esp32s3 fft4real hangs if fft4r related codes are removed. (DSP-117)

[wuyuanyi135]

Environment

• Development Kit:
• Kit version (for WroverKit/PicoKit/DevKitC):
• Module or chip used: ESP32-S3-WROOM-1-N4
• IDF version (run git describe --tags to find it): v5.2-dev-2383-g82cceabc6e
• Build System: CMake
• Compiler version (run xtensa-esp32-elf-gcc --version to find it): (crosstool-NG esp-12.2.0_20230208) 12.2.0
• Operating System: Windows
• Power Supply: USB

Problem Description

I modified the fft for real example to only use fft2r functions. The program hangs.

Expected Behavior

fft2r should work

Actual Behavior

after remove fft4r related code, the code hangs.

Steps to repropduce

see the code below

Code to reproduce this issue

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "esp_system.h"
#include "driver/spi_master.h"
#include "soc/gpio_struct.h"
#include "driver/gpio.h"
#include "driver/uart.h"
#include "soc/uart_struct.h"
#include <math.h>

#include "esp_dsp.h"

static const char *TAG = "main";

// This example shows how to use FFT from esp-dsp library

#define N_SAMPLES 2048 // Amount of real input samples
int N = N_SAMPLES;
// Input test array
__attribute__((aligned(16)))
float x1[N_SAMPLES];
// Window coefficients
__attribute__((aligned(16)))
float wind[N_SAMPLES];
// Pointers to result arrays
float* y1_cf = &x1[0];

extern "C" void app_main()
{
  esp_err_t ret;
  ESP_LOGI(TAG, "Start Example.");
  ret = dsps_fft2r_init_fc32(NULL, N>>1);
  if (ret  != ESP_OK)
  {
    ESP_LOGE(TAG, "Not possible to initialize FFT2R. Error = %i", ret);
    return;
  }

  // Generate hann window
  dsps_wind_hann_f32(wind, N);
  // Generate input signal for x1 A=1 , F=0.1
  dsps_tone_gen_f32(x1, N, 1.0, 0.16,  0);

  // Convert two input vectors to one complex vector
  for (int i=0 ; i< N ; i++)
  {
    x1[i] = x1[i] * wind[i];
  }
  // FFT Radix-2
  unsigned int start_r2 = dsp_get_cpu_cycle_count();
  dsps_fft2r_fc32(x1, N>>1);
  // Bit reverse
  dsps_bit_rev2r_fc32(x1, N>>1);
  // Convert one complex vector with length N/2 to one real spectrum vector with length N/2
  dsps_cplx2real_fc32(x1, N>>1);
  unsigned int end_r2 = dsp_get_cpu_cycle_count();

  for (int i = 0 ; i < N/2 ; i++) {
    x1[i] = 10 * log10f((x1[i * 2 + 0] * x1[i * 2 + 0] + x1[i * 2 + 1] * x1[i * 2 + 1] + 0.0000001)/N);
  }

  // Show power spectrum in 64x10 window from -100 to 0 dB from 0..N/4 samples
  ESP_LOGW(TAG, "Signal x1");
  dsps_view(x1, N/2, 64, 10,  -60, 40, '|');
  ESP_LOGI(TAG, "FFT Radix 2 for %i complex points take %i cycles", N/2, end_r2 - start_r2);

  ESP_LOGI(TAG, "End Example.");
}

Debug Logs

I (27) boot: ESP-IDF v5.2-dev-2383-g82cceabc6e 2nd stage bootloader
I (27) boot: compile time Sep  4 2023 10:24:09
I (27) boot: Multicore bootloader
I (32) boot: chip revision: v0.1
I (35) boot.esp32s3: Boot SPI Speed : 80MHz
I (40) boot.esp32s3: SPI Mode       : DIO
I (45) boot.esp32s3: SPI Flash Size : 2MB
I (50) boot: Enabling RNG early entropy source...
I (55) boot: Partition Table:
I (59) boot: ## Label            Usage          Type ST Offset   Length
I (66) boot:  0 nvs              WiFi data        01 02 00009000 00006000
I (73) boot:  1 phy_init         RF data          01 01 0000f000 00001000
I (81) boot:  2 factory          factory app      00 00 00010000 00100000
I (88) boot: End of partition table
I (92) esp_image: segment 0: paddr=00010020 vaddr=3c050020 size=202d4h (131796) map
I (125) esp_image: segment 1: paddr=000302fc vaddr=3fc90100 size=027e8h ( 10216) load
I (127) esp_image: segment 2: paddr=00032aec vaddr=40374000 size=0c048h ( 49224) load
I (141) esp_image: segment 3: paddr=0003eb3c vaddr=00000000 size=014dch (  5340)
I (143) esp_image: segment 4: paddr=00040020 vaddr=42000020 size=48ed8h (298712) map
I (206) boot: Loaded app from partition at offset 0x10000
I (206) boot: Disabling RNG early entropy source...
I (217) cpu_start: Multicore app
I (217) cpu_start: Pro cpu up.
I (218) cpu_start: Starting app cpu, entry point is 0x403750cc
0x403750cc: call_start_cpu1 at C:/Espressif/frameworks/esp-idf-master/components/esp_system/port/cpu_start.c:170

I (0) cpu_start: App cpu up.
I (235) cpu_start: Pro cpu start user code
I (236) cpu_start: cpu freq: 240000000 Hz
I (236) cpu_start: Application information:
I (239) cpu_start: Project name:     aaar_wifi_tempvib
I (244) cpu_start: App version:      1
I (249) cpu_start: Compile time:     Sep  4 2023 10:24:00
I (255) cpu_start: ELF file SHA256:  cf4ffb33c...
I (260) cpu_start: ESP-IDF:          v5.2-dev-2383-g82cceabc6e
I (267) cpu_start: Min chip rev:     v0.0
I (271) cpu_start: Max chip rev:     v0.99
I (276) cpu_start: Chip rev:         v0.1
I (281) heap_init: Initializing. RAM available for dynamic allocation:
I (288) heap_init: At 3FC987D8 len 00050F38 (323 KiB): DRAM
I (294) heap_init: At 3FCE9710 len 00005724 (21 KiB): STACK/DRAM
I (301) heap_init: At 3FCF0000 len 00008000 (32 KiB): DRAM
I (307) heap_init: At 600FE010 len 00001FD8 (7 KiB): RTCRAM
I (314) spi_flash: detected chip: generic
I (318) spi_flash: flash io: dio
W (322) spi_flash: Detected size(4096k) larger than the size in the binary image header(2048k). Using the size in the binary image header.
I (337) sleep: Configure to isolate all GPIO pins in sleep state
I (342) sleep: Enable automatic switching of GPIO sleep configuration
I (349) app_start: Starting scheduler on CPU0
I (354) app_start: Starting scheduler on CPU1
I (354) main_task: Started on CPU0
I (364) main_task: Calling app_main()
I (364) main: Start Example.

***** NO MORE LINES *****

Other items if possible

• sdkconfig file (attach the sdkconfig file from your project folder)
• elf file in the build folder (note this may contain all the code details and symbols of your project.)
• coredump (This provides stacks of tasks.)

[peter-marcisovsky]

Hello @wuyuanyi135 could you please leave the dsps_fft4r_init_fc32 in your code, and see if you could run it?

[wuyuanyi135]

Hello @wuyuanyi135 could you please leave the dsps_fft4r_init_fc32 in your code, and see if you could run it?

Yes. with dsps_fft4r_init_fc32 the example runs. I found dsps_cplx2real_fc32 is actually a part of fft4r. I guess this is the issue why uninitialized fft4r caused problem.

[wuyuanyi135]

@peter-marcisovsky sorry to bother you. Does dsps_fft4r_init_fc32 affecting radix 2 APIs rings any alarm to you?

[wuyuanyi135]

Anyone could provide some insights?