Arsène Pérard-Gayot

OK, I have found a simple improvement in `resolve_loads` that partially addresses this problem. After the call to `side_effect`, information about `data` is lost. But even before the call, assuming...

To address the issue in the general case, we need a much smarter optimization than `resolve_loads`, probably implemented using the technique described in the paper: [Composing dataflow analyses and transformations](https://dl.acm.org/citation.cfm?id=503298)....

We normally build our projects with CMake. The [anydsl meta-repository](https://github.com/AnyDSL/anydsl) automates this process. We don't have to copy files and libraries since the CMake logic of the runtime allows to...

Is this fixed? Can we close the issue?

The v2 branch now incorporates a custom thread pool based on `std::thread`.

Other than the point above, it looks good to me, ready to be merged. You may have to rebase first though to avoid a merge commit.

It should be possible to implement it, and by the way feel free to submit a PR for review. The issue though is that this requires some sort of knowledge...

For this specific use case, I don't think you need the ability to cancel the build per se. I would imagine that through OS services you can send a kill...

If you don't want to kill the current process but want to continue executing the program, you could use a subprocess to build the BVH as a work around. This...

See https://github.com/AnyDSL/thorin/issues/95. What you need here is to _lift_ `loop` (looking at its free variables and making them arguments), then rewrite calls to it. Lower2cff should be rewritten to do...