/*
 * ESP32_I2S_PARALLEL_DMA (Version 3)
 * 
 * Author:      Mrfaptastic @ https://github.com/mrfaptastic/
 * 
 * Description: Multi-ESP32 product DMA setup functions for ESP32 C3/H2 RISC-V chips
 * 
 * ESP32C series doesn't support LCD mode / parallel DMA! 
 *
 */

// Header
#include "esp32_i2s_parallel_dma.h" 

#if defined(ESP32_CXXX)


#include <esp_err.h>
 
 // Turn on and off a periphal
#include <driver/periph_ctrl.h>

// GPIO
#include <soc/gpio_periph.h>
#include <hal/gpio_types.h>
#include <driver/gpio.h>
#include <driver/periph_ctrl.h>

#include <rom/gpio.h>
#include <soc/gpio_sig_map.h>

// DMA Linked List Struct
#include <soc/lldesc.h>
#include <soc/io_mux_reg.h>

// I2S
#include <soc/i2s_struct.h>
#include <soc/i2s_reg.h>

// GDMA
#include <soc/gdma_channel.h>
#include <soc/gdma_periph.h>
#include <soc/gdma_reg.h>
#include <soc/gdma_struct.h>

// For I2S state management.
static i2s_parallel_state_t *i2s_state 	= NULL;

// ESP32-S2,S3,C3 only has IS20
// Original ESP32 has two I2S's, but we'll stick with the lowest common denominator.
static i2s_dev_t* I2S 					= &I2S0; // Device to use for this library, change if you want.

callback shiftCompleteCallback;
void setShiftCompleteCallback(callback f) {
    shiftCompleteCallback = f;
}

volatile bool previousBufferFree = true;

static void IRAM_ATTR gdma_irq_handler(void* arg) { // if we use I2S1 (default)

    GDMA.intr[0].clr.out_eof = 1;

    // at this point, the previously active buffer is free, go ahead and write to it
    // previousBufferFree = true;
    // if(shiftCompleteCallback) // we've defined a callback function ?
    // shiftCompleteCallback();

    // at this point, the previously active buffer is free, go ahead and write to it
    previousBufferFree = true;

    if(shiftCompleteCallback) // we've defined a callback function ?
        shiftCompleteCallback();
		
} // end irq_hndlr


// For peripheral setup and configuration

static inline int get_bus_width(i2s_parallel_cfg_bits_t width) {
  switch(width) {
    case I2S_PARALLEL_WIDTH_8:
      return 8;
    case I2S_PARALLEL_WIDTH_16:
      return 16;
    case I2S_PARALLEL_WIDTH_24:
      return 24;
    default:
      return -ESP_ERR_INVALID_ARG;
  }
}

static void gpio_setup_out(int gpio, int sig) {
  if(gpio < 0) {
    return;
  }

    // Configure GPIO
    // https://github.com/espressif/esp-idf/blob/d5f58ab13551cd883e8d8478ba367b6e4543ffec/examples/peripherals/gpio/generic_gpio/main/gpio_example_main.c    
    gpio_config_t io_conf = {};
    io_conf.intr_type       = GPIO_INTR_DISABLE;
    io_conf.mode            = GPIO_MODE_DEF_OUTPUT;
    io_conf.pin_bit_mask    = (1ULL << gpio);
    io_conf.pull_down_en    = 1;
    io_conf.pull_up_en      = 0;

    gpio_config(&io_conf);

    // Set IOMUX to GPIO
    //gpio_iomux_out(gpio, sig, false); // ?? is sig right?
    gpio_matrix_out(gpio, sig, false, false);

    // Drive Strength to MAX
    gpio_set_drive_capability((gpio_num_t)gpio, (gpio_drive_cap_t)3);  
}


// DMA Linked List
// Size must be less than DMA_MAX - need to handle breaking long transfer into two descriptors before call
// DMA_MAX by the way is the maximum data packet size you can hold in one chunk
void link_dma_desc(volatile lldesc_t *dmadesc, volatile lldesc_t *prevdmadesc, void *memory, size_t size) 
{
    if(size > DMA_MAX) size = DMA_MAX;

    dmadesc->size = size;
    dmadesc->length = size;
    dmadesc->buf = memory;
    dmadesc->eof = 0;
    dmadesc->sosf = 0;
    dmadesc->owner = 1;
    dmadesc->qe.stqe_next = 0;  // will need to set this elsewhere
    dmadesc->offset = 0;

    // link previous to current
    if(prevdmadesc)
        prevdmadesc->qe.stqe_next = (lldesc_t*)dmadesc;
}



esp_err_t i2s_parallel_driver_install(i2s_port_t port, i2s_parallel_config_t* cfg) {
  
  port = I2S_NUM_0; /// override.
  
  if(port < I2S_NUM_0 || port >= I2S_NUM_MAX) {
    return ESP_ERR_INVALID_ARG;
  }
  if(cfg->sample_width < I2S_PARALLEL_WIDTH_8 || cfg->sample_width >= I2S_PARALLEL_WIDTH_MAX) {
    return ESP_ERR_INVALID_ARG;
  }
  if(cfg->sample_rate > I2S_PARALLEL_CLOCK_HZ || cfg->sample_rate < 1) {
    return ESP_ERR_INVALID_ARG;
  }
  uint32_t clk_div_main = I2S_PARALLEL_CLOCK_HZ / cfg->sample_rate / i2s_parallel_get_memory_width(port, cfg->sample_width);
  if(clk_div_main < 2 || clk_div_main > 0xFF) {
    return ESP_ERR_INVALID_ARG;
  }
  
    // Setup GPIOs
    int bus_width = get_bus_width(cfg->sample_width);  

    i2s_dev_t *i2s_dev = I2S; // There's only one I2S device on C3....

    //Figure out which signal numbers to use for routing
    //printf("Setting up parallel I2S bus at I2S%d\n", i2snum(dev));
    int sig_data_base =  I2SO_SD_OUT_IDX;
    int sig_clk       =  I2SO_WS_OUT_IDX;

    //Route the signals
    for (int x=0; x < bus_width; x++) {
        gpio_setup_out(cfg->gpio_bus[x], sig_data_base+x);
    }

    //ToDo: Clk/WS may need inversion?
    gpio_setup_out(cfg->gpio_clk, sig_clk);
    
    // Power on I2S1 (or 0)
    periph_module_enable(PERIPH_I2S1_MODULE);
    
    // Now it's apparently I2S0 !?
    //i2s_dev_t *i2s_dev = &I2S0;

    // Reset RX (not that we use it)
    i2s_dev->rx_conf.val = 0;
    i2s_dev->rx_conf.rx_reset=1;         i2s_dev->rx_conf.rx_reset=0;
    i2s_dev->rx_conf.rx_fifo_reset=1;    i2s_dev->rx_conf.rx_fifo_reset=0;

    // Reset TX (what we care about)
    i2s_dev->tx_conf.val = 0;
    i2s_dev->tx_conf.tx_reset=1;         i2s_dev->tx_conf.tx_reset=0;
    i2s_dev->tx_conf.tx_fifo_reset=1;    i2s_dev->tx_conf.tx_fifo_reset=0;
    i2s_dev->tx_conf.tx_chan_equal=1;

    // Device setup
    i2s_dev->tx_conf1.val = 0;    
    i2s_dev->tx_conf1.tx_bits_mod=16;//cfg->bits;

    i2s_dev->rx_conf1.val = 0;    
    i2s_dev->rx_conf1.rx_bits_mod=16; //cfg->bits;

    i2s_dev->tx_conf1.tx_bck_div_num=2;
    i2s_dev->rx_conf1.rx_bck_div_num=2;

    i2s_dev->tx_clkm_conf.val=0;
    i2s_dev->tx_clkm_conf.tx_clk_sel=2; // 160mhz

    // clock speed
    i2s_dev->tx_clkm_conf.tx_clkm_div_num=160/16; // 10Mhz
    //i2s_dev->tx_clkm_div_conf.val = 0; // 
    i2s_dev->tx_clkm_div_conf.tx_clkm_div_x = 0; // > ?
    i2s_dev->tx_clkm_div_conf.tx_clkm_div_y = 0; // 
    i2s_dev->tx_clkm_div_conf.tx_clkm_div_yn1 = 0;
    i2s_dev->tx_clkm_div_conf.tx_clkm_div_z = 0;

    i2s_dev->tx_clkm_conf.tx_clk_active=1;  // Start

    
    //Allocate DMA descriptors
    i2s_state = malloc(sizeof(i2s_parallel_state_t));
    assert(i2s_state != NULL);
    i2s_parallel_state_t *st= i2s_state;

    st->desccount_a = cfg->desccount_a;
    st->desccount_b = cfg->desccount_b;
    st->dmadesc_a = cfg->lldesc_a;
    st->dmadesc_b = cfg->lldesc_b;

    // setup I2S Interrupt
    // SET_PERI_REG_BITS(I2S_INT_ENA_REG(1), I2S_OUT_EOF_INT_ENA_V, 1, I2S_OUT_EOF_INT_ENA_S);
    // allocate a level 1 intterupt: lowest priority, as ISR isn't urgent and may take a long time to complete

    esp_intr_alloc(ETS_I2S1_INTR_SOURCE, (int)(ESP_INTR_FLAG_IRAM | ESP_INTR_FLAG_LEVEL1), gdma_irq_handler, NULL, NULL);
    GDMA.intr[0].ena.out_eof  = 1; //?

 
    // Reset GDMA device
    GDMA.channel[0].out.out_conf0.out_rst   = 1; // REG_SET_BIT(DMA_OUT_CONF0_CH0_REG, DMA_OUT_RST_CH0);
    GDMA.channel[0].out.out_conf0.out_rst   = 0; // REG_CLR_BIT(DMA_OUT_CONF0_CH0_REG, DMA_OUT_RST_CH0);
    // GDMA.channel[0].out.out_conf0.out_eof_mode = 1; ?
    GDMA.misc_conf.ahbm_rst_inter = 1;
    GDMA.misc_conf.ahbm_rst_inter = 0;

    // Setup interrupt

    // Setup outlink
    GDMA.channel[0].out.out_link.addr       = ((uint32_t)(&st->dmadesc_a[0]));// Set a vlaue here
    GDMA.channel[0].out.out_peri_sel.sel    = SOC_GDMA_TRIG_PERIPH_I2S0; // 3 = I2S0
    GDMA.channel[0].out.out_conf0.out_data_burst_en = 1;
    GDMA.channel[0].out.out_conf0.outdscr_burst_en  = 1;
    GDMA.channel[0].out.out_link.start = 1;

    while (!GDMA.intr->raw.out_eof) { } // check status    

    i2s_dev->tx_conf.tx_start = 1;
	
    return ESP_OK;
}

 esp_err_t i2s_parallel_stop_dma(i2s_port_t port) {
  if(port < I2S_NUM_0 || port >= I2S_NUM_MAX) {
    return ESP_ERR_INVALID_ARG;
  }
	
   // Not implemented
   return ESP_OK;
}


 esp_err_t i2s_parallel_send_dma(i2s_port_t port, lldesc_t* dma_descriptor) {
  if(port < I2S_NUM_0 || port >= I2S_NUM_MAX) {
    return ESP_ERR_INVALID_ARG;
  }
  
   // Not implemented
  return ESP_OK;
}

i2s_dev_t* i2s_parallel_get_dev(i2s_port_t port) {
  if(port < I2S_NUM_0 || port >= I2S_NUM_MAX) {
    return NULL;
  }
  return I2S; // HARCODE THIS TO RETURN &I2S0
}

// Double buffering flipping
// Flip to a buffer: 0 for bufa, 1 for bufb
void i2s_parallel_flip_to_buffer(i2s_port_t port, int buffer_id) {

    if (i2s_state == NULL) {
      return; // :-()
    }

    lldesc_t *active_dma_chain;
    if (buffer_id == 0) {
        active_dma_chain=(lldesc_t*)&i2s_state->dmadesc_a[0];
    } else {
        active_dma_chain=(lldesc_t*)&i2s_state->dmadesc_b[0];
    }

    // setup linked list to refresh from new buffer (continuously) when the end of the current list has been reached
    i2s_state->dmadesc_a[i2s_state->desccount_a-1].qe.stqe_next = active_dma_chain;
    i2s_state->dmadesc_b[i2s_state->desccount_b-1].qe.stqe_next = active_dma_chain;

    // we're still shifting out the buffer, so it shouldn't be written to yet.
    previousBufferFree = false;
}

bool i2s_parallel_is_previous_buffer_free() {
    return previousBufferFree;
}

// End ESP32 original / S2, S3 check
#endif