Files
clang-p2996/flang/runtime/internal-unit.cpp
Peter Klausler da25f968a9 [flang] Runtime performance improvements to real formatted input
Profiling a basic internal real input read benchmark shows some
hot spots in the code used to prepare input for decimal-to-binary
conversion, which is of course where the time should be spent.
The library that implements decimal to/from binary conversions has
been optimized, but not the code in the Fortran runtime that calls it,
and there are some obvious light changes worth making here.

Move some member functions from *.cpp files into the class definitions
of Descriptor and IoStatementState to enable inlining and specialization.

Make GetNextInputBytes() the new basic input API within the
runtime, replacing GetCurrentChar() -- which is rewritten in terms of
GetNextInputBytes -- so that input routines can have the
ability to acquire more than one input character at a time
and amortize overhead.

These changes speed up the time to read 1M random reals
using internal I/O from a character array from 1.29s to 0.54s
on my machine, which on par with Intel Fortran and much faster than
GNU Fortran.

Differential Revision: https://reviews.llvm.org/D113697
2021-11-12 11:40:02 -08:00

156 lines
5.0 KiB
C++

//===-- runtime/internal-unit.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 "internal-unit.h"
#include "io-error.h"
#include "flang/Runtime/descriptor.h"
#include <algorithm>
#include <type_traits>
namespace Fortran::runtime::io {
template <Direction DIR>
InternalDescriptorUnit<DIR>::InternalDescriptorUnit(
Scalar scalar, std::size_t length) {
isFixedRecordLength = true;
recordLength = length;
endfileRecordNumber = 2;
void *pointer{reinterpret_cast<void *>(const_cast<char *>(scalar))};
descriptor().Establish(TypeCode{CFI_type_char}, length, pointer, 0, nullptr,
CFI_attribute_pointer);
}
template <Direction DIR>
InternalDescriptorUnit<DIR>::InternalDescriptorUnit(
const Descriptor &that, const Terminator &terminator) {
RUNTIME_CHECK(terminator, that.type().IsCharacter());
Descriptor &d{descriptor()};
RUNTIME_CHECK(
terminator, that.SizeInBytes() <= d.SizeInBytes(maxRank, true, 0));
new (&d) Descriptor{that};
d.Check();
isFixedRecordLength = true;
recordLength = d.ElementBytes();
endfileRecordNumber = d.Elements() + 1;
}
template <Direction DIR> void InternalDescriptorUnit<DIR>::EndIoStatement() {
if constexpr (DIR == Direction::Output) { // blank fill
while (char *record{CurrentRecord()}) {
if (furthestPositionInRecord <
recordLength.value_or(furthestPositionInRecord)) {
std::fill_n(record + furthestPositionInRecord,
*recordLength - furthestPositionInRecord, ' ');
}
furthestPositionInRecord = 0;
++currentRecordNumber;
}
}
}
template <Direction DIR>
bool InternalDescriptorUnit<DIR>::Emit(
const char *data, std::size_t bytes, IoErrorHandler &handler) {
if constexpr (DIR == Direction::Input) {
handler.Crash("InternalDescriptorUnit<Direction::Input>::Emit() called");
return false && data[bytes] != 0; // bogus compare silences GCC warning
} else {
if (bytes <= 0) {
return true;
}
char *record{CurrentRecord()};
if (!record) {
handler.SignalError(IostatInternalWriteOverrun);
return false;
}
auto furthestAfter{std::max(furthestPositionInRecord,
positionInRecord + static_cast<std::int64_t>(bytes))};
bool ok{true};
if (furthestAfter > static_cast<std::int64_t>(recordLength.value_or(0))) {
handler.SignalError(IostatRecordWriteOverrun);
furthestAfter = recordLength.value_or(0);
bytes = std::max(std::int64_t{0}, furthestAfter - positionInRecord);
ok = false;
} else if (positionInRecord > furthestPositionInRecord) {
std::fill_n(record + furthestPositionInRecord,
positionInRecord - furthestPositionInRecord, ' ');
}
std::memcpy(record + positionInRecord, data, bytes);
positionInRecord += bytes;
furthestPositionInRecord = furthestAfter;
return ok;
}
}
template <Direction DIR>
std::size_t InternalDescriptorUnit<DIR>::GetNextInputBytes(
const char *&p, IoErrorHandler &handler) {
if constexpr (DIR == Direction::Output) {
handler.Crash("InternalDescriptorUnit<Direction::Output>::"
"GetNextInputBytes() called");
return 0;
} else {
const char *record{CurrentRecord()};
if (!record) {
handler.SignalEnd();
return 0;
} else if (positionInRecord >= recordLength.value_or(positionInRecord)) {
return 0;
} else {
p = &record[positionInRecord];
return *recordLength - positionInRecord;
}
}
}
template <Direction DIR>
std::optional<char32_t> InternalDescriptorUnit<DIR>::GetCurrentChar(
IoErrorHandler &handler) {
const char *p{nullptr};
std::size_t bytes{GetNextInputBytes(p, handler)};
if (bytes == 0) {
return std::nullopt;
} else {
if (isUTF8) {
// TODO: UTF-8 decoding
}
return *p;
}
}
template <Direction DIR>
bool InternalDescriptorUnit<DIR>::AdvanceRecord(IoErrorHandler &handler) {
if (currentRecordNumber >= endfileRecordNumber.value_or(0)) {
handler.SignalEnd();
return false;
}
if constexpr (DIR == Direction::Output) { // blank fill
if (furthestPositionInRecord <
recordLength.value_or(furthestPositionInRecord)) {
char *record{CurrentRecord()};
RUNTIME_CHECK(handler, record != nullptr);
std::fill_n(record + furthestPositionInRecord,
*recordLength - furthestPositionInRecord, ' ');
}
}
++currentRecordNumber;
BeginRecord();
return true;
}
template <Direction DIR>
void InternalDescriptorUnit<DIR>::BackspaceRecord(IoErrorHandler &handler) {
RUNTIME_CHECK(handler, currentRecordNumber > 1);
--currentRecordNumber;
BeginRecord();
}
template class InternalDescriptorUnit<Direction::Output>;
template class InternalDescriptorUnit<Direction::Input>;
} // namespace Fortran::runtime::io