Message ID | 20220608182147.4123587-1-ppalka@redhat.com |
---|---|
State | New |
Headers |
Return-Path: <gcc-patches-bounces+patchwork=sourceware.org@gcc.gnu.org> X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id AD3F6388B5B5 for <patchwork@sourceware.org>; Wed, 8 Jun 2022 18:22:54 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org AD3F6388B5B5 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org; s=default; t=1654712574; bh=LFiuKo0xM7ecTyxVzpQdiiIvAaIj9UlssxdAR6AtkUs=; h=To:Subject:Date:List-Id:List-Unsubscribe:List-Archive:List-Post: List-Help:List-Subscribe:From:Reply-To:From; b=TkpSjM35nRB6BY/tgff+DD0IY+DuytoexBmhw/YC9WpzEqlJjTKR09Z/9Z5IezuaE rYLS0L0FTF5Us+q+GYtKDn3CdNE/eIRrAmAXnV0ovZ28xb25mHG7god+XnDv1OWCtE 2l+xzFrm7Of/feAYrSIfH07HywVzPI7ImXfyUmwk= X-Original-To: gcc-patches@gcc.gnu.org Delivered-To: gcc-patches@gcc.gnu.org Received: from us-smtp-delivery-124.mimecast.com (us-smtp-delivery-124.mimecast.com [170.10.133.124]) by sourceware.org (Postfix) with ESMTPS id 6B5FA388B5A2 for <gcc-patches@gcc.gnu.org>; Wed, 8 Jun 2022 18:21:52 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 6B5FA388B5A2 Received: from mail-qv1-f69.google.com (mail-qv1-f69.google.com [209.85.219.69]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id us-mta-112-XVNfHzOQMhi6fOr5ASY79w-1; Wed, 08 Jun 2022 14:21:51 -0400 X-MC-Unique: XVNfHzOQMhi6fOr5ASY79w-1 Received: by mail-qv1-f69.google.com with SMTP id x17-20020a0cfe11000000b004645917e45cso13496550qvr.4 for <gcc-patches@gcc.gnu.org>; Wed, 08 Jun 2022 11:21:51 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:cc:subject:date:message-id:mime-version :content-transfer-encoding; bh=LFiuKo0xM7ecTyxVzpQdiiIvAaIj9UlssxdAR6AtkUs=; b=Dvs1qJh5Sqv3h3+gAkNpqNtm2tNTF5EXfyDtLb+YJx4hMldb5AtPFtzHRsPWtpjOsp 7uIjQ3eIwsKfEy8zN3l5mizXboNSq7QaRHtY57bEhRiHX11NBdRMryWx+lkkUJPU9IJv 8ZFTgZwUdOIyAFffuOaHD9+NnyiuRX6yb3F8oBgdCVf21oXBobZoWBGIOW7zL55nAAuE yN3sM8Nm1GE/XeJHzB8T5UhXSspeloMASiAhKQxBYZV9hcipCHbm3wkYx8evxwwcgFhC Lo/ftkLww409WNYrUszS7JOPeNkl79QUEhOxPMX+Co7LD/ehPdl/uBDYCcFYBqdIxTVn WyTw== X-Gm-Message-State: AOAM530s95jbr10xfPvr7zNbSIAXmo7Xk3J2CqEz/CfnGnvWAycfqTQJ rm86n5vzqeFvY8mETsiSv5c/5ULBCMWsbL7dBO8VKfub58kqNs5NSKp6Zg3UqR+PmfLY0FoOj2x IEZ0ZLb+UTao087ILvYmoZ9OSmh8mgQAPXrScvaXasqGk2vLpqgGMzUH53EmU2bZKhIU= X-Received: by 2002:a05:6214:2aac:b0:46b:b44c:8e4d with SMTP id js12-20020a0562142aac00b0046bb44c8e4dmr9073308qvb.124.1654712510483; Wed, 08 Jun 2022 11:21:50 -0700 (PDT) X-Google-Smtp-Source: ABdhPJxiVlGGlWZY06yWkFutG6vURJxj2q1JB3YXFN3zXfUetDOyDjByWV/+SAlmyfW0kCEM84VH4Q== X-Received: by 2002:a05:6214:2aac:b0:46b:b44c:8e4d with SMTP id js12-20020a0562142aac00b0046bb44c8e4dmr9073288qvb.124.1654712510178; Wed, 08 Jun 2022 11:21:50 -0700 (PDT) Received: from localhost.localdomain (ool-457670bb.dyn.optonline.net. [69.118.112.187]) by smtp.gmail.com with ESMTPSA id bm32-20020a05620a19a000b006a6d20386f6sm5792921qkb.42.2022.06.08.11.21.49 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Wed, 08 Jun 2022 11:21:49 -0700 (PDT) To: gcc-patches@gcc.gnu.org Subject: [PATCH] c++: optimize specialization of nested class templates Date: Wed, 8 Jun 2022 14:21:47 -0400 Message-Id: <20220608182147.4123587-1-ppalka@redhat.com> X-Mailer: git-send-email 2.36.1.363.g9c897eef06 MIME-Version: 1.0 X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: redhat.com Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset="US-ASCII"; x-default=true X-Spam-Status: No, score=-14.5 required=5.0 tests=BAYES_00, DKIMWL_WL_HIGH, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_NONE, 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: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list <gcc-patches.gcc.gnu.org> List-Unsubscribe: <https://gcc.gnu.org/mailman/options/gcc-patches>, <mailto:gcc-patches-request@gcc.gnu.org?subject=unsubscribe> List-Archive: <https://gcc.gnu.org/pipermail/gcc-patches/> List-Post: <mailto:gcc-patches@gcc.gnu.org> List-Help: <mailto:gcc-patches-request@gcc.gnu.org?subject=help> List-Subscribe: <https://gcc.gnu.org/mailman/listinfo/gcc-patches>, <mailto:gcc-patches-request@gcc.gnu.org?subject=subscribe> From: Patrick Palka via Gcc-patches <gcc-patches@gcc.gnu.org> Reply-To: Patrick Palka <ppalka@redhat.com> Errors-To: gcc-patches-bounces+patchwork=sourceware.org@gcc.gnu.org Sender: "Gcc-patches" <gcc-patches-bounces+patchwork=sourceware.org@gcc.gnu.org> |
Series |
c++: optimize specialization of nested class templates
|
|
Commit Message
Patrick Palka
June 8, 2022, 6:21 p.m. UTC
When substituting a class template specialization, tsubst_aggr_type substitutes the TYPE_CONTEXT before passing it to lookup_template_class. This appears to be unnecessary, however, because the the initial value of lookup_template_class's context parameter is unused outside of the IDENTIFIER_NODE case, and l_t_c performs its own substitution of the context, anyway. So this patch removes the redundant substitution in tsubst_aggr_type. Doing so causes us to ICE on template/nested5.C because during lookup_template_class for A<T>::C::D<S> with T=E and S=S, we substitute and complete the context A<T>::C with T=E, which in turn registers the desired dependent specialization of D for us and we end up trying to register it again. This patch fixes this by checking the specializations table again after completion of the context. This patch also implements a couple of other optimizations: * In lookup_template_class, if the context of the partially instantiated template is already non-dependent, then we could reuse that instead of substituting the context of the most general template. * When substituting the TYPE_DECL for an injected-class-name in tsubst_decl, we can avoid substituting its TREE_TYPE and DECL_TI_ARGS. Together these optimizations improve memory usage for the range-v3 testcase test/view/split.cc by about 5%. The improvement is probably more significant when dealing with deeply nested class templates. Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for trunk? gcc/cp/ChangeLog: * pt.cc (lookup_template_class): Remove dead stores to context parameter. Don't substitute the context of the most general template if that of the partially instantiated template is non-dependent. Check the specializations table again after completing the context of a nested dependent specialization. (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute TYPE_CONTEXT or pass it to lookup_template_class. (tsubst_decl) <case TYPE_DECL>: Avoid substituting the TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. --- gcc/cp/pt.cc | 69 +++++++++++++++++++++++++++++++--------------------- 1 file changed, 41 insertions(+), 28 deletions(-)
Comments
On 6/8/22 14:21, Patrick Palka wrote: > When substituting a class template specialization, tsubst_aggr_type > substitutes the TYPE_CONTEXT before passing it to lookup_template_class. > This appears to be unnecessary, however, because the the initial value > of lookup_template_class's context parameter is unused outside of the > IDENTIFIER_NODE case, and l_t_c performs its own substitution of the > context, anyway. So this patch removes the redundant substitution in > tsubst_aggr_type. Doing so causes us to ICE on template/nested5.C > because during lookup_template_class for A<T>::C::D<S> with T=E and S=S, > we substitute and complete the context A<T>::C with T=E, which in turn > registers the desired dependent specialization of D for us and we end up > trying to register it again. This patch fixes this by checking the > specializations table again after completion of the context. > > This patch also implements a couple of other optimizations: > > * In lookup_template_class, if the context of the partially > instantiated template is already non-dependent, then we could > reuse that instead of substituting the context of the most > general template. > * When substituting the TYPE_DECL for an injected-class-name > in tsubst_decl, we can avoid substituting its TREE_TYPE and > DECL_TI_ARGS. > > Together these optimizations improve memory usage for the range-v3 > testcase test/view/split.cc by about 5%. The improvement is probably > more significant when dealing with deeply nested class templates. > > Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for > trunk? > > gcc/cp/ChangeLog: > > * pt.cc (lookup_template_class): Remove dead stores to > context parameter. Don't substitute the context of the > most general template if that of the partially instantiated > template is non-dependent. Check the specializations table > again after completing the context of a nested dependent > specialization. > (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute > TYPE_CONTEXT or pass it to lookup_template_class. > (tsubst_decl) <case TYPE_DECL>: Avoid substituting the > TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. > --- > gcc/cp/pt.cc | 69 +++++++++++++++++++++++++++++++--------------------- > 1 file changed, 41 insertions(+), 28 deletions(-) > > diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc > index 59b94317e88..28023d60684 100644 > --- a/gcc/cp/pt.cc > +++ b/gcc/cp/pt.cc > @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, > if (context) > pop_decl_namespace (); > } > - if (templ) > - context = DECL_CONTEXT (templ); > } > else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE (d1))) > { > @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, > { > templ = d1; > d1 = DECL_NAME (templ); > - context = DECL_CONTEXT (templ); > } > else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) > { > @@ -10059,8 +10056,25 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, > context = DECL_CONTEXT (gen_tmpl); > if (context && TYPE_P (context)) > { > - context = tsubst_aggr_type (context, arglist, complain, in_decl, true); > - context = complete_type (context); > + if (!uses_template_parms (DECL_CONTEXT (templ))) > + /* If the context of the partially instantiated template is > + already non-dependent, then we might as well use it. */ > + context = DECL_CONTEXT (templ); > + else > + { > + context = tsubst_aggr_type (context, arglist, complain, in_decl, true); > + context = complete_type (context); > + if (is_dependent_type && arg_depth > 1) > + { > + /* If this is a dependent nested specialization such as > + A<int>::B<T>, then completion of A<int> might have > + registered this specialization of B for us, so check > + the table again (33959). */ > + entry = type_specializations->find_with_hash (&elt, hash); > + if (entry) > + return entry->spec; > + } > + } > } > else > context = tsubst (context, arglist, complain, in_decl); > @@ -13711,25 +13725,12 @@ tsubst_aggr_type (tree t, > if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) > { > tree argvec; > - tree context; > tree r; > > /* In "sizeof(X<I>)" we need to evaluate "I". */ > cp_evaluated ev; > > - /* First, determine the context for the type we are looking > - up. */ > - context = TYPE_CONTEXT (t); > - if (context && TYPE_P (context)) > - { > - context = tsubst_aggr_type (context, args, complain, > - in_decl, /*entering_scope=*/1); > - /* If context is a nested class inside a class template, > - it may still need to be instantiated (c++/33959). */ > - context = complete_type (context); > - } > - > - /* Then, figure out what arguments are appropriate for the > + /* Figure out what arguments are appropriate for the > type we are trying to find. For example, given: > > template <class T> struct S; > @@ -13744,7 +13745,7 @@ tsubst_aggr_type (tree t, > r = error_mark_node; > else > { > - r = lookup_template_class (t, argvec, in_decl, context, > + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, > entering_scope, complain); > r = cp_build_qualified_type (r, cp_type_quals (t), complain); > } > @@ -14880,6 +14881,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) > ctx = tsubst_aggr_type (ctx, args, > complain, > in_decl, /*entering_scope=*/1); > + if (DECL_SELF_REFERENCE_P (t)) > + /* The context and type of a injected-class-name are > + the same, so we don't need to substitute both. */ > + type = ctx; > /* If CTX is unchanged, then T is in fact the > specialization we want. That situation occurs when > referencing a static data member within in its own > @@ -14900,14 +14905,22 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) > { > tmpl = DECL_TI_TEMPLATE (t); > gen_tmpl = most_general_template (tmpl); > - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); > - if (argvec != error_mark_node) > - argvec = (coerce_innermost_template_parms > - (DECL_TEMPLATE_PARMS (gen_tmpl), > - argvec, t, complain, > - /*all*/true, /*defarg*/true)); > - if (argvec == error_mark_node) > - RETURN (error_mark_node); > + if (DECL_SELF_REFERENCE_P (t)) > + /* The DECL_TI_ARGS for the injected-class-name are the > + generic template arguments for the class template, so > + substitution/coercion is just the identity mapping. */ > + argvec = args; Would it make sense to extend this to any TEMPLATE_DECL for which DECL_PRIMARY_TEMPLATE is the class template? So, anything that gets here except an alias template. > + else > + { > + argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); > + if (argvec != error_mark_node) > + argvec = (coerce_innermost_template_parms > + (DECL_TEMPLATE_PARMS (gen_tmpl), > + argvec, t, complain, > + /*all*/true, /*defarg*/true)); > + if (argvec == error_mark_node) > + RETURN (error_mark_node); > + } > hash = hash_tmpl_and_args (gen_tmpl, argvec); > spec = retrieve_specialization (gen_tmpl, argvec, hash); > }
On Thu, 9 Jun 2022, Jason Merrill wrote: > On 6/8/22 14:21, Patrick Palka wrote: > > When substituting a class template specialization, tsubst_aggr_type > > substitutes the TYPE_CONTEXT before passing it to lookup_template_class. > > This appears to be unnecessary, however, because the the initial value > > of lookup_template_class's context parameter is unused outside of the > > IDENTIFIER_NODE case, and l_t_c performs its own substitution of the > > context, anyway. So this patch removes the redundant substitution in > > tsubst_aggr_type. Doing so causes us to ICE on template/nested5.C > > because during lookup_template_class for A<T>::C::D<S> with T=E and S=S, > > we substitute and complete the context A<T>::C with T=E, which in turn > > registers the desired dependent specialization of D for us and we end up > > trying to register it again. This patch fixes this by checking the > > specializations table again after completion of the context. > > > > This patch also implements a couple of other optimizations: > > > > * In lookup_template_class, if the context of the partially > > instantiated template is already non-dependent, then we could > > reuse that instead of substituting the context of the most > > general template. > > * When substituting the TYPE_DECL for an injected-class-name > > in tsubst_decl, we can avoid substituting its TREE_TYPE and > > DECL_TI_ARGS. > > > > Together these optimizations improve memory usage for the range-v3 > > testcase test/view/split.cc by about 5%. The improvement is probably > > more significant when dealing with deeply nested class templates. > > > > Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for > > trunk? > > > > gcc/cp/ChangeLog: > > > > * pt.cc (lookup_template_class): Remove dead stores to > > context parameter. Don't substitute the context of the > > most general template if that of the partially instantiated > > template is non-dependent. Check the specializations table > > again after completing the context of a nested dependent > > specialization. > > (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute > > TYPE_CONTEXT or pass it to lookup_template_class. > > (tsubst_decl) <case TYPE_DECL>: Avoid substituting the > > TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. > > --- > > gcc/cp/pt.cc | 69 +++++++++++++++++++++++++++++++--------------------- > > 1 file changed, 41 insertions(+), 28 deletions(-) > > > > diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc > > index 59b94317e88..28023d60684 100644 > > --- a/gcc/cp/pt.cc > > +++ b/gcc/cp/pt.cc > > @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree > > in_decl, tree context, > > if (context) > > pop_decl_namespace (); > > } > > - if (templ) > > - context = DECL_CONTEXT (templ); > > } > > else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE > > (d1))) > > { > > @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree > > in_decl, tree context, > > { > > templ = d1; > > d1 = DECL_NAME (templ); > > - context = DECL_CONTEXT (templ); > > } > > else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) > > { > > @@ -10059,8 +10056,25 @@ lookup_template_class (tree d1, tree arglist, tree > > in_decl, tree context, > > context = DECL_CONTEXT (gen_tmpl); > > if (context && TYPE_P (context)) > > { > > - context = tsubst_aggr_type (context, arglist, complain, in_decl, > > true); > > - context = complete_type (context); > > + if (!uses_template_parms (DECL_CONTEXT (templ))) > > + /* If the context of the partially instantiated template is > > + already non-dependent, then we might as well use it. */ > > + context = DECL_CONTEXT (templ); > > + else > > + { > > + context = tsubst_aggr_type (context, arglist, complain, in_decl, > > true); > > + context = complete_type (context); > > + if (is_dependent_type && arg_depth > 1) > > + { > > + /* If this is a dependent nested specialization such as > > + A<int>::B<T>, then completion of A<int> might have > > + registered this specialization of B for us, so check > > + the table again (33959). */ > > + entry = type_specializations->find_with_hash (&elt, hash); > > + if (entry) > > + return entry->spec; > > + } > > + } > > } > > else > > context = tsubst (context, arglist, complain, in_decl); > > @@ -13711,25 +13725,12 @@ tsubst_aggr_type (tree t, > > if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) > > { > > tree argvec; > > - tree context; > > tree r; > > /* In "sizeof(X<I>)" we need to evaluate "I". */ > > cp_evaluated ev; > > - /* First, determine the context for the type we are looking > > - up. */ > > - context = TYPE_CONTEXT (t); > > - if (context && TYPE_P (context)) > > - { > > - context = tsubst_aggr_type (context, args, complain, > > - in_decl, /*entering_scope=*/1); > > - /* If context is a nested class inside a class template, > > - it may still need to be instantiated (c++/33959). */ > > - context = complete_type (context); > > - } > > - > > - /* Then, figure out what arguments are appropriate for the > > + /* Figure out what arguments are appropriate for the > > type we are trying to find. For example, given: > > template <class T> struct S; > > @@ -13744,7 +13745,7 @@ tsubst_aggr_type (tree t, > > r = error_mark_node; > > else > > { > > - r = lookup_template_class (t, argvec, in_decl, context, > > + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, > > entering_scope, complain); > > r = cp_build_qualified_type (r, cp_type_quals (t), complain); > > } > > @@ -14880,6 +14881,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t > > complain) > > ctx = tsubst_aggr_type (ctx, args, > > complain, > > in_decl, /*entering_scope=*/1); > > + if (DECL_SELF_REFERENCE_P (t)) > > + /* The context and type of a injected-class-name are > > + the same, so we don't need to substitute both. */ > > + type = ctx; > > /* If CTX is unchanged, then T is in fact the > > specialization we want. That situation occurs when > > referencing a static data member within in its own > > @@ -14900,14 +14905,22 @@ tsubst_decl (tree t, tree args, tsubst_flags_t > > complain) > > { > > tmpl = DECL_TI_TEMPLATE (t); > > gen_tmpl = most_general_template (tmpl); > > - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); > > - if (argvec != error_mark_node) > > - argvec = (coerce_innermost_template_parms > > - (DECL_TEMPLATE_PARMS (gen_tmpl), > > - argvec, t, complain, > > - /*all*/true, /*defarg*/true)); > > - if (argvec == error_mark_node) > > - RETURN (error_mark_node); > > + if (DECL_SELF_REFERENCE_P (t)) > > + /* The DECL_TI_ARGS for the injected-class-name are the > > + generic template arguments for the class template, so > > + substitution/coercion is just the identity mapping. */ > > + argvec = args; > > Would it make sense to extend this to any TEMPLATE_DECL for which > DECL_PRIMARY_TEMPLATE is the class template? So, anything that gets here > except an alias template. Ah nice, it does look like we could extend this to any TEMPLATE_DECL that satisfies DECL_CLASS_SCOPE_P && !DECL_MEMBER_TEMPLATE_P (so that we also exclude static data member templates), i.e. any templated non-template member. Is that equivalent to what you had in mind? Based on some light testing, if we do this, it seems we also need to handle 'args' having greater depth than 'DECL_TI_ARGS' here, something which could happen during satisfaction. I can work on extending this as a followup patch. > > > + else > > + { > > + argvec = tsubst (DECL_TI_ARGS (t), args, complain, > > in_decl); > > + if (argvec != error_mark_node) > > + argvec = (coerce_innermost_template_parms > > + (DECL_TEMPLATE_PARMS (gen_tmpl), > > + argvec, t, complain, > > + /*all*/true, /*defarg*/true)); > > + if (argvec == error_mark_node) > > + RETURN (error_mark_node); > > + } > > hash = hash_tmpl_and_args (gen_tmpl, argvec); > > spec = retrieve_specialization (gen_tmpl, argvec, hash); > > } > >
On Thu, 9 Jun 2022, Patrick Palka wrote: > On Thu, 9 Jun 2022, Jason Merrill wrote: > > > On 6/8/22 14:21, Patrick Palka wrote: > > > When substituting a class template specialization, tsubst_aggr_type > > > substitutes the TYPE_CONTEXT before passing it to lookup_template_class. > > > This appears to be unnecessary, however, because the the initial value > > > of lookup_template_class's context parameter is unused outside of the > > > IDENTIFIER_NODE case, and l_t_c performs its own substitution of the > > > context, anyway. So this patch removes the redundant substitution in > > > tsubst_aggr_type. Doing so causes us to ICE on template/nested5.C > > > because during lookup_template_class for A<T>::C::D<S> with T=E and S=S, > > > we substitute and complete the context A<T>::C with T=E, which in turn > > > registers the desired dependent specialization of D for us and we end up > > > trying to register it again. This patch fixes this by checking the > > > specializations table again after completion of the context. > > > > > > This patch also implements a couple of other optimizations: > > > > > > * In lookup_template_class, if the context of the partially > > > instantiated template is already non-dependent, then we could > > > reuse that instead of substituting the context of the most > > > general template. > > > * When substituting the TYPE_DECL for an injected-class-name > > > in tsubst_decl, we can avoid substituting its TREE_TYPE and > > > DECL_TI_ARGS. > > > > > > Together these optimizations improve memory usage for the range-v3 > > > testcase test/view/split.cc by about 5%. The improvement is probably > > > more significant when dealing with deeply nested class templates. > > > > > > Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for > > > trunk? > > > > > > gcc/cp/ChangeLog: > > > > > > * pt.cc (lookup_template_class): Remove dead stores to > > > context parameter. Don't substitute the context of the > > > most general template if that of the partially instantiated > > > template is non-dependent. Check the specializations table > > > again after completing the context of a nested dependent > > > specialization. > > > (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute > > > TYPE_CONTEXT or pass it to lookup_template_class. > > > (tsubst_decl) <case TYPE_DECL>: Avoid substituting the > > > TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. > > > --- > > > gcc/cp/pt.cc | 69 +++++++++++++++++++++++++++++++--------------------- > > > 1 file changed, 41 insertions(+), 28 deletions(-) > > > > > > diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc > > > index 59b94317e88..28023d60684 100644 > > > --- a/gcc/cp/pt.cc > > > +++ b/gcc/cp/pt.cc > > > @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree > > > in_decl, tree context, > > > if (context) > > > pop_decl_namespace (); > > > } > > > - if (templ) > > > - context = DECL_CONTEXT (templ); > > > } > > > else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE > > > (d1))) > > > { > > > @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree > > > in_decl, tree context, > > > { > > > templ = d1; > > > d1 = DECL_NAME (templ); > > > - context = DECL_CONTEXT (templ); > > > } > > > else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) > > > { > > > @@ -10059,8 +10056,25 @@ lookup_template_class (tree d1, tree arglist, tree > > > in_decl, tree context, > > > context = DECL_CONTEXT (gen_tmpl); > > > if (context && TYPE_P (context)) > > > { > > > - context = tsubst_aggr_type (context, arglist, complain, in_decl, > > > true); > > > - context = complete_type (context); > > > + if (!uses_template_parms (DECL_CONTEXT (templ))) > > > + /* If the context of the partially instantiated template is > > > + already non-dependent, then we might as well use it. */ > > > + context = DECL_CONTEXT (templ); > > > + else > > > + { > > > + context = tsubst_aggr_type (context, arglist, complain, in_decl, > > > true); > > > + context = complete_type (context); > > > + if (is_dependent_type && arg_depth > 1) > > > + { > > > + /* If this is a dependent nested specialization such as > > > + A<int>::B<T>, then completion of A<int> might have > > > + registered this specialization of B for us, so check > > > + the table again (33959). */ > > > + entry = type_specializations->find_with_hash (&elt, hash); > > > + if (entry) > > > + return entry->spec; > > > + } > > > + } > > > } > > > else > > > context = tsubst (context, arglist, complain, in_decl); > > > @@ -13711,25 +13725,12 @@ tsubst_aggr_type (tree t, > > > if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) > > > { > > > tree argvec; > > > - tree context; > > > tree r; > > > /* In "sizeof(X<I>)" we need to evaluate "I". */ > > > cp_evaluated ev; > > > - /* First, determine the context for the type we are looking > > > - up. */ > > > - context = TYPE_CONTEXT (t); > > > - if (context && TYPE_P (context)) > > > - { > > > - context = tsubst_aggr_type (context, args, complain, > > > - in_decl, /*entering_scope=*/1); > > > - /* If context is a nested class inside a class template, > > > - it may still need to be instantiated (c++/33959). */ > > > - context = complete_type (context); > > > - } > > > - > > > - /* Then, figure out what arguments are appropriate for the > > > + /* Figure out what arguments are appropriate for the > > > type we are trying to find. For example, given: > > > template <class T> struct S; > > > @@ -13744,7 +13745,7 @@ tsubst_aggr_type (tree t, > > > r = error_mark_node; > > > else > > > { > > > - r = lookup_template_class (t, argvec, in_decl, context, > > > + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, > > > entering_scope, complain); > > > r = cp_build_qualified_type (r, cp_type_quals (t), complain); > > > } > > > @@ -14880,6 +14881,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t > > > complain) > > > ctx = tsubst_aggr_type (ctx, args, > > > complain, > > > in_decl, /*entering_scope=*/1); > > > + if (DECL_SELF_REFERENCE_P (t)) > > > + /* The context and type of a injected-class-name are > > > + the same, so we don't need to substitute both. */ > > > + type = ctx; > > > /* If CTX is unchanged, then T is in fact the > > > specialization we want. That situation occurs when > > > referencing a static data member within in its own > > > @@ -14900,14 +14905,22 @@ tsubst_decl (tree t, tree args, tsubst_flags_t > > > complain) > > > { > > > tmpl = DECL_TI_TEMPLATE (t); > > > gen_tmpl = most_general_template (tmpl); > > > - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); > > > - if (argvec != error_mark_node) > > > - argvec = (coerce_innermost_template_parms > > > - (DECL_TEMPLATE_PARMS (gen_tmpl), > > > - argvec, t, complain, > > > - /*all*/true, /*defarg*/true)); > > > - if (argvec == error_mark_node) > > > - RETURN (error_mark_node); > > > + if (DECL_SELF_REFERENCE_P (t)) > > > + /* The DECL_TI_ARGS for the injected-class-name are the > > > + generic template arguments for the class template, so > > > + substitution/coercion is just the identity mapping. */ > > > + argvec = args; > > > > Would it make sense to extend this to any TEMPLATE_DECL for which > > DECL_PRIMARY_TEMPLATE is the class template? So, anything that gets here > > except an alias template. > > Ah nice, it does look like we could extend this to any TEMPLATE_DECL that > satisfies DECL_CLASS_SCOPE_P && !DECL_MEMBER_TEMPLATE_P (so that we also > exclude static data member templates), i.e. any templated non-template > member. Is that equivalent to what you had in mind? > > Based on some light testing, if we do this, it seems we also need to > handle 'args' having greater depth than 'DECL_TI_ARGS' here, something > which could happen during satisfaction. ... or we could just restrict the optimization to when the argument depths match, like so: -- >8 -- Subject: [PATCH] c++: optimize specialization of nested templated classes gcc/cp/ChangeLog: * pt.cc (lookup_template_class): Remove dead stores to context parameter. Don't substitute the context of the most general template if that of the partially instantiated template is non-dependent. Check the specializations table again after completing the context of a nested dependent specialization. (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute TYPE_CONTEXT or pass it to lookup_template_class. (tsubst_decl) <case TYPE_DECL>: Avoid substituting the TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. --- gcc/cp/pt.cc | 73 ++++++++++++++++++++++++++++++++-------------------- 1 file changed, 45 insertions(+), 28 deletions(-) diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc index 3154186ac20..ebd822373db 100644 --- a/gcc/cp/pt.cc +++ b/gcc/cp/pt.cc @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, if (context) pop_decl_namespace (); } - if (templ) - context = DECL_CONTEXT (templ); } else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE (d1))) { @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, { templ = d1; d1 = DECL_NAME (templ); - context = DECL_CONTEXT (templ); } else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) { @@ -10059,8 +10056,26 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, context = DECL_CONTEXT (gen_tmpl); if (context && TYPE_P (context)) { - context = tsubst_aggr_type (context, arglist, complain, in_decl, true); - context = complete_type (context); + if (!uses_template_parms (DECL_CONTEXT (templ))) + /* If the context of the partially instantiated template is + already non-dependent, then we might as well use it. */ + context = DECL_CONTEXT (templ); + else + { + context = tsubst_aggr_type (context, arglist, + complain, in_decl, true); + context = complete_type (context); + if (is_dependent_type && arg_depth > 1) + { + /* If this is a dependent nested specialization such as + A<int>::B<T>, then completion of A<int> might have + registered this specialization of B for us, so check + the table again (33959). */ + entry = type_specializations->find_with_hash (&elt, hash); + if (entry) + return entry->spec; + } + } } else context = tsubst (context, arglist, complain, in_decl); @@ -13739,25 +13754,12 @@ tsubst_aggr_type (tree t, if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) { tree argvec; - tree context; tree r; /* In "sizeof(X<I>)" we need to evaluate "I". */ cp_evaluated ev; - /* First, determine the context for the type we are looking - up. */ - context = TYPE_CONTEXT (t); - if (context && TYPE_P (context)) - { - context = tsubst_aggr_type (context, args, complain, - in_decl, /*entering_scope=*/1); - /* If context is a nested class inside a class template, - it may still need to be instantiated (c++/33959). */ - context = complete_type (context); - } - - /* Then, figure out what arguments are appropriate for the + /* Figure out what arguments are appropriate for the type we are trying to find. For example, given: template <class T> struct S; @@ -13772,7 +13774,7 @@ tsubst_aggr_type (tree t, r = error_mark_node; else { - r = lookup_template_class (t, argvec, in_decl, context, + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, entering_scope, complain); r = cp_build_qualified_type (r, cp_type_quals (t), complain); } @@ -14913,6 +14915,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) ctx = tsubst_aggr_type (ctx, args, complain, in_decl, /*entering_scope=*/1); + if (DECL_SELF_REFERENCE_P (t)) + /* The context and type of an injected-class-name are + the same, so we don't need to substitute both. */ + type = ctx; /* If CTX is unchanged, then T is in fact the specialization we want. That situation occurs when referencing a static data member within in its own @@ -14933,14 +14939,25 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) { tmpl = DECL_TI_TEMPLATE (t); gen_tmpl = most_general_template (tmpl); - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); - if (argvec != error_mark_node) - argvec = (coerce_innermost_template_parms - (DECL_TEMPLATE_PARMS (gen_tmpl), - argvec, t, complain, - /*all*/true, /*defarg*/true)); - if (argvec == error_mark_node) - RETURN (error_mark_node); + if (DECL_CLASS_SCOPE_P (tmpl) + && !DECL_MEMBER_TEMPLATE_P (tmpl) + && (TMPL_ARGS_DEPTH (args) + == TMPL_ARGS_DEPTH (DECL_TI_ARGS (t)))) + /* The DECL_TI_ARGS in this case are the generic template + arguments for the class template, so substitution/coercion + is just the identity mapping. */ + argvec = args; + else + { + argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); + if (argvec != error_mark_node) + argvec = (coerce_innermost_template_parms + (DECL_TEMPLATE_PARMS (gen_tmpl), + argvec, t, complain, + /*all*/true, /*defarg*/true)); + if (argvec == error_mark_node) + RETURN (error_mark_node); + } hash = hash_tmpl_and_args (gen_tmpl, argvec); spec = retrieve_specialization (gen_tmpl, argvec, hash); }
On Fri, 10 Jun 2022, Patrick Palka wrote: > On Thu, 9 Jun 2022, Patrick Palka wrote: > > > On Thu, 9 Jun 2022, Jason Merrill wrote: > > > > > On 6/8/22 14:21, Patrick Palka wrote: > > > > When substituting a class template specialization, tsubst_aggr_type > > > > substitutes the TYPE_CONTEXT before passing it to lookup_template_class. > > > > This appears to be unnecessary, however, because the the initial value > > > > of lookup_template_class's context parameter is unused outside of the > > > > IDENTIFIER_NODE case, and l_t_c performs its own substitution of the > > > > context, anyway. So this patch removes the redundant substitution in > > > > tsubst_aggr_type. Doing so causes us to ICE on template/nested5.C > > > > because during lookup_template_class for A<T>::C::D<S> with T=E and S=S, > > > > we substitute and complete the context A<T>::C with T=E, which in turn > > > > registers the desired dependent specialization of D for us and we end up > > > > trying to register it again. This patch fixes this by checking the > > > > specializations table again after completion of the context. > > > > > > > > This patch also implements a couple of other optimizations: > > > > > > > > * In lookup_template_class, if the context of the partially > > > > instantiated template is already non-dependent, then we could > > > > reuse that instead of substituting the context of the most > > > > general template. > > > > * When substituting the TYPE_DECL for an injected-class-name > > > > in tsubst_decl, we can avoid substituting its TREE_TYPE and > > > > DECL_TI_ARGS. > > > > > > > > Together these optimizations improve memory usage for the range-v3 > > > > testcase test/view/split.cc by about 5%. The improvement is probably > > > > more significant when dealing with deeply nested class templates. > > > > > > > > Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for > > > > trunk? > > > > > > > > gcc/cp/ChangeLog: > > > > > > > > * pt.cc (lookup_template_class): Remove dead stores to > > > > context parameter. Don't substitute the context of the > > > > most general template if that of the partially instantiated > > > > template is non-dependent. Check the specializations table > > > > again after completing the context of a nested dependent > > > > specialization. > > > > (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute > > > > TYPE_CONTEXT or pass it to lookup_template_class. > > > > (tsubst_decl) <case TYPE_DECL>: Avoid substituting the > > > > TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. > > > > --- > > > > gcc/cp/pt.cc | 69 +++++++++++++++++++++++++++++++--------------------- > > > > 1 file changed, 41 insertions(+), 28 deletions(-) > > > > > > > > diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc > > > > index 59b94317e88..28023d60684 100644 > > > > --- a/gcc/cp/pt.cc > > > > +++ b/gcc/cp/pt.cc > > > > @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree > > > > in_decl, tree context, > > > > if (context) > > > > pop_decl_namespace (); > > > > } > > > > - if (templ) > > > > - context = DECL_CONTEXT (templ); > > > > } > > > > else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE > > > > (d1))) > > > > { > > > > @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree > > > > in_decl, tree context, > > > > { > > > > templ = d1; > > > > d1 = DECL_NAME (templ); > > > > - context = DECL_CONTEXT (templ); > > > > } > > > > else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) > > > > { > > > > @@ -10059,8 +10056,25 @@ lookup_template_class (tree d1, tree arglist, tree > > > > in_decl, tree context, > > > > context = DECL_CONTEXT (gen_tmpl); > > > > if (context && TYPE_P (context)) > > > > { > > > > - context = tsubst_aggr_type (context, arglist, complain, in_decl, > > > > true); > > > > - context = complete_type (context); > > > > + if (!uses_template_parms (DECL_CONTEXT (templ))) > > > > + /* If the context of the partially instantiated template is > > > > + already non-dependent, then we might as well use it. */ > > > > + context = DECL_CONTEXT (templ); > > > > + else > > > > + { > > > > + context = tsubst_aggr_type (context, arglist, complain, in_decl, > > > > true); > > > > + context = complete_type (context); > > > > + if (is_dependent_type && arg_depth > 1) > > > > + { > > > > + /* If this is a dependent nested specialization such as > > > > + A<int>::B<T>, then completion of A<int> might have > > > > + registered this specialization of B for us, so check > > > > + the table again (33959). */ > > > > + entry = type_specializations->find_with_hash (&elt, hash); > > > > + if (entry) > > > > + return entry->spec; > > > > + } > > > > + } > > > > } > > > > else > > > > context = tsubst (context, arglist, complain, in_decl); > > > > @@ -13711,25 +13725,12 @@ tsubst_aggr_type (tree t, > > > > if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) > > > > { > > > > tree argvec; > > > > - tree context; > > > > tree r; > > > > /* In "sizeof(X<I>)" we need to evaluate "I". */ > > > > cp_evaluated ev; > > > > - /* First, determine the context for the type we are looking > > > > - up. */ > > > > - context = TYPE_CONTEXT (t); > > > > - if (context && TYPE_P (context)) > > > > - { > > > > - context = tsubst_aggr_type (context, args, complain, > > > > - in_decl, /*entering_scope=*/1); > > > > - /* If context is a nested class inside a class template, > > > > - it may still need to be instantiated (c++/33959). */ > > > > - context = complete_type (context); > > > > - } > > > > - > > > > - /* Then, figure out what arguments are appropriate for the > > > > + /* Figure out what arguments are appropriate for the > > > > type we are trying to find. For example, given: > > > > template <class T> struct S; > > > > @@ -13744,7 +13745,7 @@ tsubst_aggr_type (tree t, > > > > r = error_mark_node; > > > > else > > > > { > > > > - r = lookup_template_class (t, argvec, in_decl, context, > > > > + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, > > > > entering_scope, complain); > > > > r = cp_build_qualified_type (r, cp_type_quals (t), complain); > > > > } > > > > @@ -14880,6 +14881,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t > > > > complain) > > > > ctx = tsubst_aggr_type (ctx, args, > > > > complain, > > > > in_decl, /*entering_scope=*/1); > > > > + if (DECL_SELF_REFERENCE_P (t)) > > > > + /* The context and type of a injected-class-name are > > > > + the same, so we don't need to substitute both. */ > > > > + type = ctx; > > > > /* If CTX is unchanged, then T is in fact the > > > > specialization we want. That situation occurs when > > > > referencing a static data member within in its own > > > > @@ -14900,14 +14905,22 @@ tsubst_decl (tree t, tree args, tsubst_flags_t > > > > complain) > > > > { > > > > tmpl = DECL_TI_TEMPLATE (t); > > > > gen_tmpl = most_general_template (tmpl); > > > > - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); > > > > - if (argvec != error_mark_node) > > > > - argvec = (coerce_innermost_template_parms > > > > - (DECL_TEMPLATE_PARMS (gen_tmpl), > > > > - argvec, t, complain, > > > > - /*all*/true, /*defarg*/true)); > > > > - if (argvec == error_mark_node) > > > > - RETURN (error_mark_node); > > > > + if (DECL_SELF_REFERENCE_P (t)) > > > > + /* The DECL_TI_ARGS for the injected-class-name are the > > > > + generic template arguments for the class template, so > > > > + substitution/coercion is just the identity mapping. */ > > > > + argvec = args; > > > > > > Would it make sense to extend this to any TEMPLATE_DECL for which > > > DECL_PRIMARY_TEMPLATE is the class template? So, anything that gets here > > > except an alias template. > > > > Ah nice, it does look like we could extend this to any TEMPLATE_DECL that > > satisfies DECL_CLASS_SCOPE_P && !DECL_MEMBER_TEMPLATE_P (so that we also > > exclude static data member templates), i.e. any templated non-template > > member. Is that equivalent to what you had in mind? > > > > Based on some light testing, if we do this, it seems we also need to > > handle 'args' having greater depth than 'DECL_TI_ARGS' here, something > > which could happen during satisfaction. > > ... or we could just restrict the optimization to when the argument > depths match, like so: > > -- >8 -- > > Subject: [PATCH] c++: optimize specialization of nested templated classes > > gcc/cp/ChangeLog: > > * pt.cc (lookup_template_class): Remove dead stores to > context parameter. Don't substitute the context of the > most general template if that of the partially instantiated > template is non-dependent. Check the specializations table > again after completing the context of a nested dependent > specialization. > (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute > TYPE_CONTEXT or pass it to lookup_template_class. > (tsubst_decl) <case TYPE_DECL>: Avoid substituting the > TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. > --- > gcc/cp/pt.cc | 73 ++++++++++++++++++++++++++++++++-------------------- > 1 file changed, 45 insertions(+), 28 deletions(-) > > diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc > index 3154186ac20..ebd822373db 100644 > --- a/gcc/cp/pt.cc > +++ b/gcc/cp/pt.cc > @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, > if (context) > pop_decl_namespace (); > } > - if (templ) > - context = DECL_CONTEXT (templ); > } > else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE (d1))) > { > @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, > { > templ = d1; > d1 = DECL_NAME (templ); > - context = DECL_CONTEXT (templ); > } > else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) > { > @@ -10059,8 +10056,26 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, > context = DECL_CONTEXT (gen_tmpl); > if (context && TYPE_P (context)) > { > - context = tsubst_aggr_type (context, arglist, complain, in_decl, true); > - context = complete_type (context); > + if (!uses_template_parms (DECL_CONTEXT (templ))) > + /* If the context of the partially instantiated template is > + already non-dependent, then we might as well use it. */ > + context = DECL_CONTEXT (templ); > + else > + { > + context = tsubst_aggr_type (context, arglist, > + complain, in_decl, true); > + context = complete_type (context); > + if (is_dependent_type && arg_depth > 1) > + { > + /* If this is a dependent nested specialization such as > + A<int>::B<T>, then completion of A<int> might have > + registered this specialization of B for us, so check > + the table again (33959). */ > + entry = type_specializations->find_with_hash (&elt, hash); > + if (entry) > + return entry->spec; > + } > + } > } > else > context = tsubst (context, arglist, complain, in_decl); > @@ -13739,25 +13754,12 @@ tsubst_aggr_type (tree t, > if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) > { > tree argvec; > - tree context; > tree r; > > /* In "sizeof(X<I>)" we need to evaluate "I". */ > cp_evaluated ev; > > - /* First, determine the context for the type we are looking > - up. */ > - context = TYPE_CONTEXT (t); > - if (context && TYPE_P (context)) > - { > - context = tsubst_aggr_type (context, args, complain, > - in_decl, /*entering_scope=*/1); > - /* If context is a nested class inside a class template, > - it may still need to be instantiated (c++/33959). */ > - context = complete_type (context); > - } > - > - /* Then, figure out what arguments are appropriate for the > + /* Figure out what arguments are appropriate for the > type we are trying to find. For example, given: > > template <class T> struct S; > @@ -13772,7 +13774,7 @@ tsubst_aggr_type (tree t, > r = error_mark_node; > else > { > - r = lookup_template_class (t, argvec, in_decl, context, > + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, > entering_scope, complain); > r = cp_build_qualified_type (r, cp_type_quals (t), complain); > } > @@ -14913,6 +14915,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) > ctx = tsubst_aggr_type (ctx, args, > complain, > in_decl, /*entering_scope=*/1); > + if (DECL_SELF_REFERENCE_P (t)) > + /* The context and type of an injected-class-name are > + the same, so we don't need to substitute both. */ > + type = ctx; > /* If CTX is unchanged, then T is in fact the > specialization we want. That situation occurs when > referencing a static data member within in its own > @@ -14933,14 +14939,25 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) > { > tmpl = DECL_TI_TEMPLATE (t); > gen_tmpl = most_general_template (tmpl); > - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); > - if (argvec != error_mark_node) > - argvec = (coerce_innermost_template_parms > - (DECL_TEMPLATE_PARMS (gen_tmpl), > - argvec, t, complain, > - /*all*/true, /*defarg*/true)); > - if (argvec == error_mark_node) > - RETURN (error_mark_node); > + if (DECL_CLASS_SCOPE_P (tmpl) > + && !DECL_MEMBER_TEMPLATE_P (tmpl) > + && (TMPL_ARGS_DEPTH (args) > + == TMPL_ARGS_DEPTH (DECL_TI_ARGS (t)))) > + /* The DECL_TI_ARGS in this case are the generic template > + arguments for the class template, so substitution/coercion > + is just the identity mapping. */ > + argvec = args; > + else > + { > + argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); > + if (argvec != error_mark_node) > + argvec = (coerce_innermost_template_parms > + (DECL_TEMPLATE_PARMS (gen_tmpl), > + argvec, t, complain, > + /*all*/true, /*defarg*/true)); > + if (argvec == error_mark_node) > + RETURN (error_mark_node); > + } Seems we could be a bit cleverer here by trying to avoid the coercion even if we have to do the substitution. When the argument depth is less than the parameter depth, then the innermost substituted arguments are just lowered arguments from DECL_TI_ARGS, for which coercion would be a no-op, so we can just skip it. This optimization seems to trigger 90+% of the time and compile time/memory by about 1%/0.5% in my experiments. (The call to coerce_innermost_template_parms was added by r6-829-gb237c4cbd3da7a FWIW.) -- >8 -- Subject: [PATCH] c++: optimize specialization of nested templated classes gcc/cp/ChangeLog: * pt.cc (lookup_template_class): Remove dead stores to context parameter. Don't substitute the context of the most general template if that of the partially instantiated template is non-dependent. Check the specializations table again after completing the context of a nested dependent specialization. (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute TYPE_CONTEXT or pass it to lookup_template_class. (tsubst_decl) <case TYPE_DECL>: Avoid substituting the TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. --- gcc/cp/pt.cc | 74 ++++++++++++++++++++++++++++++++-------------------- 1 file changed, 46 insertions(+), 28 deletions(-) diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc index 3154186ac20..faa5bc377cb 100644 --- a/gcc/cp/pt.cc +++ b/gcc/cp/pt.cc @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, if (context) pop_decl_namespace (); } - if (templ) - context = DECL_CONTEXT (templ); } else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE (d1))) { @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, { templ = d1; d1 = DECL_NAME (templ); - context = DECL_CONTEXT (templ); } else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) { @@ -10059,8 +10056,26 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, context = DECL_CONTEXT (gen_tmpl); if (context && TYPE_P (context)) { - context = tsubst_aggr_type (context, arglist, complain, in_decl, true); - context = complete_type (context); + if (!uses_template_parms (DECL_CONTEXT (templ))) + /* If the context of the partially instantiated template is + already non-dependent, then we might as well use it. */ + context = DECL_CONTEXT (templ); + else + { + context = tsubst_aggr_type (context, arglist, + complain, in_decl, true); + context = complete_type (context); + if (is_dependent_type && arg_depth > 1) + { + /* If this is a dependent nested specialization such as + A<int>::B<T>, then completion of A<int> might have + registered this specialization of B for us, so check + the table again (33959). */ + entry = type_specializations->find_with_hash (&elt, hash); + if (entry) + return entry->spec; + } + } } else context = tsubst (context, arglist, complain, in_decl); @@ -13739,25 +13754,12 @@ tsubst_aggr_type (tree t, if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) { tree argvec; - tree context; tree r; /* In "sizeof(X<I>)" we need to evaluate "I". */ cp_evaluated ev; - /* First, determine the context for the type we are looking - up. */ - context = TYPE_CONTEXT (t); - if (context && TYPE_P (context)) - { - context = tsubst_aggr_type (context, args, complain, - in_decl, /*entering_scope=*/1); - /* If context is a nested class inside a class template, - it may still need to be instantiated (c++/33959). */ - context = complete_type (context); - } - - /* Then, figure out what arguments are appropriate for the + /* Figure out what arguments are appropriate for the type we are trying to find. For example, given: template <class T> struct S; @@ -13772,7 +13774,7 @@ tsubst_aggr_type (tree t, r = error_mark_node; else { - r = lookup_template_class (t, argvec, in_decl, context, + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, entering_scope, complain); r = cp_build_qualified_type (r, cp_type_quals (t), complain); } @@ -14913,6 +14915,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) ctx = tsubst_aggr_type (ctx, args, complain, in_decl, /*entering_scope=*/1); + if (DECL_SELF_REFERENCE_P (t)) + /* The context and type of an injected-class-name are + the same, so we don't need to substitute both. */ + type = ctx; /* If CTX is unchanged, then T is in fact the specialization we want. That situation occurs when referencing a static data member within in its own @@ -14931,16 +14937,28 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) if (!spec) { + int args_depth = TMPL_ARGS_DEPTH (args); + int parms_depth = TMPL_ARGS_DEPTH (DECL_TI_ARGS (t)); tmpl = DECL_TI_TEMPLATE (t); gen_tmpl = most_general_template (tmpl); - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); - if (argvec != error_mark_node) - argvec = (coerce_innermost_template_parms - (DECL_TEMPLATE_PARMS (gen_tmpl), - argvec, t, complain, - /*all*/true, /*defarg*/true)); - if (argvec == error_mark_node) - RETURN (error_mark_node); + if (args_depth == parms_depth + && DECL_CLASS_SCOPE_P (tmpl) + && !DECL_MEMBER_TEMPLATE_P (tmpl)) + /* The DECL_TI_ARGS in this case are the generic template + arguments for the class template, so substitution is + just the identity mapping. */ + argvec = args; + else + { + argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); + if (args_depth >= parms_depth && argvec != error_mark_node) + argvec = (coerce_innermost_template_parms + (DECL_TEMPLATE_PARMS (gen_tmpl), + argvec, t, complain, + /*all*/true, /*defarg*/true)); + if (argvec == error_mark_node) + RETURN (error_mark_node); + } hash = hash_tmpl_and_args (gen_tmpl, argvec); spec = retrieve_specialization (gen_tmpl, argvec, hash); }
On 6/10/22 12:00, Patrick Palka wrote: > On Fri, 10 Jun 2022, Patrick Palka wrote: > >> On Thu, 9 Jun 2022, Patrick Palka wrote: >> >>> On Thu, 9 Jun 2022, Jason Merrill wrote: >>> >>>> On 6/8/22 14:21, Patrick Palka wrote: >>>>> When substituting a class template specialization, tsubst_aggr_type >>>>> substitutes the TYPE_CONTEXT before passing it to lookup_template_class. >>>>> This appears to be unnecessary, however, because the the initial value >>>>> of lookup_template_class's context parameter is unused outside of the >>>>> IDENTIFIER_NODE case, and l_t_c performs its own substitution of the >>>>> context, anyway. So this patch removes the redundant substitution in >>>>> tsubst_aggr_type. Doing so causes us to ICE on template/nested5.C >>>>> because during lookup_template_class for A<T>::C::D<S> with T=E and S=S, >>>>> we substitute and complete the context A<T>::C with T=E, which in turn >>>>> registers the desired dependent specialization of D for us and we end up >>>>> trying to register it again. This patch fixes this by checking the >>>>> specializations table again after completion of the context. >>>>> >>>>> This patch also implements a couple of other optimizations: >>>>> >>>>> * In lookup_template_class, if the context of the partially >>>>> instantiated template is already non-dependent, then we could >>>>> reuse that instead of substituting the context of the most >>>>> general template. >>>>> * When substituting the TYPE_DECL for an injected-class-name >>>>> in tsubst_decl, we can avoid substituting its TREE_TYPE and >>>>> DECL_TI_ARGS. >>>>> >>>>> Together these optimizations improve memory usage for the range-v3 >>>>> testcase test/view/split.cc by about 5%. The improvement is probably >>>>> more significant when dealing with deeply nested class templates. >>>>> >>>>> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for >>>>> trunk? >>>>> >>>>> gcc/cp/ChangeLog: >>>>> >>>>> * pt.cc (lookup_template_class): Remove dead stores to >>>>> context parameter. Don't substitute the context of the >>>>> most general template if that of the partially instantiated >>>>> template is non-dependent. Check the specializations table >>>>> again after completing the context of a nested dependent >>>>> specialization. >>>>> (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute >>>>> TYPE_CONTEXT or pass it to lookup_template_class. >>>>> (tsubst_decl) <case TYPE_DECL>: Avoid substituting the >>>>> TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. >>>>> --- >>>>> gcc/cp/pt.cc | 69 +++++++++++++++++++++++++++++++--------------------- >>>>> 1 file changed, 41 insertions(+), 28 deletions(-) >>>>> >>>>> diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc >>>>> index 59b94317e88..28023d60684 100644 >>>>> --- a/gcc/cp/pt.cc >>>>> +++ b/gcc/cp/pt.cc >>>>> @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree >>>>> in_decl, tree context, >>>>> if (context) >>>>> pop_decl_namespace (); >>>>> } >>>>> - if (templ) >>>>> - context = DECL_CONTEXT (templ); >>>>> } >>>>> else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE >>>>> (d1))) >>>>> { >>>>> @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree >>>>> in_decl, tree context, >>>>> { >>>>> templ = d1; >>>>> d1 = DECL_NAME (templ); >>>>> - context = DECL_CONTEXT (templ); >>>>> } >>>>> else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) >>>>> { >>>>> @@ -10059,8 +10056,25 @@ lookup_template_class (tree d1, tree arglist, tree >>>>> in_decl, tree context, >>>>> context = DECL_CONTEXT (gen_tmpl); >>>>> if (context && TYPE_P (context)) >>>>> { >>>>> - context = tsubst_aggr_type (context, arglist, complain, in_decl, >>>>> true); >>>>> - context = complete_type (context); >>>>> + if (!uses_template_parms (DECL_CONTEXT (templ))) >>>>> + /* If the context of the partially instantiated template is >>>>> + already non-dependent, then we might as well use it. */ >>>>> + context = DECL_CONTEXT (templ); >>>>> + else >>>>> + { >>>>> + context = tsubst_aggr_type (context, arglist, complain, in_decl, >>>>> true); >>>>> + context = complete_type (context); >>>>> + if (is_dependent_type && arg_depth > 1) >>>>> + { >>>>> + /* If this is a dependent nested specialization such as >>>>> + A<int>::B<T>, then completion of A<int> might have >>>>> + registered this specialization of B for us, so check >>>>> + the table again (33959). */ >>>>> + entry = type_specializations->find_with_hash (&elt, hash); >>>>> + if (entry) >>>>> + return entry->spec; >>>>> + } >>>>> + } >>>>> } >>>>> else >>>>> context = tsubst (context, arglist, complain, in_decl); >>>>> @@ -13711,25 +13725,12 @@ tsubst_aggr_type (tree t, >>>>> if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) >>>>> { >>>>> tree argvec; >>>>> - tree context; >>>>> tree r; >>>>> /* In "sizeof(X<I>)" we need to evaluate "I". */ >>>>> cp_evaluated ev; >>>>> - /* First, determine the context for the type we are looking >>>>> - up. */ >>>>> - context = TYPE_CONTEXT (t); >>>>> - if (context && TYPE_P (context)) >>>>> - { >>>>> - context = tsubst_aggr_type (context, args, complain, >>>>> - in_decl, /*entering_scope=*/1); >>>>> - /* If context is a nested class inside a class template, >>>>> - it may still need to be instantiated (c++/33959). */ >>>>> - context = complete_type (context); >>>>> - } >>>>> - >>>>> - /* Then, figure out what arguments are appropriate for the >>>>> + /* Figure out what arguments are appropriate for the >>>>> type we are trying to find. For example, given: >>>>> template <class T> struct S; >>>>> @@ -13744,7 +13745,7 @@ tsubst_aggr_type (tree t, >>>>> r = error_mark_node; >>>>> else >>>>> { >>>>> - r = lookup_template_class (t, argvec, in_decl, context, >>>>> + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, >>>>> entering_scope, complain); >>>>> r = cp_build_qualified_type (r, cp_type_quals (t), complain); >>>>> } >>>>> @@ -14880,6 +14881,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t >>>>> complain) >>>>> ctx = tsubst_aggr_type (ctx, args, >>>>> complain, >>>>> in_decl, /*entering_scope=*/1); >>>>> + if (DECL_SELF_REFERENCE_P (t)) >>>>> + /* The context and type of a injected-class-name are >>>>> + the same, so we don't need to substitute both. */ >>>>> + type = ctx; >>>>> /* If CTX is unchanged, then T is in fact the >>>>> specialization we want. That situation occurs when >>>>> referencing a static data member within in its own >>>>> @@ -14900,14 +14905,22 @@ tsubst_decl (tree t, tree args, tsubst_flags_t >>>>> complain) >>>>> { >>>>> tmpl = DECL_TI_TEMPLATE (t); >>>>> gen_tmpl = most_general_template (tmpl); >>>>> - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); >>>>> - if (argvec != error_mark_node) >>>>> - argvec = (coerce_innermost_template_parms >>>>> - (DECL_TEMPLATE_PARMS (gen_tmpl), >>>>> - argvec, t, complain, >>>>> - /*all*/true, /*defarg*/true)); >>>>> - if (argvec == error_mark_node) >>>>> - RETURN (error_mark_node); >>>>> + if (DECL_SELF_REFERENCE_P (t)) >>>>> + /* The DECL_TI_ARGS for the injected-class-name are the >>>>> + generic template arguments for the class template, so >>>>> + substitution/coercion is just the identity mapping. */ >>>>> + argvec = args; >>>> >>>> Would it make sense to extend this to any TEMPLATE_DECL for which >>>> DECL_PRIMARY_TEMPLATE is the class template? So, anything that gets here >>>> except an alias template. >>> >>> Ah nice, it does look like we could extend this to any TEMPLATE_DECL that >>> satisfies DECL_CLASS_SCOPE_P && !DECL_MEMBER_TEMPLATE_P (so that we also >>> exclude static data member templates), i.e. any templated non-template >>> member. Is that equivalent to what you had in mind? >>> >>> Based on some light testing, if we do this, it seems we also need to >>> handle 'args' having greater depth than 'DECL_TI_ARGS' here, something >>> which could happen during satisfaction. >> >> ... or we could just restrict the optimization to when the argument >> depths match, like so: >> >> -- >8 -- >> >> Subject: [PATCH] c++: optimize specialization of nested templated classes >> >> gcc/cp/ChangeLog: >> >> * pt.cc (lookup_template_class): Remove dead stores to >> context parameter. Don't substitute the context of the >> most general template if that of the partially instantiated >> template is non-dependent. Check the specializations table >> again after completing the context of a nested dependent >> specialization. >> (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute >> TYPE_CONTEXT or pass it to lookup_template_class. >> (tsubst_decl) <case TYPE_DECL>: Avoid substituting the >> TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. >> --- >> gcc/cp/pt.cc | 73 ++++++++++++++++++++++++++++++++-------------------- >> 1 file changed, 45 insertions(+), 28 deletions(-) >> >> diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc >> index 3154186ac20..ebd822373db 100644 >> --- a/gcc/cp/pt.cc >> +++ b/gcc/cp/pt.cc >> @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, >> if (context) >> pop_decl_namespace (); >> } >> - if (templ) >> - context = DECL_CONTEXT (templ); >> } >> else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE (d1))) >> { >> @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, >> { >> templ = d1; >> d1 = DECL_NAME (templ); >> - context = DECL_CONTEXT (templ); >> } >> else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) >> { >> @@ -10059,8 +10056,26 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, >> context = DECL_CONTEXT (gen_tmpl); >> if (context && TYPE_P (context)) >> { >> - context = tsubst_aggr_type (context, arglist, complain, in_decl, true); >> - context = complete_type (context); >> + if (!uses_template_parms (DECL_CONTEXT (templ))) >> + /* If the context of the partially instantiated template is >> + already non-dependent, then we might as well use it. */ >> + context = DECL_CONTEXT (templ); >> + else >> + { >> + context = tsubst_aggr_type (context, arglist, >> + complain, in_decl, true); >> + context = complete_type (context); >> + if (is_dependent_type && arg_depth > 1) >> + { >> + /* If this is a dependent nested specialization such as >> + A<int>::B<T>, then completion of A<int> might have >> + registered this specialization of B for us, so check >> + the table again (33959). */ >> + entry = type_specializations->find_with_hash (&elt, hash); >> + if (entry) >> + return entry->spec; >> + } >> + } >> } >> else >> context = tsubst (context, arglist, complain, in_decl); >> @@ -13739,25 +13754,12 @@ tsubst_aggr_type (tree t, >> if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) >> { >> tree argvec; >> - tree context; >> tree r; >> >> /* In "sizeof(X<I>)" we need to evaluate "I". */ >> cp_evaluated ev; >> >> - /* First, determine the context for the type we are looking >> - up. */ >> - context = TYPE_CONTEXT (t); >> - if (context && TYPE_P (context)) >> - { >> - context = tsubst_aggr_type (context, args, complain, >> - in_decl, /*entering_scope=*/1); >> - /* If context is a nested class inside a class template, >> - it may still need to be instantiated (c++/33959). */ >> - context = complete_type (context); >> - } >> - >> - /* Then, figure out what arguments are appropriate for the >> + /* Figure out what arguments are appropriate for the >> type we are trying to find. For example, given: >> >> template <class T> struct S; >> @@ -13772,7 +13774,7 @@ tsubst_aggr_type (tree t, >> r = error_mark_node; >> else >> { >> - r = lookup_template_class (t, argvec, in_decl, context, >> + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, >> entering_scope, complain); >> r = cp_build_qualified_type (r, cp_type_quals (t), complain); >> } >> @@ -14913,6 +14915,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) >> ctx = tsubst_aggr_type (ctx, args, >> complain, >> in_decl, /*entering_scope=*/1); >> + if (DECL_SELF_REFERENCE_P (t)) >> + /* The context and type of an injected-class-name are >> + the same, so we don't need to substitute both. */ >> + type = ctx; >> /* If CTX is unchanged, then T is in fact the >> specialization we want. That situation occurs when >> referencing a static data member within in its own >> @@ -14933,14 +14939,25 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) >> { >> tmpl = DECL_TI_TEMPLATE (t); >> gen_tmpl = most_general_template (tmpl); >> - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); >> - if (argvec != error_mark_node) >> - argvec = (coerce_innermost_template_parms >> - (DECL_TEMPLATE_PARMS (gen_tmpl), >> - argvec, t, complain, >> - /*all*/true, /*defarg*/true)); >> - if (argvec == error_mark_node) >> - RETURN (error_mark_node); >> + if (DECL_CLASS_SCOPE_P (tmpl) >> + && !DECL_MEMBER_TEMPLATE_P (tmpl) >> + && (TMPL_ARGS_DEPTH (args) >> + == TMPL_ARGS_DEPTH (DECL_TI_ARGS (t)))) >> + /* The DECL_TI_ARGS in this case are the generic template >> + arguments for the class template, so substitution/coercion >> + is just the identity mapping. */ >> + argvec = args; >> + else >> + { >> + argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); >> + if (argvec != error_mark_node) >> + argvec = (coerce_innermost_template_parms >> + (DECL_TEMPLATE_PARMS (gen_tmpl), >> + argvec, t, complain, >> + /*all*/true, /*defarg*/true)); >> + if (argvec == error_mark_node) >> + RETURN (error_mark_node); >> + } > > Seems we could be a bit cleverer here by trying to avoid the > coercion even if we have to do the substitution. > > When the argument depth is less than the parameter depth, then the > innermost substituted arguments are just lowered arguments from > DECL_TI_ARGS, for which coercion would be a no-op, so we can just > skip it. This optimization seems to trigger 90+% of the time and > compile time/memory by about 1%/0.5% in my experiments. (The call to > coerce_innermost_template_parms was added by r6-829-gb237c4cbd3da7a > FWIW.) > > -- >8 -- > > > Subject: [PATCH] c++: optimize specialization of nested templated classes > > gcc/cp/ChangeLog: > > * pt.cc (lookup_template_class): Remove dead stores to > context parameter. Don't substitute the context of the > most general template if that of the partially instantiated > template is non-dependent. Check the specializations table > again after completing the context of a nested dependent > specialization. > (tsubst_aggr_type) <case RECORD_TYPE>: Don't substitute > TYPE_CONTEXT or pass it to lookup_template_class. > (tsubst_decl) <case TYPE_DECL>: Avoid substituting the > TREE_TYPE and DECL_TI_ARGS when DECL_SELF_REFERENCE_P. > --- > gcc/cp/pt.cc | 74 ++++++++++++++++++++++++++++++++-------------------- > 1 file changed, 46 insertions(+), 28 deletions(-) > > diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc > index 3154186ac20..faa5bc377cb 100644 > --- a/gcc/cp/pt.cc > +++ b/gcc/cp/pt.cc > @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, > if (context) > pop_decl_namespace (); > } > - if (templ) > - context = DECL_CONTEXT (templ); > } > else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE (d1))) > { > @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, > { > templ = d1; > d1 = DECL_NAME (templ); > - context = DECL_CONTEXT (templ); > } > else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) > { > @@ -10059,8 +10056,26 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, > context = DECL_CONTEXT (gen_tmpl); > if (context && TYPE_P (context)) > { > - context = tsubst_aggr_type (context, arglist, complain, in_decl, true); > - context = complete_type (context); > + if (!uses_template_parms (DECL_CONTEXT (templ))) > + /* If the context of the partially instantiated template is > + already non-dependent, then we might as well use it. */ > + context = DECL_CONTEXT (templ); > + else > + { > + context = tsubst_aggr_type (context, arglist, > + complain, in_decl, true); > + context = complete_type (context); > + if (is_dependent_type && arg_depth > 1) > + { > + /* If this is a dependent nested specialization such as > + A<int>::B<T>, then completion of A<int> might have > + registered this specialization of B for us, so check > + the table again (33959). */ > + entry = type_specializations->find_with_hash (&elt, hash); > + if (entry) > + return entry->spec; > + } > + } > } > else > context = tsubst (context, arglist, complain, in_decl); > @@ -13739,25 +13754,12 @@ tsubst_aggr_type (tree t, > if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) > { > tree argvec; > - tree context; > tree r; > > /* In "sizeof(X<I>)" we need to evaluate "I". */ > cp_evaluated ev; > > - /* First, determine the context for the type we are looking > - up. */ > - context = TYPE_CONTEXT (t); > - if (context && TYPE_P (context)) > - { > - context = tsubst_aggr_type (context, args, complain, > - in_decl, /*entering_scope=*/1); > - /* If context is a nested class inside a class template, > - it may still need to be instantiated (c++/33959). */ > - context = complete_type (context); > - } > - > - /* Then, figure out what arguments are appropriate for the > + /* Figure out what arguments are appropriate for the > type we are trying to find. For example, given: > > template <class T> struct S; > @@ -13772,7 +13774,7 @@ tsubst_aggr_type (tree t, > r = error_mark_node; > else > { > - r = lookup_template_class (t, argvec, in_decl, context, > + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, > entering_scope, complain); > r = cp_build_qualified_type (r, cp_type_quals (t), complain); > } > @@ -14913,6 +14915,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) > ctx = tsubst_aggr_type (ctx, args, > complain, > in_decl, /*entering_scope=*/1); > + if (DECL_SELF_REFERENCE_P (t)) > + /* The context and type of an injected-class-name are > + the same, so we don't need to substitute both. */ > + type = ctx; > /* If CTX is unchanged, then T is in fact the > specialization we want. That situation occurs when > referencing a static data member within in its own > @@ -14931,16 +14937,28 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) > > if (!spec) > { > + int args_depth = TMPL_ARGS_DEPTH (args); > + int parms_depth = TMPL_ARGS_DEPTH (DECL_TI_ARGS (t)); > tmpl = DECL_TI_TEMPLATE (t); > gen_tmpl = most_general_template (tmpl); > - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); > - if (argvec != error_mark_node) > - argvec = (coerce_innermost_template_parms > - (DECL_TEMPLATE_PARMS (gen_tmpl), > - argvec, t, complain, > - /*all*/true, /*defarg*/true)); > - if (argvec == error_mark_node) > - RETURN (error_mark_node); > + if (args_depth == parms_depth > + && DECL_CLASS_SCOPE_P (tmpl) > + && !DECL_MEMBER_TEMPLATE_P (tmpl)) I think these two lines could also be !PRIMARY_TEMPLATE_P (gen_tmpl). OK either way. > + /* The DECL_TI_ARGS in this case are the generic template > + arguments for the class template, so substitution is > + just the identity mapping. */ > + argvec = args; > + else > + { > + argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); > + if (args_depth >= parms_depth && argvec != error_mark_node) Please add your note about the args < parms case as a comment. > + argvec = (coerce_innermost_template_parms > + (DECL_TEMPLATE_PARMS (gen_tmpl), > + argvec, t, complain, > + /*all*/true, /*defarg*/true)); > + if (argvec == error_mark_node) > + RETURN (error_mark_node); > + } > hash = hash_tmpl_and_args (gen_tmpl, argvec); > spec = retrieve_specialization (gen_tmpl, argvec, hash); > }
diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc index 59b94317e88..28023d60684 100644 --- a/gcc/cp/pt.cc +++ b/gcc/cp/pt.cc @@ -9840,8 +9840,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, if (context) pop_decl_namespace (); } - if (templ) - context = DECL_CONTEXT (templ); } else if (TREE_CODE (d1) == TYPE_DECL && MAYBE_CLASS_TYPE_P (TREE_TYPE (d1))) { @@ -9868,7 +9866,6 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, { templ = d1; d1 = DECL_NAME (templ); - context = DECL_CONTEXT (templ); } else if (DECL_TEMPLATE_TEMPLATE_PARM_P (d1)) { @@ -10059,8 +10056,25 @@ lookup_template_class (tree d1, tree arglist, tree in_decl, tree context, context = DECL_CONTEXT (gen_tmpl); if (context && TYPE_P (context)) { - context = tsubst_aggr_type (context, arglist, complain, in_decl, true); - context = complete_type (context); + if (!uses_template_parms (DECL_CONTEXT (templ))) + /* If the context of the partially instantiated template is + already non-dependent, then we might as well use it. */ + context = DECL_CONTEXT (templ); + else + { + context = tsubst_aggr_type (context, arglist, complain, in_decl, true); + context = complete_type (context); + if (is_dependent_type && arg_depth > 1) + { + /* If this is a dependent nested specialization such as + A<int>::B<T>, then completion of A<int> might have + registered this specialization of B for us, so check + the table again (33959). */ + entry = type_specializations->find_with_hash (&elt, hash); + if (entry) + return entry->spec; + } + } } else context = tsubst (context, arglist, complain, in_decl); @@ -13711,25 +13725,12 @@ tsubst_aggr_type (tree t, if (TYPE_TEMPLATE_INFO (t) && uses_template_parms (t)) { tree argvec; - tree context; tree r; /* In "sizeof(X<I>)" we need to evaluate "I". */ cp_evaluated ev; - /* First, determine the context for the type we are looking - up. */ - context = TYPE_CONTEXT (t); - if (context && TYPE_P (context)) - { - context = tsubst_aggr_type (context, args, complain, - in_decl, /*entering_scope=*/1); - /* If context is a nested class inside a class template, - it may still need to be instantiated (c++/33959). */ - context = complete_type (context); - } - - /* Then, figure out what arguments are appropriate for the + /* Figure out what arguments are appropriate for the type we are trying to find. For example, given: template <class T> struct S; @@ -13744,7 +13745,7 @@ tsubst_aggr_type (tree t, r = error_mark_node; else { - r = lookup_template_class (t, argvec, in_decl, context, + r = lookup_template_class (t, argvec, in_decl, NULL_TREE, entering_scope, complain); r = cp_build_qualified_type (r, cp_type_quals (t), complain); } @@ -14880,6 +14881,10 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) ctx = tsubst_aggr_type (ctx, args, complain, in_decl, /*entering_scope=*/1); + if (DECL_SELF_REFERENCE_P (t)) + /* The context and type of a injected-class-name are + the same, so we don't need to substitute both. */ + type = ctx; /* If CTX is unchanged, then T is in fact the specialization we want. That situation occurs when referencing a static data member within in its own @@ -14900,14 +14905,22 @@ tsubst_decl (tree t, tree args, tsubst_flags_t complain) { tmpl = DECL_TI_TEMPLATE (t); gen_tmpl = most_general_template (tmpl); - argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); - if (argvec != error_mark_node) - argvec = (coerce_innermost_template_parms - (DECL_TEMPLATE_PARMS (gen_tmpl), - argvec, t, complain, - /*all*/true, /*defarg*/true)); - if (argvec == error_mark_node) - RETURN (error_mark_node); + if (DECL_SELF_REFERENCE_P (t)) + /* The DECL_TI_ARGS for the injected-class-name are the + generic template arguments for the class template, so + substitution/coercion is just the identity mapping. */ + argvec = args; + else + { + argvec = tsubst (DECL_TI_ARGS (t), args, complain, in_decl); + if (argvec != error_mark_node) + argvec = (coerce_innermost_template_parms + (DECL_TEMPLATE_PARMS (gen_tmpl), + argvec, t, complain, + /*all*/true, /*defarg*/true)); + if (argvec == error_mark_node) + RETURN (error_mark_node); + } hash = hash_tmpl_and_args (gen_tmpl, argvec); spec = retrieve_specialization (gen_tmpl, argvec, hash); }