Cortex-M4 DSPlib

Yes Agreed.... I look forward to it too.

Clement

Hey all,

We do have CMSIS support, which includes CMSIS-DSP. The CMSIS_DSP library contains over 60 functions in fixed-point (fractional q7, q15, q31) and single precision floating-point (32-bit) implementation. 

Take a look at this page for further explaination:

http://www.ti.com/tool/cmsis_dsp_headers

as well as this app note to help you understand how to use it:

http://www.ti.com/litv/pdf/spma041a

If you would like to learn some more here is a link to ARMs CMSIS page:

http://www.arm.com/products/processors/cortex-m/cortex-microcontroller-software-interface-standard.php

Hello,

I think that CMSIS-DSP library support is a good idea but as mentioned in the performance tests 1024-point FFT 16 bit on the Cortex-M4 processors @ 25MHz (see link 1) consumes same time as implementatuon on the Cortex-M3 processor @ 25MHz (see link 2). So, in this regard, it is zero profit when migrating to the Cortex-M4.

See:

LINK1: Designing  advanced  DSP  applications  on  the Kinetis ARM Cortex-M4 MCU—Part 1  -

http://www.iar.com/Global/Resources/Developers_Toolbox/Building_and_debugging/Designing_advanced_DSP_applications_on_the_Kinetis_ARM_Cortex-M4_MCU_part1.pdf

LINK2: UM0585: STM32F10x DSP library   -  

http://www.st.com/internet/com/TECHNICAL_RESOURCES/TECHNICAL_LITERATURE/USER_MANUAL/CD00208762.pdf

For TI's Cortex-M3 processor I found highly optimized DSP library here (it performs 1024-point FFT 16 bit in 5 ms):

DSP Library  -

LINK3: http://www.embeddedsignals.com/ARM.htm

Performance tests -

http://www.embeddedsignals.com/FFTlibrary2bench.pdf

Alexander,

Please take a look at this document from ARM which is not vendor specific and among other things compares DSP performance on an M3 and M4 (page 7).

In all test case the M4 out performed the M3:

http://www.arm.com/files/pdf/at-_tools_for_cortex-m4.pdf

Also if you are referring to this table:

Then you may have misread it. It clearly shows that the M4 is 2.4x faster than the M3 when using the arm specific FFT.

Please let me know if I am missing something.

Dave,

I am completely agree with you. Nevertheless, I am just trying to point out that:

Cortex-M4 performs calculations of CMSIS_FFT for the time equals to that Cortex-M3 performs calculation of non CMSIS FFT like in STM32 DSPLib, NXP DSPLib or DSPLib from Embedded Signals Team, which are assembler written and highly optimized.

Of course M4 performance exceeds that M3. But while assembler written DSPLib for Cortex-M4 is not exists the users of Cortex-M3 employing vendor specific DSPLibs mentioned above have not any profit when migrating to Cortex-M4 employing only available CMSIS_DSPLib.