From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from nikam.ms.mff.cuni.cz (nikam.ms.mff.cuni.cz [195.113.20.16]) by sourceware.org (Postfix) with ESMTPS id 1C5E43858C66 for ; Fri, 14 Apr 2023 19:12:45 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 1C5E43858C66 Authentication-Results: sourceware.org; dmarc=fail (p=none dis=none) header.from=ucw.cz Authentication-Results: sourceware.org; spf=none smtp.mailfrom=kam.mff.cuni.cz Received: by nikam.ms.mff.cuni.cz (Postfix, from userid 16202) id 0589628A6F5; Fri, 14 Apr 2023 21:12:43 +0200 (CEST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=ucw.cz; s=gen1; t=1681499564; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: in-reply-to:in-reply-to:references:references; bh=zzXdMp4ZVUgbtk7iuz0wDDwWzsgChSZfbus51HuzFXQ=; b=RsMiRrpuadaDt06SmYvA8S+fItlZMIiQ7p9L5DltmEb7Q3UJsSdmAj3+a/Mbp0P2AfHMy1 Q/EelM/NrJLbhQV/rFObM8xelo1djaoDwO2RAGUYfFwGK0OxFmZbQTmIX4AnQe6ZW4O4H3 ai2ri4IGV3yqAit5oSz6XmjjkDxctJg= Date: Fri, 14 Apr 2023 21:12:43 +0200 From: Jan Hubicka To: Richard Biener Cc: gcc-patches@gcc.gnu.org Subject: Re: [PATCH] tree-optimization/109304 - properly handle instrumented aliases Message-ID: References: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: X-Spam-Status: No, score=-5.1 required=5.0 tests=BAYES_00,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,HEADER_FROM_DIFFERENT_DOMAINS,RCVD_IN_MSPIKE_H3,RCVD_IN_MSPIKE_WL,SPF_HELO_NONE,SPF_NONE,TXREP,T_SCC_BODY_TEXT_LINE autolearn=no autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: > On Tue, 4 Apr 2023, Jan Hubicka wrote: > > > > On Tue, 28 Mar 2023, Richard Biener wrote: > > > > > > > When adjusting calls to reflect instrumentation we failed to handle > > > > calls to aliases since they appear to have no body. Instead resort > > > > to symtab node availability. The patch also avoids touching > > > > internal function calls in a more obvious way (builtins might > > > > have a body available). > > > > > > > > profiledbootstrap & regtest running on x86_64-unknown-linux-gnu. > > > > > > > > Honza - does this look OK? > > > > PR tree-optimization/109304 > > > > * tree-profile.cc (tree_profiling): Use symtab node > > > > availability to decide whether to skip adjusting calls. > > > > Do not adjust calls to internal functions. > > > > @@ -842,12 +842,15 @@ tree_profiling (void) > > > > for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi)) > > > > { > > > > gcall *call = dyn_cast (gsi_stmt (gsi)); > > > > - if (!call) > > > > + if (!call || gimple_call_internal_p (call)) > > > > continue; > > > > > > > > /* We do not clear pure/const on decls without body. */ > > > > tree fndecl = gimple_call_fndecl (call); > > > > - if (fndecl && !gimple_has_body_p (fndecl)) > > > > + cgraph_node *callee; > > > > + if (fndecl > > > > + && (callee = cgraph_node::get (fndecl)) > > > > + && callee->get_availability (node) == AVAIL_NOT_AVAILABLE) > > > > As discussed earlier, the testcase I posted can be adjusted to put the > > const declared wrapper into another translation unit, so I think we will > > need to drop the visibility check completely. But as discussed, it is > > wrong code issue, but not a regression, so we may go with the > > availability check as you suggest. So the patch is OK. > > > > > > I wonder if we do not want to drop it everywhere (as we plan for next > > stage1 anyway). I think similar ICE as in the PR can be produced with > > LTO. In normal situation declaration merging will do the right thing: > > If you have unit A calling const foo externally, it won't get processed > > by the code above. However unit B declaring foo will get it downgraded > > to non-const. > > > > Now at WPA time we will read both A and B and in declaration merging B's > > definition will prevail. This won't happen if lto_symtab_merge_p > > returns false which can probably be triggered by adding warning/error > > attribute to B's declaration but not to A's. > > > > It is however really side case and I am worried about dropping > > pure/const from builtin declarations... > > Yeah, that's what I'm worried about as well. I guess we'd need to > do the demotion to non-const/pure at WPA time and have a flag > in the cgraph node indicating instrument_add_{read,write}? But > then how should we deal with C++ comdats instrumented in one TU > but not in another? Does this mean we should do instrumentation > at IPA time instead of as small IPA pass before IPA? I do not think LTO is of any help here. You can allways call non-LTO const function from outer-world and that function can will end up calling back to instrumented const function in your unit which effectively makes the extenral const function non-const. > > That said, when there's a definition of say strlen in a TU and > that's instrumented we do want to drop pure from calls but if > not then we shouldn't worry. > > Without LTO we'd still run into coverage issues but at least > with LTO we shouldn't ICE? I am not sure I see your point here... We could avoid demoting builtins to avoid ICEs and have coverage mismathces, but how LTO makes difference? Honza > > Richard.