// Licensed to the Apache Software Foundation (ASF) under one

// or more contributor license agreements.  See the NOTICE file

// distributed with this work for additional information

// regarding copyright ownership.  The ASF licenses this file

// to you 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.

// This file is copied from

// https://github.com/apache/impala/blob/branch-2.9.0/be/src/util/cpu-info.cpp

// and modified by Doris

#include "util/cpu_info.h"

#if defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))

#elif defined(__GNUC__) && defined(__ARM_NEON__)

/* GCC-compatible compiler, targeting ARM with NEON */

#include <arm_neon.h>

#elif defined(__GNUC__) && defined(__IWMMXT__)

/* GCC-compatible compiler, targeting ARM with WMMX */

#include <mmintrin.h>

#elif (defined(__GNUC__) || defined(__xlC__)) && (defined(__VEC__) || defined(__ALTIVEC__))

/* XLC or GCC-compatible compiler, targeting PowerPC with VMX/VSX */

#include <altivec.h>

#elif defined(__GNUC__) && defined(__SPE__)

/* GCC-compatible compiler, targeting PowerPC with SPE */

#include <spe.h>

#endif

#ifndef __APPLE__

#include <sys/sysinfo.h>

#else

#include <sys/sysctl.h>

#endif

#include <gen_cpp/Metrics_types.h>

#include <sched.h>

#include <stdlib.h>

#include <unistd.h>

#include <algorithm>

#include <boost/algorithm/string/predicate.hpp>

#include <boost/algorithm/string/trim.hpp>

// IWYU pragma: no_include <bits/chrono.h>

#include <chrono> // IWYU pragma: keep

#include <filesystem>

#include <fstream>

#include "common/config.h"

#include "common/env_config.h"

#include "gflags/gflags.h"

#include "gutil/stringprintf.h"

#include "gutil/strings/substitute.h"

#include "util/cgroup_util.h"

#include "util/pretty_printer.h"

using boost::algorithm::contains;

using boost::algorithm::trim;

namespace fs = std::filesystem;

using std::max;

DECLARE_bool(abort_on_config_error);

DEFINE_int32(num_cores, 0,

             "(Advanced) If > 0, it sets the number of cores available to"

             " Impala. Setting it to 0 means Impala will use all available cores on the machine"

             " according to /proc/cpuinfo.");

namespace doris {

// Helper function to warn if a given file does not contain an expected string as its

// first line. If the file cannot be opened, no error is reported.

void WarnIfFileNotEqual(const string& filename, const string& expected,

                        const string& warning_text) {

    std::ifstream file(filename);

    if (!file) return;

  Branch (82:9): [True: 0, False: 0]

    string line;

    getline(file, line);

    if (line != expected) {

  Branch (85:9): [True: 0, False: 0]

        LOG(ERROR) << "Expected " << expected << ", actual " << line << std::endl << warning_text;

    }

}

} // namespace doris

namespace doris {

bool CpuInfo::initialized_ = false;

int64_t CpuInfo::hardware_flags_ = 0;

int64_t CpuInfo::original_hardware_flags_;

int64_t CpuInfo::cycles_per_ms_;

int CpuInfo::num_cores_ = 1;

int CpuInfo::max_num_cores_ = 1;

std::string CpuInfo::model_name_ = "unknown";

int CpuInfo::max_num_numa_nodes_;

std::unique_ptr<int[]> CpuInfo::core_to_numa_node_;

std::vector<vector<int>> CpuInfo::numa_node_to_cores_;

std::vector<int> CpuInfo::numa_node_core_idx_;

static struct {

    string name;

    int64_t flag;

} flag_mappings[] = {

        {"ssse3", CpuInfo::SSSE3},   {"sse4_1", CpuInfo::SSE4_1}, {"sse4_2", CpuInfo::SSE4_2},

        {"popcnt", CpuInfo::POPCNT}, {"avx", CpuInfo::AVX},       {"avx2", CpuInfo::AVX2},

};

// Helper function to parse for hardware flags.

// values contains a list of space-separated flags.  check to see if the flags we

// care about are present.

// Returns a bitmap of flags.

int64_t ParseCPUFlags(const string& values) {

    int64_t flags = 0;

    for (auto& flag_mapping : flag_mappings) {

  Branch (119:29): [True: 72, False: 12]

        if (contains(values, flag_mapping.name)) {

  Branch (120:13): [True: 72, False: 0]

            flags |= flag_mapping.flag;

        }

    }

    return flags;

}

void CpuInfo::init() {

    if (initialized_) return;

  Branch (128:9): [True: 9, False: 1]

    string line;

    string name;

    string value;

    float max_mhz = 0;

    int physical_num_cores = 0;

    // maybe use std::thread::hardware_concurrency()?

    // Read from /proc/cpuinfo

    std::ifstream cpuinfo("/proc/cpuinfo");

    while (cpuinfo) {

  Branch (139:12): [True: 325, False: 1]

        getline(cpuinfo, line);

        size_t colon = line.find(':');

        if (colon != string::npos) {

  Branch (142:13): [True: 312, False: 13]

            name = line.substr(0, colon - 1);

            value = line.substr(colon + 1, string::npos);

            trim(name);

            trim(value);

            if (name == "flags") {

  Branch (147:17): [True: 12, False: 300]

                hardware_flags_ |= ParseCPUFlags(value);

            } else if (name == "cpu MHz") {

  Branch (149:24): [True: 12, False: 288]

                // Every core will report a different speed.  We'll take the max, assuming

                // that when impala is running, the core will not be in a lower power state.

                // TODO: is there a more robust way to do this, such as

                // Window's QueryPerformanceFrequency()

                float mhz = atof(value.c_str());

                max_mhz = max(mhz, max_mhz);

            } else if (name == "processor") {

  Branch (156:24): [True: 12, False: 276]

                ++physical_num_cores;

            } else if (name == "model name") {

  Branch (158:24): [True: 12, False: 264]

                model_name_ = value;

            }

        }

    }

#ifdef __APPLE__

    size_t len = sizeof(max_num_cores_);

    sysctlbyname("hw.physicalcpu", &physical_num_cores, &len, nullptr, 0);

#endif

    int num_cores = CGroupUtil::get_cgroup_limited_cpu_number(physical_num_cores);

    if (max_mhz != 0) {

  Branch (170:9): [True: 1, False: 0]

        cycles_per_ms_ = int64_t(max_mhz) * 1000;

    } else {

        cycles_per_ms_ = 1000000;

    }

    original_hardware_flags_ = hardware_flags_;

    if (num_cores > 0) {

  Branch (177:9): [True: 1, False: 0]

        num_cores_ = num_cores;

    } else {

        num_cores_ = 1;

    }

    if (config::num_cores > 0) {

  Branch (182:9): [True: 0, False: 1]

        num_cores_ = config::num_cores;

    }

#ifdef __APPLE__

    sysctlbyname("hw.logicalcpu", &max_num_cores_, &len, nullptr, 0);

#else

    max_num_cores_ = get_nprocs_conf();

#endif

    // Print a warning if something is wrong with sched_getcpu().

#ifdef HAVE_SCHED_GETCPU

    if (sched_getcpu() == -1) {

  Branch (194:9): [True: 0, False: 1]

        LOG(WARNING) << "Kernel does not support sched_getcpu(). Performance may be impacted.";

    }

#else

    LOG(WARNING) << "Built on a system without sched_getcpu() support. Performance may"

                 << " be impacted.";

#endif

    _init_numa();

    initialized_ = true;

}

void CpuInfo::_init_numa() {

    // Use the NUMA info in the /sys filesystem. which is part of the Linux ABI:

    // see https://www.kernel.org/doc/Documentation/ABI/stable/sysfs-devices-node and

    // https://www.kernel.org/doc/Documentation/ABI/testing/sysfs-devices-system-cpu

    // The filesystem entries are only present if the kernel was compiled with NUMA support.

    core_to_numa_node_.reset(new int[max_num_cores_]);

    if (!fs::is_directory("/sys/devices/system/node")) {

  Branch (213:9): [True: 0, False: 1]

        LOG(WARNING) << "/sys/devices/system/node is not present - no NUMA support";

        // Assume a single NUMA node.

        max_num_numa_nodes_ = 1;

        std::fill_n(core_to_numa_node_.get(), max_num_cores_, 0);

        _init_numa_node_to_cores();

        return;

    }

    // Search for node subdirectories - node0, node1, node2, etc to determine possible

    // NUMA nodes.

    fs::directory_iterator dir_it("/sys/devices/system/node");

    max_num_numa_nodes_ = 0;

    for (; dir_it != fs::directory_iterator(); ++dir_it) {

  Branch (226:12): [True: 8, False: 1]

        const string filename = dir_it->path().filename().string();

        if (filename.find("node") == 0) ++max_num_numa_nodes_;

  Branch (228:13): [True: 1, False: 7]

    }

    if (max_num_numa_nodes_ == 0) {

  Branch (230:9): [True: 0, False: 1]

        LOG(WARNING) << "Could not find nodes in /sys/devices/system/node";

        max_num_numa_nodes_ = 1;

    }

    // Check which NUMA node each core belongs to based on the existence of a symlink

    // to the node subdirectory.

    for (int core = 0; core < max_num_cores_; ++core) {

  Branch (237:24): [True: 12, False: 1]

        bool found_numa_node = false;

        for (int node = 0; node < max_num_numa_nodes_; ++node) {

  Branch (239:28): [True: 12, False: 0]

            if (fs::exists(

  Branch (240:17): [True: 12, False: 0]

                        strings::Substitute("/sys/devices/system/cpu/cpu$0/node$1", core, node))) {

                core_to_numa_node_[core] = node;

                found_numa_node = true;

                break;

            }

        }

        if (!found_numa_node) {

  Branch (247:13): [True: 0, False: 12]

            LOG(WARNING) << "Could not determine NUMA node for core " << core

                         << " from /sys/devices/system/cpu/";

            core_to_numa_node_[core] = 0;

        }

    }

    _init_numa_node_to_cores();

}

void CpuInfo::_init_fake_numa_for_test(int max_num_numa_nodes,

                                       const std::vector<int>& core_to_numa_node) {

    DCHECK_EQ(max_num_cores_, core_to_numa_node.size());

    max_num_numa_nodes_ = max_num_numa_nodes;

    for (int i = 0; i < max_num_cores_; ++i) {

  Branch (260:21): [True: 0, False: 0]

        core_to_numa_node_[i] = core_to_numa_node[i];

    }

    numa_node_to_cores_.clear();

    _init_numa_node_to_cores();

}

void CpuInfo::_init_numa_node_to_cores() {

    DCHECK(numa_node_to_cores_.empty());

    numa_node_to_cores_.resize(max_num_numa_nodes_);

    numa_node_core_idx_.resize(max_num_cores_);

    for (int core = 0; core < max_num_cores_; ++core) {

  Branch (271:24): [True: 12, False: 1]

        std::vector<int>* cores_of_node = &numa_node_to_cores_[core_to_numa_node_[core]];

        numa_node_core_idx_[core] = cores_of_node->size();

        cores_of_node->push_back(core);

    }

}

void CpuInfo::verify_cpu_requirements() {

    if (!CpuInfo::is_supported(CpuInfo::SSSE3)) {

  Branch (279:9): [True: 0, False: 0]

        LOG(ERROR) << "CPU does not support the Supplemental SSE3 (SSSE3) instruction set. "

                   << "This setup is generally unsupported and Impala might be unstable.";

    }

}

void CpuInfo::verify_performance_governor() {

    for (int cpu_id = 0; cpu_id < CpuInfo::num_cores(); ++cpu_id) {

  Branch (286:26): [True: 0, False: 0]

        const string governor_file = strings::Substitute(

                "/sys/devices/system/cpu/cpu$0/cpufreq/scaling_governor", cpu_id);

        const string warning_text = strings::Substitute(

                "WARNING: CPU $0 is not using 'performance' governor. Note that changing the "

                "governor to 'performance' will reset the no_turbo setting to 0.",

                cpu_id);

        WarnIfFileNotEqual(governor_file, "performance", warning_text);

    }

}

void CpuInfo::verify_turbo_disabled() {

    WarnIfFileNotEqual(

            "/sys/devices/system/cpu/intel_pstate/no_turbo", "1",

            "WARNING: CPU turbo is enabled. This setting can change the clock frequency of CPU "

            "cores during the benchmark run, which can lead to inaccurate results. You can "

            "disable CPU turbo by writing a 1 to "

            "/sys/devices/system/cpu/intel_pstate/no_turbo. Note that changing the governor to "

            "'performance' will reset this to 0.");

}

void CpuInfo::enable_feature(long flag, bool enable) {

    DCHECK(initialized_);

    if (!enable) {

  Branch (309:9): [True: 0, False: 0]

        hardware_flags_ &= ~flag;

    } else {

        // Can't turn something on that can't be supported

        DCHECK((original_hardware_flags_ & flag) != 0);

        hardware_flags_ |= flag;

    }

}

int CpuInfo::get_current_core() {

    // sched_getcpu() is not supported on some old kernels/glibcs (like the versions that

    // shipped with CentOS 5). In that case just pretend we're always running on CPU 0

    // so that we can build and run with degraded perf.

#ifdef HAVE_SCHED_GETCPU

    int cpu = sched_getcpu();

    if (cpu < 0) return 0;

  Branch (324:9): [True: 0, False: 0]

    if (cpu >= max_num_cores_) {

  Branch (325:9): [True: 0, False: 0]

        LOG_FIRST_N(WARNING, 5) << "sched_getcpu() return value " << cpu

                                << ", which is greater than get_nprocs_conf() retrun value "

                                << max_num_cores_ << ", now is " << get_nprocs_conf();

        cpu %= max_num_cores_;

    }

    return cpu;

#else

    return 0;

#endif

}

void CpuInfo::_get_cache_info(long cache_sizes[NUM_CACHE_LEVELS],

                              long cache_line_sizes[NUM_CACHE_LEVELS]) {

#ifdef __APPLE__

    // On Mac OS X use sysctl() to get the cache sizes

    size_t len = 0;

    sysctlbyname("hw.cachesize", nullptr, &len, nullptr, 0);

    uint64_t* data = static_cast<uint64_t*>(malloc(len));

    sysctlbyname("hw.cachesize", data, &len, nullptr, 0);

#ifndef __arm64__

    DCHECK(len / sizeof(uint64_t) >= 3);

    for (size_t i = 0; i < NUM_CACHE_LEVELS; ++i) {

        cache_sizes[i] = data[i];

    }

#else

    for (size_t i = 0; i < NUM_CACHE_LEVELS; ++i) {

        cache_sizes[i] = data[i + 1];

    }

#endif

    size_t linesize;

    size_t sizeof_linesize = sizeof(linesize);

    sysctlbyname("hw.cachelinesize", &linesize, &sizeof_linesize, nullptr, 0);

    for (size_t i = 0; i < NUM_CACHE_LEVELS; ++i) cache_line_sizes[i] = linesize;

#else

    // Call sysconf to query for the cache sizes

    // Note: on some systems (e.g. RHEL 5 on AWS EC2), this returns 0 instead of the

    // actual cache line size.

    cache_sizes[L1_CACHE] = sysconf(_SC_LEVEL1_DCACHE_SIZE);

    cache_sizes[L2_CACHE] = sysconf(_SC_LEVEL2_CACHE_SIZE);

    cache_sizes[L3_CACHE] = sysconf(_SC_LEVEL3_CACHE_SIZE);

    cache_line_sizes[L1_CACHE] = sysconf(_SC_LEVEL1_DCACHE_LINESIZE);

    cache_line_sizes[L2_CACHE] = sysconf(_SC_LEVEL2_CACHE_LINESIZE);

    cache_line_sizes[L3_CACHE] = sysconf(_SC_LEVEL3_CACHE_LINESIZE);

#endif

}

std::string CpuInfo::debug_string() {

    DCHECK(initialized_);

    std::stringstream stream;

    long cache_sizes[NUM_CACHE_LEVELS];

    long cache_line_sizes[NUM_CACHE_LEVELS];

    _get_cache_info(cache_sizes, cache_line_sizes);

    string L1 = strings::Substitute(

            "L1 Cache: $0 (Line: $1)",

            PrettyPrinter::print(static_cast<int64_t>(cache_sizes[L1_CACHE]), TUnit::BYTES),

            PrettyPrinter::print(static_cast<int64_t>(cache_line_sizes[L1_CACHE]), TUnit::BYTES));

    string L2 = strings::Substitute(

            "L2 Cache: $0 (Line: $1)",

            PrettyPrinter::print(static_cast<int64_t>(cache_sizes[L2_CACHE]), TUnit::BYTES),

            PrettyPrinter::print(static_cast<int64_t>(cache_line_sizes[L2_CACHE]), TUnit::BYTES));

    string L3 =

            cache_sizes[L3_CACHE]

  Branch (389:13): [True: 0, False: 0]

                    ? strings::Substitute(

                              "L3 Cache: $0 (Line: $1)",

                              PrettyPrinter::print(static_cast<int64_t>(cache_sizes[L3_CACHE]),

                                                   TUnit::BYTES),

                              PrettyPrinter::print(static_cast<int64_t>(cache_line_sizes[L3_CACHE]),

                                                   TUnit::BYTES))

                    : "";

    stream << "Cpu Info:" << std::endl

           << "  Model: " << model_name_ << std::endl

           << "  Cores: " << num_cores_ << std::endl

           << "  Max Possible Cores: " << max_num_cores_ << std::endl

           << "  " << L1 << std::endl

           << "  " << L2 << std::endl

           << "  " << L3 << std::endl

           << "  Hardware Supports:" << std::endl;

    for (auto& flag_mapping : flag_mappings) {

  Branch (405:29): [True: 0, False: 0]

        if (is_supported(flag_mapping.flag)) {

  Branch (406:13): [True: 0, False: 0]

            stream << "    " << flag_mapping.name << std::endl;

        }

    }

    stream << "  Numa Nodes: " << max_num_numa_nodes_ << std::endl;

    stream << "  Numa Nodes of Cores:";

    for (int core = 0; core < max_num_cores_; ++core) {

  Branch (412:24): [True: 0, False: 0]

        stream << " " << core << "->" << core_to_numa_node_[core] << " |";

    }

    stream << std::endl;

    return stream.str();

}

} // namespace doris