123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310 |
- // Copyright 2010 Google Inc. All rights reserved.
- //
- // Licensed under the Apache License, Version 2.0 (the "License");
- // you may not use this file except in compliance with the License.
- // You may obtain a copy of the License at
- //
- // http://www.apache.org/licenses/LICENSE-2.0
- //
- // Unless required by applicable law or agreed to in writing, software
- // distributed under the License is distributed on an "AS IS" BASIS,
- // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- // See the License for the specific language governing permissions and
- // limitations under the License.
- // Implements a limited set of 128-bit arithmetic operations
- // (the ones that are used by CRC) using SSE2 intrinsics.
- #ifndef CRCUTIL_UINT128_SSE2_H_
- #define CRCUTIL_UINT128_SSE2_H_
- #include "base_types.h"
- #include "crc_casts.h" // Downcast, CrcFromUint64, Uint64FromCrc
- #include "platform.h"
- #if HAVE_SSE2
- namespace crcutil {
- // Specialized functions handling __m128i.
- template<> __forceinline uint64 Downcast(const __m128i &value) {
- #if HAVE_AMD64 && defined(__GNUC__)
- // GCC 4.4.x is too smart and, instead of MOVQ, generates SSE4 PEXTRQ
- // instruction when the code is compiled with -mmsse4.
- // Fixed in 4.5 which generates conversion through memory (why?).
- // And -- yes, it makes quite measurable difference.
- uint64 temp;
- asm(SSE2_MOVQ " %[i128], %[u64]\n" : [u64] "=r" (temp) : [i128] "x" (value));
- return temp;
- #elif HAVE_AMD64 && (!defined(_MSC_FULL_VER) || _MSC_FULL_VER > 150030729)
- return static_cast<uint64>(_mm_cvtsi128_si64(value));
- #else
- // 64-bit CL 15.00.30729.1 -O2 generates incorrect code (tests fail).
- // _mm_cvtsi128_si64() is not available on i386.
- uint64 temp;
- _mm_storel_epi64(reinterpret_cast<__m128i *>(&temp), value);
- return temp;
- #endif
- }
- class uint128_sse2 {
- public:
- uint128_sse2() {}
- ~uint128_sse2() {}
- // Default casts to uint128_sse2 and assignment operator.
- __forceinline void operator =(uint64 value) {
- #if HAVE_AMD64 && defined(__GNUC__) && !GCC_VERSION_AVAILABLE(4, 5)
- // Prevent generation of SSE4 pinsrq insruction when
- // compiling with GCC 4.4.x with -msse4 flag.
- asm(SSE2_MOVQ " %[u64], %[i128]\n" : [i128] "=x" (x_) : [u64] "r" (value));
- #elif HAVE_AMD64
- x_ = _mm_cvtsi64_si128(static_cast<int64>(value));
- #else
- x_ = _mm_loadl_epi64(reinterpret_cast<const __m128i *>(&value));
- #endif
- }
- __forceinline uint128_sse2(uint64 x) {
- *this = x;
- }
- __forceinline uint128_sse2(const __m128i x) : x_(x) {
- }
- __forceinline operator __m128i() const {
- return x_;
- }
- __forceinline void operator =(const uint128_sse2 &x) {
- x_ = x.x_;
- }
- // Extracts 64 less significant bits.
- __forceinline uint64 to_uint64() const {
- return Downcast<__m128i, uint64>(x_);
- }
- // Comparisons.
- __forceinline bool operator ==(const uint128_sse2 &y) const {
- union {
- __m128i i128;
- uint64 u64[2];
- } t;
- t.i128 = _mm_xor_si128(x_, y.x_);
- return (t.u64[0] | t.u64[1]) == 0;
- }
- __forceinline bool operator ==(uint64 value) const {
- union {
- __m128i i128;
- uint64 u64[2];
- } t;
- t.i128 = x_;
- return (t.u64[0] == value && t.u64[1] == 0);
- }
- __forceinline bool operator !=(const uint128_sse2 &y) const {
- union {
- __m128i i128;
- uint64 u64[2];
- } t;
- t.i128 = _mm_xor_si128(x_, y.x_);
- return (t.u64[0] | t.u64[1]) != 0;
- }
- __forceinline bool operator !=(uint64 value) const {
- union {
- __m128i i128;
- uint64 u64[2];
- } t;
- t.i128 = x_;
- return (t.u64[0] != value || t.u64[1] != 0);
- }
- __forceinline bool operator <(const uint128_sse2 &y) const {
- union {
- __m128i i128;
- uint64 u64[2];
- } xx, yy;
- xx.i128 = x_;
- yy.i128 = y.x_;
- return (xx.u64[0] < yy.u64[0] ||
- (xx.u64[0] == yy.u64[0] && xx.u64[1] < yy.u64[1]));
- }
- // Bitwise logic operators.
- __forceinline uint128_sse2 operator ^(const uint128_sse2 &y) const {
- return _mm_xor_si128(x_, y.x_);
- }
- __forceinline uint128_sse2 operator &(const uint128_sse2 &y) const {
- return _mm_and_si128(x_, y.x_);
- }
- __forceinline uint128_sse2 operator |(const uint128_sse2 &y) const {
- return _mm_or_si128(x_, y.x_);
- }
- __forceinline void operator ^=(const uint128_sse2 &y) {
- *this = *this ^ y.x_;
- }
- __forceinline void operator &=(const uint128_sse2 &y) {
- *this = *this & y.x_;
- }
- __forceinline void operator |=(const uint128_sse2 &y) {
- *this = *this | y.x_;
- }
- // Arithmetic operators.
- __forceinline uint128_sse2 operator +(uint64 y) const {
- union {
- __m128i i128;
- uint64 u64[2];
- } temp;
- temp.i128 = x_;
- // a + b >= 2**64 iff
- // a + b > (2**64 - 1) iff
- // a > (2**64 - 1) - b iff
- // a > ~b
- if (temp.u64[0] > ~y) {
- temp.u64[1] += 1;
- }
- temp.u64[0] += y;
- return temp.i128;
- }
- __forceinline void operator +=(uint64 x) {
- *this = *this + x;
- }
- __forceinline uint128_sse2 operator -(uint64 y) const {
- union {
- __m128i i128;
- uint64 u64[2];
- } temp;
- temp.i128 = x_;
- if (temp.u64[0] < y) {
- temp.u64[1] -= 1;
- }
- temp.u64[0] -= y;
- return temp.i128;
- }
- __forceinline void operator -=(uint64 x) {
- *this = *this - x;
- }
- // Bitwise logical shifts.
- __forceinline uint128_sse2 operator >>(const int bits) const {
- if (bits == 8) {
- return _mm_srli_si128(x_, 1);
- } else if (bits == 16) {
- return _mm_srli_si128(x_, 2);
- } else if (bits == 32) {
- return _mm_srli_si128(x_, 4);
- } else if (bits == 64) {
- return _mm_srli_si128(x_, 8);
- } else {
- return long_shift_right(bits);
- }
- }
- __forceinline uint128_sse2 operator >>(const size_t bits) const {
- return *this >> static_cast<int>(bits);
- }
- __forceinline void operator >>=(const int bits) {
- *this = *this >> bits;
- }
- __forceinline void operator >>=(const size_t bits) {
- *this = *this >> static_cast<int>(bits);
- }
- __forceinline uint128_sse2 operator <<(int bits) const {
- if (bits == 8) {
- return _mm_slli_si128(x_, 1);
- } else if (bits == 16) {
- return _mm_slli_si128(x_, 2);
- } else if (bits == 32) {
- return _mm_slli_si128(x_, 4);
- } else if (bits == 64) {
- return _mm_slli_si128(x_, 8);
- } else {
- return long_shift_left(bits);
- }
- }
- __forceinline uint128_sse2 operator <<(size_t bits) const {
- return *this << static_cast<int>(bits);
- }
- __forceinline void operator <<=(int bits) {
- *this = *this << bits;
- }
- __forceinline void operator <<=(size_t bits) {
- *this = *this << static_cast<int>(bits);
- }
- protected:
- __forceinline uint128_sse2 long_shift_right(int bits) const {
- union {
- __m128i i128;
- uint64 u64[2];
- } x;
- x.i128 = x_;
- for (; bits > 0; --bits) {
- x.u64[0] >>= 1;
- if (x.u64[1] & 1) {
- x.u64[0] |= static_cast<uint64>(1) << 63;
- }
- x.u64[1] >>= 1;
- }
- return x.i128;
- }
- __forceinline uint128_sse2 long_shift_left(int bits) const {
- union {
- __m128i i128;
- int64 i64[2];
- } x;
- x.i128 = x_;
- for (; bits > 0; --bits) {
- x.i64[1] <<= 1;
- if (x.i64[0] < 0) {
- x.i64[1] |= 1;
- }
- x.i64[0] <<= 1;
- }
- return x.i128;
- }
- __m128i x_;
- } GCC_ALIGN_ATTRIBUTE(16);
- // Specialized versions.
- template<> __forceinline uint64 Downcast(const uint128_sse2 &x) {
- return x.to_uint64();
- }
- template<> __forceinline uint32 Downcast(const uint128_sse2 &x) {
- return static_cast<uint32>(x.to_uint64());
- }
- template<> __forceinline uint16 Downcast(const uint128_sse2 &x) {
- return static_cast<uint16>(x.to_uint64());
- }
- template<> __forceinline uint8 Downcast(const uint128_sse2 &x) {
- return static_cast<uint8>(x.to_uint64());
- }
- template<> __forceinline uint128_sse2 CrcFromUint64(uint64 lo, uint64 hi) {
- union {
- __m128i i128;
- uint64 u64[2];
- } temp;
- temp.u64[0] = lo;
- temp.u64[1] = hi;
- return temp.i128;
- }
- template<> __forceinline void Uint64FromCrc(const uint128_sse2 &crc,
- uint64 *lo, uint64 *hi) {
- union {
- __m128i i128;
- uint64 u64[2];
- } temp;
- temp.i128 = crc;
- *lo = temp.u64[0];
- *hi = temp.u64[1];
- }
- } // namespace crcutil
- #endif // HAVE_SSE2
- #endif // CRCUTIL_UINT128_SSE2_H_
|