thevar1able/ClickHouse
1
0
Fork 0
mirror of https://github.com/ClickHouse/ClickHouse.git synced 2024-11-23 08:02:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

dbms: slightly improved performance when reading long strings [#METR-2944].

Browse Source
This commit is contained in:
Alexey Milovidov 2015-02-15 18:18:21 +03:00
parent 24a4956671
commit ca4ca58855
1 changed files with 38 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

58
dbms/src/DataTypes/DataTypeString.cpp
View File

@ -72,22 +72,9 @@ void DataTypeString::serializeBinary(const IColumn & column, WriteBuffer & ostr,
}
void DataTypeString::deserializeBinary(IColumn & column, ReadBuffer & istr, size_t limit, double avg_value_size_hint) const
template <int UNROLL_TIMES>
static NO_INLINE void deserializeBinarySSE2(ColumnString::Chars_t & data, ColumnString::Offsets_t & offsets, ReadBuffer & istr, size_t limit)
{
ColumnString & column_string = typeid_cast<ColumnString &>(column);
ColumnString::Chars_t & data = column_string.getChars();
ColumnString::Offsets_t & offsets = column_string.getOffsets();
/// Выбрано наугад.
constexpr auto avg_value_size_hint_reserve_multiplier = 1.2;
data.reserve(data.size()
+ std::ceil(limit * (avg_value_size_hint && avg_value_size_hint > sizeof(offsets[0])
? (avg_value_size_hint - sizeof(offsets[0])) * avg_value_size_hint_reserve_multiplier
: DBMS_APPROX_STRING_SIZE)));
offsets.reserve(offsets.size() + limit);
size_t offset = data.size();
for (size_t i = 0; i < limit; ++i)
{
@ -105,17 +92,20 @@ void DataTypeString::deserializeBinary(IColumn & column, ReadBuffer & istr, size
if (size)
{
/// Оптимистичная ветка, в которой возможно более эффективное копирование.
if (offset + 16 <= data.capacity() && istr.position() + size + 16 <= istr.buffer().end())
if (offset + 16 * UNROLL_TIMES <= data.capacity() && istr.position() + size + 16 * UNROLL_TIMES <= istr.buffer().end())
{
const __m128i * sse_src_pos = reinterpret_cast<const __m128i *>(istr.position());
const __m128i * sse_src_end = sse_src_pos + (size + 15) / 16;
const __m128i * sse_src_end = sse_src_pos + (size + (16 * UNROLL_TIMES - 1)) / (16 * UNROLL_TIMES);
__m128i * sse_dst_pos = reinterpret_cast<__m128i *>(&data[offset - size - 1]);
while (sse_src_pos < sse_src_end)
{
_mm_storeu_si128(sse_dst_pos, _mm_loadu_si128(sse_src_pos));
++sse_src_pos;
++sse_dst_pos;
/// NOTE gcc 4.9.2 разворачивает цикл, но почему-то использует только один xmm регистр.
for (size_t j = 0; j < UNROLL_TIMES; ++j)
_mm_storeu_si128(sse_dst_pos + j, _mm_loadu_si128(sse_src_pos + j));
sse_src_pos += UNROLL_TIMES;
sse_dst_pos += UNROLL_TIMES;
}
istr.position() += size;
@ -131,6 +121,34 @@ void DataTypeString::deserializeBinary(IColumn & column, ReadBuffer & istr, size
}
void DataTypeString::deserializeBinary(IColumn & column, ReadBuffer & istr, size_t limit, double avg_value_size_hint) const
{
ColumnString & column_string = typeid_cast<ColumnString &>(column);
ColumnString::Chars_t & data = column_string.getChars();
ColumnString::Offsets_t & offsets = column_string.getOffsets();
/// Выбрано наугад.
constexpr auto avg_value_size_hint_reserve_multiplier = 1.2;
double avg_chars_size = (avg_value_size_hint && avg_value_size_hint > sizeof(offsets[0])
? (avg_value_size_hint - sizeof(offsets[0])) * avg_value_size_hint_reserve_multiplier
: DBMS_APPROX_STRING_SIZE);
data.reserve(data.size() + std::ceil(limit * avg_chars_size));
offsets.reserve(offsets.size() + limit);
if (avg_chars_size >= 64)
deserializeBinarySSE2<4>(data, offsets, istr, limit);
else if (avg_chars_size >= 48)
deserializeBinarySSE2<3>(data, offsets, istr, limit);
else if (avg_chars_size >= 32)
deserializeBinarySSE2<2>(data, offsets, istr, limit);
else
deserializeBinarySSE2<1>(data, offsets, istr, limit);
}
void DataTypeString::serializeText(const Field & field, WriteBuffer & ostr) const
{
writeString(get<const String &>(field), ostr);