Skip to content

Commit

Permalink
[flang] Fix rounding edge case in F output editing
Browse files Browse the repository at this point in the history 
When an Fw.d output edit descriptor has a "d" value exactly
equal to the number of zeroes after the decimal point for a value
(e.g., 0.07 with F5.1), the Fw.d output editing code needs to
do the rounding itself to either 0.0 or 0.1 after performing
a conversion without rounding (to avoid 0.04999 rounding up twice).

Differential Revision: https://reviews.llvm.org/D113698
  • Loading branch information
@klausler
klausler committed Nov 12, 2021
1 parent f46f93b commit d1b09ad
Show file tree
Hide file tree
Showing 3 changed files with 47 additions and 15 deletions.
56 changes: 42 additions & 14 deletions flang/runtime/edit-output.cpp
View file
Expand Up @@ -159,13 +159,10 @@ bool RealOutputEditingBase::EmitSuffix(const DataEdit &edit) {

template <int binaryPrecision>
decimal::ConversionToDecimalResult RealOutputEditing<binaryPrecision>::Convert(
int significantDigits, const DataEdit &edit, int flags) {
if (edit.modes.editingFlags & signPlus) {
flags |= decimal::AlwaysSign;
}
int significantDigits, enum decimal::FortranRounding rounding, int flags) {
auto converted{decimal::ConvertToDecimal<binaryPrecision>(buffer_,
sizeof buffer_, static_cast<enum decimal::DecimalConversionFlags>(flags),
significantDigits, edit.modes.round, x_)};
significantDigits, rounding, x_)};
if (!converted.str) { // overflow
io_.GetIoErrorHandler().Crash(
"RealOutputEditing::Convert : buffer size %zd was insufficient",
Expand All @@ -181,6 +178,9 @@ bool RealOutputEditing<binaryPrecision>::EditEorDOutput(const DataEdit &edit) {
int editWidth{edit.width.value_or(0)}; // 'w' field
int significantDigits{editDigits};
int flags{0};
if (edit.modes.editingFlags & signPlus) {
flags |= decimal::AlwaysSign;
}
if (editWidth == 0) { // "the processor selects the field width"
if (edit.digits.has_value()) { // E0.d
editWidth = editDigits + 6; // -.666E+ee
Expand All @@ -204,7 +204,7 @@ bool RealOutputEditing<binaryPrecision>::EditEorDOutput(const DataEdit &edit) {
// In EN editing, multiple attempts may be necessary, so it's in a loop.
while (true) {
decimal::ConversionToDecimalResult converted{
Convert(significantDigits, edit, flags)};
Convert(significantDigits, edit.modes.round, flags)};
if (IsInfOrNaN(converted)) {
return EmitPrefix(edit, converted.length, editWidth) &&
io_.Emit(converted.str, converted.length) && EmitSuffix(edit);
Expand Down Expand Up @@ -261,7 +261,11 @@ template <int binaryPrecision>
bool RealOutputEditing<binaryPrecision>::EditFOutput(const DataEdit &edit) {
int fracDigits{edit.digits.value_or(0)}; // 'd' field
const int editWidth{edit.width.value_or(0)}; // 'w' field
enum decimal::FortranRounding rounding{edit.modes.round};
int flags{0};
if (edit.modes.editingFlags & signPlus) {
flags |= decimal::AlwaysSign;
}
if (editWidth == 0) { // "the processor selects the field width"
if (!edit.digits.has_value()) { // F0
flags |= decimal::Minimize;
Expand All @@ -274,40 +278,59 @@ bool RealOutputEditing<binaryPrecision>::EditFOutput(const DataEdit &edit) {
bool canIncrease{true};
while (true) {
decimal::ConversionToDecimalResult converted{
Convert(extraDigits + fracDigits, edit, flags)};
Convert(extraDigits + fracDigits, rounding, flags)};
if (IsInfOrNaN(converted)) {
return EmitPrefix(edit, converted.length, editWidth) &&
io_.Emit(converted.str, converted.length) && EmitSuffix(edit);
}
int scale{IsZero() ? 1 : edit.modes.scale}; // kP
int expo{converted.decimalExponent + scale};
int signLength{*converted.str == '-' || *converted.str == '+' ? 1 : 0};
int convertedDigits{static_cast<int>(converted.length) - signLength};
int trailingOnes{0};
if (expo > extraDigits && extraDigits >= 0 && canIncrease) {
extraDigits = expo;
if (!edit.digits.has_value()) { // F0
fracDigits = sizeof buffer_ - extraDigits - 2; // sign & NUL
}
canIncrease = false; // only once
continue;
} else if (expo == -fracDigits && convertedDigits > 0) {
if (rounding != decimal::FortranRounding::RoundToZero) {
// Convert again without rounding so that we can round here
rounding = decimal::FortranRounding::RoundToZero;
continue;
} else if (converted.str[signLength] >= '5') {
// Value rounds up to a scaled 1 (e.g., 0.06 for F5.1 -> 0.1)
++expo;
convertedDigits = 0;
trailingOnes = 1;
} else {
// Value rounds down to zero
expo = 0;
convertedDigits = 0;
}
} else if (expo < extraDigits && extraDigits > -fracDigits) {
extraDigits = std::max(expo, -fracDigits);
continue;
}
int signLength{*converted.str == '-' || *converted.str == '+' ? 1 : 0};
int convertedDigits{static_cast<int>(converted.length) - signLength};
int digitsBeforePoint{std::max(0, std::min(expo, convertedDigits))};
int zeroesBeforePoint{std::max(0, expo - digitsBeforePoint)};
int zeroesAfterPoint{std::min(fracDigits, std::max(0, -expo))};
int digitsAfterPoint{convertedDigits - digitsBeforePoint};
int trailingZeroes{flags & decimal::Minimize
? 0
: std::max(0, fracDigits - (zeroesAfterPoint + digitsAfterPoint))};
: std::max(0,
fracDigits -
(zeroesAfterPoint + digitsAfterPoint + trailingOnes))};
if (digitsBeforePoint + zeroesBeforePoint + zeroesAfterPoint +
digitsAfterPoint + trailingZeroes ==
digitsAfterPoint + trailingOnes + trailingZeroes ==
0) {
zeroesBeforePoint = 1; // "." -> "0."
}
int totalLength{signLength + digitsBeforePoint + zeroesBeforePoint +
1 /*'.'*/ + zeroesAfterPoint + digitsAfterPoint + trailingZeroes};
1 /*'.'*/ + zeroesAfterPoint + digitsAfterPoint + trailingOnes +
trailingZeroes};
int width{editWidth > 0 ? editWidth : totalLength};
if (totalLength > width) {
return io_.EmitRepeated('*', width);
Expand All @@ -323,6 +346,7 @@ bool RealOutputEditing<binaryPrecision>::EditFOutput(const DataEdit &edit) {
io_.EmitRepeated('0', zeroesAfterPoint) &&
io_.Emit(
converted.str + signLength + digitsBeforePoint, digitsAfterPoint) &&
io_.EmitRepeated('1', trailingOnes) &&
io_.EmitRepeated('0', trailingZeroes) &&
io_.EmitRepeated(' ', trailingBlanks_) && EmitSuffix(edit);
}
Expand All @@ -337,8 +361,12 @@ DataEdit RealOutputEditing<binaryPrecision>::EditForGOutput(DataEdit edit) {
if (!edit.width.has_value() || (*edit.width > 0 && significantDigits == 0)) {
return edit; // Gw.0 -> Ew.0 for w > 0
}
int flags{0};
if (edit.modes.editingFlags & signPlus) {
flags |= decimal::AlwaysSign;
}
decimal::ConversionToDecimalResult converted{
Convert(significantDigits, edit)};
Convert(significantDigits, edit.modes.round, flags)};
if (IsInfOrNaN(converted)) {
return edit;
}
Expand All @@ -365,7 +393,7 @@ DataEdit RealOutputEditing<binaryPrecision>::EditForGOutput(DataEdit edit) {
template <int binaryPrecision>
bool RealOutputEditing<binaryPrecision>::EditListDirectedOutput(
const DataEdit &edit) {
decimal::ConversionToDecimalResult converted{Convert(1, edit)};
decimal::ConversionToDecimalResult converted{Convert(1, edit.modes.round)};
if (IsInfOrNaN(converted)) {
return EditEorDOutput(edit);
}
Expand Down
2 changes: 1 addition & 1 deletion flang/runtime/edit-output.h
View file
Expand Up @@ -84,7 +84,7 @@ template <int KIND> class RealOutputEditing : public RealOutputEditingBase {
bool IsZero() const { return x_.IsZero(); }

decimal::ConversionToDecimalResult Convert(
int significantDigits, const DataEdit &, int flags = 0);
int significantDigits, enum decimal::FortranRounding, int flags = 0);

BinaryFloatingPoint x_;
char buffer_[BinaryFloatingPoint::maxDecimalConversionDigits +
Expand Down
4 changes: 4 additions & 0 deletions flang/unittests/Runtime/NumericalFormatTest.cpp
View file
Expand Up @@ -633,13 +633,17 @@ TEST(IOApiTests, FormatDoubleValues) {
{"(F5.3,';')", 0.099999, "0.100;"},
{"(F5.3,';')", 0.0099999, "0.010;"},
{"(F5.3,';')", 0.00099999, "0.001;"},
{"(F5.3,';')", 0.0005, "0.001;"},
{"(F5.3,';')", 0.00049999, "0.000;"},
{"(F5.3,';')", 0.000099999, "0.000;"},
{"(F5.3,';')", -99.999, "*****;"},
{"(F5.3,';')", -9.9999, "*****;"},
{"(F5.3,';')", -0.99999, "*****;"},
{"(F5.3,';')", -0.099999, "-.100;"},
{"(F5.3,';')", -0.0099999, "-.010;"},
{"(F5.3,';')", -0.00099999, "-.001;"},
{"(F5.3,';')", -0.0005, "-.001;"},
{"(F5.3,';')", -0.00049999, "-.000;"},
{"(F5.3,';')", -0.000099999, "-.000;"},
};

Expand Down

0 comments on commit d1b09ad

Please sign in to comment.