view src/hotspot/cpu/x86/x86.ad @ 49389:b242a1e3f9cf

8199421: Add support for vector popcount Reviewed-by: kvn
author rlupusoru
date Tue, 13 Mar 2018 10:22:15 -0700
parents 1a0499fd252e
children 767cdb97f103
line wrap: on
line source
//
// Copyright (c) 2011, 2017, Oracle and/or its affiliates. All rights reserved.
// DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
//
// This code is free software; you can redistribute it and/or modify it
// under the terms of the GNU General Public License version 2 only, as
// published by the Free Software Foundation.
//
// This code is distributed in the hope that it will be useful, but WITHOUT
// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
// FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
// version 2 for more details (a copy is included in the LICENSE file that
// accompanied this code).
//
// You should have received a copy of the GNU General Public License version
// 2 along with this work; if not, write to the Free Software Foundation,
// Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
//
// Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
// or visit www.oracle.com if you need additional information or have any
// questions.
//
//

// X86 Common Architecture Description File

//----------REGISTER DEFINITION BLOCK------------------------------------------
// This information is used by the matcher and the register allocator to
// describe individual registers and classes of registers within the target
// archtecture.

register %{
//----------Architecture Description Register Definitions----------------------
// General Registers
// "reg_def"  name ( register save type, C convention save type,
//                   ideal register type, encoding );
// Register Save Types:
//
// NS  = No-Save:       The register allocator assumes that these registers
//                      can be used without saving upon entry to the method, &
//                      that they do not need to be saved at call sites.
//
// SOC = Save-On-Call:  The register allocator assumes that these registers
//                      can be used without saving upon entry to the method,
//                      but that they must be saved at call sites.
//
// SOE = Save-On-Entry: The register allocator assumes that these registers
//                      must be saved before using them upon entry to the
//                      method, but they do not need to be saved at call
//                      sites.
//
// AS  = Always-Save:   The register allocator assumes that these registers
//                      must be saved before using them upon entry to the
//                      method, & that they must be saved at call sites.
//
// Ideal Register Type is used to determine how to save & restore a
// register.  Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
// spilled with LoadP/StoreP.  If the register supports both, use Op_RegI.
//
// The encoding number is the actual bit-pattern placed into the opcodes.

// XMM registers.  512-bit registers or 8 words each, labeled (a)-p.
// Word a in each register holds a Float, words ab hold a Double.
// The whole registers are used in SSE4.2 version intrinsics,
// array copy stubs and superword operations (see UseSSE42Intrinsics,
// UseXMMForArrayCopy and UseSuperword flags).
// For pre EVEX enabled architectures:
//      XMM8-XMM15 must be encoded with REX (VEX for UseAVX)
// For EVEX enabled architectures:
//      XMM8-XMM31 must be encoded with REX (EVEX for UseAVX).
//
// Linux ABI:   No register preserved across function calls
//              XMM0-XMM7 might hold parameters
// Windows ABI: XMM6-XMM31 preserved across function calls
//              XMM0-XMM3 might hold parameters

reg_def XMM0 ( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg());
reg_def XMM0b( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(1));
reg_def XMM0c( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(2));
reg_def XMM0d( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(3));
reg_def XMM0e( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(4));
reg_def XMM0f( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(5));
reg_def XMM0g( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(6));
reg_def XMM0h( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(7));
reg_def XMM0i( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(8));
reg_def XMM0j( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(9));
reg_def XMM0k( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(10));
reg_def XMM0l( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(11));
reg_def XMM0m( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(12));
reg_def XMM0n( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(13));
reg_def XMM0o( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(14));
reg_def XMM0p( SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next(15));

reg_def XMM1 ( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg());
reg_def XMM1b( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(1));
reg_def XMM1c( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(2));
reg_def XMM1d( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(3));
reg_def XMM1e( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(4));
reg_def XMM1f( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(5));
reg_def XMM1g( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(6));
reg_def XMM1h( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(7));
reg_def XMM1i( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(8));
reg_def XMM1j( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(9));
reg_def XMM1k( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(10));
reg_def XMM1l( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(11));
reg_def XMM1m( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(12));
reg_def XMM1n( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(13));
reg_def XMM1o( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(14));
reg_def XMM1p( SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next(15));

reg_def XMM2 ( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg());
reg_def XMM2b( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(1));
reg_def XMM2c( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(2));
reg_def XMM2d( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(3));
reg_def XMM2e( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(4));
reg_def XMM2f( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(5));
reg_def XMM2g( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(6));
reg_def XMM2h( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(7));
reg_def XMM2i( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(8));
reg_def XMM2j( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(9));
reg_def XMM2k( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(10));
reg_def XMM2l( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(11));
reg_def XMM2m( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(12));
reg_def XMM2n( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(13));
reg_def XMM2o( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(14));
reg_def XMM2p( SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next(15));

reg_def XMM3 ( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg());
reg_def XMM3b( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(1));
reg_def XMM3c( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(2));
reg_def XMM3d( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(3));
reg_def XMM3e( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(4));
reg_def XMM3f( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(5));
reg_def XMM3g( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(6));
reg_def XMM3h( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(7));
reg_def XMM3i( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(8));
reg_def XMM3j( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(9));
reg_def XMM3k( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(10));
reg_def XMM3l( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(11));
reg_def XMM3m( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(12));
reg_def XMM3n( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(13));
reg_def XMM3o( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(14));
reg_def XMM3p( SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next(15));

reg_def XMM4 ( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg());
reg_def XMM4b( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(1));
reg_def XMM4c( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(2));
reg_def XMM4d( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(3));
reg_def XMM4e( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(4));
reg_def XMM4f( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(5));
reg_def XMM4g( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(6));
reg_def XMM4h( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(7));
reg_def XMM4i( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(8));
reg_def XMM4j( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(9));
reg_def XMM4k( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(10));
reg_def XMM4l( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(11));
reg_def XMM4m( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(12));
reg_def XMM4n( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(13));
reg_def XMM4o( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(14));
reg_def XMM4p( SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next(15));

reg_def XMM5 ( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg());
reg_def XMM5b( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(1));
reg_def XMM5c( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(2));
reg_def XMM5d( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(3));
reg_def XMM5e( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(4));
reg_def XMM5f( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(5));
reg_def XMM5g( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(6));
reg_def XMM5h( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(7));
reg_def XMM5i( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(8));
reg_def XMM5j( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(9));
reg_def XMM5k( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(10));
reg_def XMM5l( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(11));
reg_def XMM5m( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(12));
reg_def XMM5n( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(13));
reg_def XMM5o( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(14));
reg_def XMM5p( SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next(15));

reg_def XMM6 ( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg());
reg_def XMM6b( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(1));
reg_def XMM6c( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(2));
reg_def XMM6d( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(3));
reg_def XMM6e( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(4));
reg_def XMM6f( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(5));
reg_def XMM6g( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(6));
reg_def XMM6h( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(7));
reg_def XMM6i( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(8));
reg_def XMM6j( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(9));
reg_def XMM6k( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(10));
reg_def XMM6l( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(11));
reg_def XMM6m( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(12));
reg_def XMM6n( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(13));
reg_def XMM6o( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(14));
reg_def XMM6p( SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next(15));

reg_def XMM7 ( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg());
reg_def XMM7b( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(1));
reg_def XMM7c( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(2));
reg_def XMM7d( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(3));
reg_def XMM7e( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(4));
reg_def XMM7f( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(5));
reg_def XMM7g( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(6));
reg_def XMM7h( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(7));
reg_def XMM7i( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(8));
reg_def XMM7j( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(9));
reg_def XMM7k( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(10));
reg_def XMM7l( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(11));
reg_def XMM7m( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(12));
reg_def XMM7n( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(13));
reg_def XMM7o( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(14));
reg_def XMM7p( SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next(15));

#ifdef _LP64

reg_def XMM8 ( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg());
reg_def XMM8b( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(1));
reg_def XMM8c( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(2));
reg_def XMM8d( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(3));
reg_def XMM8e( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(4));
reg_def XMM8f( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(5));
reg_def XMM8g( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(6));
reg_def XMM8h( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(7));
reg_def XMM8i( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(8));
reg_def XMM8j( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(9));
reg_def XMM8k( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(10));
reg_def XMM8l( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(11));
reg_def XMM8m( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(12));
reg_def XMM8n( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(13));
reg_def XMM8o( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(14));
reg_def XMM8p( SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next(15));

reg_def XMM9 ( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg());
reg_def XMM9b( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(1));
reg_def XMM9c( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(2));
reg_def XMM9d( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(3));
reg_def XMM9e( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(4));
reg_def XMM9f( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(5));
reg_def XMM9g( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(6));
reg_def XMM9h( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(7));
reg_def XMM9i( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(8));
reg_def XMM9j( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(9));
reg_def XMM9k( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(10));
reg_def XMM9l( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(11));
reg_def XMM9m( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(12));
reg_def XMM9n( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(13));
reg_def XMM9o( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(14));
reg_def XMM9p( SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next(15));

reg_def XMM10 ( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg());
reg_def XMM10b( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(1));
reg_def XMM10c( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(2));
reg_def XMM10d( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(3));
reg_def XMM10e( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(4));
reg_def XMM10f( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(5));
reg_def XMM10g( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(6));
reg_def XMM10h( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(7));
reg_def XMM10i( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(8));
reg_def XMM10j( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(9));
reg_def XMM10k( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(10));
reg_def XMM10l( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(11));
reg_def XMM10m( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(12));
reg_def XMM10n( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(13));
reg_def XMM10o( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(14));
reg_def XMM10p( SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next(15));

reg_def XMM11 ( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg());
reg_def XMM11b( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(1));
reg_def XMM11c( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(2));
reg_def XMM11d( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(3));
reg_def XMM11e( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(4));
reg_def XMM11f( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(5));
reg_def XMM11g( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(6));
reg_def XMM11h( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(7));
reg_def XMM11i( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(8));
reg_def XMM11j( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(9));
reg_def XMM11k( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(10));
reg_def XMM11l( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(11));
reg_def XMM11m( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(12));
reg_def XMM11n( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(13));
reg_def XMM11o( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(14));
reg_def XMM11p( SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next(15));

reg_def XMM12 ( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg());
reg_def XMM12b( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(1));
reg_def XMM12c( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(2));
reg_def XMM12d( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(3));
reg_def XMM12e( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(4));
reg_def XMM12f( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(5));
reg_def XMM12g( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(6));
reg_def XMM12h( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(7));
reg_def XMM12i( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(8));
reg_def XMM12j( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(9));
reg_def XMM12k( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(10));
reg_def XMM12l( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(11));
reg_def XMM12m( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(12));
reg_def XMM12n( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(13));
reg_def XMM12o( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(14));
reg_def XMM12p( SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next(15));

reg_def XMM13 ( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg());
reg_def XMM13b( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(1));
reg_def XMM13c( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(2));
reg_def XMM13d( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(3));
reg_def XMM13e( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(4));
reg_def XMM13f( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(5));
reg_def XMM13g( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(6));
reg_def XMM13h( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(7));
reg_def XMM13i( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(8));
reg_def XMM13j( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(9));
reg_def XMM13k( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(10));
reg_def XMM13l( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(11));
reg_def XMM13m( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(12));
reg_def XMM13n( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(13));
reg_def XMM13o( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(14));
reg_def XMM13p( SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next(15));

reg_def XMM14 ( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg());
reg_def XMM14b( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(1));
reg_def XMM14c( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(2));
reg_def XMM14d( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(3));
reg_def XMM14e( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(4));
reg_def XMM14f( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(5));
reg_def XMM14g( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(6));
reg_def XMM14h( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(7));
reg_def XMM14i( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(8));
reg_def XMM14j( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(9));
reg_def XMM14k( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(10));
reg_def XMM14l( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(11));
reg_def XMM14m( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(12));
reg_def XMM14n( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(13));
reg_def XMM14o( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(14));
reg_def XMM14p( SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next(15));

reg_def XMM15 ( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg());
reg_def XMM15b( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(1));
reg_def XMM15c( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(2));
reg_def XMM15d( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(3));
reg_def XMM15e( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(4));
reg_def XMM15f( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(5));
reg_def XMM15g( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(6));
reg_def XMM15h( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(7));
reg_def XMM15i( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(8));
reg_def XMM15j( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(9));
reg_def XMM15k( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(10));
reg_def XMM15l( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(11));
reg_def XMM15m( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(12));
reg_def XMM15n( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(13));
reg_def XMM15o( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(14));
reg_def XMM15p( SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next(15));

reg_def XMM16 ( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg());
reg_def XMM16b( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(1));
reg_def XMM16c( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(2));
reg_def XMM16d( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(3));
reg_def XMM16e( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(4));
reg_def XMM16f( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(5));
reg_def XMM16g( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(6));
reg_def XMM16h( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(7));
reg_def XMM16i( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(8));
reg_def XMM16j( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(9));
reg_def XMM16k( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(10));
reg_def XMM16l( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(11));
reg_def XMM16m( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(12));
reg_def XMM16n( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(13));
reg_def XMM16o( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(14));
reg_def XMM16p( SOC, SOC, Op_RegF, 16, xmm16->as_VMReg()->next(15));

reg_def XMM17 ( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg());
reg_def XMM17b( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(1));
reg_def XMM17c( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(2));
reg_def XMM17d( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(3));
reg_def XMM17e( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(4));
reg_def XMM17f( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(5));
reg_def XMM17g( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(6));
reg_def XMM17h( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(7));
reg_def XMM17i( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(8));
reg_def XMM17j( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(9));
reg_def XMM17k( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(10));
reg_def XMM17l( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(11));
reg_def XMM17m( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(12));
reg_def XMM17n( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(13));
reg_def XMM17o( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(14));
reg_def XMM17p( SOC, SOC, Op_RegF, 17, xmm17->as_VMReg()->next(15));

reg_def XMM18 ( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg());
reg_def XMM18b( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(1));
reg_def XMM18c( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(2));
reg_def XMM18d( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(3));
reg_def XMM18e( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(4));
reg_def XMM18f( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(5));
reg_def XMM18g( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(6));
reg_def XMM18h( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(7));
reg_def XMM18i( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(8));
reg_def XMM18j( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(9));
reg_def XMM18k( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(10));
reg_def XMM18l( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(11));
reg_def XMM18m( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(12));
reg_def XMM18n( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(13));
reg_def XMM18o( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(14));
reg_def XMM18p( SOC, SOC, Op_RegF, 18, xmm18->as_VMReg()->next(15));

reg_def XMM19 ( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg());
reg_def XMM19b( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(1));
reg_def XMM19c( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(2));
reg_def XMM19d( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(3));
reg_def XMM19e( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(4));
reg_def XMM19f( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(5));
reg_def XMM19g( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(6));
reg_def XMM19h( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(7));
reg_def XMM19i( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(8));
reg_def XMM19j( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(9));
reg_def XMM19k( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(10));
reg_def XMM19l( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(11));
reg_def XMM19m( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(12));
reg_def XMM19n( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(13));
reg_def XMM19o( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(14));
reg_def XMM19p( SOC, SOC, Op_RegF, 19, xmm19->as_VMReg()->next(15));

reg_def XMM20 ( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg());
reg_def XMM20b( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(1));
reg_def XMM20c( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(2));
reg_def XMM20d( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(3));
reg_def XMM20e( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(4));
reg_def XMM20f( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(5));
reg_def XMM20g( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(6));
reg_def XMM20h( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(7));
reg_def XMM20i( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(8));
reg_def XMM20j( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(9));
reg_def XMM20k( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(10));
reg_def XMM20l( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(11));
reg_def XMM20m( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(12));
reg_def XMM20n( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(13));
reg_def XMM20o( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(14));
reg_def XMM20p( SOC, SOC, Op_RegF, 20, xmm20->as_VMReg()->next(15));

reg_def XMM21 ( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg());
reg_def XMM21b( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(1));
reg_def XMM21c( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(2));
reg_def XMM21d( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(3));
reg_def XMM21e( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(4));
reg_def XMM21f( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(5));
reg_def XMM21g( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(6));
reg_def XMM21h( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(7));
reg_def XMM21i( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(8));
reg_def XMM21j( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(9));
reg_def XMM21k( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(10));
reg_def XMM21l( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(11));
reg_def XMM21m( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(12));
reg_def XMM21n( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(13));
reg_def XMM21o( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(14));
reg_def XMM21p( SOC, SOC, Op_RegF, 21, xmm21->as_VMReg()->next(15));

reg_def XMM22 ( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg());
reg_def XMM22b( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(1));
reg_def XMM22c( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(2));
reg_def XMM22d( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(3));
reg_def XMM22e( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(4));
reg_def XMM22f( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(5));
reg_def XMM22g( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(6));
reg_def XMM22h( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(7));
reg_def XMM22i( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(8));
reg_def XMM22j( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(9));
reg_def XMM22k( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(10));
reg_def XMM22l( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(11));
reg_def XMM22m( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(12));
reg_def XMM22n( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(13));
reg_def XMM22o( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(14));
reg_def XMM22p( SOC, SOC, Op_RegF, 22, xmm22->as_VMReg()->next(15));

reg_def XMM23 ( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg());
reg_def XMM23b( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(1));
reg_def XMM23c( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(2));
reg_def XMM23d( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(3));
reg_def XMM23e( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(4));
reg_def XMM23f( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(5));
reg_def XMM23g( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(6));
reg_def XMM23h( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(7));
reg_def XMM23i( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(8));
reg_def XMM23j( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(9));
reg_def XMM23k( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(10));
reg_def XMM23l( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(11));
reg_def XMM23m( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(12));
reg_def XMM23n( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(13));
reg_def XMM23o( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(14));
reg_def XMM23p( SOC, SOC, Op_RegF, 23, xmm23->as_VMReg()->next(15));

reg_def XMM24 ( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg());
reg_def XMM24b( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(1));
reg_def XMM24c( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(2));
reg_def XMM24d( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(3));
reg_def XMM24e( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(4));
reg_def XMM24f( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(5));
reg_def XMM24g( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(6));
reg_def XMM24h( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(7));
reg_def XMM24i( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(8));
reg_def XMM24j( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(9));
reg_def XMM24k( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(10));
reg_def XMM24l( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(11));
reg_def XMM24m( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(12));
reg_def XMM24n( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(13));
reg_def XMM24o( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(14));
reg_def XMM24p( SOC, SOC, Op_RegF, 24, xmm24->as_VMReg()->next(15));

reg_def XMM25 ( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg());
reg_def XMM25b( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(1));
reg_def XMM25c( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(2));
reg_def XMM25d( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(3));
reg_def XMM25e( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(4));
reg_def XMM25f( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(5));
reg_def XMM25g( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(6));
reg_def XMM25h( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(7));
reg_def XMM25i( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(8));
reg_def XMM25j( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(9));
reg_def XMM25k( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(10));
reg_def XMM25l( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(11));
reg_def XMM25m( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(12));
reg_def XMM25n( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(13));
reg_def XMM25o( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(14));
reg_def XMM25p( SOC, SOC, Op_RegF, 25, xmm25->as_VMReg()->next(15));

reg_def XMM26 ( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg());
reg_def XMM26b( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(1));
reg_def XMM26c( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(2));
reg_def XMM26d( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(3));
reg_def XMM26e( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(4));
reg_def XMM26f( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(5));
reg_def XMM26g( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(6));
reg_def XMM26h( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(7));
reg_def XMM26i( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(8));
reg_def XMM26j( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(9));
reg_def XMM26k( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(10));
reg_def XMM26l( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(11));
reg_def XMM26m( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(12));
reg_def XMM26n( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(13));
reg_def XMM26o( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(14));
reg_def XMM26p( SOC, SOC, Op_RegF, 26, xmm26->as_VMReg()->next(15));

reg_def XMM27 ( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg());
reg_def XMM27b( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(1));
reg_def XMM27c( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(2));
reg_def XMM27d( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(3));
reg_def XMM27e( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(4));
reg_def XMM27f( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(5));
reg_def XMM27g( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(6));
reg_def XMM27h( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(7));
reg_def XMM27i( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(8));
reg_def XMM27j( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(9));
reg_def XMM27k( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(10));
reg_def XMM27l( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(11));
reg_def XMM27m( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(12));
reg_def XMM27n( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(13));
reg_def XMM27o( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(14));
reg_def XMM27p( SOC, SOC, Op_RegF, 27, xmm27->as_VMReg()->next(15));

reg_def XMM28 ( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg());
reg_def XMM28b( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(1));
reg_def XMM28c( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(2));
reg_def XMM28d( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(3));
reg_def XMM28e( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(4));
reg_def XMM28f( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(5));
reg_def XMM28g( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(6));
reg_def XMM28h( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(7));
reg_def XMM28i( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(8));
reg_def XMM28j( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(9));
reg_def XMM28k( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(10));
reg_def XMM28l( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(11));
reg_def XMM28m( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(12));
reg_def XMM28n( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(13));
reg_def XMM28o( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(14));
reg_def XMM28p( SOC, SOC, Op_RegF, 28, xmm28->as_VMReg()->next(15));

reg_def XMM29 ( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg());
reg_def XMM29b( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(1));
reg_def XMM29c( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(2));
reg_def XMM29d( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(3));
reg_def XMM29e( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(4));
reg_def XMM29f( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(5));
reg_def XMM29g( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(6));
reg_def XMM29h( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(7));
reg_def XMM29i( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(8));
reg_def XMM29j( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(9));
reg_def XMM29k( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(10));
reg_def XMM29l( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(11));
reg_def XMM29m( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(12));
reg_def XMM29n( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(13));
reg_def XMM29o( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(14));
reg_def XMM29p( SOC, SOC, Op_RegF, 29, xmm29->as_VMReg()->next(15));

reg_def XMM30 ( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg());
reg_def XMM30b( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(1));
reg_def XMM30c( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(2));
reg_def XMM30d( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(3));
reg_def XMM30e( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(4));
reg_def XMM30f( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(5));
reg_def XMM30g( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(6));
reg_def XMM30h( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(7));
reg_def XMM30i( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(8));
reg_def XMM30j( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(9));
reg_def XMM30k( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(10));
reg_def XMM30l( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(11));
reg_def XMM30m( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(12));
reg_def XMM30n( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(13));
reg_def XMM30o( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(14));
reg_def XMM30p( SOC, SOC, Op_RegF, 30, xmm30->as_VMReg()->next(15));

reg_def XMM31 ( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg());
reg_def XMM31b( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(1));
reg_def XMM31c( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(2));
reg_def XMM31d( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(3));
reg_def XMM31e( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(4));
reg_def XMM31f( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(5));
reg_def XMM31g( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(6));
reg_def XMM31h( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(7));
reg_def XMM31i( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(8));
reg_def XMM31j( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(9));
reg_def XMM31k( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(10));
reg_def XMM31l( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(11));
reg_def XMM31m( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(12));
reg_def XMM31n( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(13));
reg_def XMM31o( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(14));
reg_def XMM31p( SOC, SOC, Op_RegF, 31, xmm31->as_VMReg()->next(15));

#endif // _LP64

#ifdef _LP64
reg_def RFLAGS(SOC, SOC, 0, 16, VMRegImpl::Bad());
#else
reg_def RFLAGS(SOC, SOC, 0, 8, VMRegImpl::Bad());
#endif // _LP64

alloc_class chunk1(XMM0,  XMM0b,  XMM0c,  XMM0d,  XMM0e,  XMM0f,  XMM0g,  XMM0h,  XMM0i,  XMM0j,  XMM0k,  XMM0l,  XMM0m,  XMM0n,  XMM0o,  XMM0p,
                   XMM1,  XMM1b,  XMM1c,  XMM1d,  XMM1e,  XMM1f,  XMM1g,  XMM1h,  XMM1i,  XMM1j,  XMM1k,  XMM1l,  XMM1m,  XMM1n,  XMM1o,  XMM1p,
                   XMM2,  XMM2b,  XMM2c,  XMM2d,  XMM2e,  XMM2f,  XMM2g,  XMM2h,  XMM2i,  XMM2j,  XMM2k,  XMM2l,  XMM2m,  XMM2n,  XMM2o,  XMM2p,
                   XMM3,  XMM3b,  XMM3c,  XMM3d,  XMM3e,  XMM3f,  XMM3g,  XMM3h,  XMM3i,  XMM3j,  XMM3k,  XMM3l,  XMM3m,  XMM3n,  XMM3o,  XMM3p,
                   XMM4,  XMM4b,  XMM4c,  XMM4d,  XMM4e,  XMM4f,  XMM4g,  XMM4h,  XMM4i,  XMM4j,  XMM4k,  XMM4l,  XMM4m,  XMM4n,  XMM4o,  XMM4p,
                   XMM5,  XMM5b,  XMM5c,  XMM5d,  XMM5e,  XMM5f,  XMM5g,  XMM5h,  XMM5i,  XMM5j,  XMM5k,  XMM5l,  XMM5m,  XMM5n,  XMM5o,  XMM5p,
                   XMM6,  XMM6b,  XMM6c,  XMM6d,  XMM6e,  XMM6f,  XMM6g,  XMM6h,  XMM6i,  XMM6j,  XMM6k,  XMM6l,  XMM6m,  XMM6n,  XMM6o,  XMM6p,
                   XMM7,  XMM7b,  XMM7c,  XMM7d,  XMM7e,  XMM7f,  XMM7g,  XMM7h,  XMM7i,  XMM7j,  XMM7k,  XMM7l,  XMM7m,  XMM7n,  XMM7o,  XMM7p
#ifdef _LP64
                  ,XMM8,  XMM8b,  XMM8c,  XMM8d,  XMM8e,  XMM8f,  XMM8g,  XMM8h,  XMM8i,  XMM8j,  XMM8k,  XMM8l,  XMM8m,  XMM8n,  XMM8o,  XMM8p,
                   XMM9,  XMM9b,  XMM9c,  XMM9d,  XMM9e,  XMM9f,  XMM9g,  XMM9h,  XMM9i,  XMM9j,  XMM9k,  XMM9l,  XMM9m,  XMM9n,  XMM9o,  XMM9p,
                   XMM10, XMM10b, XMM10c, XMM10d, XMM10e, XMM10f, XMM10g, XMM10h, XMM10i, XMM10j, XMM10k, XMM10l, XMM10m, XMM10n, XMM10o, XMM10p,
                   XMM11, XMM11b, XMM11c, XMM11d, XMM11e, XMM11f, XMM11g, XMM11h, XMM11i, XMM11j, XMM11k, XMM11l, XMM11m, XMM11n, XMM11o, XMM11p,
                   XMM12, XMM12b, XMM12c, XMM12d, XMM12e, XMM12f, XMM12g, XMM12h, XMM12i, XMM12j, XMM12k, XMM12l, XMM12m, XMM12n, XMM12o, XMM12p,
                   XMM13, XMM13b, XMM13c, XMM13d, XMM13e, XMM13f, XMM13g, XMM13h, XMM13i, XMM13j, XMM13k, XMM13l, XMM13m, XMM13n, XMM13o, XMM13p,
                   XMM14, XMM14b, XMM14c, XMM14d, XMM14e, XMM14f, XMM14g, XMM14h, XMM14i, XMM14j, XMM14k, XMM14l, XMM14m, XMM14n, XMM14o, XMM14p,
                   XMM15, XMM15b, XMM15c, XMM15d, XMM15e, XMM15f, XMM15g, XMM15h, XMM15i, XMM15j, XMM15k, XMM15l, XMM15m, XMM15n, XMM15o, XMM15p
                  ,XMM16, XMM16b, XMM16c, XMM16d, XMM16e, XMM16f, XMM16g, XMM16h, XMM16i, XMM16j, XMM16k, XMM16l, XMM16m, XMM16n, XMM16o, XMM16p,
                   XMM17, XMM17b, XMM17c, XMM17d, XMM17e, XMM17f, XMM17g, XMM17h, XMM17i, XMM17j, XMM17k, XMM17l, XMM17m, XMM17n, XMM17o, XMM17p,
                   XMM18, XMM18b, XMM18c, XMM18d, XMM18e, XMM18f, XMM18g, XMM18h, XMM18i, XMM18j, XMM18k, XMM18l, XMM18m, XMM18n, XMM18o, XMM18p,
                   XMM19, XMM19b, XMM19c, XMM19d, XMM19e, XMM19f, XMM19g, XMM19h, XMM19i, XMM19j, XMM19k, XMM19l, XMM19m, XMM19n, XMM19o, XMM19p,
                   XMM20, XMM20b, XMM20c, XMM20d, XMM20e, XMM20f, XMM20g, XMM20h, XMM20i, XMM20j, XMM20k, XMM20l, XMM20m, XMM20n, XMM20o, XMM20p,
                   XMM21, XMM21b, XMM21c, XMM21d, XMM21e, XMM21f, XMM21g, XMM21h, XMM21i, XMM21j, XMM21k, XMM21l, XMM21m, XMM21n, XMM21o, XMM21p,
                   XMM22, XMM22b, XMM22c, XMM22d, XMM22e, XMM22f, XMM22g, XMM22h, XMM22i, XMM22j, XMM22k, XMM22l, XMM22m, XMM22n, XMM22o, XMM22p,
                   XMM23, XMM23b, XMM23c, XMM23d, XMM23e, XMM23f, XMM23g, XMM23h, XMM23i, XMM23j, XMM23k, XMM23l, XMM23m, XMM23n, XMM23o, XMM23p,
                   XMM24, XMM24b, XMM24c, XMM24d, XMM24e, XMM24f, XMM24g, XMM24h, XMM24i, XMM24j, XMM24k, XMM24l, XMM24m, XMM24n, XMM24o, XMM24p,
                   XMM25, XMM25b, XMM25c, XMM25d, XMM25e, XMM25f, XMM25g, XMM25h, XMM25i, XMM25j, XMM25k, XMM25l, XMM25m, XMM25n, XMM25o, XMM25p,
                   XMM26, XMM26b, XMM26c, XMM26d, XMM26e, XMM26f, XMM26g, XMM26h, XMM26i, XMM26j, XMM26k, XMM26l, XMM26m, XMM26n, XMM26o, XMM26p,
                   XMM27, XMM27b, XMM27c, XMM27d, XMM27e, XMM27f, XMM27g, XMM27h, XMM27i, XMM27j, XMM27k, XMM27l, XMM27m, XMM27n, XMM27o, XMM27p,
                   XMM28, XMM28b, XMM28c, XMM28d, XMM28e, XMM28f, XMM28g, XMM28h, XMM28i, XMM28j, XMM28k, XMM28l, XMM28m, XMM28n, XMM28o, XMM28p,
                   XMM29, XMM29b, XMM29c, XMM29d, XMM29e, XMM29f, XMM29g, XMM29h, XMM29i, XMM29j, XMM29k, XMM29l, XMM29m, XMM29n, XMM29o, XMM29p,
                   XMM30, XMM30b, XMM30c, XMM30d, XMM30e, XMM30f, XMM30g, XMM30h, XMM30i, XMM30j, XMM30k, XMM30l, XMM30m, XMM30n, XMM30o, XMM30p,
                   XMM31, XMM31b, XMM31c, XMM31d, XMM31e, XMM31f, XMM31g, XMM31h, XMM31i, XMM31j, XMM31k, XMM31l, XMM31m, XMM31n, XMM31o, XMM31p
#endif
                      );

// flags allocation class should be last.
alloc_class chunk2(RFLAGS);

// Singleton class for condition codes
reg_class int_flags(RFLAGS);

// Class for pre evex float registers
reg_class float_reg_legacy(XMM0,
                    XMM1,
                    XMM2,
                    XMM3,
                    XMM4,
                    XMM5,
                    XMM6,
                    XMM7
#ifdef _LP64
                   ,XMM8,
                    XMM9,
                    XMM10,
                    XMM11,
                    XMM12,
                    XMM13,
                    XMM14,
                    XMM15
#endif
                    );

// Class for evex float registers
reg_class float_reg_evex(XMM0,
                    XMM1,
                    XMM2,
                    XMM3,
                    XMM4,
                    XMM5,
                    XMM6,
                    XMM7
#ifdef _LP64
                   ,XMM8,
                    XMM9,
                    XMM10,
                    XMM11,
                    XMM12,
                    XMM13,
                    XMM14,
                    XMM15,
                    XMM16,
                    XMM17,
                    XMM18,
                    XMM19,
                    XMM20,
                    XMM21,
                    XMM22,
                    XMM23,
                    XMM24,
                    XMM25,
                    XMM26,
                    XMM27,
                    XMM28,
                    XMM29,
                    XMM30,
                    XMM31
#endif
                    );

reg_class_dynamic float_reg(float_reg_evex, float_reg_legacy, %{ VM_Version::supports_evex() %} );

// Class for pre evex double registers
reg_class double_reg_legacy(XMM0,  XMM0b,
                     XMM1,  XMM1b,
                     XMM2,  XMM2b,
                     XMM3,  XMM3b,
                     XMM4,  XMM4b,
                     XMM5,  XMM5b,
                     XMM6,  XMM6b,
                     XMM7,  XMM7b
#ifdef _LP64
                    ,XMM8,  XMM8b,
                     XMM9,  XMM9b,
                     XMM10, XMM10b,
                     XMM11, XMM11b,
                     XMM12, XMM12b,
                     XMM13, XMM13b,
                     XMM14, XMM14b,
                     XMM15, XMM15b
#endif
                     );

// Class for evex double registers
reg_class double_reg_evex(XMM0,  XMM0b,
                     XMM1,  XMM1b,
                     XMM2,  XMM2b,
                     XMM3,  XMM3b,
                     XMM4,  XMM4b,
                     XMM5,  XMM5b,
                     XMM6,  XMM6b,
                     XMM7,  XMM7b
#ifdef _LP64
                    ,XMM8,  XMM8b,
                     XMM9,  XMM9b,
                     XMM10, XMM10b,
                     XMM11, XMM11b,
                     XMM12, XMM12b,
                     XMM13, XMM13b,
                     XMM14, XMM14b,
                     XMM15, XMM15b,
                     XMM16, XMM16b,
                     XMM17, XMM17b,
                     XMM18, XMM18b,
                     XMM19, XMM19b,
                     XMM20, XMM20b,
                     XMM21, XMM21b,
                     XMM22, XMM22b,
                     XMM23, XMM23b,
                     XMM24, XMM24b,
                     XMM25, XMM25b,
                     XMM26, XMM26b,
                     XMM27, XMM27b,
                     XMM28, XMM28b,
                     XMM29, XMM29b,
                     XMM30, XMM30b,
                     XMM31, XMM31b
#endif
                     );

reg_class_dynamic double_reg(double_reg_evex, double_reg_legacy, %{ VM_Version::supports_evex() %} );

// Class for pre evex 32bit vector registers
reg_class vectors_reg_legacy(XMM0,
                      XMM1,
                      XMM2,
                      XMM3,
                      XMM4,
                      XMM5,
                      XMM6,
                      XMM7
#ifdef _LP64
                     ,XMM8,
                      XMM9,
                      XMM10,
                      XMM11,
                      XMM12,
                      XMM13,
                      XMM14,
                      XMM15
#endif
                      );

// Class for evex 32bit vector registers
reg_class vectors_reg_evex(XMM0,
                      XMM1,
                      XMM2,
                      XMM3,
                      XMM4,
                      XMM5,
                      XMM6,
                      XMM7
#ifdef _LP64
                     ,XMM8,
                      XMM9,
                      XMM10,
                      XMM11,
                      XMM12,
                      XMM13,
                      XMM14,
                      XMM15,
                      XMM16,
                      XMM17,
                      XMM18,
                      XMM19,
                      XMM20,
                      XMM21,
                      XMM22,
                      XMM23,
                      XMM24,
                      XMM25,
                      XMM26,
                      XMM27,
                      XMM28,
                      XMM29,
                      XMM30,
                      XMM31
#endif
                      );

reg_class_dynamic vectors_reg(vectors_reg_evex, vectors_reg_legacy, %{ VM_Version::supports_evex() %} );

// Class for all 64bit vector registers
reg_class vectord_reg_legacy(XMM0,  XMM0b,
                      XMM1,  XMM1b,
                      XMM2,  XMM2b,
                      XMM3,  XMM3b,
                      XMM4,  XMM4b,
                      XMM5,  XMM5b,
                      XMM6,  XMM6b,
                      XMM7,  XMM7b
#ifdef _LP64
                     ,XMM8,  XMM8b,
                      XMM9,  XMM9b,
                      XMM10, XMM10b,
                      XMM11, XMM11b,
                      XMM12, XMM12b,
                      XMM13, XMM13b,
                      XMM14, XMM14b,
                      XMM15, XMM15b
#endif
                      );

// Class for all 64bit vector registers
reg_class vectord_reg_evex(XMM0,  XMM0b,
                      XMM1,  XMM1b,
                      XMM2,  XMM2b,
                      XMM3,  XMM3b,
                      XMM4,  XMM4b,
                      XMM5,  XMM5b,
                      XMM6,  XMM6b,
                      XMM7,  XMM7b
#ifdef _LP64
                     ,XMM8,  XMM8b,
                      XMM9,  XMM9b,
                      XMM10, XMM10b,
                      XMM11, XMM11b,
                      XMM12, XMM12b,
                      XMM13, XMM13b,
                      XMM14, XMM14b,
                      XMM15, XMM15b,
                      XMM16, XMM16b,
                      XMM17, XMM17b,
                      XMM18, XMM18b,
                      XMM19, XMM19b,
                      XMM20, XMM20b,
                      XMM21, XMM21b,
                      XMM22, XMM22b,
                      XMM23, XMM23b,
                      XMM24, XMM24b,
                      XMM25, XMM25b,
                      XMM26, XMM26b,
                      XMM27, XMM27b,
                      XMM28, XMM28b,
                      XMM29, XMM29b,
                      XMM30, XMM30b,
                      XMM31, XMM31b
#endif
                      );

reg_class_dynamic vectord_reg(vectord_reg_evex, vectord_reg_legacy, %{ VM_Version::supports_evex() %} );

// Class for all 128bit vector registers
reg_class vectorx_reg_legacy(XMM0,  XMM0b,  XMM0c,  XMM0d,
                      XMM1,  XMM1b,  XMM1c,  XMM1d,
                      XMM2,  XMM2b,  XMM2c,  XMM2d,
                      XMM3,  XMM3b,  XMM3c,  XMM3d,
                      XMM4,  XMM4b,  XMM4c,  XMM4d,
                      XMM5,  XMM5b,  XMM5c,  XMM5d,
                      XMM6,  XMM6b,  XMM6c,  XMM6d,
                      XMM7,  XMM7b,  XMM7c,  XMM7d
#ifdef _LP64
                     ,XMM8,  XMM8b,  XMM8c,  XMM8d,
                      XMM9,  XMM9b,  XMM9c,  XMM9d,
                      XMM10, XMM10b, XMM10c, XMM10d,
                      XMM11, XMM11b, XMM11c, XMM11d,
                      XMM12, XMM12b, XMM12c, XMM12d,
                      XMM13, XMM13b, XMM13c, XMM13d,
                      XMM14, XMM14b, XMM14c, XMM14d,
                      XMM15, XMM15b, XMM15c, XMM15d
#endif
                      );

// Class for all 128bit vector registers
reg_class vectorx_reg_evex(XMM0,  XMM0b,  XMM0c,  XMM0d,
                      XMM1,  XMM1b,  XMM1c,  XMM1d,
                      XMM2,  XMM2b,  XMM2c,  XMM2d,
                      XMM3,  XMM3b,  XMM3c,  XMM3d,
                      XMM4,  XMM4b,  XMM4c,  XMM4d,
                      XMM5,  XMM5b,  XMM5c,  XMM5d,
                      XMM6,  XMM6b,  XMM6c,  XMM6d,
                      XMM7,  XMM7b,  XMM7c,  XMM7d
#ifdef _LP64
                     ,XMM8,  XMM8b,  XMM8c,  XMM8d,
                      XMM9,  XMM9b,  XMM9c,  XMM9d,
                      XMM10, XMM10b, XMM10c, XMM10d,
                      XMM11, XMM11b, XMM11c, XMM11d,
                      XMM12, XMM12b, XMM12c, XMM12d,
                      XMM13, XMM13b, XMM13c, XMM13d,
                      XMM14, XMM14b, XMM14c, XMM14d,
                      XMM15, XMM15b, XMM15c, XMM15d,
                      XMM16, XMM16b, XMM16c, XMM16d,
                      XMM17, XMM17b, XMM17c, XMM17d,
                      XMM18, XMM18b, XMM18c, XMM18d,
                      XMM19, XMM19b, XMM19c, XMM19d,
                      XMM20, XMM20b, XMM20c, XMM20d,
                      XMM21, XMM21b, XMM21c, XMM21d,
                      XMM22, XMM22b, XMM22c, XMM22d,
                      XMM23, XMM23b, XMM23c, XMM23d,
                      XMM24, XMM24b, XMM24c, XMM24d,
                      XMM25, XMM25b, XMM25c, XMM25d,
                      XMM26, XMM26b, XMM26c, XMM26d,
                      XMM27, XMM27b, XMM27c, XMM27d,
                      XMM28, XMM28b, XMM28c, XMM28d,
                      XMM29, XMM29b, XMM29c, XMM29d,
                      XMM30, XMM30b, XMM30c, XMM30d,
                      XMM31, XMM31b, XMM31c, XMM31d
#endif
                      );

reg_class_dynamic vectorx_reg(vectorx_reg_evex, vectorx_reg_legacy, %{ VM_Version::supports_evex() %} );

// Class for all 256bit vector registers
reg_class vectory_reg_legacy(XMM0,  XMM0b,  XMM0c,  XMM0d,  XMM0e,  XMM0f,  XMM0g,  XMM0h,
                      XMM1,  XMM1b,  XMM1c,  XMM1d,  XMM1e,  XMM1f,  XMM1g,  XMM1h,
                      XMM2,  XMM2b,  XMM2c,  XMM2d,  XMM2e,  XMM2f,  XMM2g,  XMM2h,
                      XMM3,  XMM3b,  XMM3c,  XMM3d,  XMM3e,  XMM3f,  XMM3g,  XMM3h,
                      XMM4,  XMM4b,  XMM4c,  XMM4d,  XMM4e,  XMM4f,  XMM4g,  XMM4h,
                      XMM5,  XMM5b,  XMM5c,  XMM5d,  XMM5e,  XMM5f,  XMM5g,  XMM5h,
                      XMM6,  XMM6b,  XMM6c,  XMM6d,  XMM6e,  XMM6f,  XMM6g,  XMM6h,
                      XMM7,  XMM7b,  XMM7c,  XMM7d,  XMM7e,  XMM7f,  XMM7g,  XMM7h
#ifdef _LP64
                     ,XMM8,  XMM8b,  XMM8c,  XMM8d,  XMM8e,  XMM8f,  XMM8g,  XMM8h,
                      XMM9,  XMM9b,  XMM9c,  XMM9d,  XMM9e,  XMM9f,  XMM9g,  XMM9h,
                      XMM10, XMM10b, XMM10c, XMM10d, XMM10e, XMM10f, XMM10g, XMM10h,
                      XMM11, XMM11b, XMM11c, XMM11d, XMM11e, XMM11f, XMM11g, XMM11h,
                      XMM12, XMM12b, XMM12c, XMM12d, XMM12e, XMM12f, XMM12g, XMM12h,
                      XMM13, XMM13b, XMM13c, XMM13d, XMM13e, XMM13f, XMM13g, XMM13h,
                      XMM14, XMM14b, XMM14c, XMM14d, XMM14e, XMM14f, XMM14g, XMM14h,
                      XMM15, XMM15b, XMM15c, XMM15d, XMM15e, XMM15f, XMM15g, XMM15h
#endif
                      );

// Class for all 256bit vector registers
reg_class vectory_reg_evex(XMM0,  XMM0b,  XMM0c,  XMM0d,  XMM0e,  XMM0f,  XMM0g,  XMM0h,
                      XMM1,  XMM1b,  XMM1c,  XMM1d,  XMM1e,  XMM1f,  XMM1g,  XMM1h,
                      XMM2,  XMM2b,  XMM2c,  XMM2d,  XMM2e,  XMM2f,  XMM2g,  XMM2h,
                      XMM3,  XMM3b,  XMM3c,  XMM3d,  XMM3e,  XMM3f,  XMM3g,  XMM3h,
                      XMM4,  XMM4b,  XMM4c,  XMM4d,  XMM4e,  XMM4f,  XMM4g,  XMM4h,
                      XMM5,  XMM5b,  XMM5c,  XMM5d,  XMM5e,  XMM5f,  XMM5g,  XMM5h,
                      XMM6,  XMM6b,  XMM6c,  XMM6d,  XMM6e,  XMM6f,  XMM6g,  XMM6h,
                      XMM7,  XMM7b,  XMM7c,  XMM7d,  XMM7e,  XMM7f,  XMM7g,  XMM7h
#ifdef _LP64
                     ,XMM8,  XMM8b,  XMM8c,  XMM8d,  XMM8e,  XMM8f,  XMM8g,  XMM8h,
                      XMM9,  XMM9b,  XMM9c,  XMM9d,  XMM9e,  XMM9f,  XMM9g,  XMM9h,
                      XMM10, XMM10b, XMM10c, XMM10d, XMM10e, XMM10f, XMM10g, XMM10h,
                      XMM11, XMM11b, XMM11c, XMM11d, XMM11e, XMM11f, XMM11g, XMM11h,
                      XMM12, XMM12b, XMM12c, XMM12d, XMM12e, XMM12f, XMM12g, XMM12h,
                      XMM13, XMM13b, XMM13c, XMM13d, XMM13e, XMM13f, XMM13g, XMM13h,
                      XMM14, XMM14b, XMM14c, XMM14d, XMM14e, XMM14f, XMM14g, XMM14h,
                      XMM15, XMM15b, XMM15c, XMM15d, XMM15e, XMM15f, XMM15g, XMM15h,
                      XMM16, XMM16b, XMM16c, XMM16d, XMM16e, XMM16f, XMM16g, XMM16h,
                      XMM17, XMM17b, XMM17c, XMM17d, XMM17e, XMM17f, XMM17g, XMM17h,
                      XMM18, XMM18b, XMM18c, XMM18d, XMM18e, XMM18f, XMM18g, XMM18h,
                      XMM19, XMM19b, XMM19c, XMM19d, XMM19e, XMM19f, XMM19g, XMM19h,
                      XMM20, XMM20b, XMM20c, XMM20d, XMM20e, XMM20f, XMM20g, XMM20h,
                      XMM21, XMM21b, XMM21c, XMM21d, XMM21e, XMM21f, XMM21g, XMM21h,
                      XMM22, XMM22b, XMM22c, XMM22d, XMM22e, XMM22f, XMM22g, XMM22h,
                      XMM23, XMM23b, XMM23c, XMM23d, XMM23e, XMM23f, XMM23g, XMM23h,
                      XMM24, XMM24b, XMM24c, XMM24d, XMM24e, XMM24f, XMM24g, XMM24h,
                      XMM25, XMM25b, XMM25c, XMM25d, XMM25e, XMM25f, XMM25g, XMM25h,
                      XMM26, XMM26b, XMM26c, XMM26d, XMM26e, XMM26f, XMM26g, XMM26h,
                      XMM27, XMM27b, XMM27c, XMM27d, XMM27e, XMM27f, XMM27g, XMM27h,
                      XMM28, XMM28b, XMM28c, XMM28d, XMM28e, XMM28f, XMM28g, XMM28h,
                      XMM29, XMM29b, XMM29c, XMM29d, XMM29e, XMM29f, XMM29g, XMM29h,
                      XMM30, XMM30b, XMM30c, XMM30d, XMM30e, XMM30f, XMM30g, XMM30h,
                      XMM31, XMM31b, XMM31c, XMM31d, XMM31e, XMM31f, XMM31g, XMM31h
#endif
                      );

reg_class_dynamic vectory_reg(vectory_reg_evex, vectory_reg_legacy, %{ VM_Version::supports_evex() %} );

// Class for all 512bit vector registers
reg_class vectorz_reg(XMM0,  XMM0b,  XMM0c,  XMM0d,  XMM0e,  XMM0f,  XMM0g,  XMM0h,  XMM0i,  XMM0j,  XMM0k,  XMM0l,  XMM0m,  XMM0n,  XMM0o,  XMM0p,
                      XMM1,  XMM1b,  XMM1c,  XMM1d,  XMM1e,  XMM1f,  XMM1g,  XMM1h,  XMM1i,  XMM1j,  XMM1k,  XMM1l,  XMM1m,  XMM1n,  XMM1o,  XMM1p,
                      XMM2,  XMM2b,  XMM2c,  XMM2d,  XMM2e,  XMM2f,  XMM2g,  XMM2h,  XMM2i,  XMM2j,  XMM2k,  XMM2l,  XMM2m,  XMM2n,  XMM2o,  XMM2p,
                      XMM3,  XMM3b,  XMM3c,  XMM3d,  XMM3e,  XMM3f,  XMM3g,  XMM3h,  XMM3i,  XMM3j,  XMM3k,  XMM3l,  XMM3m,  XMM3n,  XMM3o,  XMM3p,
                      XMM4,  XMM4b,  XMM4c,  XMM4d,  XMM4e,  XMM4f,  XMM4g,  XMM4h,  XMM4i,  XMM4j,  XMM4k,  XMM4l,  XMM4m,  XMM4n,  XMM4o,  XMM4p,
                      XMM5,  XMM5b,  XMM5c,  XMM5d,  XMM5e,  XMM5f,  XMM5g,  XMM5h,  XMM5i,  XMM5j,  XMM5k,  XMM5l,  XMM5m,  XMM5n,  XMM5o,  XMM5p,
                      XMM6,  XMM6b,  XMM6c,  XMM6d,  XMM6e,  XMM6f,  XMM6g,  XMM6h,  XMM6i,  XMM6j,  XMM6k,  XMM6l,  XMM6m,  XMM6n,  XMM6o,  XMM6p,
                      XMM7,  XMM7b,  XMM7c,  XMM7d,  XMM7e,  XMM7f,  XMM7g,  XMM7h,  XMM7i,  XMM7j,  XMM7k,  XMM7l,  XMM7m,  XMM7n,  XMM7o,  XMM7p
#ifdef _LP64
                     ,XMM8,  XMM8b,  XMM8c,  XMM8d,  XMM8e,  XMM8f,  XMM8g,  XMM8h,  XMM8i,  XMM8j,  XMM8k,  XMM8l,  XMM8m,  XMM8n,  XMM8o,  XMM8p,
                      XMM9,  XMM9b,  XMM9c,  XMM9d,  XMM9e,  XMM9f,  XMM9g,  XMM9h,  XMM9i,  XMM9j,  XMM9k,  XMM9l,  XMM9m,  XMM9n,  XMM9o,  XMM9p,
                      XMM10, XMM10b, XMM10c, XMM10d, XMM10e, XMM10f, XMM10g, XMM10h, XMM10i, XMM10j, XMM10k, XMM10l, XMM10m, XMM10n, XMM10o, XMM10p,
                      XMM11, XMM11b, XMM11c, XMM11d, XMM11e, XMM11f, XMM11g, XMM11h, XMM11i, XMM11j, XMM11k, XMM11l, XMM11m, XMM11n, XMM11o, XMM11p,
                      XMM12, XMM12b, XMM12c, XMM12d, XMM12e, XMM12f, XMM12g, XMM12h, XMM12i, XMM12j, XMM12k, XMM12l, XMM12m, XMM12n, XMM12o, XMM12p,
                      XMM13, XMM13b, XMM13c, XMM13d, XMM13e, XMM13f, XMM13g, XMM13h, XMM13i, XMM13j, XMM13k, XMM13l, XMM13m, XMM13n, XMM13o, XMM13p,
                      XMM14, XMM14b, XMM14c, XMM14d, XMM14e, XMM14f, XMM14g, XMM14h, XMM14i, XMM14j, XMM14k, XMM14l, XMM14m, XMM14n, XMM14o, XMM14p,
                      XMM15, XMM15b, XMM15c, XMM15d, XMM15e, XMM15f, XMM15g, XMM15h, XMM15i, XMM15j, XMM15k, XMM15l, XMM15m, XMM15n, XMM15o, XMM15p
                     ,XMM16, XMM16b, XMM16c, XMM16d, XMM16e, XMM16f, XMM16g, XMM16h, XMM16i, XMM16j, XMM16k, XMM16l, XMM16m, XMM16n, XMM16o, XMM16p,
                      XMM17, XMM17b, XMM17c, XMM17d, XMM17e, XMM17f, XMM17g, XMM17h, XMM17i, XMM17j, XMM17k, XMM17l, XMM17m, XMM17n, XMM17o, XMM17p,
                      XMM18, XMM18b, XMM18c, XMM18d, XMM18e, XMM18f, XMM18g, XMM18h, XMM18i, XMM18j, XMM18k, XMM18l, XMM18m, XMM18n, XMM18o, XMM18p,
                      XMM19, XMM19b, XMM19c, XMM19d, XMM19e, XMM19f, XMM19g, XMM19h, XMM19i, XMM19j, XMM19k, XMM19l, XMM19m, XMM19n, XMM19o, XMM19p,
                      XMM20, XMM20b, XMM20c, XMM20d, XMM20e, XMM20f, XMM20g, XMM20h, XMM20i, XMM20j, XMM20k, XMM20l, XMM20m, XMM20n, XMM20o, XMM20p,
                      XMM21, XMM21b, XMM21c, XMM21d, XMM21e, XMM21f, XMM21g, XMM21h, XMM21i, XMM21j, XMM21k, XMM21l, XMM21m, XMM21n, XMM21o, XMM21p,
                      XMM22, XMM22b, XMM22c, XMM22d, XMM22e, XMM22f, XMM22g, XMM22h, XMM22i, XMM22j, XMM22k, XMM22l, XMM22m, XMM22n, XMM22o, XMM22p,
                      XMM23, XMM23b, XMM23c, XMM23d, XMM23e, XMM23f, XMM23g, XMM23h, XMM23i, XMM23j, XMM23k, XMM23l, XMM23m, XMM23n, XMM23o, XMM23p,
                      XMM24, XMM24b, XMM24c, XMM24d, XMM24e, XMM24f, XMM24g, XMM24h, XMM24i, XMM24j, XMM24k, XMM24l, XMM24m, XMM24n, XMM24o, XMM24p,
                      XMM25, XMM25b, XMM25c, XMM25d, XMM25e, XMM25f, XMM25g, XMM25h, XMM25i, XMM25j, XMM25k, XMM25l, XMM25m, XMM25n, XMM25o, XMM25p,
                      XMM26, XMM26b, XMM26c, XMM26d, XMM26e, XMM26f, XMM26g, XMM26h, XMM26i, XMM26j, XMM26k, XMM26l, XMM26m, XMM26n, XMM26o, XMM26p,
                      XMM27, XMM27b, XMM27c, XMM27d, XMM27e, XMM27f, XMM27g, XMM27h, XMM27i, XMM27j, XMM27k, XMM27l, XMM27m, XMM27n, XMM27o, XMM27p,
                      XMM28, XMM28b, XMM28c, XMM28d, XMM28e, XMM28f, XMM28g, XMM28h, XMM28i, XMM28j, XMM28k, XMM28l, XMM28m, XMM28n, XMM28o, XMM28p,
                      XMM29, XMM29b, XMM29c, XMM29d, XMM29e, XMM29f, XMM29g, XMM29h, XMM29i, XMM29j, XMM29k, XMM29l, XMM29m, XMM29n, XMM29o, XMM29p,
                      XMM30, XMM30b, XMM30c, XMM30d, XMM30e, XMM30f, XMM30g, XMM30h, XMM30i, XMM30j, XMM30k, XMM30l, XMM30m, XMM30n, XMM30o, XMM30p,
                      XMM31, XMM31b, XMM31c, XMM31d, XMM31e, XMM31f, XMM31g, XMM31h, XMM31i, XMM31j, XMM31k, XMM31l, XMM31m, XMM31n, XMM31o, XMM31p
#endif
                      );

%}


//----------SOURCE BLOCK-------------------------------------------------------
// This is a block of C++ code which provides values, functions, and
// definitions necessary in the rest of the architecture description

source_hpp %{
// Header information of the source block.
// Method declarations/definitions which are used outside
// the ad-scope can conveniently be defined here.
//
// To keep related declarations/definitions/uses close together,
// we switch between source %{ }% and source_hpp %{ }% freely as needed.

class NativeJump;

class CallStubImpl {

  //--------------------------------------------------------------
  //---<  Used for optimization in Compile::shorten_branches  >---
  //--------------------------------------------------------------

 public:
  // Size of call trampoline stub.
  static uint size_call_trampoline() {
    return 0; // no call trampolines on this platform
  }

  // number of relocations needed by a call trampoline stub
  static uint reloc_call_trampoline() {
    return 0; // no call trampolines on this platform
  }
};

class HandlerImpl {

 public:

  static int emit_exception_handler(CodeBuffer &cbuf);
  static int emit_deopt_handler(CodeBuffer& cbuf);

  static uint size_exception_handler() {
    // NativeCall instruction size is the same as NativeJump.
    // exception handler starts out as jump and can be patched to
    // a call be deoptimization.  (4932387)
    // Note that this value is also credited (in output.cpp) to
    // the size of the code section.
    return NativeJump::instruction_size;
  }

#ifdef _LP64
  static uint size_deopt_handler() {
    // three 5 byte instructions
    return 15;
  }
#else
  static uint size_deopt_handler() {
    // NativeCall instruction size is the same as NativeJump.
    // exception handler starts out as jump and can be patched to
    // a call be deoptimization.  (4932387)
    // Note that this value is also credited (in output.cpp) to
    // the size of the code section.
    return 5 + NativeJump::instruction_size; // pushl(); jmp;
  }
#endif
};

%} // end source_hpp

source %{

#include "opto/addnode.hpp"

// Emit exception handler code.
// Stuff framesize into a register and call a VM stub routine.
int HandlerImpl::emit_exception_handler(CodeBuffer& cbuf) {

  // Note that the code buffer's insts_mark is always relative to insts.
  // That's why we must use the macroassembler to generate a handler.
  MacroAssembler _masm(&cbuf);
  address base = __ start_a_stub(size_exception_handler());
  if (base == NULL) {
    ciEnv::current()->record_failure("CodeCache is full");
    return 0;  // CodeBuffer::expand failed
  }
  int offset = __ offset();
  __ jump(RuntimeAddress(OptoRuntime::exception_blob()->entry_point()));
  assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
  __ end_a_stub();
  return offset;
}

// Emit deopt handler code.
int HandlerImpl::emit_deopt_handler(CodeBuffer& cbuf) {

  // Note that the code buffer's insts_mark is always relative to insts.
  // That's why we must use the macroassembler to generate a handler.
  MacroAssembler _masm(&cbuf);
  address base = __ start_a_stub(size_deopt_handler());
  if (base == NULL) {
    ciEnv::current()->record_failure("CodeCache is full");
    return 0;  // CodeBuffer::expand failed
  }
  int offset = __ offset();

#ifdef _LP64
  address the_pc = (address) __ pc();
  Label next;
  // push a "the_pc" on the stack without destroying any registers
  // as they all may be live.

  // push address of "next"
  __ call(next, relocInfo::none); // reloc none is fine since it is a disp32
  __ bind(next);
  // adjust it so it matches "the_pc"
  __ subptr(Address(rsp, 0), __ offset() - offset);
#else
  InternalAddress here(__ pc());
  __ pushptr(here.addr());
#endif

  __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
  assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
  __ end_a_stub();
  return offset;
}


//=============================================================================

  // Float masks come from different places depending on platform.
#ifdef _LP64
  static address float_signmask()  { return StubRoutines::x86::float_sign_mask(); }
  static address float_signflip()  { return StubRoutines::x86::float_sign_flip(); }
  static address double_signmask() { return StubRoutines::x86::double_sign_mask(); }
  static address double_signflip() { return StubRoutines::x86::double_sign_flip(); }
#else
  static address float_signmask()  { return (address)float_signmask_pool; }
  static address float_signflip()  { return (address)float_signflip_pool; }
  static address double_signmask() { return (address)double_signmask_pool; }
  static address double_signflip() { return (address)double_signflip_pool; }
#endif


const bool Matcher::match_rule_supported(int opcode) {
  if (!has_match_rule(opcode))
    return false;

  bool ret_value = true;
  switch (opcode) {
    case Op_PopCountI:
    case Op_PopCountL:
      if (!UsePopCountInstruction)
        ret_value = false;
      break;
    case Op_PopCountVI:
      if (!UsePopCountInstruction || !VM_Version::supports_vpopcntdq())
        ret_value = false;
      break;
    case Op_MulVI:
      if ((UseSSE < 4) && (UseAVX < 1)) // only with SSE4_1 or AVX
        ret_value = false;
      break;
    case Op_MulVL:
    case Op_MulReductionVL:
      if (VM_Version::supports_avx512dq() == false)
        ret_value = false;
      break;
    case Op_AddReductionVL:
      if (UseAVX < 3) // only EVEX : vector connectivity becomes an issue here
        ret_value = false;
      break;
    case Op_AddReductionVI:
      if (UseSSE < 3) // requires at least SSE3
        ret_value = false;
      break;
    case Op_MulReductionVI:
      if (UseSSE < 4) // requires at least SSE4
        ret_value = false;
      break;
    case Op_AddReductionVF:
    case Op_AddReductionVD:
    case Op_MulReductionVF:
    case Op_MulReductionVD:
      if (UseSSE < 1) // requires at least SSE
        ret_value = false;
      break;
    case Op_SqrtVD:
    case Op_SqrtVF:
      if (UseAVX < 1) // enabled for AVX only
        ret_value = false;
      break;
    case Op_CompareAndSwapL:
#ifdef _LP64
    case Op_CompareAndSwapP:
#endif
      if (!VM_Version::supports_cx8())
        ret_value = false;
      break;
    case Op_CMoveVF:
    case Op_CMoveVD:
      if (UseAVX < 1 || UseAVX > 2)
        ret_value = false;
      break;
    case Op_StrIndexOf:
      if (!UseSSE42Intrinsics)
        ret_value = false;
      break;
    case Op_StrIndexOfChar:
      if (!UseSSE42Intrinsics)
        ret_value = false;
      break;
    case Op_OnSpinWait:
      if (VM_Version::supports_on_spin_wait() == false)
        ret_value = false;
      break;
  }

  return ret_value;  // Per default match rules are supported.
}

const bool Matcher::match_rule_supported_vector(int opcode, int vlen) {
  // identify extra cases that we might want to provide match rules for
  // e.g. Op_ vector nodes and other intrinsics while guarding with vlen
  bool ret_value = match_rule_supported(opcode);
  if (ret_value) {
    switch (opcode) {
      case Op_AddVB:
      case Op_SubVB:
        if ((vlen == 64) && (VM_Version::supports_avx512bw() == false))
          ret_value = false;
        break;
      case Op_URShiftVS:
      case Op_RShiftVS:
      case Op_LShiftVS:
      case Op_MulVS:
      case Op_AddVS:
      case Op_SubVS:
        if ((vlen == 32) && (VM_Version::supports_avx512bw() == false))
          ret_value = false;
        break;
      case Op_CMoveVF:
        if (vlen != 8)
          ret_value  = false;
      case Op_CMoveVD:
        if (vlen != 4)
          ret_value  = false;
        break;
    }
  }

  return ret_value;  // Per default match rules are supported.
}

const bool Matcher::has_predicated_vectors(void) {
  bool ret_value = false;
  if (UseAVX > 2) {
    ret_value = VM_Version::supports_avx512vl();
  }

  return ret_value;
}

const int Matcher::float_pressure(int default_pressure_threshold) {
  int float_pressure_threshold = default_pressure_threshold;
#ifdef _LP64
  if (UseAVX > 2) {
    // Increase pressure threshold on machines with AVX3 which have
    // 2x more XMM registers.
    float_pressure_threshold = default_pressure_threshold * 2;
  }
#endif
  return float_pressure_threshold;
}

// Max vector size in bytes. 0 if not supported.
const int Matcher::vector_width_in_bytes(BasicType bt) {
  assert(is_java_primitive(bt), "only primitive type vectors");
  if (UseSSE < 2) return 0;
  // SSE2 supports 128bit vectors for all types.
  // AVX2 supports 256bit vectors for all types.
  // AVX2/EVEX supports 512bit vectors for all types.
  int size = (UseAVX > 1) ? (1 << UseAVX) * 8 : 16;
  // AVX1 supports 256bit vectors only for FLOAT and DOUBLE.
  if (UseAVX > 0 && (bt == T_FLOAT || bt == T_DOUBLE))
    size = (UseAVX > 2) ? 64 : 32;
  // Use flag to limit vector size.
  size = MIN2(size,(int)MaxVectorSize);
  // Minimum 2 values in vector (or 4 for bytes).
  switch (bt) {
  case T_DOUBLE:
  case T_LONG:
    if (size < 16) return 0;
    break;
  case T_FLOAT:
  case T_INT:
    if (size < 8) return 0;
    break;
  case T_BOOLEAN:
    if (size < 4) return 0;
    break;
  case T_CHAR:
    if (size < 4) return 0;
    break;
  case T_BYTE:
    if (size < 4) return 0;
    break;
  case T_SHORT:
    if (size < 4) return 0;
    break;
  default:
    ShouldNotReachHere();
  }
  return size;
}

// Limits on vector size (number of elements) loaded into vector.
const int Matcher::max_vector_size(const BasicType bt) {
  return vector_width_in_bytes(bt)/type2aelembytes(bt);
}
const int Matcher::min_vector_size(const BasicType bt) {
  int max_size = max_vector_size(bt);
  // Min size which can be loaded into vector is 4 bytes.
  int size = (type2aelembytes(bt) == 1) ? 4 : 2;
  return MIN2(size,max_size);
}

// Vector ideal reg corresponding to specidied size in bytes
const uint Matcher::vector_ideal_reg(int size) {
  assert(MaxVectorSize >= size, "");
  switch(size) {
    case  4: return Op_VecS;
    case  8: return Op_VecD;
    case 16: return Op_VecX;
    case 32: return Op_VecY;
    case 64: return Op_VecZ;
  }
  ShouldNotReachHere();
  return 0;
}

// Only lowest bits of xmm reg are used for vector shift count.
const uint Matcher::vector_shift_count_ideal_reg(int size) {
  return Op_VecS;
}

// x86 supports misaligned vectors store/load.
const bool Matcher::misaligned_vectors_ok() {
  return !AlignVector; // can be changed by flag
}

// x86 AES instructions are compatible with SunJCE expanded
// keys, hence we do not need to pass the original key to stubs
const bool Matcher::pass_original_key_for_aes() {
  return false;
}


const bool Matcher::convi2l_type_required = true;

// Check for shift by small constant as well
static bool clone_shift(Node* shift, Matcher* matcher, Matcher::MStack& mstack, VectorSet& address_visited) {
  if (shift->Opcode() == Op_LShiftX && shift->in(2)->is_Con() &&
      shift->in(2)->get_int() <= 3 &&
      // Are there other uses besides address expressions?
      !matcher->is_visited(shift)) {
    address_visited.set(shift->_idx); // Flag as address_visited
    mstack.push(shift->in(2), Matcher::Visit);
    Node *conv = shift->in(1);
#ifdef _LP64
    // Allow Matcher to match the rule which bypass
    // ConvI2L operation for an array index on LP64
    // if the index value is positive.
    if (conv->Opcode() == Op_ConvI2L &&
        conv->as_Type()->type()->is_long()->_lo >= 0 &&
        // Are there other uses besides address expressions?
        !matcher->is_visited(conv)) {
      address_visited.set(conv->_idx); // Flag as address_visited
      mstack.push(conv->in(1), Matcher::Pre_Visit);
    } else
#endif
      mstack.push(conv, Matcher::Pre_Visit);
    return true;
  }
  return false;
}

// Should the Matcher clone shifts on addressing modes, expecting them
// to be subsumed into complex addressing expressions or compute them
// into registers?
bool Matcher::clone_address_expressions(AddPNode* m, Matcher::MStack& mstack, VectorSet& address_visited) {
  Node *off = m->in(AddPNode::Offset);
  if (off->is_Con()) {
    address_visited.test_set(m->_idx); // Flag as address_visited
    Node *adr = m->in(AddPNode::Address);

    // Intel can handle 2 adds in addressing mode
    // AtomicAdd is not an addressing expression.
    // Cheap to find it by looking for screwy base.
    if (adr->is_AddP() &&
        !adr->in(AddPNode::Base)->is_top() &&
        // Are there other uses besides address expressions?
        !is_visited(adr)) {
      address_visited.set(adr->_idx); // Flag as address_visited
      Node *shift = adr->in(AddPNode::Offset);
      if (!clone_shift(shift, this, mstack, address_visited)) {
        mstack.push(shift, Pre_Visit);
      }
      mstack.push(adr->in(AddPNode::Address), Pre_Visit);
      mstack.push(adr->in(AddPNode::Base), Pre_Visit);
    } else {
      mstack.push(adr, Pre_Visit);
    }

    // Clone X+offset as it also folds into most addressing expressions
    mstack.push(off, Visit);
    mstack.push(m->in(AddPNode::Base), Pre_Visit);
    return true;
  } else if (clone_shift(off, this, mstack, address_visited)) {
    address_visited.test_set(m->_idx); // Flag as address_visited
    mstack.push(m->in(AddPNode::Address), Pre_Visit);
    mstack.push(m->in(AddPNode::Base), Pre_Visit);
    return true;
  }
  return false;
}

void Compile::reshape_address(AddPNode* addp) {
}

// Helper methods for MachSpillCopyNode::implementation().
static int vec_mov_helper(CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo,
                          int src_hi, int dst_hi, uint ireg, outputStream* st) {
  // In 64-bit VM size calculation is very complex. Emitting instructions
  // into scratch buffer is used to get size in 64-bit VM.
  LP64_ONLY( assert(!do_size, "this method calculates size only for 32-bit VM"); )
  assert(ireg == Op_VecS || // 32bit vector
         (src_lo & 1) == 0 && (src_lo + 1) == src_hi &&
         (dst_lo & 1) == 0 && (dst_lo + 1) == dst_hi,
         "no non-adjacent vector moves" );
  if (cbuf) {
    MacroAssembler _masm(cbuf);
    int offset = __ offset();
    switch (ireg) {
    case Op_VecS: // copy whole register
    case Op_VecD:
    case Op_VecX:
      __ movdqu(as_XMMRegister(Matcher::_regEncode[dst_lo]), as_XMMRegister(Matcher::_regEncode[src_lo]));
      break;
    case Op_VecY:
      __ vmovdqu(as_XMMRegister(Matcher::_regEncode[dst_lo]), as_XMMRegister(Matcher::_regEncode[src_lo]));
      break;
    case Op_VecZ:
      __ evmovdquq(as_XMMRegister(Matcher::_regEncode[dst_lo]), as_XMMRegister(Matcher::_regEncode[src_lo]), 2);
      break;
    default:
      ShouldNotReachHere();
    }
    int size = __ offset() - offset;
#ifdef ASSERT
    // VEX_2bytes prefix is used if UseAVX > 0, so it takes the same 2 bytes as SIMD prefix.
    assert(!do_size || size == 4, "incorrect size calculattion");
#endif
    return size;
#ifndef PRODUCT
  } else if (!do_size) {
    switch (ireg) {
    case Op_VecS:
    case Op_VecD:
    case Op_VecX:
      st->print("movdqu  %s,%s\t# spill",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
      break;
    case Op_VecY:
    case Op_VecZ:
      st->print("vmovdqu %s,%s\t# spill",Matcher::regName[dst_lo],Matcher::regName[src_lo]);
      break;
    default:
      ShouldNotReachHere();
    }
#endif
  }
  // VEX_2bytes prefix is used if UseAVX > 0, and it takes the same 2 bytes as SIMD prefix.
  return (UseAVX > 2) ? 6 : 4;
}

static int vec_spill_helper(CodeBuffer *cbuf, bool do_size, bool is_load,
                            int stack_offset, int reg, uint ireg, outputStream* st) {
  // In 64-bit VM size calculation is very complex. Emitting instructions
  // into scratch buffer is used to get size in 64-bit VM.
  LP64_ONLY( assert(!do_size, "this method calculates size only for 32-bit VM"); )
  if (cbuf) {
    MacroAssembler _masm(cbuf);
    int offset = __ offset();
    if (is_load) {
      switch (ireg) {
      case Op_VecS:
        __ movdl(as_XMMRegister(Matcher::_regEncode[reg]), Address(rsp, stack_offset));
        break;
      case Op_VecD:
        __ movq(as_XMMRegister(Matcher::_regEncode[reg]), Address(rsp, stack_offset));
        break;
      case Op_VecX:
        __ movdqu(as_XMMRegister(Matcher::_regEncode[reg]), Address(rsp, stack_offset));
        break;
      case Op_VecY:
        __ vmovdqu(as_XMMRegister(Matcher::_regEncode[reg]), Address(rsp, stack_offset));
        break;
      case Op_VecZ:
        __ evmovdquq(as_XMMRegister(Matcher::_regEncode[reg]), Address(rsp, stack_offset), 2);
        break;
      default:
        ShouldNotReachHere();
      }
    } else { // store
      switch (ireg) {
      case Op_VecS:
        __ movdl(Address(rsp, stack_offset), as_XMMRegister(Matcher::_regEncode[reg]));
        break;
      case Op_VecD:
        __ movq(Address(rsp, stack_offset), as_XMMRegister(Matcher::_regEncode[reg]));
        break;
      case Op_VecX:
        __ movdqu(Address(rsp, stack_offset), as_XMMRegister(Matcher::_regEncode[reg]));
        break;
      case Op_VecY:
        __ vmovdqu(Address(rsp, stack_offset), as_XMMRegister(Matcher::_regEncode[reg]));
        break;
      case Op_VecZ:
        __ evmovdquq(Address(rsp, stack_offset), as_XMMRegister(Matcher::_regEncode[reg]), 2);
        break;
      default:
        ShouldNotReachHere();
      }
    }
    int size = __ offset() - offset;
#ifdef ASSERT
    int offset_size = (stack_offset == 0) ? 0 : ((stack_offset < 0x80) ? 1 : (UseAVX > 2) ? 6 : 4);
    // VEX_2bytes prefix is used if UseAVX > 0, so it takes the same 2 bytes as SIMD prefix.
    assert(!do_size || size == (5+offset_size), "incorrect size calculattion");
#endif
    return size;
#ifndef PRODUCT
  } else if (!do_size) {
    if (is_load) {
      switch (ireg) {
      case Op_VecS:
        st->print("movd    %s,[rsp + %d]\t# spill", Matcher::regName[reg], stack_offset);
        break;
      case Op_VecD:
        st->print("movq    %s,[rsp + %d]\t# spill", Matcher::regName[reg], stack_offset);
        break;
       case Op_VecX:
        st->print("movdqu  %s,[rsp + %d]\t# spill", Matcher::regName[reg], stack_offset);
        break;
      case Op_VecY:
      case Op_VecZ:
        st->print("vmovdqu %s,[rsp + %d]\t# spill", Matcher::regName[reg], stack_offset);
        break;
      default:
        ShouldNotReachHere();
      }
    } else { // store
      switch (ireg) {
      case Op_VecS:
        st->print("movd    [rsp + %d],%s\t# spill", stack_offset, Matcher::regName[reg]);
        break;
      case Op_VecD:
        st->print("movq    [rsp + %d],%s\t# spill", stack_offset, Matcher::regName[reg]);
        break;
       case Op_VecX:
        st->print("movdqu  [rsp + %d],%s\t# spill", stack_offset, Matcher::regName[reg]);
        break;
      case Op_VecY:
      case Op_VecZ:
        st->print("vmovdqu [rsp + %d],%s\t# spill", stack_offset, Matcher::regName[reg]);
        break;
      default:
        ShouldNotReachHere();
      }
    }
#endif
  }
  bool is_single_byte = false;
  int vec_len = 0;
  if ((UseAVX > 2) && (stack_offset != 0)) {
    int tuple_type = Assembler::EVEX_FVM;
    int input_size = Assembler::EVEX_32bit;
    switch (ireg) {
    case Op_VecS:
      tuple_type = Assembler::EVEX_T1S;
      break;
    case Op_VecD:
      tuple_type = Assembler::EVEX_T1S;
      input_size = Assembler::EVEX_64bit;
      break;
    case Op_VecX:
      break;
    case Op_VecY:
      vec_len = 1;
      break;
    case Op_VecZ:
      vec_len = 2;
      break;
    }
    is_single_byte = Assembler::query_compressed_disp_byte(stack_offset, true, vec_len, tuple_type, input_size, 0);
  }
  int offset_size = 0;
  int size = 5;
  if (UseAVX > 2 ) {
    if (VM_Version::supports_avx512novl() && (vec_len == 2)) {
      offset_size = (stack_offset == 0) ? 0 : ((is_single_byte) ? 1 : 4);
      size += 2; // Need an additional two bytes for EVEX encoding
    } else if (VM_Version::supports_avx512novl() && (vec_len < 2)) {
      offset_size = (stack_offset == 0) ? 0 : ((stack_offset <= 127) ? 1 : 4);
    } else {
      offset_size = (stack_offset == 0) ? 0 : ((is_single_byte) ? 1 : 4);
      size += 2; // Need an additional two bytes for EVEX encodding
    }
  } else {
    offset_size = (stack_offset == 0) ? 0 : ((stack_offset <= 127) ? 1 : 4);
  }
  // VEX_2bytes prefix is used if UseAVX > 0, so it takes the same 2 bytes as SIMD prefix.
  return size+offset_size;
}

static inline jint replicate4_imm(int con, int width) {
  // Load a constant of "width" (in bytes) and replicate it to fill 32bit.
  assert(width == 1 || width == 2, "only byte or short types here");
  int bit_width = width * 8;
  jint val = con;
  val &= (1 << bit_width) - 1;  // mask off sign bits
  while(bit_width < 32) {
    val |= (val << bit_width);
    bit_width <<= 1;
  }
  return val;
}

static inline jlong replicate8_imm(int con, int width) {
  // Load a constant of "width" (in bytes) and replicate it to fill 64bit.
  assert(width == 1 || width == 2 || width == 4, "only byte, short or int types here");
  int bit_width = width * 8;
  jlong val = con;
  val &= (((jlong) 1) << bit_width) - 1;  // mask off sign bits
  while(bit_width < 64) {
    val |= (val << bit_width);
    bit_width <<= 1;
  }
  return val;
}

#ifndef PRODUCT
  void MachNopNode::format(PhaseRegAlloc*, outputStream* st) const {
    st->print("nop \t# %d bytes pad for loops and calls", _count);
  }
#endif

  void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc*) const {
    MacroAssembler _masm(&cbuf);
    __ nop(_count);
  }

  uint MachNopNode::size(PhaseRegAlloc*) const {
    return _count;
  }

#ifndef PRODUCT
  void MachBreakpointNode::format(PhaseRegAlloc*, outputStream* st) const {
    st->print("# breakpoint");
  }
#endif

  void MachBreakpointNode::emit(CodeBuffer &cbuf, PhaseRegAlloc* ra_) const {
    MacroAssembler _masm(&cbuf);
    __ int3();
  }

  uint MachBreakpointNode::size(PhaseRegAlloc* ra_) const {
    return MachNode::size(ra_);
  }

%}

encode %{

  enc_class call_epilog %{
    if (VerifyStackAtCalls) {
      // Check that stack depth is unchanged: find majik cookie on stack
      int framesize = ra_->reg2offset_unchecked(OptoReg::add(ra_->_matcher._old_SP, -3*VMRegImpl::slots_per_word));
      MacroAssembler _masm(&cbuf);
      Label L;
      __ cmpptr(Address(rsp, framesize), (int32_t)0xbadb100d);
      __ jccb(Assembler::equal, L);
      // Die if stack mismatch
      __ int3();
      __ bind(L);
    }
  %}

%}


//----------OPERANDS-----------------------------------------------------------
// Operand definitions must precede instruction definitions for correct parsing
// in the ADLC because operands constitute user defined types which are used in
// instruction definitions.

// This one generically applies only for evex, so only one version
operand vecZ() %{
  constraint(ALLOC_IN_RC(vectorz_reg));
  match(VecZ);

  format %{ %}
  interface(REG_INTER);
%}

// Comparison Code for FP conditional move
operand cmpOp_vcmppd() %{
  match(Bool);

  predicate(n->as_Bool()->_test._test != BoolTest::overflow &&
            n->as_Bool()->_test._test != BoolTest::no_overflow);
  format %{ "" %}
  interface(COND_INTER) %{
    equal        (0x0, "eq");
    less         (0x1, "lt");
    less_equal   (0x2, "le");
    not_equal    (0xC, "ne");
    greater_equal(0xD, "ge");
    greater      (0xE, "gt");
    //TODO cannot compile (adlc breaks) without two next lines with error:
    // x86_64.ad(13987) Syntax Error: :In operand cmpOp_vcmppd: Do not support this encode constant: ' %{
    // equal' for overflow.
    overflow     (0x20, "o");  // not really supported by the instruction
    no_overflow  (0x21, "no"); // not really supported by the instruction
  %}
%}


// INSTRUCTIONS -- Platform independent definitions (same for 32- and 64-bit)

// ============================================================================

instruct ShouldNotReachHere() %{
  match(Halt);
  format %{ "ud2\t# ShouldNotReachHere" %}
  ins_encode %{
    __ ud2();
  %}
  ins_pipe(pipe_slow);
%}

// =================================EVEX special===============================

instruct setMask(rRegI dst, rRegI src) %{
  predicate(Matcher::has_predicated_vectors());
  match(Set dst (SetVectMaskI  src));
  effect(TEMP dst);
  format %{ "setvectmask   $dst, $src" %}
  ins_encode %{
    __ setvectmask($dst$$Register, $src$$Register);
  %}
  ins_pipe(pipe_slow);
%}

// ============================================================================

instruct addF_reg(regF dst, regF src) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (AddF dst src));

  format %{ "addss   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ addss($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct addF_mem(regF dst, memory src) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (AddF dst (LoadF src)));

  format %{ "addss   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ addss($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct addF_imm(regF dst, immF con) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (AddF dst con));
  format %{ "addss   $dst, [$constantaddress]\t# load from constant table: float=$con" %}
  ins_cost(150);
  ins_encode %{
    __ addss($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct addF_reg_reg(regF dst, regF src1, regF src2) %{
  predicate(UseAVX > 0);
  match(Set dst (AddF src1 src2));

  format %{ "vaddss  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vaddss($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct addF_reg_mem(regF dst, regF src1, memory src2) %{
  predicate(UseAVX > 0);
  match(Set dst (AddF src1 (LoadF src2)));

  format %{ "vaddss  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vaddss($dst$$XMMRegister, $src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct addF_reg_imm(regF dst, regF src, immF con) %{
  predicate(UseAVX > 0);
  match(Set dst (AddF src con));

  format %{ "vaddss  $dst, $src, [$constantaddress]\t# load from constant table: float=$con" %}
  ins_cost(150);
  ins_encode %{
    __ vaddss($dst$$XMMRegister, $src$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct addD_reg(regD dst, regD src) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (AddD dst src));

  format %{ "addsd   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ addsd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct addD_mem(regD dst, memory src) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (AddD dst (LoadD src)));

  format %{ "addsd   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ addsd($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct addD_imm(regD dst, immD con) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (AddD dst con));
  format %{ "addsd   $dst, [$constantaddress]\t# load from constant table: double=$con" %}
  ins_cost(150);
  ins_encode %{
    __ addsd($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct addD_reg_reg(regD dst, regD src1, regD src2) %{
  predicate(UseAVX > 0);
  match(Set dst (AddD src1 src2));

  format %{ "vaddsd  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vaddsd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct addD_reg_mem(regD dst, regD src1, memory src2) %{
  predicate(UseAVX > 0);
  match(Set dst (AddD src1 (LoadD src2)));

  format %{ "vaddsd  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vaddsd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct addD_reg_imm(regD dst, regD src, immD con) %{
  predicate(UseAVX > 0);
  match(Set dst (AddD src con));

  format %{ "vaddsd  $dst, $src, [$constantaddress]\t# load from constant table: double=$con" %}
  ins_cost(150);
  ins_encode %{
    __ vaddsd($dst$$XMMRegister, $src$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct subF_reg(regF dst, regF src) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (SubF dst src));

  format %{ "subss   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ subss($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct subF_mem(regF dst, memory src) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (SubF dst (LoadF src)));

  format %{ "subss   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ subss($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct subF_imm(regF dst, immF con) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (SubF dst con));
  format %{ "subss   $dst, [$constantaddress]\t# load from constant table: float=$con" %}
  ins_cost(150);
  ins_encode %{
    __ subss($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct subF_reg_reg(regF dst, regF src1, regF src2) %{
  predicate(UseAVX > 0);
  match(Set dst (SubF src1 src2));

  format %{ "vsubss  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vsubss($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct subF_reg_mem(regF dst, regF src1, memory src2) %{
  predicate(UseAVX > 0);
  match(Set dst (SubF src1 (LoadF src2)));

  format %{ "vsubss  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vsubss($dst$$XMMRegister, $src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct subF_reg_imm(regF dst, regF src, immF con) %{
  predicate(UseAVX > 0);
  match(Set dst (SubF src con));

  format %{ "vsubss  $dst, $src, [$constantaddress]\t# load from constant table: float=$con" %}
  ins_cost(150);
  ins_encode %{
    __ vsubss($dst$$XMMRegister, $src$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct subD_reg(regD dst, regD src) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (SubD dst src));

  format %{ "subsd   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ subsd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct subD_mem(regD dst, memory src) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (SubD dst (LoadD src)));

  format %{ "subsd   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ subsd($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct subD_imm(regD dst, immD con) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (SubD dst con));
  format %{ "subsd   $dst, [$constantaddress]\t# load from constant table: double=$con" %}
  ins_cost(150);
  ins_encode %{
    __ subsd($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct subD_reg_reg(regD dst, regD src1, regD src2) %{
  predicate(UseAVX > 0);
  match(Set dst (SubD src1 src2));

  format %{ "vsubsd  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vsubsd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct subD_reg_mem(regD dst, regD src1, memory src2) %{
  predicate(UseAVX > 0);
  match(Set dst (SubD src1 (LoadD src2)));

  format %{ "vsubsd  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vsubsd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct subD_reg_imm(regD dst, regD src, immD con) %{
  predicate(UseAVX > 0);
  match(Set dst (SubD src con));

  format %{ "vsubsd  $dst, $src, [$constantaddress]\t# load from constant table: double=$con" %}
  ins_cost(150);
  ins_encode %{
    __ vsubsd($dst$$XMMRegister, $src$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct mulF_reg(regF dst, regF src) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (MulF dst src));

  format %{ "mulss   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ mulss($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct mulF_mem(regF dst, memory src) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (MulF dst (LoadF src)));

  format %{ "mulss   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ mulss($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct mulF_imm(regF dst, immF con) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (MulF dst con));
  format %{ "mulss   $dst, [$constantaddress]\t# load from constant table: float=$con" %}
  ins_cost(150);
  ins_encode %{
    __ mulss($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct mulF_reg_reg(regF dst, regF src1, regF src2) %{
  predicate(UseAVX > 0);
  match(Set dst (MulF src1 src2));

  format %{ "vmulss  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vmulss($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct mulF_reg_mem(regF dst, regF src1, memory src2) %{
  predicate(UseAVX > 0);
  match(Set dst (MulF src1 (LoadF src2)));

  format %{ "vmulss  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vmulss($dst$$XMMRegister, $src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct mulF_reg_imm(regF dst, regF src, immF con) %{
  predicate(UseAVX > 0);
  match(Set dst (MulF src con));

  format %{ "vmulss  $dst, $src, [$constantaddress]\t# load from constant table: float=$con" %}
  ins_cost(150);
  ins_encode %{
    __ vmulss($dst$$XMMRegister, $src$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct mulD_reg(regD dst, regD src) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (MulD dst src));

  format %{ "mulsd   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ mulsd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct mulD_mem(regD dst, memory src) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (MulD dst (LoadD src)));

  format %{ "mulsd   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ mulsd($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct mulD_imm(regD dst, immD con) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (MulD dst con));
  format %{ "mulsd   $dst, [$constantaddress]\t# load from constant table: double=$con" %}
  ins_cost(150);
  ins_encode %{
    __ mulsd($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct mulD_reg_reg(regD dst, regD src1, regD src2) %{
  predicate(UseAVX > 0);
  match(Set dst (MulD src1 src2));

  format %{ "vmulsd  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vmulsd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct mulD_reg_mem(regD dst, regD src1, memory src2) %{
  predicate(UseAVX > 0);
  match(Set dst (MulD src1 (LoadD src2)));

  format %{ "vmulsd  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vmulsd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct mulD_reg_imm(regD dst, regD src, immD con) %{
  predicate(UseAVX > 0);
  match(Set dst (MulD src con));

  format %{ "vmulsd  $dst, $src, [$constantaddress]\t# load from constant table: double=$con" %}
  ins_cost(150);
  ins_encode %{
    __ vmulsd($dst$$XMMRegister, $src$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct divF_reg(regF dst, regF src) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (DivF dst src));

  format %{ "divss   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ divss($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct divF_mem(regF dst, memory src) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (DivF dst (LoadF src)));

  format %{ "divss   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ divss($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct divF_imm(regF dst, immF con) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (DivF dst con));
  format %{ "divss   $dst, [$constantaddress]\t# load from constant table: float=$con" %}
  ins_cost(150);
  ins_encode %{
    __ divss($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct divF_reg_reg(regF dst, regF src1, regF src2) %{
  predicate(UseAVX > 0);
  match(Set dst (DivF src1 src2));

  format %{ "vdivss  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vdivss($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct divF_reg_mem(regF dst, regF src1, memory src2) %{
  predicate(UseAVX > 0);
  match(Set dst (DivF src1 (LoadF src2)));

  format %{ "vdivss  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vdivss($dst$$XMMRegister, $src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct divF_reg_imm(regF dst, regF src, immF con) %{
  predicate(UseAVX > 0);
  match(Set dst (DivF src con));

  format %{ "vdivss  $dst, $src, [$constantaddress]\t# load from constant table: float=$con" %}
  ins_cost(150);
  ins_encode %{
    __ vdivss($dst$$XMMRegister, $src$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct divD_reg(regD dst, regD src) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (DivD dst src));

  format %{ "divsd   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ divsd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct divD_mem(regD dst, memory src) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (DivD dst (LoadD src)));

  format %{ "divsd   $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ divsd($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct divD_imm(regD dst, immD con) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (DivD dst con));
  format %{ "divsd   $dst, [$constantaddress]\t# load from constant table: double=$con" %}
  ins_cost(150);
  ins_encode %{
    __ divsd($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct divD_reg_reg(regD dst, regD src1, regD src2) %{
  predicate(UseAVX > 0);
  match(Set dst (DivD src1 src2));

  format %{ "vdivsd  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vdivsd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct divD_reg_mem(regD dst, regD src1, memory src2) %{
  predicate(UseAVX > 0);
  match(Set dst (DivD src1 (LoadD src2)));

  format %{ "vdivsd  $dst, $src1, $src2" %}
  ins_cost(150);
  ins_encode %{
    __ vdivsd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct divD_reg_imm(regD dst, regD src, immD con) %{
  predicate(UseAVX > 0);
  match(Set dst (DivD src con));

  format %{ "vdivsd  $dst, $src, [$constantaddress]\t# load from constant table: double=$con" %}
  ins_cost(150);
  ins_encode %{
    __ vdivsd($dst$$XMMRegister, $src$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct absF_reg(regF dst) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (AbsF dst));
  ins_cost(150);
  format %{ "andps   $dst, [0x7fffffff]\t# abs float by sign masking" %}
  ins_encode %{
    __ andps($dst$$XMMRegister, ExternalAddress(float_signmask()));
  %}
  ins_pipe(pipe_slow);
%}

instruct absF_reg_reg(regF dst, regF src) %{
  predicate(VM_Version::supports_avxonly());
  match(Set dst (AbsF src));
  ins_cost(150);
  format %{ "vandps  $dst, $src, [0x7fffffff]\t# abs float by sign masking" %}
  ins_encode %{
    int vector_len = 0;
    __ vandps($dst$$XMMRegister, $src$$XMMRegister,
              ExternalAddress(float_signmask()), vector_len);
  %}
  ins_pipe(pipe_slow);
%}

#ifdef _LP64
instruct absF_reg_reg_evex(regF dst, regF src) %{
  predicate(UseAVX > 2 && VM_Version::supports_avx512vl());
  match(Set dst (AbsF src));
  ins_cost(150);
  format %{ "vandps  $dst, $src, [0x7fffffff]\t# abs float by sign masking" %}
  ins_encode %{
    int vector_len = 0;
    __ vandps($dst$$XMMRegister, $src$$XMMRegister,
              ExternalAddress(float_signmask()), vector_len);
  %}
  ins_pipe(pipe_slow);
%}

instruct absF_reg_reg_evex_special(regF dst, regF src1, regF src2) %{
  predicate(VM_Version::supports_avx512novl());
  match(Set dst (AbsF src1));
  effect(TEMP src2);
  ins_cost(150);
  format %{ "vabsss  $dst, $src1, $src2, [0x7fffffff]\t# abs float by sign masking" %}
  ins_encode %{
    int vector_len = 0;
    __ vabsss($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister,
              ExternalAddress(float_signmask()), vector_len);
  %}
  ins_pipe(pipe_slow);
%}
#else // _LP64
instruct absF_reg_reg_evex(regF dst, regF src) %{
  predicate(UseAVX > 2);
  match(Set dst (AbsF src));
  ins_cost(150);
  format %{ "vandps  $dst, $src, [0x7fffffff]\t# abs float by sign masking" %}
  ins_encode %{
    int vector_len = 0;
    __ vandps($dst$$XMMRegister, $src$$XMMRegister,
              ExternalAddress(float_signmask()), vector_len);
  %}
  ins_pipe(pipe_slow);
%}
#endif

instruct absD_reg(regD dst) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (AbsD dst));
  ins_cost(150);
  format %{ "andpd   $dst, [0x7fffffffffffffff]\t"
            "# abs double by sign masking" %}
  ins_encode %{
    __ andpd($dst$$XMMRegister, ExternalAddress(double_signmask()));
  %}
  ins_pipe(pipe_slow);
%}

instruct absD_reg_reg(regD dst, regD src) %{
  predicate(VM_Version::supports_avxonly());
  match(Set dst (AbsD src));
  ins_cost(150);
  format %{ "vandpd  $dst, $src, [0x7fffffffffffffff]\t"
            "# abs double by sign masking" %}
  ins_encode %{
    int vector_len = 0;
    __ vandpd($dst$$XMMRegister, $src$$XMMRegister,
              ExternalAddress(double_signmask()), vector_len);
  %}
  ins_pipe(pipe_slow);
%}

#ifdef _LP64
instruct absD_reg_reg_evex(regD dst, regD src) %{
  predicate(UseAVX > 2 && VM_Version::supports_avx512vl());
  match(Set dst (AbsD src));
  ins_cost(150);
  format %{ "vandpd  $dst, $src, [0x7fffffffffffffff]\t"
            "# abs double by sign masking" %}
  ins_encode %{
    int vector_len = 0;
    __ vandpd($dst$$XMMRegister, $src$$XMMRegister,
              ExternalAddress(double_signmask()), vector_len);
  %}
  ins_pipe(pipe_slow);
%}

instruct absD_reg_reg_evex_special(regD dst, regD src1, regD src2) %{
  predicate(VM_Version::supports_avx512novl());
  match(Set dst (AbsD src1));
  effect(TEMP src2);
  ins_cost(150);
  format %{ "vabssd  $dst, $src1, $src2, [0x7fffffffffffffff]\t# abs float by sign masking" %}
  ins_encode %{
    int vector_len = 0;
    __ vabssd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister,
              ExternalAddress(double_signmask()), vector_len);
  %}
  ins_pipe(pipe_slow);
%}
#else // _LP64
instruct absD_reg_reg_evex(regD dst, regD src) %{
  predicate(UseAVX > 2);
  match(Set dst (AbsD src));
  ins_cost(150);
  format %{ "vandpd  $dst, $src, [0x7fffffffffffffff]\t"
            "# abs double by sign masking" %}
  ins_encode %{
    int vector_len = 0;
    __ vandpd($dst$$XMMRegister, $src$$XMMRegister,
              ExternalAddress(double_signmask()), vector_len);
  %}
  ins_pipe(pipe_slow);
%}
#endif

instruct negF_reg(regF dst) %{
  predicate((UseSSE>=1) && (UseAVX == 0));
  match(Set dst (NegF dst));
  ins_cost(150);
  format %{ "xorps   $dst, [0x80000000]\t# neg float by sign flipping" %}
  ins_encode %{
    __ xorps($dst$$XMMRegister, ExternalAddress(float_signflip()));
  %}
  ins_pipe(pipe_slow);
%}

instruct negF_reg_reg(regF dst, regF src) %{
  predicate(UseAVX > 0);
  match(Set dst (NegF src));
  ins_cost(150);
  format %{ "vnegatess  $dst, $src, [0x80000000]\t# neg float by sign flipping" %}
  ins_encode %{
    __ vnegatess($dst$$XMMRegister, $src$$XMMRegister,
                 ExternalAddress(float_signflip()));
  %}
  ins_pipe(pipe_slow);
%}

instruct negD_reg(regD dst) %{
  predicate((UseSSE>=2) && (UseAVX == 0));
  match(Set dst (NegD dst));
  ins_cost(150);
  format %{ "xorpd   $dst, [0x8000000000000000]\t"
            "# neg double by sign flipping" %}
  ins_encode %{
    __ xorpd($dst$$XMMRegister, ExternalAddress(double_signflip()));
  %}
  ins_pipe(pipe_slow);
%}

instruct negD_reg_reg(regD dst, regD src) %{
  predicate(UseAVX > 0);
  match(Set dst (NegD src));
  ins_cost(150);
  format %{ "vnegatess  $dst, $src, [0x8000000000000000]\t"
            "# neg double by sign flipping" %}
  ins_encode %{
    __ vnegatesd($dst$$XMMRegister, $src$$XMMRegister,
                 ExternalAddress(double_signflip()));
  %}
  ins_pipe(pipe_slow);
%}

instruct sqrtF_reg(regF dst, regF src) %{
  predicate(UseSSE>=1);
  match(Set dst (SqrtF src));

  format %{ "sqrtss  $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ sqrtss($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct sqrtF_mem(regF dst, memory src) %{
  predicate(UseSSE>=1);
  match(Set dst (SqrtF (LoadF src)));

  format %{ "sqrtss  $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ sqrtss($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct sqrtF_imm(regF dst, immF con) %{
  predicate(UseSSE>=1);
  match(Set dst (SqrtF con));

  format %{ "sqrtss  $dst, [$constantaddress]\t# load from constant table: float=$con" %}
  ins_cost(150);
  ins_encode %{
    __ sqrtss($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct sqrtD_reg(regD dst, regD src) %{
  predicate(UseSSE>=2);
  match(Set dst (SqrtD src));

  format %{ "sqrtsd  $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ sqrtsd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe(pipe_slow);
%}

instruct sqrtD_mem(regD dst, memory src) %{
  predicate(UseSSE>=2);
  match(Set dst (SqrtD (LoadD src)));

  format %{ "sqrtsd  $dst, $src" %}
  ins_cost(150);
  ins_encode %{
    __ sqrtsd($dst$$XMMRegister, $src$$Address);
  %}
  ins_pipe(pipe_slow);
%}

instruct sqrtD_imm(regD dst, immD con) %{
  predicate(UseSSE>=2);
  match(Set dst (SqrtD con));
  format %{ "sqrtsd  $dst, [$constantaddress]\t# load from constant table: double=$con" %}
  ins_cost(150);
  ins_encode %{
    __ sqrtsd($dst$$XMMRegister, $constantaddress($con));
  %}
  ins_pipe(pipe_slow);
%}

instruct onspinwait() %{
  match(OnSpinWait);
  ins_cost(200);

  format %{
    $$template
    if (os::is_MP()) {
      $$emit$$"pause\t! membar_onspinwait"
    } else {
      $$emit$$"MEMBAR-onspinwait ! (empty encoding)"
    }
  %}
  ins_encode %{
    __ pause();
  %}
  ins_pipe(pipe_slow);
%}

// a * b + c
instruct fmaD_reg(regD a, regD b, regD c) %{
  predicate(UseFMA);
  match(Set c (FmaD  c (Binary a b)));
  format %{ "fmasd $a,$b,$c\t# $c = $a * $b + $c" %}
  ins_cost(150);
  ins_encode %{
    __ fmad($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct fmaF_reg(regF a, regF b, regF c) %{
  predicate(UseFMA);
  match(Set c (FmaF  c (Binary a b)));
  format %{ "fmass $a,$b,$c\t# $c = $a * $b + $c" %}
  ins_cost(150);
  ins_encode %{
    __ fmaf($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// ====================VECTOR INSTRUCTIONS=====================================

// Load vectors (4 bytes long)
instruct loadV4(vecS dst, memory mem) %{
  predicate(n->as_LoadVector()->memory_size() == 4);
  match(Set dst (LoadVector mem));
  ins_cost(125);
  format %{ "movd    $dst,$mem\t! load vector (4 bytes)" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $mem$$Address);
  %}
  ins_pipe( pipe_slow );
%}

// Load vectors (8 bytes long)
instruct loadV8(vecD dst, memory mem) %{
  predicate(n->as_LoadVector()->memory_size() == 8);
  match(Set dst (LoadVector mem));
  ins_cost(125);
  format %{ "movq    $dst,$mem\t! load vector (8 bytes)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $mem$$Address);
  %}
  ins_pipe( pipe_slow );
%}

// Load vectors (16 bytes long)
instruct loadV16(vecX dst, memory mem) %{
  predicate(n->as_LoadVector()->memory_size() == 16);
  match(Set dst (LoadVector mem));
  ins_cost(125);
  format %{ "movdqu  $dst,$mem\t! load vector (16 bytes)" %}
  ins_encode %{
    __ movdqu($dst$$XMMRegister, $mem$$Address);
  %}
  ins_pipe( pipe_slow );
%}

// Load vectors (32 bytes long)
instruct loadV32(vecY dst, memory mem) %{
  predicate(n->as_LoadVector()->memory_size() == 32);
  match(Set dst (LoadVector mem));
  ins_cost(125);
  format %{ "vmovdqu $dst,$mem\t! load vector (32 bytes)" %}
  ins_encode %{
    __ vmovdqu($dst$$XMMRegister, $mem$$Address);
  %}
  ins_pipe( pipe_slow );
%}

// Load vectors (64 bytes long)
instruct loadV64_dword(vecZ dst, memory mem) %{
  predicate(n->as_LoadVector()->memory_size() == 64 && n->as_LoadVector()->element_size() <= 4);
  match(Set dst (LoadVector mem));
  ins_cost(125);
  format %{ "vmovdqul $dst k0,$mem\t! load vector (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ evmovdqul($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Load vectors (64 bytes long)
instruct loadV64_qword(vecZ dst, memory mem) %{
  predicate(n->as_LoadVector()->memory_size() == 64 && n->as_LoadVector()->element_size() > 4);
  match(Set dst (LoadVector mem));
  ins_cost(125);
  format %{ "vmovdquq $dst k0,$mem\t! load vector (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ evmovdquq($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Store vectors
instruct storeV4(memory mem, vecS src) %{
  predicate(n->as_StoreVector()->memory_size() == 4);
  match(Set mem (StoreVector mem src));
  ins_cost(145);
  format %{ "movd    $mem,$src\t! store vector (4 bytes)" %}
  ins_encode %{
    __ movdl($mem$$Address, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct storeV8(memory mem, vecD src) %{
  predicate(n->as_StoreVector()->memory_size() == 8);
  match(Set mem (StoreVector mem src));
  ins_cost(145);
  format %{ "movq    $mem,$src\t! store vector (8 bytes)" %}
  ins_encode %{
    __ movq($mem$$Address, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct storeV16(memory mem, vecX src) %{
  predicate(n->as_StoreVector()->memory_size() == 16);
  match(Set mem (StoreVector mem src));
  ins_cost(145);
  format %{ "movdqu  $mem,$src\t! store vector (16 bytes)" %}
  ins_encode %{
    __ movdqu($mem$$Address, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct storeV32(memory mem, vecY src) %{
  predicate(n->as_StoreVector()->memory_size() == 32);
  match(Set mem (StoreVector mem src));
  ins_cost(145);
  format %{ "vmovdqu $mem,$src\t! store vector (32 bytes)" %}
  ins_encode %{
    __ vmovdqu($mem$$Address, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct storeV64_dword(memory mem, vecZ src) %{
  predicate(n->as_StoreVector()->memory_size() == 64 && n->as_StoreVector()->element_size() <= 4);
  match(Set mem (StoreVector mem src));
  ins_cost(145);
  format %{ "vmovdqul $mem k0,$src\t! store vector (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ evmovdqul($mem$$Address, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct storeV64_qword(memory mem, vecZ src) %{
  predicate(n->as_StoreVector()->memory_size() == 64 && n->as_StoreVector()->element_size() > 4);
  match(Set mem (StoreVector mem src));
  ins_cost(145);
  format %{ "vmovdquq $mem k0,$src\t! store vector (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ evmovdquq($mem$$Address, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// ====================LEGACY REPLICATE=======================================

instruct Repl4B_mem(vecS dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB (LoadB mem)));
  format %{ "punpcklbw $dst,$mem\n\t"
            "pshuflw $dst,$dst,0x00\t! replicate4B" %}
  ins_encode %{
    __ punpcklbw($dst$$XMMRegister, $mem$$Address);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8B_mem(vecD dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB (LoadB mem)));
  format %{ "punpcklbw $dst,$mem\n\t"
            "pshuflw $dst,$dst,0x00\t! replicate8B" %}
  ins_encode %{
    __ punpcklbw($dst$$XMMRegister, $mem$$Address);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16B(vecX dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB src));
  format %{ "movd    $dst,$src\n\t"
            "punpcklbw $dst,$dst\n\t"
            "pshuflw $dst,$dst,0x00\n\t"
            "punpcklqdq $dst,$dst\t! replicate16B" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16B_mem(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 16 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB (LoadB mem)));
  format %{ "punpcklbw $dst,$mem\n\t"
            "pshuflw $dst,$dst,0x00\n\t"
            "punpcklqdq $dst,$dst\t! replicate16B" %}
  ins_encode %{
    __ punpcklbw($dst$$XMMRegister, $mem$$Address);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32B(vecY dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 32 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB src));
  format %{ "movd    $dst,$src\n\t"
            "punpcklbw $dst,$dst\n\t"
            "pshuflw $dst,$dst,0x00\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! replicate32B" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32B_mem(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 32 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB (LoadB mem)));
  format %{ "punpcklbw $dst,$mem\n\t"
            "pshuflw $dst,$dst,0x00\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! replicate32B" %}
  ins_encode %{
    __ punpcklbw($dst$$XMMRegister, $mem$$Address);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16B_imm(vecX dst, immI con) %{
  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "punpcklqdq $dst,$dst\t! replicate16B($con)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32B_imm(vecY dst, immI con) %{
  predicate(n->as_Vector()->length() == 32 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! lreplicate32B($con)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4S(vecD dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS src));
  format %{ "movd    $dst,$src\n\t"
            "pshuflw $dst,$dst,0x00\t! replicate4S" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4S_mem(vecD dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS (LoadS mem)));
  format %{ "pshuflw $dst,$mem,0x00\t! replicate4S" %}
  ins_encode %{
    __ pshuflw($dst$$XMMRegister, $mem$$Address, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8S(vecX dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS src));
  format %{ "movd    $dst,$src\n\t"
            "pshuflw $dst,$dst,0x00\n\t"
            "punpcklqdq $dst,$dst\t! replicate8S" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8S_mem(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS (LoadS mem)));
  format %{ "pshuflw $dst,$mem,0x00\n\t"
            "punpcklqdq $dst,$dst\t! replicate8S" %}
  ins_encode %{
    __ pshuflw($dst$$XMMRegister, $mem$$Address, 0x00);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8S_imm(vecX dst, immI con) %{
  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "punpcklqdq $dst,$dst\t! replicate8S($con)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16S(vecY dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS src));
  format %{ "movd    $dst,$src\n\t"
            "pshuflw $dst,$dst,0x00\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! replicate16S" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16S_mem(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS (LoadS mem)));
  format %{ "pshuflw $dst,$mem,0x00\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! replicate16S" %}
  ins_encode %{
    __ pshuflw($dst$$XMMRegister, $mem$$Address, 0x00);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16S_imm(vecY dst, immI con) %{
  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! replicate16S($con)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4I(vecX dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI src));
  format %{ "movd    $dst,$src\n\t"
            "pshufd  $dst,$dst,0x00\t! replicate4I" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4I_mem(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && UseAVX > 0 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI (LoadI mem)));
  format %{ "pshufd  $dst,$mem,0x00\t! replicate4I" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8I(vecY dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI src));
  format %{ "movd    $dst,$src\n\t"
            "pshufd  $dst,$dst,0x00\n\t"
            "vinserti128_high $dst,$dst\t! replicate8I" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8I_mem(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI (LoadI mem)));
  format %{ "pshufd  $dst,$mem,0x00\n\t"
            "vinserti128_high $dst,$dst\t! replicate8I" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4I_imm(vecX dst, immI con) %{
  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI con));
  format %{ "movq    $dst,[$constantaddress]\t! replicate4I($con)\n\t"
            "punpcklqdq $dst,$dst" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8I_imm(vecY dst, immI con) %{
  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI con));
  format %{ "movq    $dst,[$constantaddress]\t! replicate8I($con)\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Long could be loaded into xmm register directly from memory.
instruct Repl2L_mem(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 2 && !VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateL (LoadL mem)));
  format %{ "movq    $dst,$mem\n\t"
            "punpcklqdq $dst,$dst\t! replicate2L" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $mem$$Address);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Replicate long (8 byte) scalar to be vector
#ifdef _LP64
instruct Repl4L(vecY dst, rRegL src) %{
  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateL src));
  format %{ "movdq   $dst,$src\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! replicate4L" %}
  ins_encode %{
    __ movdq($dst$$XMMRegister, $src$$Register);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}
#else // _LP64
instruct Repl4L(vecY dst, eRegL src, regD tmp) %{
  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateL src));
  effect(TEMP dst, USE src, TEMP tmp);
  format %{ "movdl   $dst,$src.lo\n\t"
            "movdl   $tmp,$src.hi\n\t"
            "punpckldq $dst,$tmp\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! replicate4L" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}
#endif // _LP64

instruct Repl4L_imm(vecY dst, immL con) %{
  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateL con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! replicate4L($con)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress($con));
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4L_mem(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateL (LoadL mem)));
  format %{ "movq    $dst,$mem\n\t"
            "punpcklqdq $dst,$dst\n\t"
            "vinserti128_high $dst,$dst\t! replicate4L" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $mem$$Address);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
    __ vinserti128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl2F_mem(vecD dst, memory mem) %{
  predicate(n->as_Vector()->length() == 2 && UseAVX > 0 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateF (LoadF mem)));
  format %{ "pshufd  $dst,$mem,0x00\t! replicate2F" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4F_mem(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && UseAVX > 0 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateF (LoadF mem)));
  format %{ "pshufd  $dst,$mem,0x00\t! replicate4F" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8F(vecY dst, regF src) %{
  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateF src));
  format %{ "pshufd  $dst,$src,0x00\n\t"
            "vinsertf128_high $dst,$dst\t! replicate8F" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x00);
    __ vinsertf128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8F_mem(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateF (LoadF mem)));
  format %{ "pshufd  $dst,$mem,0x00\n\t"
            "vinsertf128_high $dst,$dst\t! replicate8F" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
    __ vinsertf128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl2F_zero(vecD dst, immF0 zero) %{
  predicate(n->as_Vector()->length() == 2 && UseAVX < 3);
  match(Set dst (ReplicateF zero));
  format %{ "xorps   $dst,$dst\t! replicate2F zero" %}
  ins_encode %{
    __ xorps($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4F_zero(vecX dst, immF0 zero) %{
  predicate(n->as_Vector()->length() == 4 && UseAVX < 3);
  match(Set dst (ReplicateF zero));
  format %{ "xorps   $dst,$dst\t! replicate4F zero" %}
  ins_encode %{
    __ xorps($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl8F_zero(vecY dst, immF0 zero) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX < 3);
  match(Set dst (ReplicateF zero));
  format %{ "vxorps  $dst,$dst,$dst\t! replicate8F zero" %}
  ins_encode %{
    int vector_len = 1;
    __ vxorps($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl2D_mem(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 2 && UseAVX > 0 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateD (LoadD mem)));
  format %{ "pshufd  $dst,$mem,0x44\t! replicate2D" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x44);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4D(vecY dst, regD src) %{
  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateD src));
  format %{ "pshufd  $dst,$src,0x44\n\t"
            "vinsertf128_high $dst,$dst\t! replicate4D" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x44);
    __ vinsertf128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4D_mem(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
  match(Set dst (ReplicateD (LoadD mem)));
  format %{ "pshufd  $dst,$mem,0x44\n\t"
            "vinsertf128_high $dst,$dst\t! replicate4D" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x44);
    __ vinsertf128_high($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Replicate double (8 byte) scalar zero to be vector
instruct Repl2D_zero(vecX dst, immD0 zero) %{
  predicate(n->as_Vector()->length() == 2 && UseAVX < 3);
  match(Set dst (ReplicateD zero));
  format %{ "xorpd   $dst,$dst\t! replicate2D zero" %}
  ins_encode %{
    __ xorpd($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4D_zero(vecY dst, immD0 zero) %{
  predicate(n->as_Vector()->length() == 4 && UseAVX < 3);
  match(Set dst (ReplicateD zero));
  format %{ "vxorpd  $dst,$dst,$dst,vect256\t! replicate4D zero" %}
  ins_encode %{
    int vector_len = 1;
    __ vxorpd($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

// ====================GENERIC REPLICATE==========================================

// Replicate byte scalar to be vector
instruct Repl4B(vecS dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (ReplicateB src));
  format %{ "movd    $dst,$src\n\t"
            "punpcklbw $dst,$dst\n\t"
            "pshuflw $dst,$dst,0x00\t! replicate4B" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8B(vecD dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 8);
  match(Set dst (ReplicateB src));
  format %{ "movd    $dst,$src\n\t"
            "punpcklbw $dst,$dst\n\t"
            "pshuflw $dst,$dst,0x00\t! replicate8B" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

// Replicate byte scalar immediate to be vector by loading from const table.
instruct Repl4B_imm(vecS dst, immI con) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (ReplicateB con));
  format %{ "movdl   $dst,[$constantaddress]\t! replicate4B($con)" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $constantaddress(replicate4_imm($con$$constant, 1)));
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8B_imm(vecD dst, immI con) %{
  predicate(n->as_Vector()->length() == 8);
  match(Set dst (ReplicateB con));
  format %{ "movq    $dst,[$constantaddress]\t! replicate8B($con)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
  %}
  ins_pipe( pipe_slow );
%}

// Replicate byte scalar zero to be vector
instruct Repl4B_zero(vecS dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (ReplicateB zero));
  format %{ "pxor    $dst,$dst\t! replicate4B zero" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl8B_zero(vecD dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 8);
  match(Set dst (ReplicateB zero));
  format %{ "pxor    $dst,$dst\t! replicate8B zero" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl16B_zero(vecX dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 16);
  match(Set dst (ReplicateB zero));
  format %{ "pxor    $dst,$dst\t! replicate16B zero" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl32B_zero(vecY dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 32);
  match(Set dst (ReplicateB zero));
  format %{ "vpxor   $dst,$dst,$dst\t! replicate32B zero" %}
  ins_encode %{
    // Use vxorpd since AVX does not have vpxor for 256-bit (AVX2 will have it).
    int vector_len = 1;
    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

// Replicate char/short (2 byte) scalar to be vector
instruct Repl2S(vecS dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateS src));
  format %{ "movd    $dst,$src\n\t"
            "pshuflw $dst,$dst,0x00\t! replicate2S" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
  %}
  ins_pipe( fpu_reg_reg );
%}

// Replicate char/short (2 byte) scalar immediate to be vector by loading from const table.
instruct Repl2S_imm(vecS dst, immI con) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateS con));
  format %{ "movdl   $dst,[$constantaddress]\t! replicate2S($con)" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $constantaddress(replicate4_imm($con$$constant, 2)));
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4S_imm(vecD dst, immI con) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (ReplicateS con));
  format %{ "movq    $dst,[$constantaddress]\t! replicate4S($con)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
  %}
  ins_pipe( fpu_reg_reg );
%}

// Replicate char/short (2 byte) scalar zero to be vector
instruct Repl2S_zero(vecS dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateS zero));
  format %{ "pxor    $dst,$dst\t! replicate2S zero" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4S_zero(vecD dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (ReplicateS zero));
  format %{ "pxor    $dst,$dst\t! replicate4S zero" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl8S_zero(vecX dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 8);
  match(Set dst (ReplicateS zero));
  format %{ "pxor    $dst,$dst\t! replicate8S zero" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl16S_zero(vecY dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 16);
  match(Set dst (ReplicateS zero));
  format %{ "vpxor   $dst,$dst,$dst\t! replicate16S zero" %}
  ins_encode %{
    // Use vxorpd since AVX does not have vpxor for 256-bit (AVX2 will have it).
    int vector_len = 1;
    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

// Replicate integer (4 byte) scalar to be vector
instruct Repl2I(vecD dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateI src));
  format %{ "movd    $dst,$src\n\t"
            "pshufd  $dst,$dst,0x00\t! replicate2I" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
  %}
  ins_pipe( fpu_reg_reg );
%}

// Integer could be loaded into xmm register directly from memory.
instruct Repl2I_mem(vecD dst, memory mem) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateI (LoadI mem)));
  format %{ "movd    $dst,$mem\n\t"
            "pshufd  $dst,$dst,0x00\t! replicate2I" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $mem$$Address);
    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
  %}
  ins_pipe( fpu_reg_reg );
%}

// Replicate integer (4 byte) scalar immediate to be vector by loading from const table.
instruct Repl2I_imm(vecD dst, immI con) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateI con));
  format %{ "movq    $dst,[$constantaddress]\t! replicate2I($con)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
  %}
  ins_pipe( fpu_reg_reg );
%}

// Replicate integer (4 byte) scalar zero to be vector
instruct Repl2I_zero(vecD dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateI zero));
  format %{ "pxor    $dst,$dst\t! replicate2I" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4I_zero(vecX dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (ReplicateI zero));
  format %{ "pxor    $dst,$dst\t! replicate4I zero)" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl8I_zero(vecY dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 8);
  match(Set dst (ReplicateI zero));
  format %{ "vpxor   $dst,$dst,$dst\t! replicate8I zero" %}
  ins_encode %{
    // Use vxorpd since AVX does not have vpxor for 256-bit (AVX2 will have it).
    int vector_len = 1;
    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

// Replicate long (8 byte) scalar to be vector
#ifdef _LP64
instruct Repl2L(vecX dst, rRegL src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateL src));
  format %{ "movdq   $dst,$src\n\t"
            "punpcklqdq $dst,$dst\t! replicate2L" %}
  ins_encode %{
    __ movdq($dst$$XMMRegister, $src$$Register);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}
#else // _LP64
instruct Repl2L(vecX dst, eRegL src, regD tmp) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateL src));
  effect(TEMP dst, USE src, TEMP tmp);
  format %{ "movdl   $dst,$src.lo\n\t"
            "movdl   $tmp,$src.hi\n\t"
            "punpckldq $dst,$tmp\n\t"
            "punpcklqdq $dst,$dst\t! replicate2L"%}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}
#endif // _LP64

// Replicate long (8 byte) scalar immediate to be vector by loading from const table.
instruct Repl2L_imm(vecX dst, immL con) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateL con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "punpcklqdq $dst,$dst\t! replicate2L($con)" %}
  ins_encode %{
    __ movq($dst$$XMMRegister, $constantaddress($con));
    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// Replicate long (8 byte) scalar zero to be vector
instruct Repl2L_zero(vecX dst, immL0 zero) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateL zero));
  format %{ "pxor    $dst,$dst\t! replicate2L zero" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $dst$$XMMRegister);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4L_zero(vecY dst, immL0 zero) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (ReplicateL zero));
  format %{ "vpxor   $dst,$dst,$dst\t! replicate4L zero" %}
  ins_encode %{
    // Use vxorpd since AVX does not have vpxor for 256-bit (AVX2 will have it).
    int vector_len = 1;
    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

// Replicate float (4 byte) scalar to be vector
instruct Repl2F(vecD dst, regF src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateF src));
  format %{ "pshufd  $dst,$dst,0x00\t! replicate2F" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x00);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4F(vecX dst, regF src) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (ReplicateF src));
  format %{ "pshufd  $dst,$dst,0x00\t! replicate4F" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x00);
  %}
  ins_pipe( pipe_slow );
%}

// Replicate double (8 bytes) scalar to be vector
instruct Repl2D(vecX dst, regD src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (ReplicateD src));
  format %{ "pshufd  $dst,$src,0x44\t! replicate2D" %}
  ins_encode %{
    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x44);
  %}
  ins_pipe( pipe_slow );
%}

// ====================EVEX REPLICATE=============================================

instruct Repl4B_mem_evex(vecS dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB (LoadB mem)));
  format %{ "vpbroadcastb  $dst,$mem\t! replicate4B" %}
  ins_encode %{
    int vector_len = 0;
    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8B_mem_evex(vecD dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB (LoadB mem)));
  format %{ "vpbroadcastb  $dst,$mem\t! replicate8B" %}
  ins_encode %{
    int vector_len = 0;
    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16B_evex(vecX dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB src));
  format %{ "vpbroadcastb $dst,$src\t! replicate16B" %}
  ins_encode %{
   int vector_len = 0;
    __ evpbroadcastb($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16B_mem_evex(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB (LoadB mem)));
  format %{ "vpbroadcastb  $dst,$mem\t! replicate16B" %}
  ins_encode %{
    int vector_len = 0;
    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32B_evex(vecY dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 32 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB src));
  format %{ "vpbroadcastb $dst,$src\t! replicate32B" %}
  ins_encode %{
   int vector_len = 1;
    __ evpbroadcastb($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32B_mem_evex(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 32 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB (LoadB mem)));
  format %{ "vpbroadcastb  $dst,$mem\t! replicate32B" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl64B_evex(vecZ dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 64 && VM_Version::supports_avx512bw());
  match(Set dst (ReplicateB src));
  format %{ "vpbroadcastb $dst,$src\t! upper replicate64B" %}
  ins_encode %{
   int vector_len = 2;
    __ evpbroadcastb($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl64B_mem_evex(vecZ dst, memory mem) %{
  predicate(n->as_Vector()->length() == 64 && VM_Version::supports_avx512bw());
  match(Set dst (ReplicateB (LoadB mem)));
  format %{ "vpbroadcastb  $dst,$mem\t! replicate64B" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16B_imm_evex(vecX dst, immI con) %{
  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "vpbroadcastb $dst,$dst\t! replicate16B" %}
  ins_encode %{
   int vector_len = 0;
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
    __ evpbroadcastb($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32B_imm_evex(vecY dst, immI con) %{
  predicate(n->as_Vector()->length() == 32 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateB con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "vpbroadcastb $dst,$dst\t! replicate32B" %}
  ins_encode %{
   int vector_len = 1;
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
    __ evpbroadcastb($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl64B_imm_evex(vecZ dst, immI con) %{
  predicate(n->as_Vector()->length() == 64 && VM_Version::supports_avx512bw());
  match(Set dst (ReplicateB con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "vpbroadcastb $dst,$dst\t! upper replicate64B" %}
  ins_encode %{
   int vector_len = 2;
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
    __ evpbroadcastb($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl64B_zero_evex(vecZ dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 64 && UseAVX > 2);
  match(Set dst (ReplicateB zero));
  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate64B zero" %}
  ins_encode %{
    // Use vxorpd since AVX does not have vpxor for 512-bit (EVEX will have it).
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4S_evex(vecD dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS src));
  format %{ "vpbroadcastw $dst,$src\t! replicate4S" %}
  ins_encode %{
   int vector_len = 0;
    __ evpbroadcastw($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4S_mem_evex(vecD dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS (LoadS mem)));
  format %{ "vpbroadcastw  $dst,$mem\t! replicate4S" %}
  ins_encode %{
    int vector_len = 0;
    __ evpbroadcastw($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8S_evex(vecX dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS src));
  format %{ "vpbroadcastw $dst,$src\t! replicate8S" %}
  ins_encode %{
   int vector_len = 0;
    __ evpbroadcastw($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8S_mem_evex(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS (LoadS mem)));
  format %{ "vpbroadcastw  $dst,$mem\t! replicate8S" %}
  ins_encode %{
    int vector_len = 0;
    __ evpbroadcastw($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16S_evex(vecY dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS src));
  format %{ "vpbroadcastw $dst,$src\t! replicate16S" %}
  ins_encode %{
   int vector_len = 1;
    __ evpbroadcastw($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16S_mem_evex(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS (LoadS mem)));
  format %{ "vpbroadcastw  $dst,$mem\t! replicate16S" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastw($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32S_evex(vecZ dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 32 && VM_Version::supports_avx512bw());
  match(Set dst (ReplicateS src));
  format %{ "vpbroadcastw $dst,$src\t! replicate32S" %}
  ins_encode %{
   int vector_len = 2;
    __ evpbroadcastw($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32S_mem_evex(vecZ dst, memory mem) %{
  predicate(n->as_Vector()->length() == 32 && VM_Version::supports_avx512bw());
  match(Set dst (ReplicateS (LoadS mem)));
  format %{ "vpbroadcastw  $dst,$mem\t! replicate32S" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastw($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8S_imm_evex(vecX dst, immI con) %{
  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "vpbroadcastw $dst,$dst\t! replicate8S" %}
  ins_encode %{
   int vector_len = 0;
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
    __ evpbroadcastw($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16S_imm_evex(vecY dst, immI con) %{
  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
  match(Set dst (ReplicateS con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "vpbroadcastw $dst,$dst\t! replicate16S" %}
  ins_encode %{
   int vector_len = 1;
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
    __ evpbroadcastw($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32S_imm_evex(vecZ dst, immI con) %{
  predicate(n->as_Vector()->length() == 32 && VM_Version::supports_avx512bw());
  match(Set dst (ReplicateS con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "vpbroadcastw $dst,$dst\t! replicate32S" %}
  ins_encode %{
   int vector_len = 2;
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
    __ evpbroadcastw($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl32S_zero_evex(vecZ dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 32 && UseAVX > 2);
  match(Set dst (ReplicateS zero));
  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate32S zero" %}
  ins_encode %{
    // Use vxorpd since AVX does not have vpxor for 512-bit (EVEX will have it).
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4I_evex(vecX dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI src));
  format %{ "vpbroadcastd  $dst,$src\t! replicate4I" %}
  ins_encode %{
    int vector_len = 0;
    __ evpbroadcastd($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4I_mem_evex(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI (LoadI mem)));
  format %{ "vpbroadcastd  $dst,$mem\t! replicate4I" %}
  ins_encode %{
    int vector_len = 0;
    __ evpbroadcastd($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8I_evex(vecY dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI src));
  format %{ "vpbroadcastd  $dst,$src\t! replicate8I" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastd($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8I_mem_evex(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI (LoadI mem)));
  format %{ "vpbroadcastd  $dst,$mem\t! replicate8I" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastd($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16I_evex(vecZ dst, rRegI src) %{
  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
  match(Set dst (ReplicateI src));
  format %{ "vpbroadcastd  $dst,$src\t! replicate16I" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastd($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16I_mem_evex(vecZ dst, memory mem) %{
  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
  match(Set dst (ReplicateI (LoadI mem)));
  format %{ "vpbroadcastd  $dst,$mem\t! replicate16I" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastd($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4I_imm_evex(vecX dst, immI con) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI con));
  format %{ "movq    $dst,[$constantaddress]\t! replicate8I($con)\n\t"
            "vpbroadcastd  $dst,$dst\t! replicate4I" %}
  ins_encode %{
    int vector_len = 0;
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
    __ evpbroadcastd($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8I_imm_evex(vecY dst, immI con) %{
  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateI con));
  format %{ "movq    $dst,[$constantaddress]\t! replicate8I($con)\n\t"
            "vpbroadcastd  $dst,$dst\t! replicate8I" %}
  ins_encode %{
    int vector_len = 1;
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
    __ evpbroadcastd($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16I_imm_evex(vecZ dst, immI con) %{
  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
  match(Set dst (ReplicateI con));
  format %{ "movq    $dst,[$constantaddress]\t! replicate16I($con)\n\t"
            "vpbroadcastd  $dst,$dst\t! replicate16I" %}
  ins_encode %{
    int vector_len = 2;
    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
    __ evpbroadcastd($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16I_zero_evex(vecZ dst, immI0 zero) %{
  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
  match(Set dst (ReplicateI zero));
  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate16I zero" %}
  ins_encode %{
    // Use vxorpd since AVX does not have vpxor for 512-bit (AVX2 will have it).
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

// Replicate long (8 byte) scalar to be vector
#ifdef _LP64
instruct Repl4L_evex(vecY dst, rRegL src) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateL src));
  format %{ "vpbroadcastq  $dst,$src\t! replicate4L" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastq($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8L_evex(vecZ dst, rRegL src) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
  match(Set dst (ReplicateL src));
  format %{ "vpbroadcastq  $dst,$src\t! replicate8L" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastq($dst$$XMMRegister, $src$$Register, vector_len);
  %}
  ins_pipe( pipe_slow );
%}
#else // _LP64
instruct Repl4L_evex(vecY dst, eRegL src, regD tmp) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateL src));
  effect(TEMP dst, USE src, TEMP tmp);
  format %{ "movdl   $dst,$src.lo\n\t"
            "movdl   $tmp,$src.hi\n\t"
            "punpckldq $dst,$tmp\n\t"
            "vpbroadcastq  $dst,$dst\t! replicate4L" %}
  ins_encode %{
    int vector_len = 1;
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
    __ evpbroadcastq($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8L_evex(vecZ dst, eRegL src, regD tmp) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
  match(Set dst (ReplicateL src));
  effect(TEMP dst, USE src, TEMP tmp);
  format %{ "movdl   $dst,$src.lo\n\t"
            "movdl   $tmp,$src.hi\n\t"
            "punpckldq $dst,$tmp\n\t"
            "vpbroadcastq  $dst,$dst\t! replicate8L" %}
  ins_encode %{
    int vector_len = 2;
    __ movdl($dst$$XMMRegister, $src$$Register);
    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
    __ evpbroadcastq($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}
#endif // _LP64

instruct Repl4L_imm_evex(vecY dst, immL con) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateL con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "vpbroadcastq  $dst,$dst\t! replicate4L" %}
  ins_encode %{
    int vector_len = 1;
    __ movq($dst$$XMMRegister, $constantaddress($con));
    __ evpbroadcastq($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8L_imm_evex(vecZ dst, immL con) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
  match(Set dst (ReplicateL con));
  format %{ "movq    $dst,[$constantaddress]\n\t"
            "vpbroadcastq  $dst,$dst\t! replicate8L" %}
  ins_encode %{
    int vector_len = 2;
    __ movq($dst$$XMMRegister, $constantaddress($con));
    __ evpbroadcastq($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl2L_mem_evex(vecX dst, memory mem) %{
  predicate(n->as_Vector()->length() == 2 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateL (LoadL mem)));
  format %{ "vpbroadcastd  $dst,$mem\t! replicate2L" %}
  ins_encode %{
    int vector_len = 0;
    __ evpbroadcastq($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4L_mem_evex(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateL (LoadL mem)));
  format %{ "vpbroadcastd  $dst,$mem\t! replicate4L" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastq($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8L_mem_evex(vecZ dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
  match(Set dst (ReplicateL (LoadL mem)));
  format %{ "vpbroadcastd  $dst,$mem\t! replicate8L" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastq($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8L_zero_evex(vecZ dst, immL0 zero) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
  match(Set dst (ReplicateL zero));
  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate8L zero" %}
  ins_encode %{
    // Use vxorpd since AVX does not have vpxor for 512-bit (EVEX will have it).
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl8F_evex(vecY dst, regF src) %{
  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateF src));
  format %{ "vbroadcastss $dst,$src\t! replicate8F" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastss($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8F_mem_evex(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateF (LoadF mem)));
  format %{ "vbroadcastss  $dst,$mem\t! replicate8F" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastss($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16F_evex(vecZ dst, regF src) %{
  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
  match(Set dst (ReplicateF src));
  format %{ "vbroadcastss $dst,$src\t! replicate16F" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastss($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl16F_mem_evex(vecZ dst, memory mem) %{
  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
  match(Set dst (ReplicateF (LoadF mem)));
  format %{ "vbroadcastss  $dst,$mem\t! replicate16F" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastss($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl2F_zero_evex(vecD dst, immF0 zero) %{
  predicate(n->as_Vector()->length() == 2 && UseAVX > 2);
  match(Set dst (ReplicateF zero));
  format %{ "vpxor  $dst k0,$dst,$dst\t! replicate2F zero" %}
  ins_encode %{
    // Use vpxor in place of vxorps since EVEX has a constriant on dq for vxorps: this is a 512-bit operation
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4F_zero_evex(vecX dst, immF0 zero) %{
  predicate(n->as_Vector()->length() == 4 && UseAVX > 2);
  match(Set dst (ReplicateF zero));
  format %{ "vpxor  $dst k0,$dst,$dst\t! replicate4F zero" %}
  ins_encode %{
    // Use vpxor in place of vxorps since EVEX has a constriant on dq for vxorps: this is a 512-bit operation
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl8F_zero_evex(vecY dst, immF0 zero) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
  match(Set dst (ReplicateF zero));
  format %{ "vpxor  $dst k0,$dst,$dst\t! replicate8F zero" %}
  ins_encode %{
    // Use vpxor in place of vxorps since EVEX has a constriant on dq for vxorps: this is a 512-bit operation
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl16F_zero_evex(vecZ dst, immF0 zero) %{
  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
  match(Set dst (ReplicateF zero));
  format %{ "vpxor  $dst k0,$dst,$dst\t! replicate16F zero" %}
  ins_encode %{
    // Use vpxor in place of vxorps since EVEX has a constriant on dq for vxorps: this is a 512-bit operation
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4D_evex(vecY dst, regD src) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateD src));
  format %{ "vbroadcastsd $dst,$src\t! replicate4D" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastsd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl4D_mem_evex(vecY dst, memory mem) %{
  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
  match(Set dst (ReplicateD (LoadD mem)));
  format %{ "vbroadcastsd  $dst,$mem\t! replicate4D" %}
  ins_encode %{
    int vector_len = 1;
    __ evpbroadcastsd($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8D_evex(vecZ dst, regD src) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
  match(Set dst (ReplicateD src));
  format %{ "vbroadcastsd $dst,$src\t! replicate8D" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastsd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl8D_mem_evex(vecZ dst, memory mem) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
  match(Set dst (ReplicateD (LoadD mem)));
  format %{ "vbroadcastsd  $dst,$mem\t! replicate8D" %}
  ins_encode %{
    int vector_len = 2;
    __ evpbroadcastsd($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct Repl2D_zero_evex(vecX dst, immD0 zero) %{
  predicate(n->as_Vector()->length() == 2 && UseAVX > 2);
  match(Set dst (ReplicateD zero));
  format %{ "vpxor  $dst k0,$dst,$dst\t! replicate2D zero" %}
  ins_encode %{
    // Use vpxor in place of vxorpd since EVEX has a constriant on dq for vxorpd: this is a 512-bit operation
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl4D_zero_evex(vecY dst, immD0 zero) %{
  predicate(n->as_Vector()->length() == 4 && UseAVX > 2);
  match(Set dst (ReplicateD zero));
  format %{ "vpxor  $dst k0,$dst,$dst\t! replicate4D zero" %}
  ins_encode %{
    // Use vpxor in place of vxorpd since EVEX has a constriant on dq for vxorpd: this is a 512-bit operation
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

instruct Repl8D_zero_evex(vecZ dst, immD0 zero) %{
  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
  match(Set dst (ReplicateD zero));
  format %{ "vpxor  $dst k0,$dst,$dst,vect512\t! replicate8D zero" %}
  ins_encode %{
    // Use vpxor in place of vxorpd since EVEX has a constriant on dq for vxorpd: this is a 512-bit operation
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( fpu_reg_reg );
%}

// ====================REDUCTION ARITHMETIC=======================================

instruct rsadd2I_reduction_reg(rRegI dst, rRegI src1, vecD src2, regF tmp, regF tmp2) %{
  predicate(UseSSE > 2 && UseAVX == 0);
  match(Set dst (AddReductionVI src1 src2));
  effect(TEMP tmp2, TEMP tmp);
  format %{ "movdqu  $tmp2,$src2\n\t"
            "phaddd  $tmp2,$tmp2\n\t"
            "movd    $tmp,$src1\n\t"
            "paddd   $tmp,$tmp2\n\t"
            "movd    $dst,$tmp\t! add reduction2I" %}
  ins_encode %{
    __ movdqu($tmp2$$XMMRegister, $src2$$XMMRegister);
    __ phaddd($tmp2$$XMMRegister, $tmp2$$XMMRegister);
    __ movdl($tmp$$XMMRegister, $src1$$Register);
    __ paddd($tmp$$XMMRegister, $tmp2$$XMMRegister);
    __ movdl($dst$$Register, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd2I_reduction_reg(rRegI dst, rRegI src1, vecD src2, regF tmp, regF tmp2) %{
  predicate(VM_Version::supports_avxonly());
  match(Set dst (AddReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "vphaddd  $tmp,$src2,$src2\n\t"
            "movd     $tmp2,$src1\n\t"
            "vpaddd   $tmp2,$tmp2,$tmp\n\t"
            "movd     $dst,$tmp2\t! add reduction2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vphaddd($tmp$$XMMRegister, $src2$$XMMRegister, $src2$$XMMRegister, vector_len);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpaddd($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, vector_len);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd2I_reduction_reg_evex(rRegI dst, rRegI src1, vecD src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (AddReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "pshufd  $tmp2,$src2,0x1\n\t"
            "vpaddd  $tmp,$src2,$tmp2\n\t"
            "movd    $tmp2,$src1\n\t"
            "vpaddd  $tmp2,$tmp,$tmp2\n\t"
            "movd    $dst,$tmp2\t! add reduction2I" %}
  ins_encode %{
    int vector_len = 0;
    __ pshufd($tmp2$$XMMRegister, $src2$$XMMRegister, 0x1);
    __ vpaddd($tmp$$XMMRegister, $src2$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpaddd($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rsadd4I_reduction_reg(rRegI dst, rRegI src1, vecX src2, regF tmp, regF tmp2) %{
  predicate(UseSSE > 2 && UseAVX == 0);
  match(Set dst (AddReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "movdqu  $tmp,$src2\n\t"
            "phaddd  $tmp,$tmp\n\t"
            "phaddd  $tmp,$tmp\n\t"
            "movd    $tmp2,$src1\n\t"
            "paddd   $tmp2,$tmp\n\t"
            "movd    $dst,$tmp2\t! add reduction4I" %}
  ins_encode %{
    __ movdqu($tmp$$XMMRegister, $src2$$XMMRegister);
    __ phaddd($tmp$$XMMRegister, $tmp$$XMMRegister);
    __ phaddd($tmp$$XMMRegister, $tmp$$XMMRegister);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ paddd($tmp2$$XMMRegister, $tmp$$XMMRegister);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd4I_reduction_reg(rRegI dst, rRegI src1, vecX src2, regF tmp, regF tmp2) %{
  predicate(VM_Version::supports_avxonly());
  match(Set dst (AddReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "vphaddd  $tmp,$src2,$src2\n\t"
            "vphaddd  $tmp,$tmp,$tmp\n\t"
            "movd     $tmp2,$src1\n\t"
            "vpaddd   $tmp2,$tmp2,$tmp\n\t"
            "movd     $dst,$tmp2\t! add reduction4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vphaddd($tmp$$XMMRegister, $src2$$XMMRegister, $src2$$XMMRegister, vector_len);
    __ vphaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp$$XMMRegister, vector_len);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpaddd($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, vector_len);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd4I_reduction_reg_evex(rRegI dst, rRegI src1, vecX src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (AddReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "pshufd  $tmp2,$src2,0xE\n\t"
            "vpaddd  $tmp,$src2,$tmp2\n\t"
            "pshufd  $tmp2,$tmp,0x1\n\t"
            "vpaddd  $tmp,$tmp,$tmp2\n\t"
            "movd    $tmp2,$src1\n\t"
            "vpaddd  $tmp2,$tmp,$tmp2\n\t"
            "movd    $dst,$tmp2\t! add reduction4I" %}
  ins_encode %{
    int vector_len = 0;
    __ pshufd($tmp2$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vpaddd($tmp$$XMMRegister, $src2$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0x1);
    __ vpaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpaddd($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd8I_reduction_reg(rRegI dst, rRegI src1, vecY src2, regF tmp, regF tmp2) %{
  predicate(VM_Version::supports_avxonly());
  match(Set dst (AddReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "vphaddd  $tmp,$src2,$src2\n\t"
            "vphaddd  $tmp,$tmp,$tmp2\n\t"
            "vextracti128_high  $tmp2,$tmp\n\t"
            "vpaddd   $tmp,$tmp,$tmp2\n\t"
            "movd     $tmp2,$src1\n\t"
            "vpaddd   $tmp2,$tmp2,$tmp\n\t"
            "movd     $dst,$tmp2\t! add reduction8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vphaddd($tmp$$XMMRegister, $src2$$XMMRegister, $src2$$XMMRegister, vector_len);
    __ vphaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ vextracti128_high($tmp2$$XMMRegister, $tmp$$XMMRegister);
    __ vpaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpaddd($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd8I_reduction_reg_evex(rRegI dst, rRegI src1, vecY src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (AddReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "vextracti128_high  $tmp,$src2\n\t"
            "vpaddd  $tmp,$tmp,$src2\n\t"
            "pshufd  $tmp2,$tmp,0xE\n\t"
            "vpaddd  $tmp,$tmp,$tmp2\n\t"
            "pshufd  $tmp2,$tmp,0x1\n\t"
            "vpaddd  $tmp,$tmp,$tmp2\n\t"
            "movd    $tmp2,$src1\n\t"
            "vpaddd  $tmp2,$tmp,$tmp2\n\t"
            "movd    $dst,$tmp2\t! add reduction8I" %}
  ins_encode %{
    int vector_len = 0;
    __ vextracti128_high($tmp$$XMMRegister, $src2$$XMMRegister);
    __ vpaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $src2$$XMMRegister, vector_len);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0xE);
    __ vpaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0x1);
    __ vpaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpaddd($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd16I_reduction_reg_evex(rRegI dst, rRegI src1, vecZ src2, regF tmp, regF tmp2, regF tmp3) %{
  predicate(UseAVX > 2);
  match(Set dst (AddReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2, TEMP tmp3);
  format %{ "vextracti64x4_high  $tmp3,$src2\n\t"
            "vpaddd  $tmp3,$tmp3,$src2\n\t"
            "vextracti128_high  $tmp,$tmp3\n\t"
            "vpaddd  $tmp,$tmp,$tmp3\n\t"
            "pshufd  $tmp2,$tmp,0xE\n\t"
            "vpaddd  $tmp,$tmp,$tmp2\n\t"
            "pshufd  $tmp2,$tmp,0x1\n\t"
            "vpaddd  $tmp,$tmp,$tmp2\n\t"
            "movd    $tmp2,$src1\n\t"
            "vpaddd  $tmp2,$tmp,$tmp2\n\t"
            "movd    $dst,$tmp2\t! mul reduction16I" %}
  ins_encode %{
    __ vextracti64x4_high($tmp3$$XMMRegister, $src2$$XMMRegister);
    __ vpaddd($tmp3$$XMMRegister, $tmp3$$XMMRegister, $src2$$XMMRegister, 1);
    __ vextracti128_high($tmp$$XMMRegister, $tmp3$$XMMRegister);
    __ vpaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp3$$XMMRegister, 0);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0xE);
    __ vpaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0x1);
    __ vpaddd($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpaddd($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

#ifdef _LP64
instruct rvadd2L_reduction_reg(rRegL dst, rRegL src1, vecX src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (AddReductionVL src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "pshufd  $tmp2,$src2,0xE\n\t"
            "vpaddq  $tmp,$src2,$tmp2\n\t"
            "movdq   $tmp2,$src1\n\t"
            "vpaddq  $tmp2,$tmp,$tmp2\n\t"
            "movdq   $dst,$tmp2\t! add reduction2L" %}
  ins_encode %{
    __ pshufd($tmp2$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vpaddq($tmp$$XMMRegister, $src2$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ movdq($tmp2$$XMMRegister, $src1$$Register);
    __ vpaddq($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ movdq($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd4L_reduction_reg(rRegL dst, rRegL src1, vecY src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (AddReductionVL src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "vextracti128_high  $tmp,$src2\n\t"
            "vpaddq  $tmp2,$tmp,$src2\n\t"
            "pshufd  $tmp,$tmp2,0xE\n\t"
            "vpaddq  $tmp2,$tmp2,$tmp\n\t"
            "movdq   $tmp,$src1\n\t"
            "vpaddq  $tmp2,$tmp2,$tmp\n\t"
            "movdq   $dst,$tmp2\t! add reduction4L" %}
  ins_encode %{
    __ vextracti128_high($tmp$$XMMRegister, $src2$$XMMRegister);
    __ vpaddq($tmp2$$XMMRegister, $tmp$$XMMRegister, $src2$$XMMRegister, 0);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vpaddq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ movdq($tmp$$XMMRegister, $src1$$Register);
    __ vpaddq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ movdq($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd8L_reduction_reg(rRegL dst, rRegL src1, vecZ src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (AddReductionVL src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "vextracti64x4_high  $tmp2,$src2\n\t"
            "vpaddq  $tmp2,$tmp2,$src2\n\t"
            "vextracti128_high  $tmp,$tmp2\n\t"
            "vpaddq  $tmp2,$tmp2,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0xE\n\t"
            "vpaddq  $tmp2,$tmp2,$tmp\n\t"
            "movdq   $tmp,$src1\n\t"
            "vpaddq  $tmp2,$tmp2,$tmp\n\t"
            "movdq   $dst,$tmp2\t! add reduction8L" %}
  ins_encode %{
    __ vextracti64x4_high($tmp2$$XMMRegister, $src2$$XMMRegister);
    __ vpaddq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $src2$$XMMRegister, 1);
    __ vextracti128_high($tmp$$XMMRegister, $tmp2$$XMMRegister);
    __ vpaddq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vpaddq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ movdq($tmp$$XMMRegister, $src1$$Register);
    __ vpaddq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ movdq($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}
#endif

instruct rsadd2F_reduction_reg(regF dst, vecD src2, regF tmp) %{
  predicate(UseSSE >= 1 && UseAVX == 0);
  match(Set dst (AddReductionVF dst src2));
  effect(TEMP dst, TEMP tmp);
  format %{ "addss   $dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "addss   $dst,$tmp\t! add reduction2F" %}
  ins_encode %{
    __ addss($dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ addss($dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd2F_reduction_reg(regF dst, vecD src2, regF tmp) %{
  predicate(UseAVX > 0);
  match(Set dst (AddReductionVF dst src2));
  effect(TEMP dst, TEMP tmp);
  format %{ "vaddss  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "vaddss  $dst,$dst,$tmp\t! add reduction2F" %}
  ins_encode %{
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rsadd4F_reduction_reg(regF dst, vecX src2, regF tmp) %{
  predicate(UseSSE >= 1 && UseAVX == 0);
  match(Set dst (AddReductionVF dst src2));
  effect(TEMP dst, TEMP tmp);
  format %{ "addss   $dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "addss   $dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x02\n\t"
            "addss   $dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x03\n\t"
            "addss   $dst,$tmp\t! add reduction4F" %}
  ins_encode %{
    __ addss($dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ addss($dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x02);
    __ addss($dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x03);
    __ addss($dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd4F_reduction_reg(regF dst, vecX src2, regF tmp) %{
  predicate(UseAVX > 0);
  match(Set dst (AddReductionVF dst src2));
  effect(TEMP tmp, TEMP dst);
  format %{ "vaddss  $dst,dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x02\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x03\n\t"
            "vaddss  $dst,$dst,$tmp\t! add reduction4F" %}
  ins_encode %{
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x02);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x03);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct radd8F_reduction_reg(regF dst, vecY src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 0);
  match(Set dst (AddReductionVF dst src2));
  effect(TEMP tmp, TEMP dst, TEMP tmp2);
  format %{ "vaddss  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x02\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x03\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "vextractf128_high  $tmp2,$src2\n\t"
            "vaddss  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0x01\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x02\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x03\n\t"
            "vaddss  $dst,$dst,$tmp\t! add reduction8F" %}
  ins_encode %{
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x02);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x03);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf128_high($tmp2$$XMMRegister, $src2$$XMMRegister);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x01);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x02);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x03);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct radd16F_reduction_reg(regF dst, vecZ src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (AddReductionVF dst src2));
  effect(TEMP tmp, TEMP dst, TEMP tmp2);
  format %{ "vaddss  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x02\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x03\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x1\n\t"
            "vaddss  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0x01\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x02\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x03\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x2\n\t"
            "vaddss  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0x01\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x02\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x03\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x3\n\t"
            "vaddss  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0x01\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x02\n\t"
            "vaddss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x03\n\t"
            "vaddss  $dst,$dst,$tmp\t! add reduction16F" %}
  ins_encode %{
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x02);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x03);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x1);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x01);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x02);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x03);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x2);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x01);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x02);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x03);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x3);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x01);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x02);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x03);
    __ vaddss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rsadd2D_reduction_reg(regD dst, vecX src2, regD tmp) %{
  predicate(UseSSE >= 1 && UseAVX == 0);
  match(Set dst (AddReductionVD dst src2));
  effect(TEMP tmp, TEMP dst);
  format %{ "addsd   $dst,$src2\n\t"
            "pshufd  $tmp,$src2,0xE\n\t"
            "addsd   $dst,$tmp\t! add reduction2D" %}
  ins_encode %{
    __ addsd($dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ addsd($dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd2D_reduction_reg(regD dst, vecX src2, regD tmp) %{
  predicate(UseAVX > 0);
  match(Set dst (AddReductionVD dst src2));
  effect(TEMP tmp, TEMP dst);
  format %{ "vaddsd  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0xE\n\t"
            "vaddsd  $dst,$dst,$tmp\t! add reduction2D" %}
  ins_encode %{
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd4D_reduction_reg(regD dst, vecY src2, regD tmp, regD tmp2) %{
  predicate(UseAVX > 0);
  match(Set dst (AddReductionVD dst src2));
  effect(TEMP tmp, TEMP dst, TEMP tmp2);
  format %{ "vaddsd  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0xE\n\t"
            "vaddsd  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x1\n\t"
            "vaddsd  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0xE\n\t"
            "vaddsd  $dst,$dst,$tmp\t! add reduction4D" %}
  ins_encode %{
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x1);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvadd8D_reduction_reg(regD dst, vecZ src2, regD tmp, regD tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (AddReductionVD dst src2));
  effect(TEMP tmp, TEMP dst, TEMP tmp2);
  format %{ "vaddsd  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0xE\n\t"
            "vaddsd  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x1\n\t"
            "vaddsd  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0xE\n\t"
            "vaddsd  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x2\n\t"
            "vaddsd  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0xE\n\t"
            "vaddsd  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x3\n\t"
            "vaddsd  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0xE\n\t"
            "vaddsd  $dst,$dst,$tmp\t! add reduction8D" %}
  ins_encode %{
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x1);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x2);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x3);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vaddsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rsmul2I_reduction_reg(rRegI dst, rRegI src1, vecD src2, regF tmp, regF tmp2) %{
  predicate(UseSSE > 3 && UseAVX == 0);
  match(Set dst (MulReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "pshufd  $tmp2,$src2,0x1\n\t"
            "pmulld  $tmp2,$src2\n\t"
            "movd    $tmp,$src1\n\t"
            "pmulld  $tmp2,$tmp\n\t"
            "movd    $dst,$tmp2\t! mul reduction2I" %}
  ins_encode %{
    __ pshufd($tmp2$$XMMRegister, $src2$$XMMRegister, 0x1);
    __ pmulld($tmp2$$XMMRegister, $src2$$XMMRegister);
    __ movdl($tmp$$XMMRegister, $src1$$Register);
    __ pmulld($tmp2$$XMMRegister, $tmp$$XMMRegister);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul2I_reduction_reg(rRegI dst, rRegI src1, vecD src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 0);
  match(Set dst (MulReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "pshufd   $tmp2,$src2,0x1\n\t"
            "vpmulld  $tmp,$src2,$tmp2\n\t"
            "movd     $tmp2,$src1\n\t"
            "vpmulld  $tmp2,$tmp,$tmp2\n\t"
            "movd     $dst,$tmp2\t! mul reduction2I" %}
  ins_encode %{
    int vector_len = 0;
    __ pshufd($tmp2$$XMMRegister, $src2$$XMMRegister, 0x1);
    __ vpmulld($tmp$$XMMRegister, $src2$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpmulld($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rsmul4I_reduction_reg(rRegI dst, rRegI src1, vecX src2, regF tmp, regF tmp2) %{
  predicate(UseSSE > 3 && UseAVX == 0);
  match(Set dst (MulReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "pshufd  $tmp2,$src2,0xE\n\t"
            "pmulld  $tmp2,$src2\n\t"
            "pshufd  $tmp,$tmp2,0x1\n\t"
            "pmulld  $tmp2,$tmp\n\t"
            "movd    $tmp,$src1\n\t"
            "pmulld  $tmp2,$tmp\n\t"
            "movd    $dst,$tmp2\t! mul reduction4I" %}
  ins_encode %{
    __ pshufd($tmp2$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ pmulld($tmp2$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x1);
    __ pmulld($tmp2$$XMMRegister, $tmp$$XMMRegister);
    __ movdl($tmp$$XMMRegister, $src1$$Register);
    __ pmulld($tmp2$$XMMRegister, $tmp$$XMMRegister);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul4I_reduction_reg(rRegI dst, rRegI src1, vecX src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 0);
  match(Set dst (MulReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "pshufd   $tmp2,$src2,0xE\n\t"
            "vpmulld  $tmp,$src2,$tmp2\n\t"
            "pshufd   $tmp2,$tmp,0x1\n\t"
            "vpmulld  $tmp,$tmp,$tmp2\n\t"
            "movd     $tmp2,$src1\n\t"
            "vpmulld  $tmp2,$tmp,$tmp2\n\t"
            "movd     $dst,$tmp2\t! mul reduction4I" %}
  ins_encode %{
    int vector_len = 0;
    __ pshufd($tmp2$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vpmulld($tmp$$XMMRegister, $src2$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0x1);
    __ vpmulld($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpmulld($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul8I_reduction_reg(rRegI dst, rRegI src1, vecY src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 0);
  match(Set dst (MulReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "vextracti128_high  $tmp,$src2\n\t"
            "vpmulld  $tmp,$tmp,$src2\n\t"
            "pshufd   $tmp2,$tmp,0xE\n\t"
            "vpmulld  $tmp,$tmp,$tmp2\n\t"
            "pshufd   $tmp2,$tmp,0x1\n\t"
            "vpmulld  $tmp,$tmp,$tmp2\n\t"
            "movd     $tmp2,$src1\n\t"
            "vpmulld  $tmp2,$tmp,$tmp2\n\t"
            "movd     $dst,$tmp2\t! mul reduction8I" %}
  ins_encode %{
    int vector_len = 0;
    __ vextracti128_high($tmp$$XMMRegister, $src2$$XMMRegister);
    __ vpmulld($tmp$$XMMRegister, $tmp$$XMMRegister, $src2$$XMMRegister, vector_len);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0xE);
    __ vpmulld($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0x1);
    __ vpmulld($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpmulld($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, vector_len);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul16I_reduction_reg(rRegI dst, rRegI src1, vecZ src2, regF tmp, regF tmp2, regF tmp3) %{
  predicate(UseAVX > 2);
  match(Set dst (MulReductionVI src1 src2));
  effect(TEMP tmp, TEMP tmp2, TEMP tmp3);
  format %{ "vextracti64x4_high  $tmp3,$src2\n\t"
            "vpmulld  $tmp3,$tmp3,$src2\n\t"
            "vextracti128_high  $tmp,$tmp3\n\t"
            "vpmulld  $tmp,$tmp,$src2\n\t"
            "pshufd   $tmp2,$tmp,0xE\n\t"
            "vpmulld  $tmp,$tmp,$tmp2\n\t"
            "pshufd   $tmp2,$tmp,0x1\n\t"
            "vpmulld  $tmp,$tmp,$tmp2\n\t"
            "movd     $tmp2,$src1\n\t"
            "vpmulld  $tmp2,$tmp,$tmp2\n\t"
            "movd     $dst,$tmp2\t! mul reduction16I" %}
  ins_encode %{
    __ vextracti64x4_high($tmp3$$XMMRegister, $src2$$XMMRegister);
    __ vpmulld($tmp3$$XMMRegister, $tmp3$$XMMRegister, $src2$$XMMRegister, 1);
    __ vextracti128_high($tmp$$XMMRegister, $tmp3$$XMMRegister);
    __ vpmulld($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp3$$XMMRegister, 0);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0xE);
    __ vpmulld($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ pshufd($tmp2$$XMMRegister, $tmp$$XMMRegister, 0x1);
    __ vpmulld($tmp$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ movdl($tmp2$$XMMRegister, $src1$$Register);
    __ vpmulld($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ movdl($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

#ifdef _LP64
instruct rvmul2L_reduction_reg(rRegL dst, rRegL src1, vecX src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2 && VM_Version::supports_avx512dq());
  match(Set dst (MulReductionVL src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "pshufd   $tmp2,$src2,0xE\n\t"
            "vpmullq  $tmp,$src2,$tmp2\n\t"
            "movdq    $tmp2,$src1\n\t"
            "vpmullq  $tmp2,$tmp,$tmp2\n\t"
            "movdq    $dst,$tmp2\t! mul reduction2L" %}
  ins_encode %{
    __ pshufd($tmp2$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vpmullq($tmp$$XMMRegister, $src2$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ movdq($tmp2$$XMMRegister, $src1$$Register);
    __ vpmullq($tmp2$$XMMRegister, $tmp$$XMMRegister, $tmp2$$XMMRegister, 0);
    __ movdq($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul4L_reduction_reg(rRegL dst, rRegL src1, vecY src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2 && VM_Version::supports_avx512dq());
  match(Set dst (MulReductionVL src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "vextracti128_high  $tmp,$src2\n\t"
            "vpmullq  $tmp2,$tmp,$src2\n\t"
            "pshufd   $tmp,$tmp2,0xE\n\t"
            "vpmullq  $tmp2,$tmp2,$tmp\n\t"
            "movdq    $tmp,$src1\n\t"
            "vpmullq  $tmp2,$tmp2,$tmp\n\t"
            "movdq    $dst,$tmp2\t! mul reduction4L" %}
  ins_encode %{
    __ vextracti128_high($tmp$$XMMRegister, $src2$$XMMRegister);
    __ vpmullq($tmp2$$XMMRegister, $tmp$$XMMRegister, $src2$$XMMRegister, 0);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vpmullq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ movdq($tmp$$XMMRegister, $src1$$Register);
    __ vpmullq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ movdq($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul8L_reduction_reg(rRegL dst, rRegL src1, vecZ src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2 && VM_Version::supports_avx512dq());
  match(Set dst (MulReductionVL src1 src2));
  effect(TEMP tmp, TEMP tmp2);
  format %{ "vextracti64x4_high  $tmp2,$src2\n\t"
            "vpmullq  $tmp2,$tmp2,$src2\n\t"
            "vextracti128_high  $tmp,$tmp2\n\t"
            "vpmullq  $tmp2,$tmp2,$tmp\n\t"
            "pshufd   $tmp,$tmp2,0xE\n\t"
            "vpmullq  $tmp2,$tmp2,$tmp\n\t"
            "movdq    $tmp,$src1\n\t"
            "vpmullq  $tmp2,$tmp2,$tmp\n\t"
            "movdq    $dst,$tmp2\t! mul reduction8L" %}
  ins_encode %{
    __ vextracti64x4_high($tmp2$$XMMRegister, $src2$$XMMRegister);
    __ vpmullq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $src2$$XMMRegister, 1);
    __ vextracti128_high($tmp$$XMMRegister, $tmp2$$XMMRegister);
    __ vpmullq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vpmullq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ movdq($tmp$$XMMRegister, $src1$$Register);
    __ vpmullq($tmp2$$XMMRegister, $tmp2$$XMMRegister, $tmp$$XMMRegister, 0);
    __ movdq($dst$$Register, $tmp2$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}
#endif

instruct rsmul2F_reduction(regF dst, vecD src2, regF tmp) %{
  predicate(UseSSE >= 1 && UseAVX == 0);
  match(Set dst (MulReductionVF dst src2));
  effect(TEMP dst, TEMP tmp);
  format %{ "mulss   $dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "mulss   $dst,$tmp\t! mul reduction2F" %}
  ins_encode %{
    __ mulss($dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ mulss($dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul2F_reduction_reg(regF dst, vecD src2, regF tmp) %{
  predicate(UseAVX > 0);
  match(Set dst (MulReductionVF dst src2));
  effect(TEMP tmp, TEMP dst);
  format %{ "vmulss  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "vmulss  $dst,$dst,$tmp\t! mul reduction2F" %}
  ins_encode %{
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rsmul4F_reduction_reg(regF dst, vecX src2, regF tmp) %{
  predicate(UseSSE >= 1 && UseAVX == 0);
  match(Set dst (MulReductionVF dst src2));
  effect(TEMP dst, TEMP tmp);
  format %{ "mulss   $dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "mulss   $dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x02\n\t"
            "mulss   $dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x03\n\t"
            "mulss   $dst,$tmp\t! mul reduction4F" %}
  ins_encode %{
    __ mulss($dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ mulss($dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x02);
    __ mulss($dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x03);
    __ mulss($dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul4F_reduction_reg(regF dst, vecX src2, regF tmp) %{
  predicate(UseAVX > 0);
  match(Set dst (MulReductionVF dst src2));
  effect(TEMP tmp, TEMP dst);
  format %{ "vmulss  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x02\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x03\n\t"
            "vmulss  $dst,$dst,$tmp\t! mul reduction4F" %}
  ins_encode %{
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x02);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x03);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul8F_reduction_reg(regF dst, vecY src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 0);
  match(Set dst (MulReductionVF dst src2));
  effect(TEMP tmp, TEMP dst, TEMP tmp2);
  format %{ "vmulss  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x02\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x03\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "vextractf128_high  $tmp2,$src2\n\t"
            "vmulss  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0x01\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x02\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x03\n\t"
            "vmulss  $dst,$dst,$tmp\t! mul reduction8F" %}
  ins_encode %{
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x02);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x03);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf128_high($tmp2$$XMMRegister, $src2$$XMMRegister);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x01);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x02);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x03);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul16F_reduction_reg(regF dst, vecZ src2, regF tmp, regF tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (MulReductionVF dst src2));
  effect(TEMP tmp, TEMP dst, TEMP tmp2);
  format %{ "vmulss  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0x01\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x02\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$src2,0x03\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x1\n\t"
            "vmulss  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0x01\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x02\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x03\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x2\n\t"
            "vmulss  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0x01\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x02\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x03\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x3\n\t"
            "vmulss  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0x01\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x02\n\t"
            "vmulss  $dst,$dst,$tmp\n\t"
            "pshufd  $tmp,$tmp2,0x03\n\t"
            "vmulss  $dst,$dst,$tmp\t! mul reduction16F" %}
  ins_encode %{
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x01);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x02);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0x03);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x1);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x01);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x02);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x03);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x2);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x01);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x02);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x03);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x3);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x01);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x02);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0x03);
    __ vmulss($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rsmul2D_reduction_reg(regD dst, vecX src2, regD tmp) %{
  predicate(UseSSE >= 1 && UseAVX == 0);
  match(Set dst (MulReductionVD dst src2));
  effect(TEMP dst, TEMP tmp);
  format %{ "mulsd   $dst,$src2\n\t"
            "pshufd  $tmp,$src2,0xE\n\t"
            "mulsd   $dst,$tmp\t! mul reduction2D" %}
  ins_encode %{
    __ mulsd($dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ mulsd($dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul2D_reduction_reg(regD dst, vecX src2, regD tmp) %{
  predicate(UseAVX > 0);
  match(Set dst (MulReductionVD dst src2));
  effect(TEMP tmp, TEMP dst);
  format %{ "vmulsd  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0xE\n\t"
            "vmulsd  $dst,$dst,$tmp\t! mul reduction2D" %}
  ins_encode %{
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul4D_reduction_reg(regD dst, vecY src2, regD tmp, regD tmp2) %{
  predicate(UseAVX > 0);
  match(Set dst (MulReductionVD dst src2));
  effect(TEMP tmp, TEMP dst, TEMP tmp2);
  format %{ "vmulsd  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0xE\n\t"
            "vmulsd  $dst,$dst,$tmp\n\t"
            "vextractf128_high  $tmp2,$src2\n\t"
            "vmulsd  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0xE\n\t"
            "vmulsd  $dst,$dst,$tmp\t! mul reduction4D" %}
  ins_encode %{
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf128_high($tmp2$$XMMRegister, $src2$$XMMRegister);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct rvmul8D_reduction_reg(regD dst, vecZ src2, regD tmp, regD tmp2) %{
  predicate(UseAVX > 2);
  match(Set dst (MulReductionVD dst src2));
  effect(TEMP tmp, TEMP dst, TEMP tmp2);
  format %{ "vmulsd  $dst,$dst,$src2\n\t"
            "pshufd  $tmp,$src2,0xE\n\t"
            "vmulsd  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x1\n\t"
            "vmulsd  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$src2,0xE\n\t"
            "vmulsd  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x2\n\t"
            "vmulsd  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0xE\n\t"
            "vmulsd  $dst,$dst,$tmp\n\t"
            "vextractf32x4  $tmp2,$src2,0x3\n\t"
            "vmulsd  $dst,$dst,$tmp2\n\t"
            "pshufd  $tmp,$tmp2,0xE\n\t"
            "vmulsd  $dst,$dst,$tmp\t! mul reduction8D" %}
  ins_encode %{
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $src2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $src2$$XMMRegister, 0xE);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x1);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x2);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
    __ vextractf32x4($tmp2$$XMMRegister, $src2$$XMMRegister, 0x3);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp2$$XMMRegister);
    __ pshufd($tmp$$XMMRegister, $tmp2$$XMMRegister, 0xE);
    __ vmulsd($dst$$XMMRegister, $dst$$XMMRegister, $tmp$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

// ====================VECTOR ARITHMETIC=======================================

// --------------------------------- ADD --------------------------------------

// Bytes vector add
instruct vadd4B(vecS dst, vecS src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (AddVB dst src));
  format %{ "paddb   $dst,$src\t! add packed4B" %}
  ins_encode %{
    __ paddb($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4B_reg_avx(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (AddVB src1 src2));
  format %{ "vpaddb  $dst,$src1,$src2\t! add packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4B_reg_evex(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (AddVB src1 src2));
  format %{ "vpaddb  $dst,$src1,$src2\t! add packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4B_reg_evex_special(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (AddVB dst src2));
  effect(TEMP src1);
  format %{ "vpaddb  $dst,$dst,$src2\t! add packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4B_mem_avx(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (AddVB src (LoadVector mem)));
  format %{ "vpaddb  $dst,$src,$mem\t! add packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4B_mem_evex(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (AddVB src (LoadVector mem)));
  format %{ "vpaddb  $dst,$src,$mem\t! add packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4B_mem_evex_special(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (AddVB dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpaddb  $dst,$src,$mem\t! add packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8B(vecD dst, vecD src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (AddVB dst src));
  format %{ "paddb   $dst,$src\t! add packed8B" %}
  ins_encode %{
    __ paddb($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8B_reg_avx(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (AddVB src1 src2));
  format %{ "vpaddb  $dst,$src1,$src2\t! add packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8B_reg_evex(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (AddVB src1 src2));
  format %{ "vpaddb  $dst,$src1,$src2\t! add packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8B_reg_evex_special(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (AddVB dst src2));
  effect(TEMP src1);
  format %{ "vpaddb  $dst,$dst,$src2\t! add packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8B_mem_avx(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (AddVB src (LoadVector mem)));
  format %{ "vpaddb  $dst,$src,$mem\t! add packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8B_mem_evex(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (AddVB src (LoadVector mem)));
  format %{ "vpaddb  $dst,$src,$mem\t! add packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8B_mem_evex_special(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (AddVB dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpaddb  $dst,$src,$mem\t! add packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16B(vecX dst, vecX src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 16);
  match(Set dst (AddVB dst src));
  format %{ "paddb   $dst,$src\t! add packed16B" %}
  ins_encode %{
    __ paddb($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16B_reg_avx(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 16);
  match(Set dst (AddVB src1 src2));
  format %{ "vpaddb  $dst,$src1,$src2\t! add packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16B_reg_evex(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (AddVB src1 src2));
  format %{ "vpaddb  $dst,$src1,$src2\t! add packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16B_reg_evex_special(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (AddVB dst src2));
  effect(TEMP src1);
  format %{ "vpaddb  $dst,$dst,$src2\t! add packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16B_mem_avx(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 16);
  match(Set dst (AddVB src (LoadVector mem)));
  format %{ "vpaddb  $dst,$src,$mem\t! add packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16B_mem_evex(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (AddVB src (LoadVector mem)));
  format %{ "vpaddb  $dst,$src,$mem\t! add packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16B_mem_evex_special(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (AddVB dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpaddb  $dst,$src,$mem\t! add packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd32B_reg_avx(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 32);
  match(Set dst (AddVB src1 src2));
  format %{ "vpaddb  $dst,$src1,$src2\t! add packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd32B_reg_evex(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (AddVB src1 src2));
  format %{ "vpaddb  $dst,$src1,$src2\t! add packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd32B_reg_evex_special(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 32);
  match(Set dst (AddVB dst src2));
  effect(TEMP src1);
  format %{ "vpaddb  $dst,$dst,$src2\t! add packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd32B_mem_avx(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 32);
  match(Set dst (AddVB src (LoadVector mem)));
  format %{ "vpaddb  $dst,$src,$mem\t! add packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd32B_mem_evex(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (AddVB src (LoadVector mem)));
  format %{ "vpaddb  $dst,$src,$mem\t! add packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd32B_mem_evex_special(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (AddVB dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpaddb  $dst,$src,$mem\t! add packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd64B_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 64);
  match(Set dst (AddVB src1 src2));
  format %{ "vpaddb  $dst,$src1,$src2\t! add packed64B" %}
  ins_encode %{
    int vector_len = 2;
    __ vpaddb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd64B_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 64);
  match(Set dst (AddVB src (LoadVector mem)));
  format %{ "vpaddb  $dst,$src,$mem\t! add packed64B" %}
  ins_encode %{
    int vector_len = 2;
    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Shorts/Chars vector add
instruct vadd2S(vecS dst, vecS src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 2);
  match(Set dst (AddVS dst src));
  format %{ "paddw   $dst,$src\t! add packed2S" %}
  ins_encode %{
    __ paddw($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2S_reg_avx(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (AddVS src1 src2));
  format %{ "vpaddw  $dst,$src1,$src2\t! add packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2S_reg_evex(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (AddVS src1 src2));
  format %{ "vpaddw  $dst,$src1,$src2\t! add packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2S_reg_evex_special(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (AddVS dst src2));
  effect(TEMP src1);
  format %{ "vpaddw  $dst,$dst,$src2\t! add packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2S_mem_avx(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (AddVS src (LoadVector mem)));
  format %{ "vpaddw  $dst,$src,$mem\t! add packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2S_mem_evex(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (AddVS src (LoadVector mem)));
  format %{ "vpaddw  $dst,$src,$mem\t! add packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2S_mem_evex_special(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (AddVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpaddw  $dst,$src,$mem\t! add packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4S(vecD dst, vecD src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (AddVS dst src));
  format %{ "paddw   $dst,$src\t! add packed4S" %}
  ins_encode %{
    __ paddw($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4S_reg_avx(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (AddVS src1 src2));
  format %{ "vpaddw  $dst,$src1,$src2\t! add packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4S_reg_evex(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (AddVS src1 src2));
  format %{ "vpaddw  $dst,$src1,$src2\t! add packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4S_reg_evex_special(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (AddVS dst src2));
  effect(TEMP src1);
  format %{ "vpaddw  $dst,$dst,$src2\t! add packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4S_mem_avx(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (AddVS src (LoadVector mem)));
  format %{ "vpaddw  $dst,$src,$mem\t! add packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4S_mem_evex(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (AddVS src (LoadVector mem)));
  format %{ "vpaddw  $dst,$src,$mem\t! add packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4S_mem_evex_special(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (AddVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpaddw  $dst,$src,$mem\t! add packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8S(vecX dst, vecX src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (AddVS dst src));
  format %{ "paddw   $dst,$src\t! add packed8S" %}
  ins_encode %{
    __ paddw($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8S_reg_avx(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (AddVS src1 src2));
  format %{ "vpaddw  $dst,$src1,$src2\t! add packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8S_reg_evex(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (AddVS src1 src2));
  format %{ "vpaddw  $dst,$src1,$src2\t! add packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8S_reg_evex_special(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (AddVS dst src2));
  effect(TEMP src1);
  format %{ "vpaddw  $dst,$dst,$src2\t! add packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8S_mem_avx(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (AddVS src (LoadVector mem)));
  format %{ "vpaddw  $dst,$src,$mem\t! add packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8S_mem_evex(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (AddVS src (LoadVector mem)));
  format %{ "vpaddw  $dst,$src,$mem\t! add packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8S_mem_evex_special(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (AddVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpaddw  $dst,$src,$mem\t! add packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16S_reg_avx(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (AddVS src1 src2));
  format %{ "vpaddw  $dst,$src1,$src2\t! add packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16S_reg_evex(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (AddVS src1 src2));
  format %{ "vpaddw  $dst,$src1,$src2\t! add packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16S_reg_evex_special(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (AddVS dst src2));
  effect(TEMP src1);
  format %{ "vpaddw  $dst,$dst,$src2\t! add packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16S_mem_avx(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (AddVS src (LoadVector mem)));
  format %{ "vpaddw  $dst,$src,$mem\t! add packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16S_mem_evex(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (AddVS src (LoadVector mem)));
  format %{ "vpaddw  $dst,$src,$mem\t! add packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16S_mem_evex_special(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (AddVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpaddw  $dst,$src,$mem\t! add packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd32S_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (AddVS src1 src2));
  format %{ "vpaddw  $dst,$src1,$src2\t! add packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpaddw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd32S_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (AddVS src (LoadVector mem)));
  format %{ "vpaddw  $dst,$src,$mem\t! add packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Integers vector add
instruct vadd2I(vecD dst, vecD src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (AddVI dst src));
  format %{ "paddd   $dst,$src\t! add packed2I" %}
  ins_encode %{
    __ paddd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2I_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (AddVI src1 src2));
  format %{ "vpaddd  $dst,$src1,$src2\t! add packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2I_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (AddVI src (LoadVector mem)));
  format %{ "vpaddd  $dst,$src,$mem\t! add packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4I(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (AddVI dst src));
  format %{ "paddd   $dst,$src\t! add packed4I" %}
  ins_encode %{
    __ paddd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4I_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (AddVI src1 src2));
  format %{ "vpaddd  $dst,$src1,$src2\t! add packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4I_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (AddVI src (LoadVector mem)));
  format %{ "vpaddd  $dst,$src,$mem\t! add packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8I_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (AddVI src1 src2));
  format %{ "vpaddd  $dst,$src1,$src2\t! add packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8I_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (AddVI src (LoadVector mem)));
  format %{ "vpaddd  $dst,$src,$mem\t! add packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16I_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (AddVI src1 src2));
  format %{ "vpaddd  $dst,$src1,$src2\t! add packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpaddd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16I_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (AddVI src (LoadVector mem)));
  format %{ "vpaddd  $dst,$src,$mem\t! add packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpaddd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Longs vector add
instruct vadd2L(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (AddVL dst src));
  format %{ "paddq   $dst,$src\t! add packed2L" %}
  ins_encode %{
    __ paddq($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2L_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (AddVL src1 src2));
  format %{ "vpaddq  $dst,$src1,$src2\t! add packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2L_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (AddVL src (LoadVector mem)));
  format %{ "vpaddq  $dst,$src,$mem\t! add packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpaddq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4L_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 4);
  match(Set dst (AddVL src1 src2));
  format %{ "vpaddq  $dst,$src1,$src2\t! add packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4L_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 4);
  match(Set dst (AddVL src (LoadVector mem)));
  format %{ "vpaddq  $dst,$src,$mem\t! add packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpaddq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8L_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (AddVL src1 src2));
  format %{ "vpaddq  $dst,$src1,$src2\t! add packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpaddq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8L_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (AddVL src (LoadVector mem)));
  format %{ "vpaddq  $dst,$src,$mem\t! add packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpaddq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Floats vector add
instruct vadd2F(vecD dst, vecD src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (AddVF dst src));
  format %{ "addps   $dst,$src\t! add packed2F" %}
  ins_encode %{
    __ addps($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2F_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (AddVF src1 src2));
  format %{ "vaddps  $dst,$src1,$src2\t! add packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vaddps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2F_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (AddVF src (LoadVector mem)));
  format %{ "vaddps  $dst,$src,$mem\t! add packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vaddps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4F(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (AddVF dst src));
  format %{ "addps   $dst,$src\t! add packed4F" %}
  ins_encode %{
    __ addps($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4F_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (AddVF src1 src2));
  format %{ "vaddps  $dst,$src1,$src2\t! add packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vaddps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4F_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (AddVF src (LoadVector mem)));
  format %{ "vaddps  $dst,$src,$mem\t! add packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vaddps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8F_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (AddVF src1 src2));
  format %{ "vaddps  $dst,$src1,$src2\t! add packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vaddps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8F_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (AddVF src (LoadVector mem)));
  format %{ "vaddps  $dst,$src,$mem\t! add packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vaddps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16F_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (AddVF src1 src2));
  format %{ "vaddps  $dst,$src1,$src2\t! add packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vaddps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd16F_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (AddVF src (LoadVector mem)));
  format %{ "vaddps  $dst,$src,$mem\t! add packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vaddps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Doubles vector add
instruct vadd2D(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (AddVD dst src));
  format %{ "addpd   $dst,$src\t! add packed2D" %}
  ins_encode %{
    __ addpd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2D_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (AddVD src1 src2));
  format %{ "vaddpd  $dst,$src1,$src2\t! add packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vaddpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd2D_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (AddVD src (LoadVector mem)));
  format %{ "vaddpd  $dst,$src,$mem\t! add packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vaddpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4D_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (AddVD src1 src2));
  format %{ "vaddpd  $dst,$src1,$src2\t! add packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vaddpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd4D_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (AddVD src (LoadVector mem)));
  format %{ "vaddpd  $dst,$src,$mem\t! add packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vaddpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8D_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (AddVD src1 src2));
  format %{ "vaddpd  $dst,$src1,$src2\t! add packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vaddpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vadd8D_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (AddVD src (LoadVector mem)));
  format %{ "vaddpd  $dst,$src,$mem\t! add packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vaddpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// --------------------------------- SUB --------------------------------------

// Bytes vector sub
instruct vsub4B(vecS dst, vecS src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (SubVB dst src));
  format %{ "psubb   $dst,$src\t! sub packed4B" %}
  ins_encode %{
    __ psubb($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4B_reg_avx(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (SubVB src1 src2));
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4B_reg_evex(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (SubVB src1 src2));
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4B_reg_exex_special(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (SubVB dst src2));
  effect(TEMP src1);
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4B_mem_avx(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (SubVB src (LoadVector mem)));
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4B_mem_evex(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (SubVB src (LoadVector mem)));
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4B_mem_evex_special(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (SubVB dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed4B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8B(vecD dst, vecD src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (SubVB dst src));
  format %{ "psubb   $dst,$src\t! sub packed8B" %}
  ins_encode %{
    __ psubb($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8B_reg_avx(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (SubVB src1 src2));
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8B_reg_evex(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (SubVB src1 src2));
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8B_reg_evex_special(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (SubVB dst src2));
  effect(TEMP src1);
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8B_mem_avx(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (SubVB src (LoadVector mem)));
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8B_mem_evex(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (SubVB src (LoadVector mem)));
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8B_mem_evex_special(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (SubVB dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed8B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16B(vecX dst, vecX src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 16);
  match(Set dst (SubVB dst src));
  format %{ "psubb   $dst,$src\t! sub packed16B" %}
  ins_encode %{
    __ psubb($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16B_reg_avx(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 16);
  match(Set dst (SubVB src1 src2));
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16B_reg_evex(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (SubVB src1 src2));
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16B_reg_evex_special(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (SubVB dst src2));
  effect(TEMP src1);
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16B_mem_avx(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 16);
  match(Set dst (SubVB src (LoadVector mem)));
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16B_mem_evex(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (SubVB src (LoadVector mem)));
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16B_mem_evex_special(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (SubVB dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed16B" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub32B_reg_avx(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 32);
  match(Set dst (SubVB src1 src2));
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub32B_reg_evex(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (SubVB src1 src2));
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub32B_reg_evex_special(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 32);
  match(Set dst (SubVB dst src2));
  effect(TEMP src1);
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub32B_mem_avx(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 32);
  match(Set dst (SubVB src (LoadVector mem)));
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub32B_mem_evex(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (SubVB src (LoadVector mem)));
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub32B_mem_evex_special(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 32);
  match(Set dst (SubVB dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed32B" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub64B_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 64);
  match(Set dst (SubVB src1 src2));
  format %{ "vpsubb  $dst,$src1,$src2\t! sub packed64B" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsubb($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub64B_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 64);
  match(Set dst (SubVB src (LoadVector mem)));
  format %{ "vpsubb  $dst,$src,$mem\t! sub packed64B" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Shorts/Chars vector sub
instruct vsub2S(vecS dst, vecS src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 2);
  match(Set dst (SubVS dst src));
  format %{ "psubw   $dst,$src\t! sub packed2S" %}
  ins_encode %{
    __ psubw($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2S_reg_avx(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (SubVS src1 src2));
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2S_reg_evex(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (SubVS src1 src2));
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2S_reg_evex_special(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (SubVS dst src2));
  effect(TEMP src1);
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2S_mem_avx(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (SubVS src (LoadVector mem)));
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2S_mem_evex(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (SubVS src (LoadVector mem)));
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2S_mem_evex_special(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (SubVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4S(vecD dst, vecD src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (SubVS dst src));
  format %{ "psubw   $dst,$src\t! sub packed4S" %}
  ins_encode %{
    __ psubw($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4S_reg_avx(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (SubVS src1 src2));
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4S_reg_evex(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (SubVS src1 src2));
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4S_reg_evex_special(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (SubVS dst src2));
  effect(TEMP src1);
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4S_mem_avx(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (SubVS src (LoadVector mem)));
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4S_mem_evex(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (SubVS src (LoadVector mem)));
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4S_mem_evex_special(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (SubVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8S(vecX dst, vecX src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (SubVS dst src));
  format %{ "psubw   $dst,$src\t! sub packed8S" %}
  ins_encode %{
    __ psubw($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8S_reg_avx(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (SubVS src1 src2));
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8S_reg_evex(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (SubVS src1 src2));
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8S_reg_evex_special(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (SubVS dst src2));
  effect(TEMP src1);
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8S_mem_avx(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (SubVS src (LoadVector mem)));
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8S_mem_evex(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (SubVS src (LoadVector mem)));
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8S_mem_evex_special(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (SubVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16S_reg_avx(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (SubVS src1 src2));
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16S_reg_evex(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (SubVS src1 src2));
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16S_reg_evex_special(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (SubVS dst src2));
  effect(TEMP src1);
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16S_mem_avx(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (SubVS src (LoadVector mem)));
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16S_mem_evex(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (SubVS src (LoadVector mem)));
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16S_mem_evex_special(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (SubVS dst (LoadVector mem)));
   effect(TEMP src);
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub32S_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (SubVS src1 src2));
  format %{ "vpsubw  $dst,$src1,$src2\t! sub packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsubw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub32S_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (SubVS src (LoadVector mem)));
  format %{ "vpsubw  $dst,$src,$mem\t! sub packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Integers vector sub
instruct vsub2I(vecD dst, vecD src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (SubVI dst src));
  format %{ "psubd   $dst,$src\t! sub packed2I" %}
  ins_encode %{
    __ psubd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2I_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SubVI src1 src2));
  format %{ "vpsubd  $dst,$src1,$src2\t! sub packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2I_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SubVI src (LoadVector mem)));
  format %{ "vpsubd  $dst,$src,$mem\t! sub packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4I(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (SubVI dst src));
  format %{ "psubd   $dst,$src\t! sub packed4I" %}
  ins_encode %{
    __ psubd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4I_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SubVI src1 src2));
  format %{ "vpsubd  $dst,$src1,$src2\t! sub packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4I_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SubVI src (LoadVector mem)));
  format %{ "vpsubd  $dst,$src,$mem\t! sub packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8I_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (SubVI src1 src2));
  format %{ "vpsubd  $dst,$src1,$src2\t! sub packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8I_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (SubVI src (LoadVector mem)));
  format %{ "vpsubd  $dst,$src,$mem\t! sub packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16I_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (SubVI src1 src2));
  format %{ "vpsubd  $dst,$src1,$src2\t! sub packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsubd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16I_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (SubVI src (LoadVector mem)));
  format %{ "vpsubd  $dst,$src,$mem\t! sub packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsubd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Longs vector sub
instruct vsub2L(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (SubVL dst src));
  format %{ "psubq   $dst,$src\t! sub packed2L" %}
  ins_encode %{
    __ psubq($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2L_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SubVL src1 src2));
  format %{ "vpsubq  $dst,$src1,$src2\t! sub packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2L_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SubVL src (LoadVector mem)));
  format %{ "vpsubq  $dst,$src,$mem\t! sub packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsubq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4L_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 4);
  match(Set dst (SubVL src1 src2));
  format %{ "vpsubq  $dst,$src1,$src2\t! sub packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4L_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 4);
  match(Set dst (SubVL src (LoadVector mem)));
  format %{ "vpsubq  $dst,$src,$mem\t! sub packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsubq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8L_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (SubVL src1 src2));
  format %{ "vpsubq  $dst,$src1,$src2\t! sub packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsubq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8L_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (SubVL src (LoadVector mem)));
  format %{ "vpsubq  $dst,$src,$mem\t! sub packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsubq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Floats vector sub
instruct vsub2F(vecD dst, vecD src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (SubVF dst src));
  format %{ "subps   $dst,$src\t! sub packed2F" %}
  ins_encode %{
    __ subps($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2F_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SubVF src1 src2));
  format %{ "vsubps  $dst,$src1,$src2\t! sub packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vsubps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2F_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SubVF src (LoadVector mem)));
  format %{ "vsubps  $dst,$src,$mem\t! sub packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vsubps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4F(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (SubVF dst src));
  format %{ "subps   $dst,$src\t! sub packed4F" %}
  ins_encode %{
    __ subps($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4F_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SubVF src1 src2));
  format %{ "vsubps  $dst,$src1,$src2\t! sub packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vsubps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4F_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SubVF src (LoadVector mem)));
  format %{ "vsubps  $dst,$src,$mem\t! sub packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vsubps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8F_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (SubVF src1 src2));
  format %{ "vsubps  $dst,$src1,$src2\t! sub packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vsubps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8F_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (SubVF src (LoadVector mem)));
  format %{ "vsubps  $dst,$src,$mem\t! sub packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vsubps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16F_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (SubVF src1 src2));
  format %{ "vsubps  $dst,$src1,$src2\t! sub packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vsubps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub16F_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (SubVF src (LoadVector mem)));
  format %{ "vsubps  $dst,$src,$mem\t! sub packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vsubps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Doubles vector sub
instruct vsub2D(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (SubVD dst src));
  format %{ "subpd   $dst,$src\t! sub packed2D" %}
  ins_encode %{
    __ subpd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2D_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SubVD src1 src2));
  format %{ "vsubpd  $dst,$src1,$src2\t! sub packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vsubpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub2D_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SubVD src (LoadVector mem)));
  format %{ "vsubpd  $dst,$src,$mem\t! sub packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vsubpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4D_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SubVD src1 src2));
  format %{ "vsubpd  $dst,$src1,$src2\t! sub packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vsubpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub4D_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SubVD src (LoadVector mem)));
  format %{ "vsubpd  $dst,$src,$mem\t! sub packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vsubpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8D_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (SubVD src1 src2));
  format %{ "vsubpd  $dst,$src1,$src2\t! sub packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vsubpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsub8D_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (SubVD src (LoadVector mem)));
  format %{ "vsubpd  $dst,$src,$mem\t! sub packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vsubpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// --------------------------------- MUL --------------------------------------

// Shorts/Chars vector mul
instruct vmul2S(vecS dst, vecS src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 2);
  match(Set dst (MulVS dst src));
  format %{ "pmullw $dst,$src\t! mul packed2S" %}
  ins_encode %{
    __ pmullw($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2S_reg_avx(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (MulVS src1 src2));
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2S_reg_evex(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (MulVS src1 src2));
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2S_evex_special(vecS dst, vecS src1, vecS src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (MulVS dst src2));
  effect(TEMP src1);
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2S_mem_avx(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (MulVS src (LoadVector mem)));
  format %{ "vpmullw $dst,$src,$mem\t! mul packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2S_mem_evex(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (MulVS src (LoadVector mem)));
  format %{ "vpmullw $dst,$src,$mem\t! mul packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2S_mem_evex_special(vecS dst, vecS src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (MulVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpmullw $dst,$src,$mem\t! mul packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4S(vecD dst, vecD src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (MulVS dst src));
  format %{ "pmullw  $dst,$src\t! mul packed4S" %}
  ins_encode %{
    __ pmullw($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4S_reg_avx(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (MulVS src1 src2));
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4S_reg_evex(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (MulVS src1 src2));
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4S_reg_evex_special(vecD dst, vecD src1, vecD src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (MulVS dst src2));
  effect(TEMP src1);
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4S_mem_avx(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (MulVS src (LoadVector mem)));
  format %{ "vpmullw $dst,$src,$mem\t! mul packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4S_mem_evex(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (MulVS src (LoadVector mem)));
  format %{ "vpmullw $dst,$src,$mem\t! mul packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4S_mem_evex_special(vecD dst, vecD src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (MulVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpmullw $dst,$src,$mem\t! mul packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8S(vecX dst, vecX src) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (MulVS dst src));
  format %{ "pmullw  $dst,$src\t! mul packed8S" %}
  ins_encode %{
    __ pmullw($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8S_reg_avx(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (MulVS src1 src2));
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8S_reg_evex(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (MulVS src1 src2));
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8S_reg_evex_special(vecX dst, vecX src1, vecX src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (MulVS dst src2));
  effect(TEMP src1);
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8S_mem_avx(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (MulVS src (LoadVector mem)));
  format %{ "vpmullw $dst,$src,$mem\t! mul packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8S_mem_evex(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (MulVS src (LoadVector mem)));
  format %{ "vpmullw $dst,$src,$mem\t! mul packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8S_mem_evex_special(vecX dst, vecX src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (MulVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpmullw $dst,$src,$mem\t! mul packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16S_reg_avx(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (MulVS src1 src2));
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16S_reg_evex(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (MulVS src1 src2));
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16S_reg_evex_special(vecY dst, vecY src1, vecY src2) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (MulVS dst src2));
  effect(TEMP src1);
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16S_mem_avx(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (MulVS src (LoadVector mem)));
  format %{ "vpmullw $dst,$src,$mem\t! mul packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16S_mem_evex(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (MulVS src (LoadVector mem)));
  format %{ "vpmullw $dst,$src,$mem\t! mul packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16S_mem_evex_special(vecY dst, vecY src, memory mem) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (MulVS dst (LoadVector mem)));
  effect(TEMP src);
  format %{ "vpmullw $dst,$src,$mem\t! mul packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul32S_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (MulVS src1 src2));
  format %{ "vpmullw $dst,$src1,$src2\t! mul packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpmullw($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul32S_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (MulVS src (LoadVector mem)));
  format %{ "vpmullw $dst,$src,$mem\t! mul packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Integers vector mul (sse4_1)
instruct vmul2I(vecD dst, vecD src) %{
  predicate(UseSSE > 3 && n->as_Vector()->length() == 2);
  match(Set dst (MulVI dst src));
  format %{ "pmulld  $dst,$src\t! mul packed2I" %}
  ins_encode %{
    __ pmulld($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2I_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (MulVI src1 src2));
  format %{ "vpmulld $dst,$src1,$src2\t! mul packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmulld($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2I_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (MulVI src (LoadVector mem)));
  format %{ "vpmulld $dst,$src,$mem\t! mul packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmulld($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4I(vecX dst, vecX src) %{
  predicate(UseSSE > 3 && n->as_Vector()->length() == 4);
  match(Set dst (MulVI dst src));
  format %{ "pmulld  $dst,$src\t! mul packed4I" %}
  ins_encode %{
    __ pmulld($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4I_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (MulVI src1 src2));
  format %{ "vpmulld $dst,$src1,$src2\t! mul packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmulld($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4I_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (MulVI src (LoadVector mem)));
  format %{ "vpmulld $dst,$src,$mem\t! mul packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmulld($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2L_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 2 && VM_Version::supports_avx512dq());
  match(Set dst (MulVL src1 src2));
  format %{ "vpmullq $dst,$src1,$src2\t! mul packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2L_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 2 && VM_Version::supports_avx512dq());
  match(Set dst (MulVL src (LoadVector mem)));
  format %{ "vpmullq $dst,$src,$mem\t! mul packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpmullq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4L_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 4 && VM_Version::supports_avx512dq());
  match(Set dst (MulVL src1 src2));
  format %{ "vpmullq $dst,$src1,$src2\t! mul packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmullq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4L_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 4 && VM_Version::supports_avx512dq());
  match(Set dst (MulVL src (LoadVector mem)));
  format %{ "vpmullq $dst,$src,$mem\t! mul packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmullq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8L_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8 && VM_Version::supports_avx512dq());
  match(Set dst (MulVL src1 src2));
  format %{ "vpmullq $dst,$src1,$src2\t! mul packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpmullq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8L_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8 && VM_Version::supports_avx512dq());
  match(Set dst (MulVL src (LoadVector mem)));
  format %{ "vpmullq $dst,$src,$mem\t! mul packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpmullq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8I_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (MulVI src1 src2));
  format %{ "vpmulld $dst,$src1,$src2\t! mul packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmulld($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8I_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (MulVI src (LoadVector mem)));
  format %{ "vpmulld $dst,$src,$mem\t! mul packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpmulld($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16I_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (MulVI src1 src2));
  format %{ "vpmulld $dst,$src1,$src2\t! mul packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpmulld($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16I_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (MulVI src (LoadVector mem)));
  format %{ "vpmulld $dst,$src,$mem\t! mul packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpmulld($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Floats vector mul
instruct vmul2F(vecD dst, vecD src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (MulVF dst src));
  format %{ "mulps   $dst,$src\t! mul packed2F" %}
  ins_encode %{
    __ mulps($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2F_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (MulVF src1 src2));
  format %{ "vmulps  $dst,$src1,$src2\t! mul packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vmulps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2F_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (MulVF src (LoadVector mem)));
  format %{ "vmulps  $dst,$src,$mem\t! mul packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vmulps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4F(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (MulVF dst src));
  format %{ "mulps   $dst,$src\t! mul packed4F" %}
  ins_encode %{
    __ mulps($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4F_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (MulVF src1 src2));
  format %{ "vmulps  $dst,$src1,$src2\t! mul packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vmulps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4F_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (MulVF src (LoadVector mem)));
  format %{ "vmulps  $dst,$src,$mem\t! mul packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vmulps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8F_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (MulVF src1 src2));
  format %{ "vmulps  $dst,$src1,$src2\t! mul packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vmulps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8F_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (MulVF src (LoadVector mem)));
  format %{ "vmulps  $dst,$src,$mem\t! mul packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vmulps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16F_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (MulVF src1 src2));
  format %{ "vmulps  $dst,$src1,$src2\t! mul packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vmulps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul16F_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (MulVF src (LoadVector mem)));
  format %{ "vmulps  $dst,$src,$mem\t! mul packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vmulps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Doubles vector mul
instruct vmul2D(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (MulVD dst src));
  format %{ "mulpd   $dst,$src\t! mul packed2D" %}
  ins_encode %{
    __ mulpd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2D_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (MulVD src1 src2));
  format %{ "vmulpd  $dst,$src1,$src2\t! mul packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vmulpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul2D_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (MulVD src (LoadVector mem)));
  format %{ "vmulpd  $dst,$src,$mem\t! mul packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vmulpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4D_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (MulVD src1 src2));
  format %{ "vmulpd  $dst,$src1,$src2\t! mul packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vmulpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul4D_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (MulVD src (LoadVector mem)));
  format %{ "vmulpd  $dst,$src,$mem\t! mul packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vmulpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8D_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (MulVD src1 src2));
  format %{ "vmulpd  $dst k0,$src1,$src2\t! mul packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vmulpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vmul8D_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (MulVD src (LoadVector mem)));
  format %{ "vmulpd  $dst k0,$src,$mem\t! mul packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vmulpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vcmov8F_reg(vecY dst, vecY src1, vecY src2, immI8 cop, cmpOp_vcmppd copnd) %{
  predicate(UseAVX > 0 && UseAVX < 3 && n->as_Vector()->length() == 8);
  match(Set dst (CMoveVF (Binary copnd cop) (Binary src1 src2)));
  effect(TEMP dst, USE src1, USE src2);
  format %{ "cmpps.$copnd  $dst, $src1, $src2  ! vcmovevf, cond=$cop\n\t"
            "blendvps $dst,$src1,$src2,$dst ! vcmovevf\n\t"
         %}
  ins_encode %{
    int vector_len = 1;
    int cond = (Assembler::Condition)($copnd$$cmpcode);
    __ cmpps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, cond, vector_len);
    __ blendvps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vcmov4D_reg(vecY dst, vecY src1, vecY src2, immI8 cop, cmpOp_vcmppd copnd) %{
  predicate(UseAVX > 0 && UseAVX < 3 && n->as_Vector()->length() == 4);
  match(Set dst (CMoveVD (Binary copnd cop) (Binary src1 src2)));
  effect(TEMP dst, USE src1, USE src2);
  format %{ "cmppd.$copnd  $dst, $src1, $src2  ! vcmovevd, cond=$cop\n\t"
            "blendvpd $dst,$src1,$src2,$dst ! vcmovevd\n\t"
         %}
  ins_encode %{
    int vector_len = 1;
    int cond = (Assembler::Condition)($copnd$$cmpcode);
    __ cmppd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, cond, vector_len);
    __ blendvpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, $dst$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// --------------------------------- DIV --------------------------------------

// Floats vector div
instruct vdiv2F(vecD dst, vecD src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (DivVF dst src));
  format %{ "divps   $dst,$src\t! div packed2F" %}
  ins_encode %{
    __ divps($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv2F_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (DivVF src1 src2));
  format %{ "vdivps  $dst,$src1,$src2\t! div packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vdivps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv2F_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (DivVF src (LoadVector mem)));
  format %{ "vdivps  $dst,$src,$mem\t! div packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vdivps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv4F(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (DivVF dst src));
  format %{ "divps   $dst,$src\t! div packed4F" %}
  ins_encode %{
    __ divps($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv4F_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (DivVF src1 src2));
  format %{ "vdivps  $dst,$src1,$src2\t! div packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vdivps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv4F_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (DivVF src (LoadVector mem)));
  format %{ "vdivps  $dst,$src,$mem\t! div packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vdivps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv8F_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (DivVF src1 src2));
  format %{ "vdivps  $dst,$src1,$src2\t! div packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vdivps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv8F_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (DivVF src (LoadVector mem)));
  format %{ "vdivps  $dst,$src,$mem\t! div packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vdivps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv16F_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 16);
  match(Set dst (DivVF src1 src2));
  format %{ "vdivps  $dst,$src1,$src2\t! div packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vdivps($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv16F_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 16);
  match(Set dst (DivVF src (LoadVector mem)));
  format %{ "vdivps  $dst,$src,$mem\t! div packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vdivps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Doubles vector div
instruct vdiv2D(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (DivVD dst src));
  format %{ "divpd   $dst,$src\t! div packed2D" %}
  ins_encode %{
    __ divpd($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv2D_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (DivVD src1 src2));
  format %{ "vdivpd  $dst,$src1,$src2\t! div packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vdivpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv2D_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (DivVD src (LoadVector mem)));
  format %{ "vdivpd  $dst,$src,$mem\t! div packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vdivpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv4D_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (DivVD src1 src2));
  format %{ "vdivpd  $dst,$src1,$src2\t! div packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vdivpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv4D_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (DivVD src (LoadVector mem)));
  format %{ "vdivpd  $dst,$src,$mem\t! div packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vdivpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv8D_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (DivVD src1 src2));
  format %{ "vdivpd  $dst,$src1,$src2\t! div packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vdivpd($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vdiv8D_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (DivVD src (LoadVector mem)));
  format %{ "vdivpd  $dst,$src,$mem\t! div packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vdivpd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// ------------------------------ Shift ---------------------------------------

// Left and right shift count vectors are the same on x86
// (only lowest bits of xmm reg are used for count).
instruct vshiftcnt(vecS dst, rRegI cnt) %{
  match(Set dst (LShiftCntV cnt));
  match(Set dst (RShiftCntV cnt));
  format %{ "movd    $dst,$cnt\t! load shift count" %}
  ins_encode %{
    __ movdl($dst$$XMMRegister, $cnt$$Register);
  %}
  ins_pipe( pipe_slow );
%}

// --------------------------------- Sqrt --------------------------------------

// Floating point vector sqrt
instruct vsqrt2D_reg(vecX dst, vecX src) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SqrtVD src));
  format %{ "vsqrtpd  $dst,$src\t! sqrt packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vsqrtpd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt2D_mem(vecX dst, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SqrtVD (LoadVector mem)));
  format %{ "vsqrtpd  $dst,$mem\t! sqrt packed2D" %}
  ins_encode %{
    int vector_len = 0;
    __ vsqrtpd($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt4D_reg(vecY dst, vecY src) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SqrtVD src));
  format %{ "vsqrtpd  $dst,$src\t! sqrt packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vsqrtpd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt4D_mem(vecY dst, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SqrtVD (LoadVector mem)));
  format %{ "vsqrtpd  $dst,$mem\t! sqrt packed4D" %}
  ins_encode %{
    int vector_len = 1;
    __ vsqrtpd($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt8D_reg(vecZ dst, vecZ src) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (SqrtVD src));
  format %{ "vsqrtpd  $dst,$src\t! sqrt packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vsqrtpd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt8D_mem(vecZ dst, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (SqrtVD (LoadVector mem)));
  format %{ "vsqrtpd  $dst,$mem\t! sqrt packed8D" %}
  ins_encode %{
    int vector_len = 2;
    __ vsqrtpd($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt2F_reg(vecD dst, vecD src) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SqrtVF src));
  format %{ "vsqrtps  $dst,$src\t! sqrt packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vsqrtps($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt2F_mem(vecD dst, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (SqrtVF (LoadVector mem)));
  format %{ "vsqrtps  $dst,$mem\t! sqrt packed2F" %}
  ins_encode %{
    int vector_len = 0;
    __ vsqrtps($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt4F_reg(vecX dst, vecX src) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SqrtVF src));
  format %{ "vsqrtps  $dst,$src\t! sqrt packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vsqrtps($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt4F_mem(vecX dst, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (SqrtVF (LoadVector mem)));
  format %{ "vsqrtps  $dst,$mem\t! sqrt packed4F" %}
  ins_encode %{
    int vector_len = 0;
    __ vsqrtps($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt8F_reg(vecY dst, vecY src) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (SqrtVF src));
  format %{ "vsqrtps  $dst,$src\t! sqrt packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vsqrtps($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt8F_mem(vecY dst, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
  match(Set dst (SqrtVF (LoadVector mem)));
  format %{ "vsqrtps  $dst,$mem\t! sqrt packed8F" %}
  ins_encode %{
    int vector_len = 1;
    __ vsqrtps($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt16F_reg(vecZ dst, vecZ src) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (SqrtVF src));
  format %{ "vsqrtps  $dst,$src\t! sqrt packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vsqrtps($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsqrt16F_mem(vecZ dst, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (SqrtVF (LoadVector mem)));
  format %{ "vsqrtps  $dst,$mem\t! sqrt packed16F" %}
  ins_encode %{
    int vector_len = 2;
    __ vsqrtps($dst$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// ------------------------------ LeftShift -----------------------------------

// Shorts/Chars vector left shift
instruct vsll2S(vecS dst, vecS shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVS dst shift));
  format %{ "psllw   $dst,$shift\t! left shift packed2S" %}
  ins_encode %{
    __ psllw($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2S_imm(vecS dst, immI8 shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVS dst shift));
  format %{ "psllw   $dst,$shift\t! left shift packed2S" %}
  ins_encode %{
    __ psllw($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2S_reg_avx(vecS dst, vecS src, vecS shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2S_reg_evex(vecS dst, vecS src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2S_reg_evex_special(vecS dst, vecS src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2S_reg_imm_avx(vecS dst, vecS src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2S_reg_imm_evex(vecS dst, vecS src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2S_reg_imm_evex_special(vecS dst, vecS src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4S(vecD dst, vecS shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVS dst shift));
  format %{ "psllw   $dst,$shift\t! left shift packed4S" %}
  ins_encode %{
    __ psllw($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4S_imm(vecD dst, immI8 shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVS dst shift));
  format %{ "psllw   $dst,$shift\t! left shift packed4S" %}
  ins_encode %{
    __ psllw($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4S_reg_avx(vecD dst, vecD src, vecS shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4S_reg_evex(vecD dst, vecD src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4S_reg_evex_special(vecD dst, vecD src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4S_reg_imm_avx(vecD dst, vecD src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4S_reg_imm_evex(vecD dst, vecD src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4S_reg_imm_evex_special(vecD dst, vecD src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8S(vecX dst, vecS shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVS dst shift));
  format %{ "psllw   $dst,$shift\t! left shift packed8S" %}
  ins_encode %{
    __ psllw($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8S_imm(vecX dst, immI8 shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVS dst shift));
  format %{ "psllw   $dst,$shift\t! left shift packed8S" %}
  ins_encode %{
    __ psllw($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8S_reg_avx(vecX dst, vecX src, vecS shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8S_reg_evex(vecX dst, vecX src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8S_reg_evex_special(vecX dst, vecX src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8S_reg_imm_avx(vecX dst, vecX src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8S_reg_imm_evex(vecX dst, vecX src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8S_reg_imm_evex_special(vecX dst, vecX src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll16S_reg_avx(vecY dst, vecY src, vecS shift) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll16S_reg_evex(vecY dst, vecY src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll16S_reg_evex_special(vecY dst, vecY src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (LShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll16S_reg_imm_avx(vecY dst, vecY src, immI8 shift) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll16S_reg_imm_evex(vecY dst, vecY src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll16S_reg_imm_evex_special(vecY dst, vecY src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (LShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll32S_reg(vecZ dst, vecZ src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll32S_reg_imm(vecZ dst, vecZ src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (LShiftVS src shift));
  format %{ "vpsllw  $dst,$src,$shift\t! left shift packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsllw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Integers vector left shift
instruct vsll2I(vecD dst, vecS shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (LShiftVI dst shift));
  format %{ "pslld   $dst,$shift\t! left shift packed2I" %}
  ins_encode %{
    __ pslld($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2I_imm(vecD dst, immI8 shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (LShiftVI dst shift));
  format %{ "pslld   $dst,$shift\t! left shift packed2I" %}
  ins_encode %{
    __ pslld($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2I_reg(vecD dst, vecD src, vecS shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVI src shift));
  format %{ "vpslld  $dst,$src,$shift\t! left shift packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpslld($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2I_reg_imm(vecD dst, vecD src, immI8 shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVI src shift));
  format %{ "vpslld  $dst,$src,$shift\t! left shift packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpslld($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4I(vecX dst, vecS shift) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (LShiftVI dst shift));
  format %{ "pslld   $dst,$shift\t! left shift packed4I" %}
  ins_encode %{
    __ pslld($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4I_imm(vecX dst, immI8 shift) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (LShiftVI dst shift));
  format %{ "pslld   $dst,$shift\t! left shift packed4I" %}
  ins_encode %{
    __ pslld($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4I_reg(vecX dst, vecX src, vecS shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVI src shift));
  format %{ "vpslld  $dst,$src,$shift\t! left shift packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpslld($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4I_reg_imm(vecX dst, vecX src, immI8 shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVI src shift));
  format %{ "vpslld  $dst,$src,$shift\t! left shift packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpslld($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8I_reg(vecY dst, vecY src, vecS shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVI src shift));
  format %{ "vpslld  $dst,$src,$shift\t! left shift packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpslld($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8I_reg_imm(vecY dst, vecY src, immI8 shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVI src shift));
  format %{ "vpslld  $dst,$src,$shift\t! left shift packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpslld($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll16I_reg(vecZ dst, vecZ src, vecS shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (LShiftVI src shift));
  format %{ "vpslld  $dst,$src,$shift\t! left shift packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpslld($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll16I_reg_imm(vecZ dst, vecZ src, immI8 shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (LShiftVI src shift));
  format %{ "vpslld  $dst,$src,$shift\t! left shift packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpslld($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Longs vector left shift
instruct vsll2L(vecX dst, vecS shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (LShiftVL dst shift));
  format %{ "psllq   $dst,$shift\t! left shift packed2L" %}
  ins_encode %{
    __ psllq($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2L_imm(vecX dst, immI8 shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (LShiftVL dst shift));
  format %{ "psllq   $dst,$shift\t! left shift packed2L" %}
  ins_encode %{
    __ psllq($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2L_reg(vecX dst, vecX src, vecS shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVL src shift));
  format %{ "vpsllq  $dst,$src,$shift\t! left shift packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllq($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll2L_reg_imm(vecX dst, vecX src, immI8 shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (LShiftVL src shift));
  format %{ "vpsllq  $dst,$src,$shift\t! left shift packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsllq($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4L_reg(vecY dst, vecY src, vecS shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVL src shift));
  format %{ "vpsllq  $dst,$src,$shift\t! left shift packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsllq($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll4L_reg_imm(vecY dst, vecY src, immI8 shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 4);
  match(Set dst (LShiftVL src shift));
  format %{ "vpsllq  $dst,$src,$shift\t! left shift packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsllq($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8L_reg(vecZ dst, vecZ src, vecS shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVL src shift));
  format %{ "vpsllq  $dst,$src,$shift\t! left shift packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsllq($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsll8L_reg_imm(vecZ dst, vecZ src, immI8 shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (LShiftVL src shift));
  format %{ "vpsllq  $dst,$src,$shift\t! left shift packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsllq($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// ----------------------- LogicalRightShift -----------------------------------

// Shorts vector logical right shift produces incorrect Java result
// for negative data because java code convert short value into int with
// sign extension before a shift. But char vectors are fine since chars are
// unsigned values.

instruct vsrl2S(vecS dst, vecS shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVS dst shift));
  format %{ "psrlw   $dst,$shift\t! logical right shift packed2S" %}
  ins_encode %{
    __ psrlw($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2S_imm(vecS dst, immI8 shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVS dst shift));
  format %{ "psrlw   $dst,$shift\t! logical right shift packed2S" %}
  ins_encode %{
    __ psrlw($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2S_reg_avx(vecS dst, vecS src, vecS shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2S_reg_evex(vecS dst, vecS src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2S_reg_evex_special(vecS dst, vecS src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2S_reg_imm_avx(vecS dst, vecS src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2S_reg_imm_evex(vecS dst, vecS src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2S_reg_imm_evex_special(vecS dst, vecS src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4S(vecD dst, vecS shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVS dst shift));
  format %{ "psrlw   $dst,$shift\t! logical right shift packed4S" %}
  ins_encode %{
    __ psrlw($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4S_imm(vecD dst, immI8 shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVS dst shift));
  format %{ "psrlw   $dst,$shift\t! logical right shift packed4S" %}
  ins_encode %{
    __ psrlw($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4S_reg_avx(vecD dst, vecD src, vecS shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4S_reg_evex(vecD dst, vecD src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4S_reg_evex_special(vecD dst, vecD src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4S_reg_imm_avx(vecD dst, vecD src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4S_reg_imm_evex(vecD dst, vecD src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4S_reg_imm_evex_special(vecD dst, vecD src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8S(vecX dst, vecS shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVS dst shift));
  format %{ "psrlw   $dst,$shift\t! logical right shift packed8S" %}
  ins_encode %{
    __ psrlw($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8S_imm(vecX dst, immI8 shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVS dst shift));
  format %{ "psrlw   $dst,$shift\t! logical right shift packed8S" %}
  ins_encode %{
    __ psrlw($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8S_reg_avx(vecX dst, vecX src, vecS shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8S_reg_evex(vecX dst, vecX src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8S_reg_evex_special(vecX dst, vecX src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8S_reg_imm_avx(vecX dst, vecX src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8S_reg_imm_evex(vecX dst, vecX src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8S_reg_imm_evex_special(vecX dst, vecX src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl16S_reg_avx(vecY dst, vecY src, vecS shift) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl16S_reg_evex(vecY dst, vecY src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl16S_reg_evex_special(vecY dst, vecY src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (URShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl16S_reg_imm_avx(vecY dst, vecY src, immI8 shift) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl16S_reg_imm_evex(vecY dst, vecY src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl16S_reg_imm_evex_special(vecY dst, vecY src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (URShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl32S_reg(vecZ dst, vecZ src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl32S_reg_imm(vecZ dst, vecZ src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (URShiftVS src shift));
  format %{ "vpsrlw  $dst,$src,$shift\t! logical right shift packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsrlw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Integers vector logical right shift
instruct vsrl2I(vecD dst, vecS shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (URShiftVI dst shift));
  format %{ "psrld   $dst,$shift\t! logical right shift packed2I" %}
  ins_encode %{
    __ psrld($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2I_imm(vecD dst, immI8 shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (URShiftVI dst shift));
  format %{ "psrld   $dst,$shift\t! logical right shift packed2I" %}
  ins_encode %{
    __ psrld($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2I_reg(vecD dst, vecD src, vecS shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVI src shift));
  format %{ "vpsrld  $dst,$src,$shift\t! logical right shift packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrld($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2I_reg_imm(vecD dst, vecD src, immI8 shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVI src shift));
  format %{ "vpsrld  $dst,$src,$shift\t! logical right shift packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrld($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4I(vecX dst, vecS shift) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (URShiftVI dst shift));
  format %{ "psrld   $dst,$shift\t! logical right shift packed4I" %}
  ins_encode %{
    __ psrld($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4I_imm(vecX dst, immI8 shift) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (URShiftVI dst shift));
  format %{ "psrld   $dst,$shift\t! logical right shift packed4I" %}
  ins_encode %{
    __ psrld($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4I_reg(vecX dst, vecX src, vecS shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVI src shift));
  format %{ "vpsrld  $dst,$src,$shift\t! logical right shift packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrld($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4I_reg_imm(vecX dst, vecX src, immI8 shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVI src shift));
  format %{ "vpsrld  $dst,$src,$shift\t! logical right shift packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrld($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8I_reg(vecY dst, vecY src, vecS shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVI src shift));
  format %{ "vpsrld  $dst,$src,$shift\t! logical right shift packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrld($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8I_reg_imm(vecY dst, vecY src, immI8 shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVI src shift));
  format %{ "vpsrld  $dst,$src,$shift\t! logical right shift packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrld($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl16I_reg(vecZ dst, vecZ src, vecS shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (URShiftVI src shift));
  format %{ "vpsrld  $dst,$src,$shift\t! logical right shift packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsrld($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl16I_reg_imm(vecZ dst, vecZ src, immI8 shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (URShiftVI src shift));
  format %{ "vpsrld  $dst,$src,$shift\t! logical right shift packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsrld($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Longs vector logical right shift
instruct vsrl2L(vecX dst, vecS shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (URShiftVL dst shift));
  format %{ "psrlq   $dst,$shift\t! logical right shift packed2L" %}
  ins_encode %{
    __ psrlq($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2L_imm(vecX dst, immI8 shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (URShiftVL dst shift));
  format %{ "psrlq   $dst,$shift\t! logical right shift packed2L" %}
  ins_encode %{
    __ psrlq($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2L_reg(vecX dst, vecX src, vecS shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVL src shift));
  format %{ "vpsrlq  $dst,$src,$shift\t! logical right shift packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlq($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl2L_reg_imm(vecX dst, vecX src, immI8 shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (URShiftVL src shift));
  format %{ "vpsrlq  $dst,$src,$shift\t! logical right shift packed2L" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrlq($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4L_reg(vecY dst, vecY src, vecS shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVL src shift));
  format %{ "vpsrlq  $dst,$src,$shift\t! logical right shift packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrlq($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl4L_reg_imm(vecY dst, vecY src, immI8 shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 4);
  match(Set dst (URShiftVL src shift));
  format %{ "vpsrlq  $dst,$src,$shift\t! logical right shift packed4L" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrlq($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8L_reg(vecZ dst, vecZ src, vecS shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVL src shift));
  format %{ "vpsrlq  $dst,$src,$shift\t! logical right shift packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsrlq($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsrl8L_reg_imm(vecZ dst, vecZ src, immI8 shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 8);
  match(Set dst (URShiftVL src shift));
  format %{ "vpsrlq  $dst,$src,$shift\t! logical right shift packed8L" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsrlq($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// ------------------- ArithmeticRightShift -----------------------------------

// Shorts/Chars vector arithmetic right shift
instruct vsra2S(vecS dst, vecS shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 2);
  match(Set dst (RShiftVS dst shift));
  format %{ "psraw   $dst,$shift\t! arithmetic right shift packed2S" %}
  ins_encode %{
    __ psraw($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2S_imm(vecS dst, immI8 shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (RShiftVS dst shift));
  format %{ "psraw   $dst,$shift\t! arithmetic right shift packed2S" %}
  ins_encode %{
    __ psraw($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2S_reg_avx(vecS dst, vecS src, vecS shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2S_reg_evex(vecS dst, vecS src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2S_reg_evex_special(vecS dst, vecS src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (RShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2S_reg_imm_avx(vecS dst, vecS src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 2);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2S_reg_imm_evex(vecS dst, vecS src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 2);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2S_reg_imm_evex_special(vecS dst, vecS src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 2);
  match(Set dst (RShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed2S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4S(vecD dst, vecS shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVS dst shift));
  format %{ "psraw   $dst,$shift\t! arithmetic right shift packed4S" %}
  ins_encode %{
    __ psraw($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4S_imm(vecD dst, immI8 shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVS dst shift));
  format %{ "psraw   $dst,$shift\t! arithmetic right shift packed4S" %}
  ins_encode %{
    __ psraw($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4S_reg_avx(vecD dst, vecD src, vecS shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4S_reg_evex(vecD dst, vecD src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4S_reg_evex_special(vecD dst, vecD src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4S_reg_imm_avx(vecD dst, vecD src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4S_reg_imm_evex(vecD dst, vecD src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4S_reg_imm_evex_special(vecD dst, vecD src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed4S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8S(vecX dst, vecS shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVS dst shift));
  format %{ "psraw   $dst,$shift\t! arithmetic right shift packed8S" %}
  ins_encode %{
    __ psraw($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8S_imm(vecX dst, immI8 shift) %{
  predicate(UseAVX == 0 && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVS dst shift));
  format %{ "psraw   $dst,$shift\t! arithmetic right shift packed8S" %}
  ins_encode %{
    __ psraw($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8S_reg_avx(vecX dst, vecX src, vecS shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8S_reg_evex(vecX dst, vecX src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8S_reg_evex_special(vecX dst, vecX src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8S_reg_imm_avx(vecX dst, vecX src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8S_reg_imm_evex(vecX dst, vecX src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8S_reg_imm_evex_special(vecX dst, vecX src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed8S" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra16S_reg_avx(vecY dst, vecY src, vecS shift) %{
  predicate(VM_Version::supports_avx256only() && n->as_Vector()->length() == 16);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra16S_reg_evex(vecY dst, vecY src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra16S_reg_evex_special(vecY dst, vecY src, vecS shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (RShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra16S_reg_imm_avx(vecY dst, vecY src, immI8 shift) %{
  predicate(VM_Version::supports_avxonly() && n->as_Vector()->length() == 16);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra16S_reg_imm_evex(vecY dst, vecY src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 16);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra16S_reg_imm_evex_special(vecY dst, vecY src, immI8 shift) %{
  predicate(VM_Version::supports_avx512nobw() && n->as_Vector()->length() == 16);
  match(Set dst (RShiftVS dst shift));
  effect(TEMP src);
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed16S" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra32S_reg(vecZ dst, vecZ src, vecS shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra32S_reg_imm(vecZ dst, vecZ src, immI8 shift) %{
  predicate(VM_Version::supports_avx512bw() && n->as_Vector()->length() == 32);
  match(Set dst (RShiftVS src shift));
  format %{ "vpsraw  $dst,$src,$shift\t! arithmetic right shift packed32S" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsraw($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// Integers vector arithmetic right shift
instruct vsra2I(vecD dst, vecS shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (RShiftVI dst shift));
  format %{ "psrad   $dst,$shift\t! arithmetic right shift packed2I" %}
  ins_encode %{
    __ psrad($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2I_imm(vecD dst, immI8 shift) %{
  predicate(n->as_Vector()->length() == 2);
  match(Set dst (RShiftVI dst shift));
  format %{ "psrad   $dst,$shift\t! arithmetic right shift packed2I" %}
  ins_encode %{
    __ psrad($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2I_reg(vecD dst, vecD src, vecS shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (RShiftVI src shift));
  format %{ "vpsrad  $dst,$src,$shift\t! arithmetic right shift packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrad($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra2I_reg_imm(vecD dst, vecD src, immI8 shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
  match(Set dst (RShiftVI src shift));
  format %{ "vpsrad  $dst,$src,$shift\t! arithmetic right shift packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrad($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4I(vecX dst, vecS shift) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (RShiftVI dst shift));
  format %{ "psrad   $dst,$shift\t! arithmetic right shift packed4I" %}
  ins_encode %{
    __ psrad($dst$$XMMRegister, $shift$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4I_imm(vecX dst, immI8 shift) %{
  predicate(n->as_Vector()->length() == 4);
  match(Set dst (RShiftVI dst shift));
  format %{ "psrad   $dst,$shift\t! arithmetic right shift packed4I" %}
  ins_encode %{
    __ psrad($dst$$XMMRegister, (int)$shift$$constant);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4I_reg(vecX dst, vecX src, vecS shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVI src shift));
  format %{ "vpsrad  $dst,$src,$shift\t! arithmetic right shift packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrad($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra4I_reg_imm(vecX dst, vecX src, immI8 shift) %{
  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
  match(Set dst (RShiftVI src shift));
  format %{ "vpsrad  $dst,$src,$shift\t! arithmetic right shift packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpsrad($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8I_reg(vecY dst, vecY src, vecS shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVI src shift));
  format %{ "vpsrad  $dst,$src,$shift\t! arithmetic right shift packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrad($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra8I_reg_imm(vecY dst, vecY src, immI8 shift) %{
  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
  match(Set dst (RShiftVI src shift));
  format %{ "vpsrad  $dst,$src,$shift\t! arithmetic right shift packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpsrad($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra16I_reg(vecZ dst, vecZ src, vecS shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (RShiftVI src shift));
  format %{ "vpsrad  $dst,$src,$shift\t! arithmetic right shift packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsrad($dst$$XMMRegister, $src$$XMMRegister, $shift$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vsra16I_reg_imm(vecZ dst, vecZ src, immI8 shift) %{
  predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
  match(Set dst (RShiftVI src shift));
  format %{ "vpsrad  $dst,$src,$shift\t! arithmetic right shift packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpsrad($dst$$XMMRegister, $src$$XMMRegister, (int)$shift$$constant, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// There are no longs vector arithmetic right shift instructions.


// --------------------------------- AND --------------------------------------

instruct vand4B(vecS dst, vecS src) %{
  predicate(n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (AndV dst src));
  format %{ "pand    $dst,$src\t! and vectors (4 bytes)" %}
  ins_encode %{
    __ pand($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand4B_reg(vecS dst, vecS src1, vecS src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (AndV src1 src2));
  format %{ "vpand   $dst,$src1,$src2\t! and vectors (4 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpand($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand4B_mem(vecS dst, vecS src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (AndV src (LoadVector mem)));
  format %{ "vpand   $dst,$src,$mem\t! and vectors (4 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpand($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand8B(vecD dst, vecD src) %{
  predicate(n->as_Vector()->length_in_bytes() == 8);
  match(Set dst (AndV dst src));
  format %{ "pand    $dst,$src\t! and vectors (8 bytes)" %}
  ins_encode %{
    __ pand($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand8B_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 8);
  match(Set dst (AndV src1 src2));
  format %{ "vpand   $dst,$src1,$src2\t! and vectors (8 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpand($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand8B_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 8);
  match(Set dst (AndV src (LoadVector mem)));
  format %{ "vpand   $dst,$src,$mem\t! and vectors (8 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpand($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand16B(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length_in_bytes() == 16);
  match(Set dst (AndV dst src));
  format %{ "pand    $dst,$src\t! and vectors (16 bytes)" %}
  ins_encode %{
    __ pand($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand16B_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 16);
  match(Set dst (AndV src1 src2));
  format %{ "vpand   $dst,$src1,$src2\t! and vectors (16 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpand($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand16B_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 16);
  match(Set dst (AndV src (LoadVector mem)));
  format %{ "vpand   $dst,$src,$mem\t! and vectors (16 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpand($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand32B_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 1 && n->as_Vector()->length_in_bytes() == 32);
  match(Set dst (AndV src1 src2));
  format %{ "vpand   $dst,$src1,$src2\t! and vectors (32 bytes)" %}
  ins_encode %{
    int vector_len = 1;
    __ vpand($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand32B_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 1 && n->as_Vector()->length_in_bytes() == 32);
  match(Set dst (AndV src (LoadVector mem)));
  format %{ "vpand   $dst,$src,$mem\t! and vectors (32 bytes)" %}
  ins_encode %{
    int vector_len = 1;
    __ vpand($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand64B_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length_in_bytes() == 64);
  match(Set dst (AndV src1 src2));
  format %{ "vpand   $dst,$src1,$src2\t! and vectors (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ vpand($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vand64B_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length_in_bytes() == 64);
  match(Set dst (AndV src (LoadVector mem)));
  format %{ "vpand   $dst,$src,$mem\t! and vectors (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ vpand($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// --------------------------------- OR ---------------------------------------

instruct vor4B(vecS dst, vecS src) %{
  predicate(n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (OrV dst src));
  format %{ "por     $dst,$src\t! or vectors (4 bytes)" %}
  ins_encode %{
    __ por($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor4B_reg(vecS dst, vecS src1, vecS src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (OrV src1 src2));
  format %{ "vpor    $dst,$src1,$src2\t! or vectors (4 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor4B_mem(vecS dst, vecS src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (OrV src (LoadVector mem)));
  format %{ "vpor    $dst,$src,$mem\t! or vectors (4 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor8B(vecD dst, vecD src) %{
  predicate(n->as_Vector()->length_in_bytes() == 8);
  match(Set dst (OrV dst src));
  format %{ "por     $dst,$src\t! or vectors (8 bytes)" %}
  ins_encode %{
    __ por($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor8B_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 8);
  match(Set dst (OrV src1 src2));
  format %{ "vpor    $dst,$src1,$src2\t! or vectors (8 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor8B_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (OrV src (LoadVector mem)));
  format %{ "vpor    $dst,$src,$mem\t! or vectors (8 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor16B(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length_in_bytes() == 16);
  match(Set dst (OrV dst src));
  format %{ "por     $dst,$src\t! or vectors (16 bytes)" %}
  ins_encode %{
    __ por($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor16B_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 16);
  match(Set dst (OrV src1 src2));
  format %{ "vpor    $dst,$src1,$src2\t! or vectors (16 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor16B_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 16);
  match(Set dst (OrV src (LoadVector mem)));
  format %{ "vpor    $dst,$src,$mem\t! or vectors (16 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor32B_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 1 && n->as_Vector()->length_in_bytes() == 32);
  match(Set dst (OrV src1 src2));
  format %{ "vpor    $dst,$src1,$src2\t! or vectors (32 bytes)" %}
  ins_encode %{
    int vector_len = 1;
    __ vpor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor32B_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 1 && n->as_Vector()->length_in_bytes() == 32);
  match(Set dst (OrV src (LoadVector mem)));
  format %{ "vpor    $dst,$src,$mem\t! or vectors (32 bytes)" %}
  ins_encode %{
    int vector_len = 1;
    __ vpor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor64B_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length_in_bytes() == 64);
  match(Set dst (OrV src1 src2));
  format %{ "vpor    $dst,$src1,$src2\t! or vectors (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ vpor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vor64B_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length_in_bytes() == 64);
  match(Set dst (OrV src (LoadVector mem)));
  format %{ "vpor    $dst,$src,$mem\t! or vectors (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ vpor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// --------------------------------- XOR --------------------------------------

instruct vxor4B(vecS dst, vecS src) %{
  predicate(n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (XorV dst src));
  format %{ "pxor    $dst,$src\t! xor vectors (4 bytes)" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor4B_reg(vecS dst, vecS src1, vecS src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (XorV src1 src2));
  format %{ "vpxor   $dst,$src1,$src2\t! xor vectors (4 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpxor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor4B_mem(vecS dst, vecS src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
  match(Set dst (XorV src (LoadVector mem)));
  format %{ "vpxor   $dst,$src,$mem\t! xor vectors (4 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpxor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor8B(vecD dst, vecD src) %{
  predicate(n->as_Vector()->length_in_bytes() == 8);
  match(Set dst (XorV dst src));
  format %{ "pxor    $dst,$src\t! xor vectors (8 bytes)" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor8B_reg(vecD dst, vecD src1, vecD src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 8);
  match(Set dst (XorV src1 src2));
  format %{ "vpxor   $dst,$src1,$src2\t! xor vectors (8 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpxor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor8B_mem(vecD dst, vecD src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 8);
  match(Set dst (XorV src (LoadVector mem)));
  format %{ "vpxor   $dst,$src,$mem\t! xor vectors (8 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpxor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor16B(vecX dst, vecX src) %{
  predicate(n->as_Vector()->length_in_bytes() == 16);
  match(Set dst (XorV dst src));
  format %{ "pxor    $dst,$src\t! xor vectors (16 bytes)" %}
  ins_encode %{
    __ pxor($dst$$XMMRegister, $src$$XMMRegister);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor16B_reg(vecX dst, vecX src1, vecX src2) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 16);
  match(Set dst (XorV src1 src2));
  format %{ "vpxor   $dst,$src1,$src2\t! xor vectors (16 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpxor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor16B_mem(vecX dst, vecX src, memory mem) %{
  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 16);
  match(Set dst (XorV src (LoadVector mem)));
  format %{ "vpxor   $dst,$src,$mem\t! xor vectors (16 bytes)" %}
  ins_encode %{
    int vector_len = 0;
    __ vpxor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor32B_reg(vecY dst, vecY src1, vecY src2) %{
  predicate(UseAVX > 1 && n->as_Vector()->length_in_bytes() == 32);
  match(Set dst (XorV src1 src2));
  format %{ "vpxor   $dst,$src1,$src2\t! xor vectors (32 bytes)" %}
  ins_encode %{
    int vector_len = 1;
    __ vpxor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor32B_mem(vecY dst, vecY src, memory mem) %{
  predicate(UseAVX > 1 && n->as_Vector()->length_in_bytes() == 32);
  match(Set dst (XorV src (LoadVector mem)));
  format %{ "vpxor   $dst,$src,$mem\t! xor vectors (32 bytes)" %}
  ins_encode %{
    int vector_len = 1;
    __ vpxor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor64B_reg(vecZ dst, vecZ src1, vecZ src2) %{
  predicate(UseAVX > 2 && n->as_Vector()->length_in_bytes() == 64);
  match(Set dst (XorV src1 src2));
  format %{ "vpxor   $dst,$src1,$src2\t! xor vectors (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vxor64B_mem(vecZ dst, vecZ src, memory mem) %{
  predicate(UseAVX > 2 && n->as_Vector()->length_in_bytes() == 64);
  match(Set dst (XorV src (LoadVector mem)));
  format %{ "vpxor   $dst,$src,$mem\t! xor vectors (64 bytes)" %}
  ins_encode %{
    int vector_len = 2;
    __ vpxor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// --------------------------------- FMA --------------------------------------

// a * b + c
instruct vfma2D_reg(vecX a, vecX b, vecX c) %{
  predicate(UseFMA && n->as_Vector()->length() == 2);
  match(Set c (FmaVD  c (Binary a b)));
  format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed2D" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 0;
    __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma2D_mem(vecX a, memory b, vecX c) %{
  predicate(UseFMA && n->as_Vector()->length() == 2);
  match(Set c (FmaVD  c (Binary a (LoadVector b))));
  format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed2D" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 0;
    __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}


// a * b + c
instruct vfma4D_reg(vecY a, vecY b, vecY c) %{
  predicate(UseFMA && n->as_Vector()->length() == 4);
  match(Set c (FmaVD  c (Binary a b)));
  format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed4D" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 1;
    __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma4D_mem(vecY a, memory b, vecY c) %{
  predicate(UseFMA && n->as_Vector()->length() == 4);
  match(Set c (FmaVD  c (Binary a (LoadVector b))));
  format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed4D" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 1;
    __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma8D_reg(vecZ a, vecZ b, vecZ c) %{
  predicate(UseFMA && n->as_Vector()->length() == 8);
  match(Set c (FmaVD  c (Binary a b)));
  format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed8D" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 2;
    __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma8D_mem(vecZ a, memory b, vecZ c) %{
  predicate(UseFMA && n->as_Vector()->length() == 8);
  match(Set c (FmaVD  c (Binary a (LoadVector b))));
  format %{ "fmapd $a,$b,$c\t# $c = $a * $b + $c fma packed8D" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 2;
    __ vfmad($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma4F_reg(vecX a, vecX b, vecX c) %{
  predicate(UseFMA && n->as_Vector()->length() == 4);
  match(Set c (FmaVF  c (Binary a b)));
  format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed4F" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 0;
    __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma4F_mem(vecX a, memory b, vecX c) %{
  predicate(UseFMA && n->as_Vector()->length() == 4);
  match(Set c (FmaVF  c (Binary a (LoadVector b))));
  format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed4F" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 0;
    __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma8F_reg(vecY a, vecY b, vecY c) %{
  predicate(UseFMA && n->as_Vector()->length() == 8);
  match(Set c (FmaVF  c (Binary a b)));
  format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed8F" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 1;
    __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma8F_mem(vecY a, memory b, vecY c) %{
  predicate(UseFMA && n->as_Vector()->length() == 8);
  match(Set c (FmaVF  c (Binary a (LoadVector b))));
  format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed8F" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 1;
    __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma16F_reg(vecZ a, vecZ b, vecZ c) %{
  predicate(UseFMA && n->as_Vector()->length() == 16);
  match(Set c (FmaVF  c (Binary a b)));
  format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed16F" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 2;
    __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// a * b + c
instruct vfma16F_mem(vecZ a, memory b, vecZ c) %{
  predicate(UseFMA && n->as_Vector()->length() == 16);
  match(Set c (FmaVF  c (Binary a (LoadVector b))));
  format %{ "fmaps $a,$b,$c\t# $c = $a * $b + $c fma packed16F" %}
  ins_cost(150);
  ins_encode %{
    int vector_len = 2;
    __ vfmaf($c$$XMMRegister, $a$$XMMRegister, $b$$Address, $c$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

// --------------------------------- PopCount --------------------------------------

instruct vpopcount2I(vecD dst, vecD src) %{
  predicate(VM_Version::supports_vpopcntdq() && UsePopCountInstruction && n->as_Vector()->length() == 2);
  match(Set dst (PopCountVI src));
  format %{ "vpopcntd  $dst,$src\t! vector popcount packed2I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpopcntd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vpopcount4I(vecX dst, vecX src) %{
  predicate(VM_Version::supports_vpopcntdq() && UsePopCountInstruction && n->as_Vector()->length() == 4);
  match(Set dst (PopCountVI src));
  format %{ "vpopcntd  $dst,$src\t! vector popcount packed4I" %}
  ins_encode %{
    int vector_len = 0;
    __ vpopcntd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vpopcount8I(vecY dst, vecY src) %{
  predicate(VM_Version::supports_vpopcntdq() && UsePopCountInstruction && n->as_Vector()->length() == 8);
  match(Set dst (PopCountVI src));
  format %{ "vpopcntd  $dst,$src\t! vector popcount packed8I" %}
  ins_encode %{
    int vector_len = 1;
    __ vpopcntd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}

instruct vpopcount16I(vecZ dst, vecZ src) %{
  predicate(VM_Version::supports_vpopcntdq() && UsePopCountInstruction && n->as_Vector()->length() == 16);
  match(Set dst (PopCountVI src));
  format %{ "vpopcntd  $dst,$src\t! vector popcount packed16I" %}
  ins_encode %{
    int vector_len = 2;
    __ vpopcntd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
  %}
  ins_pipe( pipe_slow );
%}