On Wed, May 26, 2021 at 10:44 AM Uros Bizjak <ubiz...@gmail.com> wrote: > > On Tue, May 25, 2021 at 4:29 PM Richard Biener > <richard.guent...@gmail.com> wrote: > > > > On Fri, May 21, 2021 at 5:00 PM Uros Bizjak via Gcc-patches > > <gcc-patches@gcc.gnu.org> wrote: > > > > > > Here it is, the patch that enables auto-vectorization for 32bit modes. > > > > > > Sent as RFC, because the patch fails some vectorizer scans, as it > > > obviously enables more vectorization to happen: > > > > > > Running target unix > > > FAIL: gcc.dg/vect/pr71264.c -flto -ffat-lto-objects scan-tree-dump > > > vect "vectorized 1 loops in function" > > > FAIL: gcc.dg/vect/pr71264.c scan-tree-dump vect "vectorized 1 loops in > > > function" > > > FAIL: gcc.dg/vect/slp-28.c -flto -ffat-lto-objects > > > scan-tree-dump-times vect "vectorized 1 loops" 1 > > > FAIL: gcc.dg/vect/slp-28.c -flto -ffat-lto-objects > > > scan-tree-dump-times vect "vectorizing stmts using SLP" 1 > > > FAIL: gcc.dg/vect/slp-28.c scan-tree-dump-times vect "vectorized 1 loops" > > > 1 > > > FAIL: gcc.dg/vect/slp-28.c scan-tree-dump-times vect "vectorizing > > > stmts using SLP" 1 > > > FAIL: gcc.dg/vect/slp-3.c -flto -ffat-lto-objects > > > scan-tree-dump-times vect "vectorized 3 loops" 1 > > > FAIL: gcc.dg/vect/slp-3.c -flto -ffat-lto-objects > > > scan-tree-dump-times vect "vectorizing stmts using SLP" 3 > > > FAIL: gcc.dg/vect/slp-3.c scan-tree-dump-times vect "vectorized 3 loops" 1 > > > FAIL: gcc.dg/vect/slp-3.c scan-tree-dump-times vect "vectorizing stmts > > > using SLP" 3 > > > > > > > > > Running target unix/-m32 > > > FAIL: gcc.dg/vect/no-vfa-vect-101.c scan-tree-dump-times vect "can't > > > determine dependence" 1 > > > FAIL: gcc.dg/vect/no-vfa-vect-102.c scan-tree-dump-times vect > > > "possible dependence between data-refs" 1 > > > FAIL: gcc.dg/vect/no-vfa-vect-102a.c scan-tree-dump-times vect > > > "possible dependence between data-refs" 1 > > > FAIL: gcc.dg/vect/no-vfa-vect-37.c scan-tree-dump-times vect "can't > > > determine dependence" 2 > > > FAIL: gcc.dg/vect/pr71264.c -flto -ffat-lto-objects scan-tree-dump > > > vect "vectorized 1 loops in function" > > > FAIL: gcc.dg/vect/pr71264.c scan-tree-dump vect "vectorized 1 loops in > > > function" > > > FAIL: gcc.dg/vect/slp-28.c -flto -ffat-lto-objects > > > scan-tree-dump-times vect "vectorized 1 loops" 1 > > > FAIL: gcc.dg/vect/slp-28.c -flto -ffat-lto-objects > > > scan-tree-dump-times vect "vectorizing stmts using SLP" 1 > > > FAIL: gcc.dg/vect/slp-28.c scan-tree-dump-times vect "vectorized 1 loops" > > > 1 > > > FAIL: gcc.dg/vect/slp-28.c scan-tree-dump-times vect "vectorizing > > > stmts using SLP" 1 > > > FAIL: gcc.dg/vect/slp-3.c -flto -ffat-lto-objects > > > scan-tree-dump-times vect "vectorized 3 loops" 1 > > > FAIL: gcc.dg/vect/slp-3.c -flto -ffat-lto-objects > > > scan-tree-dump-times vect "vectorizing stmts using SLP" 3 > > > FAIL: gcc.dg/vect/slp-3.c scan-tree-dump-times vect "vectorized 3 loops" 1 > > > FAIL: gcc.dg/vect/slp-3.c scan-tree-dump-times vect "vectorizing stmts > > > using SLP" 3 > > > FAIL: gcc.dg/vect/vect-104.c -flto -ffat-lto-objects > > > scan-tree-dump-times vect "possible dependence between data-refs" 1 > > > FAIL: gcc.dg/vect/vect-104.c scan-tree-dump-times vect "possible > > > dependence between data-refs" 1 > > > > Yeah, it's a bit iffy to adjust expectations. If there's a way to > > disable vectorization > > for 32bit modes on x86 that might be a way to "fix" them, otherwise we're > > lacking a way to query for available vector modes/sizes in the dejagnu vect > > targets. There's available_vector_sizes but it's implementation is hardly > > complete nor is size the only important thing (FP vs. INT). At least > > one could add a vect32 predicate similar to the existing vect64 one. > > I went the way you proposed above. By adding 32bit vector size to > available_vector_sizes only two testcases fails. The attached patch > fixes all vect scan failures (the remaining failure in > vect_epilogues.c is just the case of missing uavg<mode>3_ceil pattern > for V4QI epilogue vectorization - I plan to add the insn in the > follow-up patch). > > The patch also xfails pr71264.c, the case of missing re-vectorization > of 32bit vectors. > > WDYT?
LGTM. Richard. > Uros.
