cfft_offplace() missing from 55xdsp.lib; .asm won't build without serious warnings

Hi Brian,

1) There is no cfft_offplace() routine in the dsplib. However, you can perform the complex bit-reversal with the cbrev() routine in either in-place mode or off-place mode.

Usage:    void cbrev (DATA *x, DATA *r, ushort n)

* If x==r, in-place bit reversal is performed.

* Input vector x[ ] and output vector r[ ] must be aligned on 32−bit boundary

E.g.

#pragma DATA_ALIGN(r, 4);             // aligns r to 32-bit boundary

DATA r[2*NX];                                  // contains bit-reversed version of x

* The DSPLIB documentation says that in-place bit-reversal has better performance.

* Make sure you have the current C55x DSPLIB source and documentation:
http://focus.ti.com/docs/toolsw/folders/print/sprc100.html

2) The types of warnings (or remarks) you are seeing are only problematic for specific silicon revisions under the conditions described in the warning. Depending on your DSP, many of these warnings may disappear if you specify a Custom Target to Code Composer. I can help you figure out what to put in the Custom Target field if you tell me what C55x part you are using and what revision it is.

To set the custom target field: Open the Project Build Options > Click on “Compiler” tab > Click on “Basic” under Category > Enter “custom target string” into the Custom Target (-v) field.

For example, the TMS320VC5505 should use custom target “cpu:3.3”.

3) You can access the twiddle memory table by including twiddle.asm with the project source code (located at <dsplib_dir>\twiddle\twiddle.asm).

Remove 55xdsp.lib from the list of included libraries so that your project links to your twiddle.asm instead of the library: Open the Project Build Options > Click on “Linker” tab > Click on “Libraries” under Category > Delete “55xdsp.lib” and any excess semicolons.

You can then place the twiddle code anywhere in memory by specifying
.sect ".twiddle"   (in twiddle.asm) and

.twiddle : {} > SARAM3 PAGE 0  (in the “SECTION” block of linker command file –  cfft.cmd)

Note that the twiddle table must be located in internal memory since it is accelerated by the C55x coefficient bus.

For the best performance:

* Input data in DARAM block.

* Twiddle table in SARAM block or DARAM block different than the DARAM block that contains the input data.

4) If you are looking for significant speed-up with the FFT routine, I recommend trying the TMS320VC5505 (http://focus.ti.com/docs/prod/folders/print/tms320vc5505.html) for its FFT hardware accelerator. The FFT accelerator is 4 ~ 6x more energy efficient and 2.2 ~ 3.8x faster than CPU alone.

Hope this helps,

Mark