Vybrid FPU Floating point performance?

キャンセル
次の結果を表示 
表示  限定  | 次の代わりに検索 
もしかして: 

Vybrid FPU Floating point performance?

ソリューションへジャンプ
2,502件の閲覧回数
future_boston
Contributor II

Have any floating point metrics for Vybrid VF6?  Peak MFLOPs using register-based data?

using SRAM source/destination? using DRAM source/destination? performance of

single precision vs double precision impact due to bus bandwidth? Floating point comparison of VF6 relative to

iMX6, MobilGT, Qorivva, or QorIQ?

ラベル(5)
1 解決策
1,608件の閲覧回数
michaeljohnson
Contributor II

Hi Ioseph,

It's just Whetstone. I ran it using CrossWorks (www.rowley.co.uk) on the Vybrid TWR board. I would expect similar numbers (certainly ratios) using IAR/Keil etc.

Regards

Michael

元の投稿で解決策を見る

0 件の賞賛
返信
11 返答(返信)
1,608件の閲覧回数
jackdean
Contributor I

Hi,

I cannot reproduce these results.

If I run whetstone at the 396MHz, I can only run around 4000 loops in 53 seconds.

None of these results are reproducible.

Would you please post how you were able to get 10,000 loops in 53 seconds?

Even at 500 MHz it won't do more than about 5500 Whetstone loops in 53 seconds.

Thank you,

0 件の賞賛
返信
1,608件の閲覧回数
michaeljohnson
Contributor II

Hi Mike,

Not sure if this helps but anyhoo

//Use 396MHz CA5, 396 MHz DDR, 132MHz CM4, and 66MHz bus

#if (CLOCK_SETUP == 0)

Double/VFP

Loops: 10000, Iterations: 1, Duration: 6 sec.

C Converted Double Precision Whetstones: 166.7 MIPS

Float/VFP

Loops: 10000, Iterations: 1, Duration: 4 sec.

C Converted Double Precision Whetstones: 250.0 MIPS

Double/No VFP

Loops: 10000, Iterations: 1, Duration: 53 sec.

C Converted Double Precision Whetstones: 18.9 MIPS

Float/No VFP

Loops: 10000, Iterations: 1, Duration: 24 sec.

C Converted Double Precision Whetstones: 41.7 MIPS

So the VFP works...

The numbers are the same running in OCRAM/DDR and QuadSPI - so the cache/MMU works as well...

Regards

Michael

1,608件の閲覧回数
ioseph_martinez
NXP Employee
NXP Employee

Michael, could you share the project where you did these tests? would be nice to have a sample to work with this benchmark.

Thanks,

Ioseph

0 件の賞賛
返信
1,609件の閲覧回数
michaeljohnson
Contributor II

Hi Ioseph,

It's just Whetstone. I ran it using CrossWorks (www.rowley.co.uk) on the Vybrid TWR board. I would expect similar numbers (certainly ratios) using IAR/Keil etc.

Regards

Michael

0 件の賞賛
返信
1,608件の閲覧回数
karina_valencia
NXP Apps Support
NXP Apps Support

ioseph_martinez canyou continue with the follow up?

0 件の賞賛
返信
1,608件の閲覧回数
ioseph_martinez
NXP Employee
NXP Employee

Karina, Michael Johnson is the right answer.

We don't have comparison tables for the other products.

0 件の賞賛
返信
1,608件の閲覧回数
karina_valencia
NXP Apps Support
NXP Apps Support

Get it, thanks.

0 件の賞賛
返信
1,608件の閲覧回数
karina_valencia
NXP Apps Support
NXP Apps Support

timesyssupport can you help on this case?

0 件の賞賛
返信
1,608件の閲覧回数
timesyssupport
Senior Contributor II

At this time, these performance metrics have not been collected. We are currently reviewing the scope and timeline for providing such metrics. As they pertain to other Freescale families, would you have metrics available on those karinavalencia?

Regards,

Timesys Support

0 件の賞賛
返信
1,608件の閲覧回数
yibinjin
Contributor III

hi,how do i know if VFP is working?

if i compile my test funtion as follows : (i did not enable Enable ARM hard-float support in the desktop Factory)

armv7l-timesys-linux-gnueabi-gcc -static vfp.c -lm -o vfp

and run it in my board , does the VFP works ? or just soft FP works ?

thanks so much

0 件の賞賛
返信
1,608件の閲覧回数
karina_valencia
NXP Apps Support
NXP Apps Support

Mike, about   your question:  Floating point comparison of VF6 relative to iMX6, MobilGT, Qorivva, or QorIQ? I dont  think we have this  information. ioseph_martinez do you know if  we have this kind of  details?

0 件の賞賛
返信