01_processor_core/01_polynominal

Hands-on: Polynomials with many floating-point operations

• Choose either C/C++ (cpp) or Fortran (fortran) samples. Both of them are fine, as well.

C/C++

How to compile and how to execute

1. Run a script of creating working space

• edit task.sh (modify --gname option)
• See objst in create_project.sh.
  • It indicates a list of the settings of compile options.
• Running create_project.sh, the executable file (run.x) is generated in Obj_*** directory.
  • This directory is also automatically generated.
  • When you desire to use the own compiler, set MAKE_DIR variable in create_project.sh from config to config.own.
• Example:

$ cd cpp
$ vi task.sh
$ bash create_project.sh 
$ ls 
Obj_clang  Obj_clang.512.swp  Obj_clang.inlv2  Obj_clang.novec  

2. Run program

• A job script to execute the program is located in Obj_***/results.
• You can run the program either:

## To run as a batch job
$ cd Obj_clang/results
$ pjsub task.sh  
## Or, to run in an interactive job
$ cd Obj_clang/results
$ bash task.sh  

• Each of the samples in the Exercises will be completed within 1-2 minutes.
  • For safety, we set the upper limit of job elapsed time as 3 minutes.

Exercises A

• E1: Check differences between Makefile.clang and Makefile.clang.novec. Examine the meaning of the compiler options related to the differences.
• E2: Check which of options are related to generating a compiler report in FJ (*.lst) and diagnosis message.
• E3: Read the compiler report (*.lst) in each of Obj_*** directories. Also, check the diagnosis message in a log file of the compiler (make.log).
• E4: Compare the performance (FLOP/s) in Obj_clang to that in Obj_clang.novec. You can find the value of FLOP/s in the standard output of the executable file.
• E5: Check the behaviors when setting the compiler options related to optimization on instruction scheduling.
  • Repeat the above exercises, when using Makefile.clang.inlv2 and Makefile.clang.512.swp.
  • Compare the performance (FLOP/s) to that in Obj_clang. Which of the compiler options is the best? Furthermore, compare the best result to the peak FLOP/s of a single core in A64FX.

Exercises B (advanced)

• E6: Test the cases when superword-level parallelism (SLP) vectorization is enable (i.e., -fslp-vectorize).
• E7: Test the cases when using Trad mode (e.g., Makefile.trad.).
• E8: Measure the cases in cpp_ExpressionTemplate. Here, we use a C++ template technique, Expression Template. This technique includes a sort of inlining functions in compiling time.
• E9: In cpp.fapp, performance the basic CPUPA analysis using fapp. How does the data of cycle accounting change varying compiler options?
• E10: Measure the cases using the normal mode of Fugagku (freq=2000, eco_state=0), and compare the performance (FLOP/s) with the peak performance of a single core of A64FX in the normal mode.

Note

• We summarize the main compiler options of each Makefile.

Makefile.clang         #Clang mode. Auto-vectorization with scalable vector length is allowed. 
Makefile.clang.512     #Clang mode. Auto-vectorization with 512-bit vector length is allowed.
Makefile.clang.512.swp #Clang mode. Auto-vectorization with 512-bit vector length and software pipelining are allowed.
Makefile.clang.inlv2   #Clang mode. Auto-vectorization with scalable vector length and interleaving with count=2 are allowed.
Makefile.clang.novec   #Clang mode, without vectorization. 
Makefile.trad          #Trad mode. Auto-vectorization (w/ 512-bit vector length) is allowed, but software pipelining is not allowed.
Makefile.trad.nosimd   #Trad mode, without vectorization.
Makefile.trad.swp      #Trad mode. Auto-vectorization (w/ 512-bit vector length) and software pipelining are allowed.

Fortran

How to compile and how to execute

1. Run a script of creating working space

• edit task.sh (modify --gname option)
• See objst in create_project.sh.
  • It indicates a list of the settings of compile options.
• Running create_project.sh, the executable file (run.x) is generated inObj_*** directory.
  • This directory is also automatically generated.
  • When you desire to use the own compiler, set MAKE_DIR variable in create_project.sh from config to config.own.
• Example:

$ cd fortran
$ vi task.sh
$ bash create_project.sh 
$ ls 
Obj_nosimd  Obj_simd  Obj_swp  

2. Run program

• A job script to execute the program is located in Obj_***/results.
• You can run the program either:

## To run as a batch job
$ cd Obj_simd/results
$ pjsub task.sh  
## Or, to run in an interactive job
$ cd Obj_simd/results
$ bash task.sh  

• Each of the samples in the Exercises will be completed within 1-2 minutes.
  • For safety, we set the upper limit of job elapsed time as 3 minutes.

Exercises A

• E1: Check differences between Makefile.simd and Makefile.nosimd. Examine the meaning of the compiler options related to the differences.
• E2: Check which of options are related to generating a compiler report in FJ (*.lst) and diagnosis message.
• E3: Read the compiler report (*.lst) in each of Obj_*** directories. Also, check the diagnosis message in a log file of the compiler (make.log).
• E4: Compare the performance (FLOP/s) in Obj_simd to that in Obj_nosimd. You can find the value of FLOP/s in the standard output of the executable file.
• E5: Check the behaviors when setting the compiler options related to optimization on instruction scheduling.
  • Repeat the above exercises, when using Makefile.swp.
  • Compare the performance (FLOP/s) to that in Obj_simd. Which of the compiler options is the best? Furthermore, compare the best result to the peak FLOP/s of a single core in A64FX.

Exercises B (advanced)

• E6: Test the cases when the policy of software pipelining is changed; -Kswp_policy=large and -Kswp_policy=small, for example.
• E7: Measure the cases in fortran_DoConcurrent. Here, we use do concurrent statement in Fortran 2008 standard.
• E8: Measure the cases in fortran_AssumedShapeArray. Here, we write the kernel in a modern Fortran manner.
• E9: In fortran.fapp, performance the basic CPUPA analysis using fapp. How does the data of cycle accounting change, varying compiler options?
• E10: Measure the cases using the normal mode of Fugagku (freq=2000, eco_state=0), and compare the performance (FLOP/s) with the peak performance of a single core of A64FX in the normal mode.

Note

• We summarize the main compiler options of each Makefile.

Makefile.nosimd  #Auto-vectorization is not allowed. 
Makefile.simd    #Auto-vectorization is allowed.
Makefile.swp     #Auto-vectorization and software pipelining are allowed.