Ignoring parameters with ID==0 when loading

…nto master

Added division support to CAN tx

This adds a simple blocking delay function to assist with
timing sensitive SPI drivers. The delay is fixed for
STM32F103 processors running at 72MHz and no attempt is
made to compensate for different speeds. This should be
OK for all openinverter boards.

Tests:
 - Host in a standalone libopencm3 application and toggle a
   GPIO pin with a uDelay(1000) call.
 - Test at short 10 usec and long 0.5 sec delays and ensure
   that at least the minimum time is achieved.

This adds an 8-bit CRC calculation with the CCITT
polynomial (x^8+x^2+X+1). It uses a 256 byte lookup table
which is pre-calculated using the C pre-preprocessor.
This could have been implemented with constexpr but that
would require c++17 to enable use of std::array in constexpr
functions. Using the pre-processor is more portable for
now.

Code originally comes from a stack overflow answer:
https://stackoverflow.com/a/27843120/6353

Tests:
 - Check against incoming and outgoing logic analyser
   captures of STGAP1AS isolated gate drivers which use this
   algorithm.

Tesla m3 gate driver

Allow strongly typed enum parameters to be stored. We
just cast to an int with no validation.

Added simple averaging to ADC module
Added proportional only function to PI controller

…write to last/second-to-last page

This fixes a compile warning when building with LTO (which
fails for other reasons). The mistmatch doesn't cause
problems in regular builds because the linker doesn't check
types.

a flash page between param_save and can_save.
The module that writes to the (potential) start of the page must be
called first!
PARAM_BLKNUM defines on which block, counting backwards from end of flash
the parameters are saved.
CAN_BLKNUM does the same for CAN.

Update the definition of putchar to match normal Linux glibc
to allow compilation on Linux as well as STM32.

Fix a type problem with character pointer format strings in
the printf implementation to allow compilation on Linux
x86_64.

Tests:
 - Build on Linux x86_64 and STM32
 - Run STM32 build on Bluepill and verify parameter display
   is working as before

Allow the posted messages to be reset to a default state
and check whether an error has been posted. These are
useful when including ErrorMessage in unit tests which
need to ensure the error state is consistent.

Tests:
 - Build on both STM32 and Linux
 - Verify correct operation in several independent unit tests
   that raise errors

Basic CMake build script for the parts of libopeninv needed
to build unit tests.

The FP_TOFLT() macro is the reverse of FP_FROMFLT() and
intended to be used in test code.

Tests:
 - Build as part of expanded unit tests and verify values
   returned in the debugger

Remove all of the hand rolled compile options and instead
pull them from a global_options library from elsewhere in
the build system.

Tests:
 - Build as part of a refactored CMake build system

… is invalid

Expand the CMake build to include a new
libopeninv_stm32f1 library. This pulls in all of the CPP
modules that were missing from the libopeninv library. The
new library is conditional on the "stm32f1" platform being
specified.

CMakeList.txt reformatted with cmake-format.

Tests:
 - Build and link into working STM32 binary

On the C2000 platform an "int" is 16 bits not 32-bits as it is
on ARM and x86_64. This changes fp_ln() to use a type
guaranteed to be 32-bit.

Fixes an error on the C2000 build with:
   error: shift count is too large

params.h/params.cpp:
   C2000 doesn't have an 8-bit type so prefer an at-least 8-
   bit type for flags. This will be 16-bits on C2000 but 8-bits
   on other architectures.

sine_core.h/sine_core.cpp:
   On C2000 an int is 16-bits and constants are ints unless
   otherwise specified. We specify constants as long where
   they need to be 32-bits.

foc.cpp:
  FOC::fpsqrt() mixes signed and unsigned ints which upsets
  the C2000 compiler. Force it to only take signed fixed point
  values.

Using the "register" keyword causes a warning when
building on C++17 Linux host builds. Remove this redundant
use as the optimiser on all our compilers make their own
mind up anyway.

Tests:
 - Build on all platforms and verify no warnings especially on
   x86_64 Linux.

Remove all use of non-length specific int in favour of int32_t in the params class. This changes the return type of IsParam to be bool which matches its use in the rest of the code.

Tests:
 - Build on all platforms and run unit tests on Linux

The modMax constant of 2/sqrt(3) in the FOC calculation
has a signed integer overflow. This results in undefined
behaviour and ends up with a different on C2000/x86_64
compared to ARM. The fix is to evaluate in an external
calulator and store as a floating point constant similar to the
other constants for FOC.

The fpsqrt method was forcing all input values to be
INT32_MAX rather than clamping values above this to
INT32_MAX. Fixed by using MIN() rather than MAX().

Tests:
 - Run against new FOC test suite

Remove the SQRT3 #define in preference to the sqrt3 constant. This removes an unused symbol warning when building for C2000.

Tests:
 - Build for all platforms
 - Verify unit tests pass apart from expected
   TestFocPwmGeneration.NormalModeRun and
   ManulModeRun tests

The foc module increases CST_DIGITS to 15. When using
FP_FROMINT with constants this will cause the C2000 to
experience an integer overflow as constants are int unless
specified. Explicitly specify constants as long to avoid the
overflow. This should cause any issues on 32-bit ARM and
x86_64 systems.

Tests:
 - Build on all platforms. Verify unit tests still pass on Linux.
 - No tests on C2000 yet.

Explicitly specify the size of the integer type used for
calculations in the PiController. This doesn't matter for 32-
bit ARM or x86_64 but does for the C2000 which will opt for
it's default 16-bit int unless told otherwise.

Tests:
 - Run unit tests and verify no failures
 - Run pwm simulator and verify output on C2000 is
   unaffected. It still differs from the Linux build signficantly
  though.

The FP_FROMFLT() macro overflows when used with
CST_DIGITS = 15 in the FOC module. This changes the
FRAC_FAC to use a long constant to prevent overflow.

Tests:
 - Verify in the C2000 debugger that
    FOC::GetMaximumModulationIndex() returns 37837 the
    same as on Linux

…ation

Allow redefining CAN limits in Makefile (max messages etc.)

Mostly automatic merge of the upstream "float" branch.
Some manual merging of params.h due to additional
Set/GetEnum() functions in our branch.

Tests:
 - Run against updated unit tests
 - Compile on all supported platforms

With the C2000 doesn't support uint8_t. Crudely expand
the crc8 implementation to use uint16_t. Mask of out of
range inputs so only the lower 8-bits of the uint16_t ever
contain data.

Tests:
 - Add unit tests in the stm32_sine project to test

…o float

Pull in the latest upstream changes. Johannes has added an FP_TOFLOAT() macro equivalent to the FP_TOFLT() macro. To ease future merges perfer the upstream FP_TOFLOAT() and remove FP_TOFLT().

  - move watchdog pmic driver to inverter lib
  - make a template off it for easier test/mock
  - break out specific platform dependent SPI class
  - add a small scheduler class used to strobe watchdog
  - add a small test program targeting inverter board

 - seperate out lock/unlock sequence, add wait
 - add resetting registers during setup
 - fix DEVCTRL/N inversion
 - add skip result reading for testing purposes
 - remove SPI driver read/write, replace with transfer

Update the c2000 inverter application to use the gate driver.
This involves turning on the gate drive PSU before initialising
the gate drivers for each phase. The fault state of the gate
drivers is reported to stdout every second.

Add a simple heartbeat LED toggle to the main loop to
compensate for the gate drive fault LED no longer being on (!)
as a way of determinig if the inverter is powered up.

Tests:
 - Run on the Tesla M3 inverter and verify it operates with the
   gate drive fault LED extinguished and not gate drive fault is
   reported.

Prior to merging this repo into the parent c2000-inverter repo we move the contents into a sub-directory with the same name.

Follows the guide at https://medium.com/walkme-engineering/how-to-merge-a-git-submodule-into-its-main-repository-d83a215a319c

…rtable-cpp

This moves the main STM32F1 inverter code into a platform
directory so it follows the style and layout of the C2000 port.

STM32 platform specific code remains in libopeninv.

Tests:
 - Build all variants of the CMake build process
 - Build FOC and SINE variants with STM32 Makefile

This change moves all of the STM32 specific parts of libopeninv into the platform/stm32f1 directory. This leaves the libopeninv directory in the top-level to contain only cross-platform code.

No code changes were required only build system updates.

Tests:
 - Build all CMake variants
 - Build FOC and SINE variants using the old stm32_sine
   Makefile

The C2000 inverter erroneously reported "Gate Drive PSU
OFF" when it was in fact turned on. The PSU was in the
desired state it was just the message.

Tests:
 - Run on the Tesla M3 inverter and verify the updated
   message is displayed when running the inverter