llvm-capstone/flang/runtime/sum.cpp
peter klausler c375ec8613 [flang] Use a "double-double" accumulator in SUM
Use a "double-double" accumulator, a/k/a Kahan summation,
in the SUM intrinsic in the runtime for real & complex.
This seems to be the best-recommended technique for reducing
error, as opposed to the initial implementation of SUM's
distinct accumulators for positive and negative items.

Differential Revision: https://reviews.llvm.org/D104338
2021-06-16 14:29:39 -07:00

181 lines
7.1 KiB
C++

//===-- runtime/sum.cpp ---------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
// Implements SUM for all required operand types and shapes.
//
// Real and complex SUM reductions attempt to reduce floating-point
// cancellation on intermediate results by using "Kahan summation"
// (basically the same as manual "double-double").
#include "reduction-templates.h"
#include "reduction.h"
#include "flang/Common/long-double.h"
#include <cinttypes>
#include <complex>
namespace Fortran::runtime {
template <typename INTERMEDIATE> class IntegerSumAccumulator {
public:
explicit IntegerSumAccumulator(const Descriptor &array) : array_{array} {}
void Reinitialize() { sum_ = 0; }
template <typename A> void GetResult(A *p, int /*zeroBasedDim*/ = -1) const {
*p = static_cast<A>(sum_);
}
template <typename A> bool AccumulateAt(const SubscriptValue at[]) {
sum_ += *array_.Element<A>(at);
return true;
}
private:
const Descriptor &array_;
INTERMEDIATE sum_{0};
};
template <typename INTERMEDIATE> class RealSumAccumulator {
public:
explicit RealSumAccumulator(const Descriptor &array) : array_{array} {}
void Reinitialize() { sum_ = correction_ = 0; }
template <typename A> A Result() const { return sum_; }
template <typename A> void GetResult(A *p, int /*zeroBasedDim*/ = -1) const {
*p = Result<A>();
}
template <typename A> bool Accumulate(A x) {
// Kahan summation
auto next{x + correction_};
auto oldSum{sum_};
sum_ += next;
correction_ = (sum_ - oldSum) - next; // algebraically zero
return true;
}
template <typename A> bool AccumulateAt(const SubscriptValue at[]) {
return Accumulate(*array_.Element<A>(at));
}
private:
const Descriptor &array_;
INTERMEDIATE sum_{0.0}, correction_{0.0};
};
template <typename PART> class ComplexSumAccumulator {
public:
explicit ComplexSumAccumulator(const Descriptor &array) : array_{array} {}
void Reinitialize() {
reals_.Reinitialize();
imaginaries_.Reinitialize();
}
template <typename A> void GetResult(A *p, int /*zeroBasedDim*/ = -1) const {
using ResultPart = typename A::value_type;
*p = {reals_.template Result<ResultPart>(),
imaginaries_.template Result<ResultPart>()};
}
template <typename A> bool Accumulate(const A &z) {
reals_.Accumulate(z.real());
imaginaries_.Accumulate(z.imag());
return true;
}
template <typename A> bool AccumulateAt(const SubscriptValue at[]) {
return Accumulate(*array_.Element<A>(at));
}
private:
const Descriptor &array_;
RealSumAccumulator<PART> reals_{array_}, imaginaries_{array_};
};
extern "C" {
CppTypeFor<TypeCategory::Integer, 1> RTNAME(SumInteger1)(const Descriptor &x,
const char *source, int line, int dim, const Descriptor *mask) {
return GetTotalReduction<TypeCategory::Integer, 1>(x, source, line, dim, mask,
IntegerSumAccumulator<CppTypeFor<TypeCategory::Integer, 4>>{x}, "SUM");
}
CppTypeFor<TypeCategory::Integer, 2> RTNAME(SumInteger2)(const Descriptor &x,
const char *source, int line, int dim, const Descriptor *mask) {
return GetTotalReduction<TypeCategory::Integer, 2>(x, source, line, dim, mask,
IntegerSumAccumulator<CppTypeFor<TypeCategory::Integer, 4>>{x}, "SUM");
}
CppTypeFor<TypeCategory::Integer, 4> RTNAME(SumInteger4)(const Descriptor &x,
const char *source, int line, int dim, const Descriptor *mask) {
return GetTotalReduction<TypeCategory::Integer, 4>(x, source, line, dim, mask,
IntegerSumAccumulator<CppTypeFor<TypeCategory::Integer, 4>>{x}, "SUM");
}
CppTypeFor<TypeCategory::Integer, 8> RTNAME(SumInteger8)(const Descriptor &x,
const char *source, int line, int dim, const Descriptor *mask) {
return GetTotalReduction<TypeCategory::Integer, 8>(x, source, line, dim, mask,
IntegerSumAccumulator<CppTypeFor<TypeCategory::Integer, 8>>{x}, "SUM");
}
#ifdef __SIZEOF_INT128__
CppTypeFor<TypeCategory::Integer, 16> RTNAME(SumInteger16)(const Descriptor &x,
const char *source, int line, int dim, const Descriptor *mask) {
return GetTotalReduction<TypeCategory::Integer, 16>(x, source, line, dim,
mask, IntegerSumAccumulator<CppTypeFor<TypeCategory::Integer, 16>>{x},
"SUM");
}
#endif
// TODO: real/complex(2 & 3)
CppTypeFor<TypeCategory::Real, 4> RTNAME(SumReal4)(const Descriptor &x,
const char *source, int line, int dim, const Descriptor *mask) {
return GetTotalReduction<TypeCategory::Real, 4>(
x, source, line, dim, mask, RealSumAccumulator<double>{x}, "SUM");
}
CppTypeFor<TypeCategory::Real, 8> RTNAME(SumReal8)(const Descriptor &x,
const char *source, int line, int dim, const Descriptor *mask) {
return GetTotalReduction<TypeCategory::Real, 8>(
x, source, line, dim, mask, RealSumAccumulator<double>{x}, "SUM");
}
#if LONG_DOUBLE == 80
CppTypeFor<TypeCategory::Real, 10> RTNAME(SumReal10)(const Descriptor &x,
const char *source, int line, int dim, const Descriptor *mask) {
return GetTotalReduction<TypeCategory::Real, 10>(
x, source, line, dim, mask, RealSumAccumulator<long double>{x}, "SUM");
}
#elif LONG_DOUBLE == 128
CppTypeFor<TypeCategory::Real, 16> RTNAME(SumReal16)(const Descriptor &x,
const char *source, int line, int dim, const Descriptor *mask) {
return GetTotalReduction<TypeCategory::Real, 16>(
x, source, line, dim, mask, RealSumAccumulator<long double>{x}, "SUM");
}
#endif
void RTNAME(CppSumComplex4)(CppTypeFor<TypeCategory::Complex, 4> &result,
const Descriptor &x, const char *source, int line, int dim,
const Descriptor *mask) {
result = GetTotalReduction<TypeCategory::Complex, 4>(
x, source, line, dim, mask, ComplexSumAccumulator<double>{x}, "SUM");
}
void RTNAME(CppSumComplex8)(CppTypeFor<TypeCategory::Complex, 8> &result,
const Descriptor &x, const char *source, int line, int dim,
const Descriptor *mask) {
result = GetTotalReduction<TypeCategory::Complex, 8>(
x, source, line, dim, mask, ComplexSumAccumulator<double>{x}, "SUM");
}
#if LONG_DOUBLE == 80
void RTNAME(CppSumComplex10)(CppTypeFor<TypeCategory::Complex, 10> &result,
const Descriptor &x, const char *source, int line, int dim,
const Descriptor *mask) {
result = GetTotalReduction<TypeCategory::Complex, 10>(
x, source, line, dim, mask, ComplexSumAccumulator<long double>{x}, "SUM");
}
#elif LONG_DOUBLE == 128
void RTNAME(CppSumComplex16)(CppTypeFor<TypeCategory::Complex, 16> &result,
const Descriptor &x, const char *source, int line, int dim,
const Descriptor *mask) {
result = GetTotalReduction<TypeCategory::Complex, 16>(
x, source, line, dim, mask, ComplexSumAccumulator<long double>{x}, "SUM");
}
#endif
void RTNAME(SumDim)(Descriptor &result, const Descriptor &x, int dim,
const char *source, int line, const Descriptor *mask) {
TypedPartialNumericReduction<IntegerSumAccumulator, RealSumAccumulator,
ComplexSumAccumulator>(result, x, dim, source, line, mask, "SUM");
}
} // extern "C"
} // namespace Fortran::runtime