Re: gm2 frontend in gcc 14 (fedora 40) bug

[Michael Riedl]

Hallo Gerald,

then it's definitely buggy "A x (solution vector)" should give back approximately the original RHS of the linear system as shown in

 A times solution

 +1        1.000000000000000000000000000000
 +2        2.000000000000000000000000000000
 +3        3.000000000000000444089209850063
 +4        4.000000000000000000000000000000
 +5        5.000000000000000000000000000000

If I find time I will check my code with 14.0.1 as well - never saw such a kind of bug before, interesting to see what went wrong when the issue is fixed.

Gruß

Michael

PS: Attached another version to test (if you like) with two of my library procedures (adopted from LinPack). Would be interesting to see if they fail as well ...

Am 29.04.24 um 08:49 schrieb Gerald Kasner:

Hi Michael,

here are the outputs of your test:

gm2 version 13.2.1

      125.000000000000369482222595252097
    -2880.000000000007275957614183425903
    14490.000000000032741809263825416565
   -24640.000000000047293724492192268372
    13230.000000000023646862246096134186

 A times solution

 +1        1.000000000000000000000000000000
 +2        2.000000000000000000000000000000
 +3        3.000000000000000444089209850063
 +4        4.000000000000000000000000000000
 +5        5.000000000000000000000000000000

 A times GM2 14.0.1 result

 +1    -3511.861111111118134431308135390282
 +2    -2721.708333333338941883994266390800
 +3    -2215.725000000004456524038687348366
 +4    -1865.505555555559340064064599573612
 +5    -1609.383333333336622672504745423794

 gm2 version 14.0.1

        1.000000000000000000000000000000
        1.500000000000000000000000000000
        1.166666666666666740681534975010
   -24640.000000000047293724492192268372
    13230.000000000023646862246096134186

 A times solution

  1    -3511.861111111118589178659021854401
  2    -2721.708333333338941883994266390800
  3    -2215.725000000004456524038687348366
  4    -1865.505555555559567437740042805672
  5    -1609.383333333336850046180188655853

 A times GM2 14.0.1 result

  1    -3511.861111111118134431308135390282
  2    -2721.708333333338941883994266390800
  3    -2215.725000000004456524038687348366
  4    -1865.505555555559340064064599573612
  5    -1609.383333333336622672504745423794

Best,

Gerald

Gerald,

may I ask you to run the code attached (amended your code with a short test) - just to check if we are not fooled by faulty IO-Routines.

Don't assume this is the case, but just to be sure.

And is the result "reproducible" if you use another right or left hand side for the linear system ?

Gruß

Michael

PS: Gaius, if needed, I can send you the updated code (and test routines) from my library to check other numerical code.

testgauss3.mod
Description: audio/mod