Compiler/TMS320F28075: Error in composing a long from 2 words

Some more information.
I compile with the following optimization options:

Optimization level (--opt_level, -O): 2 - Global Optimization
Speed vs. size trade-offs ( --opt_for_speed, -mf): 4

Hello Demis, 

Demis Biscaro said:

    dataC = ((((unsigned long)dataA) & 0x0000FFFFUL) << 8) | (dataB & 0x00FF);

Try: dataC = ((unsigned long) dataA << 16) | (unsigned long) dataB;

Hello Tom,
thank you for your answer, but I don't understand it.

I don't want to shift dataA by 16 bits but only by 8 bits, because I want insert only the least significant byte of dataB.

Why should I shift dataA by 16 bits?

Best regards,

Demis

I've changed the function in the following way

void ApplGetParamInfo(unsigned short dataA, unsigned short dataB)
{
volatile unsigned long dataC; // <----------------------------- VOLATILE attribute inserted!



/* DataA is cast from word (16 bits) to long (32 bits);
Data A is masked and left-shifted by 8 bit;
DataB is masked;
DataA and DataB are composed together with a bitwise OR and the result is assigned to dataC
*/

dataC = ((((unsigned long)dataA) & 0x0000FFFFUL) << 8) | (dataB & 0x00FF);

/* Other stuff */

}

and now I get the correct result, i.e. if I have

dataA = 0x1234;

dataB = 0x0056;

I get

dataC = 0x00123456;

I think maybe there's an issue in compiler beahviour when those optimization options are active.

Isn't it?

Best Regards,

Demis

I am unable to reproduce the problem.  For the source file which contains the problem expression, please follow the directions in the article How to Submit a Compiler Test Case.

Thanks and regards,

-George

I have not checked the comments in your code. My mistake.

George Mock said:

I am unable to reproduce the problem.  For the source file which contains the problem expression, please follow the directions in the article How to Submit a Compiler Test Case.

Thanks and regards,

-George

Hello George,

thank you for the link, it's the first time I report a (seeming) compiler issue, so I didn't know there was a specific procedure to report it.

The original source file is Appl.c so I've generated the Preprocess source file Appl.pp (I've renamed as Appl_pp.txt because the forum doesn't let me to attach a .pp file...):

/cfs-file/__key/communityserver-discussions-components-files/81/Appl_5F00_pp.txt

 

Then I've copied the output from the build console to get all Compiler options. I've pasted to compiler_options_Appl.txt:

/cfs-file/__key/communityserver-discussions-components-files/81/compiler_5F00_options_5F00_Appl.txt

 

Finally the Compiler version is:

v16.9.9.LTS

You find the critical instruction in file App.pp at line 31662:

    parameterIndexSubindex = ((((UINT32)parameterIndex) & 0x0000FFFFUL) << 8) | (parameterSubindex & 0x00FF);

If I have

parameterIndex = 0x1234

parameterSubindex = 0x0056

I get

parameterIndexSubindex = 0x00003456

instead of

parameterIndexSubindex = 0x00123456

which is tha value I expected.

One more consideration.

As you can see in the file App.pp, parameterIndexSubindex is defined as follows:

    UINT32 parameterIndexSubindex;

However, yesterday I've made some tests to find a temporary workaround. I've found that if I define parameterIndexSubindex in the following way

  volatile UINT32 parameterIndexSubindex;

the issue disappears, i.e. I get the expected value

parameterIndexSubindex = 0x00123456

This led me to think of it as a compiler problem, and specifically as an optimization options problem.

Anyway, I hope you have all the information to reproduce the issue on your system.

Best regards,

Demis

Thank you for the test case.  Even though I have everything I need, I am still unable to work out what happened.  Before I pass this issue on to other experts, I have these questions.

Exactly how do you see the problem?  Is it possible that this is only a problem in how CCS displays the values?  If you run all the way through the function with the problem statement, does the caller to the function act as if it received bad results?

Thanks and regards,

-George

Hello George

George Mock said:

xactly how do you see the problem?  Is it possible that this is only a problem in how CCS displays the values?  If you run all the way through the function with the problem statement, does the caller to the function act as if it received bad results?

Yes, it does, unfortunately.

I use parameterIndexSubindex as a key to seek a specific item in a list, but the function doesn't find it because the key value is wrong.

I stumbled on this issue just because the caller to the function got a bad result: "item not found".

On the opposite, when I put the "volatile" attribute in parameterIndexSubindex definition the function finds the item, the caller gets the correct result and the software runs correctly.

To be sincere, I should add the "volatile" attribute both to parameterIndexSubindex and dictionaryParameterIndexSubindex, otherwise even dictionaryParameterIndexSubindex has a wrong value, i.e. the assignment

dictionaryParameterIndexSubindex = CAN_VOCAB[middleIndex].Indirizzo;

transfers only the least significant word from CAN_VOCAB[middleIndex].Indirizzo to dictionaryParameterIndexSubindex.

For instance, suppose I define

 UINT32 dictionaryParameterIndexSubindex;

and I have

CAN_VOCAB[middleIndex].Indirizzo = 0x00123456

After the assignment

dictionaryParameterIndexSubindex = CAN_VOCAB[middleIndex].Indirizzo;

I get

dictionaryParameterIndexSubindex = 0x00003456

instead of

dictionaryParameterIndexSubindex = 0x00123456

Therefore I should define

    volatile UINT32 parameterIndexSubindex;
    volatile UINT32 dictionaryParameterIndexSubindex;

to get the whole function run correctly.

Very, very odd.

I attach a file with the assembly code I get when I compile with and without the "volatile" attribute.

/cfs-file/__key/communityserver-discussions-components-files/81/assembly_5F00_code.txt

Don't hesitate to contact me if you need additional information.

Thank you.

Best regards,

Demis

Hello George,
I've done some more tests and I've made a recap of the whole issue history.

Some days ago, I saw that the function returned "item not found", so I inspected variable contents in "Variables" watch window and I found that parameterIndexSubindex and dictionaryParameterIndexSubindex had wrong values.

Then I compiled the funcion with the "volatile" attribute and in this way I saw the expected values of parameterIndexSubindex and dictionaryParameterIndexSubindex in "Variables" watch window.
However, function returned "item not found" because there was no item with the 0x00123456 key in my list.

So I added the item with the 0x00123456 key in my list and I saw that function returned "item found" correctly.

This morning I've removed the "volatile" attribute and I've done some new tests.
I've seen that function returns "item found" correctly even if I see the wrong values of parameterIndexSubindex and dictionaryParameterIndexSubindex in "Variables" watch window!

Then I've defined a global array

volatile INT32 sd[32];

and I've put these assignment in my function

...
...
parameterIndexSubindex = ((((UINT32)parameterIndex) & 0x0000FFFFUL) << 8) | (parameterSubindex & 0x00FF);

sd[10] = parameterIndexSubindex; // <---------- new assignment
...
...
dictionaryParameterIndexSubindex = CAN_VOCAB[middleIndex].Indirizzo;

sd[11] = dictionaryParameterIndexSubindex; // <--------------------- new assignment
...
...

In Expressions watch window I've seen
sd[10] = 0x00123456
sd[11] = 0x00123456

i.e. the correct values!


In conclusion:
- code works correctly both with and without the "volatile" attribute;
- Variables watch window shows the correct values only with the "volatile" attribute.

Then you're right, the issue isn't about the compiler but it concerns only Variables watch window.

I do apologise for my mistake, I'm verry sorry.

Thank you very much for your support.

Best regards,

Demis

Hello Ki-Soo,
I've reported your request to my boss.
As you can easily imagine I can't send you the whole project because it's confidential. To strip this project isn't so easy because I've got a lot of tasks to accomplish and the deadline is getting closer and closer. I'll try to find some time to do it but I can't promise you that.

Thanks.

Demis

I understand your situation - it is a common one because of IP and the not so trivial task of stripping down a full project. If you can find the time to put together a test case, it would be greatly appreciated.

Thanks
ki