1 NPB介绍和安装
NPB介绍和安装详见本人的另外一篇博客,NPB(NAS Parallel Benchmarks)使用、安装和配置。
本文使用的NPB版本是NPB-3.3。
2 mpiP介绍和安装
2.1 mpiP介绍
mpiP是一个用于MPI应用程序的轻量级、伸缩性良好的MPI profiling库。由于mpiP只收集关于MPI函数的统计信息,因此与跟踪工具相比,它生成的开销和数据要少得多。mpiP捕获的所有信息都是任务本地的。它只在报告生成期间(通常在实验结束时)使用通信来将所有任务的结果合并到一个输出文件中。
可以从http://sourceforge.net/projects/mpip下载mpiP的当前版本。本文使用的mpiP版本是3.1.2。
2.2 mpiP安装
mpiP配置安装命令如下
cd mpiP-3.1.2
sudo ./configure
sudo make
3 NPB和mpiP整合
使用mpiP非常简单。因为它通过MPI分析层收集MPI信息,所以mpiP是一个链接时库。也就是说,不必重新编译应用程序来使用mpiP。将NPB和mpiP进行整合,需要将mpiP的链接库链接到NPB的配置文件中。所以在NPB的配置文件中修改FMPI_LIB和CMPI_LIB的内容:
cd NPB3.3-MPI
cd config
vim make.def
链接的格式为:
-L${mpiP_root}/lib -lmpiP -lm -lbfd -liberty -lunwind
其中${mpiP_root}就是mpiP所在地址相对于 NPB3.3-MPI的相对路径。
在我的系统具体的修改内容如下:
39行:FMPI_LIB = -L../../../mpiP-3.1.2/ -lmpiP -lbfd -liberty -lm -lunwind
87行:CMPI_LIB = -L../../../mpiP-3.1.2/ -lmpiP -lbfd -liberty -lm -lunwind
4 验证NPB和mpiP的整合结果
cd NPB-3.3/NPB-3.3-MPI
make ft NPROCS=4 CLASS=A
运行结果如下:
=========================================
= NAS Parallel Benchmarks 3.3 =
= MPI/F77/C =
=========================================
cd FT; make NPROCS=4 CLASS=A
make[1]: Entering directory '/home/dadan/Downloads/NPB3.3/NPB3.3-MPI/FT'
make[2]: Entering directory '/home/dadan/Downloads/NPB3.3/NPB3.3-MPI/sys'
cc -g -o setparams setparams.c
make[2]: Leaving directory '/home/dadan/Downloads/NPB3.3/NPB3.3-MPI/sys'
../sys/setparams ft 4 A
make.def modified. Rebuilding npbparams.h just in case
rm -f npbparams.h
../sys/setparams ft 4 A
mpif77 -c -I/usr/local/mpich/include -O ft.f
mpif77 -O -o ../bin/ft.A.4 ft.o ../common/randi8.o ../common/print_results.o ../common/timers.o -L../../../mpiP-3.1.2/ -lmpiP -lbfd -liberty -lm -lunwind
make[1]: Leaving directory '/home/dadan/Downloads/NPB3.3/NPB3.3-MPI/FT'
mpirun -np 4 bin/ft.A.4
运行结果为:
mpiP:
mpiP:
mpiP: mpiP V3.1.2 (Build Jan 8 2020/10:22:40)
mpiP: Direct questions and errors to mpip-help@lists.sourceforge.net
mpiP:
NAS Parallel Benchmarks 3.3 -- FT Benchmark
No input file inputft.data. Using compiled defaults
Size : 256x 256x 128
Iterations : 6
Number of processes : 4
Processor array : 1x 4
Layout type : 1D
T = 1 Checksum = 5.046735008193D+02 5.114047905510D+02
T = 2 Checksum = 5.059412319734D+02 5.098809666433D+02
T = 3 Checksum = 5.069376896287D+02 5.098144042213D+02
T = 4 Checksum = 5.077892868474D+02 5.101336130759D+02
T = 5 Checksum = 5.085233095391D+02 5.104914655194D+02
T = 6 Checksum = 5.091487099959D+02 5.107917842803D+02
Result verification successful
class = A
FT Benchmark Completed.
Class = A
Size = 256x 256x 128
Iterations = 6
Time in seconds = 1.08
Total processes = 4
Compiled procs = 4
Mop/s total = 6605.74
Mop/s/process = 1651.44
Operation type = floating point
Verification = SUCCESSFUL
Version = 3.3
Compile date = 08 Jan 2020
Compile options:
MPIF77 = mpif77
FLINK = $(MPIF77)
FMPI_LIB = -L../../../mpiP-3.1.2/ -lmpiP -lbfd -libert...
FMPI_INC = -I/usr/local/mpich/include
FFLAGS = -O
FLINKFLAGS = -O
RAND = randi8
Please send the results of this run to:
NPB Development Team
Internet: npb@nas.nasa.gov
If email is not available, send this to:
MS T27A-1
NASA Ames Research Center
Moffett Field, CA 94035-1000
Fax: 650-604-3957
mpiP:
mpiP: Storing mpiP output in [./ft.A.4.4.18246.1.mpiP].
mpiP:
检测在NPB3.3目录下生成的ft.A.4.4.18246.1.mpiP文件