public inbox for gcc-cvs@sourceware.org help / color / mirror / Atom feed
From: Philipp Tomsich <ptomsich@gcc.gnu.org> To: gcc-cvs@gcc.gnu.org Subject: [gcc(refs/vendors/vrull/heads/slp-improvements)] Fix issues and improve vect_slp_optimize_permutes. Date: Wed, 17 Jan 2024 19:14:26 +0000 (GMT) [thread overview] Message-ID: <20240117191426.297C83858409@sourceware.org> (raw) https://gcc.gnu.org/g:6c6e798decb95bfd002bb24abb134a22b9a78e48 commit 6c6e798decb95bfd002bb24abb134a22b9a78e48 Author: Manolis Tsamis <manolis.tsamis@vrull.eu> Date: Thu Nov 23 11:28:44 2023 +0100 Fix issues and improve vect_slp_optimize_permutes. * Don't use tree_to_vec_perm_builder to get the elements of the permute constant vectors. This function uses vector_cst_encoded_nelts internally which may return fewer elements than VECTOR_CST_NELTS and in turn fewer elements than what is observed on the GIMPLE statement. Because we iterate based on NELTS this would cause memory corruption issues. The resulting code is also more clean and performant. * Call update_stmt after rewriting permutes because otherwise we may get SSA imm_use lists issues. Ref #346 Diff: --- gcc/tree-vect-slp.cc | 39 +++++++++++++++++++++++---------------- 1 file changed, 23 insertions(+), 16 deletions(-) diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc index 58543629756..20acfba887d 100644 --- a/gcc/tree-vect-slp.cc +++ b/gcc/tree-vect-slp.cc @@ -8256,33 +8256,36 @@ vect_slp_optimize_permutes (function *fun) continue; vec_perm_builder sel (nelts, nelts, 1); - vec_perm_builder sel1_1, sel1_2, sel2_1, sel2_2; - if (!tree_to_vec_perm_builder (&sel1_1, gimple_assign_rhs3 (src1_1)) - || !tree_to_vec_perm_builder (&sel1_2, gimple_assign_rhs3 (src1_2)) - || !tree_to_vec_perm_builder (&sel2_1, gimple_assign_rhs3 (src2_1)) - || !tree_to_vec_perm_builder (&sel2_2, gimple_assign_rhs3 (src2_2))) - continue; + vec_perm_builder sel_1 (nelts, nelts, 1); + vec_perm_builder sel_2 (nelts, nelts, 1); /* Rewrite the permutations based on MASK1_LANES/MASK2_LANES. */ for (i = 0; i < nelts; i++) { + /* Calculate new mask for STMT2. */ tree val = VECTOR_CST_ELT (mask2, i); - unsigned HOST_WIDE_INT j + unsigned HOST_WIDE_INT lane = TREE_INT_CST_LOW (val) & (2 * nelts - 1); - unsigned off = (j < nelts) ? 0 : nelts; - unsigned HOST_WIDE_INT new_lane = *mask2_lanes.get (j - off) + off; + unsigned off = (lane < nelts) ? 0 : nelts; + unsigned HOST_WIDE_INT new_lane + = *mask2_lanes.get (lane - off) + off; sel.quick_push (new_lane); - if (!mask1_lanes.get (i)) - { - sel1_1[i] = nelts + sel2_1[i]; - sel1_2[i] = nelts + sel2_2[i]; - } + /* Calculate new masks for SRC1_1 and SRC1_2. */ + bool use_src1 = mask1_lanes.get (i); + tree mask1 = gimple_assign_rhs3 (use_src1 ? src1_1 : src2_1); + tree mask2 = gimple_assign_rhs3 (use_src1 ? src1_2 : src2_2); + unsigned HOST_WIDE_INT lane1 + = TREE_INT_CST_LOW (VECTOR_CST_ELT (mask1, i)) & (nelts - 1); + unsigned HOST_WIDE_INT lane2 + = TREE_INT_CST_LOW (VECTOR_CST_ELT (mask2, i)) & (nelts - 1); + sel_1.quick_push (lane1 + (use_src1 ? 0 : nelts)); + sel_2.quick_push (lane2 + (use_src1 ? 0 : nelts)); } vec_perm_indices indices (sel, 2, nelts); - vec_perm_indices indices1_1 (sel1_1, 2, nelts); - vec_perm_indices indices1_2 (sel1_2, 2, nelts); + vec_perm_indices indices1_1 (sel_1, 2, nelts); + vec_perm_indices indices1_2 (sel_2, 2, nelts); tree vectype = TREE_TYPE (gimple_assign_lhs (stmt2)); if (!can_vec_perm_const_p (TYPE_MODE (vectype), TYPE_MODE (vectype), indices) @@ -8300,6 +8303,10 @@ vect_slp_optimize_permutes (function *fun) gimple_assign_set_rhs3 (src1_1, vect_gen_perm_mask_checked (vectype, indices1_1)); gimple_assign_set_rhs3 (src1_2, vect_gen_perm_mask_checked (vectype, indices1_2)); + + update_stmt (stmt2); + update_stmt (src1_1); + update_stmt (src1_2); } }
next reply other threads:[~2024-01-17 19:14 UTC|newest] Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top 2024-01-17 19:14 Philipp Tomsich [this message] -- strict thread matches above, loose matches on Subject: below -- 2024-01-23 20:57 Philipp Tomsich 2023-11-28 13:35 Philipp Tomsich
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20240117191426.297C83858409@sourceware.org \ --to=ptomsich@gcc.gnu.org \ --cc=gcc-cvs@gcc.gnu.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).