Skip to content

brw: Use the common NIR lowering for fquantize.

Emma Anholt requested to merge anholt/mesa:intel-fquantize into main

This generates one extra instruction to set the rounding mode to RTE due to f2f16_rtne in the lowering. If the rounding mode should be undef, as was previously being done in the backend, then we should probably fix up the core lowering as well.

Edited by Emma Anholt

Merge request reports