PPC
Steps:
1
to Launch the tool
- The file with the sparse matrix
for this exercise must be in the Virtual Directory of MD.
- Open the PPC Viewer directly
from Migrating Desktop. In the main window, it appears four menus.
·
Grid:
get/set the Grid parameters
·
Kernel:
select the kernel to be predicted
·
Display:
show prediction results
·
Help
2
to get and show grid related
information
- Get the
information provided by JIMS (Grid menu → Monitoring → Get
data)
- Show the grid information (Grid menu → Monitoring → Show
data)
·
Show the information of a
specific node (Select a node number)
3
to select the kernel
- Select the kernel to be predicted (in this exercise, the
Iterative Solvers)
- Load the matrix (Kernel menu → Iterative solvers → Load
matrix)
- Select the matrix in the
Virtual Directory of MD
- Select the Conjugate Gradient iterative solver (Kernel menu → Iterative solver
→ CG)
- In the pop-up window, select
preconditioner (ILU0) and scaling (Symmetric)
- In the new pop-up window, select the number of rows/columns of
the processor mesh
4
to display the results
- Display the results (Display menu)
- Display the number of FLOPs (Floating Point Operations) for a
specific processor (Display menu → Computations)
- FLOPs for different parts of
the kernel can be showed:
- Iterative method itself
- Scaling of the matrix
- Computation of the
preconditioner
- Application of the
preconditioner
- Total FLOPs for the whole
kernel
- Display the predicted load balance (Display menu → Load balance)
- Show Kiviat Diagrams in terms of:
- FLOPs
- Memory
- Executing time