aquiles viza
06/02/2024, 6:52 AMtests
directory. The regression ones are fine but the rest of them (bsim3, jfet, mos6) fails or passes with a lot of warnings.
I've used the compiled ngspice to run simulations of sky130 and ihp pdks and it works fine, it gives the expected result.
The evaluation system runs Ubuntu 24.04 over a docker container on windows.
This are the summarized results for the non-regression tests:
make -C tests/bsim1 check # failed 1/1
make -C tests/bsim2 check # failed 1/1
make -C tests/bsim3 check # Pass, most of checks out of the ranges
make -C tests/bsim3soidd check # failed 2/5
make -C tests/bsim3soifd check # failed 3/5
make -C tests/bsim3soipd check # failed 2/5
make -C tests/bsim4 check # Pass, most of checks out of the ranges
make -C tests/bsimbulk check # No rule to make target "check"
make -C tests/bsimsoi check # Pass, most of checks out of the ranges
make -C tests/filters check # failed 1/1
make -C tests/general check # failed 8/8
make -C tests/hfet check # failed 2/2
make -C tests/hicum2 check # Pass, most of checks out of the ranges
make -C tests/hisim check # Pass, most of checks out of the ranges
make -C tests/hisimhv1 check # Pass, most of checks out of the ranges
make -C tests/hisimhv2 check # Pass, most of checks out of the ranges
make -C tests/jfet check # failed 1/1
make -C tests/mes check # failed 1/1
make -C tests/mesa check # failed 7/9
make -C tests/mos6 check # failed 2/2
make -C tests/polezero check # failed 6/6
make -C tests/resistance check # failed 3/3
make -C tests/sensitivity check # failed 1/1
make -C tests/transient check # failed 1/1
make -C tests/transmission check # failed 4/6
make -C tests/vbic check # failed 4/6
make -C tests/xspice check # Pass
It seems that the tests with ****** Checking test (ngspice):
always passes, despite the errors.
Some of the maximum relative error:
• standard: 99%
• m: 4.88%
• Flip_N: 93.9%
On the regression tests some of them writes warnings and errors, but all pass.
lib-processing:
• On `scope-3.cir`:
Warning: Nesting of subcircuits with parameters is only marginally supported!
misc:
• On `asrc-tc-1.cir`: Writes some warnings and errors.
Warning from checkvalid: vector v2#branch is not available or has zero length.
Error: RHS "i(v2)" invalid
Error: no such vector v(3)
Error: Can't evaluate "v(3)"
Warning from checkvalid: vector iac2 is not available or has zero length.
Error: RHS "iac2/iac2_gold - 1" invalid
Warning from checkvalid: vector vac3 is not available or has zero length.
Error: RHS "vac3/vac3_gold - 1" invalid
Warning from checkvalid: vector err3 is not available or has zero length.
• On `asrc-tc-2.cir`: Writes some of warnings and errors.
Warning from checkvalid: vector v2#branch is not available or has zero length.
Error: RHS "i(v2)" invalid
Warning from checkvalid: vector v3#branch is not available or has zero length.
Error: RHS "i(v3)" invalid
Warning from checkvalid: vector iac2 is not available or has zero length.
Error: RHS "iac2/iac_gold - 1" invalid
Warning from checkvalid: vector iac3 is not available or has zero length.
Error: RHS "iac3/iac_gold - 1" invalid
Warning from checkvalid: vector err3 is not available or has zero length.
pipe:
• on `alter-1.cmd`: Seems to require X-server, despite of this the test doesn't increment the fail_count variable so I think this is fine
Does this indicates that the ngspice binary is wrong? Some of the tests that are not on the regression/
directory have more than 10 years without modifications. Maybe they are not valid on ngspice-42.
Since all regression tests pass. Should I care about the warnings and errors on misc
, lib-processing
and pipe
?
Thanks for the help
@Mitch Bailey The tip of using make check
on the build directory worked.
Is Holger Vogt on this slack?