Most efficient implementation of FFT on TMS320F280025C

Thank you very much for providing answer to each question. I have no more queries regarding points 1,2 &4. However for point number 3 it will be nice to have more details. The example folder contains implementation for 2833x series. I will try to take guide lines from the linker command files of this controller. However it will be much better if there is some reference linker command file for 28002x series that contains mapping for the IQMath sections as well as FFT sections.

You can refer to the IQMath examples as well. The linker cmd files are provided in the IQMath library. You only need to make sure that the tables and libraries are properly allocated for both FFT and IQMath. Let me know if you face any difficulty.

Dear Shanty thankyou very much for the continued support. The documentation has provided sample allocation as below:
FFTipcb ALIGN(256) : { } > RAML4, PAGE 1
FFTipcbsrc : { } > RAML5, PAGE 1
FFTmag > RAML6, PAGE 1
FFTtf > RAML7, PAGE 1

The linker command file that I am using was modified from the example file for TMS320F280025C launchpad. In this linker command file the RAM sectors are combined to a single block and already associated to different sections. Please see below:

MEMORY
{
BEGIN : origin = 0x080000, length = 0x000002
BOOT_RSVD : origin = 0x00000002, length = 0x00000126
RAMM0 : origin = 0x00000128, length = 0x000002D8
RAMM1 : origin = 0x00000400, length = 0x000003F8 /* on-chip RAM block M1 */
// RAMM1_RSVD : origin = 0x000007F8, length = 0x00000008 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */

/* RAMLS4 : origin = 0x0000A000, length = 0x00000800
RAMLS5 : origin = 0x0000A800, length = 0x00000800
RAMLS6 : origin = 0x0000B000, length = 0x00000800
RAMLS7 : origin = 0x0000B800, length = 0x00000800*/

/* Combining all the LS RAMs */
RAMLS4567 : origin = 0x0000A000, length = 0x00002000
RAMGS0 : origin = 0x0000C000, length = 0x000007F8
// RAMGS0_RSVD : origin = 0x0000C7F8, length = 0x00000008 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */

// FLASHBANK1 : origin = 0x00080000, length = 0x0000FFF0
// FLASH_BANK1_RSVD : origin = 0x0008FFF0, length = 0x00000010 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */
BOOTROM : origin = 0x003F0000, length = 0x00008000
BOOTROM_EXT : origin = 0x003F8000, length = 0x00007FC0
RESET : origin = 0x003FFFC0, length = 0x00000002
/* Flash sectors */
/* BANK 0 */
FLASH_BANK0_SEC0 : origin = 0x080002, length = 0x000FFE /* on-chip Flash */
FLASH_BANK0_SEC1 : origin = 0x081000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC2 : origin = 0x082000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC3 : origin = 0x083000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC4 : origin = 0x084000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC5 : origin = 0x085000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC6_7 : origin = 0x086000, length = 0x002000 /* on-chip Flash */
FLASH_BANK0_SEC8 : origin = 0x088000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC9 : origin = 0x089000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC10 : origin = 0x08A000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC11 : origin = 0x08B000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC12 : origin = 0x08C000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC13 : origin = 0x08D000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC14 : origin = 0x08E000, length = 0x001000 /* on-chip Flash */
FLASH_BANK0_SEC15 : origin = 0x08F000, length = 0x000FF0 /* on-chip Flash */
// FLASH_BANK0_SEC15_RSVD : origin = 0x08FFF0, length = 0x000010 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */

}

SECTIONS
{
codestart : > BEGIN, ALIGN(8)
.text : >> FLASH_BANK0_SEC2 | FLASH_BANK0_SEC3 | FLASH_BANK0_SEC4 | FLASH_BANK0_SEC5 , ALIGN(8)
.cinit : > FLASH_BANK0_SEC1, ALIGN(8)
.switch : > FLASH_BANK0_SEC1, ALIGN(8)
.reset : > RESET, TYPE = DSECT /* not used, */

.stack : > RAMM1

.init_array : > FLASH_BANK0_SEC1, ALIGN(8)
.bss : > RAMLS4567
.bss:output : > RAMLS4567
.bss:cio : > RAMGS0
.const : > FLASH_BANK0_SEC6_7, ALIGN(8)
.data : > RAMLS4567
.sysmem : > RAMLS4567

ramgs0 : > RAMGS0

/* Allocate IQ math areas: */
IQmath : > FLASH_BANK0_SEC1, PAGE = 0, ALIGN(4) /* Math Code */
IQmathTables : > FLASH_BANK0_SEC2, PAGE = 0, ALIGN(4)

.TI.ramfunc : LOAD = FLASH_BANK0_SEC1,
RUN = RAMGS0,
LOAD_START(RamfuncsLoadStart),
LOAD_SIZE(RamfuncsLoadSize),
LOAD_END(RamfuncsLoadEnd),
RUN_START(RamfuncsRunStart),
RUN_SIZE(RamfuncsRunSize),
RUN_END(RamfuncsRunEnd),
ALIGN(8)

}
/*
//===========================================================================
// End of file.
//===========================================================================
*/

I am not very much familiar with the modifications for the linker command files. Please give me guidelines on how to accommodate the FFT sections in the above linker command file. It will be very nice if you can edit the above linker file for same.

Hi,

You are right. You can allocate the FPU FFT math tables to RAMLS4567 in F28002x the same way. 

However, if you would like to allocate tables (IQMath or FFT) to FLASH, please make sure to memcpy them to RAM in your code as demonstrated by the examples. 

-Shantanu

Why is it required to memcpy IQMath /FFT to RAM?. Is it for fast execution?. Secondly the FFT twiddle factors are taking three sectors of the FLASH. This is very costly for my project. I understand that this is due to 4096 point FFT twiddle factors. Can I reduce it  considering that I will be using only 64 points RFFT. It will be grateful if you can check if the following linker command has proper assignments

MEMORY
{
   BEGIN           	: origin = 0x080000, length = 0x000002
   BOOT_RSVD		: origin = 0x00000002, length = 0x00000126
   RAMM0           	: origin = 0x00000128, length = 0x000002D8
   RAMM1            : origin = 0x00000400, length = 0x000003F8     /* on-chip RAM block M1 */
// RAMM1_RSVD       : origin = 0x000007F8, length = 0x00000008 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */

/* RAMLS4           : origin = 0x0000A000, length = 0x00000800
   RAMLS5           : origin = 0x0000A800, length = 0x00000800
   RAMLS6           : origin = 0x0000B000, length = 0x00000800
   RAMLS7           : origin = 0x0000B800, length = 0x00000800*/

   /* Combining all the LS RAMs */
   RAMLS4567        : origin = 0x0000A000, length = 0x00002000
   RAMGS0           : origin = 0x0000C000, length = 0x000007F8
// RAMGS0_RSVD      : origin = 0x0000C7F8, length = 0x00000008 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */


//  FLASHBANK1       : origin = 0x00080000, length = 0x0000FFF0
//  FLASH_BANK1_RSVD : origin = 0x0008FFF0, length = 0x00000010 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */
   BOOTROM          : origin = 0x003F0000, length = 0x00008000
   BOOTROM_EXT      : origin = 0x003F8000, length = 0x00007FC0
   RESET            : origin = 0x003FFFC0, length = 0x00000002
   /* Flash sectors */
   /* BANK 0 */
   FLASH_BANK0_SEC0  : origin = 0x080002, length = 0x000FFE	/* on-chip Flash */
   FLASH_BANK0_SEC1  : origin = 0x081000, length = 0x001000	/* on-chip Flash */
   FLASH_BANK0_SEC2  : origin = 0x082000, length = 0x001000	/* on-chip Flash */
   FLASH_BANK0_SEC3_9  : origin = 0x083000, length = 0x007000	/* on-chip Flash */
   FLASH_BANK0_SEC10_11 : origin = 0x08A000, length = 0x002000	/* on-chip Flash */
   FLASH_BANK0_SEC12_15 : origin = 0x08C000, length = 0x003FF0	/* on-chip Flash */
   // FLASH_BANK0_SEC15_RSVD     : origin = 0x08FFF0, length = 0x000010  /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */

}


SECTIONS
{
   codestart        : > BEGIN, ALIGN(8)
   .text            : >> FLASH_BANK0_SEC2 | FLASH_BANK0_SEC3_9 ,   ALIGN(8)
   .cinit           : > FLASH_BANK0_SEC1,  ALIGN(8)
   .switch          : > FLASH_BANK0_SEC1,  ALIGN(8)
   .reset           : > RESET,                  TYPE = DSECT /* not used, */

   .stack           : > RAMM1

   .init_array      : > FLASH_BANK0_SEC1,  ALIGN(8)
   .bss             : > RAMLS4567
   .bss:output      : > RAMLS4567
   .bss:cio         : > RAMGS0
   .const           : > FLASH_BANK0_SEC10_11,  ALIGN(8)
   .data            : > RAMLS4567
   .sysmem          : > RAMLS4567

    ramgs0 : > RAMGS0


   /* Allocate IQ math areas: */
   IQmath : > FLASH_BANK0_SEC1, PAGE = 0, ALIGN(4) /* Math Code */
   IQmathTables : > FLASH_BANK0_SEC2, PAGE = 0, ALIGN(4)

   /* Allocating FFT areas: */
   FFTipcb 		: >> RAMGS0, ALIGN(256) //64 points FFT
   FFTipcbsrc	: >> RAMLS4567
   FFTtf        : > FLASH_BANK0_SEC12_15

  .TI.ramfunc      : LOAD = FLASH_BANK0_SEC1,
                  RUN = RAMGS0,
                  LOAD_START(RamfuncsLoadStart),
                  LOAD_SIZE(RamfuncsLoadSize),
                  LOAD_END(RamfuncsLoadEnd),
                  RUN_START(RamfuncsRunStart),
                  RUN_SIZE(RamfuncsRunSize),
                  RUN_END(RamfuncsRunEnd),
                  ALIGN(8)

}
/*
//===========================================================================
// End of file.
//===========================================================================
*/

Hi,

You can also use the FFT and IQMath tables from ROM. You do not need to actually load them if space is an issue. Please refer to the RAM_ROMTABLES configurations as well in the examples. You need to make sure that the symbols library is included. Refer to the documentation.

It is not necessary to put tables in FLASH, but as specified in the examples, this is how the allocation to FLASH is to be done:

FPU64fftTables : LOAD = FLASHB,
RUN = RAMGS2,
RUN_START(FPU64fftTablesRunStart),
LOAD_START(FPU64fftTablesLoadStart),
LOAD_SIZE(FPU64fftTablesLoadSize),
PAGE = 1

Since you are specifying that you are running from RAM, you need to memcpy because memory access times are lower. 

Hope this helps.

-Shantanu

Dear Shanty,

  I have used example code as a reference to make a small code for checking RFFT on TMS320F280025C. My linker command file is as below:

MEMORY
{
   BEGIN           	: origin = 0x080000, length = 0x000002
   BOOT_RSVD		: origin = 0x00000002, length = 0x00000126
   RAMM0           	: origin = 0x00000128, length = 0x000002D8
   RAMM1            : origin = 0x00000400, length = 0x000003F8     /* on-chip RAM block M1 */
// RAMM1_RSVD       : origin = 0x000007F8, length = 0x00000008 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */

/* RAMLS4           : origin = 0x0000A000, length = 0x00000800
   RAMLS5           : origin = 0x0000A800, length = 0x00000800
   RAMLS6           : origin = 0x0000B000, length = 0x00000800
   RAMLS7           : origin = 0x0000B800, length = 0x00000800*/

   /* Combining all the LS RAMs */
   RAMLS4567        : origin = 0x0000A000, length = 0x00002000
   RAMGS0           : origin = 0x0000C000, length = 0x000007F8
// RAMGS0_RSVD      : origin = 0x0000C7F8, length = 0x00000008 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */


//  FLASHBANK1       : origin = 0x00080000, length = 0x0000FFF0
//  FLASH_BANK1_RSVD : origin = 0x0008FFF0, length = 0x00000010 /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */
   BOOTROM          : origin = 0x003F0000, length = 0x00008000
   BOOTROM_EXT      : origin = 0x003F8000, length = 0x00007FC0
   RESET            : origin = 0x003FFFC0, length = 0x00000002
   /*IQMath table already exists in the BOOTROM section*/
   //IQTABLES 		: origin = 0x3F1402, length = 0x0166D
   /* Flash sectors */
   /* BANK 0 */
   FLASH_BANK0_SEC0  : origin = 0x080002, length = 0x000FFE	/* on-chip Flash */
   FLASH_BANK0_SEC1  : origin = 0x081000, length = 0x001000	/* on-chip Flash */
   FLASH_BANK0_SEC2  : origin = 0x082000, length = 0x001000	/* on-chip Flash */
   FLASH_BANK0_SEC3_9  : origin = 0x083000, length = 0x007000	/* on-chip Flash */
   FLASH_BANK0_SEC10_11 : origin = 0x08A000, length = 0x002000	/* on-chip Flash */
   FLASH_BANK0_SEC12_15 : origin = 0x08C000, length = 0x003FF0	/* on-chip Flash */
   // FLASH_BANK0_SEC15_RSVD     : origin = 0x08FFF0, length = 0x000010  /* Reserve and do not use for code as per the errata advisory "Memory: Prefetching Beyond Valid Memory" */

}


SECTIONS
{
   codestart        : > BEGIN, ALIGN(8)
   .text            : >> FLASH_BANK0_SEC2 | FLASH_BANK0_SEC3_9 ,   ALIGN(8)//executable code >> operator splits code
   .cinit           : > FLASH_BANK0_SEC1,  ALIGN(8)//table that initializes global variables
   .switch          : > FLASH_BANK0_SEC1,  ALIGN(8)
   .reset           : > RESET,                  TYPE = DSECT /* not used, Dummy Section*/

   .stack           : > RAMM1

   .init_array      : > FLASH_BANK0_SEC1,  ALIGN(8)
   .bss             : > RAMLS4567 //uninitialized global variables
   .bss:output      : > RAMLS4567
   .bss:cio         : > RAMGS0
   .const           : > FLASH_BANK0_SEC10_11,  ALIGN(8)
   .data            : > RAMLS4567 //initialized data
   .sysmem          : > RAMLS4567

    ramgs0 : > RAMGS0


   /* Allocate IQ math areas: */
   IQmath : > FLASH_BANK0_SEC1, PAGE = 0, ALIGN(4) /* Math Code */
   //IQmathTables : > FLASH_BANK0_SEC2, PAGE = 0, ALIGN(4)/*This was working*/
   IQmathTables : > FLASH_BANK0_SEC1, PAGE = 0, ALIGN(4)/*Moved to sector 1 to run from RAM.Need verification*/

   /* Allocating FFT areas: */
   FFTipcb 		: >> RAMGS0, ALIGN(256) //64 points FFT
   FFTipcbsrc	: >> RAMLS4567
   FFTtf        : > FLASH_BANK0_SEC12_15

  .TI.ramfunc      : LOAD = FLASH_BANK0_SEC1, //Flash bank0 sector 1 contaings time critical functions required to run from RAM
                  RUN = RAMGS0,
                  LOAD_START(RamfuncsLoadStart),
                  LOAD_SIZE(RamfuncsLoadSize),
                  LOAD_END(RamfuncsLoadEnd),
                  RUN_START(RamfuncsRunStart),
                  RUN_SIZE(RamfuncsRunSize),
                  RUN_END(RamfuncsRunEnd),
                  ALIGN(8)

}
/*
//===========================================================================
// End of file.
//===========================================================================
*/

 There are two issues with the result.

1- The values in ipcbsrc array should be large values but these are very much scaled down (printed as sample inputs). However it is still a one cycle sine wave.

2- The FFT result is all zero.

Kindly check if the linker command file is Okay. Also check the code if I am missing some calculation step.

The snippets of the FFT code are as below (I have removed non FFT code for simplicity purposes).

#define FFT_SIZE    64

#include "fft.h"
#include "fft_hamming_Q31.h"
#include "math.h"

//Defining FFT sections
#pragma DATA_SECTION(ipcb, "FFTipcb");
int32_t ipcb[FFT_SIZE+2];         //Calculations buffer
#pragma DATA_SECTION(ipcbsrc, "FFTipcbsrc");
int32_t ipcbsrc[FFT_SIZE];        //Sampled input signal
//Defining real fft object for 64 point calculations
RFFT32  rfft = RFFT32_64P_DEFAULTS;



void main(void)
{for (i=0;i<FFT_SIZE;i++)
    {
        ipcbsrc[i]=(long)(2147483648*sinf(Rad));//Q31 format
        Rad=Rad+0.09817;
    }
    printf("\r\nSample Input:\r\n");
    for (i = 0; i < FFT_SIZE; i++)


        printf("%ld\r\n", ipcbsrc[i]);
    }
    RFFT32_brev(ipcbsrc, ipcb, FFT_SIZE); // real FFT bit reversing

    rfft.ipcbptr = ipcb;                  // FFT computation buffer
    rfft.magptr  = ipcbsrc;               // Magnitude output buffer
    rfft.winptr  = (long *)win;           // Window coefficient array
    rfft.init(&rfft);                     // Twiddle factor pointer initialization

    rfft.calc(&rfft);                     // Compute the FFT
    rfft.split(&rfft);                    // Post processing to get the correct spectrum
    rfft.mag(&rfft);                      // Q31 format (abs(ipcbsrc)/2^16).^2


    printf("\r\nFFT Result:\r\n");
    for (i = 0; i < FFT_SIZE; i++)
    {
         printf("%ld\r\n", ipcbsrc[i]);
    }
}

Following is the output of the code as printed on serial terminal:

Sample Input:
0
31
63
94
123
152
179
205
229
250
269
285
299
310
317
322
324
322
317
310
299
285
269
250
229
205
180
152
124
94
63
31
0
-31
-63
-94
-123
-152
-179
-205
-229
-250
-269
-285
-299
-310
-317
-322
-324
-322
-317
-310
-299
-285
-269
-250
-229
-205
-180
-152
-124
-94
-63
-31

FFT Result:
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
-31
-63
-94
-123
-152
-179
-205
-229
-250
-269
-285
-299
-310
-317
-322
-324
-322
-317
-310
-299
-285
-269
-250
-229
-205
-180
-152
-124
-94
-63
-31

Hi,
Sorry for the delay in response. If you look at the rfft example provided, the input and output buffers are interchanged. Please provide the ipcb buffer as well.

IN addition, please do the following checks:

1) In the map file, make sure all tables are allocated

2) During runtime, make sure that all buffers have the appropriate values

3) Try copying to RAM using the LOAD, RUN parameters. It should look similar to this in the cmd file:

FPUmathTables : LOAD = FLASH,
RUN = RAMGS3,
RUN_START(FPUmathTablesRunStart),
LOAD_START(FPUmathTablesLoadStart),
LOAD_SIZE(FPUmathTablesLoadSize),

IN the code, you will have to memcpy the tables by  MemCopy(&FPUmathTablesLoadStart, &FPUmathTablesLoadEnd, &FPUmathTablesRunStart);

4) You can also use the tables already provided in the Boot ROM. You do not need to allocate them in FLASH or RAM. Consult the RAM-ROMTABLES config and the documentation/

Hope this helps

-Shantanu

Dear Shantanu,

   I am sorry that I missed your reply some how. I have created a new post in which I have provided more details. Let me mark this thread as resolved as the initial intention was to get advise regarding initial implementation queries that have been addressed in this thread. Please guide on the other thread regarding implementation issues.

Asad,

You can mark this thread as resolved in that case. Please send the link of the new thread.

-Shantanu

e2e.ti.com/.../tms320f280025c-rfft-code-calculations-giving-all-zeros-and-lots-of-warnings