Brian Taylor

plot v(3) shows incorrect behavior compared with ngspice 45.2

Looking at delTemp values where DEVlimitlog is called in VDMOSload you can see that those values rapidly become wild. Many delTemp values (degrees Kelvin?) are nonsense eventually becoming nan. I hope this helps fixing the problem in the vdmos code. Some extra checks in com_fft might be useful before: length = (plot_cur->pl_scale)->v_length; time = (plot_cur->pl_scale)->v_realdata;

This test case now gives timestep too small errors then generates a segmentation fault in com_fft. I am using the latest pre-master-46 (dec 29).

The ignore_shared_libs.supp in paranoia_parallel suppresses some memory leaks which it should not. The attached suppressions file does not hide some leaks which are fixed in commit aadb0c. I was able to find the leaks using valgrind 3.22 on Ubuntu 24.04, and also on macos (with a different suppressions file). You may need to modify the suppressions depending on your linux build environment.

There is a version of valgrind available for macos. It is: https://github.com/LouisBrunner/valgrind-macos It needs compiling from source. I have used this successfully with paranoia_parallel tests. You need a non-stripped debug build of ngspice using clang to get accurate results. A minor problem is that call stacks do not show symbols only hex addresses. If anyone is interested, I have a script which decodes the call stacks using lldb.

Russell, I have attached two of the files needed for Uros' motorforce example to run on Windows using the Visual Studio compiler (cl). Your subckt d_process instance has one input and one output. These are single bits, and I can't tell if they match your mcu.exe program. The example Uros mentions has wider inputs and outputs.

Russell, if you can provide your test case, and which OS and ngspice version, that will help me to debug the problem.

I have made a few tweaks to enable CIDER with KLU for OP, DC, and TRAN analyses. AC is not supported. All those years of hard work by Francesco have paid off. See what you think about including this in ngspice-46. commit af4da75f1deb2240b0ab363f1b9b6701149fae63 Author: Brian Taylor lbwnet@comcast.net Date: Wed Sep 3 16:56:36 2025 -0700 Enable CIDER with KLU for DC, OP, and TRAN analyses. Small signal AC analysis is not yet supported for CIDER complex valued device KLU matrices. The examples/cider...