Missing Detail in C5515 HWAFFT Documentation

Jason51114

I've found that the built-in routines for the HWAFFT module on the C5505/C5515 do not return the correct results if the input/output and scratch buffers provided by the caller do not lie in the same data page. Looking at the corresponding assembly code, it becomes apparent why. Between each stage of FFT computation, there are steps like:

; Start 1st double stage
AR0 = AR4
AR1 = AR5

; Start second double stage
AR2 = AR5
AR3 = AR4

And so on. The two pointers bounce back and forth between the input/output and scratch buffers. However, since the move instruction only references the auxiliary registers and not the extended auxiliary registers (XARn), only the bottom 16 bits are copied. As long as the two buffers are in the same page (so their most-significant 8 bits are the same), then everything works. However, if they are on different pages, data ends up getting fetched from the wrong address and the results are garbage.

Obviously this can't be changed in ROM, but it would probably be a good idea if the HWAFFT documentation specified this requirement.

Jason

over 15 years ago

Steve Tsang over 15 years ago

TI__Mastermind 23785 points

Sure. We will update the doc.

Thanks,

Steve

Micky52259 over 14 years ago in reply to Steve Tsang

Intellectual 765 points

This hasnot yet been done!!!

Cong Van Nguyen over 13 years ago in reply to Micky52259

Intellectual 970 points

The FFT assembly routines burned in C5515 ROM seems to support only small memory model!

In order to use HWAFFT for large/huge memory models and allow allocating HWAFFT buffers in different data pages, I created the following wrapper:

In a C header file:

typedef Uint16 (*hwafft_func_t)(Int32 *data,
                                Int32 *scratch,
                                Uint16 fft_flag,
                                Uint16 scale_flag);
Uint16 hwafft_wrapper( Int32 *data,
                        Int32 *scratch,
                        Uint16 fft_flag,
                        Uint16 scale_flag,
                        hwafft_func_t fft_func);

In an ASM source file:

    .sect ".onchip_code"
    .def _hwafft_wrapper

; HWAFFT code burned in C5515 ROM seems to support small memory model only.
; The following wrapper helps to use HWAFFT for large/huge memory models.
; CVN 20110920
_hwafft_wrapper:
    XAR2 = XAR1
    XAR3 = XAR0
    call AC0
    return

Cong Van Nguyen over 13 years ago in reply to Cong Van Nguyen

Intellectual 970 points

Another possible workaround is to disassembly all FFT routines from ROM, replace all occurrences of

AR2 = AR5

AR3 = AR4

XAR2 = XAR5

XAR3 = XAR4

and then allocate these modified routines in RAM instead. This workaround would consume 4KB of RAM.

Ras over 13 years ago in reply to Cong Van Nguyen

Intellectual 675 points

Is any change required for the following lines if using huge memory model?

AR0 = AR4

AR1 = AR5

Antony Rix over 12 years ago

Intellectual 265 points

Just a quick note to thank all of you who contributed to this thread, otherwise I wouldn't have found how to get this working because I was working on documentation downloaded a few months ago.

TI have finally (!!!!) updated the documentation on the HWAFFT to point to a new update to the silicon errata for C5515 and C5505.

See http://www.ti.com/dsp/docs/litabsmultiplefilelist.tsp?sectionId=3&tabId=409&literatureNumber=sprabb6b&docCategoryId=1&familyId=325

I can confirm that placing both the data and scratch buffers in the first 64kByte of memory appears to work correctly.

Antony

Anand Chandrasekhar over 12 years ago in reply to Cong Van Nguyen

Intellectual 525 points

Some1 please help me out !!

But, when i am running the code, I am not able to get the FFT computed. As soon as the code reaches the FFT computation part, it just stands still.
So , I stopped the running to see what is getting executed. The code gets stuck here

8e49bdeb1110             COPR #16,AC0,dbl(*AR2-),AC1 :: MOV AC1,dbl(*(AR3+T0))
8e49bdeb1110             COPR #16,AC0,dbl(*AR2-),AC1 :: MOV AC1,dbl(*(AR3+T0))
8e49bdeb1110             COPR #16,AC0,dbl(*AR2-),AC1 :: MOV AC1,dbl(*(AR3+T0))
8e49bdeb1110             COPR #16,AC0,dbl(*AR2-),AC1 :: MOV AC1,dbl(*(AR3+T0))

I am not sure , what exactly these statements does. But all the lines are same !!

If i step debug this, it loops arround these lines.

But, If i give a run and stop , it get stuck @

040009 BCC #0x3d12c3,AC0 == #0

and if I give a run again : I get an error message saying I cant do tht !!!!!!

I am struggling with this issue for the past 2-3 days. It will be a great help, If you can tell me how to solve this.
I have a breif code structure, which shows how I am computing FFT. Please have a look.

This is my code
==================================================================================================================================
#include "stdio.h"
#include "usbstk5515.h"
#include "usbstk5515_led.h"
#include "usbstk5515_gpio.h"
#include "usbstk5515_i2c.h"
#include "sar.h"
#include "lcd.h"
#include "hwafft.h"
#include "math.h"

Int32 Input_Signal1024];
extern Uint16 hwafft_wrapper( Int32 *data,Int32 *scratch,Uint16 fft_flag,Uint16 scale_flag,hwafft_func_t fft_func);
Int32* fft(Int32 *data);

void main( void )
{
   Int16 result = 0;Int32 *fft_data = 0;
    USBSTK5515_init( );
  Input_Signal = Aquire_signal_from_ADC_for_10s();
    fft_data = fft( Input_Signal);
    while(1);
}

Int32* fft(Int32 *data)
{
   Int32 *result = 0;
   Int32 *data_br = 0;
   Uint16 out_sel = 0;
   Int32 *scratch= 0;
   hwafft_br(data,data_br,DATA_LEN_1024); // bit reversal
   data = data_br;
   out_sel = hwafft_wrapper(data,scratch,FFT_FLAG,SCALE_FLAG,hwafft_1024pts); // I am using this because, I am using large memory model
    if (out_sel == OUT_SEL_DATA)
   {
       result = data;
   }
   else
   {
       result = scratch;
   }
   return(result);
}
===================================================================================================================================
===================================================================================================================================

I have defined the below code in an assembly file. I got this code snippet from TI forum. I was asked to use this since I am using large memory model. It seems
FFT accelerator supports only small memory model

         .sect ".onchip_code"
        .def _hwafft_wrapper
       ; HWAFFT code burned in C5515 ROM seems to support small memory model only.
       ; The following wrapper helps to use HWAFFT for large/huge memory models.
       ; CVN 20110920
_hwafft_wrapper:
    XAR2 = XAR1
    XAR3 = XAR0
    call AC0
    return

I have defined the below code in a C header file.(USBSTK5515.h)

===================================================================================================================================
===================================================================================================================================

The following is my .cmd file

-stack    0x400                /* PRIMARY STACK SIZE    */
-sysstack 0x400                /* SECONDARY STACK SIZE */
-heap     0x400               /* HEAP AREA SIZE        */
MEMORY
{
   MMR     (RW) : origin = 0000000h length = 0000c0h    /* MMRs */
   DARAM_0 (RW) : origin = 00000c0h length = 001f40h   /* on-chip DARAM 0 */
    DARAM_1 (RW) : origin = 0002000h length = 002000h   /* on-chip DARAM 1 */
    DARAM_2 (RW) : origin = 0004000h length = 002000h   /* on-chip DARAM 2 */
    DARAM_3 (RW) : origin = 0006000h length = 002000h   /* on-chip DARAM 3 */
    DARAM_4 (RW) : origin = 0008000h length = 002000h   /* on-chip DARAM 4 */
    DARAM_5 (RW) : origin = 000a000h length = 002000h   /* on-chip DARAM 5 */
    DARAM_6 (RW) : origin = 000c000h length = 002000h   /* on-chip DARAM 6 */
    DARAM_7 (RW) : origin = 000e000h length = 002000h   /* on-chip DARAM 7 */
    SARAM   (RW) : origin = 0010000h length = 040000h   /* on-chip SARAM */
    SAROM_0 (RX) : origin = 0fe0000h length = 008000h    /* on-chip ROM 0 */
    SAROM_1 (RX) : origin = 0fe8000h length = 008000h    /* on-chip ROM 1 */
    SAROM_2 (RX) : origin = 0ff0000h length = 008000h    /* on-chip ROM 2 */
    SAROM_3 (RX) : origin = 0ff8000h length = 007f00h    /* on-chip ROM 3 */
   VECS    (RX) : origin = 0ffff00h length = 000100h   /* on-chip ROM vectors */
    EMIF_CS0 (RW) : origin = 0050000h length = 07B0000h   /* mSDR */
   EMIF_CS2 (RW) : origin = 0800000h length = 0400000h   /* ASYNC1 : NAND */
   EMIF_CS3 (RW) : origin = 0C00000h length = 0200000h   /* ASYNC2 : NAND */
   EMIF_CS4 (RW) : origin = 0E00000h length = 0100000h   /* ASYNC3 : NOR */
   EMIF_CS5 (RW) : origin = 0F00000h length = 00E0000h   /* ASYNC4 : SRAM */
}
SECTIONS
{
    vectors (NOLOAD)
    .intvec                         : > DARAM_0       ALIGN = 256
    .bss                             : > SARAM        /*, fill = 0 */
    .stack       : > DARAM_1     ALIGN = 1024
    .sysstack                    : > DARAM_1    ALIGN = 1024
    .heap          : > DARAM_1       ALIGN = 1024
    .text            : > SARAM       ALIGN = 1024
    .onchip_code            : > SARAM
   .cinit          : > DARAM_2
   .cio              : > DARAM_2
   .const          : > DARAM_2

// The Bit-Reverse destination buffer data_br_buf requires an address with
// at least 4+log2(FFT_LENGTH) least significant binary zeros

   data_br_buf    : > DARAM_1       ALIGN = 1024
    scratch_buf       : > DARAM_1       ALIGN = 1024
   convolved_buf   : > DARAM_2
   coeffs_fft_buf   : >   DARAM_2

   RcvL1           : > DARAM_3
   RcvL2           : > DARAM_3
   RcvR1           : > DARAM_4
   RcvR2           : > DARAM_4

   FilterOut   : > DARAM_5
   OverlapL   : > DARAM_5
   OverlapR   : > DARAM_5

   XmitL1       : > DARAM_6
   XmitL2       : > DARAM_6
   XmitR1       : > DARAM_7
   XmitR2       : > DARAM_7

   RcvL1_copy   : > SARAM
   RcvL2_copy   : > SARAM
   RcvR1_copy   : > SARAM
   RcvR2_copy   : > SARAM

   .emif_cs0   : > EMIF_CS0
   .emif_cs2   : > EMIF_CS2
   .emif_cs3   : > EMIF_CS3
   .emif_cs4   : > EMIF_CS4
   .emif_cs5   : > EMIF_CS5
}

Following is my console output:

**** Build of configuration Debug for project first_Project ****

E:\TI\ccsv4\utils\gmake\gmake -k all
'Building file: ../csl_irqplug.asm'
'Invoking: Compiler'
"E:/TI/ccsv4/tools/compiler/c5500/bin/cl55" -vcpu:3.3 -g --include_path="E:/TI/xdais_6_25_01_08/packages/ti/xdais" --include_path="E:/C5505/CD/C5505/Chapter 1/Application 1 Getting Started" --include_path="E:/TI/ccsv4/emulation/boards/usbstk5515_v1/include" --include_path="E:/TI/ccsv4/tools/compiler/c5500/include" --include_path="E:/TI/bios_5_41_02_14" --include_path="E:/TI/bios_6_21_00_13" --include_path="E:/C5505/CSL/build" --include_path="E:/C5505/CSL/inc" --include_path="E:/C5505/CSL/src" --diag_warning=225 --ptrdiff_size=32 --algebraic --memory_model=large --asm_source=algebraic --preproc_with_compile --preproc_dependency="csl_irqplug.pp" "../csl_irqplug.asm"
'Finished building: ../csl_irqplug.asm'
' '
'Building file: ../main.c'
'Invoking: Compiler'
"E:/TI/ccsv4/tools/compiler/c5500/bin/cl55" -vcpu:3.3 -g --include_path="E:/TI/xdais_6_25_01_08/packages/ti/xdais" --include_path="E:/C5505/CD/C5505/Chapter 1/Application 1 Getting Started" --include_path="E:/TI/ccsv4/emulation/boards/usbstk5515_v1/include" --include_path="E:/TI/ccsv4/tools/compiler/c5500/include" --include_path="E:/TI/bios_5_41_02_14" --include_path="E:/TI/bios_6_21_00_13" --include_path="E:/C5505/CSL/build" --include_path="E:/C5505/CSL/inc" --include_path="E:/C5505/CSL/src" --diag_warning=225 --ptrdiff_size=32 --algebraic --memory_model=large --asm_source=algebraic --preproc_with_compile --preproc_dependency="main.pp" "../main.c"
"E:\TI\ccsv4\emulation\boards\usbstk5515_v1\include\usbstk5515.h", line 144: warning: typedef name has already been declared (with same type)
"E:\TI\ccsv4\emulation\boards\usbstk5515_v1\include\usbstk5515.h", line 144: warning: typedef name has already been declared (with same type)
"E:\TI\ccsv4\emulation\boards\usbstk5515_v1\include\usbstk5515.h", line 144: warning: typedef name has already been declared (with same type)
"E:/TI/ccsv4/emulation/boards/usbstk5515_v1/include/usbstk5515.h", line 144: warning: typedef name has already been declared (with same type)
"../main.c", line 33: warning: integer conversion resulted in a change of sign
"../main.c", line 34: warning: integer conversion resulted in a change of sign
"../main.c", line 341: warning: last line of file ends without a newline
'Finished building: ../main.c'
' '
'Building file: ../sample_code.c'
'Invoking: Compiler'
"E:/TI/ccsv4/tools/compiler/c5500/bin/cl55" -vcpu:3.3 -g --include_path="E:/TI/xdais_6_25_01_08/packages/ti/xdais" --include_path="E:/C5505/CD/C5505/Chapter 1/Application 1 Getting Started" --include_path="E:/TI/ccsv4/emulation/boards/usbstk5515_v1/include" --include_path="E:/TI/ccsv4/tools/compiler/c5500/include" --include_path="E:/TI/bios_5_41_02_14" --include_path="E:/TI/bios_6_21_00_13" --include_path="E:/C5505/CSL/build" --include_path="E:/C5505/CSL/inc" --include_path="E:/C5505/CSL/src" --diag_warning=225 --ptrdiff_size=32 --algebraic --memory_model=large --asm_source=algebraic --preproc_with_compile --preproc_dependency="sample_code.pp" "../sample_code.c"
At end of source: warning: a translation unit must contain at least one declaration
'Finished building: ../sample_code.c'
' '
'Building file: ../vectors.asm'
'Invoking: Compiler'
"E:/TI/ccsv4/tools/compiler/c5500/bin/cl55" -vcpu:3.3 -g --include_path="E:/TI/xdais_6_25_01_08/packages/ti/xdais" --include_path="E:/C5505/CD/C5505/Chapter 1/Application 1 Getting Started" --include_path="E:/TI/ccsv4/emulation/boards/usbstk5515_v1/include" --include_path="E:/TI/ccsv4/tools/compiler/c5500/include" --include_path="E:/TI/bios_5_41_02_14" --include_path="E:/TI/bios_6_21_00_13" --include_path="E:/C5505/CSL/build" --include_path="E:/C5505/CSL/inc" --include_path="E:/C5505/CSL/src" --diag_warning=225 --ptrdiff_size=32 --algebraic --memory_model=large --asm_source=algebraic --preproc_with_compile --preproc_dependency="vectors.pp" "../vectors.asm"
'Finished building: ../vectors.asm'
' '
'Building target: first_Project.out'
'Invoking: Linker'
"E:/TI/ccsv4/tools/compiler/c5500/bin/cl55" -vcpu:3.3 -g --diag_warning=225 --ptrdiff_size=32 --algebraic --memory_model=large --asm_source=algebraic -z -m"first_Project.map" --warn_sections -i"E:/TI/ccsv4/tools/compiler/c5500/lib" -i"E:/C5505/CSL/first_Project/../../lib" -i"E:/TI/ccsv4/tools/compiler/c5500/include" --reread_libs --rom_model -o "first_Project.out" "./vectors.obj" "./sample_code.obj" "./main.obj" "./csl_irqplug.obj" -l"E:\TI\ccsv4\tools\compiler\c5500\lib\rts55x.lib" -l"E:\C5505\CD\C5505\Chapter 1\Application 1 Getting Started\55xdsph.lib" -l"E:\C5505\CSL\build\cslVC5505.lib" -l"E:\TI\ccsv4\tools\compiler\c5500\lib\lib\usbstk5515bsl.lib" "../C5515.cmd"
<Linking>
'Finished building target: first_Project.out'
' '
Build complete for project first_Project

Antony Rix over 12 years ago in reply to Anand Chandrasekhar

Intellectual 265 points

Anand,

Sounds like you are probably not initialising the device clocks correctly - I suggest you search on this site or take a look at the processor's wiki page for further details of the small number of instructions you need to drop into the start of main() to set it up.

Antony

Processors

Processors forum

Missing Detail in C5515 HWAFFT Documentation