How fast can I transmit another SPI frame calling spi_device_polling_transmit() again?

[mad_b747]

I have been experimenting with SPI;
Set and init for VSPI ok, then considering the code below:
(Ignoring that it will trigger watchdog in some seconds and etc.)
-Or working dedicated without interrupts enabled.

Code: [Select all] [Expand/Collapse]

1. // SPI configuration
2. #define PIN_NUM_MISO GPIO_NUM_19
3. #define PIN_NUM_MOSI GPIO_NUM_23
4. #define PIN_NUM_CLK  GPIO_NUM_18
5. #define PIN_NUM_CS   GPIO_NUM_5
6.  
7.     spi_bus_config_t bus_config;
8.     bus_config.miso_io_num = PIN_NUM_MISO;
9.     bus_config.mosi_io_num = PIN_NUM_MOSI;
10.     bus_config.sclk_io_num = PIN_NUM_CLK;
11.     bus_config.quadwp_io_num = -1;
12.     bus_config.quadhd_io_num = -1;
13.     bus_config.max_transfer_sz = 0;
14.  
15.     spi_device_interface_config_t dev_config;
16.     dev_config.clock_speed_hz = 1000000;  // Clock speed (1 MHz)
17.     dev_config.spics_io_num = PIN_NUM_CS;
18.     dev_config.queue_size = 1;
19.     dev_config.mode = 0;  // SPI mode 0
20.  
21.     spi_device_handle_t spi_handle;
22.  
23.     // Initialize the SPI bus
24.     esp_err_t ret;
25.     ret = spi_bus_initialize(VSPI_HOST, &bus_config, SPI_DMA_DISABLED);
26.     assert(ret == ESP_OK);
27.  
28.     // Attach the device to the SPI bus
29.     ret = spi_bus_add_device(VSPI_HOST, &dev_config, &spi_handle);
30.     assert(ret == ESP_OK);
31.  
32.     // Data to send
33.     spi_transaction_t trans;
34.     trans.flags = 0;
35.     trans.length = 64;  // Data length in bits
36.     trans.tx_buffer = data_to_send0;
37.  
38.     uint8_t data_to_send0[8] = { 0xFA, 0x5E, 0x01, 0xD0, 0xDA, 0xD0, 0xCA, 0xB0 };
39.     uint8_t data_to_send1[8] = { 0xFA, 0x5E, 0x02, 0xD0, 0xDA, 0xD0, 0xCA, 0xB1 };
40.  
41.     while (1)
42.     {
43.       trans.tx_buffer = data_to_send0;
44.       ret = spi_device_polling_transmit(spi_handle, &trans);
45.       //trans.tx_buffer = data_to_send1;
46.       //ret = spi_device_polling_transmit(spi_handle, &trans);
47.     }
48.  

GeSHi © Codebox Plus Extension

I have a couple of questions on this PoC/test code.
(This is not checking status of return, it previously was and it calls and works ok)

1-Even as it is above, with no scheduler interrupts, I checked in the logic analyzer to have a ~14 us delay between the data_to_send0 frames of data. I expect/need no delay in between calls, is something behind scenes that consumes the time after a transmit ends?

2-If I uncomment the second transmit, code breaks. Always. Immediately. So I cannot have double buffering for this? I tried also to let same references and refil a single data_to_send buffer, or have it source of data updated; is there some data lock or similar behind this, causing an access violation or whatever?

I do not want to reinvent the SPI and make a bit banging procedure for it to attend my need, what else could I do, in case someone have seen similar issues?
I need to feed a device near 1MHz limit of SPI functioning, few KB of data for a couple of seconds each time or so.

Thanks
PS. This runs in the second core, a single task, no interrupts. Disabled task watchdog.
PS2. Tried both using
trans.tx_buffer = data_to_send0;
and
trans.tx_buffer = &data_to_send0;
No change in results fir both working and not working cases.

[ESP_Sprite]

First of all, please don't initialize your structs like that; make sure they're zeroed, e.g.

Code: Select all

    spi_device_interface_config_t dev_config={0};

[mad_b747]

[Edit] Thanks, this seems to have solved the issue.

I presume that, for the compiler, when the next example statement is made, the non mentioned fields are by default initialized to zero

Code: Select all

spi_bus_config_t buscfg={
        .miso_io_num = -1,
        .mosi_io_num = MOSI_PIN,
        .sclk_io_num = CLK_PIN,
        .quadwp_io_num = -1,
        .quadhd_io_num = -1,
        .max_transfer_sz = 32,
    };

[ESP_Sprite]

Yes, if you initialize it using that method, you'll zero-initialize unmentioned members.