From patchwork Tue Jan 27 12:29:37 2015 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ilya Enkovich X-Patchwork-Id: 433450 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 03A7214012A for ; Tue, 27 Jan 2015 23:30:14 +1100 (AEDT) DomainKey-Signature: a=rsa-sha1; c=nofws; d=gcc.gnu.org; h=list-id :list-unsubscribe:list-archive:list-post:list-help:sender:date :from:to:cc:subject:message-id:references:mime-version :content-type:in-reply-to; q=dns; s=default; b=w5NHtIai0Int5n8Hx rFZZm2VErexpCMkOPJ4+g/yR+JM55MAsQLqiRn9E4RmLtLPd/sIArmKb+G6uSyN2 w93SvkY7ETUiV81GiPtkOvPppFmLG39N6Y8lpgtqvjIgAiKpZc0JmQkwOzng5JLJ opQsJhoebAHqZK83sjL2g0zMPQ= DKIM-Signature: v=1; a=rsa-sha1; c=relaxed; d=gcc.gnu.org; h=list-id :list-unsubscribe:list-archive:list-post:list-help:sender:date :from:to:cc:subject:message-id:references:mime-version :content-type:in-reply-to; s=default; bh=/rkfYxhbwZ6zNKd2gZZIBGM lCOY=; b=P6qSBNwz50MOXK7cOE4phjKILE11umwwfcF6LYQdSQ4pusUXxqufdlf yHDOJ+16ot6wETkmRWZL1NPw/8SvNpDkiS78nX+m+1KKc8lMbLk0UzldHds3OWcw kDY0WSc3mj7m9dOBX0juUC+o9MUSWyKKflhXVtdhgzeSh3lPCn6g= Received: (qmail 26670 invoked by alias); 27 Jan 2015 12:30:05 -0000 Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org Delivered-To: mailing list gcc-patches@gcc.gnu.org Received: (qmail 26590 invoked by uid 89); 27 Jan 2015 12:29:52 -0000 Authentication-Results: sourceware.org; auth=none X-Spam-SWARE-Status: No, score=-1.0 required=5.0 tests=AWL, FREEMAIL_FROM, RCVD_IN_DNSWL_LOW, SPF_PASS autolearn=ham version=3.3.2 X-HELO: mail-qa0-f47.google.com Received: from mail-qa0-f47.google.com (HELO mail-qa0-f47.google.com) (209.85.216.47) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with (AES128-SHA encrypted) ESMTPS; Tue, 27 Jan 2015 12:29:50 +0000 Received: by mail-qa0-f47.google.com with SMTP id n8so11115805qaq.6 for ; Tue, 27 Jan 2015 04:29:48 -0800 (PST) X-Received: by 10.229.83.196 with SMTP id g4mr1580725qcl.18.1422361788366; Tue, 27 Jan 2015 04:29:48 -0800 (PST) Received: from msticlxl57.ims.intel.com (fmdmzpr03-ext.fm.intel.com. [192.55.54.38]) by mx.google.com with ESMTPSA id i46sm927151qgd.35.2015.01.27.04.29.45 (version=TLSv1 cipher=ECDHE-RSA-AES128-SHA bits=128/128); Tue, 27 Jan 2015 04:29:47 -0800 (PST) Date: Tue, 27 Jan 2015 15:29:37 +0300 From: Ilya Enkovich To: Richard Biener Cc: GCC Patches Subject: Re: [PATCH, PR tree-optimization/64277] Improve loop iterations count estimation Message-ID: <20150127122937.GD47190@msticlxl57.ims.intel.com> References: <20150127094046.GB47190@msticlxl57.ims.intel.com> <20150127104729.GC47190@msticlxl57.ims.intel.com> MIME-Version: 1.0 Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) X-IsSubscribed: yes On 27 Jan 12:29, Richard Biener wrote: > On Tue, Jan 27, 2015 at 11:47 AM, Ilya Enkovich wrote: > > On 27 Jan 12:40, Ilya Enkovich wrote: > >> Hi, > >> > >> This patch was supposed to fix PR tree-optimization/64277. Tracker is now fixed by warnings disabling but I think patch is still useful to avoid dead code generated by complete unroll. > >> > >> Bootstrapped and tested on x86_64-unknown-linux-gnu. > >> > >> Thanks, > >> Ilya > >> -- > >> gcc/ > >> > >> 2015-01-27 Ilya Enkovich > >> > >> * tree-ssa-loop-niter.c (record_nonwrapping_iv): Use base > >> range info when possible to refine estimation. > >> > >> gcc/testsuite/ > >> > >> 2015-01-27 Ilya Enkovich > >> > >> * gcc.dg/pr64277.c: New. > >> > >> > > > > Here is a new version fixed according to comments in the tracker. I also fixed a test to scan cunroll dumps. Does it look OK? > > Minor comments below. > > > What are possible branches for this patch? > > You can probably create a testcase that shows code-size regressions > against a version that didn't peel completely (GCC 4.7). Thus I'd say > it would apply to 4.9 as well (4.8 doesn't have range information). > > > Thanks, > > Ilya > > -- > > diff --git a/gcc/testsuite/gcc.dg/pr64277.c b/gcc/testsuite/gcc.dg/pr64277.c > > new file mode 100644 > > index 0000000..c6ef331 > > --- /dev/null > > +++ b/gcc/testsuite/gcc.dg/pr64277.c > > @@ -0,0 +1,23 @@ > > +/* PR tree-optimization/64277 */ > > +/* { dg-do compile } */ > > +/* { dg-options "-O3 -Wall -Werror -fdump-tree-cunroll-details" } */ > > +/* { dg-final { scan-tree-dump "loop with 5 iterations completely unrolled" "cunroll" } } */ > > +/* { dg-final { scan-tree-dump "loop with 6 iterations completely unrolled" "cunroll" } } */ > > +/* { dg-final { cleanup-tree-dump "cunroll" } } */ > > + > > +int f1[10]; > > +void test1 (short a[], short m, unsigned short l) > > +{ > > + int i = l; > > + for (i = i + 5; i < m; i++) > > + f1[i] = a[i]++; > > +} > > + > > +void test2 (short a[], short m, short l) > > +{ > > + int i; > > + if (m > 5) > > + m = 5; > > + for (i = m; i > l; i--) > > + f1[i] = a[i]++; > > +} > > diff --git a/gcc/tree-ssa-loop-niter.c b/gcc/tree-ssa-loop-niter.c > > index 919f5c0..1cd297d 100644 > > --- a/gcc/tree-ssa-loop-niter.c > > +++ b/gcc/tree-ssa-loop-niter.c > > @@ -2754,6 +2754,7 @@ record_nonwrapping_iv (struct loop *loop, tree base, tree step, gimple stmt, > > { > > tree niter_bound, extreme, delta; > > tree type = TREE_TYPE (base), unsigned_type; > > + tree orig_base = base; > > > > if (TREE_CODE (step) != INTEGER_CST || integer_zerop (step)) > > return; > > @@ -2777,16 +2778,30 @@ record_nonwrapping_iv (struct loop *loop, tree base, tree step, gimple stmt, > > > > if (tree_int_cst_sign_bit (step)) > > { > > + wide_int min, max; > > extreme = fold_convert (unsigned_type, low); > > - if (TREE_CODE (base) != INTEGER_CST) > > + if (TREE_CODE (orig_base) == SSA_NAME > > + && TREE_CODE (high) == INTEGER_CST > > + && INTEGRAL_TYPE_P (TREE_TYPE (orig_base)) > > + && get_range_info (orig_base, &min, &max) == VR_RANGE > > + && wi::gts_p (wide_int (high), max)) > > For me a simple wi::gts_p (high, max) worked fine. > > > + base = wide_int_to_tree (unsigned_type, max); > > + else if (TREE_CODE (base) != INTEGER_CST) > > base = fold_convert (unsigned_type, high); > > delta = fold_build2 (MINUS_EXPR, unsigned_type, base, extreme); > > step = fold_build1 (NEGATE_EXPR, unsigned_type, step); > > } > > else > > { > > + wide_int min, max; > > extreme = fold_convert (unsigned_type, high); > > - if (TREE_CODE (base) != INTEGER_CST) > > + if (TREE_CODE (orig_base) == SSA_NAME > > + && TREE_CODE (low) == INTEGER_CST > > + && INTEGRAL_TYPE_P (TREE_TYPE (orig_base)) > > + && get_range_info (orig_base, &min, &max) == VR_RANGE > > + && wi::gts_p (min, wide_int (low))) > > Likewise. > > Ok for trunk with that changes. For the 4.9 branch you need to adjust > the patch to not use wide-ints. I'd leave it on trunk for a while and > eventually open a bugreport for the size regression to keep track of it. > > Thanks, > Richard. > Thanks a lot for review! Here is a final version for GCC 5.0. Will prepare 4.9 version later. Thanks, Ilya diff --git a/gcc/testsuite/gcc.dg/pr64277.c b/gcc/testsuite/gcc.dg/pr64277.c new file mode 100644 index 0000000..c6ef331 --- /dev/null +++ b/gcc/testsuite/gcc.dg/pr64277.c @@ -0,0 +1,23 @@ +/* PR tree-optimization/64277 */ +/* { dg-do compile } */ +/* { dg-options "-O3 -Wall -Werror -fdump-tree-cunroll-details" } */ +/* { dg-final { scan-tree-dump "loop with 5 iterations completely unrolled" "cunroll" } } */ +/* { dg-final { scan-tree-dump "loop with 6 iterations completely unrolled" "cunroll" } } */ +/* { dg-final { cleanup-tree-dump "cunroll" } } */ + +int f1[10]; +void test1 (short a[], short m, unsigned short l) +{ + int i = l; + for (i = i + 5; i < m; i++) + f1[i] = a[i]++; +} + +void test2 (short a[], short m, short l) +{ + int i; + if (m > 5) + m = 5; + for (i = m; i > l; i--) + f1[i] = a[i]++; +} diff --git a/gcc/tree-ssa-loop-niter.c b/gcc/tree-ssa-loop-niter.c index 919f5c0..a677958 100644 --- a/gcc/tree-ssa-loop-niter.c +++ b/gcc/tree-ssa-loop-niter.c @@ -2754,6 +2754,7 @@ record_nonwrapping_iv (struct loop *loop, tree base, tree step, gimple stmt, { tree niter_bound, extreme, delta; tree type = TREE_TYPE (base), unsigned_type; + tree orig_base = base; if (TREE_CODE (step) != INTEGER_CST || integer_zerop (step)) return; @@ -2777,16 +2778,30 @@ record_nonwrapping_iv (struct loop *loop, tree base, tree step, gimple stmt, if (tree_int_cst_sign_bit (step)) { + wide_int min, max; extreme = fold_convert (unsigned_type, low); - if (TREE_CODE (base) != INTEGER_CST) + if (TREE_CODE (orig_base) == SSA_NAME + && TREE_CODE (high) == INTEGER_CST + && INTEGRAL_TYPE_P (TREE_TYPE (orig_base)) + && get_range_info (orig_base, &min, &max) == VR_RANGE + && wi::gts_p (high, max)) + base = wide_int_to_tree (unsigned_type, max); + else if (TREE_CODE (base) != INTEGER_CST) base = fold_convert (unsigned_type, high); delta = fold_build2 (MINUS_EXPR, unsigned_type, base, extreme); step = fold_build1 (NEGATE_EXPR, unsigned_type, step); } else { + wide_int min, max; extreme = fold_convert (unsigned_type, high); - if (TREE_CODE (base) != INTEGER_CST) + if (TREE_CODE (orig_base) == SSA_NAME + && TREE_CODE (low) == INTEGER_CST + && INTEGRAL_TYPE_P (TREE_TYPE (orig_base)) + && get_range_info (orig_base, &min, &max) == VR_RANGE + && wi::gts_p (min, low)) + base = wide_int_to_tree (unsigned_type, min); + else if (TREE_CODE (base) != INTEGER_CST) base = fold_convert (unsigned_type, low); delta = fold_build2 (MINUS_EXPR, unsigned_type, extreme, base); }