These are @tpalli's patches, but he's on vacation. I think @majanes found that this had a significant impact on DG2 unit test results.
I reviewed bspec 46211, and the text that Tapani quoted appeared to apply for gen12+, so I made that change to his patches.
For DG2 these patches were tested for performance impacts, and didn't appear to cause significant regressions for the several workloads tested.
Cc: @kwg, @currojerez