Support single precision about cktso HOT 5 CLOSED

This is just the input error. Errors accumulate during LU factorization. Using floats can easily lead to larger errors in the output, or even INF or NAN output. Typically for circuit simulation, we need the accuracy at least 1e-12, or the simulated circuit response may be incorrect. If you application can tolerate large errors, and the matrix is very well-conditioned (condition number is very small), using floats may generate correct LU factorization results. But this is just a special case.

from cktso.

float has much lower precision and smaller value range so it can easily lead to large round-off errors and numerical overflow. float has almost no practical use in sparse solvers.

from cktso.

For a 3-decimal calculation, does it have to be double precision?

from cktso.

No relation with how simple input data are. For a simple example, you can print 1.1 to see what it is. For float, it is 1.1000000238418579.... For double, it is 1.1000000000000001...

from cktso.

To me, the error is absolutely acceptable …

from cktso.