Dear dac & All,
On Mon, Oct 25, 2021 at 02:56:50PM -0600, David A Case wrote:
> On Mon, Oct 25, 2021, Thomas Zeiser wrote:
> >
> > gcc-11.2.0 fails to compile src/pmemd/src/cuda/gpu.cpp line 2852
> > with "error: ordered comparison of pointer with integer zero".
> >
> > That line 2852 also look suspicious:
> > if (nmrat[i][0] >= 0 && nmrat[i][1] >= 0 && nmrat[i][2] >= 0 && nmrat[3] >= 0) {
>
> It seems quite clear that one should replace "nmrat[3]" above with
> "nmrat[i][3]". Can you give that a try?
with the change, it does compile successfully using gcc-11.2 (as to
be expected).
When running ./test_amber_cuda_serial.sh on an Nvidia A40 GPU the
change does not have any significant effect; the failed test remain
the same - but with slightly different numbers.
With the original sources, compiled with system gcc-8.4.1 on an
RHEL8 clone running on AMD Milan CPUs and using CUDA-11.5 on Nvida
A40 with diver 495.29.05, 243 file comparisons passed 7 file
comparisons failed (2 of which can be ignored).
With the modified lince, compiled with SPACK-built gcc-11.2.0 on an
RHEL8 clone running on AMD Milan CPUs and using CUDA-11.5 on Nvidia
A40 with driver 495.29.05, again 243 file comparisons passed and
the same 7 file comparisons failed (2 of which can be ignored).
irest1_ntt0_igb7_ntc2.out.dif slightly changed, "Maximum absolute
error in matching lines = 6.12e+00 at line 132 field 3" is the same
in both cases; "Maximum relative error in matching lines = 4.05e-01
at line 168 field 3" changed to "4.01e-01".
irest1_ntt0_igb8_ntc2.out.dif:
Maximum absolute error in matching lines = 4.93e+00 at line 141
field 3 // 4.90e+00
Maximum relative error in matching lines = 3.07e-01 at line 177
field 3 // 3.05e-01
myoglobin_md_igb7.out.dif:
Maximum absolute error in matching lines = 7.80e+02 at line 62
field 3 // 8.02e+02
Maximum relative error in matching lines = 9.64e-01 at line 86
field 3 // 9.91e-01
myoglobin_md_igb8.out.dif:
Maximum absolute error in matching lines = 6.55e+02 at line 71
field 3 // 6.75e+02
Maximum relative error in matching lines = 6.51e-01 at line 201
field 4 // 6.02e-01
myoglobin_md_igb8_gbsa.out.dif:
Maximum absolute error in matching lines = 6.55e+02 at line 113
field 3 // 6.75e+02
Maximum relative error in matching lines = 9.67e-01 at line 154
field 2 // 9.41e-01
(ignored) check myoglobin_md_igb8_gbsa3.out.dif
Maximum absolute error in matching lines = 6.55e+02 at line 113
field 3 // 6.75e+02
Maximum relative error in matching lines = 7.22e-01 at line 152
field 4 // 6.65e-01
(ignored) check mdout.mcwat.dif:
Maximum absolute error in matching lines = 2.12e-02 at line 150
field 3 // 2.12e-02
Maximum relative error in matching lines = 5.64e-02 at line 100
field 7 // 5.64e-02
Best regards
thomas
> Thanks for the report.....dac
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
--
Dr.-Ing. Thomas Zeiser, Head of HPC Systems & Services
Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU)
Zentrum für Nationales Hochleistungsrechnen Erlangen (NHR.FAU)
c/o Regionales Rechenzentrum Erlangen (RRZE)
Martensstraße 1, 91058 Erlangen, Germany
https://hpc.fau.de/ & https://www.rrze.de/
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Oct 25 2021 - 23:30:02 PDT
