search for: rl224742

...iddly probably being instruction selection converting appropriate IR to the right target-specific functionality. ## LLVM commits * The SROA (scalar replacement of aggregates) pass has seen some refactoring to, in the future, allow for more intelligent rewriting. [r224742](http://reviews.llvm.org/rL224742), [r224798](http://reviews.llvm.org/rL224798). * The masked load and store intrinsics have been documented. [r224832](http://reviews.llvm.org/rL224832). * CodeGenPrepare learned to speculate calls to llvm.cttz/ctlz (count trailing/leading zeroes) if `isCheapToSpeculateCtlz/isCheapToSpeculatCttz`...