thevar1able/ClickHouse
1
0
Fork 0
mirror of https://github.com/ClickHouse/ClickHouse.git synced 2024-11-22 23:52:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

T64 codec optimisation: unroll cycles

Browse Source
This commit is contained in:
chertus 2019-06-12 16:03:51 +03:00
parent f9abc8194a
commit 08983b0669
1 changed files with 58 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

78
dbms/src/Compression/CompressionCodecT64.cpp
View File

@ -126,21 +126,71 @@ void revTranspose64x8(const UInt64 * src, UInt64 * dst, UInt32)
}
}
template <typename _T>
void transposeBytes(_T value, UInt64 * mx, UInt32 col)
{
UInt8 * mx8 = reinterpret_cast<UInt8 *>(mx);
const UInt8 * value8 = reinterpret_cast<const UInt8 *>(&value);
if constexpr (sizeof(_T) > 4)
{
mx8[64 * 7 + col] = value8[7];
mx8[64 * 6 + col] = value8[6];
mx8[64 * 5 + col] = value8[5];
mx8[64 * 4 + col] = value8[4];
}
if constexpr (sizeof(_T) > 2)
{
mx8[64 * 3 + col] = value8[3];
mx8[64 * 2 + col] = value8[2];
}
if constexpr (sizeof(_T) > 1)
mx8[64 * 1 + col] = value8[1];
mx8[64 * 0 + col] = value8[0];
}
template <typename _T>
void revTransposeBytes(const UInt64 * mx, UInt32 col, _T & value)
{
auto * mx8 = reinterpret_cast<const UInt8 *>(mx);
if constexpr (sizeof(_T) > 4)
{
value |= UInt64(mx8[64 * 7 + col]) << (8 * 7);
value |= UInt64(mx8[64 * 6 + col]) << (8 * 6);
value |= UInt64(mx8[64 * 5 + col]) << (8 * 5);
value |= UInt64(mx8[64 * 4 + col]) << (8 * 4);
}
if constexpr (sizeof(_T) > 2)
{
value |= UInt32(mx8[64 * 3 + col]) << (8 * 3);
value |= UInt32(mx8[64 * 2 + col]) << (8 * 2);
}
if constexpr (sizeof(_T) > 1)
value |= UInt32(mx8[64 * 1 + col]) << (8 * 1);
value |= UInt32(mx8[col]);
}
/// UIntX[64] -> UInt64[N] transposed matrix, N <= X
template <typename _T>
void transpose(const char * src, char * dst, UInt32 num_bits, UInt32 tail = 64)
{
UInt32 full_bytes = num_bits / 8;
UInt32 part_bits = num_bits % 8;
UInt32 meaning_bytes = full_bytes + (part_bits ? 1 : 0);
UInt64 mx[64] = {};
for (UInt32 row = 0; row < meaning_bytes; ++row)
const char * ptr = src;
for (UInt32 col = 0; col < tail; ++col, ptr += sizeof(_T))
{
UInt8 * same_byte_data = reinterpret_cast<UInt8 *>(&mx[row * 8]);
const char * byte = src + row;
for (UInt32 col = 0; col < tail; ++col, byte += sizeof(_T))
same_byte_data[col] = unalignedLoad<UInt8>(byte);
_T value = unalignedLoad<_T>(ptr);
transposeBytes(value, mx, col);
}
UInt32 full_size = sizeof(UInt64) * (num_bits - part_bits);
@ -157,17 +207,6 @@ void transpose(const char * src, char * dst, UInt32 num_bits, UInt32 tail = 64)
}
}
template <typename _T>
void revTransposeBytes(const UInt64 * mx, UInt32 col, UInt32 meaning_bytes, _T & value)
{
using UnsignedT = std::conditional_t<(sizeof(_T) <= 4), UInt32, UInt64>;
auto * mx8 = reinterpret_cast<const UInt8 *>(mx);
for (UInt32 row = 0; row < meaning_bytes; ++row)
value |= UnsignedT(mx8[64 * row + col]) << (8 * row);
}
/// UInt64[N] transposed matrix -> UIntX[64]
template <typename _T, typename _MinMaxT = std::conditional_t<std::is_signed_v<_T>, Int64, UInt64>>
void revTranspose(const char * src, char * dst, UInt32 num_bits, _MinMaxT min, _MinMaxT max [[maybe_unused]], UInt32 tail = 64)
@ -177,7 +216,6 @@ void revTranspose(const char * src, char * dst, UInt32 num_bits, _MinMaxT min, _
UInt32 full_bytes = num_bits / 8;
UInt32 part_bits = num_bits % 8;
UInt32 meaning_bytes = full_bytes + (part_bits ? 1 : 0);
UInt64 * partial = &mx[full_bytes * 8];
if (part_bits)
@ -204,7 +242,7 @@ void revTranspose(const char * src, char * dst, UInt32 num_bits, _MinMaxT min, _
for (UInt32 col = 0; col < tail; ++col)
{
_T & value = buf[col];
revTransposeBytes(mx, col, meaning_bytes, value);
revTransposeBytes(mx, col, value);
if (value & sign_bit)
value |= upper_min;
@ -220,7 +258,7 @@ void revTranspose(const char * src, char * dst, UInt32 num_bits, _MinMaxT min, _
for (UInt32 col = 0; col < tail; ++col)
{
_T & value = buf[col];
revTransposeBytes(mx, col, meaning_bytes, value);
revTransposeBytes(mx, col, value);
value |= upper_min;
}