test-fsytrf: ERROR: A != P^T U^T D U P

[d-torrance]

I got the following during a PPA build of the master branch on amd64 in Ubuntu 20.10:

FAIL: test-fsytrf
=================

with seed = 589378027
Checking Modular_implem<float, float, uint32_t> modulo 2693 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<float, float, uint32_t> modulo 1493 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<float, float, uint32_t> modulo 839 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<float, float, uint32_t> modulo 4027 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<float, float, uint32_t> modulo 1399 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<float, float, uint32_t> modulo 3433 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<double, double, uint64_t> modulo 28722047 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<double, double, uint64_t> modulo 22497637 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<double, double, uint64_t> modulo 94102597 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<double, double, uint64_t> modulo 58002821 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<double, double, uint64_t> modulo 21295159 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking Modular_implem<double, double, uint64_t> modulo 57182443 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..PASSED 
Checking ..........ModularBalanced<float> mod 677 ... GRP..Up...Low...RPM..Random dim...GRP..Up...Low...RPM..ERROR: A != P^T U^T D U P
PTUTDUP =
   0    0    0    0    0    0    0    0    0    0    0  262  135  188  -21  122  181  132   64  207  315  -90  192  -86  208  169   43 -119  233   75  -55  161  186 -136  163 -292   57  230 -103 -250  -31  -52  274  208  -41  162  -71  124  -57 -307 -125  232  109  184   61  -88
... [snipping large matrix] ...

[ClementPernet]

Can't reproduce, neither on my Linux Mint 20 with g++-10, nor on a Ubuntu Xenial with g++-8. I tried using the exact same compilation arguments passed to g++ and running with the seed: 589378027 reproduces the same sequence of modulos: the run for ModularBalanced mod 677 passes fine on both instances. I don't know what's going on here!