AMD EPYC (Zen 4) 向けの C++ ビルド環境を作成して OpenBLAS で行列計算を試してみた

sudo dnf update -y

# Development Tools グループのインストール
sudo dnf groupinstall -y "Development Tools"

# C++ コンパイラとビルドツール
sudo dnf install -y gcc gcc-c++ cmake make git

# OpenBLAS ライブラリ
sudo dnf install -y openblas-devel lapack-devel

gcc --version

gcc (GCC) 11.5.0 20240719 (Red Hat 11.5.0-5)
Copyright (C) 2021 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

g++ --version

g++ (GCC) 11.5.0 20240719 (Red Hat 11.5.0-5)
Copyright (C) 2021 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

cmake --version

cmake version 3.22.2

CMake suite maintained and supported by Kitware (kitware.com/cmake).

make --version

GNU Make 4.3
Built for x86_64-amazon-linux-gnu
Copyright (C) 1988-2020 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.

rpm -q openblas openblas-devel

openblas-0.3.18-1.amzn2023.0.3.x86_64
openblas-devel-0.3.18-1.amzn2023.0.3.x86_6

cat /proc/cpuinfo | grep flags | head -n 1

flags		: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch topoext perfctr_core invpcid_single ssbd perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx512_bf16 clzero xsaveerptr rdpru wbnoinvd arat avx512vbmi pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid flush_l1d

lscpu | grep -E "Architecture|Model name|CPU\(s\)|Thread|Core|Socket|NUMA"

Architecture:                            x86_64
CPU(s):                                  4
On-line CPU(s) list:                     0-3
Model name:                              AMD EPYC 9R14
Thread(s) per core:                      1
Core(s) per socket:                      4
Socket(s):                               1
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-3

#include <stdio.h>

extern "C" {
    char* openblas_get_config(void);
    char* openblas_get_corename(void);
    int openblas_get_parallel(void);
    int openblas_get_num_threads(void);
}

int main() {
    printf("=== OpenBLAS Configuration ===\n");
    printf("%s\n", openblas_get_config());
    printf("\n=== Runtime Information ===\n");
    printf("Core Name: %s\n", openblas_get_corename());
    printf("Parallel: %d\n", openblas_get_parallel());
    printf("Threads: %d\n", openblas_get_num_threads());
    return 0;
}

# コンパイル
g++ openblas_info.cpp -o openblas_info -lopenblas

# 実行
./openblas_info

=== OpenBLAS Configuration ===
OpenBLAS 0.3.18 DYNAMIC_ARCH NO_AFFINITY Zen SINGLE_THREADED

=== Runtime Information ===
Core Name: Zen
Parallel: 0
Threads: 1

#include <cblas.h>
#include <chrono>
#include <iostream>
#include <vector>
#include <unistd.h>

void run_benchmark(const char* label) {
    const int N = 10000;
    std::vector<double> A(N*N, 1.0);
    std::vector<double> B(N*N, 2.0);
    std::vector<double> C(N*N, 0.0);

    auto start = std::chrono::high_resolution_clock::now();

    cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans,
                N, N, N, 1.0, A.data(), N, B.data(), N, 0.0, C.data(), N);

    auto end = std::chrono::high_resolution_clock::now();
    auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(end - start);

    std::cout << label << ": " << duration.count() << " ms" << std::endl;
}

int main() {
    std::cout << "Available CPUs: " << sysconf(_SC_NPROCESSORS_ONLN) << std::endl;

    const char* num_threads = getenv("OPENBLAS_NUM_THREADS");
    std::cout << "OPENBLAS_NUM_THREADS: "
              << (num_threads ? num_threads : "not set") << std::endl;

    std::cout << "\nRunning 10000x10000 matrix multiplication...\n" << std::endl;

    run_benchmark("Test 1");
    run_benchmark("Test 2");
    run_benchmark("Test 3");

    return 0;
}

# Serial 版のコンパイル
g++ -O3 performance_test.cpp -o test_serial -lopenblas

# OpenMP 版のコンパイル（マルチスレッド）
g++ -O3 performance_test.cpp -o test_parallel -lopenblaso -fopenmp

export OPENBLAS_NUM_THREADS=1
./test_serial

Available CPUs: 4
OPENBLAS_NUM_THREADS: 1

Running 10000x10000 matrix multiplication...

Test 1: 36785 ms
Test 2: 36775 ms
Test 3: 36772 ms

export OPENBLAS_NUM_THREADS=2
./test_parallel

Available CPUs: 4
OPENBLAS_NUM_THREADS: 2

Running 10000x10000 matrix multiplication...

Test 1: 18319 ms
Test 2: 18311 ms
Test 3: 18311 ms

export OPENBLAS_NUM_THREADS=4
./test_parallel

Available CPUs: 4
OPENBLAS_NUM_THREADS: 4

Running 10000x10000 matrix multiplication...

Test 1: 9192 ms
Test 2: 9200 ms
Test 3: 9210 ms

g++ -O3 -march=native -mtune=native performance_test.cpp -o test_native -lopenblaso -fopenmp

export OPENBLAS_NUM_THREADS=4
./test_native

Available CPUs: 4
OPENBLAS_NUM_THREADS: 4

Running 10000x10000 matrix multiplication...

Test 1: 9185 ms
Test 2: 9203 ms
Test 3: 9193 ms

g++ -O3 -march=znver3 -mtune=znver3 performance_test.cpp -o test_znver3 -lopenblaso -fopenmp

export OPENBLAS_NUM_THREADS=4
./test_znver3

Available CPUs: 4
OPENBLAS_NUM_THREADS: 4

Running 10000x10000 matrix multiplication...

Test 1: 9304 ms
Test 2: 9309 ms
Test 3: 9301 ms

sudo dnf install -y clang19

clang++-19 --version

clang version 19.1.7 (AWS 19.1.7-13.amzn2023.0.1)
Target: x86_64-amazon-linux-gnu
Thread model: posix
InstalledDir: /usr/lib64/llvm19/bin
Configuration file: /etc/clang19/x86_64-amazon-linux-gnu-clang++.cfg

clang++-19 -O3 -march=znver4 -mtune=znver4 performance_test.cpp -o test_znver4 -lopenblaso -fopenmp

export OPENBLAS_NUM_THREADS=4
./test_znver4

Available CPUs: 4
OPENBLAS_NUM_THREADS: 4

Running 10000x10000 matrix multiplication...

Test 1: 9226 ms
Test 2: 9216 ms
Test 3: 9213 ms