Repository navigation
Add nnz argument to set_csr_data in sparse CG examples - #2696
chuanyuf merged 2 commits into
Conversation
Sorry, something went wrong.
spencerpatty
left a comment
There was a problem hiding this comment.
LGTM thansk for making this update (we should denote that it should coincide with 2025.2 release ... do we need to keep both solutions there with macro conditionals that check what version of MKL it is built against ?
Sorry, something went wrong.
Done. The 2025.3 version of the example will build and run without warning even when using previous oneMKL releases. This is not required but nice to have, and it makes it visible that the set_csr_data API changed from 2025.3. |
Sorry, something went wrong.
…se_cg_example Add nnz argument to set_csr_data in sparse CG examples
Change in sparse_cg sample examples required for the 2025.3 tag
Description
Add nnz (number of non-zero elements in the sparse matrix) to the set_csr_data call, to be aligned with the API's updated signature in the oneMKL 2025.3 release, and to prevent build warnings when using the deprecated API without nnz.
Type of change
How Has This Been Tested?
I used make on the command line to build and run the example on a PVC Linux system, using a oneMKL 2025.3 Release Candidate build.
Note that:
running the example without this fix against oneMKL 2025.3 will work but there will be warnings:
sparse_cg.cpp:383:44: warning: 'set_csr_data' is deprecated: Use oneapi::mkl::sparse::set_csr_data(queue, spmat, nrows, ncols, nnz, ...) instead..
running the example with this fix against oneMKL 2025.2 (or previous releases) will cause a build failure.
Command Line
oneapi-cli
Visual Studio
Eclipse IDE
VSCode
Building:
Running:
./sparse_cg ######################################################################## # Sparse Preconditioned Conjugate Gradient Solver with USM # # Uses the preconditioned conjugate gradient algorithm to # iteratively solve the symmetric linear system # # A * x = b # # where A is a symmetric sparse matrix in CSR format, and # x and b are dense vectors. # # Uses the symmetric Gauss-Seidel preconditioner. # # alpha and beta constants in PCG algorithm are host side. # ######################################################################## Running tests on Intel(R) Data Center GPU Max 1550. Running with single precision real data type: sparse PCG parameters: A size: (4096, 4096) Preconditioner = Symmetric Gauss-Seidel max iterations = 500 relative tolerance limit = 1e-05 absolute tolerance limit = 0.0005 relative norm of residual on 1 iteration: 0.178532 relative norm of residual on 2 iteration: 0.0280123 relative norm of residual on 3 iteration: 0.0048948 relative norm of residual on 4 iteration: 0.000796108 relative norm of residual on 5 iteration: 0.000119025 relative norm of residual on 6 iteration: 1.86945e-05 absolute norm of residual on 6 iteration: 0.000149556 Preconditioned CG process has successfully converged in absolute error in 6 steps with relative error ||r||_2 / ||r_0||_2 = 1.86945e-05 > 1e-05 absolute error ||r||_2 = 0.000149556 < 0.0005 Running with double precision real data type: sparse PCG parameters: A size: (4096, 4096) Preconditioner = Symmetric Gauss-Seidel max iterations = 500 relative tolerance limit = 1e-05 absolute tolerance limit = 0.0005 relative norm of residual on 1 iteration: 0.178532 relative norm of residual on 2 iteration: 0.0280123 relative norm of residual on 3 iteration: 0.0048948 relative norm of residual on 4 iteration: 0.000796108 relative norm of residual on 5 iteration: 0.000119025 relative norm of residual on 6 iteration: 1.86945e-05 absolute norm of residual on 6 iteration: 0.000149556 Preconditioned CG process has successfully converged in absolute error in 6 steps with relative error ||r||_2 / ||r_0||_2 = 1.86945e-05 > 1e-05 absolute error ||r||_2 = 0.000149556 < 0.0005 ./sparse_cg2 ######################################################################## # Sparse Preconditioned Conjugate Gradient Solver with USM 2 # # Uses the preconditioned conjugate gradient algorithm to # iteratively solve the symmetric linear system # # A * x = b # # where A is a symmetric sparse matrix in CSR format, and # x and b are dense vectors. # # Uses the symmetric Gauss-Seidel preconditioner. # # alpha and beta constants in PCG algorithm are kept # device side. # ######################################################################## Running tests on Intel(R) Data Center GPU Max 1550. Running with single precision real data type: sparse PCG parameters: A size: (4096, 4096) Preconditioner = Symmetric Gauss-Seidel max iterations = 500 relative tolerance limit = 1e-05 absolute tolerance limit = 0.0005 relative norm of residual on 1 iteration: 0.178532 relative norm of residual on 2 iteration: 0.0280123 relative norm of residual on 3 iteration: 0.0048948 relative norm of residual on 4 iteration: 0.000796109 relative norm of residual on 5 iteration: 0.000119025 relative norm of residual on 6 iteration: 1.86945e-05 absolute norm of residual on 6 iteration: 0.000149556 Preconditioned CG process has successfully converged in absolute error in 6 steps with relative error ||r||_2 / ||r_0||_2 = 1.86945e-05 > 1e-05 absolute error ||r||_2 = 0.000149556 < 0.0005 Running with double precision real data type: sparse PCG parameters: A size: (4096, 4096) Preconditioner = Symmetric Gauss-Seidel max iterations = 500 relative tolerance limit = 1e-05 absolute tolerance limit = 0.0005 relative norm of residual on 1 iteration: 0.178532 relative norm of residual on 2 iteration: 0.0280123 relative norm of residual on 3 iteration: 0.0048948 relative norm of residual on 4 iteration: 0.000796108 relative norm of residual on 5 iteration: 0.000119025 relative norm of residual on 6 iteration: 1.86945e-05 absolute norm of residual on 6 iteration: 0.000149556 Preconditioned CG process has successfully converged in absolute error in 6 steps with relative error ||r||_2 / ||r_0||_2 = 1.86945e-05 > 1e-05 absolute error ||r||_2 = 0.000149556 < 0.0005