164. stream run on x86¶
- 查询GNU
compiler
得知
-O和-O1优化级别相同。stream测试结果也相同。 -O和-O1测试结果较好，stream官方例子带-o- 不指定编译优化选项，或者指定编译优化选项为
-O0时，测试结果最差 - stream官方要求数组大小使每个数组大于cache的4倍。数组大小满足这个需求之后，增大数组，cache-misses增加，但是性能测试结果不变
164.1. 硬件信息¶
cpu E5-2697A v4 @ 2.60GHz 64核 L3 cache 40MB
[root@localhost ~]# lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 64
On-line CPU(s) list: 0-63
Thread(s) per core: 2
Core(s) per socket: 16
Socket(s): 2
NUMA node(s): 2
Vendor ID: GenuineIntel
CPU family: 6
Model: 79
Model name: Intel(R) Xeon(R) CPU E5-2697A v4 @ 2.60GHz
Stepping: 1
CPU MHz: 1199.196
CPU max MHz: 3600.0000
CPU min MHz: 1200.0000
BogoMIPS: 5188.11
Virtualization: VT-x
L1d cache: 32K
L1i cache: 32K
L2 cache: 256K
L3 cache: 40960K
NUMA node0 CPU(s): 0-15,32-47
NUMA node1 CPU(s): 16-31,48-63
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single pti intel_ppin tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm rdt_a rdseed adx smap intel_pt xsaveopt cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts
内存
32G*16条 = 512G内存 频率2133MT/s
Handle 0x0017, DMI type 17, 40 bytes
Memory Device
Array Handle: 0x0000
Error Information Handle: Not Provided
Total Width: 72 bits
Data Width: 64 bits
Size: 32 GB
Form Factor: DIMM
Set: None
Locator: DIMM130
Bank Locator: _Node1_Channel3_Dimm0
Type: DRAM
Type Detail: Synchronous Registered (Buffered)
Speed: 2133 MT/s
Manufacturer: Hynix
Serial Number: 0x116E4F85
Asset Tag: NO DIMM
Part Number: HMA84GR7MFR4N-TF
Rank: 2
Configured Clock Speed: 2133 MT/s
Minimum Voltage: 1.2 V
Maximum Voltage: 1.2 V
Configured Voltage: 1.2 V
[root@localhost ~]# free -h
total used free shared buff/cache available
Mem: 503G 926M 499G 26M 2.8G 499G
Swap: 4.0G 0B 4.0G
164.2. 软件信息¶
164.2.1. OS centos 7¶
[root@localhost ~]# cat /etc/os-release
NAME="CentOS Linux"
VERSION="7 (Core)"
ID="centos"
ID_LIKE="rhel fedora"
VERSION_ID="7"
PRETTY_NAME="CentOS Linux 7 (Core)"
ANSI_COLOR="0;31"
CPE_NAME="cpe:/o:centos:centos:7"
HOME_URL="https://www.centos.org/"
BUG_REPORT_URL="https://bugs.centos.org/"
CENTOS_MANTISBT_PROJECT="CentOS-7"
CENTOS_MANTISBT_PROJECT_VERSION="7"
REDHAT_SUPPORT_PRODUCT="centos"
REDHAT_SUPPORT_PRODUCT_VERSION="7"
164.2.2. stream 5.10¶
/* Program: STREAM */
/* Revision: $Id: stream.c,v 5.10 2013/01/17 16:01:06 mccalpin Exp mccalpin $ */
/*
164.2.3. perf 3.10.0¶
Installed Packages
Name : perf
Arch : x86_64
Version : 3.10.0
Release : 957.5.1.el7
Size : 5.4 M
Repo : installed
From repo : updates
Summary : Performance monitoring for the Linux kernel
URL : http://www.kernel.org/
License : GPLv2
Description : This package contains the perf tool, which enables performance monitoring
: of the Linux kernel.
164.2.4. gcc 4.8.5¶
[root@localhost stream]# gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/x86_64-redhat-linux/4.8.5/lto-wrapper
Target: x86_64-redhat-linux
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --with-bugurl=http://bugzilla.redhat.com/bugzilla --enable-bootstrap --enable-shared --enable-threads=posix --enable-checking=release --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions --enable-gnu-unique-object --enable-linker-build-id --with-linker-hash-style=gnu --enable-languages=c,c++,objc,obj-c++,java,fortran,ada,go,lto --enable-plugin --enable-initfini-array --disable-libgcj --with-isl=/builddir/build/BUILD/gcc-4.8.5-20150702/obj-x86_64-redhat-linux/isl-install --with-cloog=/builddir/build/BUILD/gcc-4.8.5-20150702/obj-x86_64-redhat-linux/cloog-install --enable-gnu-indirect-function --with-tune=generic --with-arch_32=x86-64 --build=x86_64-redhat-linux
Thread model: posix
gcc version 4.8.5 20150623 (Red Hat 4.8.5-36) (GCC)
164.3. 执行结果¶
164.4. 数组10000000，选项无¶
[root@localhost stream]# gcc stream.c -o stream
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 10000000 (elements), Offset = 0 (elements)
Memory per array = 76.3 MiB (= 0.1 GiB).
Total memory required = 228.9 MiB (= 0.2 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 27683 microseconds.
(= 27683 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 6063.6 0.026436 0.026387 0.026485
Scale: 5873.7 0.027301 0.027240 0.027391
Add: 8484.2 0.028379 0.028288 0.028467
Triad: 7965.8 0.030200 0.030129 0.030277
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
111,963,498 cache-misses
1.291149781 seconds time elapsed
164.5. 数组20000000，选项无¶
gcc -DSTREAM_ARRAY_SIZE=20000000 stream.c -o stream
为了让数组大小大于L3cache的4倍,应该设置20000000个数组元素
200000000*8/1024/1024 = 152 MB
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 20000000 (elements), Offset = 0 (elements)
Memory per array = 152.6 MiB (= 0.1 GiB).
Total memory required = 457.8 MiB (= 0.4 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 50360 microseconds.
(= 50360 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 5960.6 0.053784 0.053686 0.054260
Scale: 5867.8 0.054635 0.054535 0.055155
Add: 8444.3 0.056898 0.056843 0.056956
Triad: 7965.9 0.060358 0.060257 0.060863
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
212,489,174 cache-misses
2.579120788 seconds time elapsed
[root@localhost stream]#
结果相差不多，默认数组大小在x86上执行结果正确。
164.6. 数组20000000，选项-O1¶
指导文档使用-O，经查，等于-O1
gcc -O -DSTREAM_ARRAY_SIZE=20000000 stream.c -o stream
[root@localhost stream]# gcc -O -DSTREAM_ARRAY_SIZE=20000000 stream.c -o stream
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 20000000 (elements), Offset = 0 (elements)
Memory per array = 152.6 MiB (= 0.1 GiB).
Total memory required = 457.8 MiB (= 0.4 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 18355 microseconds.
(= 18355 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 10046.5 0.031868 0.031852 0.031885
Scale: 10236.7 0.031280 0.031260 0.031298
Add: 10847.5 0.044293 0.044250 0.044328
Triad: 11011.7 0.043612 0.043590 0.043641
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
163,072,098 cache-misses
1.749581755 seconds time elapsed
[root@localhost stream]#
164.7. 数组20000000，选项-O1¶
[root@localhost stream]# gcc -O1 -DSTREAM_ARRAY_SIZE=20000000 stream.c -o stream
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 20000000 (elements), Offset = 0 (elements)
Memory per array = 152.6 MiB (= 0.1 GiB).
Total memory required = 457.8 MiB (= 0.4 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 18549 microseconds.
(= 18549 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 10058.8 0.031857 0.031813 0.031907
Scale: 10222.4 0.031368 0.031304 0.031422
Add: 10832.0 0.044360 0.044313 0.044405
Triad: 10977.7 0.043773 0.043725 0.043835
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
162,980,110 cache-misses
1.757360340 seconds time elapsed
[root@localhost stream]#
164.8. 数组20000000，选项-O2¶
[root@localhost stream]# gcc -O2 -DSTREAM_ARRAY_SIZE=20000000 stream.c -o stream
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 20000000 (elements), Offset = 0 (elements)
Memory per array = 152.6 MiB (= 0.1 GiB).
Total memory required = 457.8 MiB (= 0.4 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 18338 microseconds.
(= 18338 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 10048.3 0.031864 0.031846 0.031882
Scale: 10144.9 0.031571 0.031543 0.031592
Add: 10861.4 0.044214 0.044193 0.044234
Triad: 10896.2 0.044092 0.044052 0.044117
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
163,743,497 cache-misses
1.761638820 seconds time elapsed
164.9. 数组20000000，选项-O3¶
[root@localhost stream]# gcc -O3 -DSTREAM_ARRAY_SIZE=20000000 stream.c -o stream
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 20000000 (elements), Offset = 0 (elements)
Memory per array = 152.6 MiB (= 0.1 GiB).
Total memory required = 457.8 MiB (= 0.4 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 18628 microseconds.
(= 18628 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 16874.0 0.018975 0.018964 0.018988
Scale: 9966.6 0.032122 0.032107 0.032143
Add: 10795.3 0.044488 0.044464 0.044501
Triad: 10761.4 0.044620 0.044604 0.044649
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
155,187,006 cache-misses
1.653922727 seconds time elapsed
[root@localhost stream]#
164.10. 数组20000000，选项-O0¶
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 20000000 (elements), Offset = 0 (elements)
Memory per array = 152.6 MiB (= 0.1 GiB).
Total memory required = 457.8 MiB (= 0.4 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 50331 microseconds.
(= 50331 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 5956.9 0.053873 0.053719 0.054393
Scale: 5870.6 0.054687 0.054509 0.055268
Add: 8448.0 0.056944 0.056818 0.057079
Triad: 7960.3 0.060478 0.060299 0.061003
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
212,044,019 cache-misses
2.581445722 seconds time elapsed
[root@localhost stream]#
164.11. 数组30000000，选项-O¶
[root@localhost stream]# gcc -O -DSTREAM_ARRAY_SIZE=30000000 stream.c -o stream
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 30000000 (elements), Offset = 0 (elements)
Memory per array = 228.9 MiB (= 0.2 GiB).
Total memory required = 686.6 MiB (= 0.7 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 26309 microseconds.
(= 26309 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 10593.9 0.045330 0.045309 0.045347
Scale: 10608.0 0.045292 0.045249 0.045328
Add: 11352.4 0.063469 0.063423 0.063617
Triad: 11288.3 0.063819 0.063783 0.063925
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
249,967,744 cache-misses
2.531537686 seconds time elapsed
164.12. 数组40000000，选项-O¶
[root@localhost stream]# gcc -O -DSTREAM_ARRAY_SIZE=40000000 stream.c -o stream
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 40000000 (elements), Offset = 0 (elements)
Memory per array = 305.2 MiB (= 0.3 GiB).
Total memory required = 915.5 MiB (= 0.9 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 35411 microseconds.
(= 35411 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 10221.9 0.062669 0.062611 0.062891
Scale: 10435.7 0.061434 0.061328 0.061619
Add: 10980.6 0.087577 0.087427 0.087796
Triad: 11013.6 0.087207 0.087165 0.087377
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
334,073,899 cache-misses
3.469917205 seconds time elapsed
164.13. 数组50000000，选项-O¶
[root@localhost stream]# gcc -O -DSTREAM_ARRAY_SIZE=50000000 stream.c -o stream
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 50000000 (elements), Offset = 0 (elements)
Memory per array = 381.5 MiB (= 0.4 GiB).
Total memory required = 1144.4 MiB (= 1.1 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 44100 microseconds.
(= 44100 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 10957.6 0.073081 0.073009 0.073334
Scale: 10329.6 0.077563 0.077447 0.077736
Add: 11045.7 0.108870 0.108640 0.109140
Triad: 11196.7 0.107286 0.107175 0.107587
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
426,955,169 cache-misses
4.245849035 seconds time elapsed
164.14. 数组60000000，选项-O¶
[root@localhost stream]# gcc -O -DSTREAM_ARRAY_SIZE=60000000 stream.c -o stream
[root@localhost stream]# perf stat -e cache-misses ./stream
-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 60000000 (elements), Offset = 0 (elements)
Memory per array = 457.8 MiB (= 0.4 GiB).
Total memory required = 1373.3 MiB (= 1.3 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 52796 microseconds.
(= 52796 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 10130.0 0.094858 0.094768 0.095173
Scale: 10631.8 0.090408 0.090295 0.090570
Add: 11216.3 0.128531 0.128385 0.128746
Triad: 11289.0 0.127709 0.127558 0.127884
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
Performance counter stats for './stream':
495,551,653 cache-misses
5.104867523 seconds time elapsed
[root@localhost stream]#