Files
clang-p2996/flang/lib/Evaluate/complex.cpp
Peter Klausler 2b5cd3778c [flang] Use naive algorithm for folding complex division when it doesn't over/underflow
f18 unconditionally uses a scaling algorithm for complex/complex division
that avoids needless overflows and underflows when computing the sum of
the squares of the components of the denominator -- but testing has shown
some 1 ULP differences relative to the naive calculation due to the
extra operations and roundings.  So use the scaling algorithm only when
the naive calculation actually would overflow or underflow.

Differential Revision: https://reviews.llvm.org/D132164
2022-08-18 15:11:34 -07:00

126 lines
5.0 KiB
C++

//===-- lib/Evaluate/complex.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
//
//===----------------------------------------------------------------------===//
#include "flang/Evaluate/complex.h"
#include "llvm/Support/raw_ostream.h"
namespace Fortran::evaluate::value {
template <typename R>
ValueWithRealFlags<Complex<R>> Complex<R>::Add(
const Complex &that, Rounding rounding) const {
RealFlags flags;
Part reSum{re_.Add(that.re_, rounding).AccumulateFlags(flags)};
Part imSum{im_.Add(that.im_, rounding).AccumulateFlags(flags)};
return {Complex{reSum, imSum}, flags};
}
template <typename R>
ValueWithRealFlags<Complex<R>> Complex<R>::Subtract(
const Complex &that, Rounding rounding) const {
RealFlags flags;
Part reDiff{re_.Subtract(that.re_, rounding).AccumulateFlags(flags)};
Part imDiff{im_.Subtract(that.im_, rounding).AccumulateFlags(flags)};
return {Complex{reDiff, imDiff}, flags};
}
template <typename R>
ValueWithRealFlags<Complex<R>> Complex<R>::Multiply(
const Complex &that, Rounding rounding) const {
// (a + ib)*(c + id) -> ac - bd + i(ad + bc)
RealFlags flags;
Part ac{re_.Multiply(that.re_, rounding).AccumulateFlags(flags)};
Part bd{im_.Multiply(that.im_, rounding).AccumulateFlags(flags)};
Part ad{re_.Multiply(that.im_, rounding).AccumulateFlags(flags)};
Part bc{im_.Multiply(that.re_, rounding).AccumulateFlags(flags)};
Part acbd{ac.Subtract(bd, rounding).AccumulateFlags(flags)};
Part adbc{ad.Add(bc, rounding).AccumulateFlags(flags)};
return {Complex{acbd, adbc}, flags};
}
template <typename R>
ValueWithRealFlags<Complex<R>> Complex<R>::Divide(
const Complex &that, Rounding rounding) const {
// (a + ib)/(c + id) -> [(a+ib)*(c-id)] / [(c+id)*(c-id)]
// -> [ac+bd+i(bc-ad)] / (cc+dd) -- note (cc+dd) is real
// -> ((ac+bd)/(cc+dd)) + i((bc-ad)/(cc+dd))
RealFlags flags;
Part cc{that.re_.Multiply(that.re_, rounding).AccumulateFlags(flags)};
Part dd{that.im_.Multiply(that.im_, rounding).AccumulateFlags(flags)};
Part ccPdd{cc.Add(dd, rounding).AccumulateFlags(flags)};
if (!flags.test(RealFlag::Overflow) && !flags.test(RealFlag::Underflow)) {
// den = (cc+dd) did not overflow or underflow; try the naive
// sequence without scaling to avoid extra roundings.
Part ac{re_.Multiply(that.re_, rounding).AccumulateFlags(flags)};
Part ad{re_.Multiply(that.im_, rounding).AccumulateFlags(flags)};
Part bc{im_.Multiply(that.re_, rounding).AccumulateFlags(flags)};
Part bd{im_.Multiply(that.im_, rounding).AccumulateFlags(flags)};
Part acPbd{ac.Add(bd, rounding).AccumulateFlags(flags)};
Part bcSad{bc.Subtract(ad, rounding).AccumulateFlags(flags)};
Part re{acPbd.Divide(ccPdd, rounding).AccumulateFlags(flags)};
Part im{bcSad.Divide(ccPdd, rounding).AccumulateFlags(flags)};
if (!flags.test(RealFlag::Overflow) && !flags.test(RealFlag::Underflow)) {
return {Complex{re, im}, flags};
}
}
// Scale numerator and denominator by d/c (if c>=d) or c/d (if c<d)
flags.clear();
Part scale; // will be <= 1.0 in magnitude
bool cGEd{that.re_.ABS().Compare(that.im_.ABS()) != Relation::Less};
if (cGEd) {
scale = that.im_.Divide(that.re_, rounding).AccumulateFlags(flags);
} else {
scale = that.re_.Divide(that.im_, rounding).AccumulateFlags(flags);
}
Part den;
if (cGEd) {
Part dS{scale.Multiply(that.im_, rounding).AccumulateFlags(flags)};
den = dS.Add(that.re_, rounding).AccumulateFlags(flags);
} else {
Part cS{scale.Multiply(that.re_, rounding).AccumulateFlags(flags)};
den = cS.Add(that.im_, rounding).AccumulateFlags(flags);
}
Part aS{scale.Multiply(re_, rounding).AccumulateFlags(flags)};
Part bS{scale.Multiply(im_, rounding).AccumulateFlags(flags)};
Part re1, im1;
if (cGEd) {
re1 = re_.Add(bS, rounding).AccumulateFlags(flags);
im1 = im_.Subtract(aS, rounding).AccumulateFlags(flags);
} else {
re1 = aS.Add(im_, rounding).AccumulateFlags(flags);
im1 = bS.Subtract(re_, rounding).AccumulateFlags(flags);
}
Part re{re1.Divide(den, rounding).AccumulateFlags(flags)};
Part im{im1.Divide(den, rounding).AccumulateFlags(flags)};
return {Complex{re, im}, flags};
}
template <typename R> std::string Complex<R>::DumpHexadecimal() const {
std::string result{'('};
result += re_.DumpHexadecimal();
result += ',';
result += im_.DumpHexadecimal();
result += ')';
return result;
}
template <typename R>
llvm::raw_ostream &Complex<R>::AsFortran(llvm::raw_ostream &o, int kind) const {
re_.AsFortran(o << '(', kind);
im_.AsFortran(o << ',', kind);
return o << ')';
}
template class Complex<Real<Integer<16>, 11>>;
template class Complex<Real<Integer<16>, 8>>;
template class Complex<Real<Integer<32>, 24>>;
template class Complex<Real<Integer<64>, 53>>;
template class Complex<Real<Integer<80>, 64>>;
template class Complex<Real<Integer<128>, 113>>;
} // namespace Fortran::evaluate::value