From patchwork Fri Apr 1 06:43:43 2011 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Ira Rosen X-Patchwork-Id: 872 Return-Path: Delivered-To: unknown Received: from imap.gmail.com (74.125.159.109) by localhost6.localdomain6 with IMAP4-SSL; 08 Jun 2011 14:46:51 -0000 Delivered-To: patches@linaro.org Received: by 10.42.161.68 with SMTP id s4cs189601icx; Thu, 31 Mar 2011 23:43:45 -0700 (PDT) Received: by 10.150.93.4 with SMTP id q4mr3926886ybb.261.1301640224939; Thu, 31 Mar 2011 23:43:44 -0700 (PDT) Received: from mail-gy0-f178.google.com (mail-gy0-f178.google.com [209.85.160.178]) by mx.google.com with ESMTPS id u16si4618528yba.57.2011.03.31.23.43.43 (version=TLSv1/SSLv3 cipher=OTHER); Thu, 31 Mar 2011 23:43:43 -0700 (PDT) Received-SPF: neutral (google.com: 209.85.160.178 is neither permitted nor denied by best guess record for domain of ira.rosen@linaro.org) client-ip=209.85.160.178; Authentication-Results: mx.google.com; spf=neutral (google.com: 209.85.160.178 is neither permitted nor denied by best guess record for domain of ira.rosen@linaro.org) smtp.mail=ira.rosen@linaro.org Received: by gyd12 with SMTP id 12so1494180gyd.37 for ; Thu, 31 Mar 2011 23:43:43 -0700 (PDT) MIME-Version: 1.0 Received: by 10.150.135.15 with SMTP id i15mr3458431ybd.158.1301640223324; Thu, 31 Mar 2011 23:43:43 -0700 (PDT) Received: by 10.150.92.11 with HTTP; Thu, 31 Mar 2011 23:43:43 -0700 (PDT) In-Reply-To: References: Date: Fri, 1 Apr 2011 08:43:43 +0200 Message-ID: Subject: Re: [RFC][patch] If-conversion of COMPONENT_REFs From: Ira Rosen To: Richard Guenther Cc: gcc-patches@gcc.gnu.org, Patch Tracking On 30 March 2011 15:09, Ira Rosen wrote: > > I don't see any :) I'll test your version. > Bootstrapped on powerpc64-suse-linux and tested on x86_64-suse-linux the attached patch. OK to apply? Thanks, Ira ChangeLog:         * tree-if-conv.c (memrefs_read_or_written_unconditionally): Strip all non-variable offsets and compare the remaining bases of the two accesses instead of looking for exact same data-ref. testsuite/ChangeLog:         * gcc.dg/vect/if-cvt-stores-vect-ifcvt-18.c: New test.         * gcc.dg/vect/vect.exp: Run if-cvt-stores-vect* tests with         -ftree-loop-if-convert-stores. Index: testsuite/gcc.dg/vect/if-cvt-stores-vect-ifcvt-18.c =================================================================== --- testsuite/gcc.dg/vect/if-cvt-stores-vect-ifcvt-18.c (revision 0) +++ testsuite/gcc.dg/vect/if-cvt-stores-vect-ifcvt-18.c (revision 0) @@ -0,0 +1,69 @@ +/* { dg-require-effective-target vect_int } */ + +#include +#include "tree-vect.h" + +#define N 50 + +typedef struct { + short a; + short b; +} data; + +data in1[N], in2[N], out[N]; +short result[N*2] = {10,-7,11,-6,12,-5,13,-4,14,-3,15,-2,16,-1,17,0,18,1,19,2,20,3,21,4,22,5,23,6,24,7,25,8,26,9,27,10,28,11,29,12,30,13,31,14,32,15,33,16,34,17,35,18,36,19,37,20,38,21,39,22,40,23,41,24,42,25,43,26,44,27,45,28,46,29,47,30,48,31,49,32,50,33,51,34,52,35,53,36,54,37,55,38,56,39,57,40,58,41,59,42}; +short out1[N], out2[N]; + +__attribute__ ((noinline)) void +foo () +{ + int i; + short c, d; + + for (i = 0; i < N; i++) + { + c = in1[i].b; + d = in2[i].b; + + if (c >= d) + { + out[i].b = in1[i].a; + out[i].a = d + 5; + } + else + { + out[i].b = d - 12; + out[i].a = in2[i].a + d; + } + } +} + +int +main (void) +{ + int i; + + check_vect (); + + for (i = 0; i < N; i++) + { + in1[i].a = i; + in1[i].b = i + 2; + in2[i].a = 5; + in2[i].b = i + 5; + __asm__ volatile (""); + } + + foo (); + + for (i = 0; i < N; i++) + { + if (out[i].a != result[2*i] || out[i].b != result[2*i+1]) + abort (); + } + + return 0; +} + +/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail { vect_no_align || {! vect_strided } } } } } */ +/* { dg-final { cleanup-tree-dump "vect" } } */ Index: testsuite/gcc.dg/vect/vect.exp =================================================================== --- testsuite/gcc.dg/vect/vect.exp (revision 171716) +++ testsuite/gcc.dg/vect/vect.exp (working copy) @@ -210,6 +210,12 @@ lappend DEFAULT_VECTCFLAGS "--param" "gg dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/ggc-*.\[cS\]]] \ "" $DEFAULT_VECTCFLAGS +# -ftree-loop-if-convert-stores +set DEFAULT_VECTCFLAGS $SAVED_DEFAULT_VECTCFLAGS +lappend DEFAULT_VECTCFLAGS "-ftree-loop-if-convert-stores" +dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/if-cvt-stores-vect-*.\[cS\]]] \ + "" $DEFAULT_VECTCFLAGS + # With -O3. # Don't allow IPA cloning, because it throws our counts out of whack. set DEFAULT_VECTCFLAGS $SAVED_DEFAULT_VECTCFLAGS Index: tree-if-conv.c =================================================================== --- tree-if-conv.c (revision 171716) +++ tree-if-conv.c (working copy) @@ -464,8 +464,8 @@ struct ifc_dr { /* Returns true when the memory references of STMT are read or written unconditionally. In other words, this function returns true when for every data reference A in STMT there exist other accesses to - the same data reference with predicates that add up (OR-up) to the - true predicate: this ensures that the data reference A is touched + a data reference with the same base with predicates that add up (OR-up) to + the true predicate: this ensures that the data reference A is touched (read or written) on every iteration of the if-converted loop. */ static bool @@ -489,21 +489,38 @@ memrefs_read_or_written_unconditionally continue; for (j = 0; VEC_iterate (data_reference_p, drs, j, b); j++) - if (DR_STMT (b) != stmt - && same_data_refs (a, b)) - { - tree cb = bb_predicate (gimple_bb (DR_STMT (b))); + { + tree ref_base_a = DR_REF (a); + tree ref_base_b = DR_REF (b); + + if (DR_STMT (b) == stmt) + continue; + + while (TREE_CODE (ref_base_a) == COMPONENT_REF + || TREE_CODE (ref_base_a) == IMAGPART_EXPR + || TREE_CODE (ref_base_a) == REALPART_EXPR) + ref_base_a = TREE_OPERAND (ref_base_a, 0); + + while (TREE_CODE (ref_base_b) == COMPONENT_REF + || TREE_CODE (ref_base_b) == IMAGPART_EXPR + || TREE_CODE (ref_base_b) == REALPART_EXPR) + ref_base_b = TREE_OPERAND (ref_base_b, 0); - if (DR_RW_UNCONDITIONALLY (b) == 1 - || is_true_predicate (cb) - || is_true_predicate (ca = fold_or_predicates (EXPR_LOCATION (cb), - ca, cb))) - { - DR_RW_UNCONDITIONALLY (a) = 1; - DR_RW_UNCONDITIONALLY (b) = 1; - found = true; - break; - } + if (!operand_equal_p (ref_base_a, ref_base_b, 0)) + { + tree cb = bb_predicate (gimple_bb (DR_STMT (b))); + + if (DR_RW_UNCONDITIONALLY (b) == 1 + || is_true_predicate (cb) + || is_true_predicate (ca + = fold_or_predicates (EXPR_LOCATION (cb), ca, cb))) + { + DR_RW_UNCONDITIONALLY (a) = 1; + DR_RW_UNCONDITIONALLY (b) = 1; + found = true; + break; + } + } } if (!found)