#include <avr/io.h>
#include <util/delay_basic.h>
#include <stdbool.h>
#include "i2c.h"

#define SCL_BIT PD3
#define SDA_BIT PD2
#define SCL_M   (1 << SCL_BIT)
#define SDA_M   (1 << SDA_BIT)

/* Open drain: PORT bits stay 0, DDR=1 drives low, DDR=0 releases (pull-up). */
#define SDA_LOW()  (DDRD |=  SDA_M)
#define SDA_HIGH() (DDRD &= ~SDA_M)
#define SCL_LOW()  (DDRD |=  SCL_M)
#define SCL_HIGH() (DDRD &= ~SCL_M)

/*
 * Bit timing (16 MHz). Data bits are inlined with cycle-exact pads so that
 * delay 0 gives at least T_BIT cycles per SCL period (400 kHz, Fast mode) with
 * tLOW >= 1.3 us and tHIGH >= 0.6 us. Each delay unit adds 3 cycles to each
 * half period: the period is 40 cycles at delay 0 and 38 + 6 * delay above:
 *   0 = 400 kHz, 5 = 235 kHz, 10 = 163 kHz, 47 = 50 kHz.
 * Reads run somewhat slower (~314 kHz at delay 0); both stay within Fast mode.
 * The pads below are tuned against the generated code (see wr()/rd()); the
 * SCL edge is followed by a read-back, so slow rise or clock stretching only
 * ever lengthens the high phase.
 */
static uint8_t dly = 0;
static bool    bus_err;

/* START/STOP/recovery only: not on the per-bit path, >= 0.8 us at delay 0 */
static void d(void) { if (dly) _delay_loop_1(dly); }

/* 3 cycles per unit; n stays in a register for the whole byte */
static inline __attribute__((always_inline)) void xdelay(uint8_t n)
{
    if (n)
        __asm__ volatile ("1: dec %0\n\tbrne 1b" : "+r" (n));
}

void i2c_set_delay(uint8_t v) { dly = v; }

void i2c_init(void)
{
    PORTD &= ~(SCL_M | SDA_M);
    DDRD  &= ~(SCL_M | SDA_M);
}

static void __attribute__((noinline)) scl_wait(void)    /* stretched or slow rise */
{
    uint16_t t = 0;              /* ~16 ms max */
    while (!(PIND & SCL_M)) if (--t == 0) { bus_err = true; break; }
}

/*
 * Release SCL and honour clock stretching. The input synchronizer needs a
 * cycle before PIND shows the released pin, hence the nop: without it the
 * check always failed and every bit took the slow path.
 */
static inline __attribute__((always_inline)) void scl_up(void)
{
    SCL_HIGH();
    __asm__ volatile ("nop");
    if (!(PIND & SCL_M)) scl_wait();
}

static void scl_release(void) { scl_up(); }

/*
 * One SCL pulse for a data or ACK bit, SDA already set up. Returns the SDA
 * level sampled at the end of the high phase. PAD_LOW/PAD_HIGH make up the
 * difference between the instruction overhead and the timing targets.
 */
/*
 * Counted on avr-gcc 15 -Os at delay 0, sbi/cbi = 2 cycles, from the SCL
 * edge to the next one. wr(): data low 13 (bit 1) / 16 (bit 0), ACK low 9 +
 * ACK_PAD, high 10. rd(): see there. Recount after compiler or code changes.
 */
#define PAD_LOW   12u   /* data low 25 / 28, ACK low 24 */
#define PAD_HIGH  6u    /* high 16 */
#define ACK_PAD   3u
static inline __attribute__((always_inline)) uint8_t clock_bit(uint8_t n)
{
    __builtin_avr_delay_cycles(PAD_LOW);
    xdelay(n);
    scl_up();
    __builtin_avr_delay_cycles(PAD_HIGH);
    xdelay(n);
    uint8_t sda = PIND & SDA_M;
    SCL_LOW();
    return sda;
}

static void start(void)
{
    SDA_HIGH(); d();
    scl_release(); d();
    SDA_LOW(); d();
    SCL_LOW();
}

static void stop(void)
{
    SDA_LOW(); d();
    scl_release(); d();
    SDA_HIGH(); d();
}

static bool wr(uint8_t b)        /* returns true on ACK */
{
    uint8_t n = dly;

    for (uint8_t i = 0; i < 8; i++) {
        if (b & 0x80) SDA_HIGH(); else SDA_LOW();
        b <<= 1;
        clock_bit(n);
    }
    SDA_HIGH();
    __builtin_avr_delay_cycles(ACK_PAD);    /* this path is shorter than a data bit */
    return !clock_bit(n);
}

static uint8_t rd(bool ack)
{
    uint8_t n = dly, v = 0;

    SDA_HIGH();
    for (uint8_t i = 0; i < 8; i++)
        v = (v << 1) | (clock_bit(n) ? 1 : 0);
    if (ack) SDA_LOW(); else SDA_HIGH();
    clock_bit(n);
    SDA_HIGH();
    return v;
}

/* Diagnostic: optionally enable weak internal pull-ups, return bit0=SCL, bit1=SDA levels */
uint8_t i2c_lines(uint8_t pullup)
{
    if (pullup) PORTD |= (SCL_M | SDA_M); else PORTD &= ~(SCL_M | SDA_M);
    DDRD &= ~(SCL_M | SDA_M);
    d(); d();
    uint8_t p = PIND;
    return ((p & SCL_M) ? 1 : 0) | ((p & SDA_M) ? 2 : 0);
}

uint8_t i2c_recover(void)
{
    bus_err = false;
    for (uint8_t i = 0; i < 9 && !(PIND & SDA_M); i++) {
        SCL_LOW(); d(); d();
        scl_release(); d(); d();
    }
    stop();
    return (bus_err || !(PIND & SDA_M) || !(PIND & SCL_M)) ? I2C_BUS_ERR : I2C_OK;
}

/* ---- split-phase API (one transaction spread over several USB requests) ---- */
uint8_t i2c_start(uint8_t addr_rw, bool repeated)
{
    if (!repeated) {
        bus_err = false;
        if (!(PIND & SDA_M) || !(PIND & SCL_M)) {
            if (i2c_recover() != I2C_OK) return I2C_BUS_ERR;
        }
    }
    start();
    bool ack = wr(addr_rw);
    if (bus_err) return I2C_BUS_ERR;
    return ack ? I2C_OK : I2C_NACK_ADDR;
}

uint8_t i2c_write(uint8_t b)
{
    bool ack = wr(b);
    if (bus_err) return I2C_BUS_ERR;
    return ack ? I2C_OK : I2C_NACK_DATA;
}

uint8_t i2c_read(bool ack)     { return rd(ack); }
bool    i2c_bus_error(void)    { return bus_err; }
void    i2c_stop(void)         { stop(); }

uint8_t i2c_xfer(uint8_t addr, const uint8_t *w, uint8_t wlen, uint8_t *r, uint8_t rlen)
{
    uint8_t st = I2C_OK;
    bus_err = false;

    if (!(PIND & SDA_M) || !(PIND & SCL_M)) {
        if (i2c_recover() != I2C_OK) return I2C_BUS_ERR;
    }

    if (wlen || !rlen) {
        start();
        if (!wr(addr << 1)) { st = I2C_NACK_ADDR; goto done; }
        for (uint8_t i = 0; i < wlen; i++)
            if (!wr(w[i])) { st = I2C_NACK_DATA; goto done; }
    }
    if (rlen) {
        start();                               /* (repeated) START */
        if (!wr((addr << 1) | 1)) { st = I2C_NACK_ADDR; goto done; }
        for (uint8_t i = 0; i < rlen; i++)
            r[i] = rd(i < rlen - 1);
    }
done:
    stop();
    return bus_err ? I2C_BUS_ERR : st;
}