2 * Copyright 2015 Facebook, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #ifndef FOLLY_EXPERIMENTAL_INSTRUCTIONS_H
18 #define FOLLY_EXPERIMENTAL_INSTRUCTIONS_H
20 #include <folly/CpuId.h>
22 namespace folly { namespace compression { namespace instructions {
24 // NOTE: It's recommended to compile EF coding with -msse4.2, starting
25 // with Nehalem, Intel CPUs support POPCNT instruction and gcc will emit
26 // it for __builtin_popcountll intrinsic.
27 // But we provide an alternative way for the client code: it can switch to
28 // the appropriate version of EliasFanoReader<> in realtime (client should
29 // implement this switching logic itself) by specifying instruction set to
33 static bool supported(const folly::CpuId& cpuId = {}) {
36 static inline uint64_t popcount(uint64_t value) {
37 return __builtin_popcountll(value);
39 static inline int ctz(uint64_t value) {
41 return __builtin_ctzll(value);
43 static inline int clz(uint64_t value) {
45 return __builtin_clzll(value);
47 static inline uint64_t blsr(uint64_t value) {
48 return value & (value - 1);
52 struct Nehalem : public Default {
53 static bool supported(const folly::CpuId& cpuId = {}) {
54 return cpuId.popcnt();
56 static inline uint64_t popcount(uint64_t value) {
57 // POPCNT is supported starting with Intel Nehalem, AMD K10.
59 asm ("popcntq %1, %0" : "=r" (result) : "r" (value));
64 struct Haswell : public Nehalem {
65 static bool supported(const folly::CpuId& cpuId = {}) {
66 return Nehalem::supported(cpuId) && cpuId.bmi1();
68 static inline uint64_t blsr(uint64_t value) {
69 // BMI1 is supported starting with Intel Haswell, AMD Piledriver.
70 // BLSR combines two instuctions into one and reduces register pressure.
72 asm ("blsrq %1, %0" : "=r" (result) : "r" (value));
79 #endif // FOLLY_EXPERIMENTAL_INSTRUCTIONS_H