Renamed cast member of the constants struct to magic to clearly identify
magic numbers.
Introduced nlconst struct inheriting from plib::constants<nl_fptype> to
make code better understandable.
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.
Added ability to compile using float instead of double. Specifically the
the solver as well as the infrastructure now can have their own floating
point type. Currently this is only an academic exercise since
numerically demanding circuits like kidniki only work with double/double
support. Using float here is pushing numerical stability over the
limits.
The long term design goal is too have the matrix type (double/float)
being a parameter.
- 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
The builtin preprocessor now behaves closer to cpp:
- supports macro parameters, i.e. define x(a) a
- supports stringification, i.e. define x(a) #a
- supports concatenation, i.e. define x(a) a ## _ext
In addition, error reporting now provides a source context including the
include history.
- Added support for line markers to the preprocessor and parser.
- Added support for include processing to the preprocessor.
- Moved sources base type to plib to be used for preprocessor includes.
This enables to include e.g. from rom memory regions.
- Renamed some defines
- moved netlists out of driver code into audio/ or machine/ as
nl_xxx.cpp files.
- identified and documented extended validation
- updated arcade, mess and nl targets
- converted NL_MAX_LINK_RESOLVE_LOOPS into a netlist parameter.
- Reduced potential bit-rot.
- nltool -v --version now displays values of all compile time defines.
There are still far too many compile time defines. However, most of them
ensure and test future scalability.
Runtime performance statistics can now be enabled with nltool
option "-s". To enable those with MAME you need to run
NL_STATS=1 ./mamenl64 -v -oslog game
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.
Memory management in plib is now alignment-aware. All allocations
respect c++11 alignas. Selected classes like parray and aligned_vector
also provide hints (__builtin_assume_aligned) to g++ and clang.
The alignment optimizations have little impact on the current use cases.
They only become effective on bigger data processing.
What has a measurable impact is memory pooling. This speeds up netlist
games like breakout and pong by about 5%.
Tested with linux, macosx and windows cross builds. All features are
disabled since I can not rule out they may temporarily break more exotic
builds.
Set USE_MEMPOOL to 1 to try this (max 5% performance increase).
For mingw, there is no alignment support. This triggers -Wattribute
errors which due to -Werror crash the build.
- 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.
- Removed dead code.
- nltool now adds a define NLTOOL_VERSION. This can be tested in
netlists. It is used in kidniki to ensure I stop committing
debug parameters.
- Optimized the proposal for no-deactivate hints.
- Documented in breakout that hints were manually optimized.
- Minor optimizations in the order of 2% enhancement.
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.
