Float point performance

Jiri Dobry

Hello,

I am not able found any information about Floating point unit. TMS570 datasheets have not any information about it.

Generic ARM documentation about Cortex R4F tell me only information from programmers point of view. It is VFPv3-D16 instruction/feature set, implementation is optimized for single precision.

But nothing about performance. And we need some information about used VFP performance BEFORE we will start project.

Do anybody have some more information or measurement (ex MFLOPS)?? Single and double precision. Some URL?

Many thanks.

over 13 years ago

0 Jiri Dobry over 13 years ago

Expert 1810 points

Another relevant problem is RunFast Mode of VFPv3. What it impact to performance?

0 Zhaohong Zhang over 13 years ago in reply to Jiri Dobry

TI__Mastermind 22715 points

Jiri,

Thank you very much for your interest in TI devices. I have forwarded you question to TI CPU/architecture expert and will get back to you as soon as possible.

Thanks and regards,

Zhaohong

0 B Chavali over 13 years ago in reply to Zhaohong Zhang

TI__Prodigy 255 points

Hi Jiri,

Documentation on Cortex R4F VFP can be found in R4 TRM. Cortex R4F supports both single precision and double precision.

ARM R4 TRM is available at http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0363e/index.html .

TRM also documents cycle timing behavior of floating point instructions. Cycle count information is present in Chapter 14. Cycle Timings and Interlock Behavior. Additional information can be found in ARM architecture reference manual.

We currently dont have the MFLOPS benchmark results. Can you explain what are your key care abouts with VFP?

Also our compiler 5.0 beta has significant enhancements and should be used for float benchmarking.

Sincerely,

B Chavali

0 Jiri Dobry over 13 years ago in reply to B Chavali

Expert 1810 points

Many thanks to answer. I was not found VFP instruction timing in chapter 14 before (thanks again).

Explanation for my VFP care is simple. We have application with FFT and/or some mathematical models. And we need to know if this platform have sufficient power before we start implementation.

Arm-based microcontrollers

Arm-based microcontrollers forum

Float point performance