tanakamura

AMD Ryzen 7 1700X Eight-Core Processor         
4k page
parallel rand read : 0.464111[ns/read], range=64[KB]
parallel rand read : 0.494720[ns/read], range=128[KB]
parallel rand read : 0.567612[ns/read], range=256[KB]
parallel rand read : 0.642448[ns/read], range=512[KB]
parallel rand read : 0.713230[ns/read], range=1024[KB]
parallel rand read : 0.737040[ns/read], range=2048[KB]
parallel rand read : 0.773684[ns/read], range=4096[KB]
parallel rand read : 2.559083[ns/read], range=8192[KB]

tanakamura

#define _GNU_SOURCE
#include <stdio.h>
#include <sys/time.h>
#include <time.h>
#include <stdlib.h>
#include <cpuid.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
#include <sys/mman.h>

tanakamura

#include <stdio.h>
#include <x86intrin.h>
#include <math.h>
#include <unistd.h>
#include <time.h>
#include <sys/time.h>

static double calib_ops() {
    long long n = 1024*1024*4;
    long long core_cycle = 64 * n;

tanakamura

static double getsec()
{
    struct timespec ts;

    clock_gettime(CLOCK_MONOTONIC, &ts);

    return (ts.tv_nsec / 1e9) + ts.tv_sec;
}
static double get_sec() {
    struct timespec ts;

tanakamura

#include <stdio.h>
#include <unistd.h>
#include <time.h>
#include <sys/mman.h>
#include <getopt.h>
#include <sys/time.h>
#include <stdlib.h>
#include <string.h>
#include <immintrin.h>
#include <pthread.h>

tanakamura

num_thread = 1
libc-memset      :     1024[ B] 42.105008[GB/s]
libc-memset      :     2048[ B] 42.338436[GB/s]
libc-memset      :     4096[ B] 49.077192[GB/s]
libc-memset      :     8192[ B] 48.563454[GB/s]
libc-memset      :       16[KB] 50.124791[GB/s]
libc-memset      :       32[KB] 49.960671[GB/s]
libc-memset      :       64[KB] 49.425556[GB/s]
libc-memset      :      128[KB] 49.101868[GB/s]
libc-memset      :      256[KB] 47.760674[GB/s]

tanakamura

/* 
 * clzero の後にstoreするとmemsetよりはるかに悪くなる 
 *
 *  * clzeroのみ      :30[GB/s] 
 *  * clzero + ストア :3[GB/s]
 *  * memset          :12.5[GB/s]
 */
void *amd_clzero(void *dst, const void *src, size_t sz)
{
    size_t line_size = 64;

tanakamura

#include <stdio.h>
#include <time.h>
#include <sys/time.h>

double getsec() {
  struct timespec tv;
  clock_gettime(CLOCK_MONOTONIC, &tv);

  return tv.tv_sec + tv.tv_nsec/1e9;
}

tanakamura

Ryzen 7 3700x vs i7-6700

============= LATENCY ==============================================================================
                              instruction |     IPC         (   rel[%]),     CPI         (   rel[%])
------------------------------------------+---------------------------------------------------------
    m128                            addps |    0.33-0.25    (  33.3[%]),    3.00-4.00    ( -25.0[%])
    m128                           aesdec |    0.25-0.25    (   0.0[%]),    4.00-4.00    (  -0.0[%])
    m128                       aesdeclast |    0.25-0.25    (  -0.0[%]),    4.00-4.00    (   0.0[%])
    m128                           aesenc |    0.25-0.25    (   0.0[%]),    4.00-4.00    (  -0.0[%])
    m128                       aesenclast |    0.25-0.25    (  -0.0[%]),    4.00-4.00    (   0.0[%])

tanakamura

num_thread = 1
libc-memset      :     1024[ B] 109.797060[GB/s]
libc-memset      :     2048[ B] 119.469974[GB/s]
libc-memset      :     4096[ B] 124.585380[GB/s]
libc-memset      :     8192[ B] 127.374256[GB/s]
libc-memset      :       16[KB] 126.546911[GB/s]
libc-memset      :       32[KB] 125.730241[GB/s]
libc-memset      :       64[KB] 120.244104[GB/s]
libc-memset      :      128[KB] 118.669685[GB/s]
libc-memset      :      256[KB] 112.864779[GB/s]