| Message ID | 03a1edb5-f1d2-9285-70d3-ea008a765bd0@arm.com |
|---|---|
| State | New |
| Headers | show |
| Series | [vect] PR92429: do not fold when updating epilogue statements | expand |
On Fri, 10 Jan 2020, Andre Vieira (lists) wrote: > Hi, > > This patch addresses the problem reported in PR92429. When creating an > epilogue for vectorization we have to replace the SSA_NAMEs in the > PATTERN_DEF_SEQs and RELATED_STMTs of the epilogue's loop_vec_infos. When > doing this we were using simplify_replace_tree which always folds the > replacement. This may lead to a different tree-node than the one which was > analyzed in vect_loop_analyze. In turn the new tree-node may require a > different vectorization than the one we had prepared for which caused the ICE > in question. > > This patch adds a parameter to the simplify_replace_tree function such that we > can optionally disable folding as this patch does in > update_epilogue_loop_vinfo. > > Bootstrapped and regression tested on x86_64. > > Is this OK for trunk? OK. Thanks, Richard. > gcc/ChangeLog: > 2020-01-10 Andre Vieira <andre.simoesdiasvieira@arm.com> > > PR tree-optimization/92429 > * tree-ssa-loop-niter.h (simplify_replace_tree): Add parameter. > * tree-ssa-loop-niter.c (simplify_replace_tree): Add parameter > to control folding. > * tree-vect-loop.c (update_epilogue_vinfo): Do not fold when > replacing tree. > > gcc/testsuite/ChangeLog: > 2020-01-10 Andre Vieira <andre.simoesdiasvieira@arm.com> > > PR tree-optimization/92429 > * gcc.dg/vect/pr92429.c > >
diff --git a/gcc/testsuite/gcc.dg/vect/pr92429.c b/gcc/testsuite/gcc.dg/vect/pr92429.c new file mode 100644 index 0000000000000000000000000000000000000000..7885cd6e4c2639f553a68edcbd70203d1b7b13bb --- /dev/null +++ b/gcc/testsuite/gcc.dg/vect/pr92429.c @@ -0,0 +1,14 @@ +/* PR92429 */ +/* { dg-do compile } */ +/* { dg-additional-options "-O1 -fno-tree-fre" } */ +/* { dg-additional-options "-mavx2" { target { x86_64-*-* i?86-*-* } } } */ + +void +as (int *gl, int k1) +{ + while (k1 < 1) + { + gl[k1] = gl[k1] * gl[k1] / 2; + ++k1; + } +} diff --git a/gcc/tree-ssa-loop-niter.h b/gcc/tree-ssa-loop-niter.h index 621e2c2e28dfbeda279687dcae78e9c4fbe7f559..eb8d1579479cf6df66f9486c7ce6f9ac20a02e2a 100644 --- a/gcc/tree-ssa-loop-niter.h +++ b/gcc/tree-ssa-loop-niter.h @@ -58,7 +58,7 @@ extern void free_numbers_of_iterations_estimates (class loop *); extern void free_numbers_of_iterations_estimates (function *); extern tree simplify_replace_tree (tree, tree, tree, tree (*)(tree, void *) = NULL, - void * = NULL); + void * = NULL, bool do_fold = true); extern void substitute_in_loop_info (class loop *, tree, tree); #endif /* GCC_TREE_SSA_LOOP_NITER_H */ diff --git a/gcc/tree-ssa-loop-niter.c b/gcc/tree-ssa-loop-niter.c index 4d5e04945118880d0c0cfb00d9c5841cc58e46f5..6e6df0bfdb898de948c917640ff488b19cf7911f 100644 --- a/gcc/tree-ssa-loop-niter.c +++ b/gcc/tree-ssa-loop-niter.c @@ -1934,7 +1934,8 @@ number_of_iterations_cond (class loop *loop, tree simplify_replace_tree (tree expr, tree old, tree new_tree, - tree (*valueize) (tree, void*), void *context) + tree (*valueize) (tree, void*), void *context, + bool do_fold) { unsigned i, n; tree ret = NULL_TREE, e, se; @@ -1966,7 +1967,7 @@ simplify_replace_tree (tree expr, tree old, tree new_tree, for (i = 0; i < n; i++) { e = TREE_OPERAND (expr, i); - se = simplify_replace_tree (e, old, new_tree, valueize, context); + se = simplify_replace_tree (e, old, new_tree, valueize, context, do_fold); if (e == se) continue; @@ -1976,7 +1977,7 @@ simplify_replace_tree (tree expr, tree old, tree new_tree, TREE_OPERAND (ret, i) = se; } - return (ret ? fold (ret) : expr); + return (ret ? (do_fold ? fold (ret) : ret) : expr); } /* Expand definitions of ssa names in EXPR as long as they are simple diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c index cf9b7bcb9c41d678b326b41a4167f1e9c00cc5a4..f30e104e9958c0ebaf170004747ff436b1c1d855 100644 --- a/gcc/tree-vect-loop.c +++ b/gcc/tree-vect-loop.c @@ -8440,8 +8440,13 @@ update_epilogue_loop_vinfo (class loop *epilogue, tree advance, gimple_set_op (stmt, j, *new_op); else { + /* PR92429: The last argument of simplify_replace_tree disables + folding when replacing arguments. This is required as + otherwise you might end up with different statements than the + ones analyzed in vect_loop_analyze, leading to different + vectorization. */ op = simplify_replace_tree (op, NULL_TREE, NULL_TREE, - &find_in_mapping, &mapping); + &find_in_mapping, &mapping, false); gimple_set_op (stmt, j, op); } }
Hi, This patch addresses the problem reported in PR92429. When creating an epilogue for vectorization we have to replace the SSA_NAMEs in the PATTERN_DEF_SEQs and RELATED_STMTs of the epilogue's loop_vec_infos. When doing this we were using simplify_replace_tree which always folds the replacement. This may lead to a different tree-node than the one which was analyzed in vect_loop_analyze. In turn the new tree-node may require a different vectorization than the one we had prepared for which caused the ICE in question. This patch adds a parameter to the simplify_replace_tree function such that we can optionally disable folding as this patch does in update_epilogue_loop_vinfo. Bootstrapped and regression tested on x86_64. Is this OK for trunk? gcc/ChangeLog: 2020-01-10 Andre Vieira <andre.simoesdiasvieira@arm.com> PR tree-optimization/92429 * tree-ssa-loop-niter.h (simplify_replace_tree): Add parameter. * tree-ssa-loop-niter.c (simplify_replace_tree): Add parameter to control folding. * tree-vect-loop.c (update_epilogue_vinfo): Do not fold when replacing tree. gcc/testsuite/ChangeLog: 2020-01-10 Andre Vieira <andre.simoesdiasvieira@arm.com> PR tree-optimization/92429 * gcc.dg/vect/pr92429.c