You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Since I wired up the original build support, I added a mode where this project could have a hard dependency on iree for getting the iree-dialects (we have frontend dialects and standalone, not-yet-upstreamed things in this standalone project). However, given the goals of primarily enabling interop and code flow of things that aren't ready to be upstreamed or promoted in another way, I think I should bias towards simplicity: any objection to me just copying the iree-dialects project in tree and maintaining it by copying snapshots as needed? That seems wholly more usable than trying to keep yet another LLVM-derived project in sync, etc. This would allow me to adapt more of IREE's input pipelines to directly target more interesting whole programs at the sandbox, and it would also expose ops for IREE's parallelism and distribution concepts that we could experiment with.
The text was updated successfully, but these errors were encountered:
+1, the LLVM dependency mismatch turned out to be quite annoying in the past weeks of multi-days integrate.
Will be nice to iterate to that minimal IREE + pluggable codegen world.
Since I wired up the original build support, I added a mode where this project could have a hard dependency on iree for getting the iree-dialects (we have frontend dialects and standalone, not-yet-upstreamed things in this standalone project). However, given the goals of primarily enabling interop and code flow of things that aren't ready to be upstreamed or promoted in another way, I think I should bias towards simplicity: any objection to me just copying the iree-dialects project in tree and maintaining it by copying snapshots as needed? That seems wholly more usable than trying to keep yet another LLVM-derived project in sync, etc. This would allow me to adapt more of IREE's input pipelines to directly target more interesting whole programs at the sandbox, and it would also expose ops for IREE's parallelism and distribution concepts that we could experiment with.
The text was updated successfully, but these errors were encountered: