norm2 test failing on Kokkos impl

[mzuzek]

@mhoemmen @fnrizzi
norm2 native test fails when executed on Kokkos implementation (with LINALG_ENABLE_KOKKOS_DEFAULT=ON), e.g. see this CI job:

      Start 14: norm2
15/36 Test #14: norm2 ................................***Failed    0.01 sec
Running main() from /home/runner/work/stdBLAS/stdBLAS/build/stdblas/_deps/googletest-src/googletest/src/gtest_main.cc
[==========] Running 4 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 4 tests from BLAS1_norm2
[ RUN      ] BLAS1_norm2.mdspan_zero
[       OK ] BLAS1_norm2.mdspan_zero (0 ms)
[ RUN      ] BLAS1_norm2.mdspan_one
/home/runner/work/stdBLAS/stdBLAS/src/stdblas/tests/native/norm2.cpp:85: Failure
Expected equality of these values:
  expectedNormResult + mag_t(1.0)
    Which is: 4
  normResultPlusOne
    Which is: 3.16228
[  FAILED  ] BLAS1_norm2.mdspan_one (0 ms)
[ RUN      ] BLAS1_norm2.mdspan_double
[       OK ] BLAS1_norm2.mdspan_double (0 ms)
[ RUN      ] BLAS1_norm2.mdspan_complex_double
[       OK ] BLAS1_norm2.mdspan_complex_double (0 ms)
[----------] 4 tests from BLAS1_norm2 (0 ms total)

[----------] Global test environment tear-down
[==========] 4 tests from 1 test suite ran. (0 ms total)
[  PASSED  ] 3 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] BLAS1_norm2.mdspan_one

 1 FAILED TEST

[mzuzek]

@mhoemmen @fnrizzi

According to P1673R8, norm2 should use:

vector_sum_of_squares(x, {.scaling_factor=1.0, .scaled_sum_of_squares=init})

Instead the current serial implementation uses:

vector_sum_of_squares(x, {.scaling_factor=0.0, .scaled_sum_of_squares=1.0})

and then includes init outside sqrt()...

That could be easy to fix, but just below we have the current serial implementation of vector_sum_of_squares() which is doing even stranger things in the calculation (e.g. accumulating squares of reversed values: 1/abs(x[i])), so it's not trivial/immediate for me to address.

[mzuzek]

@mhoemmen @fnrizzi

In P1673R8 we have:

1. sounds like init should be squared too;
2. there is no A in norm2 - should it be v ?

[fnrizzi]

Issue 141 is opened and related to this

[mzuzek]

Thanks @fnrizzi!
If that's a known bug (and out of current scope ?), maybe we should disable norm2 test for now and add re-enabling it to the TODOs here in the issue?