From patchwork Thu Nov 16 16:49:55 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Martin Jambor X-Patchwork-Id: 165898 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a59:b909:0:b0:403:3b70:6f57 with SMTP id t9csp3342122vqg; Thu, 16 Nov 2023 08:50:49 -0800 (PST) X-Google-Smtp-Source: AGHT+IF8PMVaCYREhgO1N67A54LqsOXkUSMe5ndBqOANbQG2A9isJmk9qKkRFLidcY8at1f8mjBL X-Received: by 2002:ad4:5149:0:b0:671:3f49:c8b2 with SMTP id g9-20020ad45149000000b006713f49c8b2mr9309761qvq.15.1700153449095; Thu, 16 Nov 2023 08:50:49 -0800 (PST) ARC-Seal: i=2; a=rsa-sha256; t=1700153449; cv=pass; d=google.com; s=arc-20160816; b=ZT4r5bwQCnGacrgeSxnGnd0WA6WeOc3TrB0quhTHGvjB/UwGY0WIGKVYJj0u66Z9gb tW6Ku3FcuYGjYezdOB23+9qb/tD7b86nqked9qswCzK24eMVB4URkDbm+iIqmgpBzU8r Pr3oR8W8o0/ORwS6ZpEJlXox/bvdbr+UZWAqqd1x0rbbFhH2gKapnPaNsNywmrmMqYtV giL7G9zSR9DcQ7/xxGOiLnf7ucxgUtfCAxtOyhScJQTb8PZCnj5MuriLdkvwKLfAuKLM THJOdOCJqzv+9GIlt+6q3GgLlM6k/+P5KeQDEpruW3LLFJYiCat6B6mB4JS4jk2YG4k5 tX0g== ARC-Message-Signature: i=2; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=errors-to:list-subscribe:list-help:list-post:list-archive :list-unsubscribe:list-id:precedence:mime-version:date:user-agent :subject:cc:to:from:dkim-signature:dkim-signature:arc-filter :dmarc-filter:delivered-to:message-id; bh=p9GRp8Nr7SUb68eZupfppf1NjR+x8O3lAy1h0o02QYM=; fh=A3Wf0JmQRmrpDq4HcJatpWHDr5j5KRjG4sAEmNkNl4c=; b=MIxck9ivrHyiEmOsnJEXkm29Ki4lBjj9jlSnlC9W3hopj5u1edHqe1I5eEMZTx2MP+ YJhjCZH/t07vP/q6j24JQ5wybBlowRzoEIM1qkRm6r/mT7dVqhF972lRs7RsPBNnx5hs nD6gwjmm2HVHLHVCpv3JPa52s4yiS97P9Z+DtuqSg64ooEMXNwlK1EfnYrLUP1NtlJV3 RMfysHXtrNQXjeEgrokKzjLyLNp8WNDs3M/8G03/bP6v8CzJkSU/2DL6cgOEnG2IkQgg WJlREAlj4iEN07q1UloTEW30lWHeOudxo6wbqZFM8U6/kHFcO+wzjx5ddoPtUV4nVkRs 3Yqw== ARC-Authentication-Results: i=2; mx.google.com; dkim=pass header.i=@suse.cz header.s=susede2_rsa header.b=mskRnaPi; dkim=neutral (no key) header.i=@suse.cz; arc=pass (i=1); spf=pass (google.com: domain of gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org designates 8.43.85.97 as permitted sender) smtp.mailfrom="gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org" Received: from server2.sourceware.org (server2.sourceware.org. [8.43.85.97]) by mx.google.com with ESMTPS id s21-20020a05621412d500b0065b1ded1828si11372484qvv.202.2023.11.16.08.50.49 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Thu, 16 Nov 2023 08:50:49 -0800 (PST) Received-SPF: pass (google.com: domain of gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org designates 8.43.85.97 as permitted sender) client-ip=8.43.85.97; Authentication-Results: mx.google.com; dkim=pass header.i=@suse.cz header.s=susede2_rsa header.b=mskRnaPi; dkim=neutral (no key) header.i=@suse.cz; arc=pass (i=1); spf=pass (google.com: domain of gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org designates 8.43.85.97 as permitted sender) smtp.mailfrom="gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org" Message-ID: <65564869.050a0220.fc37f.b9bfSMTPIN_ADDED_BROKEN@mx.google.com> X-Google-Original-Message-ID: Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id BE1A8385AC1B for ; Thu, 16 Nov 2023 16:50:31 +0000 (GMT) X-Original-To: gcc-patches@gcc.gnu.org Delivered-To: gcc-patches@gcc.gnu.org Received: from smtp-out2.suse.de (smtp-out2.suse.de [IPv6:2001:67c:2178:6::1d]) by sourceware.org (Postfix) with ESMTPS id 758093858415 for ; Thu, 16 Nov 2023 16:50:01 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 758093858415 Authentication-Results: sourceware.org; dmarc=none (p=none dis=none) header.from=suse.cz Authentication-Results: sourceware.org; spf=fail smtp.mailfrom=suse.cz ARC-Filter: OpenARC Filter v1.0.0 sourceware.org 758093858415 Authentication-Results: server2.sourceware.org; arc=none smtp.remote-ip=2001:67c:2178:6::1d ARC-Seal: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1700153405; cv=none; b=IiO776yYjXYckIWUUBPTEt06QA7S2cA+0hPDrGzxTiUtADLJHPSntEFJk3QylxMSttFiwXMRmB6h6XOaKvXonsoEU5aJVQwRePKoIazkfxqBfm5Ut9w/CApe6yVy3XqU7y9/iGmXveRXdEwYMd4EXX2phK6Dgt3c8DEYdyD/3Ck= ARC-Message-Signature: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1700153405; c=relaxed/simple; bh=Akse1ncKA4zFXSbBK0Lqo4UEXknZOd4tXFMecdVobSU=; h=DKIM-Signature:DKIM-Signature:From:To:Subject:Date:Message-ID: MIME-Version; b=ZT2+n07xXvv2cYDy9z3OJBY6h60tNwtImIpWHM/kQ5bZr4xcgn6uOZVAdoQVOREFEA7moHJPgNJgdmYciclaznNW+yOPIu1zPqco7/LNs2DeRPJY/4HNL1Yuu4Bvia/6E2jvK/ErQAVS0e5wBgDRyx6aOOVv66nuh4bgVE3298c= ARC-Authentication-Results: i=1; server2.sourceware.org Received: from imap2.suse-dmz.suse.de (imap2.suse-dmz.suse.de [192.168.254.74]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature ECDSA (P-521) server-digest SHA512) (No client certificate requested) by smtp-out2.suse.de (Postfix) with ESMTPS id DD53220502; Thu, 16 Nov 2023 16:49:59 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=suse.cz; s=susede2_rsa; t=1700153399; h=from:from:reply-to:date:date:message-id:message-id:to:to:cc:cc: mime-version:mime-version:content-type:content-type; bh=p9GRp8Nr7SUb68eZupfppf1NjR+x8O3lAy1h0o02QYM=; b=mskRnaPik8o+lr1Yz5nbNHrKR0So3reYxUL1Tr0u6PsJ8jT1SzRDPf4lp2qxYjYq2GNGFU DE3PCpmyU/Bov+b4R4fbFtYMFIrDvQ+Y5bhxoaqJG0h6QxSnPO5cMIL+qdo9YJRw5hja3+ KZTIL39LCOLlK1Kisrld3uWMnQm5U40= DKIM-Signature: v=1; a=ed25519-sha256; c=relaxed/relaxed; d=suse.cz; s=susede2_ed25519; t=1700153399; h=from:from:reply-to:date:date:message-id:message-id:to:to:cc:cc: mime-version:mime-version:content-type:content-type; bh=p9GRp8Nr7SUb68eZupfppf1NjR+x8O3lAy1h0o02QYM=; b=RN8yqZtjBC1ImyaDbISP/AxI0lC0Bzc1+lAkgkJ8NXI5dRvtk5USTVVzzXzdM7s96L54Bc p3Af8hAl23Fz5BAA== Received: from imap2.suse-dmz.suse.de (imap2.suse-dmz.suse.de [192.168.254.74]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature ECDSA (P-521) server-digest SHA512) (No client certificate requested) by imap2.suse-dmz.suse.de (Postfix) with ESMTPS id D50681377E; Thu, 16 Nov 2023 16:49:59 +0000 (UTC) Received: from dovecot-director2.suse.de ([192.168.254.65]) by imap2.suse-dmz.suse.de with ESMTPSA id c1/7MzdIVmW2BAAAMHmgww (envelope-from ); Thu, 16 Nov 2023 16:49:59 +0000 From: Martin Jambor To: GCC Patches Cc: Richard Biener , Jan Hubicka Subject: [PATCH] sra: SRA of non-escaped aggregates passed by reference to calls User-Agent: Notmuch/0.37 (https://notmuchmail.org) Emacs/29.1 (x86_64-suse-linux-gnu) Date: Thu, 16 Nov 2023 17:49:55 +0100 MIME-Version: 1.0 Authentication-Results: smtp-out2.suse.de; none X-Spam-Level: X-Spam-Score: -2.10 X-Spamd-Result: default: False [-2.10 / 50.00]; ARC_NA(0.00)[]; RCVD_VIA_SMTP_AUTH(0.00)[]; FROM_HAS_DN(0.00)[]; RCPT_COUNT_THREE(0.00)[3]; TO_MATCH_ENVRCPT_ALL(0.00)[]; NEURAL_HAM_LONG(-1.00)[-1.000]; MIME_GOOD(-0.10)[text/plain]; DKIM_SIGNED(0.00)[suse.cz:s=susede2_rsa,suse.cz:s=susede2_ed25519]; TO_DN_ALL(0.00)[]; NEURAL_HAM_SHORT(-0.20)[-0.994]; INVALID_MSGID(1.70)[]; FUZZY_BLOCKED(0.00)[rspamd.com]; FROM_EQ_ENVFROM(0.00)[]; MIME_TRACE(0.00)[0:+]; MID_RHS_NOT_FQDN(0.50)[]; RCVD_COUNT_TWO(0.00)[2]; RCVD_TLS_ALL(0.00)[]; BAYES_HAM(-3.00)[100.00%] X-Spam-Status: No, score=-11.0 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, INVALID_MSGID, SPF_HELO_NONE, SPF_SOFTFAIL, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.30 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: gcc-patches-bounces+ouuuleilei=gmail.com@gcc.gnu.org X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1782740103238353419 X-GMAIL-MSGID: 1782740103238353419 Hello, PR109849 shows that a loop that heavily pushes and pops from a stack implemented by a C++ std::vec results in slow code, mainly because the vector structure is not split by SRA and so we end up in many loads and stores into it. This is because it is passed by reference to (re)allocation methods and so needs to live in memory, even though it does not escape from them and so we could SRA it if we re-constructed it before the call and then separated it to distinct replacements afterwards. This patch does exactly that, first relaxing the selection of candidates to also include those which are addressable but do not escape and then adding code to deal with the calls. The micro-benchmark that is also the (scan-dump) testcase in this patch runs twice as fast with it than with current trunk. Honza measured its effect on the libjxl benchmark and it almost closes the performance gap between Clang and GCC while not requiring excessive inlining and thus code growth. The patch disallows creation of replacements for such aggregates which are also accessed with a precision smaller than their size because I have observed that this led to excessive zero-extending of data leading to slow-downs of perlbench (on some CPUs). Apart from this case I have not noticed any regressions, at least not so far. Gimple call argument flags can tell if an argument is unused (and then we do not need to generate any statements for it) or if it is not written to and then we do not need to generate statements loading replacements from the original aggregate after the call statement. Unfortunately, we cannot symmetrically use flags that an aggregate is not read because to avoid re-constructing the aggregate before the call because flags don't tell which what parts of aggregates were not written to, so we load all replacements, and so all need to have the correct value before the call. The patch passes bootstrap, lto-bootstrap and profiled-lto-bootstrap on x86_64-linux and a very similar patch has also passed bootstrap and testing on Aarch64-linux and ppc64le-linux (I'm re-running both on these two architectures but as I'm sending this). OK for master? Thanks, Martin gcc/ChangeLog: 2023-11-16 Martin Jambor PR middle-end/109849 * tree-sra.cc (passed_by_ref_in_call): New. (sra_initialize): Allocate passed_by_ref_in_call. (sra_deinitialize): Free passed_by_ref_in_call. (create_access): Add decl pool candidates only if they are not already candidates. (build_access_from_expr_1): Bail out on ADDR_EXPRs. (build_access_from_call_arg): New function. (asm_visit_addr): Rename to scan_visit_addr, change the disqualification dump message. (scan_function): Check taken addresses for all non-call statements, including phi nodes. Process all call arguments, including the static chain, build_access_from_call_arg. (maybe_add_sra_candidate): Relax need_to_live_in_memory check to allow non-escaped local variables. (sort_and_splice_var_accesses): Disallow smaller-than-precision replacements for aggregates passed by reference to functions. (sra_modify_expr): Use a separate stmt iterator for adding satements before the processed statement and after it. (sra_modify_call_arg): New function. (sra_modify_assign): Adjust calls to sra_modify_expr. (sra_modify_function_body): Likewise, use sra_modify_call_arg to process call arguments, including the static chain. gcc/testsuite/ChangeLog: 2023-11-03 Martin Jambor PR middle-end/109849 * g++.dg/tree-ssa/pr109849.C: New test. * gfortran.dg/pr43984.f90: Added -fno-tree-sra to dg-options. --- gcc/testsuite/g++.dg/tree-ssa/pr109849.C | 31 +++ gcc/testsuite/gfortran.dg/pr43984.f90 | 2 +- gcc/tree-sra.cc | 244 ++++++++++++++++++----- 3 files changed, 231 insertions(+), 46 deletions(-) create mode 100644 gcc/testsuite/g++.dg/tree-ssa/pr109849.C diff --git a/gcc/testsuite/g++.dg/tree-ssa/pr109849.C b/gcc/testsuite/g++.dg/tree-ssa/pr109849.C new file mode 100644 index 00000000000..cd348c0f590 --- /dev/null +++ b/gcc/testsuite/g++.dg/tree-ssa/pr109849.C @@ -0,0 +1,31 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-sra" } */ + +#include +typedef unsigned int uint32_t; +std::pair pair; +void +test() +{ + std::vector > stack; + stack.push_back (pair); + while (!stack.empty()) { + std::pair cur = stack.back(); + stack.pop_back(); + if (!cur.first) + { + cur.second++; + stack.push_back (cur); + } + if (cur.second > 10000) + break; + } +} +int +main() +{ + for (int i = 0; i < 10000; i++) + test(); +} + +/* { dg-final { scan-tree-dump "Created a replacement for stack offset" "sra"} } */ diff --git a/gcc/testsuite/gfortran.dg/pr43984.f90 b/gcc/testsuite/gfortran.dg/pr43984.f90 index 130d114462c..dce26b0ef3b 100644 --- a/gcc/testsuite/gfortran.dg/pr43984.f90 +++ b/gcc/testsuite/gfortran.dg/pr43984.f90 @@ -1,5 +1,5 @@ ! { dg-do compile } -! { dg-options "-O2 -fno-tree-dominator-opts -fdump-tree-pre" } +! { dg-options "-O2 -fno-tree-dominator-opts -fdump-tree-pre -fno-tree-sra" } module test type shell1quartet_type diff --git a/gcc/tree-sra.cc b/gcc/tree-sra.cc index b985dee6964..676931d5383 100644 --- a/gcc/tree-sra.cc +++ b/gcc/tree-sra.cc @@ -337,6 +337,9 @@ static bitmap should_scalarize_away_bitmap, cannot_scalarize_away_bitmap; because this would produce non-constant expressions (e.g. Ada). */ static bitmap disqualified_constants; +/* Bitmap of candidates which are passed by reference in call arguments. */ +static bitmap passed_by_ref_in_call; + /* Obstack for creation of fancy names. */ static struct obstack name_obstack; @@ -717,6 +720,7 @@ sra_initialize (void) should_scalarize_away_bitmap = BITMAP_ALLOC (NULL); cannot_scalarize_away_bitmap = BITMAP_ALLOC (NULL); disqualified_constants = BITMAP_ALLOC (NULL); + passed_by_ref_in_call = BITMAP_ALLOC (NULL); gcc_obstack_init (&name_obstack); base_access_vec = new hash_map >; memset (&sra_stats, 0, sizeof (sra_stats)); @@ -733,6 +737,7 @@ sra_deinitialize (void) BITMAP_FREE (should_scalarize_away_bitmap); BITMAP_FREE (cannot_scalarize_away_bitmap); BITMAP_FREE (disqualified_constants); + BITMAP_FREE (passed_by_ref_in_call); access_pool.release (); assign_link_pool.release (); obstack_free (&name_obstack, NULL); @@ -905,9 +910,9 @@ create_access (tree expr, gimple *stmt, bool write) &reverse); /* For constant-pool entries, check we can substitute the constant value. */ - if (constant_decl_p (base)) + if (constant_decl_p (base) + && !bitmap_bit_p (disqualified_constants, DECL_UID (base))) { - gcc_assert (!bitmap_bit_p (disqualified_constants, DECL_UID (base))); if (expr != base && !is_gimple_reg_type (TREE_TYPE (expr)) && dump_file && (dump_flags & TDF_DETAILS)) @@ -1135,6 +1140,17 @@ sra_handled_bf_read_p (tree expr) static struct access * build_access_from_expr_1 (tree expr, gimple *stmt, bool write) { + /* We only allow ADDR_EXPRs in arguments of function calls and those must + have been dealt with in build_access_from_call_arg. Any other address + taking should have been caught by scan_visit_addr. */ + if (TREE_CODE (expr) == ADDR_EXPR) + { + tree base = get_base_address (TREE_OPERAND (expr, 0)); + gcc_assert (!DECL_P (base) + || !bitmap_bit_p (candidate_bitmap, DECL_UID (base))); + return NULL; + } + struct access *ret = NULL; bool partial_ref; @@ -1223,6 +1239,40 @@ build_access_from_expr (tree expr, gimple *stmt, bool write) return false; } +/* Scan expression EXPR which is an argument of a call and create access + structures for all accesses to candidates for scalarization. Return true if + any access has been inserted. STMT must be the statement from which the + expression is taken. */ + +static bool +build_access_from_call_arg (tree expr, gimple *stmt) +{ + if (TREE_CODE (expr) == ADDR_EXPR) + { + tree base = get_base_address (TREE_OPERAND (expr, 0)); + bool read = build_access_from_expr (base, stmt, false); + bool write = build_access_from_expr (base, stmt, true); + if (read || write) + { + if (dump_file && (dump_flags & TDF_DETAILS)) + { + fprintf (dump_file, "Allowed ADDR_EXPR of "); + print_generic_expr (dump_file, base); + fprintf (dump_file, " because of "); + print_gimple_stmt (dump_file, stmt, 0); + fprintf (dump_file, "\n"); + } + bitmap_set_bit (passed_by_ref_in_call, DECL_UID (base)); + return true; + } + else + return false; + } + + return build_access_from_expr (expr, stmt, false); +} + + /* Return the single non-EH successor edge of BB or NULL if there is none or more than one. */ @@ -1364,16 +1414,18 @@ build_accesses_from_assign (gimple *stmt) return lacc || racc; } -/* Callback of walk_stmt_load_store_addr_ops visit_addr used to determine - GIMPLE_ASM operands with memory constrains which cannot be scalarized. */ +/* Callback of walk_stmt_load_store_addr_ops visit_addr used to detect taking + addresses of candidates a places which are not call arguments. Such + candidates are disqalified from SRA. This also applies to GIMPLE_ASM + operands with memory constrains which cannot be scalarized. */ static bool -asm_visit_addr (gimple *, tree op, tree, void *) +scan_visit_addr (gimple *, tree op, tree, void *) { op = get_base_address (op); if (op && DECL_P (op)) - disqualify_candidate (op, "Non-scalarizable GIMPLE_ASM operand."); + disqualify_candidate (op, "Address taken in a non-call-argument context."); return false; } @@ -1390,12 +1442,20 @@ scan_function (void) FOR_EACH_BB_FN (bb, cfun) { gimple_stmt_iterator gsi; + for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi)) + walk_stmt_load_store_addr_ops (gsi_stmt (gsi), NULL, NULL, NULL, + scan_visit_addr); + for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi)) { gimple *stmt = gsi_stmt (gsi); tree t; unsigned i; + if (gimple_code (stmt) != GIMPLE_CALL) + walk_stmt_load_store_addr_ops (stmt, NULL, NULL, NULL, + scan_visit_addr); + switch (gimple_code (stmt)) { case GIMPLE_RETURN: @@ -1410,8 +1470,11 @@ scan_function (void) case GIMPLE_CALL: for (i = 0; i < gimple_call_num_args (stmt); i++) - ret |= build_access_from_expr (gimple_call_arg (stmt, i), - stmt, false); + ret |= build_access_from_call_arg (gimple_call_arg (stmt, i), + stmt); + if (gimple_call_chain(stmt)) + ret |= build_access_from_call_arg (gimple_call_chain(stmt), + stmt); t = gimple_call_lhs (stmt); if (t && !disqualify_if_bad_bb_terminating_stmt (stmt, t, NULL)) @@ -1428,8 +1491,6 @@ scan_function (void) case GIMPLE_ASM: { gasm *asm_stmt = as_a (stmt); - walk_stmt_load_store_addr_ops (asm_stmt, NULL, NULL, NULL, - asm_visit_addr); for (i = 0; i < gimple_asm_ninputs (asm_stmt); i++) { t = TREE_VALUE (gimple_asm_input_op (asm_stmt, i)); @@ -1920,10 +1981,19 @@ maybe_add_sra_candidate (tree var) reject (var, "not aggregate"); return false; } - /* Allow constant-pool entries that "need to live in memory". */ - if (needs_to_live_in_memory (var) && !constant_decl_p (var)) + + if ((is_global_var (var) + /* There are cases where non-addressable variables fail the + pt_solutions_check test, e.g in gcc.dg/uninit-40.c. */ + || (TREE_ADDRESSABLE (var) + && pt_solution_includes (&cfun->gimple_df->escaped, var)) + || (TREE_CODE (var) == RESULT_DECL + && !DECL_BY_REFERENCE (var) + && aggregate_value_p (var, current_function_decl))) + /* Allow constant-pool entries that "need to live in memory". */ + && !constant_decl_p (var)) { - reject (var, "needs to live in memory"); + reject (var, "needs to live in memory and escapes or global"); return false; } if (TREE_THIS_VOLATILE (var)) @@ -2122,6 +2192,21 @@ sort_and_splice_var_accesses (tree var) gcc_assert (access->offset >= low && access->offset + access->size <= high); + if (INTEGRAL_TYPE_P (access->type) + && TYPE_PRECISION (access->type) != access->size + && bitmap_bit_p (passed_by_ref_in_call, DECL_UID (access->base))) + { + if (dump_file && (dump_flags & TDF_DETAILS)) + { + fprintf (dump_file, "Won't scalarize "); + print_generic_expr (dump_file, access->base); + fprintf (dump_file, "(%d), it is passed by reference to a call " + "and there are accesses with presicion not covering " + "their type size.", DECL_UID (access->base)); + } + return NULL; + } + grp_same_access_path = path_comparable_for_same_access (access->expr); j = i + 1; @@ -3774,12 +3859,18 @@ get_access_for_expr (tree expr) /* Replace the expression EXPR with a scalar replacement if there is one and generate other statements to do type conversion or subtree copying if - necessary. GSI is used to place newly created statements, WRITE is true if - the expression is being written to (it is on a LHS of a statement or output - in an assembly statement). */ + necessary. WRITE is true if the expression is being written to (it is on a + LHS of a statement or output in an assembly statement). STMT_GSI is used to + place newly created statements before the processed statement, REFRESH_GSI + is used to place them afterwards - unless the processed statement must end a + BB in which case it is placed on the outgoing non-EH edge. REFRESH_GSI and + is then used to continue iteration over the BB. If sra_modify_expr is + called only once with WRITE equal to true on a given statement, both + iterator parameters can point to the same one. */ static bool -sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write) +sra_modify_expr (tree *expr, bool write, gimple_stmt_iterator *stmt_gsi, + gimple_stmt_iterator *refresh_gsi) { location_t loc; struct access *access; @@ -3806,12 +3897,12 @@ sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write) type = TREE_TYPE (*expr); orig_expr = *expr; - loc = gimple_location (gsi_stmt (*gsi)); + loc = gimple_location (gsi_stmt (*stmt_gsi)); gimple_stmt_iterator alt_gsi = gsi_none (); - if (write && stmt_ends_bb_p (gsi_stmt (*gsi))) + if (write && stmt_ends_bb_p (gsi_stmt (*stmt_gsi))) { - alt_gsi = gsi_start_edge (single_non_eh_succ (gsi_bb (*gsi))); - gsi = &alt_gsi; + alt_gsi = gsi_start_edge (single_non_eh_succ (gsi_bb (*stmt_gsi))); + refresh_gsi = &alt_gsi; } if (access->grp_to_be_replaced) @@ -3831,7 +3922,8 @@ sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write) { tree ref; - ref = build_ref_for_model (loc, orig_expr, 0, access, gsi, false); + ref = build_ref_for_model (loc, orig_expr, 0, access, stmt_gsi, + false); if (partial_cplx_access) { @@ -3847,7 +3939,7 @@ sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write) tree tmp = make_ssa_name (type); gassign *stmt = gimple_build_assign (tmp, t); /* This is always a read. */ - gsi_insert_before (gsi, stmt, GSI_SAME_STMT); + gsi_insert_before (stmt_gsi, stmt, GSI_SAME_STMT); t = tmp; } *expr = t; @@ -3857,22 +3949,23 @@ sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write) gassign *stmt; if (access->grp_partial_lhs) - ref = force_gimple_operand_gsi (gsi, ref, true, NULL_TREE, - false, GSI_NEW_STMT); + ref = force_gimple_operand_gsi (refresh_gsi, ref, true, + NULL_TREE, false, GSI_NEW_STMT); stmt = gimple_build_assign (repl, ref); gimple_set_location (stmt, loc); - gsi_insert_after (gsi, stmt, GSI_NEW_STMT); + gsi_insert_after (refresh_gsi, stmt, GSI_NEW_STMT); } else { gassign *stmt; if (access->grp_partial_lhs) - repl = force_gimple_operand_gsi (gsi, repl, true, NULL_TREE, - true, GSI_SAME_STMT); + repl = force_gimple_operand_gsi (stmt_gsi, repl, true, + NULL_TREE, true, + GSI_SAME_STMT); stmt = gimple_build_assign (ref, repl); gimple_set_location (stmt, loc); - gsi_insert_before (gsi, stmt, GSI_SAME_STMT); + gsi_insert_before (stmt_gsi, stmt, GSI_SAME_STMT); } } else @@ -3899,8 +3992,8 @@ sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write) { gdebug *ds = gimple_build_debug_bind (get_access_replacement (access), NULL_TREE, - gsi_stmt (*gsi)); - gsi_insert_after (gsi, ds, GSI_NEW_STMT); + gsi_stmt (*stmt_gsi)); + gsi_insert_after (stmt_gsi, ds, GSI_NEW_STMT); } if (access->first_child && !TREE_READONLY (access->base)) @@ -3918,8 +4011,58 @@ sra_modify_expr (tree *expr, gimple_stmt_iterator *gsi, bool write) start_offset = chunk_size = 0; generate_subtree_copies (access->first_child, orig_expr, access->offset, - start_offset, chunk_size, gsi, write, write, - loc); + start_offset, chunk_size, + write ? refresh_gsi : stmt_gsi, + write, write, loc); + } + return true; +} + +/* If EXPR, which must be a call argument, is an ADDR_EXPR, generate writes and + reads from its base before and after the call statement given in CALL_GSI + and return true if any copying took place. Otherwise call sra_modify_expr + on EXPR and return its value. FLAGS is what the gimple_call_arg_flags + return for the given parameter. */ + +static bool +sra_modify_call_arg (tree *expr, gimple_stmt_iterator *call_gsi, + gimple_stmt_iterator *refresh_gsi, int flags) +{ + if (TREE_CODE (*expr) != ADDR_EXPR) + return sra_modify_expr (expr, false, call_gsi, refresh_gsi); + + if (flags & EAF_UNUSED) + return false; + + tree base = get_base_address (TREE_OPERAND (*expr, 0)); + if (!DECL_P (base)) + return false; + struct access *access = get_access_for_expr (base); + if (!access) + return false; + + gimple *stmt = gsi_stmt (*call_gsi); + location_t loc = gimple_location (stmt); + generate_subtree_copies (access, base, 0, 0, 0, call_gsi, false, false, + loc); + + if ((flags & (EAF_NO_DIRECT_CLOBBER | EAF_NO_INDIRECT_CLOBBER)) + == (EAF_NO_DIRECT_CLOBBER | EAF_NO_INDIRECT_CLOBBER)) + return true; + + if (!stmt_ends_bb_p (stmt)) + generate_subtree_copies (access, base, 0, 0, 0, refresh_gsi, true, + true, loc); + else + { + edge e; + edge_iterator ei; + FOR_EACH_EDGE (e, ei, gsi_bb (*call_gsi)->succs) + { + gimple_stmt_iterator alt_gsi = gsi_start_edge (e); + generate_subtree_copies (access, base, 0, 0, 0, &alt_gsi, true, + true, loc); + } } return true; } @@ -4279,9 +4422,9 @@ sra_modify_assign (gimple *stmt, gimple_stmt_iterator *gsi) || TREE_CODE (lhs) == BIT_FIELD_REF) { modify_this_stmt = sra_modify_expr (gimple_assign_rhs1_ptr (stmt), - gsi, false); + false, gsi, gsi); modify_this_stmt |= sra_modify_expr (gimple_assign_lhs_ptr (stmt), - gsi, true); + true, gsi, gsi); return modify_this_stmt ? SRA_AM_MODIFIED : SRA_AM_NONE; } @@ -4603,7 +4746,7 @@ sra_modify_function_body (void) case GIMPLE_RETURN: t = gimple_return_retval_ptr (as_a (stmt)); if (*t != NULL_TREE) - modified |= sra_modify_expr (t, &gsi, false); + modified |= sra_modify_expr (t, false, &gsi, &gsi); break; case GIMPLE_ASSIGN: @@ -4622,33 +4765,44 @@ sra_modify_function_body (void) } else { + gcall *call = as_a (stmt); + gimple_stmt_iterator call_gsi = gsi; + /* Operands must be processed before the lhs. */ - for (i = 0; i < gimple_call_num_args (stmt); i++) + for (i = 0; i < gimple_call_num_args (call); i++) { - t = gimple_call_arg_ptr (stmt, i); - modified |= sra_modify_expr (t, &gsi, false); + int flags = gimple_call_arg_flags (call, i); + t = gimple_call_arg_ptr (call, i); + modified |= sra_modify_call_arg (t, &call_gsi, &gsi, flags); } - - if (gimple_call_lhs (stmt)) + if (gimple_call_chain (call)) + { + t = gimple_call_chain_ptr (call); + int flags = gimple_call_static_chain_flags (call); + modified |= sra_modify_call_arg (t, &call_gsi, &gsi, + flags); + } + if (gimple_call_lhs (call)) { - t = gimple_call_lhs_ptr (stmt); - modified |= sra_modify_expr (t, &gsi, true); + t = gimple_call_lhs_ptr (call); + modified |= sra_modify_expr (t, true, &call_gsi, &gsi); } } break; case GIMPLE_ASM: { + gimple_stmt_iterator stmt_gsi = gsi; gasm *asm_stmt = as_a (stmt); for (i = 0; i < gimple_asm_ninputs (asm_stmt); i++) { t = &TREE_VALUE (gimple_asm_input_op (asm_stmt, i)); - modified |= sra_modify_expr (t, &gsi, false); + modified |= sra_modify_expr (t, false, &stmt_gsi, &gsi); } for (i = 0; i < gimple_asm_noutputs (asm_stmt); i++) { t = &TREE_VALUE (gimple_asm_output_op (asm_stmt, i)); - modified |= sra_modify_expr (t, &gsi, true); + modified |= sra_modify_expr (t, true, &stmt_gsi, &gsi); } } break;