Page 3 of 10
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 8:31 am
by Pito
Nope. No other special settings.
1. the MFLOPS numbers we get are still a bit high to my taste.. But it could be the gcc makes miracles.. I would expect MFLOPs at 1/3 or 1/4 of clock. I got 82MFlops at 168MHz, that is really unbelievable high..
2. the -O3 is not good as it optimizes off the N3 measurement (seee the result).
I would better stay with -O2
3, The FPU is on, it seems, based on the results.
4, there is a link with hundreds of results off this test, you may see F407/F411 outperforms Pentium/Cyrix/AMD chips from 90ties..
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 8:36 am
by Bingo600
W. STM Core (untouched settings) and a F4011 i get -O2 : 40 30 30
I don't know the clock settings (yet)
Seems a bit more realistic , the FPU doc said all FLOPS (Except divide) was single cycle on a M4.
But as you mentioned .. We also have to load/save the results ... Takes cycles too.
How do you know your F407 runs 168MHz ??
/Bingo
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 8:45 am
by Pito
F407 is usually set to 168MHz, so it is my assumption the STM core sets that.
You clock looks like 72MHz.
The FPU does in 1 or 3 or 14 cycles, 2+ cycles is load/store, I think it has got some combined/fused operations like MAC (3 cycles)..
Not sure such FPU optimised operations are available off gcc, there are special CMSIS libraries available for DSP and FPU advanced topics, see below.
DSP+FPU intro:
https://www.youtube.com/watch?v=tGMsLvz7bBY
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 9:45 am
by Bingo600
ST Core
Just dug out an original STM F407-Discovery (Old board)
-Os
Code: Select all
##########################################
Single Precision C Whetstone Benchmark
Calibrate
0.14 Seconds 1 Passes (x 100)
0.71 Seconds 5 Passes (x 100)
3.57 Seconds 25 Passes (x 100)
Use 70 passes (x 100)
Single Precision C/C++ Whetstone Benchmark
Loop content Result MFLOPS MOPS Seconds
N1 floating point -1.12475013732910156 79.059 0.017
N2 floating point -1.12274742126464844 57.366 0.164
N3 if then else 1.00000000000000000 83.276 0.087
N4 fixed point 12.00000000000000000 253.449 0.087
N5 sin,cos etc. 0.49909299612045288 1.284 4.537
N6 floating point 0.99999982118606567 15.739 2.399
N7 assignments 3.00000000000000000 13.617 0.950
N8 exp,sqrt etc. 0.75110614299774170 1.492 1.745
MWIPS 70.098 9.986
-O2
Code: Select all
##########################################
Single Precision C Whetstone Benchmark
Calibrate
0.10 Seconds 1 Passes (x 100)
0.51 Seconds 5 Passes (x 100)
2.57 Seconds 25 Passes (x 100)
Use 97 passes (x 100)
Single Precision C/C++ Whetstone Benchmark
Loop content Result MFLOPS MOPS Seconds
N1 floating point -1.12475013732910156 81.684 0.023
N2 floating point -1.12274742126464844 57.685 0.226
N3 if then else 1.00000000000000000 167.325 0.060
N4 fixed point 12.00000000000000000 315.000 0.097
N5 sin,cos etc. 0.49909299612045288 1.292 6.247
N6 floating point 0.99999982118606567 59.255 0.883
N7 assignments 3.00000000000000000 71.990 0.249
N8 exp,sqrt etc. 0.75110614299774170 1.654 2.182
MWIPS 97.323 9.967
-O3
Code: Select all
##########################################
Single Precision C Whetstone Benchmark
Calibrate
0.10 Seconds 1 Passes (x 100)
0.50 Seconds 5 Passes (x 100)
2.49 Seconds 25 Passes (x 100)
Use 100 passes (x 100)
Single Precision C/C++ Whetstone Benchmark
Loop content Result MFLOPS MOPS Seconds
N1 floating point -1.12475013732910156 83.843 0.023
N2 floating point -1.12274742126464844 64.306 0.209
N3 if then else 1.00000000000000000 0.000 0.000
N4 fixed point 12.00000000000000000 362.069 0.087
N5 sin,cos etc. 0.49909299612045288 1.330 6.255
N6 floating point 0.99999982118606567 59.275 0.910
N7 assignments 3.00000000000000000 83.620 0.221
N8 exp,sqrt etc. 0.75110614299774170 1.655 2.248
MWIPS 100.473 9.953
Had to put a delay(3000); , after the last SerialUSB.printf("%39.3f%19.3f\n\n", mwips, TimeUsed);
Else it didn't show ... It's prob bad karma to have an empty loop()
/Bingo
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 9:51 am
by Bingo600
@Pito
Your (-O2) N2 is quite different from mine
viewtopic.php?p=891#p891
You get 86 , i get 81
/Bingo
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 9:52 am
by Pito
What device have you selected for the Discovery F407?
Try with BlackF407ZE
It could be the boards have different gcc setting..
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 10:03 am
by Bingo600
I selected Disco
- Disco.png (49.61 KiB) Viewed 11586 times
Hmm the forum doesn't like Colons in the upload filename
Menu_2020012608:18:35.png is flagged as invalid
/Bingo
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 10:07 am
by Bingo600
I have a Black F407
https://www.aliexpress.com/item/4000103742703.html
How do you upload to that one ?
DFU or SWD ??
I t has some buttons : (Left) RST + K_UP - (Right) KD + ??? (bad silkscreen)
I tried RST + K_UP (thought upload) .. But it doesnt go into DFU mode.
Edit: Ahh K_UP = Wakeup
I'm lazy - SWD means read pin-doc
/Bingo
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 10:14 am
by Bingo600
- 407VE.png (57.37 KiB) Viewed 11584 times
-O2
Code: Select all
##########################################
Single Precision C Whetstone Benchmark
Calibrate
0.10 Seconds 1 Passes (x 100)
0.51 Seconds 5 Passes (x 100)
2.57 Seconds 25 Passes (x 100)
Use 97 passes (x 100)
Single Precision C/C++ Whetstone Benchmark
Loop content Result MFLOPS MOPS Seconds
N1 floating point -1.12475013732910156 81.327 0.023
N2 floating point -1.12274742126464844 57.431 0.227
N3 if then else 1.00000000000000000 167.325 0.060
N4 fixed point 12.00000000000000000 315.000 0.097
N5 sin,cos etc. 0.49909299612045288 1.292 6.246
N6 floating point 0.99999982118606567 59.255 0.883
N7 assignments 3.00000000000000000 71.702 0.250
N8 exp,sqrt etc. 0.75110614299774170 1.654 2.181
MWIPS 97.322 9.967
Re: dhrystone and whetstone benchmarks
Posted: Sun Jan 26, 2020 10:20 am
by Bingo600
- F407ZE.png (57.3 KiB) Viewed 11583 times
F407ZE -O2
Code: Select all
##########################################
Single Precision C Whetstone Benchmark
Calibrate
0.10 Seconds 1 Passes (x 100)
0.51 Seconds 5 Passes (x 100)
2.57 Seconds 25 Passes (x 100)
Use 97 passes (x 100)
Single Precision C/C++ Whetstone Benchmark
Loop content Result MFLOPS MOPS Seconds
N1 floating point -1.12475013732910156 81.327 0.023
N2 floating point -1.12274742126464844 57.685 0.226
N3 if then else 1.00000000000000000 167.325 0.060
N4 fixed point 12.00000000000000000 315.000 0.097
N5 sin,cos etc. 0.49909299612045288 1.292 6.246
N6 floating point 0.99999982118606567 59.255 0.883
N7 assignments 3.00000000000000000 71.990 0.249
N8 exp,sqrt etc. 0.75110614299774170 1.654 2.182
MWIPS 97.332 9.966