目前是初版，基本功能有了。後續還會再加入主要函式 sync function 等。
Based on MultiTimers。

// Esp8266MultiPwms ver.0.1
// https://waterfalls.ddns.net
// by Ken Woo
// 2020.12.03

// PWM class: using MultiTimersV0.4 to generate four 192us timers which are sequentially sync-offset by 48us to one another.
// which are approaching to and are regarded as 200us/50us. Such a facility forms a single 50us timer with 4 sequential ISRs.
// PWM waveforms whereas ought to be transition of the processes will be evenly settled in the 4 ISRs,
// unless there are specified ones to be explicitly offset or synced to another, which determines the position.
// note that explicitly sync offset is restricted only to same period PWM.
// duty cycle 70% means headed high 70% first then low, inverted duty cycle 70% means headed low 70% then high.
// sync or offset counts for headed beginning.
// so, PWM periods are restricted to multiples of 200us(and at least 200us, at most 0.8s; in order to have fixed positions),
// in addition, the duty cycle step must be multiple of 50us, is restricted too.
// for example, 2.2ms PWM with 45 steps is allowed; 2200/200=11 is integer, 2200/44/50=1 is integer,
// step0 0%, step1 <=2.27%(100%/44), step2(<=100%*2/44)..., step44(>100%*43/44) 100%.
// or with 12 steps, each step is 200us, and the like.
// the highest one is 200us/5kHz with 5 steps(0%, 0%< <=25%, 25%< <=50%, 50%< <=75%, 75%< <=100%).
// the number of PWMs depends on bits of id; you can rewrite it for unlimited PWMs theoretically.


#include"Esp8266HwSwTimers.h"

unsigned z1, z2, z3, z4;

class cMultiPwm{

    typedef struct sPwmObj{
        unsigned counter:       12; // counter, reload to it.
        unsigned rsv1_not_use:  20; // reserved 1. it should be 0 and do not use.

        unsigned rsv2:          20; // reserved 2.
        unsigned reload_high:   12; // reload count for high level. 200us is one-round, 200x4096=0.8s.

        unsigned freeze:        1;  // will not access this obj.
        unsigned stopit:        1;  // will delete this obj.
        unsigned accepted:      1;  // this bit will be set if isr accepted freeze or stopit.
        unsigned is_fixed_pos:  1;  // is it a fixed position by a synced PWM; the whom synced must fixed too.
        unsigned is_inverted:   1;  // we use it at the final waveform, so it affects nothing.
        unsigned is_high_level: 1;  // the current level counted is high? it will be toggling.
        unsigned stop_h_or_l:   1;  // high or low when stopped.
        unsigned tr_high_pos:   2;  // the position at timer ISR[] for transiting to high level, it exists with low-counting.
        unsigned tr_low_pos:    2;  // the position at timer ISR[] for transiting to low level, coexists with high-counting.
        unsigned gpio:          4;  // the gpio pin number
        unsigned id:            5;  // id in order to search. for unlimited new/delete, must maintain it. but i prefer not.
        unsigned reload_low:    12; // reload count for low level. 200us is one-round, 200x4096=0.8s.
        // note the relationship of is_high_level, tr_high_pos, tr_low_pos.
        // is_high_level indicats currently is high counting or low. tr_high_pos: currently is low level counting,
        // it is going to transit to high level, so it is positioned at tr_high_pos, and vice versa.
        // so, the duty-cycle count allocates on reload_high, and when which positioning at tr_high_pos, is active, it is counting low,
        // when count-up, loading the reload_high to counter, transits the level from low to high, and finally hands over the control
        // to tr_low_pos for it active. after a while the count is up again, it is responsible for transiting from high to low,
        // then reload reload_low, and then changing position to tr_high_pos again alternatively.

        sPwmObj(){
            (reinterpret_cast<unsigned*>(this))[0]=0;
            (reinterpret_cast<unsigned*>(this))[1]=0;
            (reinterpret_cast<unsigned*>(this))[2]=0;
        }
    } sPwmObj;

    typedef struct sNode{
        sPwmObj data;
        sNode *next;

        sNode(): next(0){};
    } sNode;

    IRAM_ATTR static sNode** node_get_conn_pt(sNode **head){ // get the tail-next address so we can store data in it.
        sNode *a=*head;
        while (a){
            head=&(a->next);
            a=a->next;
        }
        return head;
    };

    int node_get_length(sNode *head){ // evaluate the linkedlist length
        int i=0;
        for (; head; i++, head=head->next);
        return i;
    };

    void node_add(sNode **head, sNode *a){ // attach a node to the tail
        while (*head) head=&((*head)->next);
        *head=a;
    };

    void node_delete(sNode *head){ // delete entire linkedlist
        for (sNode *i; head; i=head->next, delete head, head=i);
    };

    IRAM_ATTR static sNode** node_pwm_dec_cnt(sNode **head){ // possibly process several times by caller.
        sNode *a=*head;
        for (; a && (a->data).counter>1; --(a->data).counter, head=&(a->next), a=a->next); // 1 is minimum.
        if (a) return head;
        return 0;
    };

    sNode** node_pwm_find_parent(unsigned id, unsigned pos){ // find the parent of the node having this id.
        sNode **i;
        for (i=&(pwms[pos]); *i && ((*i)->data).id!=id; i=&((*i)->next));
        if (!*i) return 0;
        return i;
    };

    IRAM_ATTR static void node_pwm_move_to(sNode **src, unsigned new_pos){ // move this src node to attach to pwms[new_pos].
        sNode *obj=*src;
        *src=obj->next;
        *node_get_conn_pt(&(pwms[new_pos]))=obj;
        obj->next=0;
        /// priorly excluded.
        /// sNode **a=node_get_conn_pt(&(pwms[new_pos]));
        /// if (&((*src)->next)!=&((*a)->next)){*a=*src;*src=(*src)->next;(*a)->next=0;}
    };

    unsigned wellAdd(sNode *a){ // add to proper ISR position, return the position.
        unsigned i=0, j=0, pos=0, min=-1;
        for (; i<4; i++){
            if (pwms[i]){
                if ((j=node_get_length(pwms[i]))<min){
                    pos=i;
                    min=j;
                }
            }
            else {pos=i; break;} // empty
        }
        node_add(&(pwms[pos]), a);
        return pos;
    };

    void configTimer(){
        timers[0].setTimer(192, cMultiPwm::timerISR0);
        timers[1].setTimer(192, cMultiPwm::timerISR1);
        timers[2].setTimer(192, cMultiPwm::timerISR2);
        timers[3].setTimer(192, cMultiPwm::timerISR3);
        timers[1].ForceHaltForSync(timers[0], 48);
        timers[2].ForceHaltForSync(timers[1], 48);
        timers[3].ForceHaltForSync(timers[2], 48);
    };

    void* getMemory(int obj_size){ // use sizeof(unsigned) and alignment; little endian.
        int a=(obj_size+sizeof(unsigned)-1)/sizeof(unsigned);
        void *b=malloc(sizeof(unsigned)*a);
        if (b && (unsigned(b)/sizeof(unsigned)*sizeof(unsigned)==unsigned(b))){
            for (; a--; ((unsigned*)b)[a]=0);
            return b;
        }
        free(b);
        return 0;
    };

    void freeMemory(void *b){free(b);};

    static cHwTimer timers[4];
    static sNode* pwms[4];
    static unsigned pause_all; // to avoid conflict
    static unsigned gid;

    sNode *owner;
    unsigned period;
    unsigned steps; // 0% is not counted. however we need it be a step.
    unsigned uid;

    public:

    cMultiPwm(): owner(0), uid(0){};
    ~cMultiPwm(){freeMemory(owner);};

    cMultiPwm(unsigned gpio, unsigned period_us, unsigned set_steps,\
        bool inverted=false, bool stop_high=false): owner(0), uid(0){ // if a step is 1%, use 100 steps, not 101 steps.

        if (period_us%200 || !set_steps || period_us%set_steps || period_us/set_steps%50) return;

        period=period_us;
        steps=set_steps;
        uid=++gid;

        owner=(sNode*)getMemory(sizeof(sNode));
        if (!owner) return;

        (owner->data).id=uid;
        (owner->data).is_inverted=!!inverted;
        (owner->data).is_high_level=0; // at the outset, it is going to high, so it is under low before beginning.
        (owner->data).stop_h_or_l=!!stop_high;
        (owner->data).gpio=gpio;
        (owner->data).freeze=1; // paused
        (owner->data).tr_high_pos=wellAdd(owner); // at the outset, raise high then immed change pos to tr_low_pos.
    };

    bool start(){if (!uid) return false; (owner->data).freeze=0; return true;};
    bool pause(){if (!uid) return false; (owner->data).freeze=1; return true;};
    bool ack(){return (owner->data).accepted;};
    void nack(){(owner->data).accepted=0;};

    void setInvert(bool invert_or_not){
        if (!uid) return;
        (owner->data).is_inverted=!!invert_or_not;
    };

    void setDC(float percentage, unsigned set_by_steps=0){ // will by steps if it is nonzero
        if (!uid) return;
        if (!set_by_steps){
            (percentage*=steps)/=100.0f;
            if ((set_by_steps=unsigned(percentage))<percentage) set_by_steps++;
        }
        else if (set_by_steps>steps) set_by_steps=steps;

        set_by_steps*=period/steps/50; // times the magnifier to get the native count

        // important note that the counting for high, time-up is decided at tr_low_pos, but time starts at tr_high_pos.
        // different position means different time, that is the problem.
        // integer quotient, 0 remainder, means integer rounds starts from pos and ends at pos.
        // remainder if any means the last round that is not a complete round and ends at tr_pos other than pos.
        // so, quotient+1 would be the final count.
        // however one more to consider, if remainder is 0, expected time-up and count-up are exactly matched at the same pos,
        // but what if in such case, we take other position for end? yes, beyond or behind the expected time when count is up.
        // so how to do right the code here(tr_high_pos and tr_low_pos are diff pos, could the COUNT handle it all correctly)?
        // that is right, nothing to do about this problem since tr_low_pos of its position had taken care, think about it.
        // yet the other problem to think about, the boundary condition. when counter becomes 1 which is minimum because
        // we either add 1 if remainder or quotient is nonzero/because 0% is excluded, is for tr_low_pos to make decision,
        // hence the 1 means time is up; the 1 represents time spent from tr_high_pos to tr_low_pos under a round(equal if same pos).
        // however what about the time span for low level at the condition of high level counts is only 1 and high-counts > low-counts?
        // it could happen for example period=200us, step=4, dc=25~75%/count-always-1, low should be 0 by calculation.
        // as a whole if calculation lead to 0 for low level, our decision making uses >1 in node_pwm_dec_cnt could cover each condition,
        // that is, time spent is correct.

        this->pause();
        while (timers[0].isActive() && !this->ack()) delayMicroseconds(25);

        if (!set_by_steps){ // 0%, it paused for isr ignores.
            digitalWrite((owner->data).gpio, LOW);
            return;
        }
        else if (set_by_steps==steps){ // 100%, it paused for isr ignores.
            digitalWrite((owner->data).gpio, HIGH);
            return;
        }

        int pos=(owner->data).tr_high_pos;
        int tr_pos=set_by_steps%4;
        int count=set_by_steps/4;

        if (!tr_pos) tr_pos=pos;
        else {
            count++;
            tr_pos+=pos;
            if (tr_pos>3) tr_pos-=4;
        }

        (owner->data).tr_low_pos=tr_pos; // this is the "duty-cycle count" cast into "tr_low_pos".
        (owner->data).reload_high=count;
        (owner->data).reload_low=period/200-count;
        (owner->data).counter=1; // force to reload for starting.

        this->start();
        this->nack();
        if (!timers[0].isActive()) configTimer();
    };

    static void traverseLinkedList(){
        int c=0;
        for (int i=0; i<4; i++){
            sNode *a=pwms[i];
            while (a){
                printf("level, tr_high_pos, tr_low_pos, gpio, id, rld_high, rld_low, (%d, %d, %d, % 2d, % 2d, %d, %d)\r\n",\
                    (*a).data.is_high_level,\
                    (*a).data.tr_high_pos,\
                    (*a).data.tr_low_pos,\
                    (*a).data.gpio,\
                    (*a).data.id,\
                    (*a).data.reload_high,\
                    (*a).data.reload_low);
                a=a->next;

                c++;
            }
            printf("(%d) ----\r\n", i);
        }
        printf("The 4 ISRs (%d nodes in it) each costs us time (%d %d %d %d)\r\n\r\n", c, z1, z2, z3, z4);
    };

    IRAM_ATTR static void trigger(sNode **u){
        while (u=node_pwm_dec_cnt(u)){
            sPwmObj *v=&((*u)->data);
            if (v->freeze){ // and for stop check
                v->accepted=1;
            }
            else if (v->is_high_level^=1){ // if going to high counting
                digitalWrite(v->gpio, !v->is_inverted);
                v->counter=v->reload_high;
                if (v->tr_high_pos!=v->tr_low_pos)
                    node_pwm_move_to(u, v->tr_low_pos);
            }
            else {
                digitalWrite(v->gpio, v->is_inverted);
                v->counter=v->reload_low;
                if (v->tr_high_pos!=v->tr_low_pos)
                    node_pwm_move_to(u, v->tr_high_pos);
            }
            if (*u) u=&((*u)->next);
        }
    };

    IRAM_ATTR static void timerISR0(){
        static int x;
        x=micros();
        trigger(&(pwms[0]));
        z1=micros()-x;
    };

    IRAM_ATTR static void timerISR1(){
        static int x;
        x=micros();
        trigger(&(pwms[1]));
        z2=micros()-x;
    };

    IRAM_ATTR static void timerISR2(){
        static int x;
        x=micros();
        trigger(&(pwms[2]));
        z3=micros()-x;
    };

    IRAM_ATTR static void timerISR3(){
        static int x;
        x=micros();
        trigger(&(pwms[3]));
        z4=micros()-x;
    };
};


cHwTimer cMultiPwm::timers[4];
IRAM_ATTR cMultiPwm::sNode* cMultiPwm::pwms[4]={0, 0, 0, 0};
unsigned cMultiPwm::pause_all;
unsigned cMultiPwm::gid;


cMultiPwm sss[16]={
    {D2, 200, 4},
    {D3, 4000, 10},
    {D4, 20000, 4},
    {D5, 2000, 8},

    {D6, 120000, 10},
    {D7, 126000, 10},
    {D8, 130000, 10},
    {D1, 136000, 10},

    {D1, 322000, 10},
    {D1, 332000, 10},
    {D1, 342000, 10},
    {D1, 352000, 10},

    {D6, 200, 4},
    {D7, 200, 4},
    {D8, 200, 4},
    {D1, 200, 4},
};


bool Timeout_3s(){
    static int current_time = millis();
    int new_time = millis();
    if (new_time < (current_time + 3000)) return false;
    current_time = new_time;
    return true;
}


void setup() {
    Serial.begin(115200);

    pinMode(D2, OUTPUT);
    pinMode(D3, OUTPUT);
    pinMode(D4, OUTPUT);
    pinMode(D5, OUTPUT);

    delay(5000);


    sss[0].setDC(51);
    sss[1].setDC(51, 2);
    sss[2].setDC(77, 1);
    sss[3].setDC(0, 7);
    
    sss[4].setDC(0, 1);
    sss[5].setDC(0, 2);
    sss[6].setDC(0, 3);
    sss[7].setDC(0, 4);
    
    sss[8].setDC(0, 5);
    sss[9].setDC(0, 6);
    sss[10].setDC(0, 7);
    sss[11].setDC(0, 8);
    
    sss[12].setDC(0, 2);
    sss[13].setDC(0, 1);
    sss[14].setDC(0, 3);
    sss[15].setDC(0, 4);
}


void loop() {

    if (Timeout_3s()){



        cMultiPwm::traverseLinkedList();

        static unsigned y;
        sss[0].setDC(0, y);
        if (++y>4) y=0;




    }
}