https://github.com/jhuber6 commented:
I'm not actually sure why this assertion is here, it contradicts my
understanding of the OffloadAction. It's mainly used to combine actions so the
driver can traverse the tree and propagate information. In the device-only case
it will have dependencies for each architecture you passed, which if one if you
only passed on.
```
+- 0: input, "/dev/null", cuda, (device-cuda, sm_89)
+- 1: preprocessor, {0}, cuda-cpp-output, (device-cuda, sm_89)
+- 2: compiler, {1}, ir, (device-cuda, sm_89)
+- 3: backend, {2}, assembler, (device-cuda, sm_89)
+- 4: assembler, {3}, object, (device-cuda, sm_89)
5: offload, "device-cuda (nvptx64-nvidia-cuda:sm_89)" {4}, none
```
>From that it feels like this assertion is too strict, but I've never really
>worked with the tooling before so I don't know exactly what they were trying
>to test.
https://github.com/llvm/llvm-project/pull/173762
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
