From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dedi548.your-server.de (dedi548.your-server.de [85.10.215.148]) by sourceware.org (Postfix) with ESMTPS id D80FB3834E32 for ; Fri, 16 Dec 2022 10:25:28 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org D80FB3834E32 Authentication-Results: sourceware.org; dmarc=none (p=none dis=none) header.from=embedded-brains.de Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=embedded-brains.de Received: from sslproxy01.your-server.de ([78.46.139.224]) by dedi548.your-server.de with esmtpsa (TLS1.3) tls TLS_AES_256_GCM_SHA384 (Exim 4.94.2) (envelope-from ) id 1p67uA-000H8u-Hk; Fri, 16 Dec 2022 11:25:26 +0100 Received: from [82.100.198.138] (helo=mail.embedded-brains.de) by sslproxy01.your-server.de with esmtpsa (TLSv1.3:TLS_AES_256_GCM_SHA384:256) (Exim 4.92) (envelope-from ) id 1p67uA-0007dv-EM; Fri, 16 Dec 2022 11:25:26 +0100 Received: from localhost (localhost [127.0.0.1]) by mail.embedded-brains.de (Postfix) with ESMTP id 2383E4800AA; Fri, 16 Dec 2022 11:25:26 +0100 (CET) Received: from mail.embedded-brains.de ([127.0.0.1]) by localhost (zimbra.eb.localhost [127.0.0.1]) (amavisd-new, port 10032) with ESMTP id Ac6Pdeb0lZIG; Fri, 16 Dec 2022 11:25:25 +0100 (CET) Received: from localhost (localhost [127.0.0.1]) by mail.embedded-brains.de (Postfix) with ESMTP id A90AA480181; Fri, 16 Dec 2022 11:25:25 +0100 (CET) X-Virus-Scanned: amavisd-new at zimbra.eb.localhost Received: from mail.embedded-brains.de ([127.0.0.1]) by localhost (zimbra.eb.localhost [127.0.0.1]) (amavisd-new, port 10026) with ESMTP id Z2t9owFjcHTN; Fri, 16 Dec 2022 11:25:25 +0100 (CET) Received: from zimbra.eb.localhost (unknown [192.168.96.242]) by mail.embedded-brains.de (Postfix) with ESMTPSA id 888C24800AA; Fri, 16 Dec 2022 11:25:25 +0100 (CET) From: Sebastian Huber To: gcc-patches@gcc.gnu.org Cc: Richard Biener Subject: [PATCH v2] gcov: Fix -fprofile-update=atomic Date: Fri, 16 Dec 2022 11:25:21 +0100 Message-Id: <20221216102521.73271-1-sebastian.huber@embedded-brains.de> X-Mailer: git-send-email 2.35.3 MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable X-Authenticated-Sender: smtp-embedded@poldinet.de X-Virus-Scanned: Clear (ClamAV 0.103.7/26752/Fri Dec 16 09:25:27 2022) X-Spam-Status: No, score=-11.4 required=5.0 tests=BAYES_00,GIT_PATCH_0,KAM_DMARC_STATUS,SPF_HELO_NONE,SPF_PASS,TXREP autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: The code coverage support uses counters to determine which edges in the c= ontrol flow graph were executed. If a counter overflows, then the code coverage information is invalid. Therefore the counter type should be a 64-bit in= teger. In multithreaded applications, it is important that the counter increment= s are atomic. This is not the case by default. The user can enable atomic cou= nter increments through the -fprofile-update=3Datomic and -fprofile-update=3Dprefer-atomic options. If the hardware supports 64-bit atomic operations, then everything is fin= e. If not and -fprofile-update=3Dprefer-atomic was chosen by the user, then non= -atomic counter increments will be used. However, if the hardware does not suppo= rt the required atomic operations and -fprofile-atomic=3Dupdate was chosen by th= e user, then a warning was issued and as a forced fallback to non-atomic operatio= ns was done. This is probably not what a user wants. There is still hardware o= n the market which does not have atomic operations and is used for multithreade= d applications. A user which selects -fprofile-update=3Datomic wants consi= stent code coverage data and not random data. This patch removes the fallback to non-atomic operations for -fprofile-update=3Datomic. If atomic operations in hardware are not avai= lable, then a library call to libatomic is emitted. To mitigate potential perfo= rmance issues an optimization for systems which only support 32-bit atomic opera= tions is provided. Here, the edge counter increments are done like this: low =3D __atomic_add_fetch_4 (&counter.low, 1, MEMMODEL_RELAXED); high_inc =3D low =3D=3D 0 ? 1 : 0; __atomic_add_fetch_4 (&counter.high, high_inc, MEMMODEL_RELAXED); In gimple_gen_time_profiler() this split operation cannot be used, since = the updated counter value is also required. Here, a library call is emitted.= This is not a performance issue since the update is only done if counters[0] =3D= =3D 0. gcc/ChangeLog: * tree-profile.cc (split_atomic_increment): New. (gimple_gen_edge_profiler): Split the atomic edge counter increment in two 32-bit atomic operations if necessary. (tree_profiling): Remove profile update warning and fallback. Set split_atomic_increment if necessary. * doc/invoke.texi (-fprofile-update): Clarify default method. Document the atomic method behaviour. --- v2: * Check gcov_type_size for split_atomic_increment. * Update documentation. gcc/doc/invoke.texi | 16 ++++++++- gcc/tree-profile.cc | 81 +++++++++++++++++++++++++++++++++------------ 2 files changed, 74 insertions(+), 23 deletions(-) diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi index a50417a4ab7..6b32b659e50 100644 --- a/gcc/doc/invoke.texi +++ b/gcc/doc/invoke.texi @@ -16457,7 +16457,21 @@ while the second one prevents profile corruption= by emitting thread-safe code. Using @samp{prefer-atomic} would be transformed either to @samp{atomic}, when supported by a target, or to @samp{single} otherwise. The GCC driv= er automatically selects @samp{prefer-atomic} when @option{-pthread} -is present in the command line. +is present in the command line, otherwise the default method is @samp{si= ngle}. + +If @samp{atomic} is selected, then the profile information is updated us= ing +atomic operations. If the target does not support atomic operations in +hardware, then function calls to @code{__atomic_fetch_add_4} or +@code{__atomic_fetch_add_8} are emitted. These functions are usually pr= ovided +by the @file{libatomic} runtime library. Not all targets provide the +@file{libatomic} runtime library. If it is not available for the target= , then +a linker error may happen. Using function calls to update the profiling +information may be a performance issue. For targets which use 64-bit co= unters +for the profiling information and support only 32-bit atomic operations,= the +performance critical profiling updates are done using two 32-bit atomic +operations for each counter update. If a signal interrupts these two +operations updating a counter, then the profiling information may be in = an +inconsistent state. =20 @item -fprofile-filter-files=3D@var{regex} @opindex fprofile-filter-files diff --git a/gcc/tree-profile.cc b/gcc/tree-profile.cc index 2beb49241f2..49c8caeae18 100644 --- a/gcc/tree-profile.cc +++ b/gcc/tree-profile.cc @@ -73,6 +73,17 @@ static GTY(()) tree ic_tuple_var; static GTY(()) tree ic_tuple_counters_field; static GTY(()) tree ic_tuple_callee_field; =20 +/* If the user selected atomic profile counter updates + (-fprofile-update=3Datomic), then the counter updates will be done at= omically. + Ideally, this is done through atomic operations in hardware. If the + hardware supports only 32-bit atomic increments and gcov_type_node is= a + 64-bit integer type, then for the profile edge counters the increment= is + performed through two separate 32-bit atomic increments. This case i= s + indicated by the split_atomic_increment variable begin true. If the + hardware does not support atomic operations at all, then a library ca= ll to + libatomic is emitted. */ +static bool split_atomic_increment; + /* Do initialization work for the edge profiler. */ =20 /* Add code: @@ -242,30 +253,59 @@ gimple_init_gcov_profiler (void) void gimple_gen_edge_profiler (int edgeno, edge e) { - tree one; - - one =3D build_int_cst (gcov_type_node, 1); + const char *name =3D "PROF_edge_counter"; + tree ref =3D tree_coverage_counter_ref (GCOV_COUNTER_ARCS, edgeno); + tree one =3D build_int_cst (gcov_type_node, 1); =20 if (flag_profile_update =3D=3D PROFILE_UPDATE_ATOMIC) { - /* __atomic_fetch_add (&counter, 1, MEMMODEL_RELAXED); */ - tree addr =3D tree_coverage_counter_addr (GCOV_COUNTER_ARCS, edgen= o); - tree f =3D builtin_decl_explicit (TYPE_PRECISION (gcov_type_node) = > 32 - ? BUILT_IN_ATOMIC_FETCH_ADD_8: - BUILT_IN_ATOMIC_FETCH_ADD_4); - gcall *stmt =3D gimple_build_call (f, 3, addr, one, - build_int_cst (integer_type_node, - MEMMODEL_RELAXED)); - gsi_insert_on_edge (e, stmt); + tree addr =3D build_fold_addr_expr (ref); + tree relaxed =3D build_int_cst (integer_type_node, MEMMODEL_RELAXE= D); + if (!split_atomic_increment) + { + /* __atomic_fetch_add (&counter, 1, MEMMODEL_RELAXED); */ + tree f =3D builtin_decl_explicit (TYPE_PRECISION (gcov_type_node) > 3= 2 + ? BUILT_IN_ATOMIC_FETCH_ADD_8: + BUILT_IN_ATOMIC_FETCH_ADD_4); + gcall *stmt =3D gimple_build_call (f, 3, addr, one, relaxed); + gsi_insert_on_edge (e, stmt); + } + else + { + /* low =3D __atomic_add_fetch_4 (addr, 1, MEMMODEL_RELAXED); + high_inc =3D low =3D=3D 0 ? 1 : 0; + __atomic_add_fetch_4 (addr_high, high_inc, MEMMODEL_RELAXED); */ + tree zero32 =3D build_zero_cst (uint32_type_node); + tree one32 =3D build_one_cst (uint32_type_node); + tree addr_high =3D make_temp_ssa_name (TREE_TYPE (addr), NULL, name); + gimple *stmt =3D gimple_build_assign (addr_high, POINTER_PLUS_EXPR, + addr, + build_int_cst (size_type_node, + 4)); + gsi_insert_on_edge (e, stmt); + if (WORDS_BIG_ENDIAN) + std::swap (addr, addr_high); + tree f =3D builtin_decl_explicit (BUILT_IN_ATOMIC_ADD_FETCH_4); + stmt =3D gimple_build_call (f, 3, addr, one, relaxed); + tree low =3D make_temp_ssa_name (uint32_type_node, NULL, name); + gimple_call_set_lhs (stmt, low); + gsi_insert_on_edge (e, stmt); + tree is_zero =3D make_temp_ssa_name (boolean_type_node, NULL, name); + stmt =3D gimple_build_assign (is_zero, EQ_EXPR, low, zero32); + gsi_insert_on_edge (e, stmt); + tree high_inc =3D make_temp_ssa_name (uint32_type_node, NULL, name); + stmt =3D gimple_build_assign (high_inc, COND_EXPR, is_zero, one32, + zero32); + gsi_insert_on_edge (e, stmt); + stmt =3D gimple_build_call (f, 3, addr_high, high_inc, relaxed); + gsi_insert_on_edge (e, stmt); + } } else { - tree ref =3D tree_coverage_counter_ref (GCOV_COUNTER_ARCS, edgeno)= ; - tree gcov_type_tmp_var =3D make_temp_ssa_name (gcov_type_node, - NULL, "PROF_edge_counter"); + tree gcov_type_tmp_var =3D make_temp_ssa_name (gcov_type_node, NUL= L, name); gassign *stmt1 =3D gimple_build_assign (gcov_type_tmp_var, ref); - gcov_type_tmp_var =3D make_temp_ssa_name (gcov_type_node, - NULL, "PROF_edge_counter"); + gcov_type_tmp_var =3D make_temp_ssa_name (gcov_type_node, NULL, na= me); gassign *stmt2 =3D gimple_build_assign (gcov_type_tmp_var, PLUS_EX= PR, gimple_assign_lhs (stmt1), one); gassign *stmt3 =3D gimple_build_assign (unshare_expr (ref), @@ -710,11 +750,8 @@ tree_profiling (void) =20 if (flag_profile_update =3D=3D PROFILE_UPDATE_ATOMIC && !can_support_atomic) - { - warning (0, "target does not support atomic profile update, " - "single mode is selected"); - flag_profile_update =3D PROFILE_UPDATE_SINGLE; - } + split_atomic_increment =3D gcov_type_size =3D=3D 8 + && (HAVE_sync_compare_and_swapsi || HAVE_atomic_compare_and_swapsi= ); else if (flag_profile_update =3D=3D PROFILE_UPDATE_PREFER_ATOMIC) flag_profile_update =3D can_support_atomic ? PROFILE_UPDATE_ATOMIC : PROFILE_UPDATE_SINGLE; --=20 2.35.3