[Mlir-commits] [mlir] [AMD][ROCDL][AMDGPU] Support packed conversions fp8/bf8->bf16 and fp8/bf8->fp32 (PR #131850)

[Krzysztof Drewniak]

krzysz00 wrote:

Ah

Then the PR title confused me - I thought you were plumbing those through the lowering

So ... yeah, I'll take another look and then lift my hold on the assumption that the bf16 conversions'll be followup work

(You might want to fiddle with the PR title though to make it clearer that you're not supporting packed ext to bf16 with this change)

https://github.com/llvm/llvm-project/pull/131850