Global state in SystemC

[nmeum]

I am currently working on a specialized SystemC project where I want to restart the SystemC simulation process multiple times. I have a working proof of concept which does this by providing a custom main function in which I reset the global SystemC simulation context (sc_core::sc_curr_simcontext) and call sc_core::sc_elab_and_sim in a loop. As an example, consider the following code:

int main(int argc, char **argv) {
        // …

	for (i = 0; !exit_simulation; i++) {
		printf("\n##\n# %zuth execution\n##\n", i);

		// Reset SystemC simulation context
		if (sc_core::sc_curr_simcontext)
			delete sc_core::sc_curr_simcontext;
		sc_core::sc_curr_simcontext = NULL;

		if ((ret = sc_core::sc_elab_and_sim(argc, argv)))
			return ret;
	}

        // …
}

I am aware that this code relies on specific implementation details of the sc_get_curr_simcontext() function and is thus not conforming to IEEE Std 1666-2001. In fact, the code above only works because sc_get_curr_simcontext() currently creates a new simulation context if the global sc_core::sc_curr_simcontext is NULL:

systemc/src/sysc/kernel/sc_simcontext.h

Lines 420 to 423 in c81e0d9

	if( sc_curr_simcontext == 0 ) {
	sc_default_global_context = new sc_simcontext;
	sc_curr_simcontext = sc_default_global_context;
	}

Apart from relying on implementation-defined behaviour, an additional problem with the code above is that SystemC contains global variables which are independent of the simulation context. Some of these variables even hold references to the simulation context which cannot be updated and thus cause a double free on program termination. One example of such variables is the sc_event::none variable which was added in 55da81d:

systemc/src/sysc/kernel/sc_event.h

Lines 297 to 298 in 55da81d

	// never notified event
	static const sc_event none;

As sc_event::none is a global variable, it is initialized on program startup and holds a reference to a simulation context (created through the invocation of sc_get_curr_simcontext() by the sc_event constructor). This reference is not updated by the loop from above, in fact in cannot be updated as the m_simc member of sc_event is private. An additional global variable with the same problem is sc_process_handle::non_event:

systemc/src/sysc/kernel/sc_process.cpp

Line 50 in f567c9c

sc_event sc_process_handle::non_event( sc_event::kernel_event );

I am currently considering working on a modified version of SystemC which addresses issue related to such global variables. I created this issue to figure out if there is an interest in reducing global state in SystemC and if patches in this regard would be accepted.

[mattgately]

I agree that this is a useful feature request. I am also doing something where I reset the simcontext to enable rerunning of the simulation kernel.

[wltr]

Has there be any progress on this? Not being able to restart the simulation is the biggest limitation of SystemC imho and I'd be very interested in this.

[nmeum]

We are using the patches from #9 as part of a SystemC-based symbolic execution engine, there is some room for improvement but the general idea does seem to work.

[AndrewGoodrich]

There are number of global structures that really should be reset when trying to use a "fresh" simulation context. I think we might be better off if we had a reinitialize() method in the sc_simcontext class, that would allow the code to properly reinitialize those global structures.

[thatTPR]

Is this solved? If not I would appreciate pointers on helping make this feature available.

[maehne]

@AndrewGoodrich: Might you be able to provide some pointers to @thatTPR?

[AndrewGoodrich]

The original mention of the issue is in #8 in accelera-official. It is issue #435 in OSCI-wg/systemc. accelera-official issue #20 describes what is in internal #435. I closed #20 when I created #435. I have code with the change present, and will generate a pull request in the near future. Andy

…

On Nov 11, 2021, at 5:43 PM, Torsten Maehne ***@***.***> wrote: @AndrewGoodrich <https://github.com/AndrewGoodrich>: Might you be able to provide some pointers to @thatTPR <https://github.com/thatTPR>? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#8 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGQR2PBHBYFPJ6NZEUISU3ULRBJ3ANCNFSM4RVSJQRA>. Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.

[wz108578656]

So does this issue haave a proper solution for now? If yes, many appriciate for anybody to show an example on how to multiple restart the simulation.

[maehne]

@AndrewGoodrich: Could you comment on the state in the released SystemC 2.3.4? As far as I can see from its release notes, your mentioned improvements are not yet part of it.

[AndrewGoodrich]

The change for this was put into the master branch and the issue #435 closed on 14 December 2021. If sc_event::none() appears in sysc/communication/sc_interface.cpp the changes are present. Andy

…

On Aug 29, 2023, at 4:56 AM, Torsten Maehne ***@***.***> wrote: @AndrewGoodrich <https://github.com/AndrewGoodrich>: Could you comment on the state in the released SystemC 2.3.4? As far as I can see from its release notes, your mentioned improvements are not yet part of it. — Reply to this email directly, view it on GitHub <#8 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGQR2PXGXAPFLTRSWZKW3DXXWVFHANCNFSM4RVSJQRA>. You are receiving this because you were mentioned.

[AmeyaVS]

I don't these changes in this repo here:

systemc/src/sysc/communication/sc_interface.cpp

Line 55 in 3974007

return sc_event::none;

[AmeyaVS]

Seems like the release process to public release is broken.

[AndrewGoodrich]

So that suggests the change from December 2021 was not migrated into 2.3.4.

…

On Aug 29, 2023, at 10:16 AM, Ameya Vikram Singh ***@***.***> wrote: I don't these changes in this repo here: https://github.com/accellera-official/systemc/blob/39740075f47fedbce242808d357fcfb6d3d33957/src/sysc/communication/sc_interface.cpp#L55 — Reply to this email directly, view it on GitHub <#8 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGQR2LDPN244BCLLUTBNSTXXX2SLANCNFSM4RVSJQRA>. You are receiving this because you were mentioned.

[jakobs]

Any possibility to raise the issue again in OSCI-wg/systemc? Resetting a simulation state I find quite relevant, also for unit testing implementations.

[jakobs]

@AndrewGoodrich would it be possible that you could include the PR for OSCI-wg/systemc #435 in this public repository? It seems cleaner than replicating your changes based on the issue trail available.

[AndrewGoodrich]

I have made the changes in a copy of 2.3.4 in a sandbox on my machine, should I just do a push to get the changes into the repository? Thanks, Andy

…

On Oct 20, 2023, at 6:51 AM, Jakob Schwendner ***@***.***> wrote: @AndrewGoodrich <https://github.com/AndrewGoodrich> would it be possible that you could include the PR for OSCI-wg/systemc #435 in this public repository? It seems cleaner than replicating your changes based on the issue trail available. — Reply to this email directly, view it on GitHub <#8 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGQR2OIIH75BEGITECBDU3YAJJT3AVCNFSM4RVSJQRKU5DIOJSWCZC7NNSXTN2JONZXKZKDN5WW2ZLOOQ5TCNZXGI2TCOJWGM2Q>. You are receiving this because you were mentioned.

[jakobs]

I have made the changes in a copy of 2.3.4 in a sandbox on my machine, should I just do a push to get the changes into the repository? Thanks, Andy

@AndrewGoodrich I don't know how this public repository relates to the OSCI-wg/systemc internal one, but I would guess the cleanest way would be for you to clone this public repository, push your changes to that clone and open a Pull Request in this repository. Also it would be great if somehow these changes could make it into an official release.
I much appreciate your support!

[AndrewGoodrich]

Its pull request #59

…

On Oct 21, 2023, at 7:11 AM, Jakob Schwendner ***@***.***> wrote: I have made the changes in a copy of 2.3.4 in a sandbox on my machine, should I just do a push to get the changes into the repository? Thanks, Andy @AndrewGoodrich <https://github.com/AndrewGoodrich> I don't know how this public repository relates to the OSCI-wg/systemc internal one, but I would guess the cleanest way would be for you to clone this public repository, push your changes to that clone and open a Pull Request in this repository. Also it would be great if somehow these changes could make it into an official release. I much appreciate your support! — Reply to this email directly, view it on GitHub <#8 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGQR2NABDY3AELLUZASQDTYAOUVPAVCNFSM4RVSJQRKU5DIOJSWCZC7NNSXTN2JONZXKZKDN5WW2ZLOOQ5TCNZXGM3TKOBVGA2A>. You are receiving this because you were mentioned.

[chenhaoc]

There is a simple way to avoid double free when exit. That is to force sc_curr_simcontext=NULL before sc_elab_and_sim.
btw: 3.0.0 has removed the global state static const sc_event none;.

if(cold_start) {
    sc_curr_simcontext = NULL; // force to be null before sc_elab
    ret = sc_elab_and_sim(argc,argv);
    cold_start = false;
}
else {
    delete sc_curr_simcontext;
    sc_curr_simcontext = NULL; 
    ret = sc_elab_and_sim(argc,argv);
}