Xmc4500 flops

[#1]

Does anybody have some floating point performance measurements at hand? I have testet a loop with a single instruction and got these results:

Instruction / MFLOPS:
a = b * c:

7,5

a += b * c:

6,3

a = b + c:

7,5

a += b + c:

6,3

a = b / c:

4,1

a += b / c:

3,7

The variables a,b,c are floats (32bit). I compiled the test program without optimizations (-O0).
Are these values reasonable?

Best regards, Uli

[Travis]

Uli wrote:
Does anybody have some floating point performance measurements at hand? I have testet a loop with a single instruction and got these results:

Instruction / MFLOPS:
a = b * c:

7,5

a += b * c:

6,3

a = b + c:

7,5

a += b + c:

6,3

a = b / c:

4,1

a += b / c:

3,7

The variables a,b,c are floats (32bit). I compiled the test program without optimizations (-O0).
Are these values reasonable?

Best regards, Uli

This is pretty subjective as calculation equation is different for different user. However it would be good to pull an IO high before calculation and pull the IO Low after the calculation to understand the duration of the calculation.

[#3]

Thanks for your input. Why would this type of calculation be subjective? I can't think of any reasons except compiler options. Or do you have concerns that the method of timing these operations is not reliable?

[Travis]

Uli wrote:
Thanks for your input. Why would this type of calculation be subjective? I can't think of any reasons except compiler options. Or do you have concerns that the method of timing these operations is not reliable?

I guess you are doing some kinds of performance evaluation with the above mentioned calculations. What I mean is that some customer has other method of performance evaluation (eg. Doing square root etc) to suit their needs, therefore it is subjective.

[#5]

I see, but I think the difference is, that the Cortex-M4 has intrinsics for multiply–accumulate operations (MAC), whereas sqrt() has to be calculated in an iterative way. I want to know, if the XMC4500 is well-suited for tasks that involve some filtering. Unfortunately, I haven't got yet requirements to test against. So any help on performance benchmarking is appreciated!

[#6]

Doesn't Infineon provide some kind of DSP performance benchmark for the XMC4000 series?