From patchwork Mon Sep 2 17:54:45 2013 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Richard Henderson X-Patchwork-Id: 272030 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from lists.gnu.org (lists.gnu.org [IPv6:2001:4830:134:3::11]) (using TLSv1 with cipher AES256-SHA (256/256 bits)) (Client did not present a certificate) by ozlabs.org (Postfix) with ESMTPS id 3B7E72C009A for ; Tue, 3 Sep 2013 04:03:49 +1000 (EST) Received: from localhost ([::1]:41425 helo=lists.gnu.org) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VGYSL-0006Yp-KC for incoming@patchwork.ozlabs.org; Mon, 02 Sep 2013 14:02:29 -0400 Received: from eggs.gnu.org ([2001:4830:134:3::10]:56660) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VGYLe-0005wK-8t for qemu-devel@nongnu.org; Mon, 02 Sep 2013 13:55:40 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1VGYLY-00017t-BO for qemu-devel@nongnu.org; Mon, 02 Sep 2013 13:55:34 -0400 Received: from mail-pa0-x22d.google.com ([2607:f8b0:400e:c03::22d]:59281) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VGYLX-00017h-Vb for qemu-devel@nongnu.org; Mon, 02 Sep 2013 13:55:28 -0400 Received: by mail-pa0-f45.google.com with SMTP id bg4so5428277pad.18 for ; Mon, 02 Sep 2013 10:55:27 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=sender:from:to:cc:subject:date:message-id:in-reply-to:references; bh=BU5aGiiWhD6z4mNjjHvkcqjQRjKRuriW3vLqBrTqMNc=; b=uZNo/Uf3PvzienBTslm4wbtlu7eMXBYJDp0nJlPMmix8bzNb8+++SD5YBgrWtgF7Ot UdGKhlNLihQybty47rEaTpYErsLJanhVt40UdyH+K2wM9rbsVCU+idvzJmuHpdOChVpb xSF/CtgZkf5PTR+EMuTwfSH7CDBuFzndPToNRbybFdiP2LhVMsBSdeFWzzFnGvMrgX4y M07wBxXLu8uB+9X+6kH6aMktNr1EVQf7eqR0Zu+emGwkFsgDcwmuBAakG1+IwhiKhDrm hf76zRoLpZa3FB1wuIZrBBX+K28RAR1nk0XpSW22JjecLmGZh/8/VYj0nd/ZKAHvca2A dMiQ== X-Received: by 10.66.121.234 with SMTP id ln10mr27173909pab.20.1378144526924; Mon, 02 Sep 2013 10:55:26 -0700 (PDT) Received: from pebble.twiddle.net (50-194-63-110-static.hfc.comcastbusiness.net. [50.194.63.110]) by mx.google.com with ESMTPSA id tr10sm17218114pbc.22.1969.12.31.16.00.00 (version=TLSv1.2 cipher=RC4-SHA bits=128/128); Mon, 02 Sep 2013 10:55:26 -0700 (PDT) From: Richard Henderson To: qemu-devel@nongnu.org Date: Mon, 2 Sep 2013 10:54:45 -0700 Message-Id: <1378144503-15808-12-git-send-email-rth@twiddle.net> X-Mailer: git-send-email 1.8.3.1 In-Reply-To: <1378144503-15808-1-git-send-email-rth@twiddle.net> References: <1378144503-15808-1-git-send-email-rth@twiddle.net> X-detected-operating-system: by eggs.gnu.org: Error: Malformed IPv6 address (bad octet value). X-Received-From: 2607:f8b0:400e:c03::22d Cc: claudio.fontana@huawei.com, Richard Henderson Subject: [Qemu-devel] [PATCH v3 11/29] tcg-aarch64: Handle constant operands to and, or, xor X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.14 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: qemu-devel-bounces+incoming=patchwork.ozlabs.org@nongnu.org Sender: qemu-devel-bounces+incoming=patchwork.ozlabs.org@nongnu.org Handle a simplified set of logical immediates for the moment. The way gcc and binutils do it, with 52k worth of tables, and a binary search depth of log2(5334) = 13, seems slow for the most common cases. Signed-off-by: Richard Henderson --- tcg/aarch64/tcg-target.c | 148 +++++++++++++++++++++++++++++++---------------- 1 file changed, 99 insertions(+), 49 deletions(-) diff --git a/tcg/aarch64/tcg-target.c b/tcg/aarch64/tcg-target.c index dbb1c45..9324185 100644 --- a/tcg/aarch64/tcg-target.c +++ b/tcg/aarch64/tcg-target.c @@ -108,6 +108,7 @@ static inline void patch_reloc(uint8_t *code_ptr, int type, #define TCG_CT_CONST_IS32 0x100 #define TCG_CT_CONST_AIMM 0x200 +#define TCG_CT_CONST_LIMM 0x400 /* parse target specific constraints */ static int target_parse_constraint(TCGArgConstraint *ct, @@ -138,6 +139,9 @@ static int target_parse_constraint(TCGArgConstraint *ct, case 'A': /* Valid for arithmetic immediate (positive or negative). */ ct->ct |= TCG_CT_CONST_AIMM; break; + case 'L': /* Valid for logical immediate. */ + ct->ct |= TCG_CT_CONST_LIMM; + break; default: return -1; } @@ -152,6 +156,26 @@ static inline bool is_aimm(uint64_t val) return (val & ~0xfff) == 0 || (val & ~0xfff000) == 0; } +static inline bool is_limm(uint64_t val) +{ + /* Taking a simplified view of the logical immediates for now, ignoring + the replication that can happen across the field. Match bit patterns + of the forms + 0....01....1 + 0..01..10..0 + and their inverses. */ + + /* Make things easier below, by testing the form with msb clear. */ + if ((int64_t)val < 0) { + val = ~val; + } + if (val == 0) { + return false; + } + val += val & -val; + return (val & (val - 1)) == 0; +} + static int tcg_target_const_match(tcg_target_long val, const TCGArgConstraint *arg_ct) { @@ -166,6 +190,9 @@ static int tcg_target_const_match(tcg_target_long val, if ((ct & TCG_CT_CONST_AIMM) && (is_aimm(val) || is_aimm(-val))) { return 1; } + if ((ct & TCG_CT_CONST_LIMM) && is_limm(val)) { + return 1; + } return 0; } @@ -220,6 +247,11 @@ enum aarch64_ldst_op_type { /* type of operation */ }; typedef enum { + /* Logical immediate instructions */ + INSN_ANDI = 0x12000000, + INSN_ORRI = 0x32000000, + INSN_EORI = 0x52000000, + /* Logical shifted register instructions */ INSN_AND = 0x0a000000, INSN_ORR = 0x2a000000, @@ -359,6 +391,41 @@ static inline void tcg_fmt_Rdn_aimm(TCGContext *s, AArch64Insn insn, bool ext, tcg_out32(s, insn | ext << 31 | aimm << 10 | rn << 5 | rd); } +static inline void tcg_fmt_Rdn_r_s(TCGContext *s, AArch64Insn insn, bool ext, + TCGReg rd, TCGReg rn, int r, int m) +{ + tcg_out32(s, insn | ext * 0x80400000 | r << 16 | m << 10 | rn << 5 | rd); +} + +static void tcg_fmt_Rdn_limm(TCGContext *s, AArch64Insn insn, bool ext, + TCGReg rd, TCGReg rn, uint64_t limm) +{ + unsigned h, l, r, c; + + /* See is_limm comment about simplified logical immediates. */ + assert(is_limm(limm)); + + h = clz64(limm); + l = ctz64(limm); + if (l == 0) { + r = 0; /* form 0....01....1 */ + c = ctz64(~limm) - 1; + if (h == 0) { + r = clz64(~limm); /* form 1..10..01..1 */ + c += r; + } + } else { + r = 64 - l; /* form 1....10....0 or 0..01..10..0 */ + c = r - h - 1; + } + if (!ext) { + r &= 31; + c &= 31; + } + + tcg_fmt_Rdn_r_s(s, insn, ext, rd, rn, r, c); +} + static inline void tcg_out_ldst_9(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type, @@ -665,40 +732,6 @@ static inline void tcg_out_call(TCGContext *s, tcg_target_long target) } } -/* encode a logical immediate, mapping user parameter - M=set bits pattern length to S=M-1 */ -static inline unsigned int -aarch64_limm(unsigned int m, unsigned int r) -{ - assert(m > 0); - return r << 16 | (m - 1) << 10; -} - -/* test a register against an immediate bit pattern made of - M set bits rotated right by R. - Examples: - to test a 32/64 reg against 0x00000007, pass M = 3, R = 0. - to test a 32/64 reg against 0x000000ff, pass M = 8, R = 0. - to test a 32bit reg against 0xff000000, pass M = 8, R = 8. - to test a 32bit reg against 0xff0000ff, pass M = 16, R = 8. - */ -static inline void tcg_out_tst(TCGContext *s, bool ext, TCGReg rn, - unsigned int m, unsigned int r) -{ - /* using TST alias of ANDS XZR, Xn,#bimm64 0x7200001f */ - unsigned int base = ext ? 0xf240001f : 0x7200001f; - tcg_out32(s, base | aarch64_limm(m, r) | rn << 5); -} - -/* and a register with a bit pattern, similarly to TST, no flags change */ -static inline void tcg_out_andi(TCGContext *s, bool ext, TCGReg rd, TCGReg rn, - unsigned int m, unsigned int r) -{ - /* using AND 0x12000000 */ - unsigned int base = ext ? 0x92400000 : 0x12000000; - tcg_out32(s, base | aarch64_limm(m, r) | rn << 5 | rd); -} - static inline void tcg_out_ret(TCGContext *s) { /* emit RET { LR } */ @@ -901,9 +934,8 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, /* Store the page mask part of the address and the low s_bits into X3. Later this allows checking for equality and alignment at the same time. X3 = addr_reg & (PAGE_MASK | ((1 << s_bits) - 1)) */ - tcg_out_andi(s, (TARGET_LONG_BITS == 64), TCG_REG_X3, addr_reg, - (TARGET_LONG_BITS - TARGET_PAGE_BITS) + s_bits, - (TARGET_LONG_BITS - TARGET_PAGE_BITS)); + tcg_fmt_Rdn_limm(s, INSN_ANDI, TARGET_LONG_BITS == 64, TCG_REG_X3, + addr_reg, TARGET_PAGE_MASK | ((1 << s_bits) - 1)); /* Add any "high bits" from the tlb offset to the env address into X2, to take advantage of the LSL12 form of the ADDI instruction. X2 = env + (tlb_offset & 0xfff000) */ @@ -1207,19 +1239,37 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, } break; - case INDEX_op_and_i64: case INDEX_op_and_i32: - tcg_fmt_Rdnm(s, INSN_AND, ext, a0, a1, a2); + a2 = (int32_t)a2; + /* FALLTHRU */ + case INDEX_op_and_i64: + if (c2) { + tcg_fmt_Rdn_limm(s, INSN_ANDI, ext, a0, a1, a2); + } else { + tcg_fmt_Rdnm(s, INSN_AND, ext, a0, a1, a2); + } break; - case INDEX_op_or_i64: case INDEX_op_or_i32: - tcg_fmt_Rdnm(s, INSN_ORR, ext, a0, a1, a2); + a2 = (int32_t)a2; + /* FALLTHRU */ + case INDEX_op_or_i64: + if (c2) { + tcg_fmt_Rdn_limm(s, INSN_ORRI, ext, a0, a1, a2); + } else { + tcg_fmt_Rdnm(s, INSN_ORR, ext, a0, a1, a2); + } break; - case INDEX_op_xor_i64: case INDEX_op_xor_i32: - tcg_fmt_Rdnm(s, INSN_EOR, ext, a0, a1, a2); + a2 = (int32_t)a2; + /* FALLTHRU */ + case INDEX_op_xor_i64: + if (c2) { + tcg_fmt_Rdn_limm(s, INSN_EORI, ext, a0, a1, a2); + } else { + tcg_fmt_Rdnm(s, INSN_EOR, ext, a0, a1, a2); + } break; case INDEX_op_mul_i64: @@ -1412,12 +1462,12 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_sub_i64, { "r", "r", "rA" } }, { INDEX_op_mul_i32, { "r", "r", "r" } }, { INDEX_op_mul_i64, { "r", "r", "r" } }, - { INDEX_op_and_i32, { "r", "r", "r" } }, - { INDEX_op_and_i64, { "r", "r", "r" } }, - { INDEX_op_or_i32, { "r", "r", "r" } }, - { INDEX_op_or_i64, { "r", "r", "r" } }, - { INDEX_op_xor_i32, { "r", "r", "r" } }, - { INDEX_op_xor_i64, { "r", "r", "r" } }, + { INDEX_op_and_i32, { "r", "r", "rwL" } }, + { INDEX_op_and_i64, { "r", "r", "rL" } }, + { INDEX_op_or_i32, { "r", "r", "rwL" } }, + { INDEX_op_or_i64, { "r", "r", "rL" } }, + { INDEX_op_xor_i32, { "r", "r", "rwL" } }, + { INDEX_op_xor_i64, { "r", "r", "rL" } }, { INDEX_op_shl_i32, { "r", "r", "ri" } }, { INDEX_op_shr_i32, { "r", "r", "ri" } },