On Fri, Nov 15, 2019 at 12:33 PM Richard Sandiford <richard.sandif...@arm.com> wrote: > > Richard Biener <richard.guent...@gmail.com> writes: > > On Mon, Nov 11, 2019 at 7:47 PM Richard Sandiford > > <richard.sandif...@arm.com> wrote: > >> > >> This patch adds a bunch of flags to dr_with_seg_len_pair_t, > >> for use by later patches. The update to tree-loop-distribution.c > >> is conservatively correct, but might be tweakable later. > > > > Does this all work with interleaved SLP loads/stores like > > > > a[i] = b[i]; > > tem1 = b[i+1]; > > tem2 = b[i+2]; > > a[i+2] = tem2; > > a[i+1] = tem1; > > > > where we don't preserve the scalar order but emit code at the > > latest stmt of the grouped access? > > Yeah. vect-alias-check-9.c is a less sophisticated example of that. > In both cases vect_preserves_scalar_order_p is false. > > > That is, what does "preserve scalar oder" actually mean? > > It's supposed to mean that if a memory access A from the first > group and a memory access B from the second group occur within > the same scalar iteration, the order of the accesses in the > vector loop body will be the same as it was in the scalar loop body. > In other words, the order of the vector stmts honours any dependencies > between the accesses that occur within one iteration of the scalar loop.
OK, I see. The patch is OK then. Thanks, Richard. > Thanks, > Richard
