speeduino/speeduino/crankMaths.cpp

86 lines
3.9 KiB
C++
Raw Permalink Normal View History

2018-07-19 06:26:31 -07:00
#include "globals.h"
#include "crankMaths.h"
#include "bit_shifts.h"
2018-07-19 06:26:31 -07:00
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
#define SECOND_DERIV_ENABLED 0
//These are only part of the experimental 2nd deriv calcs
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
#if SECOND_DERIV_ENABLED!=0
byte deltaToothCount = 0; //The last tooth that was used with the deltaV calc
int rpmDelta;
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
#endif
typedef uint32_t UQ24X8_t;
static constexpr uint8_t UQ24X8_Shift = 8U;
/** @brief uS per degree at current RPM in UQ24.8 fixed point */
static UQ24X8_t microsPerDegree;
static constexpr uint8_t microsPerDegree_Shift = UQ24X8_Shift;
typedef uint16_t UQ1X15_t;
static constexpr uint8_t UQ1X15_Shift = 15U;
/** @brief Degrees per uS in UQ1.15 fixed point.
*
* Ranges from 8 (0.000246) at MIN_RPM to 3542 (0.108) at MAX_RPM
*/
static UQ1X15_t degreesPerMicro;
static constexpr uint8_t degreesPerMicro_Shift = UQ1X15_Shift;
void setAngleConverterRevolutionTime(uint32_t revolutionTime) {
microsPerDegree = div360(lshift<microsPerDegree_Shift>(revolutionTime));
degreesPerMicro = (uint16_t)UDIV_ROUND_CLOSEST(lshift<degreesPerMicro_Shift>(UINT32_C(360)), revolutionTime, uint32_t);
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
}
uint32_t angleToTimeMicroSecPerDegree(uint16_t angle) {
UQ24X8_t micros = (uint32_t)angle * (uint32_t)microsPerDegree;
return rshift_round<microsPerDegree_Shift>(micros);
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
}
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
uint16_t timeToAngleDegPerMicroSec(uint32_t time) {
uint32_t degFixed = time * (uint32_t)degreesPerMicro;
return rshift_round<degreesPerMicro_Shift>(degFixed);
}
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
#if SECOND_DERIV_ENABLED!=0
2022-11-05 15:43:29 -07:00
void doCrankSpeedCalcs(void)
{
//********************************************************
//How fast are we going? Need to know how long (uS) it will take to get from one tooth to the next. We then use that to estimate how far we are between the last tooth and the next one
//We use a 1st Deriv acceleration prediction, but only when there is an even spacing between primary sensor teeth
//Any decoder that has uneven spacing has its triggerToothAngle set to 0
2018-09-03 22:13:14 -07:00
//THIS IS CURRENTLY DISABLED FOR ALL DECODERS! It needs more work.
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
if( (BIT_CHECK(decoderState, BIT_DECODER_2ND_DERIV)) && (toothHistoryIndex >= 3) && (currentStatus.RPM < 2000) ) //toothHistoryIndex must be greater than or equal to 3 as we need the last 3 entries. Currently this mode only runs below 3000 rpm
{
//Only recalculate deltaV if the tooth has changed since last time (DeltaV stays the same until the next tooth)
//if (deltaToothCount != toothCurrentCount)
{
deltaToothCount = toothCurrentCount;
int angle1, angle2; //These represent the crank angles that are travelled for the last 2 pulses
if(configPage4.TrigPattern == 4)
{
//Special case for 70/110 pattern on 4g63
angle2 = triggerToothAngle; //Angle 2 is the most recent
if (angle2 == 70) { angle1 = 110; }
else { angle1 = 70; }
}
else if(configPage4.TrigPattern == 0)
{
//Special case for missing tooth decoder where the missing tooth was one of the last 2 seen
if(toothCurrentCount == 1) { angle2 = 2*triggerToothAngle; angle1 = triggerToothAngle; }
else if(toothCurrentCount == 2) { angle1 = 2*triggerToothAngle; angle2 = triggerToothAngle; }
else { angle1 = triggerToothAngle; angle2 = triggerToothAngle; }
}
else { angle1 = triggerToothAngle; angle2 = triggerToothAngle; }
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
uint32_t toothDeltaV = (MICROS_PER_SEC * angle2 / toothHistory[toothHistoryIndex]) - (MICROS_PER_SEC * angle1 / toothHistory[toothHistoryIndex-1]);
2019-12-01 19:47:30 -08:00
uint32_t toothDeltaT = toothHistory[toothHistoryIndex];
//long timeToLastTooth = micros() - toothLastToothTime;
rpmDelta = lshift<10>(toothDeltaV) / (6 * toothDeltaT);
}
timePerDegreex16 = ldiv( 2666656L, currentStatus.RPM + rpmDelta).quot; //This gives accuracy down to 0.1 of a degree and can provide noticeably better timing results on low resolution triggers
}
}
Performance: optimize division (#1082) * Add udiv_32_16 * Apply udiv_32_16() where possible * Convert udiv_32_16 to assembler It's worth 20 loop/s * Remove unused functions * Remove degreesPeruSx2048 - unused * Remove angleToTime - replace with direct calls 1. Drop angleToTime() It's slow, only partially implemented and adds zero value (and has MISRA violations) 2. Consistent function naming 3. Doxygen * triggerPri_Nissan360 shouldn't set timePerDegree. It will be overwritten every loop by doCrankSpeedCalcs() * Use angleToTimeMicroSecPerDegree() instead of timePerDegree No loss in performance Increased injection open/close time accuracy (so unit test values must change) Can remove timePerDegree global. * Hide (encapsulate) crank math globals. * Base all angle to time conversions on decoder computed variables. This is within 2us of the revolution based method and is much faster - which is essentially zero percent change. * Performance: move calculation of degreesPeruSx32768 into decoders. Remove doCrankSpeedCalcs() - it's doing nothing at the moment. * Apply libdivide to triggerSetEndTeeth functions. Since triggerToothAngle is set once at initialization time, we can generate the libdivide struct once and reuse it many times. * Remove lastToothCalcAdvance - unused * Replace 16-bit division with shift * Replace 32-bit divison with 16-bit division * Avoid 32-bit division; use div100() * inline percentage() * Optimize div100() * MISRA fixes * Replace magic numbers with #defs * Replace libdivide structs with inline constants No perf or memory changes * Use fixed types for PWM max count variables * Accurate rounded integer division * Formalise rounding behavior (DIV_ROUND_CORRECT) * Apply DIV_ROUND_CORRECT to DIV_ROUND_CLOSEST(), UDIV_ROUND_CLOSEST(), div100(), div360(), percentage() & halfPercentage() * Add, fix & improve unit tests * Add udiv_32_16_closest() * Perf: Limit percentage calculations to 16-bits * MISRA fixes * Add compare_executiontime() to encapsulate common perf testing code * Signed to unsigned division * Convert ignitionLimits() to an inline function. Slight speed up, probably due to removing multiple evaluations of macro arguments. * Split unit tests up. * udiv_32_16 - check for valid parameters
2023-11-05 14:10:08 -08:00
#endif