Description
PR #1743 ran into errors in GpuLower::analysis. In the short term, the workaround is to skip Mma specific validation in analysis.
Ideally if intend to use the EE, then we should not be compiling a fusion / launching a kernel (already skipped after #1723) . This will require us to modify some downstream calls and FusionExecutor interface to integrate using EE alongside nvFuser kernels seamlessly.
Metadata
Metadata
Assignees
Labels
No labels