TMS320F28377S: Error in computations when compared between C28x results and CLA results.

Asad Hameed

I have just started working on the CLA system. To validate if I have setup everything correctly for the CLA, I did some calculations using test vectors. The calculations were done on C28x processor and the CLA processor. It is observed that there is some error between the results obtained from both processors. To have a reference for comparison the same calculations were done in the excel sheet. The formula is as given below:

Dout =( ((Var1 - Var2)*(70/6.25))-Var3)/360; where var1,var2 and var3 are test vectors of 500 values each.

The C28x code implementation is as below:

D_C28[i]= (((Iref[i]-Iinv[i])*11.2)-Van[i])/360;

The same computation is performed in the CLA task as shown below:

Out1 =(((IBatt-V_batt)*(11.2))-Vbus)/360.0;

The CLA task is called 500 times from the main loop by passing values of IBatt (same as Iref),V_batt (same as Iinv)and Vbus (same as Van) defined in CPU to CLA message RAM. The output Out1 ,defined in CLA to CPU message RAM is read and stored in an array D_CLA[500] for comparing with D_C28[500].

The difference between the C28x and excel results is as below:

The difference between the CLA and excel results is as below:

The actual output as computed by excel is as below:

Apparently there is significant error in the CLA calculations. Please guide if I might be doing something incorrectly or if there is logical explanation to this error.

Note: I am using the CLAmath library: cla1_math_library_datarom_fpu32_eabi.lib & C2000Ware4.1.0.0. The processor options in the project are as below:

over 3 years ago

0 Veena Kamath over 3 years ago

TI__Mastermind 32695 points

Hi Asad Hameed,

Can you please provide one set of sample values to replicate the issue at my end?

Regards,

Veena

0 Asad Hameed over 3 years ago in reply to Veena Kamath

Intellectual 735 points

Hi Veena Kamath,

Please refer to the table below. I have highlighted cases where CLA has shown significant error. Let me know if more information is required.

Sr.#	Iref	Iinv	Van	CLA Computed	Excel	CLA Diff.	C28 Computed	C28 Diff.
1	0	0.109474182	5.527175903	-0.018759126	-0.01875913	4.04033E-09	-0.01875913	2.40333E-10
2	0.001800537	0.242996216	11.06726074	-0.038246252	-0.038246257	4.90372E-09	-0.038246255	1.10372E-09
3	-0.033561707	0.401107788	16.59013367	-0.05960675	-0.059606756	5.98324E-09	-0.059606753	2.28324E-09
4	0.36013794	0.572662354	22.11946106	-0.068054803	-0.068054818	1.53467E-08	-0.068054818	4.46667E-10
5	0.955368042	0.736412048	27.62727356	-0.069930449	-0.069930462	1.32071E-08	-0.069930464	-1.69289E-09
6	1.555969238	0.880348206	33.13723755	-0.071028553	-0.071028561	8.17533E-09	-0.07102856	6.75333E-10
7	2.03629303	0.999649048	38.62138367	-0.075030468	-0.075030475	6.79578E-09	-0.075030476	-7.04222E-10
8	2.345573425	1.096282959	44.1033783	-0.083642565	-0.08364257	4.96889E-09	-0.083642565	4.96889E-09
9	2.484107971	1.174629211	49.55525208	-0.09691412	-0.096914139	1.88E-08	-0.096914142	-3.6E-09
10	2.496147156	1.240829468	55.00067139	-0.13285768	-0.113725315	-0.019132365	-0.113725312	2.67889E-09
11	2.423217773	1.299552917	60.41381836	-0.152732983	-0.1328577	-0.019875283	-0.132857695	4.92444E-09
12	2.32169342	1.35471344	65.8140564	-0.152732983	-0.152733002	1.87333E-08	-0.152732998	3.73333E-09
13	2.222137451	1.407653809	71.17771912	-0.172376364	-0.172376395	3.136E-08	-0.172376394	1.36E-09
14	2.140731812	1.459281921	76.52632141	-0.191372424	-0.191372452	2.77522E-08	-0.191372439	1.27522E-08
15	2.077789307	1.509109497	81.83404541	-0.209624499	-0.209624532	3.305E-08	-0.209624514	1.805E-08
16	2.039047241	1.55696106	87.12025452	-0.227002442	-0.22700247	2.82578E-08	-0.227002457	1.32578E-08
17	2.020942688	1.601631165	92.36343384	-0.243519828	-0.243519847	1.86178E-08	-0.243519843	3.61778E-09
18	2.010910034	1.642799377	97.58079529	-0.259605408	-0.259605433	2.51433E-08	-0.259605408	2.51433E-08
19	1.995674133	1.67993927	102.7486725	-0.275590092	-0.275590117	2.47622E-08	-0.275590092	2.47622E-08
20	1.983726501	1.713516235	107.8907318	-0.291289896	-0.291289936	3.96133E-08	-0.291289926	9.61333E-09
21	1.969070435	1.742958069	112.9768524	-0.306789935	-0.306789983	4.80578E-08	-0.306789994	-1.09422E-08
22	1.945930481	1.769058228	118.0350037	-0.322372288	-0.322372318	2.99622E-08	-0.322372288	2.99622E-08
23	1.916435242	1.791725159	123.0329132	-0.353095263	-0.337878223	-0.01521704	-0.337878227	-4.02667E-09
24	1.89037323	1.81161499	127.9963989	-0.353095263	-0.353095296	3.31444E-08	-0.353095293	3.14444E-09
25	1.863014221	1.828300476	132.8974915	-0.368079692	-0.368079715	2.34333E-08	-0.368079722	-6.56667E-09
26	1.836868286	1.842330933	137.7598572	-0.382836163	-0.382836219	5.60178E-08	-0.382836223	-3.98222E-09
27	1.807197571	1.853416443	142.5576782	-0.397431403	-0.397431471	6.80178E-08	-0.397431463	8.01778E-09
28	1.776725769	1.862136841	147.310318	-0.425944746	-0.411852561	-0.014092185	-0.411852568	-6.87111E-09
29	1.748214722	1.868202209	151.9962616	-0.425944746	-0.425944782	3.58178E-08	-0.425944775	6.81778E-09

0 Veena Kamath over 3 years ago in reply to Asad Hameed

TI__Mastermind 32695 points

Hi,

I am unable to recreate the issue.

I added the following code in C28x -

float IBatt = 2.496147156;
float V_batt = 1.240829468;
float Vbus = 55.00067139;

fVal = (((IBatt-V_batt)*(11.2))-Vbus)/360.0;

and following in CLA -

float IBatt = 2.496147156;
float V_batt = 1.240829468;
float Vbus = 55.00067139;

fResult = (((IBatt-V_batt)*(11.2))-Vbus)/360.0;

Result -

0 Asad Hameed over 3 years ago in reply to Veena Kamath

Intellectual 735 points

There is only one difference between your test and mine. I passed the values to CLA using message RAMS instead of directly defining in the CLA. Let me try with your approach and compare results with passing through message RAM. The data was captured using the watch window of debug mode with XDS110 debug probe. I assume that it should not be an issue. Let me do this and then get back to you.

+1 Asad Hameed over 3 years ago in reply to Asad Hameed

Intellectual 735 points

Hi Veena,

I have figured out the issue. It was an overlook at my end. I was updating values in the CLA Message RAMs and calling the CLA task back to back without waiting for the previous CLA task completion. Now I have added an exaggerated delay of around 10uSec between the two CLA task calls and the results match. Some results are shown below:

Sr.#	C28	CLA
1	-0.01875913	-0.018759126
2	-0.038246255	-0.038246252
3	-0.059606753	-0.05960675
4	-0.068054818	-0.068054803
5	-0.069930464	-0.069930449
6	-0.07102856	-0.071028553
7	-0.075030476	-0.075030468
8	-0.083642565	-0.083642565
9	-0.096914142	-0.09691412
10	-0.113725312	-0.113725305
11	-0.132857695	-0.13285768
12	-0.152732998	-0.152732983
13	-0.172376394	-0.172376364
14	-0.191372439	-0.191372424
15	-0.209624514	-0.209624499
16	-0.227002457	-0.227002442
17	-0.243519843	-0.243519828
18	-0.259605408	-0.259605408
19	-0.275590092	-0.275590092
20	-0.291289926	-0.291289896

There are some minor differences between the two results but I think that these are acceptable. Is it logical to expect exactly same outputs from both processors or some differences can be expected?. Otherwise we may close this thread if the results are fine.

0 Veena Kamath over 3 years ago in reply to Asad Hameed

TI__Mastermind 32695 points

Hi,

I just realized that I had used fp_mode = strict. Changing to relaxed to match your setting, I too see minor differences

You can find more details about this option in the compiler guide. It would be faster, but might affect the accuracy.

Regards,

Veena

C2000™︎ microcontrollers

C2000 microcontrollers forum

TMS320F28377S: Error in computations when compared between C28x results and CLA results.