thevar1able/ClickHouse
1
0
Fork 0
mirror of https://github.com/ClickHouse/ClickHouse.git synced 2024-11-29 11:02:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

lz4 decompress: dynamic dispatch with TargetSpecific

Browse Source
This commit is contained in:
Wangyang Guo 2022-07-15 16:00:09 +08:00
parent 397a4ac888
commit c9e15017ac
1 changed files with 18 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
src/Compression/LZ4_decompress_faster.cpp
View File

@ -4,7 +4,7 @@
#include <iostream>
#include <Core/Defines.h>
#include <Common/Stopwatch.h>
#include <Common/CpuId.h>
#include <Common/TargetSpecific.h>
#include <base/types.h>
#include <base/unaligned.h>
@ -16,7 +16,7 @@
#include <tmmintrin.h>
#endif
#if defined(__AVX512VBMI__)
#if USE_MULTITARGET_CODE
#include <immintrin.h>
#endif
@ -408,9 +408,9 @@ inline void copyOverlap32(UInt8 * op, const UInt8 *& match, const size_t offset)
match += shift4[offset];
}
DECLARE_AVX512VBMI_SPECIFIC_CODE(
inline void copyOverlap32Shuffle(UInt8 * op, const UInt8 *& match, const size_t offset)
{
#if defined(__AVX512VBMI__) && !defined(MEMORY_SANITIZER)
static constexpr UInt8 __attribute__((__aligned__(32))) masks[] =
{
0, 1, 2, 2, 4, 2, 2, 4, 8, 5, 2, 10, 8, 6, 4, 2, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, /* offset=0, shift amount index. */
@ -447,28 +447,25 @@ inline void copyOverlap32Shuffle(UInt8 * op, const UInt8 *& match, const size_t
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 0,
};
if (DB::Cpu::CpuFlagsCache::have_AVX512VBMI)
{
_mm256_storeu_si256(reinterpret_cast<__m256i *>(op),
_mm256_permutexvar_epi8(
_mm256_load_si256(reinterpret_cast<const __m256i *>(masks) + offset),
_mm256_loadu_si256(reinterpret_cast<const __m256i *>(match))));
match += masks[offset];
}
else
{
copyOverlap32(op, match, offset);
}
#else
copyOverlap32(op, match, offset);
#endif
}
) /// DECLARE_AVX512VBMI_SPECIFIC_CODE
template <> void inline copy<32>(UInt8 * dst, const UInt8 * src) { copy32(dst, src); }
template <> void inline wildCopy<32>(UInt8 * dst, const UInt8 * src, UInt8 * dst_end) { wildCopy32(dst, src, dst_end); }
template <> void inline copyOverlap<32, false>(UInt8 * op, const UInt8 *& match, const size_t offset) { copyOverlap32(op, match, offset); }
template <> void inline copyOverlap<32, true>(UInt8 * op, const UInt8 *& match, const size_t offset) { copyOverlap32Shuffle(op, match, offset); }
template <> void inline copyOverlap<32, true>(UInt8 * op, const UInt8 *& match, const size_t offset) {
#if USE_MULTITARGET_CODE
TargetSpecific::AVX512VBMI::copyOverlap32Shuffle(op, match, offset);
#else
copyOverlap32(op, match, offset);
#endif
}
/// See also https://stackoverflow.com/a/30669632
@ -641,9 +638,9 @@ bool decompress(
if (dest_size >= 32768)
{
size_t variant_size = 4;
#if defined(__AVX512VBMI__) && !defined(MEMORY_SANITIZER)
#if USE_MULTITARGET_CODE && !defined(MEMORY_SANITIZER)
/// best_variant == 4 only valid when AVX512VBMI available
if (DB::Cpu::CpuFlagsCache::have_AVX512VBMI)
if (isArchSupported(DB::TargetArch::AVX512VBMI))
variant_size = 5;
#endif
size_t best_variant = statistics.select(variant_size);