diff options
author | Wladimir J. van der Laan <laanwj@gmail.com> | 2017-06-13 19:38:37 +0200 |
---|---|---|
committer | Wladimir J. van der Laan <laanwj@gmail.com> | 2017-06-13 19:48:13 +0200 |
commit | a4fe07714da104205a1376343d67b7f576f08d26 (patch) | |
tree | e0dfa9a78fb5cb1956d7ef9d48c26f1449990669 /src/leveldb/port/port_posix_sse.cc | |
parent | 22ec7688388694289cf5ffdc39c69ee3b12c8a23 (diff) | |
parent | 3ee3d04374894ad56ac00b654aef70d70fe6466e (diff) | |
download | bitcoin-a4fe07714da104205a1376343d67b7f576f08d26.tar.xz |
Merge #10544: Update to LevelDB 1.20
3ee3d04 Add extra LevelDB source to Makefile (MarcoFalke)
2424989 leveldb: enable runtime-detected crc32 instructions (Cory Fields)
cf44e4c Squashed 'src/leveldb/' changes from a31c8aa40..196962ff0 (Pieter Wuille)
Tree-SHA512: 19ade77e3f6265507b3ab7b9aa5150d378aa0751e24ac7a61567b0f720a566cedc6c3d3336da17a3bd2b5d068ee86600d96a15228f78bd20ccf98c8fc9041a91
Diffstat (limited to 'src/leveldb/port/port_posix_sse.cc')
-rw-r--r-- | src/leveldb/port/port_posix_sse.cc | 129 |
1 files changed, 129 insertions, 0 deletions
diff --git a/src/leveldb/port/port_posix_sse.cc b/src/leveldb/port/port_posix_sse.cc new file mode 100644 index 0000000000..1e519ba0b6 --- /dev/null +++ b/src/leveldb/port/port_posix_sse.cc @@ -0,0 +1,129 @@ +// Copyright 2016 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. +// +// A portable implementation of crc32c, optimized to handle +// four bytes at a time. +// +// In a separate source file to allow this accelerated CRC32C function to be +// compiled with the appropriate compiler flags to enable x86 SSE 4.2 +// instructions. + +#include <stdint.h> +#include <string.h> +#include "port/port.h" + +#if defined(LEVELDB_PLATFORM_POSIX_SSE) + +#if defined(_MSC_VER) +#include <intrin.h> +#elif defined(__GNUC__) && defined(__SSE4_2__) +#include <nmmintrin.h> +#include <cpuid.h> +#endif + +#endif // defined(LEVELDB_PLATFORM_POSIX_SSE) + +namespace leveldb { +namespace port { + +#if defined(LEVELDB_PLATFORM_POSIX_SSE) + +// Used to fetch a naturally-aligned 32-bit word in little endian byte-order +static inline uint32_t LE_LOAD32(const uint8_t *p) { + // SSE is x86 only, so ensured that |p| is always little-endian. + uint32_t word; + memcpy(&word, p, sizeof(word)); + return word; +} + +#if defined(_M_X64) || defined(__x86_64__) // LE_LOAD64 is only used on x64. + +// Used to fetch a naturally-aligned 64-bit word in little endian byte-order +static inline uint64_t LE_LOAD64(const uint8_t *p) { + uint64_t dword; + memcpy(&dword, p, sizeof(dword)); + return dword; +} + +#endif // defined(_M_X64) || defined(__x86_64__) + +static inline bool HaveSSE42() { +#if defined(_MSC_VER) + int cpu_info[4]; + __cpuid(cpu_info, 1); + return (cpu_info[2] & (1 << 20)) != 0; +#elif defined(__GNUC__) + unsigned int eax, ebx, ecx, edx; + __get_cpuid(1, &eax, &ebx, &ecx, &edx); + return (ecx & (1 << 20)) != 0; +#else + return false; +#endif +} + +#endif // defined(LEVELDB_PLATFORM_POSIX_SSE) + +// For further improvements see Intel publication at: +// http://download.intel.com/design/intarch/papers/323405.pdf +uint32_t AcceleratedCRC32C(uint32_t crc, const char* buf, size_t size) { +#if !defined(LEVELDB_PLATFORM_POSIX_SSE) + return 0; +#else + static bool have = HaveSSE42(); + if (!have) { + return 0; + } + + const uint8_t *p = reinterpret_cast<const uint8_t *>(buf); + const uint8_t *e = p + size; + uint32_t l = crc ^ 0xffffffffu; + +#define STEP1 do { \ + l = _mm_crc32_u8(l, *p++); \ +} while (0) +#define STEP4 do { \ + l = _mm_crc32_u32(l, LE_LOAD32(p)); \ + p += 4; \ +} while (0) +#define STEP8 do { \ + l = _mm_crc32_u64(l, LE_LOAD64(p)); \ + p += 8; \ +} while (0) + + if (size > 16) { + // Process unaligned bytes + for (unsigned int i = reinterpret_cast<uintptr_t>(p) % 8; i; --i) { + STEP1; + } + + // _mm_crc32_u64 is only available on x64. +#if defined(_M_X64) || defined(__x86_64__) + // Process 8 bytes at a time + while ((e-p) >= 8) { + STEP8; + } + // Process 4 bytes at a time + if ((e-p) >= 4) { + STEP4; + } +#else // !(defined(_M_X64) || defined(__x86_64__)) + // Process 4 bytes at a time + while ((e-p) >= 4) { + STEP4; + } +#endif // defined(_M_X64) || defined(__x86_64__) + } + // Process the last few bytes + while (p != e) { + STEP1; + } +#undef STEP8 +#undef STEP4 +#undef STEP1 + return l ^ 0xffffffffu; +#endif // defined(LEVELDB_PLATFORM_POSIX_SSE) +} + +} // namespace port +} // namespace leveldb |