From patchwork Tue Nov 21 21:25:17 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Richard Henderson X-Patchwork-Id: 840204 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Authentication-Results: ozlabs.org; spf=pass (mailfrom) smtp.mailfrom=nongnu.org (client-ip=2001:4830:134:3::11; helo=lists.gnu.org; envelope-from=qemu-devel-bounces+incoming=patchwork.ozlabs.org@nongnu.org; receiver=) Authentication-Results: ozlabs.org; dkim=fail reason="signature verification failed" (1024-bit key; unprotected) header.d=linaro.org header.i=@linaro.org header.b="HtykI9Zf"; dkim-atps=neutral Received: from lists.gnu.org (lists.gnu.org [IPv6:2001:4830:134:3::11]) (using TLSv1 with cipher AES256-SHA (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 3yhJrw62fgz9t3s for ; Wed, 22 Nov 2017 08:39:44 +1100 (AEDT) Received: from localhost ([::1]:36636 helo=lists.gnu.org) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1eHGGc-0005yj-Si for incoming@patchwork.ozlabs.org; Tue, 21 Nov 2017 16:39:42 -0500 Received: from eggs.gnu.org ([2001:4830:134:3::10]:54025) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1eHG59-0004Mr-0Z for qemu-devel@nongnu.org; Tue, 21 Nov 2017 16:27:52 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1eHG57-0007ho-Ff for qemu-devel@nongnu.org; Tue, 21 Nov 2017 16:27:50 -0500 Received: from mail-wm0-x241.google.com ([2a00:1450:400c:c09::241]:38057) by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_128_CBC_SHA1:16) (Exim 4.71) (envelope-from ) id 1eHG57-0007hL-52 for qemu-devel@nongnu.org; Tue, 21 Nov 2017 16:27:49 -0500 Received: by mail-wm0-x241.google.com with SMTP id 128so6267641wmo.3 for ; Tue, 21 Nov 2017 13:27:49 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linaro.org; s=google; h=from:to:subject:date:message-id:in-reply-to:references:mime-version :content-transfer-encoding; bh=Rrd4S97eVhWKdenLugOq5tq9MgHu7wRf2kAbYFyPVy8=; b=HtykI9Zfzh+eKhbql3LZf4BPx/ybwj056POeUARhUso5xlSz/r/rApVGMYNXgpF45y 8JlVtXN0St+v/f0EdUovVHDjtWiRKkmdUKcItE0o/xb67POaWcX9fHgkTqxjlpxhmNug 5tn4jIRu8ra6fajVeUrTGbfPINmhL6pbdFgbc= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=Rrd4S97eVhWKdenLugOq5tq9MgHu7wRf2kAbYFyPVy8=; b=JE/D7TorPtPv3wT8lXVC/vp+DQWgh81GKq7tnoACpGdsgeHuCOso+eYF2VHeMIGIJd quiUwey2SFF0ha8Pw93vnt5K1PUQgVaVPlRxuqaXUpXOiMsDyveCNUnNlyUY0fiaYvYE w8ckUAKIsoFowtpXVR/oDDSBdUj1sUfDCHKDsrRjR3dlsxm5OyBckeJYxJTtloRQzO0s Eeez5SvbOgHfMJ85u6TmO3MjkBEtDC9WQphlcthWDruU8rpkjNs3n3vTcfMunO2VWUQm u0UHZ2eo8K7sR1V/5KxXLphsjS+zZFTftYsWJdeILIL1Js725CcE+xupBo+GLbI1mJJ6 jU5Q== X-Gm-Message-State: AJaThX6+1ll/1g++BYcyUL/WuOcpPSwYZ2QdkjLwsXDUny+TdbaJS/ol ZeQn/uixQ/zdK7VVji3ZGln8Yt5gYdE= X-Google-Smtp-Source: AGs4zMb4rIvCQAk/LL8EApAfkdH1o1rFDrtwW7IPlz19CRCRy64W3sTtYWl5OXJDy1Y4MDp/RdPbZA== X-Received: by 10.28.238.74 with SMTP id m71mr2287517wmh.83.1511299667871; Tue, 21 Nov 2017 13:27:47 -0800 (PST) Received: from cloudburst.twiddle.net (70.red-37-158-60.dynamicip.rima-tde.net. [37.158.60.70]) by smtp.gmail.com with ESMTPSA id e124sm706517wmg.34.2017.11.21.13.27.46 for (version=TLS1_2 cipher=ECDHE-RSA-CHACHA20-POLY1305 bits=256/256); Tue, 21 Nov 2017 13:27:47 -0800 (PST) From: Richard Henderson To: qemu-devel@nongnu.org Date: Tue, 21 Nov 2017 22:25:17 +0100 Message-Id: <20171121212534.5177-10-richard.henderson@linaro.org> X-Mailer: git-send-email 2.13.6 In-Reply-To: <20171121212534.5177-1-richard.henderson@linaro.org> References: <20171121212534.5177-1-richard.henderson@linaro.org> MIME-Version: 1.0 X-detected-operating-system: by eggs.gnu.org: Genre and OS details not recognized. X-Received-From: 2a00:1450:400c:c09::241 Subject: [Qemu-devel] [PATCH v6 09/26] target/arm: Use vector infrastructure for aa64 add/sub/logic X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.21 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" Reviewed-by: Alex Bennée Signed-off-by: Richard Henderson --- target/arm/translate-a64.c | 207 +++++++++++++++++++++++++++++---------------- 1 file changed, 134 insertions(+), 73 deletions(-) diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c index ba94f7d045..572af456d1 100644 --- a/target/arm/translate-a64.c +++ b/target/arm/translate-a64.c @@ -21,6 +21,7 @@ #include "cpu.h" #include "exec/exec-all.h" #include "tcg-op.h" +#include "tcg-op-gvec.h" #include "qemu/log.h" #include "arm_ldst.h" #include "translate.h" @@ -83,6 +84,10 @@ typedef void NeonGenOneOpFn(TCGv_i64, TCGv_i64); typedef void CryptoTwoOpEnvFn(TCGv_ptr, TCGv_i32, TCGv_i32); typedef void CryptoThreeOpEnvFn(TCGv_ptr, TCGv_i32, TCGv_i32, TCGv_i32); +/* Note that the gvec expanders operate on offsets + sizes. */ +typedef void GVecGen3Fn(unsigned, uint32_t, uint32_t, + uint32_t, uint32_t, uint32_t); + /* initialize TCG globals. */ void a64_translate_init(void) { @@ -535,6 +540,21 @@ static inline int vec_reg_offset(DisasContext *s, int regno, return offs; } +/* Return the offset info CPUARMState of the "whole" vector register Qn. */ +static inline int vec_full_reg_offset(DisasContext *s, int regno) +{ + assert_fp_access_checked(s); + return offsetof(CPUARMState, vfp.regs[regno * 2]); +} + +/* Return the byte size of the "whole" vector register, VL / 8. */ +static inline int vec_full_reg_size(DisasContext *s) +{ + /* FIXME SVE: We should put the composite ZCR_EL* value into tb->flags. + In the meantime this is just the AdvSIMD length of 128. */ + return 128 / 8; +} + /* Return the offset into CPUARMState of a slice (from * the least significant end) of FP register Qn (ie * Dn, Sn, Hn or Bn). @@ -9048,85 +9068,125 @@ static void disas_simd_three_reg_diff(DisasContext *s, uint32_t insn) } } +static void gen_bsl_i64(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm) +{ + tcg_gen_xor_i64(rn, rn, rm); + tcg_gen_and_i64(rn, rn, rd); + tcg_gen_xor_i64(rd, rm, rn); +} + +static void gen_bit_i64(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm) +{ + tcg_gen_xor_i64(rn, rn, rd); + tcg_gen_and_i64(rn, rn, rm); + tcg_gen_xor_i64(rd, rd, rn); +} + +static void gen_bif_i64(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm) +{ + tcg_gen_xor_i64(rn, rn, rd); + tcg_gen_andc_i64(rn, rn, rm); + tcg_gen_xor_i64(rd, rd, rn); +} + +static void gen_bsl_vec(unsigned vece, TCGv_vec rd, TCGv_vec rn, TCGv_vec rm) +{ + tcg_gen_xor_vec(vece, rn, rn, rm); + tcg_gen_and_vec(vece, rn, rn, rd); + tcg_gen_xor_vec(vece, rd, rm, rn); +} + +static void gen_bit_vec(unsigned vece, TCGv_vec rd, TCGv_vec rn, TCGv_vec rm) +{ + tcg_gen_xor_vec(vece, rn, rn, rd); + tcg_gen_and_vec(vece, rn, rn, rm); + tcg_gen_xor_vec(vece, rd, rd, rn); +} + +static void gen_bif_vec(unsigned vece, TCGv_vec rd, TCGv_vec rn, TCGv_vec rm) +{ + tcg_gen_xor_vec(vece, rn, rn, rd); + tcg_gen_andc_vec(vece, rn, rn, rm); + tcg_gen_xor_vec(vece, rd, rd, rn); +} + /* Logic op (opcode == 3) subgroup of C3.6.16. */ static void disas_simd_3same_logic(DisasContext *s, uint32_t insn) { + static const GVecGen3 bsl_op = { + .fni8 = gen_bsl_i64, + .fniv = gen_bsl_vec, + .prefer_i64 = TCG_TARGET_REG_BITS == 64, + .load_dest = true + }; + static const GVecGen3 bit_op = { + .fni8 = gen_bit_i64, + .fniv = gen_bit_vec, + .prefer_i64 = TCG_TARGET_REG_BITS == 64, + .load_dest = true + }; + static const GVecGen3 bif_op = { + .fni8 = gen_bif_i64, + .fniv = gen_bif_vec, + .prefer_i64 = TCG_TARGET_REG_BITS == 64, + .load_dest = true + }; + int rd = extract32(insn, 0, 5); int rn = extract32(insn, 5, 5); int rm = extract32(insn, 16, 5); int size = extract32(insn, 22, 2); bool is_u = extract32(insn, 29, 1); bool is_q = extract32(insn, 30, 1); - TCGv_i64 tcg_op1, tcg_op2, tcg_res[2]; - int pass; + GVecGen3Fn *gvec_fn; + const GVecGen3 *gvec_op; if (!fp_access_check(s)) { return; } - tcg_op1 = tcg_temp_new_i64(); - tcg_op2 = tcg_temp_new_i64(); - tcg_res[0] = tcg_temp_new_i64(); - tcg_res[1] = tcg_temp_new_i64(); - - for (pass = 0; pass < (is_q ? 2 : 1); pass++) { - read_vec_element(s, tcg_op1, rn, pass, MO_64); - read_vec_element(s, tcg_op2, rm, pass, MO_64); - - if (!is_u) { - switch (size) { - case 0: /* AND */ - tcg_gen_and_i64(tcg_res[pass], tcg_op1, tcg_op2); - break; - case 1: /* BIC */ - tcg_gen_andc_i64(tcg_res[pass], tcg_op1, tcg_op2); - break; - case 2: /* ORR */ - tcg_gen_or_i64(tcg_res[pass], tcg_op1, tcg_op2); - break; - case 3: /* ORN */ - tcg_gen_orc_i64(tcg_res[pass], tcg_op1, tcg_op2); - break; - } - } else { - if (size != 0) { - /* B* ops need res loaded to operate on */ - read_vec_element(s, tcg_res[pass], rd, pass, MO_64); - } - - switch (size) { - case 0: /* EOR */ - tcg_gen_xor_i64(tcg_res[pass], tcg_op1, tcg_op2); - break; - case 1: /* BSL bitwise select */ - tcg_gen_xor_i64(tcg_op1, tcg_op1, tcg_op2); - tcg_gen_and_i64(tcg_op1, tcg_op1, tcg_res[pass]); - tcg_gen_xor_i64(tcg_res[pass], tcg_op2, tcg_op1); - break; - case 2: /* BIT, bitwise insert if true */ - tcg_gen_xor_i64(tcg_op1, tcg_op1, tcg_res[pass]); - tcg_gen_and_i64(tcg_op1, tcg_op1, tcg_op2); - tcg_gen_xor_i64(tcg_res[pass], tcg_res[pass], tcg_op1); - break; - case 3: /* BIF, bitwise insert if false */ - tcg_gen_xor_i64(tcg_op1, tcg_op1, tcg_res[pass]); - tcg_gen_andc_i64(tcg_op1, tcg_op1, tcg_op2); - tcg_gen_xor_i64(tcg_res[pass], tcg_res[pass], tcg_op1); - break; - } - } - } + switch (size + 4 * is_u) { + case 0: /* AND */ + gvec_fn = tcg_gen_gvec_and; + goto do_fn; + case 1: /* BIC */ + gvec_fn = tcg_gen_gvec_andc; + goto do_fn; + case 2: /* ORR */ + gvec_fn = tcg_gen_gvec_or; + goto do_fn; + case 3: /* ORN */ + gvec_fn = tcg_gen_gvec_orc; + goto do_fn; + case 4: /* EOR */ + gvec_fn = tcg_gen_gvec_xor; + goto do_fn; + do_fn: + gvec_fn(0, vec_full_reg_offset(s, rd), + vec_full_reg_offset(s, rn), + vec_full_reg_offset(s, rm), + is_q ? 16 : 8, vec_full_reg_size(s)); + return; + + case 5: /* BSL bitwise select */ + gvec_op = &bsl_op; + goto do_op; + case 6: /* BIT, bitwise insert if true */ + gvec_op = &bit_op; + goto do_op; + case 7: /* BIF, bitwise insert if false */ + gvec_op = &bif_op; + goto do_op; + do_op: + tcg_gen_gvec_3(vec_full_reg_offset(s, rd), + vec_full_reg_offset(s, rn), + vec_full_reg_offset(s, rm), + is_q ? 16 : 8, vec_full_reg_size(s), gvec_op); + return; - write_vec_element(s, tcg_res[0], rd, 0, MO_64); - if (!is_q) { - tcg_gen_movi_i64(tcg_res[1], 0); + default: + g_assert_not_reached(); } - write_vec_element(s, tcg_res[1], rd, 1, MO_64); - - tcg_temp_free_i64(tcg_op1); - tcg_temp_free_i64(tcg_op2); - tcg_temp_free_i64(tcg_res[0]); - tcg_temp_free_i64(tcg_res[1]); } /* Helper functions for 32 bit comparisons */ @@ -9387,6 +9447,7 @@ static void disas_simd_3same_int(DisasContext *s, uint32_t insn) int rn = extract32(insn, 5, 5); int rd = extract32(insn, 0, 5); int pass; + GVecGen3Fn *gvec_op; switch (opcode) { case 0x13: /* MUL, PMUL */ @@ -9426,6 +9487,16 @@ static void disas_simd_3same_int(DisasContext *s, uint32_t insn) return; } + switch (opcode) { + case 0x10: /* ADD, SUB */ + gvec_op = u ? tcg_gen_gvec_sub : tcg_gen_gvec_add; + gvec_op(size, vec_full_reg_offset(s, rd), + vec_full_reg_offset(s, rn), + vec_full_reg_offset(s, rm), + is_q ? 16 : 8, vec_full_reg_size(s)); + return; + } + if (size == 3) { assert(is_q); for (pass = 0; pass < 2; pass++) { @@ -9598,16 +9669,6 @@ static void disas_simd_3same_int(DisasContext *s, uint32_t insn) genfn = fns[size][u]; break; } - case 0x10: /* ADD, SUB */ - { - static NeonGenTwoOpFn * const fns[3][2] = { - { gen_helper_neon_add_u8, gen_helper_neon_sub_u8 }, - { gen_helper_neon_add_u16, gen_helper_neon_sub_u16 }, - { tcg_gen_add_i32, tcg_gen_sub_i32 }, - }; - genfn = fns[size][u]; - break; - } case 0x11: /* CMTST, CMEQ */ { static NeonGenTwoOpFn * const fns[3][2] = {