Scott Todd
Scott Todd
Are the tests in https://github.com/iree-org/iree/blob/main/tests/e2e/linalg_ext_ops/scan.mlir representative enough?
Seeing some errors after syncing: https://github.com/iree-org/iree/actions/runs/9164893455/job/25197219527?pr=15878#step:4:1068 ``` /work/runtime/bindings/python/io.cc:217:26: error: no member named 'bool_' in namespace 'nanobind' if (py::cast(self_object.attr("is_host_allocation"))) { ~~~~^ /work/runtime/bindings/python/io.cc:217:32: error: type name requires a specifier or qualifier if...
Ping, @sogartar I think you wanted to build off of this?
Ping? (Just noticed these tests still not being included in the lightweight 'runtime' builds on https://github.com/iree-org/iree/pull/17959)
Hah yeah. They were close but had some gaps. I'll sync this then merge. Thanks for looking!
before: https://github.com/iree-org/iree/actions/runs/10001221131/job/27644479262?pr=17959#step:9:1067 ``` iree/modules/check/test = 0.00 sec*proc (4 tests) iree/task = 1.95 sec*proc (8 tests) ``` after: https://github.com/iree-org/iree/actions/runs/10002430990/job/27647713462?pr=15878#step:9:1081 ``` iree/modules/check/test = 0.00 sec*proc (4 tests) iree/runtime/bindings/python = 1.73 sec*proc (7...
Looks like triu is not implemented in https://github.com/llvm/torch-mlir . We also have test failures for `onnx.Trilu` that would use a similar lowering path (those look like the same math 🤔).
Ah, some triu tests did start passing at https://github.com/iree-org/iree/commit/440c87019240e225bafce595be5cf61ce344e0c8. Also yes, we should push a new version to pip soon...
The long term fix is to refactor the benchmark jobs to only depend on release artifacts. That would currently mean anchoring on the newer "pkgci" jobs, or forking off into...