArenaAllocator* allocator,
HInstruction* base,
HInstruction* index,
HInstruction* value,
DataType::Type packed_type,
SideEffects side_effects,
size_t vector_length,
uint32_t dex_pc)
: HVecMemoryOperation(kVecStore,
allocator,
packed_type,
side_effects,
/* number_of_inputs= */ 3,
vector_length,
dex_pc) {
DCHECK(HasConsistentPackedTypes(value, packed_type));
SetRawInputAt(0, base);
SetRawInputAt(1, index);
SetRawInputAt(2, value);
}
// A store needs to stay in place.
bool CanBeMoved() const override { return false; }
DECLARE_INSTRUCTION(VecStore);
protected:
DEFAULT_COPY_CONSTRUCTOR(VecStore)
};
} // namespace art
#endif // ART_COMPILER_OPTIMIZING_NODES_VECTOR_H_