From patchwork Mon Mar 16 16:21:02 2015 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jakub Jelinek X-Patchwork-Id: 450649 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 56BDE140077 for ; Tue, 17 Mar 2015 03:21:21 +1100 (AEDT) Authentication-Results: ozlabs.org; dkim=pass reason="1024-bit key; unprotected key" header.d=gcc.gnu.org header.i=@gcc.gnu.org header.b=OV4GXs7X; dkim-adsp=none (unprotected policy); dkim-atps=neutral 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:reply-to:mime-version :content-type; q=dns; s=default; b=ao1NUNy8gQhWN1cGfiGLTRtb7GpuO 3r5BSgTckRE05u4qAYx/uk+FfWTUKQudhptrNcd2CQmi1qQrZgP5OP+54CbaGNzO xv5DsLXvEjFBXIbcbObE+R3PHtIGgZ3gnGGpWtbQPUAoUxeQjXOMvuHt/e8lyIjw WzPx1x/nhftpKg= 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:reply-to:mime-version :content-type; s=default; bh=G/3NxzlP4pfx7wx3t/nnGe59Grk=; b=OV4 GXs7X4ZGWHTZb/Je7X6u4ez/pQiMYBtYjtmYRzqgMxacdv2jTzNpDZgSGckiehNy rch+Tx7Tm7rlX4bu97QdxOFignTj1LI9cmZv7OT66OF8I7bHPJbogTvXT+D2+gA7 5wGUVF3bLBVUTCp0Q8/fQBWGQnQ2FItaYZq15BdU= Received: (qmail 48832 invoked by alias); 16 Mar 2015 16:21:13 -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 48819 invoked by uid 89); 16 Mar 2015 16:21:11 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-2.0 required=5.0 tests=AWL, BAYES_00, SPF_HELO_PASS, T_RP_MATCHES_RCVD autolearn=ham version=3.3.2 X-HELO: mx1.redhat.com Received: from mx1.redhat.com (HELO mx1.redhat.com) (209.132.183.28) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with (AES256-GCM-SHA384 encrypted) ESMTPS; Mon, 16 Mar 2015 16:21:10 +0000 Received: from int-mx13.intmail.prod.int.phx2.redhat.com (int-mx13.intmail.prod.int.phx2.redhat.com [10.5.11.26]) by mx1.redhat.com (8.14.4/8.14.4) with ESMTP id t2GGL8Y7004499 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-GCM-SHA384 bits=256 verify=FAIL); Mon, 16 Mar 2015 12:21:08 -0400 Received: from tucnak.zalov.cz (ovpn-116-63.ams2.redhat.com [10.36.116.63]) by int-mx13.intmail.prod.int.phx2.redhat.com (8.14.4/8.14.4) with ESMTP id t2GGL6j6002296 (version=TLSv1/SSLv3 cipher=AES128-GCM-SHA256 bits=128 verify=NO); Mon, 16 Mar 2015 12:21:07 -0400 Received: from tucnak.zalov.cz (localhost [127.0.0.1]) by tucnak.zalov.cz (8.14.9/8.14.9) with ESMTP id t2GGL4Th002184; Mon, 16 Mar 2015 17:21:04 +0100 Received: (from jakub@localhost) by tucnak.zalov.cz (8.14.9/8.14.9/Submit) id t2GGL2NS002183; Mon, 16 Mar 2015 17:21:02 +0100 Date: Mon, 16 Mar 2015 17:21:02 +0100 From: Jakub Jelinek To: Richard Biener Cc: gcc-patches@gcc.gnu.org Subject: [PATCH] Handle vector COND_EXPRs in vector genericization (PR tree-optimization/65427) Message-ID: <20150316162102.GR1746@tucnak.redhat.com> Reply-To: Jakub Jelinek MIME-Version: 1.0 Content-Disposition: inline User-Agent: Mutt/1.5.23 (2014-03-12) X-IsSubscribed: yes Hi! On the following testcase, gimple LIM creates a vector COND_EXPR (scalar condition, vector lhs, rhs2 and rhs3), but if we don't have corresponding vector mode for it, we ICE trying to expand the BLKmode COND_EXPR, as it is unprepared for that. This patch lowers those (parallel or piecewise). Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? 2015-03-16 Jakub Jelinek PR tree-optimization/65427 * tree-vect-generic.c (do_cond, expand_vector_scalar_condition): New functions. (expand_vector_operations_1): Handle BLKmode vector COND_EXPR. * gcc.c-torture/execute/pr65427.c: New test. Jakub --- gcc/tree-vect-generic.c.jj 2015-01-15 20:25:40.000000000 +0100 +++ gcc/tree-vect-generic.c 2015-03-16 14:25:37.391932269 +0100 @@ -1417,6 +1417,57 @@ count_type_subparts (tree type) return VECTOR_TYPE_P (type) ? TYPE_VECTOR_SUBPARTS (type) : 1; } +static tree +do_cond (gimple_stmt_iterator *gsi, tree inner_type, tree a, tree b, + tree bitpos, tree bitsize, enum tree_code code) +{ + if (TREE_CODE (TREE_TYPE (a)) == VECTOR_TYPE) + a = tree_vec_extract (gsi, inner_type, a, bitsize, bitpos); + if (TREE_CODE (TREE_TYPE (b)) == VECTOR_TYPE) + b = tree_vec_extract (gsi, inner_type, b, bitsize, bitpos); + tree cond = gimple_assign_rhs1 (gsi_stmt (*gsi)); + return gimplify_build3 (gsi, code, inner_type, cond, a, b); +} + +/* Expand a vector COND_EXPR to scalars, piecewise. */ +static void +expand_vector_scalar_condition (gimple_stmt_iterator *gsi) +{ + gassign *stmt = as_a (gsi_stmt (*gsi)); + tree type = gimple_expr_type (stmt); + tree compute_type = get_compute_type (COND_EXPR, mov_optab, type); + machine_mode compute_mode = TYPE_MODE (compute_type); + gcc_assert (compute_mode != BLKmode); + tree lhs = gimple_assign_lhs (stmt); + tree rhs2 = gimple_assign_rhs2 (stmt); + tree rhs3 = gimple_assign_rhs3 (stmt); + tree new_rhs; + + /* If the compute mode is not a vector mode (hence we are not decomposing + a BLKmode vector to smaller, hardware-supported vectors), we may want + to expand the operations in parallel. */ + if (GET_MODE_CLASS (compute_mode) != MODE_VECTOR_INT + && GET_MODE_CLASS (compute_mode) != MODE_VECTOR_FLOAT + && GET_MODE_CLASS (compute_mode) != MODE_VECTOR_FRACT + && GET_MODE_CLASS (compute_mode) != MODE_VECTOR_UFRACT + && GET_MODE_CLASS (compute_mode) != MODE_VECTOR_ACCUM + && GET_MODE_CLASS (compute_mode) != MODE_VECTOR_UACCUM) + new_rhs = expand_vector_parallel (gsi, do_cond, type, rhs2, rhs3, + COND_EXPR); + else + new_rhs = expand_vector_piecewise (gsi, do_cond, type, compute_type, + rhs2, rhs3, COND_EXPR); + if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (new_rhs))) + new_rhs = gimplify_build1 (gsi, VIEW_CONVERT_EXPR, TREE_TYPE (lhs), + new_rhs); + + /* NOTE: We should avoid using gimple_assign_set_rhs_from_tree. One + way to do it is change expand_vector_operation and its callees to + return a tree_code, RHS1 and RHS2 instead of a tree. */ + gimple_assign_set_rhs_from_tree (gsi, new_rhs); + update_stmt (gsi_stmt (*gsi)); +} + /* Process one statement. If we identify a vector operation, expand it. */ static void @@ -1449,6 +1500,14 @@ expand_vector_operations_1 (gimple_stmt_ return; } + if (code == COND_EXPR + && TREE_CODE (TREE_TYPE (gimple_assign_lhs (stmt))) == VECTOR_TYPE + && TYPE_MODE (TREE_TYPE (gimple_assign_lhs (stmt))) == BLKmode) + { + expand_vector_scalar_condition (gsi); + return; + } + if (code == CONSTRUCTOR && TREE_CODE (lhs) == SSA_NAME && VECTOR_MODE_P (TYPE_MODE (TREE_TYPE (lhs))) --- gcc/testsuite/gcc.c-torture/execute/pr65427.c.jj 2015-03-16 14:36:29.489254701 +0100 +++ gcc/testsuite/gcc.c-torture/execute/pr65427.c 2015-03-16 14:40:58.789851433 +0100 @@ -0,0 +1,34 @@ +/* PR tree-optimization/65427 */ + +typedef int V __attribute__ ((vector_size (8 * sizeof (int)))); +V a, b, c, d, e, f; + +__attribute__((noinline, noclone)) void +foo (int x, int y) +{ + do + { + if (x) + d = a ^ c; + else + d = a ^ b; + } + while (y); +} + +int +main () +{ + a = (V) { 1, 2, 3, 4, 5, 6, 7, 8 }; + b = (V) { 0x40, 0x80, 0x40, 0x80, 0x40, 0x80, 0x40, 0x80 }; + e = (V) { 0x41, 0x82, 0x43, 0x84, 0x45, 0x86, 0x47, 0x88 }; + foo (0, 0); + if (__builtin_memcmp (&d, &e, sizeof (V)) != 0) + __builtin_abort (); + c = (V) { 0x80, 0x40, 0x80, 0x40, 0x80, 0x40, 0x80, 0x40 }; + f = (V) { 0x81, 0x42, 0x83, 0x44, 0x85, 0x46, 0x87, 0x48 }; + foo (1, 0); + if (__builtin_memcmp (&d, &f, sizeof (V)) != 0) + __builtin_abort (); + return 0; +}