Quad: command line flag

[Konstantin8105]

Hello, Input data:

%%MatrixMarket matrix coordinate real general
2 2 4 1 0
1 1 1.0000000000000000e+00
1 2 2.0000000000000000e+00
2 1 1.0000000000000000e+00
2 2 1.0000000000000000e+00
1 3.0000000000000000e+00
2 2.0000000000000000e+00

Run:

~/lis/bin/lsolve /tmp/103353238/input 0 1.txt 2.txt

solution:

%%MatrixMarket vector coordinate real general
2
1   1.00000000000000022204e+00
2   9.99999999999999888978e-01

But if I try run with flag -f quad:

~/lis/bin/lsolve /tmp/103353238/input 0 1.txt 2.txt -f quad

number of processes = 1
matrix size = 2 x 2 (4 nonzero entries)

initial vector x      : all components set to 0
precision             : quad
linear solver         : BiCG
preconditioner        : none
convergence condition : ||b-Ax||_2 <= 1.0e-12 * ||b-Ax_0||_2
matrix storage format : CSR
linear solver status  : normal end

BiCG: number of iterations = 1
BiCG:   double             = 0
BiCG:   quad               = 1
BiCG: elapsed time         = 3.619194e-04 sec.
BiCG:   preconditioner     = 3.814697e-06 sec.
BiCG:     matrix creation  = 1.907349e-06 sec.
BiCG:   linear solver      = 3.581047e-04 sec.
BiCG: relative residual    = 0.000000e+00

solution is not correct:

%%MatrixMarket vector coordinate real general
2
1   3.00000000000000000000e+00
2   2.00000000000000000000e+00

Us I check, it is happen because I don't add flag ./configure --enable-quad. After recompilation with that flag - All is ok.

In my point of view, if I compile without flag --enable-quad and use flag -f quad, then result is Error but not linear solver status : normal end.

[anishida]

Hi,

Thank you for your information. We released a new version 2.0.16, in which option '-f quad' is disabled when quad precision is not enabled. We appreciate it if you could check it.

Sincerely, Akira Nishida

From: Konstantin [mailto:notifications@github.com] Sent: Wednesday, September 26, 2018 5:35 AM To: anishida/lis lis@noreply.github.com Cc: Subscribed subscribed@noreply.github.com Subject: [anishida/lis] Quad: command line flag (#19)

Input data: %%MatrixMarket matrix coordinate real general 2 2 4 1 0 1 1 1.0000000000000000e+00 1 2 2.0000000000000000e+00 2 1 1.0000000000000000e+00 2 2 1.0000000000000000e+00 1 3.0000000000000000e+00 2 2.0000000000000000e+00 Run: ~/lis/bin/lsolve /tmp/103353238/input 0 1.txt 2.txt solution: %%MatrixMarket vector coordinate real general 2 1 1.00000000000000022204e+00 2 9.99999999999999888978e-01 But if I try run with flag -f quad: ~/lis/bin/lsolve /tmp/103353238/input 0 1.txt 2.txt -f quad

number of processes = 1 matrix size = 2 x 2 (4 nonzero entries)

initial vector x : all components set to 0 precision : quad linear solver : BiCG preconditioner : none convergence condition : ||b-Ax||_2 <= 1.0e-12 * ||b-Ax_0||_2 matrix storage format : CSR linear solver status : normal end

BiCG: number of iterations = 1 BiCG: double = 0 BiCG: quad = 1 BiCG: elapsed time = 3.619194e-04 sec. BiCG: preconditioner = 3.814697e-06 sec. BiCG: matrix creation = 1.907349e-06 sec. BiCG: linear solver = 3.581047e-04 sec. BiCG: relative residual = 0.000000e+00 solution is not correct: %%MatrixMarket vector coordinate real general 2 1 3.00000000000000000000e+00 2 2.00000000000000000000e+00 Us I check, it is happen because I don't add flag ./configure --enable-quad. After recompilation with that flag - All is ok. In my point of view, if I compile without flag --enable-quad and use flag -f quad, then result is Error but not linear solver status : normal end. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[Konstantin8105]

Thank you for feedback.

[Konstantin8105]

Thank you, I see PR about that.