You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Rank-k and Rank-2k preconditions are wrong and other problems #274
Hi @rasolca ! Thanks for reporting this! We plan to report this as an LWG issue. If you would like credit for this in the public record, please contact me offline, using my e-mail address on the proposal. Thanks!
[linalg.algs.blas2.gemv]
I agree with your suggestion.
[linalg.algs.blas3.rankk] and [linalg.algs.blas3.rank2k]
I agree.
[linalg.algs.blas3.inplacetrsm]
Hold on, let me look at this (will update)
[linalg.algs.blas3.trsm] 6
I agree. (Note to self: This refers only to the left solve. The complexity of the right solve in (13) looks OK.)
[linalg.algs.blas3.inplacetrsm] 13
I agree. (Note to self: This refers only to the right solve. The complexity of the left solve in (6) looks OK.)
Please see also https://isocpp.org/files/papers/P3371R0.html , "Fix C++26 by making the symmetric and Hermitian rank-k and rank-2k updates consistent with the BLAS," which will come out in the next mailing.
Activity
mhoemmen commentedon Jul 5, 2024
Thanks for reporting! We'll take a look as soon as we can.
rasolca commentedon Jul 9, 2024
I had an extra look to the proposal and found other problems.
I summarize all the problems I found here:
GEMV
possibly-addable<decltype(x), decltype(y), decltype(z)>()istruepossibly-addable<decltype(y), decltype(y), decltype(z)>()istrueaddable(x,y,z)istrueaddable(y,y,z)is trueRANKK
compatible-static-extents<decltype(A), decltype(A)>(0, 1)istrue<delete>A.extent(0)equalsA.extent(1)<delete>RANK2K
possibly-addable<decltype(A), decltype(B), decltype(C)>()istruepossibly-multipliable<decltype(C), decltype(A), decltype(B)>()istruecompatible-static-extents<decltype(A), decltype(A)>(0, 1)istruecompatible-static-extents<decltype(C), decltype(C)>(0, 1)istrueaddable(A, B, C)istruemultipliable(C, A, B)istrueA.extent(0)equalsA.extent(1)C.extent(0)equalsC.extent(1)TRMM in-place
linalg.algs.blas3.inplacetrsm (left)linalg.algs.blas3.trmm (left)
B = A XC <- A C
A(n,n), X(m,n) B(m,n)A(m,m), C(m,n)
O(A.extent(0)×A.extent(1)×C.extent(0))O(A.extent(0)×A.extent(1)×C.extent(1))TRSM
O(A.extent(0)×X.extent(1)×X.extent(1))O(A.extent(0)×X.extent(0)×X.extent(1))TRSM in-place
O(A.extent(0)×A.extent(1)×B.extent(1))O(B.extent(0)×A.extent(0)×A.extent(1))[-]Rank-k and Rank-2k preconditions are wrong.[/-][+]Rank-k and Rank-2k preconditions are wrong and other problems[/+]mhoemmen commentedon Jul 16, 2024
Hi @rasolca ! Thanks for reporting this! We plan to report this as an LWG issue. If you would like credit for this in the public record, please contact me offline, using my e-mail address on the proposal. Thanks!
[linalg.algs.blas2.gemv]
I agree with your suggestion.
[linalg.algs.blas3.rankk] and [linalg.algs.blas3.rank2k]
I agree.
[linalg.algs.blas3.inplacetrsm]
Hold on, let me look at this (will update)
[linalg.algs.blas3.trsm] 6
I agree. (Note to self: This refers only to the left solve. The complexity of the right solve in (13) looks OK.)
[linalg.algs.blas3.inplacetrsm] 13
I agree. (Note to self: This refers only to the right solve. The complexity of the left solve in (6) looks OK.)
rasolca commentedon Jul 17, 2024
Sorry, for the third case (the one you said you will have a look), I made a couple of copy-paste typos.
I updated the table above.
mhoemmen commentedon Jul 18, 2024
@rasolca Thanks for the update! Here are my updated corrections: ORNL/cpp-proposals-pub#464 (comment) .
mhoemmen commentedon Aug 13, 2024
@rasolca Thanks so much for your help! I've submitted an LWG issue for these: https://cplusplus.github.io/LWG/issue4137 . Please let me know if there are any further issues; thanks! : - )
Please see also https://isocpp.org/files/papers/P3371R0.html , "Fix C++26 by making the symmetric and Hermitian rank-k and rank-2k updates consistent with the BLAS," which will come out in the next mailing.