OPAMP: Generic opamp model. This does all the annoying calculations.
Just pass the the datasheet values.
LVCCS: A limited current voltage control current source. This will allow
slew rate limiting going forward.
In addition:
- add a (small) parallel conductance to all capacitors to improve
convergence.
- some initial work to use "long double".
boils down to a 87x87 matrix. This is due to a total of 6 opamps which
all are submodels and thus add their own internal nets.
Gauss Seidel iterative solving comes to it's limits.
nltool runs this at about 50% speed on my machine. Given the complexity
this is quite good. Yet, any m62 game currently will not be playable.
Time for a new cpu :-)
[Andrew Gardner, Couriersud]
NETLIST_START(lib)
TRUTHTABLE_START(TTL_7400A_NAND, 2, 1, 0, "+A,B")
TT_HEAD(" A , B | Q ")
TT_LINE(" 0 , X | 1 |22")
TT_LINE(" X , 0 | 1 |22")
TT_LINE(" 1 , 1 | 0 |15")
TRUTHTABLE_END()
NETLIST_END()
This enables the addition of devices without changing the netlist
source code and allows the creation of libraries. Used pong.c as a
proof of concept for the time being. [Couriersud]
This is a first step to ease synchronisation with a stand alone, e.g.
outside mame, netlist implementation. More signed/unsigned cleanups and
started work on generic truthtable devices. (nw)
- Pong Doubles now working
Thanks a lot to Adam Bousley for dual licensing DICE netlists.
Hooked up player inputs
- Minor performance tweaks
- Separated game netlists from driver. Dice is making extensive use of macros which may conflict
if these are in one source file.
- Added some rescap.h macros to netlist
New games added or promoted from NOT_WORKING status
---------------------------------------------------
Pong Doubles [DICE Team, Couriersud]
- fixed a bug in which the fallback solver was called instead of gauss - seidel.
- matrix solvers are now subdevices of the solver devices
- matrix solvers can now be scheduled independently
- Rename RESCHED_LOOPS to GS_LOOPS (Gauss Seidel)
- Added paragmenter NR_LOOPS (Newton Raphson)
The implementation also supports nesting.
Opamp emulation now is as easy as
/* Opamp wired as impedance changer */
SUBMODEL(op, opamp)
NET_C(op.GND, GND)
NET_C(op.PLUS, clk)
NET_C(op.MINUS, op.OUT)
The missing bit now is a central submodel repository. I'll start a discussion soon on the list.
nl_examples/opamp.c is an example of a impedance changer stage followed by a 1:2 amplifier stage.
System size (= number of voltage levels to be calculated) is between 20 - 30.
Using a simple, yet better opamp model than usually implemented in the old discrete core, I get
./nltool -f nl_examples/opamp.c -t 30
startup time ==> 0.002
runnning ...
30.000000 seconds emulation took 0.438599 real time ==> 6839.96%
Which leaves quite some buffer to emulate even complex mixing and filtering stages in real-time :-)
- Working Ebers Moll model. That's a significant step ahead.
- Simple 2x2|RHS SPICE stamps now supported by two terminal devices.
This was implicitly contained in the design, but set_mat now shows how
a two-terminal device fits into a SPICE approach.
- Introduced direct solvers for net groups with 1 or 2 nets.
- Introduced specialized solvers for N=1,2,3,4,5 by using templates.
- nice performance increase for BJTs
The examples have a ".c" suffix. In eclipse, I get automatic syntax parsing and error notifications. The parser treats "#" preprocessor defines/includes just as comments.
All of these examples can be run through nltool:
./nltool -f nl_examples/opamp.c -t 1 -l OUT
runs the opamp example for 1 second of emulation time and logs the terminal named "OUT" to "netlist_log_OUT.log".
I'll post a simple script to the list to visualize those logs using gnuplot.
