From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 17762 invoked by alias); 26 Sep 2011 14:56:24 -0000 Received: (qmail 17750 invoked by uid 22791); 26 Sep 2011 14:56:21 -0000 X-SWARE-Spam-Status: No, hits=-6.8 required=5.0 tests=AWL,BAYES_00,RCVD_IN_DNSWL_HI,RP_MATCHES_RCVD,SPF_HELO_PASS X-Spam-Check-By: sourceware.org Received: from mx1.redhat.com (HELO mx1.redhat.com) (209.132.183.28) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Mon, 26 Sep 2011 14:56:06 +0000 Received: from int-mx02.intmail.prod.int.phx2.redhat.com (int-mx02.intmail.prod.int.phx2.redhat.com [10.5.11.12]) by mx1.redhat.com (8.14.4/8.14.4) with ESMTP id p8QEtrS2030041 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=OK); Mon, 26 Sep 2011 10:55:53 -0400 Received: from tyan-ft48-01.lab.bos.redhat.com (tyan-ft48-01.lab.bos.redhat.com [10.16.42.4]) by int-mx02.intmail.prod.int.phx2.redhat.com (8.13.8/8.13.8) with ESMTP id p8QEtq0L015584 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Mon, 26 Sep 2011 10:55:52 -0400 Received: from tyan-ft48-01.lab.bos.redhat.com (localhost.localdomain [127.0.0.1]) by tyan-ft48-01.lab.bos.redhat.com (8.14.4/8.14.4) with ESMTP id p8QEtpRp010523; Mon, 26 Sep 2011 16:55:51 +0200 Received: (from jakub@localhost) by tyan-ft48-01.lab.bos.redhat.com (8.14.4/8.14.4/Submit) id p8QEto6Q010522; Mon, 26 Sep 2011 16:55:50 +0200 Date: Mon, 26 Sep 2011 15:55:00 -0000 From: Jakub Jelinek To: Richard Guenther , Eric Botcazou , gcc-patches@gcc.gnu.org, richard.sandiford@linaro.org Subject: [PATCH] Optimize in RTL vector AND { -1, -1, ... }, IOR { -1, -1, ... } and XOR { -1, -1, ... } (take 2) Message-ID: <20110926145550.GG2687@tyan-ft48-01.lab.bos.redhat.com> Reply-To: Jakub Jelinek References: <20110926083144.GE2687@tyan-ft48-01.lab.bos.redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) X-IsSubscribed: yes Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org X-SW-Source: 2011-09/txt/msg01614.txt.bz2 On Mon, Sep 26, 2011 at 10:18:43AM +0100, Richard Sandiford wrote: > Jakub Jelinek writes: > > optimize all ones vectors in simplify-rtx.c (and i386 expansion): > > http://gcc.gnu.org/ml/gcc-patches/2011-09/msg01364.html > > I wonder whether we should have a CONSTM1_RTX(MODE). It seems > inconsistent to have vector 0s and 1s, but only have integer -1s. > > As far as simplify-rtx.c goes, I think all_ones_cst should cover the > existing CONST_INT_P case too. CONST_INTs have to be represented as > sign-extended, and "trueop1" should still have mode "mode", so something's > wrong if the current code matches things that aren't constm1_rtx. > The simplify-rtx.c change looks good otherwise. Here is an updated patch that introduces CONSTM1_RTX (for MODE_INT and MODE_VECTOR_INT modes) and uses it in simplify-rtx.c/i386.c instead. I'm testing INTEGRAL_MODE_P just in case in the future CONSTM1_RTX is filled also for MODE_FLOAT_INT etc. modes, because for them -1 is not all ones mask. Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? 2011-09-26 Jakub Jelinek * rtl.h (const_tiny_rtx): Change into array of 4 x MAX_MACHINE_MODE from 3 x MAX_MACHINE_MODE. (CONSTM1_RTX): Define. * emit-rtl.c (const_tiny_rtx): Change into array of 4 x MAX_MACHINE_MODE from 3 x MAX_MACHINE_MODE. (gen_rtx_CONST_VECTOR): Use CONSTM1_RTX if all inner constants are CONSTM1_RTX. (init_emit_once): Initialize CONSTM1_RTX for MODE_INT and MODE_VECTOR_INT modes. * simplify-rtx.c (simplify_binary_operation_1) : Optimize if one operand is CONSTM1_RTX. * config/i386/i386.c (ix86_expand_sse_movcc): Optimize mask ? -1 : x into mask | x. --- gcc/rtl.h.jj 2011-08-28 12:36:57.000000000 +0200 +++ gcc/rtl.h 2011-09-26 12:57:50.000000000 +0200 @@ -2074,17 +2074,18 @@ extern GTY(()) rtx const_int_rtx[MAX_SAV #define constm1_rtx (const_int_rtx[MAX_SAVED_CONST_INT-1]) extern GTY(()) rtx const_true_rtx; -extern GTY(()) rtx const_tiny_rtx[3][(int) MAX_MACHINE_MODE]; +extern GTY(()) rtx const_tiny_rtx[4][(int) MAX_MACHINE_MODE]; /* Returns a constant 0 rtx in mode MODE. Integer modes are treated the same as VOIDmode. */ #define CONST0_RTX(MODE) (const_tiny_rtx[0][(int) (MODE)]) -/* Likewise, for the constants 1 and 2. */ +/* Likewise, for the constants 1 and 2 and -1. */ #define CONST1_RTX(MODE) (const_tiny_rtx[1][(int) (MODE)]) #define CONST2_RTX(MODE) (const_tiny_rtx[2][(int) (MODE)]) +#define CONSTM1_RTX(MODE) (const_tiny_rtx[3][(int) (MODE)]) /* If HARD_FRAME_POINTER_REGNUM is defined, then a special dummy reg is used to represent the frame pointer. This is because the --- gcc/emit-rtl.c.jj 2011-09-20 21:43:07.000000000 +0200 +++ gcc/emit-rtl.c 2011-09-26 12:56:33.000000000 +0200 @@ -93,9 +93,10 @@ static GTY(()) int label_num = 1; /* We record floating-point CONST_DOUBLEs in each floating-point mode for the values of 0, 1, and 2. For the integer entries and VOIDmode, we - record a copy of const[012]_rtx. */ + record a copy of const[012]_rtx and constm1_rtx. CONSTM1_RTX + is set only for MODE_INT and MODE_VECTOR_INT modes. */ -rtx const_tiny_rtx[3][(int) MAX_MACHINE_MODE]; +rtx const_tiny_rtx[4][(int) MAX_MACHINE_MODE]; rtx const_true_rtx; @@ -5514,6 +5515,8 @@ gen_rtx_CONST_VECTOR (enum machine_mode return CONST0_RTX (mode); else if (x == CONST1_RTX (inner)) return CONST1_RTX (mode); + else if (x == CONSTM1_RTX (inner)) + return CONSTM1_RTX (mode); } return gen_rtx_raw_CONST_VECTOR (mode, v); @@ -5674,7 +5677,7 @@ init_emit_once (void) dconsthalf = dconst1; SET_REAL_EXP (&dconsthalf, REAL_EXP (&dconsthalf) - 1); - for (i = 0; i < (int) ARRAY_SIZE (const_tiny_rtx); i++) + for (i = 0; i < 3; i++) { const REAL_VALUE_TYPE *const r = (i == 0 ? &dconst0 : i == 1 ? &dconst1 : &dconst2); @@ -5704,6 +5707,13 @@ init_emit_once (void) const_tiny_rtx[i][(int) mode] = GEN_INT (i); } + const_tiny_rtx[3][(int) VOIDmode] = constm1_rtx; + + for (mode = GET_CLASS_NARROWEST_MODE (MODE_INT); + mode != VOIDmode; + mode = GET_MODE_WIDER_MODE (mode)) + const_tiny_rtx[3][(int) mode] = constm1_rtx; + for (mode = GET_CLASS_NARROWEST_MODE (MODE_COMPLEX_INT); mode != VOIDmode; mode = GET_MODE_WIDER_MODE (mode)) @@ -5726,6 +5736,7 @@ init_emit_once (void) { const_tiny_rtx[0][(int) mode] = gen_const_vector (mode, 0); const_tiny_rtx[1][(int) mode] = gen_const_vector (mode, 1); + const_tiny_rtx[3][(int) mode] = gen_const_vector (mode, 3); } for (mode = GET_CLASS_NARROWEST_MODE (MODE_VECTOR_FLOAT); --- gcc/simplify-rtx.c.jj 2011-09-15 12:18:54.000000000 +0200 +++ gcc/simplify-rtx.c 2011-09-26 13:17:33.000000000 +0200 @@ -2431,9 +2431,7 @@ simplify_binary_operation_1 (enum rtx_co case IOR: if (trueop1 == CONST0_RTX (mode)) return op0; - if (CONST_INT_P (trueop1) - && ((UINTVAL (trueop1) & GET_MODE_MASK (mode)) - == GET_MODE_MASK (mode))) + if (INTEGRAL_MODE_P (mode) && trueop1 == CONSTM1_RTX (mode)) return op1; if (rtx_equal_p (trueop0, trueop1) && ! side_effects_p (op0)) return op0; @@ -2573,9 +2571,7 @@ simplify_binary_operation_1 (enum rtx_co case XOR: if (trueop1 == CONST0_RTX (mode)) return op0; - if (CONST_INT_P (trueop1) - && ((UINTVAL (trueop1) & GET_MODE_MASK (mode)) - == GET_MODE_MASK (mode))) + if (INTEGRAL_MODE_P (mode) && trueop1 == CONSTM1_RTX (mode)) return simplify_gen_unary (NOT, mode, op0, mode); if (rtx_equal_p (trueop0, trueop1) && ! side_effects_p (op0) @@ -2721,6 +2717,8 @@ simplify_binary_operation_1 (enum rtx_co case AND: if (trueop1 == CONST0_RTX (mode) && ! side_effects_p (op0)) return trueop1; + if (INTEGRAL_MODE_P (mode) && trueop1 == CONSTM1_RTX (mode)) + return op0; if (HWI_COMPUTABLE_MODE_P (mode)) { HOST_WIDE_INT nzop0 = nonzero_bits (trueop0, mode); --- gcc/config/i386/i386.c.jj 2011-09-22 18:37:00.000000000 +0200 +++ gcc/config/i386/i386.c 2011-09-26 13:20:49.000000000 +0200 @@ -18899,6 +18899,12 @@ ix86_expand_sse_movcc (rtx dest, rtx cmp x = gen_rtx_AND (mode, x, op_false); emit_insn (gen_rtx_SET (VOIDmode, dest, x)); } + else if (INTEGRAL_MODE_P (mode) && op_true == CONSTM1_RTX (mode)) + { + op_false = force_reg (mode, op_false); + x = gen_rtx_IOR (mode, cmp, op_false); + emit_insn (gen_rtx_SET (VOIDmode, dest, x)); + } else if (TARGET_XOP) { op_true = force_reg (mode, op_true); Jakub