Invalid loop bounds in rijndael benchmarks

[Emoun]

The two rijndael benchmarks rijndael_dec and rijndael_enc have invalid loop bounds that mean they cannot be WCET analyzed.

For rijndael_dec, the problem is in aes.c's rijndael_dec_set_key function where a do-while loop (line 90) is given the bounds _Pragma( "loopbound min 0 max 0" ). The same happens again on line 103.
This cannot be ignored as this function is directly called from rijndael_dec_main and so are part of the benchmarked code.
A 0 bound is meaningless and can therefore not be used as a basis for WCET analysis.

rijndael_enc has the exact same problem with its rijndael_enc_set_key function.

The true bounds for these loops must be found so the code can be corrected.

[blisper]

Dear all, I don't know about the Tacle-bench functions, but loop bounds that are zero can indeed be meaningful. Such bounds typically indicate that the loop is dead code that never is executed. Best, Björn Lisper

…

________________________________ From: Emad Jacob Maroun ***@***.***> Sent: Friday, April 5, 2024 4:34 PM To: tacle/tacle-bench ***@***.***> Cc: Subscribed ***@***.***> Subject: [tacle/tacle-bench] Invalid loop bounds in rijndael benchmarks (Issue #44) [EXTERNAL MAIL - MDU] VARNING: Detta epostmeddelande har sitt ursprung utanför MDU. Klicka inte på länkar och öppna inte bifogade filer om du inte känner igen avsändaren och vet att innehållet är säkert. CAUTION: This email originated from outside of MDU. Do not click links or open attachments unless you recognize the sender and know the content is safe. The two rijndael benchmarks rijndael_dec and rijndael_enc have invalid loop bounds that mean they cannot be WCET analyzed. For rijndael_dec, the problem is in aes.c's rijndael_dec_set_keyfunction where a do-while loop (line 90) is given the bounds_Pragma( "loopbound min 0 max 0" ). The same happens again on line 103. This cannot be ignored as this function is directly called from rijndael_dec_main` and so are part of the benchmarked code. A 0 bound is meaningless and can therefore not be used as a basis for WCET analysis. rijndael_enc has the exact same problem with its rijndael_enc_set_key function. The true bounds for these loops must be found so the code can be corrected. — Reply to this email directly, view it on GitHub<#44>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ADBJJSVQAFABCTHNK2JCPRDY32Y57AVCNFSM6AAAAABFZIRPLKVHI2DSMVQWIX3LMV43ASLTON2WKOZSGIZDQMJYGM4DKMA>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Emoun]

@blisper could you elaborate on the usefulness of allowing upper bounds of 0? In which cases would a 0-bounded loop be preferable to just removing the code from the program?
I'm here thinking only of statically known bounds of 0. I of course see the use of dynamic loop bounds (e.g. that an analyzer would find) evaluating to 0 under certain conditions.

In this case specifically, the 0-bounds are set on do-while loops. do-while loops' bodies will be run at least once, contradicting the bound. Our Patmos compiler disallows do-while loops with a static 0-bound so that such contradictions can be caught at compile time. Therefore, I still feel a refinement of the bounds in these programs is preferable.

[blisper]

Hi again all, I was just pointing out that there are situations where zero loop bounds can make sense. Whether they are useful is another issue. And I have no idea whether the rijndael bounds are valid or not. Best, Björn

…

________________________________ From: Emad Jacob Maroun ***@***.***> Sent: Thursday, April 11, 2024 11:25 AM To: tacle/tacle-bench ***@***.***> Cc: Björn Lisper ***@***.***>; Mention ***@***.***> Subject: Re: [tacle/tacle-bench] Invalid loop bounds in rijndael benchmarks (Issue #44) [EXTERNAL MAIL - MDU] VARNING: Detta epostmeddelande har sitt ursprung utanför MDU. Klicka inte på länkar och öppna inte bifogade filer om du inte känner igen avsändaren och vet att innehållet är säkert. CAUTION: This email originated from outside of MDU. Do not click links or open attachments unless you recognize the sender and know the content is safe. @blisper<https://github.com/blisper> could you elaborate on the usefulness of allowing upper bounds of 0? In which cases would a 0-bounded loop be preferable to just removing the code from the program? I'm here thinking only of statically known bounds of 0. I of course see the use of dynamic loop bounds (e.g. that an analyzer would find) evaluating to 0 under certain conditions. In this case specifically, the 0-bounds are set on do-while loops. do-while loops' bodies will be run at least once, contradicting the bound. Our Patmos compiler disallows do-while loops with a static 0-bound so that such contradictions can be caught at compile time. Therefore, I still feel a refinement of the bounds in these programs is preferable. — Reply to this email directly, view it on GitHub<#44 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ADBJJSUWJL6DPBOPMVMXVC3Y4ZJKBAVCNFSM6AAAAABFZIRPLKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANBZGI4DIMZUGM>. You are receiving this because you were mentioned.Message ID: ***@***.***>