GEMMTR changes the inputs in CBLAS tests

[cenewcombe]

This is probably not a priority given the temporary nature of the gemmtr interface, but similar to #5111, CBLAS tests fail for all S/D/C/Z as follows due to changed parameters:

 cblas_sgemmtr PASSED THE TESTS OF ERROR-EXITS

 ******* FATAL ERROR - PARAMETER NUMBER 12 WAS CHANGED INCORRECTLY *******
 ******* cblas_sgemmtr FAILED ON CALL NUMBER:
   4861: cblas_sgemmtr( CblasColMajor,CblasUpper    ,  CblasNoTrans,  CblasNoTrans,
            9,  0,  0.0 , A, 10, B,  1,  0.0 , C, 10).
 ******* cblas_sgemmtr FAILED ON CALL NUMBER:
    973: cblas_sgemmtr( CblasRowMajor,CblasUpper    ,  CblasNoTrans,  CblasNoTrans,
            1,  0,  0.0 , A,  2, B,  1,  0.0 , C,  2).

[martin-frbg]

Are you testing with a post-PR#5143 checkout of the develop branch ? I cannot reproduce this (on x86_64 at least) with the updated CBLAS tests from Reference-LAPACK (that I still need to merge here)

[cenewcombe]

My original description was wrong, I am only seeing this with S/D gemmtr, whereas the fix was for complex types. I don't see this issue when I run the BLATs, only the CBLATs. I have also borrowed tests from Reference-Lapack so I will double check my testing setup. I'm using GCC 13 on x86-64 to reproduce this. Do your CBLAS level 3 tests pass for single and double real? Thanks

[martin-frbg]

They did/do pass in a GNU Make build (gcc 14.2 here but I doubt it would matter for such a fundamental fault) - I realized yesterday that the CMake build was broken, using the regular GEMM source to build GEMMTR (though "GEMMT" was using the correct sources except for SBGEMMT).