__float128 is a gnu extension delivering true 128 bit floating point
support. Currently not supported by clang. In addition, the quadmath
library needs to be linked. For the time being therefore disabled.
The newly added RELTOL and VNTOL parameters implement Newton convergence
checks comparable following other SPICE implementations.
The ACCURACY solver parameter now is only used for convergence checks in
iterative solvers.
In addition, type safety was significantly improved and a lot of "magic"
numbers are identifiable now.
- Added new solver parameter FPTYPE. This determines in which floating
point domain the linear system is solved. May be one of "FLOAT",
"DOUBLE" or "LONGDOUBLE"
- Added option "--fperr" to nltool. This enables floating point
exceptions. This helps debugging the code under gdb.
The purpose of this going forward is to have more choice in
optimization. Non-dynamic systems should be just fine in the float
domain. Dynamic systems (i.e. diodes, bjts, mosfets) should in general
work with double. Certain edge cases may require long double resolution.
- solver: align matrix population along the various solvers
- solver: delete dead code
- renamed nl_double to nl_fptype and use nl_fptype where previously
double has been used.
- renamed param_double_t to param_fp_t
- Fix SUBMODEL
- move to strongly typed matrix sort constant
- extend maximum matrix size to 512x512
- optionally do parallel processing based on total operations
- templatize GMRES solver loops
This effectively reverts b380514764 and
c24473ddff, restoring the state at
598cd52272.
Before pushing, please check that what you're about to push is sane.
Check your local commit log and ensure there isn't anything out-of-place
before pushing to mainline. When things like this happen, it wastes
everyone's time. I really don't need this in a week when real work™ is
busting my balls and I'm behind where I want to be with preparing for
MAME release.
- convert macros to c++ code.
- order of device creation should not depend on std lib.
- some state saving cleanup.
- added support for clang-tidy to makefile.
- modifications triggered by clang-tidy-9.
Still some work ahead to separate interface from execution. This is a
preparation to switch to another sparse matrix format easily which may
be better suited for parallel processing.
On the linear algebra side there are some nice additions:
- Two additional sort modes: One tries to obtain a upper left identity
matrix, the other prefers a diagonal band matrix structure. Both deliver
slightly better performance than just sorting.
- Parallel execution analysis for Gaussian elimination and LU solve.
This determines which operations may be done independently.
All of this is not really useful right now. The matrix sizes are below
100 nets. I estimate that we at least need four times more so that CPU
parallel processing overhead pays off. For GPU, add another order. But
it's nice to have code which may scale.
This is an effort to separate netlist creation from netlist execution.
The primary target is to avoid that code which will only run during
execution is able to call setup code and thus create ugly hacks.
- more use of c++ features
- some CRTP in pfmtlog
- demangled code for truthtables
- use more constexpr
- rewrite main loop
- use default constructors and assignment operators were applicable.
- optimized 7448 and 9316
All of this has decreased startup time by approx. 25% to 30%. Complex
netlists like pong or kidniki are parsed, analyzed and constructed in
around 15 ms. Run performance has increased by about 5%.
All in all not to bad. A game like pong uses a clock of 7 MHz (after
division by 2). Thats 14 MHz clock invocations. Running at over 200%, 28
MHz. On a 3.9 GHz Machine about 140 cycles/clock change.
[Couriersud]
Fixed -Winconsistent-missing-destructor-override warnings. Made
some constructors of template classes and classes with virtual .. = 0
methods protected. Fixed src/lib/netlist/build/makefile (nw)
- Removed trampolines (OUTLOGIC, INPLOGIC and friends).
- Started using doxygen comment and documentation style. Added doxygen
files to documentation folder.
- Refactored code triggered by doxygen output.
- Moved internal and support classes into namespace detail.
- Use an anordered map in parser.
- -Wconversion fixes - All done now.
- Fixed -Wold-style-cast warnings in netlist code.
- Added iterators to pstring.
- Moved two macros, added more RAII and improved exceptions. Fixed some
bugs in parser code.
- Fixed a number of bugs in parser code and exception handling.
[Couriersud]
- nltool now accepts -Ddefine=value to pass on to netlists
- improved option handling and added "dummy" options to add grouping and
examples in help output.
- improved --cmd=listdevices output
- Fix dynamic timestepping. This will work with breakout using real
capacitor modelling instead of delay devices. Really slow, but very
useful to calibrate timings.
- Fix an awful bug in timing for delay devices.
- Switched to clang 3.8 and made code compile with
-Weverything -Werror -Wno-old-style-cast -Wno-padded -Wno-weak-vtables
-Wno-missing-variable-declarations -Wno-conversion -Wno-c++98-compat
-Wno-float-equal -Wno-cast-align -Wno-global-constructors
-Wno-c++98-compat-pedantic -Wno-exit-time-destructors
-Wno-format-nonliteral -Wno-weak-template-vtables
This was a helpful exercise since it brought forward some
serious issues with implicit constructors.
[Couriersud]
Object model optimisation.
Merge remote-tracking branch 'origin/master' into netlist_dev
Fix a merge issue.
#if ==> #elif. Ouch.
Default PHAS_PMF_INTERNAL=0 for 32bit windows mingw.
Change UINT8 to uint_[fast|least|8_t.
Move state_var so it can be used by base devices as well.
Remove last traces of ATTR_ALIGN.
Refactored netlist_time into a template.
Removed implicit double assignment to netlist. Doomed to produce
bugs.
Instead, use netlist_time::from_double.
Switch to using proper (i.e. bool type) param_logic_t.
Formally differentiate between logic inputs (e.g. switches) and int
inputs (e.g. resistor ladders or selection switches).
Added parameter USE_DEACTIVATE to truthtable devices.
Added more constexpr to netlist_time.
Fixed some ifdef code paths.
- More c++.
- Simplify main processing loop. As a nice side-effect that squeezed
out some cycles.
- More cycle squeezing.
- Removed pvector_t.
- Use std::sort.
- Refactored netlist state manager.
- Introduction of state_var object template to be used on device
state
members.
- Changed remaining save occurrences to state_var.
- Rewrote nltool's listdevices command. This allowed removal of one
member from devices which served solely for listdevices.
- Remove hashmap_t. Fix kidniki regression.
- Dead code removal and minor refactoring.
- Simplify. Align naming with stl. Fix somed pedantic warnings.
- More STL compatability.
- Remove ATTR_HOT and ATTR_COLD. Refactored netlist_time.
- Fix long standing workaround which would ignore policy of change-only"
propagation.
- Rewrote for loops to use auto : semantics.
- Truthtable cleanup. (nw)
- Get rid of nl_math. Remove nl_util.h and moved contents to
plib/putil.h.
- Fix standalone build. Refactor ptypes.h.
[Couriersud]
allow in-place creation (for increased locality) of netlist classes.
Main use is in truthtable class.
- Remove PLIB_NAMESPACE macros.
- Remove namespace macros. Use explicit namespace declarations.
- Moved device definitions into cpp files.
- Moved more device definitions into cpp files.
- New prefix "nlid" for include files flags purely internal include
files not to be leaked into userland.
- Fix factory code.
- added a new solver using compressed row format
- fixed sorting
As a result, netlist performance on kidniki nearly doubled. The
performance increase is mainly due to the fact that sorting decreases
the number of operations for gaussian elimination of the kidniki matrix
from ~7800 to 707. In addition, compressed row format improves L1 usage.
[Couriersud]
