49 lines
1.3 KiB
Text
49 lines
1.3 KiB
Text
UPDATE 2025-12-18
|
|
Ex2_Second_Attempt:
|
|
* Makefile missing [GH: added]
|
|
* execution::par : correct results and seems 6x faster on 8 cores.
|
|
|
|
Ex5_Second_Attempt:
|
|
* FEM assembling parallel OK
|
|
* no parallelization of GetDiag
|
|
* extra _omp version of JacobiSolve and two others
|
|
This is not necessary. Parallelize directly in the functions, also in vdop.cpp
|
|
See pdf Ex5_Second_Attempt.pdf
|
|
----------------------------
|
|
Summery of your results?!
|
|
|
|
g++ -O3 -fopenmp mainEx1.cpp mylib.cpp -o dotprod
|
|
I added Makefile, CLANG_default.mk GCC_default.mk to your repository
|
|
|
|
Linux: Using g++
|
|
> make run EX=Ex5
|
|
Using clang++
|
|
> make run EX=Ex5 COMPILER=CLANG_
|
|
|
|
Added
|
|
#include <algorithm> // GH: transform()
|
|
in mylib.pp:2
|
|
|
|
1:
|
|
no scheduling tested
|
|
reduction_vec_append() implemented but never tested.
|
|
|
|
|
|
2: mainEx2.cpp
|
|
no parallelization at all. neighter OpenMP nor C++ execution policies
|
|
|
|
3: mainEx3.cpp
|
|
nested parallelization in count_goldbach(), single_goldbach()
|
|
Did that pay off?!
|
|
|
|
4:
|
|
Try
|
|
collapse(2)
|
|
in bench_funcs.cpp:75
|
|
(faster in my code for Mat-Mat-Mult)
|
|
|
|
5:
|
|
Why not using my provided 2D FEM code? That was already required in exercise 3.
|
|
Code only 1D FEM whioch is quite simple to parallelize because of the very simple matrix pattern.
|
|
jacobi_par_parallel()
|
|
build_fem_system_atomic() OK
|