Open
Description
Right now SFU math ops generate a MOV to the SFU from their source. We could coalesce the SFU write with an ALU op generating the source, like we do for texture coordinates.
Would this be a flag on the temp write indicating which SFU op should be done, so that the SFU result is still register allocated like we do today? Or would we want to move to something like TEX_RESULT, where the collection from SFU is separate from the write to SFU?