[Mlir-commits] [mlir] [mlir][tosa] Fix regression in mlir StableHLO tests due to PR #203583 (PR #207995)

[Luke Hutton]

lhutton1 wrote:

Thanks for the fix @mydatascience and apologies for introducing these regressions. I'm curious to understand more about the failures you see in StableHLO? This fix makes the error message very generic "op operand #0 must be tosa-conformant tensor of number values" - it seems unfortunate to lose the ability to inform a user exactly what went wrong when constructing a block_scaled tensor

https://github.com/llvm/llvm-project/pull/207995