llvm-project

768182df - [X86] Declare 128/256-bit funnel shifts legal on VBMI2 + NOVLX targets (#184634)

[X86] Declare 128/256-bit funnel shifts legal on VBMI2 + NOVLX targets (#184634) Add tablegen patterns to widen 128/256-bit funnel shift to 512-bit Similar to what we do for 128/256-bit AVX512F rotates (and a lot of other instructions) already - and I've take the opportunity to create similar macros to reduce duplication in the rotate patterns Another step towards #184002