Thread ID Cleanup, main branch (2025.01.09.)

[krasznaa]

Following #808, I thought it would be time to clean up the rest of the device functions as well with how they receive thread identifiers.

• Made all functions that need a "global index" use traccc::device::global_index_t;
  • This allowed removing a number of static_cast-s from these functions;
• All functions that receive a thread_id and/or barrier object, now do so using constant references;
  • All such types provide only const functions. So I didn't see much reason for using non-const references here. 🤔

After this, I went and harmonized the CUDA, SYCL and ALPAKA codes a little as well.

• Moved all thread_id types to be private headers in their libraries;
• Added automatic, compile-time checks that they would all fulfill the traccc::device::concepts::thread_id1 concept;
• Introduced traccc::cuda::details::global_index1() and traccc::sycl::details::global_index(...) as helper functions for generating traccc::device::global_index_t values;

While doing all of this, I tried to fix up the includes in all the touched files a bit. Since many of them were doing very questionable things. (Including way more files than necessary, hiding missing includes in some of the common device headers.)

I'm pretty happy with these updates myself, but am interested in your opinions.

[stephenswat]

I generally dislike this idea that because vecmem only supports 32-bit integers we should be downcasting all out accesses in a 64-bit memory space to 32-bit at the generation site of those indices, rather than generating them at the appropriate size and only downcasting them necessary.

[stephenswat]

Also, I don't think we need this; the thread identifier classes already serve this purpose. 😕

[krasznaa]

I thought about it. But note that global_index_t is sort of "its own thing" in the code, it's not directly tied to the thread_id classes. Those are still allowed to return any integer if they really need to. (That's another story by itself.)

This just seemed like a nicely readable way of expressing what we want. 🤔 In both the CUDA and SYCL code.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud