|
|
|
|
Quellcode-Bibliothek riscv.ad
Sprache: unbekannt
|
|
Spracherkennung für: .ad vermutete Sprache: Unknown {[0] [0] [0]} [Methode: Schwerpunktbildung, einfache Gewichte, sechs Dimensionen]
//
// Copyright (c) 2003, 2022, Oracle and/or its affiliates. All rights reserved.
// Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
// Copyright (c) 2020, 2022, Huawei Technologies Co., Ltd. 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.
//
//
// RISCV 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
// architecture.
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.
// We must define the 64 bit int registers in two 32 bit halves, the
// real lower register and a virtual upper half register. upper halves
// are used by the register allocator but are not actually supplied as
// operands to memory ops.
//
// follow the C1 compiler in making registers
//
// x7, x9-x17, x27-x31 volatile (caller save)
// x0-x4, x8, x23 system (no save, no allocate)
// x5-x6 non-allocatable (so we can use them as temporary regs)
//
// as regards Java usage. we don't use any callee save registers
// because this makes it difficult to de-optimise a frame (see comment
// in x86 implementation of Deoptimization::unwind_callee_save_values)
//
// General Registers
reg_def R0 ( NS, NS, Op_RegI, 0, x0->as_VMReg() ); // zr
reg_def R0_H ( NS, NS, Op_RegI, 0, x0->as_VMReg()->next() );
reg_def R1 ( NS, SOC, Op_RegI, 1, x1->as_VMReg() ); // ra
reg_def R1_H ( NS, SOC, Op_RegI, 1, x1->as_VMReg()->next() );
reg_def R2 ( NS, SOE, Op_RegI, 2, x2->as_VMReg() ); // sp
reg_def R2_H ( NS, SOE, Op_RegI, 2, x2->as_VMReg()->next() );
reg_def R3 ( NS, NS, Op_RegI, 3, x3->as_VMReg() ); // gp
reg_def R3_H ( NS, NS, Op_RegI, 3, x3->as_VMReg()->next() );
reg_def R4 ( NS, NS, Op_RegI, 4, x4->as_VMReg() ); // tp
reg_def R4_H ( NS, NS, Op_RegI, 4, x4->as_VMReg()->next() );
reg_def R7 ( SOC, SOC, Op_RegI, 7, x7->as_VMReg() );
reg_def R7_H ( SOC, SOC, Op_RegI, 7, x7->as_VMReg()->next() );
reg_def R8 ( NS, SOE, Op_RegI, 8, x8->as_VMReg() ); // fp
reg_def R8_H ( NS, SOE, Op_RegI, 8, x8->as_VMReg()->next() );
reg_def R9 ( SOC, SOE, Op_RegI, 9, x9->as_VMReg() );
reg_def R9_H ( SOC, SOE, Op_RegI, 9, x9->as_VMReg()->next() );
reg_def R10 ( SOC, SOC, Op_RegI, 10, x10->as_VMReg() );
reg_def R10_H ( SOC, SOC, Op_RegI, 10, x10->as_VMReg()->next());
reg_def R11 ( SOC, SOC, Op_RegI, 11, x11->as_VMReg() );
reg_def R11_H ( SOC, SOC, Op_RegI, 11, x11->as_VMReg()->next());
reg_def R12 ( SOC, SOC, Op_RegI, 12, x12->as_VMReg() );
reg_def R12_H ( SOC, SOC, Op_RegI, 12, x12->as_VMReg()->next());
reg_def R13 ( SOC, SOC, Op_RegI, 13, x13->as_VMReg() );
reg_def R13_H ( SOC, SOC, Op_RegI, 13, x13->as_VMReg()->next());
reg_def R14 ( SOC, SOC, Op_RegI, 14, x14->as_VMReg() );
reg_def R14_H ( SOC, SOC, Op_RegI, 14, x14->as_VMReg()->next());
reg_def R15 ( SOC, SOC, Op_RegI, 15, x15->as_VMReg() );
reg_def R15_H ( SOC, SOC, Op_RegI, 15, x15->as_VMReg()->next());
reg_def R16 ( SOC, SOC, Op_RegI, 16, x16->as_VMReg() );
reg_def R16_H ( SOC, SOC, Op_RegI, 16, x16->as_VMReg()->next());
reg_def R17 ( SOC, SOC, Op_RegI, 17, x17->as_VMReg() );
reg_def R17_H ( SOC, SOC, Op_RegI, 17, x17->as_VMReg()->next());
reg_def R18 ( SOC, SOE, Op_RegI, 18, x18->as_VMReg() );
reg_def R18_H ( SOC, SOE, Op_RegI, 18, x18->as_VMReg()->next());
reg_def R19 ( SOC, SOE, Op_RegI, 19, x19->as_VMReg() );
reg_def R19_H ( SOC, SOE, Op_RegI, 19, x19->as_VMReg()->next());
reg_def R20 ( SOC, SOE, Op_RegI, 20, x20->as_VMReg() ); // caller esp
reg_def R20_H ( SOC, SOE, Op_RegI, 20, x20->as_VMReg()->next());
reg_def R21 ( SOC, SOE, Op_RegI, 21, x21->as_VMReg() );
reg_def R21_H ( SOC, SOE, Op_RegI, 21, x21->as_VMReg()->next());
reg_def R22 ( SOC, SOE, Op_RegI, 22, x22->as_VMReg() );
reg_def R22_H ( SOC, SOE, Op_RegI, 22, x22->as_VMReg()->next());
reg_def R23 ( NS, SOE, Op_RegI, 23, x23->as_VMReg() ); // java thread
reg_def R23_H ( NS, SOE, Op_RegI, 23, x23->as_VMReg()->next());
reg_def R24 ( SOC, SOE, Op_RegI, 24, x24->as_VMReg() );
reg_def R24_H ( SOC, SOE, Op_RegI, 24, x24->as_VMReg()->next());
reg_def R25 ( SOC, SOE, Op_RegI, 25, x25->as_VMReg() );
reg_def R25_H ( SOC, SOE, Op_RegI, 25, x25->as_VMReg()->next());
reg_def R26 ( SOC, SOE, Op_RegI, 26, x26->as_VMReg() );
reg_def R26_H ( SOC, SOE, Op_RegI, 26, x26->as_VMReg()->next());
reg_def R27 ( SOC, SOE, Op_RegI, 27, x27->as_VMReg() ); // heapbase
reg_def R27_H ( SOC, SOE, Op_RegI, 27, x27->as_VMReg()->next());
reg_def R28 ( SOC, SOC, Op_RegI, 28, x28->as_VMReg() );
reg_def R28_H ( SOC, SOC, Op_RegI, 28, x28->as_VMReg()->next());
reg_def R29 ( SOC, SOC, Op_RegI, 29, x29->as_VMReg() );
reg_def R29_H ( SOC, SOC, Op_RegI, 29, x29->as_VMReg()->next());
reg_def R30 ( SOC, SOC, Op_RegI, 30, x30->as_VMReg() );
reg_def R30_H ( SOC, SOC, Op_RegI, 30, x30->as_VMReg()->next());
reg_def R31 ( SOC, SOC, Op_RegI, 31, x31->as_VMReg() );
reg_def R31_H ( SOC, SOC, Op_RegI, 31, x31->as_VMReg()->next());
// ----------------------------
// Float/Double Registers
// ----------------------------
// Double Registers
// The rules of ADL require that double registers be defined in pairs.
// Each pair must be two 32-bit values, but not necessarily a pair of
// single float registers. In each pair, ADLC-assigned register numbers
// must be adjacent, with the lower number even. Finally, when the
// CPU stores such a register pair to memory, the word associated with
// the lower ADLC-assigned number must be stored to the lower address.
// RISCV has 32 floating-point registers. Each can store a single
// or double precision floating-point value.
// for Java use float registers f0-f31 are always save on call whereas
// the platform ABI treats f8-f9 and f18-f27 as callee save). Other
// float registers are SOC as per the platform spec
reg_def F0 ( SOC, SOC, Op_RegF, 0, f0->as_VMReg() );
reg_def F0_H ( SOC, SOC, Op_RegF, 0, f0->as_VMReg()->next() );
reg_def F1 ( SOC, SOC, Op_RegF, 1, f1->as_VMReg() );
reg_def F1_H ( SOC, SOC, Op_RegF, 1, f1->as_VMReg()->next() );
reg_def F2 ( SOC, SOC, Op_RegF, 2, f2->as_VMReg() );
reg_def F2_H ( SOC, SOC, Op_RegF, 2, f2->as_VMReg()->next() );
reg_def F3 ( SOC, SOC, Op_RegF, 3, f3->as_VMReg() );
reg_def F3_H ( SOC, SOC, Op_RegF, 3, f3->as_VMReg()->next() );
reg_def F4 ( SOC, SOC, Op_RegF, 4, f4->as_VMReg() );
reg_def F4_H ( SOC, SOC, Op_RegF, 4, f4->as_VMReg()->next() );
reg_def F5 ( SOC, SOC, Op_RegF, 5, f5->as_VMReg() );
reg_def F5_H ( SOC, SOC, Op_RegF, 5, f5->as_VMReg()->next() );
reg_def F6 ( SOC, SOC, Op_RegF, 6, f6->as_VMReg() );
reg_def F6_H ( SOC, SOC, Op_RegF, 6, f6->as_VMReg()->next() );
reg_def F7 ( SOC, SOC, Op_RegF, 7, f7->as_VMReg() );
reg_def F7_H ( SOC, SOC, Op_RegF, 7, f7->as_VMReg()->next() );
reg_def F8 ( SOC, SOE, Op_RegF, 8, f8->as_VMReg() );
reg_def F8_H ( SOC, SOE, Op_RegF, 8, f8->as_VMReg()->next() );
reg_def F9 ( SOC, SOE, Op_RegF, 9, f9->as_VMReg() );
reg_def F9_H ( SOC, SOE, Op_RegF, 9, f9->as_VMReg()->next() );
reg_def F10 ( SOC, SOC, Op_RegF, 10, f10->as_VMReg() );
reg_def F10_H ( SOC, SOC, Op_RegF, 10, f10->as_VMReg()->next() );
reg_def F11 ( SOC, SOC, Op_RegF, 11, f11->as_VMReg() );
reg_def F11_H ( SOC, SOC, Op_RegF, 11, f11->as_VMReg()->next() );
reg_def F12 ( SOC, SOC, Op_RegF, 12, f12->as_VMReg() );
reg_def F12_H ( SOC, SOC, Op_RegF, 12, f12->as_VMReg()->next() );
reg_def F13 ( SOC, SOC, Op_RegF, 13, f13->as_VMReg() );
reg_def F13_H ( SOC, SOC, Op_RegF, 13, f13->as_VMReg()->next() );
reg_def F14 ( SOC, SOC, Op_RegF, 14, f14->as_VMReg() );
reg_def F14_H ( SOC, SOC, Op_RegF, 14, f14->as_VMReg()->next() );
reg_def F15 ( SOC, SOC, Op_RegF, 15, f15->as_VMReg() );
reg_def F15_H ( SOC, SOC, Op_RegF, 15, f15->as_VMReg()->next() );
reg_def F16 ( SOC, SOC, Op_RegF, 16, f16->as_VMReg() );
reg_def F16_H ( SOC, SOC, Op_RegF, 16, f16->as_VMReg()->next() );
reg_def F17 ( SOC, SOC, Op_RegF, 17, f17->as_VMReg() );
reg_def F17_H ( SOC, SOC, Op_RegF, 17, f17->as_VMReg()->next() );
reg_def F18 ( SOC, SOE, Op_RegF, 18, f18->as_VMReg() );
reg_def F18_H ( SOC, SOE, Op_RegF, 18, f18->as_VMReg()->next() );
reg_def F19 ( SOC, SOE, Op_RegF, 19, f19->as_VMReg() );
reg_def F19_H ( SOC, SOE, Op_RegF, 19, f19->as_VMReg()->next() );
reg_def F20 ( SOC, SOE, Op_RegF, 20, f20->as_VMReg() );
reg_def F20_H ( SOC, SOE, Op_RegF, 20, f20->as_VMReg()->next() );
reg_def F21 ( SOC, SOE, Op_RegF, 21, f21->as_VMReg() );
reg_def F21_H ( SOC, SOE, Op_RegF, 21, f21->as_VMReg()->next() );
reg_def F22 ( SOC, SOE, Op_RegF, 22, f22->as_VMReg() );
reg_def F22_H ( SOC, SOE, Op_RegF, 22, f22->as_VMReg()->next() );
reg_def F23 ( SOC, SOE, Op_RegF, 23, f23->as_VMReg() );
reg_def F23_H ( SOC, SOE, Op_RegF, 23, f23->as_VMReg()->next() );
reg_def F24 ( SOC, SOE, Op_RegF, 24, f24->as_VMReg() );
reg_def F24_H ( SOC, SOE, Op_RegF, 24, f24->as_VMReg()->next() );
reg_def F25 ( SOC, SOE, Op_RegF, 25, f25->as_VMReg() );
reg_def F25_H ( SOC, SOE, Op_RegF, 25, f25->as_VMReg()->next() );
reg_def F26 ( SOC, SOE, Op_RegF, 26, f26->as_VMReg() );
reg_def F26_H ( SOC, SOE, Op_RegF, 26, f26->as_VMReg()->next() );
reg_def F27 ( SOC, SOE, Op_RegF, 27, f27->as_VMReg() );
reg_def F27_H ( SOC, SOE, Op_RegF, 27, f27->as_VMReg()->next() );
reg_def F28 ( SOC, SOC, Op_RegF, 28, f28->as_VMReg() );
reg_def F28_H ( SOC, SOC, Op_RegF, 28, f28->as_VMReg()->next() );
reg_def F29 ( SOC, SOC, Op_RegF, 29, f29->as_VMReg() );
reg_def F29_H ( SOC, SOC, Op_RegF, 29, f29->as_VMReg()->next() );
reg_def F30 ( SOC, SOC, Op_RegF, 30, f30->as_VMReg() );
reg_def F30_H ( SOC, SOC, Op_RegF, 30, f30->as_VMReg()->next() );
reg_def F31 ( SOC, SOC, Op_RegF, 31, f31->as_VMReg() );
reg_def F31_H ( SOC, SOC, Op_RegF, 31, f31->as_VMReg()->next() );
// ----------------------------
// Vector Registers
// ----------------------------
// For RVV vector registers, we simply extend vector register size to 4
// 'logical' slots. This is nominally 128 bits but it actually covers
// all possible 'physical' RVV vector register lengths from 128 ~ 1024
// bits. The 'physical' RVV vector register length is detected during
// startup, so the register allocator is able to identify the correct
// number of bytes needed for an RVV spill/unspill.
reg_def V0 ( SOC, SOC, Op_VecA, 0, v0->as_VMReg() );
reg_def V0_H ( SOC, SOC, Op_VecA, 0, v0->as_VMReg()->next() );
reg_def V0_J ( SOC, SOC, Op_VecA, 0, v0->as_VMReg()->next(2) );
reg_def V0_K ( SOC, SOC, Op_VecA, 0, v0->as_VMReg()->next(3) );
reg_def V1 ( SOC, SOC, Op_VecA, 1, v1->as_VMReg() );
reg_def V1_H ( SOC, SOC, Op_VecA, 1, v1->as_VMReg()->next() );
reg_def V1_J ( SOC, SOC, Op_VecA, 1, v1->as_VMReg()->next(2) );
reg_def V1_K ( SOC, SOC, Op_VecA, 1, v1->as_VMReg()->next(3) );
reg_def V2 ( SOC, SOC, Op_VecA, 2, v2->as_VMReg() );
reg_def V2_H ( SOC, SOC, Op_VecA, 2, v2->as_VMReg()->next() );
reg_def V2_J ( SOC, SOC, Op_VecA, 2, v2->as_VMReg()->next(2) );
reg_def V2_K ( SOC, SOC, Op_VecA, 2, v2->as_VMReg()->next(3) );
reg_def V3 ( SOC, SOC, Op_VecA, 3, v3->as_VMReg() );
reg_def V3_H ( SOC, SOC, Op_VecA, 3, v3->as_VMReg()->next() );
reg_def V3_J ( SOC, SOC, Op_VecA, 3, v3->as_VMReg()->next(2) );
reg_def V3_K ( SOC, SOC, Op_VecA, 3, v3->as_VMReg()->next(3) );
reg_def V4 ( SOC, SOC, Op_VecA, 4, v4->as_VMReg() );
reg_def V4_H ( SOC, SOC, Op_VecA, 4, v4->as_VMReg()->next() );
reg_def V4_J ( SOC, SOC, Op_VecA, 4, v4->as_VMReg()->next(2) );
reg_def V4_K ( SOC, SOC, Op_VecA, 4, v4->as_VMReg()->next(3) );
reg_def V5 ( SOC, SOC, Op_VecA, 5, v5->as_VMReg() );
reg_def V5_H ( SOC, SOC, Op_VecA, 5, v5->as_VMReg()->next() );
reg_def V5_J ( SOC, SOC, Op_VecA, 5, v5->as_VMReg()->next(2) );
reg_def V5_K ( SOC, SOC, Op_VecA, 5, v5->as_VMReg()->next(3) );
reg_def V6 ( SOC, SOC, Op_VecA, 6, v6->as_VMReg() );
reg_def V6_H ( SOC, SOC, Op_VecA, 6, v6->as_VMReg()->next() );
reg_def V6_J ( SOC, SOC, Op_VecA, 6, v6->as_VMReg()->next(2) );
reg_def V6_K ( SOC, SOC, Op_VecA, 6, v6->as_VMReg()->next(3) );
reg_def V7 ( SOC, SOC, Op_VecA, 7, v7->as_VMReg() );
reg_def V7_H ( SOC, SOC, Op_VecA, 7, v7->as_VMReg()->next() );
reg_def V7_J ( SOC, SOC, Op_VecA, 7, v7->as_VMReg()->next(2) );
reg_def V7_K ( SOC, SOC, Op_VecA, 7, v7->as_VMReg()->next(3) );
reg_def V8 ( SOC, SOC, Op_VecA, 8, v8->as_VMReg() );
reg_def V8_H ( SOC, SOC, Op_VecA, 8, v8->as_VMReg()->next() );
reg_def V8_J ( SOC, SOC, Op_VecA, 8, v8->as_VMReg()->next(2) );
reg_def V8_K ( SOC, SOC, Op_VecA, 8, v8->as_VMReg()->next(3) );
reg_def V9 ( SOC, SOC, Op_VecA, 9, v9->as_VMReg() );
reg_def V9_H ( SOC, SOC, Op_VecA, 9, v9->as_VMReg()->next() );
reg_def V9_J ( SOC, SOC, Op_VecA, 9, v9->as_VMReg()->next(2) );
reg_def V9_K ( SOC, SOC, Op_VecA, 9, v9->as_VMReg()->next(3) );
reg_def V10 ( SOC, SOC, Op_VecA, 10, v10->as_VMReg() );
reg_def V10_H ( SOC, SOC, Op_VecA, 10, v10->as_VMReg()->next() );
reg_def V10_J ( SOC, SOC, Op_VecA, 10, v10->as_VMReg()->next(2) );
reg_def V10_K ( SOC, SOC, Op_VecA, 10, v10->as_VMReg()->next(3) );
reg_def V11 ( SOC, SOC, Op_VecA, 11, v11->as_VMReg() );
reg_def V11_H ( SOC, SOC, Op_VecA, 11, v11->as_VMReg()->next() );
reg_def V11_J ( SOC, SOC, Op_VecA, 11, v11->as_VMReg()->next(2) );
reg_def V11_K ( SOC, SOC, Op_VecA, 11, v11->as_VMReg()->next(3) );
reg_def V12 ( SOC, SOC, Op_VecA, 12, v12->as_VMReg() );
reg_def V12_H ( SOC, SOC, Op_VecA, 12, v12->as_VMReg()->next() );
reg_def V12_J ( SOC, SOC, Op_VecA, 12, v12->as_VMReg()->next(2) );
reg_def V12_K ( SOC, SOC, Op_VecA, 12, v12->as_VMReg()->next(3) );
reg_def V13 ( SOC, SOC, Op_VecA, 13, v13->as_VMReg() );
reg_def V13_H ( SOC, SOC, Op_VecA, 13, v13->as_VMReg()->next() );
reg_def V13_J ( SOC, SOC, Op_VecA, 13, v13->as_VMReg()->next(2) );
reg_def V13_K ( SOC, SOC, Op_VecA, 13, v13->as_VMReg()->next(3) );
reg_def V14 ( SOC, SOC, Op_VecA, 14, v14->as_VMReg() );
reg_def V14_H ( SOC, SOC, Op_VecA, 14, v14->as_VMReg()->next() );
reg_def V14_J ( SOC, SOC, Op_VecA, 14, v14->as_VMReg()->next(2) );
reg_def V14_K ( SOC, SOC, Op_VecA, 14, v14->as_VMReg()->next(3) );
reg_def V15 ( SOC, SOC, Op_VecA, 15, v15->as_VMReg() );
reg_def V15_H ( SOC, SOC, Op_VecA, 15, v15->as_VMReg()->next() );
reg_def V15_J ( SOC, SOC, Op_VecA, 15, v15->as_VMReg()->next(2) );
reg_def V15_K ( SOC, SOC, Op_VecA, 15, v15->as_VMReg()->next(3) );
reg_def V16 ( SOC, SOC, Op_VecA, 16, v16->as_VMReg() );
reg_def V16_H ( SOC, SOC, Op_VecA, 16, v16->as_VMReg()->next() );
reg_def V16_J ( SOC, SOC, Op_VecA, 16, v16->as_VMReg()->next(2) );
reg_def V16_K ( SOC, SOC, Op_VecA, 16, v16->as_VMReg()->next(3) );
reg_def V17 ( SOC, SOC, Op_VecA, 17, v17->as_VMReg() );
reg_def V17_H ( SOC, SOC, Op_VecA, 17, v17->as_VMReg()->next() );
reg_def V17_J ( SOC, SOC, Op_VecA, 17, v17->as_VMReg()->next(2) );
reg_def V17_K ( SOC, SOC, Op_VecA, 17, v17->as_VMReg()->next(3) );
reg_def V18 ( SOC, SOC, Op_VecA, 18, v18->as_VMReg() );
reg_def V18_H ( SOC, SOC, Op_VecA, 18, v18->as_VMReg()->next() );
reg_def V18_J ( SOC, SOC, Op_VecA, 18, v18->as_VMReg()->next(2) );
reg_def V18_K ( SOC, SOC, Op_VecA, 18, v18->as_VMReg()->next(3) );
reg_def V19 ( SOC, SOC, Op_VecA, 19, v19->as_VMReg() );
reg_def V19_H ( SOC, SOC, Op_VecA, 19, v19->as_VMReg()->next() );
reg_def V19_J ( SOC, SOC, Op_VecA, 19, v19->as_VMReg()->next(2) );
reg_def V19_K ( SOC, SOC, Op_VecA, 19, v19->as_VMReg()->next(3) );
reg_def V20 ( SOC, SOC, Op_VecA, 20, v20->as_VMReg() );
reg_def V20_H ( SOC, SOC, Op_VecA, 20, v20->as_VMReg()->next() );
reg_def V20_J ( SOC, SOC, Op_VecA, 20, v20->as_VMReg()->next(2) );
reg_def V20_K ( SOC, SOC, Op_VecA, 20, v20->as_VMReg()->next(3) );
reg_def V21 ( SOC, SOC, Op_VecA, 21, v21->as_VMReg() );
reg_def V21_H ( SOC, SOC, Op_VecA, 21, v21->as_VMReg()->next() );
reg_def V21_J ( SOC, SOC, Op_VecA, 21, v21->as_VMReg()->next(2) );
reg_def V21_K ( SOC, SOC, Op_VecA, 21, v21->as_VMReg()->next(3) );
reg_def V22 ( SOC, SOC, Op_VecA, 22, v22->as_VMReg() );
reg_def V22_H ( SOC, SOC, Op_VecA, 22, v22->as_VMReg()->next() );
reg_def V22_J ( SOC, SOC, Op_VecA, 22, v22->as_VMReg()->next(2) );
reg_def V22_K ( SOC, SOC, Op_VecA, 22, v22->as_VMReg()->next(3) );
reg_def V23 ( SOC, SOC, Op_VecA, 23, v23->as_VMReg() );
reg_def V23_H ( SOC, SOC, Op_VecA, 23, v23->as_VMReg()->next() );
reg_def V23_J ( SOC, SOC, Op_VecA, 23, v23->as_VMReg()->next(2) );
reg_def V23_K ( SOC, SOC, Op_VecA, 23, v23->as_VMReg()->next(3) );
reg_def V24 ( SOC, SOC, Op_VecA, 24, v24->as_VMReg() );
reg_def V24_H ( SOC, SOC, Op_VecA, 24, v24->as_VMReg()->next() );
reg_def V24_J ( SOC, SOC, Op_VecA, 24, v24->as_VMReg()->next(2) );
reg_def V24_K ( SOC, SOC, Op_VecA, 24, v24->as_VMReg()->next(3) );
reg_def V25 ( SOC, SOC, Op_VecA, 25, v25->as_VMReg() );
reg_def V25_H ( SOC, SOC, Op_VecA, 25, v25->as_VMReg()->next() );
reg_def V25_J ( SOC, SOC, Op_VecA, 25, v25->as_VMReg()->next(2) );
reg_def V25_K ( SOC, SOC, Op_VecA, 25, v25->as_VMReg()->next(3) );
reg_def V26 ( SOC, SOC, Op_VecA, 26, v26->as_VMReg() );
reg_def V26_H ( SOC, SOC, Op_VecA, 26, v26->as_VMReg()->next() );
reg_def V26_J ( SOC, SOC, Op_VecA, 26, v26->as_VMReg()->next(2) );
reg_def V26_K ( SOC, SOC, Op_VecA, 26, v26->as_VMReg()->next(3) );
reg_def V27 ( SOC, SOC, Op_VecA, 27, v27->as_VMReg() );
reg_def V27_H ( SOC, SOC, Op_VecA, 27, v27->as_VMReg()->next() );
reg_def V27_J ( SOC, SOC, Op_VecA, 27, v27->as_VMReg()->next(2) );
reg_def V27_K ( SOC, SOC, Op_VecA, 27, v27->as_VMReg()->next(3) );
reg_def V28 ( SOC, SOC, Op_VecA, 28, v28->as_VMReg() );
reg_def V28_H ( SOC, SOC, Op_VecA, 28, v28->as_VMReg()->next() );
reg_def V28_J ( SOC, SOC, Op_VecA, 28, v28->as_VMReg()->next(2) );
reg_def V28_K ( SOC, SOC, Op_VecA, 28, v28->as_VMReg()->next(3) );
reg_def V29 ( SOC, SOC, Op_VecA, 29, v29->as_VMReg() );
reg_def V29_H ( SOC, SOC, Op_VecA, 29, v29->as_VMReg()->next() );
reg_def V29_J ( SOC, SOC, Op_VecA, 29, v29->as_VMReg()->next(2) );
reg_def V29_K ( SOC, SOC, Op_VecA, 29, v29->as_VMReg()->next(3) );
reg_def V30 ( SOC, SOC, Op_VecA, 30, v30->as_VMReg() );
reg_def V30_H ( SOC, SOC, Op_VecA, 30, v30->as_VMReg()->next() );
reg_def V30_J ( SOC, SOC, Op_VecA, 30, v30->as_VMReg()->next(2) );
reg_def V30_K ( SOC, SOC, Op_VecA, 30, v30->as_VMReg()->next(3) );
reg_def V31 ( SOC, SOC, Op_VecA, 31, v31->as_VMReg() );
reg_def V31_H ( SOC, SOC, Op_VecA, 31, v31->as_VMReg()->next() );
reg_def V31_J ( SOC, SOC, Op_VecA, 31, v31->as_VMReg()->next(2) );
reg_def V31_K ( SOC, SOC, Op_VecA, 31, v31->as_VMReg()->next(3) );
// ----------------------------
// Special Registers
// ----------------------------
// On riscv, the physical flag register is missing, so we use t1 instead,
// to bridge the RegFlag semantics in share/opto
reg_def RFLAGS (SOC, SOC, Op_RegFlags, 6, x6->as_VMReg() );
// Specify priority of register selection within phases of register
// allocation. Highest priority is first. A useful heuristic is to
// give registers a low priority when they are required by machine
// instructions, like EAX and EDX on I486, and choose no-save registers
// before save-on-call, & save-on-call before save-on-entry. Registers
// which participate in fixed calling sequences should come last.
// Registers which are used as pairs must fall on an even boundary.
alloc_class chunk0(
// volatiles
R7, R7_H,
R28, R28_H,
R29, R29_H,
R30, R30_H,
R31, R31_H,
// arg registers
R10, R10_H,
R11, R11_H,
R12, R12_H,
R13, R13_H,
R14, R14_H,
R15, R15_H,
R16, R16_H,
R17, R17_H,
// non-volatiles
R9, R9_H,
R18, R18_H,
R19, R19_H,
R20, R20_H,
R21, R21_H,
R22, R22_H,
R24, R24_H,
R25, R25_H,
R26, R26_H,
// non-allocatable registers
R23, R23_H, // java thread
R27, R27_H, // heapbase
R4, R4_H, // thread
R8, R8_H, // fp
R0, R0_H, // zero
R1, R1_H, // ra
R2, R2_H, // sp
R3, R3_H, // gp
);
alloc_class chunk1(
// no save
F0, F0_H,
F1, F1_H,
F2, F2_H,
F3, F3_H,
F4, F4_H,
F5, F5_H,
F6, F6_H,
F7, F7_H,
F28, F28_H,
F29, F29_H,
F30, F30_H,
F31, F31_H,
// arg registers
F10, F10_H,
F11, F11_H,
F12, F12_H,
F13, F13_H,
F14, F14_H,
F15, F15_H,
F16, F16_H,
F17, F17_H,
// non-volatiles
F8, F8_H,
F9, F9_H,
F18, F18_H,
F19, F19_H,
F20, F20_H,
F21, F21_H,
F22, F22_H,
F23, F23_H,
F24, F24_H,
F25, F25_H,
F26, F26_H,
F27, F27_H,
);
alloc_class chunk2(
V0, V0_H, V0_J, V0_K,
V1, V1_H, V1_J, V1_K,
V2, V2_H, V2_J, V2_K,
V3, V3_H, V3_J, V3_K,
V4, V4_H, V4_J, V4_K,
V5, V5_H, V5_J, V5_K,
V6, V6_H, V6_J, V6_K,
V7, V7_H, V7_J, V7_K,
V8, V8_H, V8_J, V8_K,
V9, V9_H, V9_J, V9_K,
V10, V10_H, V10_J, V10_K,
V11, V11_H, V11_J, V11_K,
V12, V12_H, V12_J, V12_K,
V13, V13_H, V13_J, V13_K,
V14, V14_H, V14_J, V14_K,
V15, V15_H, V15_J, V15_K,
V16, V16_H, V16_J, V16_K,
V17, V17_H, V17_J, V17_K,
V18, V18_H, V18_J, V18_K,
V19, V19_H, V19_J, V19_K,
V20, V20_H, V20_J, V20_K,
V21, V21_H, V21_J, V21_K,
V22, V22_H, V22_J, V22_K,
V23, V23_H, V23_J, V23_K,
V24, V24_H, V24_J, V24_K,
V25, V25_H, V25_J, V25_K,
V26, V26_H, V26_J, V26_K,
V27, V27_H, V27_J, V27_K,
V28, V28_H, V28_J, V28_K,
V29, V29_H, V29_J, V29_K,
V30, V30_H, V30_J, V30_K,
V31, V31_H, V31_J, V31_K,
);
alloc_class chunk3(RFLAGS);
//----------Architecture Description Register Classes--------------------------
// Several register classes are automatically defined based upon information in
// this architecture description.
// 1) reg_class inline_cache_reg ( /* as def'd in frame section */ )
// 2) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
//
// Class for all 32 bit general purpose registers
reg_class all_reg32(
R0,
R1,
R2,
R3,
R4,
R7,
R8,
R9,
R10,
R11,
R12,
R13,
R14,
R15,
R16,
R17,
R18,
R19,
R20,
R21,
R22,
R23,
R24,
R25,
R26,
R27,
R28,
R29,
R30,
R31
);
// Class for any 32 bit integer registers (excluding zr)
reg_class any_reg32 %{
return _ANY_REG32_mask;
%}
// Singleton class for R10 int register
reg_class int_r10_reg(R10);
// Singleton class for R12 int register
reg_class int_r12_reg(R12);
// Singleton class for R13 int register
reg_class int_r13_reg(R13);
// Singleton class for R14 int register
reg_class int_r14_reg(R14);
// Class for all long integer registers
reg_class all_reg(
R0, R0_H,
R1, R1_H,
R2, R2_H,
R3, R3_H,
R4, R4_H,
R7, R7_H,
R8, R8_H,
R9, R9_H,
R10, R10_H,
R11, R11_H,
R12, R12_H,
R13, R13_H,
R14, R14_H,
R15, R15_H,
R16, R16_H,
R17, R17_H,
R18, R18_H,
R19, R19_H,
R20, R20_H,
R21, R21_H,
R22, R22_H,
R23, R23_H,
R24, R24_H,
R25, R25_H,
R26, R26_H,
R27, R27_H,
R28, R28_H,
R29, R29_H,
R30, R30_H,
R31, R31_H
);
// Class for all long integer registers (excluding zr)
reg_class any_reg %{
return _ANY_REG_mask;
%}
// Class for non-allocatable 32 bit registers
reg_class non_allocatable_reg32(
R0, // zr
R1, // ra
R2, // sp
R3, // gp
R4, // tp
R23 // java thread
);
// Class for non-allocatable 64 bit registers
reg_class non_allocatable_reg(
R0, R0_H, // zr
R1, R1_H, // ra
R2, R2_H, // sp
R3, R3_H, // gp
R4, R4_H, // tp
R23, R23_H // java thread
);
reg_class no_special_reg32 %{
return _NO_SPECIAL_REG32_mask;
%}
reg_class no_special_reg %{
return _NO_SPECIAL_REG_mask;
%}
reg_class ptr_reg %{
return _PTR_REG_mask;
%}
reg_class no_special_ptr_reg %{
return _NO_SPECIAL_PTR_REG_mask;
%}
// Class for 64 bit register r10
reg_class r10_reg(
R10, R10_H
);
// Class for 64 bit register r11
reg_class r11_reg(
R11, R11_H
);
// Class for 64 bit register r12
reg_class r12_reg(
R12, R12_H
);
// Class for 64 bit register r13
reg_class r13_reg(
R13, R13_H
);
// Class for 64 bit register r14
reg_class r14_reg(
R14, R14_H
);
// Class for 64 bit register r15
reg_class r15_reg(
R15, R15_H
);
// Class for 64 bit register r16
reg_class r16_reg(
R16, R16_H
);
// Class for method register
reg_class method_reg(
R31, R31_H
);
// Class for heapbase register
reg_class heapbase_reg(
R27, R27_H
);
// Class for java thread register
reg_class java_thread_reg(
R23, R23_H
);
reg_class r28_reg(
R28, R28_H
);
reg_class r29_reg(
R29, R29_H
);
reg_class r30_reg(
R30, R30_H
);
reg_class r31_reg(
R31, R31_H
);
// Class for zero registesr
reg_class zr_reg(
R0, R0_H
);
// Class for thread register
reg_class thread_reg(
R4, R4_H
);
// Class for frame pointer register
reg_class fp_reg(
R8, R8_H
);
// Class for link register
reg_class ra_reg(
R1, R1_H
);
// Class for long sp register
reg_class sp_reg(
R2, R2_H
);
// Class for all float registers
reg_class float_reg(
F0,
F1,
F2,
F3,
F4,
F5,
F6,
F7,
F8,
F9,
F10,
F11,
F12,
F13,
F14,
F15,
F16,
F17,
F18,
F19,
F20,
F21,
F22,
F23,
F24,
F25,
F26,
F27,
F28,
F29,
F30,
F31
);
// Double precision float registers have virtual `high halves' that
// are needed by the allocator.
// Class for all double registers
reg_class double_reg(
F0, F0_H,
F1, F1_H,
F2, F2_H,
F3, F3_H,
F4, F4_H,
F5, F5_H,
F6, F6_H,
F7, F7_H,
F8, F8_H,
F9, F9_H,
F10, F10_H,
F11, F11_H,
F12, F12_H,
F13, F13_H,
F14, F14_H,
F15, F15_H,
F16, F16_H,
F17, F17_H,
F18, F18_H,
F19, F19_H,
F20, F20_H,
F21, F21_H,
F22, F22_H,
F23, F23_H,
F24, F24_H,
F25, F25_H,
F26, F26_H,
F27, F27_H,
F28, F28_H,
F29, F29_H,
F30, F30_H,
F31, F31_H
);
// Class for all RVV vector registers
reg_class vectora_reg(
V1, V1_H, V1_J, V1_K,
V2, V2_H, V2_J, V2_K,
V3, V3_H, V3_J, V3_K,
V4, V4_H, V4_J, V4_K,
V5, V5_H, V5_J, V5_K,
V6, V6_H, V6_J, V6_K,
V7, V7_H, V7_J, V7_K,
V8, V8_H, V8_J, V8_K,
V9, V9_H, V9_J, V9_K,
V10, V10_H, V10_J, V10_K,
V11, V11_H, V11_J, V11_K,
V12, V12_H, V12_J, V12_K,
V13, V13_H, V13_J, V13_K,
V14, V14_H, V14_J, V14_K,
V15, V15_H, V15_J, V15_K,
V16, V16_H, V16_J, V16_K,
V17, V17_H, V17_J, V17_K,
V18, V18_H, V18_J, V18_K,
V19, V19_H, V19_J, V19_K,
V20, V20_H, V20_J, V20_K,
V21, V21_H, V21_J, V21_K,
V22, V22_H, V22_J, V22_K,
V23, V23_H, V23_J, V23_K,
V24, V24_H, V24_J, V24_K,
V25, V25_H, V25_J, V25_K,
V26, V26_H, V26_J, V26_K,
V27, V27_H, V27_J, V27_K,
V28, V28_H, V28_J, V28_K,
V29, V29_H, V29_J, V29_K,
V30, V30_H, V30_J, V30_K,
V31, V31_H, V31_J, V31_K
);
// Class for 64 bit register f0
reg_class f0_reg(
F0, F0_H
);
// Class for 64 bit register f1
reg_class f1_reg(
F1, F1_H
);
// Class for 64 bit register f2
reg_class f2_reg(
F2, F2_H
);
// Class for 64 bit register f3
reg_class f3_reg(
F3, F3_H
);
// class for vector register v1
reg_class v1_reg(
V1, V1_H, V1_J, V1_K
);
// class for vector register v2
reg_class v2_reg(
V2, V2_H, V2_J, V2_K
);
// class for vector register v3
reg_class v3_reg(
V3, V3_H, V3_J, V3_K
);
// class for vector register v4
reg_class v4_reg(
V4, V4_H, V4_J, V4_K
);
// class for vector register v5
reg_class v5_reg(
V5, V5_H, V5_J, V5_K
);
// class for condition codes
reg_class reg_flags(RFLAGS);
%}
//----------DEFINITION BLOCK---------------------------------------------------
// Define name --> value mappings to inform the ADLC of an integer valued name
// Current support includes integer values in the range [0, 0x7FFFFFFF]
// Format:
// int_def <name> ( <int_value>, <expression>);
// Generated Code in ad_<arch>.hpp
// #define <name> (<expression>)
// // value == <int_value>
// Generated code in ad_<arch>.cpp adlc_verification()
// assert( <name> == <int_value>, "Expect (<expression>) to equal <int_value>");
//
// we follow the ppc-aix port in using a simple cost model which ranks
// register operations as cheap, memory ops as more expensive and
// branches as most expensive. the first two have a low as well as a
// normal cost. huge cost appears to be a way of saying don't do
// something
definitions %{
// The default cost (of a register move instruction).
int_def DEFAULT_COST ( 100, 100);
int_def ALU_COST ( 100, 1 * DEFAULT_COST); // unknown, const, arith, shift, slt,
// multi, auipc, nop, logical, move
int_def LOAD_COST ( 300, 3 * DEFAULT_COST); // load, fpload
int_def STORE_COST ( 100, 1 * DEFAULT_COST); // store, fpstore
int_def XFER_COST ( 300, 3 * DEFAULT_COST); // mfc, mtc, fcvt, fmove, fcmp
int_def BRANCH_COST ( 100, 1 * DEFAULT_COST); // branch, jmp, call
int_def IMUL_COST ( 1000, 10 * DEFAULT_COST); // imul
int_def IDIVSI_COST ( 3400, 34 * DEFAULT_COST); // idivdi
int_def IDIVDI_COST ( 6600, 66 * DEFAULT_COST); // idivsi
int_def FMUL_SINGLE_COST ( 500, 5 * DEFAULT_COST); // fadd, fmul, fmadd
int_def FMUL_DOUBLE_COST ( 700, 7 * DEFAULT_COST); // fadd, fmul, fmadd
int_def FDIV_COST ( 2000, 20 * DEFAULT_COST); // fdiv
int_def FSQRT_COST ( 2500, 25 * DEFAULT_COST); // fsqrt
int_def VOLATILE_REF_COST ( 1000, 10 * DEFAULT_COST);
%}
//----------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 %{
#include "asm/macroAssembler.hpp"
#include "gc/shared/barrierSetAssembler.hpp"
#include "gc/shared/cardTable.hpp"
#include "gc/shared/cardTableBarrierSet.hpp"
#include "gc/shared/collectedHeap.hpp"
#include "opto/addnode.hpp"
#include "opto/convertnode.hpp"
#include "runtime/objectMonitor.hpp"
extern RegMask _ANY_REG32_mask;
extern RegMask _ANY_REG_mask;
extern RegMask _PTR_REG_mask;
extern RegMask _NO_SPECIAL_REG32_mask;
extern RegMask _NO_SPECIAL_REG_mask;
extern RegMask _NO_SPECIAL_PTR_REG_mask;
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() {
return MacroAssembler::far_branch_size();
}
static uint size_deopt_handler() {
// count auipc + far branch
return NativeInstruction::instruction_size + MacroAssembler::far_branch_size();
}
};
class Node::PD {
public:
enum NodeFlags {
_last_flag = Node::_last_flag
};
};
bool is_CAS(int opcode, bool maybe_volatile);
// predicate controlling translation of CompareAndSwapX
bool needs_acquiring_load_reserved(const Node *load);
// predicate controlling addressing modes
bool size_fits_all_mem_uses(AddPNode* addp, int shift);
%}
source %{
// Derived RegMask with conditionally allocatable registers
RegMask _ANY_REG32_mask;
RegMask _ANY_REG_mask;
RegMask _PTR_REG_mask;
RegMask _NO_SPECIAL_REG32_mask;
RegMask _NO_SPECIAL_REG_mask;
RegMask _NO_SPECIAL_PTR_REG_mask;
void reg_mask_init() {
_ANY_REG32_mask = _ALL_REG32_mask;
_ANY_REG32_mask.Remove(OptoReg::as_OptoReg(x0->as_VMReg()));
_ANY_REG_mask = _ALL_REG_mask;
_ANY_REG_mask.SUBTRACT(_ZR_REG_mask);
_PTR_REG_mask = _ALL_REG_mask;
_PTR_REG_mask.SUBTRACT(_ZR_REG_mask);
_NO_SPECIAL_REG32_mask = _ALL_REG32_mask;
_NO_SPECIAL_REG32_mask.SUBTRACT(_NON_ALLOCATABLE_REG32_mask);
_NO_SPECIAL_REG_mask = _ALL_REG_mask;
_NO_SPECIAL_REG_mask.SUBTRACT(_NON_ALLOCATABLE_REG_mask);
_NO_SPECIAL_PTR_REG_mask = _ALL_REG_mask;
_NO_SPECIAL_PTR_REG_mask.SUBTRACT(_NON_ALLOCATABLE_REG_mask);
// x27 is not allocatable when compressed oops is on
if (UseCompressedOops) {
_NO_SPECIAL_REG32_mask.Remove(OptoReg::as_OptoReg(x27->as_VMReg()));
_NO_SPECIAL_REG_mask.SUBTRACT(_HEAPBASE_REG_mask);
_NO_SPECIAL_PTR_REG_mask.SUBTRACT(_HEAPBASE_REG_mask);
}
// x8 is not allocatable when PreserveFramePointer is on
if (PreserveFramePointer) {
_NO_SPECIAL_REG32_mask.Remove(OptoReg::as_OptoReg(x8->as_VMReg()));
_NO_SPECIAL_REG_mask.SUBTRACT(_FP_REG_mask);
_NO_SPECIAL_PTR_REG_mask.SUBTRACT(_FP_REG_mask);
}
}
void PhaseOutput::pd_perform_mach_node_analysis() {
}
int MachNode::pd_alignment_required() const {
return 1;
}
int MachNode::compute_padding(int current_offset) const {
return 0;
}
// is_CAS(int opcode, bool maybe_volatile)
//
// return true if opcode is one of the possible CompareAndSwapX
// values otherwise false.
bool is_CAS(int opcode, bool maybe_volatile)
{
switch (opcode) {
// We handle these
case Op_CompareAndSwapI:
case Op_CompareAndSwapL:
case Op_CompareAndSwapP:
case Op_CompareAndSwapN:
case Op_ShenandoahCompareAndSwapP:
case Op_ShenandoahCompareAndSwapN:
case Op_CompareAndSwapB:
case Op_CompareAndSwapS:
case Op_GetAndSetI:
case Op_GetAndSetL:
case Op_GetAndSetP:
case Op_GetAndSetN:
case Op_GetAndAddI:
case Op_GetAndAddL:
return true;
case Op_CompareAndExchangeI:
case Op_CompareAndExchangeN:
case Op_CompareAndExchangeB:
case Op_CompareAndExchangeS:
case Op_CompareAndExchangeL:
case Op_CompareAndExchangeP:
case Op_WeakCompareAndSwapB:
case Op_WeakCompareAndSwapS:
case Op_WeakCompareAndSwapI:
case Op_WeakCompareAndSwapL:
case Op_WeakCompareAndSwapP:
case Op_WeakCompareAndSwapN:
case Op_ShenandoahWeakCompareAndSwapP:
case Op_ShenandoahWeakCompareAndSwapN:
case Op_ShenandoahCompareAndExchangeP:
case Op_ShenandoahCompareAndExchangeN:
return maybe_volatile;
default:
return false;
}
}
// predicate controlling translation of CAS
//
// returns true if CAS needs to use an acquiring load otherwise false
bool needs_acquiring_load_reserved(const Node *n)
{
assert(n != NULL && is_CAS(n->Opcode(), true), "expecting a compare and swap");
LoadStoreNode* ldst = n->as_LoadStore();
if (n != NULL && is_CAS(n->Opcode(), false)) {
assert(ldst != NULL && ldst->trailing_membar() != NULL, "expected trailing membar");
} else {
return ldst != NULL && ldst->trailing_membar() != NULL;
}
// so we can just return true here
return true;
}
#define __ _masm.
// advance declarations for helper functions to convert register
// indices to register objects
// the ad file has to provide implementations of certain methods
// expected by the generic code
//
// REQUIRED FUNCTIONALITY
//=============================================================================
// !!!!! Special hack to get all types of calls to specify the byte offset
// from the start of the call to the point where the return address
// will point.
int MachCallStaticJavaNode::ret_addr_offset()
{
// jal
return 1 * NativeInstruction::instruction_size;
}
int MachCallDynamicJavaNode::ret_addr_offset()
{
return 7 * NativeInstruction::instruction_size; // movptr, jal
}
int MachCallRuntimeNode::ret_addr_offset() {
// for generated stubs the call will be
// jal(addr)
// or with far branches
// jal(trampoline_stub)
// for real runtime callouts it will be 11 instructions
// see riscv_enc_java_to_runtime
// la(t1, retaddr) -> auipc + addi
// la(t0, RuntimeAddress(addr)) -> lui + addi + slli + addi + slli + addi
// addi(sp, sp, -2 * wordSize) -> addi
// sd(t1, Address(sp, wordSize)) -> sd
// jalr(t0) -> jalr
CodeBlob *cb = CodeCache::find_blob(_entry_point);
if (cb != NULL) {
return 1 * NativeInstruction::instruction_size;
} else {
return 11 * NativeInstruction::instruction_size;
}
}
//
// Compute padding required for nodes which need alignment
//
// With RVC a call instruction may get 2-byte aligned.
// The address of the call instruction needs to be 4-byte aligned to
// ensure that it does not span a cache line so that it can be patched.
int CallStaticJavaDirectNode::compute_padding(int current_offset) const
{
// to make sure the address of jal 4-byte aligned.
return align_up(current_offset, alignment_required()) - current_offset;
}
// With RVC a call instruction may get 2-byte aligned.
// The address of the call instruction needs to be 4-byte aligned to
// ensure that it does not span a cache line so that it can be patched.
int CallDynamicJavaDirectNode::compute_padding(int current_offset) const
{
// skip the movptr in MacroAssembler::ic_call():
// lui + addi + slli + addi + slli + addi
// Though movptr() has already 4-byte aligned with or without RVC,
// We need to prevent from further changes by explicitly calculating the size.
const int movptr_size = 6 * NativeInstruction::instruction_size;
current_offset += movptr_size;
// to make sure the address of jal 4-byte aligned.
return align_up(current_offset, alignment_required()) - current_offset;
}
//=============================================================================
#ifndef PRODUCT
void MachBreakpointNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
assert_cond(st != NULL);
st->print("BREAKPOINT");
}
#endif
void MachBreakpointNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
C2_MacroAssembler _masm(&cbuf);
__ ebreak();
}
uint MachBreakpointNode::size(PhaseRegAlloc *ra_) const {
return MachNode::size(ra_);
}
//=============================================================================
#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 {
C2_MacroAssembler _masm(&cbuf);
Assembler::CompressibleRegion cr(&_masm); // nops shall be 2-byte under RVC for alignment purposes.
for (int i = 0; i < _count; i++) {
__ nop();
}
}
uint MachNopNode::size(PhaseRegAlloc*) const {
return _count * (UseRVC ? NativeInstruction::compressed_instruction_size : NativeInstru ction::instruction_size);
}
//=============================================================================
const RegMask& MachConstantBaseNode::_out_RegMask = RegMask::Empty;
int ConstantTable::calculate_table_base_offset() const {
return 0; // absolute addressing, no offset
}
bool MachConstantBaseNode::requires_postalloc_expand() const { return false; }
void MachConstantBaseNode::postalloc_expand(GrowableArray <Node *> *nodes, PhaseRegAlloc *ra_) {
ShouldNotReachHere();
}
void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
// Empty encoding
}
uint MachConstantBaseNode::size(PhaseRegAlloc* ra_) const {
return 0;
}
#ifndef PRODUCT
void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
assert_cond(st != NULL);
st->print("-- \t// MachConstantBaseNode (empty encoding)");
}
#endif
#ifndef PRODUCT
void MachPrologNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
assert_cond(st != NULL && ra_ != NULL);
Compile* C = ra_->C;
int framesize = C->output()->frame_slots() << LogBytesPerInt;
if (C->output()->need_stack_bang(framesize)) {
st->print("# stack bang size=%d\n\t", framesize);
}
st->print("sd fp, [sp, #%d]\n\t", - 2 * wordSize);
st->print("sd ra, [sp, #%d]\n\t", - wordSize);
if (PreserveFramePointer) { st->print("sub fp, sp, #%d\n\t", 2 * wordSize); }
st->print("sub sp, sp, #%d\n\t", framesize);
if (C->stub_function() == NULL && BarrierSet::barrier_set()->barrier_set_nmethod() != NULL) {
st->print("ld t0, [guard]\n\t");
st->print("membar LoadLoad\n\t");
st->print("ld t1, [xthread, #thread_disarmed_offset]\n\t");
st->print("beq t0, t1, skip\n\t");
st->print("jalr #nmethod_entry_barrier_stub\n\t");
st->print("j skip\n\t");
st->print("guard: int\n\t");
st->print("skip:\n\t");
}
}
#endif
void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
assert_cond(ra_ != NULL);
Compile* C = ra_->C;
C2_MacroAssembler _masm(&cbuf);
// n.b. frame size includes space for return pc and fp
const int framesize = C->output()->frame_size_in_bytes();
// insert a nop at the start of the prolog so we can patch in a
// branch if we need to invalidate the method later
{
Assembler::IncompressibleRegion ir(&_masm); // keep the nop as 4 bytes for patching.
MacroAssembler::assert_alignment(__ pc());
__ nop(); // 4 bytes
}
assert_cond(C != NULL);
if (C->clinit_barrier_on_entry()) {
assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
Label L_skip_barrier;
__ mov_metadata(t1, C->method()->holder()->constant_encoding());
__ clinit_barrier(t1, t0, &L_skip_barrier);
__ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
__ bind(L_skip_barrier);
}
int bangsize = C->output()->bang_size_in_bytes();
if (C->output()->need_stack_bang(bangsize)) {
__ generate_stack_overflow_check(bangsize);
}
__ build_frame(framesize);
if (C->stub_function() == NULL) {
BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
if (BarrierSet::barrier_set()->barrier_set_nmethod() != NULL) {
// Dummy labels for just measuring the code size
Label dummy_slow_path;
Label dummy_continuation;
Label dummy_guard;
Label* slow_path = &dummy_slow_path;
Label* continuation = &dummy_continuation;
Label* guard = &dummy_guard;
if (!Compile::current()->output()->in_scratch_emit_size()) {
// Use real labels from actual stub when not emitting code for purpose of measuring its size
C2EntryBarrierStub* stub = Compile::current()->output()->entry_barrier_table()->add_entry_barrier();
slow_path = &stub->slow_path();
continuation = &stub->continuation();
guard = &stub->guard();
}
// In the C2 code, we move the non-hot part of nmethod entry barriers out-of-line to a stub.
bs->nmethod_entry_barrier(&_masm, slow_path, continuation, guard);
}
}
if (VerifyStackAtCalls) {
Unimplemented();
}
C->output()->set_frame_complete(cbuf.insts_size());
if (C->has_mach_constant_base_node()) {
// NOTE: We set the table base offset here because users might be
// emitted before MachConstantBaseNode.
ConstantTable& constant_table = C->output()->constant_table();
constant_table.set_table_base_offset(constant_table.calculate_table_base_offset());
}
}
uint MachPrologNode::size(PhaseRegAlloc* ra_) const
{
assert_cond(ra_ != NULL);
return MachNode::size(ra_); // too many variables; just compute it
// the hard way
}
int MachPrologNode::reloc() const
{
return 0;
}
//=============================================================================
#ifndef PRODUCT
void MachEpilogNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
assert_cond(st != NULL && ra_ != NULL);
Compile* C = ra_->C;
assert_cond(C != NULL);
int framesize = C->output()->frame_size_in_bytes();
st->print("# pop frame %d\n\t", framesize);
if (framesize == 0) {
st->print("ld ra, [sp,#%d]\n\t", (2 * wordSize));
st->print("ld fp, [sp,#%d]\n\t", (3 * wordSize));
st->print("add sp, sp, #%d\n\t", (2 * wordSize));
} else {
st->print("add sp, sp, #%d\n\t", framesize);
st->print("ld ra, [sp,#%d]\n\t", - 2 * wordSize);
st->print("ld fp, [sp,#%d]\n\t", - wordSize);
}
if (do_polling() && C->is_method_compilation()) {
st->print("# test polling word\n\t");
st->print("ld t0, [xthread,#%d]\n\t", in_bytes(JavaThread::polling_word_offset()));
st->print("bgtu sp, t0, #slow_path");
}
}
#endif
void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
assert_cond(ra_ != NULL);
Compile* C = ra_->C;
C2_MacroAssembler _masm(&cbuf);
assert_cond(C != NULL);
int framesize = C->output()->frame_size_in_bytes();
__ remove_frame(framesize);
if (StackReservedPages > 0 && C->has_reserved_stack_access()) {
__ reserved_stack_check();
}
if (do_polling() && C->is_method_compilation()) {
Label dummy_label;
Label* code_stub = &dummy_label;
if (!C->output()->in_scratch_emit_size()) {
code_stub = &C->output()->safepoint_poll_table()->add_safepoint(__ offset());
}
__ relocate(relocInfo::poll_return_type);
__ safepoint_poll(*code_stub, true /* at_return */, false /* acquire */, true /* in_nmethod */);
}
}
uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
assert_cond(ra_ != NULL);
// Variable size. Determine dynamically.
return MachNode::size(ra_);
}
int MachEpilogNode::reloc() const {
// Return number of relocatable values contained in this instruction.
return 1; // 1 for polling page.
}
const Pipeline * MachEpilogNode::pipeline() const {
return MachNode::pipeline_class();
}
//=============================================================================
// Figure out which register class each belongs in: rc_int, rc_float or
// rc_stack.
enum RC { rc_bad, rc_int, rc_float, rc_vector, rc_stack };
static enum RC rc_class(OptoReg::Name reg) {
if (reg == OptoReg::Bad) {
return rc_bad;
}
// we have 30 int registers * 2 halves
// (t0 and t1 are omitted)
int slots_of_int_registers = Register::max_slots_per_register * (Register::number_of_registers - 2);
if (reg < slots_of_int_registers) {
return rc_int;
}
// we have 32 float register * 2 halves
int slots_of_float_registers = FloatRegister::max_slots_per_register * FloatRegister::number_of_registers;
if (reg < slots_of_int_registers + slots_of_float_registers) {
return rc_float;
}
// we have 32 vector register * 4 halves
int slots_of_vector_registers = VectorRegister::max_slots_per_register * VectorRegister::number_of_registers;
if (reg < slots_of_int_registers + slots_of_float_registers + slots_of_vector_registers) {
return rc_vector;
}
// Between vector regs & stack is the flags regs.
assert(OptoReg::is_stack(reg), "blow up if spilling flags");
return rc_stack;
}
uint MachSpillCopyNode::implementation(CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, outputStream *st) const {
assert_cond(ra_ != NULL);
Compile* C = ra_->C;
// Get registers to move.
OptoReg::Name src_hi = ra_->get_reg_second(in(1));
OptoReg::Name src_lo = ra_->get_reg_first(in(1));
OptoReg::Name dst_hi = ra_->get_reg_second(this);
OptoReg::Name dst_lo = ra_->get_reg_first(this);
enum RC src_hi_rc = rc_class(src_hi);
enum RC src_lo_rc = rc_class(src_lo);
enum RC dst_hi_rc = rc_class(dst_hi);
enum RC dst_lo_rc = rc_class(dst_lo);
assert(src_lo != OptoReg::Bad && dst_lo != OptoReg::Bad, "must move at least 1 register");
if (src_hi != OptoReg::Bad) {
assert((src_lo & 1) == 0 && src_lo + 1 == src_hi &&
(dst_lo & 1) == 0 && dst_lo + 1 == dst_hi,
"expected aligned-adjacent pairs");
}
if (src_lo == dst_lo && src_hi == dst_hi) {
return 0; // Self copy, no move.
}
bool is64 = (src_lo & 1) == 0 && src_lo + 1 == src_hi &&
(dst_lo & 1) == 0 && dst_lo + 1 == dst_hi;
int src_offset = ra_->reg2offset(src_lo);
int dst_offset = ra_->reg2offset(dst_lo);
if (bottom_type()->isa_vect() != NULL) {
uint ireg = ideal_reg();
if (ireg == Op_VecA && cbuf) {
C2_MacroAssembler _masm(cbuf);
int vector_reg_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
if (src_lo_rc == rc_stack && dst_lo_rc == rc_stack) {
// stack to stack
__ spill_copy_vector_stack_to_stack(src_offset, dst_offset,
vector_reg_size_in_bytes);
} else if (src_lo_rc == rc_vector && dst_lo_rc == rc_stack) {
// vpr to stack
__ spill(as_VectorRegister(Matcher::_regEncode[src_lo]), ra_->reg2offset(dst_lo));
} else if (src_lo_rc == rc_stack && dst_lo_rc == rc_vector) {
// stack to vpr
__ unspill(as_VectorRegister(Matcher::_regEncode[dst_lo]), ra_->reg2offset(src_lo));
} else if (src_lo_rc == rc_vector && dst_lo_rc == rc_vector) {
// vpr to vpr
__ vmv1r_v(as_VectorRegister(Matcher::_regEncode[dst_lo]), as_VectorRegister(Matcher::_regEncode[src_lo]));
} else {
ShouldNotReachHere();
}
}
} else if (cbuf != NULL) {
C2_MacroAssembler _masm(cbuf);
switch (src_lo_rc) {
case rc_int:
if (dst_lo_rc == rc_int) { // gpr --> gpr copy
if (!is64 && this->ideal_reg() != Op_RegI) { // zero extended for narrow oop or klass
__ zero_extend(as_Register(Matcher::_regEncode[dst_lo]), as_Register(Matcher::_regEncode[src_lo]), 32);
} else {
__ mv(as_Register(Matcher::_regEncode[dst_lo]), as_Register(Matcher::_regEncode[src_lo]));
}
} else if (dst_lo_rc == rc_float) { // gpr --> fpr copy
if (is64) {
__ fmv_d_x(as_FloatRegister(Matcher::_regEncode[dst_lo]),
as_Register(Matcher::_regEncode[src_lo]));
} else {
__ fmv_w_x(as_FloatRegister(Matcher::_regEncode[dst_lo]),
as_Register(Matcher::_regEncode[src_lo]));
}
} else { // gpr --> stack spill
assert(dst_lo_rc == rc_stack, "spill to bad register class");
__ spill(as_Register(Matcher::_regEncode[src_lo]), is64, dst_offset);
}
break;
case rc_float:
if (dst_lo_rc == rc_int) { // fpr --> gpr copy
if (is64) {
__ fmv_x_d(as_Register(Matcher::_regEncode[dst_lo]),
as_FloatRegister(Matcher::_regEncode[src_lo]));
} else {
__ fmv_x_w(as_Register(Matcher::_regEncode[dst_lo]),
as_FloatRegister(Matcher::_regEncode[src_lo]));
}
} else if (dst_lo_rc == rc_float) { // fpr --> fpr copy
if (is64) {
__ fmv_d(as_FloatRegister(Matcher::_regEncode[dst_lo]),
as_FloatRegister(Matcher::_regEncode[src_lo]));
} else {
__ fmv_s(as_FloatRegister(Matcher::_regEncode[dst_lo]),
as_FloatRegister(Matcher::_regEncode[src_lo]));
}
} else { // fpr --> stack spill
assert(dst_lo_rc == rc_stack, "spill to bad register class");
__ spill(as_FloatRegister(Matcher::_regEncode[src_lo]),
is64, dst_offset);
}
break;
case rc_stack:
if (dst_lo_rc == rc_int) { // stack --> gpr load
if (this->ideal_reg() == Op_RegI) {
__ unspill(as_Register(Matcher::_regEncode[dst_lo]), is64, src_offset);
} else { // // zero extended for narrow oop or klass
__ unspillu(as_Register(Matcher::_regEncode[dst_lo]), is64, src_offset);
}
} else if (dst_lo_rc == rc_float) { // stack --> fpr load
__ unspill(as_FloatRegister(Matcher::_regEncode[dst_lo]),
is64, src_offset);
} else { // stack --> stack copy
assert(dst_lo_rc == rc_stack, "spill to bad register class");
if (this->ideal_reg() == Op_RegI) {
__ unspill(t0, is64, src_offset);
} else { // zero extended for narrow oop or klass
__ unspillu(t0, is64, src_offset);
}
__ spill(t0, is64, dst_offset);
}
break;
default:
ShouldNotReachHere();
}
}
if (st != NULL) {
st->print("spill ");
if (src_lo_rc == rc_stack) {
st->print("[sp, #%d] -> ", src_offset);
} else {
st->print("%s -> ", Matcher::regName[src_lo]);
}
if (dst_lo_rc == rc_stack) {
st->print("[sp, #%d]", dst_offset);
} else {
st->print("%s", Matcher::regName[dst_lo]);
}
if (bottom_type()->isa_vect() != NULL) {
int vsize = 0;
if (ideal_reg() == Op_VecA) {
vsize = Matcher::scalable_vector_reg_size(T_BYTE) * 8;
} else {
ShouldNotReachHere();
}
st->print("\t# vector spill size = %d", vsize);
} else {
st->print("\t# spill size = %d", is64 ? 64 : 32);
}
}
return 0;
}
#ifndef PRODUCT
void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
if (ra_ == NULL) {
st->print("N%d = SpillCopy(N%d)", _idx, in(1)->_idx);
} else {
implementation(NULL, ra_, false, st);
}
}
#endif
void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
implementation(&cbuf, ra_, false, NULL);
}
uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const {
return MachNode::size(ra_);
}
//=============================================================================
#ifndef PRODUCT
void BoxLockNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
assert_cond(ra_ != NULL && st != NULL);
int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
int reg = ra_->get_reg_first(this);
st->print("add %s, sp, #%d\t# box lock",
Matcher::regName[reg], offset);
}
#endif
void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
C2_MacroAssembler _masm(&cbuf);
Assembler::IncompressibleRegion ir(&_masm); // Fixed length: see BoxLockNode::size()
assert_cond(ra_ != NULL);
int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
int reg = ra_->get_encode(this);
if (is_imm_in_range(offset, 12, 0)) {
__ addi(as_Register(reg), sp, offset);
} else if (is_imm_in_range(offset, 32, 0)) {
__ li32(t0, offset);
__ add(as_Register(reg), sp, t0);
} else {
ShouldNotReachHere();
}
}
uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
// BoxLockNode is not a MachNode, so we can't just call MachNode::size(ra_).
int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
if (is_imm_in_range(offset, 12, 0)) {
return NativeInstruction::instruction_size;
} else {
return 3 * NativeInstruction::instruction_size; // lui + addiw + add;
}
}
//=============================================================================
#ifndef PRODUCT
void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const
{
assert_cond(st != NULL);
st->print_cr("# MachUEPNode");
if (UseCompressedClassPointers) {
st->print_cr("\tlwu t0, [j_rarg0, oopDesc::klass_offset_in_bytes()]\t# compressed klass");
if (CompressedKlassPointers::shift() != 0) {
st->print_cr("\tdecode_klass_not_null t0, t0");
}
} else {
st->print_cr("\tld t0, [j_rarg0, oopDesc::klass_offset_in_bytes()]\t# compressed klass");
}
st->print_cr("\tbeq t0, t1, ic_hit");
st->print_cr("\tj, SharedRuntime::_ic_miss_stub\t # Inline cache check");
st->print_cr("\tic_hit:");
}
#endif
void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
{
// This is the unverified entry point.
C2_MacroAssembler _masm(&cbuf);
Label skip;
__ cmp_klass(j_rarg0, t1, t0, t2 /* call-clobbered t2 as a tmp */, skip);
__ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
__ bind(skip);
// These NOPs are critical so that verified entry point is properly
// 4 bytes aligned for patching by NativeJump::patch_verified_entry()
__ align(NativeInstruction::instruction_size);
}
uint MachUEPNode::size(PhaseRegAlloc* ra_) const
{
assert_cond(ra_ != NULL);
return MachNode::size(ra_);
}
// REQUIRED EMIT CODE
//=============================================================================
// Emit exception handler code.
int HandlerImpl::emit_exception_handler(CodeBuffer& cbuf)
{
// la_patchable t0, #exception_blob_entry_point
// jr (offset)t0
// or
// j #exception_blob_entry_point
// 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.
C2_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();
__ far_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.
C2_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();
__ auipc(ra, 0);
__ far_jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
__ end_a_stub();
return offset;
}
// REQUIRED MATCHER CODE
//=============================================================================
const bool Matcher::match_rule_supported(int opcode) {
if (!has_match_rule(opcode)) {
return false;
}
switch (opcode) {
case Op_CacheWB: // fall through
case Op_CacheWBPreSync: // fall through
case Op_CacheWBPostSync:
if (!VM_Version::supports_data_cache_line_flush()) {
return false;
}
break;
case Op_StrCompressedCopy: // fall through
case Op_StrInflatedCopy: // fall through
case Op_CountPositives:
return UseRVV;
case Op_EncodeISOArray:
return UseRVV && SpecialEncodeISOArray;
case Op_PopCountI:
case Op_PopCountL:
return UsePopCountInstruction;
case Op_RotateRight:
case Op_RotateLeft:
case Op_CountLeadingZerosI:
case Op_CountLeadingZerosL:
case Op_CountTrailingZerosI:
case Op_CountTrailingZerosL:
return UseZbb;
}
return true; // Per default match rules are supported.
}
const bool Matcher::match_rule_supported_superword(int opcode, int vlen, BasicType bt) {
return match_rule_supported_vector(opcode, vlen, bt);
}
// Identify extra cases that we might want to provide match rules for vector nodes and
// other intrinsics guarded with vector length (vlen) and element type (bt).
const bool Matcher::match_rule_supported_vector(int opcode, int vlen, BasicType bt) {
if (!match_rule_supported(opcode) || !vector_size_supported(bt, vlen)) {
return false;
}
return op_vec_supported(opcode);
}
const bool Matcher::match_rule_supported_vector_masked(int opcode, int vlen, BasicType bt) {
return false;
}
const bool Matcher::vector_needs_partial_operations(Node* node, const TypeVect* vt) {
return false;
}
const RegMask* Matcher::predicate_reg_mask(void) {
return NULL;
}
const TypeVectMask* Matcher::predicate_reg_type(const Type* elemTy, int length) {
return NULL;
}
// Vector calling convention not yet implemented.
const bool Matcher::supports_vector_calling_convention(void) {
return false;
}
OptoRegPair Matcher::vector_return_value(uint ideal_reg) {
Unimplemented();
return OptoRegPair(0, 0);
}
// Is this branch offset short enough that a short branch can be used?
//
// NOTE: If the platform does not provide any short branch variants, then
// this method should return false for offset 0.
// |---label(L1)-----|
// |-----------------|
// |-----------------|----------eq: float-------------------
// |-----------------| // far_cmpD_branch | cmpD_branch
// |------- ---------| feq; | feq;
// |-far_cmpD_branch-| beqz done; | bnez L;
// |-----------------| j L; |
// |-----------------| bind(done); |
// |-----------------|--------------------------------------
// |-----------------| // so shortBrSize = br_size - 4;
// |-----------------| // so offs = offset - shortBrSize + 4;
// |---label(L2)-----|
bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
// The passed offset is relative to address of the branch.
int shortBrSize = br_size - 4;
int offs = offset - shortBrSize + 4;
return (-4096 <= offs && offs < 4096);
}
// Vector width in bytes.
const int Matcher::vector_width_in_bytes(BasicType bt) {
if (UseRVV) {
// The MaxVectorSize should have been set by detecting RVV max vector register size when check UseRVV.
// MaxVectorSize == VM_Version::_initial_vector_length
return MaxVectorSize;
}
return 0;
}
// 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) {
return max_vector_size(bt);
}
// Vector ideal reg.
const uint Matcher::vector_ideal_reg(int len) {
assert(MaxVectorSize >= len, "");
if (UseRVV) {
return Op_VecA;
}
ShouldNotReachHere();
return 0;
}
const int Matcher::scalable_vector_reg_size(const BasicType bt) {
return Matcher::max_vector_size(bt);
}
MachOper* Matcher::pd_specialize_generic_vector_operand(MachOper* original_opnd, uint ideal_reg, bool is_temp) {
ShouldNotReachHere(); // generic vector operands not supported
return NULL;
}
bool Matcher::is_reg2reg_move(MachNode* m) {
ShouldNotReachHere(); // generic vector operands not supported
return false;
}
bool Matcher::is_generic_vector(MachOper* opnd) {
ShouldNotReachHere(); // generic vector operands not supported
return false;
}
// Return whether or not this register is ever used as an argument.
// This function is used on startup to build the trampoline stubs in
// generateOptoStub. Registers not mentioned will be killed by the VM
// call in the trampoline, and arguments in those registers not be
// available to the callee.
bool Matcher::can_be_java_arg(int reg)
{
return
reg == R10_num || reg == R10_H_num ||
reg == R11_num || reg == R11_H_num ||
reg == R12_num || reg == R12_H_num ||
reg == R13_num || reg == R13_H_num ||
reg == R14_num || reg == R14_H_num ||
reg == R15_num || reg == R15_H_num ||
reg == R16_num || reg == R16_H_num ||
reg == R17_num || reg == R17_H_num ||
reg == F10_num || reg == F10_H_num ||
reg == F11_num || reg == F11_H_num ||
reg == F12_num || reg == F12_H_num ||
reg == F13_num || reg == F13_H_num ||
reg == F14_num || reg == F14_H_num ||
reg == F15_num || reg == F15_H_num ||
reg == F16_num || reg == F16_H_num ||
reg == F17_num || reg == F17_H_num;
}
bool Matcher::is_spillable_arg(int reg)
{
return can_be_java_arg(reg);
}
uint Matcher::int_pressure_limit()
{
// A derived pointer is live at CallNode and then is flagged by RA
// as a spilled LRG. Spilling heuristics(Spill-USE) explicitly skip
// derived pointers and lastly fail to spill after reaching maximum
// number of iterations. Lowering the default pressure threshold to
// (_NO_SPECIAL_REG32_mask.Size() minus 1) forces CallNode to become
// a high register pressure area of the code so that split_DEF can
// generate DefinitionSpillCopy for the derived pointer.
uint default_int_pressure_threshold = _NO_SPECIAL_REG32_mask.Size() - 1;
if (!PreserveFramePointer) {
// When PreserveFramePointer is off, frame pointer is allocatable,
// but different from other SOC registers, it is excluded from
// fatproj's mask because its save type is No-Save. Decrease 1 to
// ensure high pressure at fatproj when PreserveFramePointer is off.
// See check_pressure_at_fatproj().
default_int_pressure_threshold--;
}
return (INTPRESSURE == -1) ? default_int_pressure_threshold : INTPRESSURE;
}
uint Matcher::float_pressure_limit()
{
// _FLOAT_REG_mask is generated by adlc from the float_reg register class.
return (FLOATPRESSURE == -1) ? _FLOAT_REG_mask.Size() : FLOATPRESSURE;
}
bool Matcher::use_asm_for_ldiv_by_con(jlong divisor) {
return false;
}
RegMask Matcher::divI_proj_mask() {
ShouldNotReachHere();
return RegMask();
}
// Register for MODI projection of divmodI.
RegMask Matcher::modI_proj_mask() {
ShouldNotReachHere();
return RegMask();
}
// Register for DIVL projection of divmodL.
RegMask Matcher::divL_proj_mask() {
ShouldNotReachHere();
return RegMask();
}
// Register for MODL projection of divmodL.
RegMask Matcher::modL_proj_mask() {
ShouldNotReachHere();
return RegMask();
}
const RegMask Matcher::method_handle_invoke_SP_save_mask() {
return FP_REG_mask();
}
bool size_fits_all_mem_uses(AddPNode* addp, int shift) {
assert_cond(addp != NULL);
for (DUIterator_Fast imax, i = addp->fast_outs(imax); i < imax; i++) {
Node* u = addp->fast_out(i);
if (u != NULL && u->is_Mem()) {
int opsize = u->as_Mem()->memory_size();
assert(opsize > 0, "unexpected memory operand size");
if (u->as_Mem()->memory_size() != (1 << shift)) {
return false;
}
}
}
return true;
}
// Should the Matcher clone input 'm' of node 'n'?
bool Matcher::pd_clone_node(Node* n, Node* m, Matcher::MStack& mstack) {
assert_cond(m != NULL);
if (is_vshift_con_pattern(n, m)) { // ShiftV src (ShiftCntV con)
mstack.push(m, Visit); // m = ShiftCntV
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::pd_clone_address_expressions(AddPNode* m, Matcher::MStack& mstack, VectorSet& address_visited) {
return clone_base_plus_offset_address(m, mstack, address_visited);
}
%}
//----------ENCODING BLOCK-----------------------------------------------------
// This block specifies the encoding classes used by the compiler to
// output byte streams. Encoding classes are parameterized macros
// used by Machine Instruction Nodes in order to generate the bit
// encoding of the instruction. Operands specify their base encoding
// interface with the interface keyword. There are currently
// supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, &
// COND_INTER. REG_INTER causes an operand to generate a function
// which returns its register number when queried. CONST_INTER causes
// an operand to generate a function which returns the value of the
// constant when queried. MEMORY_INTER causes an operand to generate
// four functions which return the Base Register, the Index Register,
// the Scale Value, and the Offset Value of the operand when queried.
// COND_INTER causes an operand to generate six functions which return
// the encoding code (ie - encoding bits for the instruction)
// associated with each basic boolean condition for a conditional
// instruction.
//
// Instructions specify two basic values for encoding. Again, a
// function is available to check if the constant displacement is an
// oop. They use the ins_encode keyword to specify their encoding
// classes (which must be a sequence of enc_class names, and their
// parameters, specified in the encoding block), and they use the
// opcode keyword to specify, in order, their primary, secondary, and
// tertiary opcode. Only the opcode sections which a particular
// instruction needs for encoding need to be specified.
encode %{
// BEGIN Non-volatile memory access
enc_class riscv_enc_li_imm(iRegIorL dst, immIorL src) %{
C2_MacroAssembler _masm(&cbuf);
int64_t con = (int64_t)$src$$constant;
Register dst_reg = as_Register($dst$$reg);
__ mv(dst_reg, con);
%}
enc_class riscv_enc_mov_p(iRegP dst, immP src) %{
C2_MacroAssembler _masm(&cbuf);
Register dst_reg = as_Register($dst$$reg);
address con = (address)$src$$constant;
if (con == NULL || con == (address)1) {
ShouldNotReachHere();
} else {
relocInfo::relocType rtype = $src->constant_reloc();
if (rtype == relocInfo::oop_type) {
__ movoop(dst_reg, (jobject)con);
} else if (rtype == relocInfo::metadata_type) {
__ mov_metadata(dst_reg, (Metadata*)con);
} else {
assert(rtype == relocInfo::none, "unexpected reloc type");
__ mv(dst_reg, $src$$constant);
}
}
%}
enc_class riscv_enc_mov_p1(iRegP dst) %{
C2_MacroAssembler _masm(&cbuf);
Register dst_reg = as_Register($dst$$reg);
__ mv(dst_reg, 1);
%}
enc_class riscv_enc_mov_byte_map_base(iRegP dst) %{
C2_MacroAssembler _masm(&cbuf);
__ load_byte_map_base($dst$$Register);
%}
enc_class riscv_enc_mov_n(iRegN dst, immN src) %{
C2_MacroAssembler _masm(&cbuf);
Register dst_reg = as_Register($dst$$reg);
address con = (address)$src$$constant;
if (con == NULL) {
ShouldNotReachHere();
} else {
relocInfo::relocType rtype = $src->constant_reloc();
assert(rtype == relocInfo::oop_type, "unexpected reloc type");
__ set_narrow_oop(dst_reg, (jobject)con);
}
%}
enc_class riscv_enc_mov_zero(iRegNorP dst) %{
C2_MacroAssembler _masm(&cbuf);
Register dst_reg = as_Register($dst$$reg);
__ mv(dst_reg, zr);
%}
enc_class riscv_enc_mov_nk(iRegN dst, immNKlass src) %{
C2_MacroAssembler _masm(&cbuf);
Register dst_reg = as_Register($dst$$reg);
address con = (address)$src$$constant;
if (con == NULL) {
ShouldNotReachHere();
} else {
relocInfo::relocType rtype = $src->constant_reloc();
assert(rtype == relocInfo::metadata_type, "unexpected reloc type");
__ set_narrow_klass(dst_reg, (Klass *)con);
}
%}
enc_class riscv_enc_cmpxchgw(iRegINoSp res, memory mem, iRegI oldval, iRegI newval) %{
C2_MacroAssembler _masm(&cbuf);
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int32,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register,
/*result as bool*/ true);
%}
enc_class riscv_enc_cmpxchgn(iRegINoSp res, memory mem, iRegI oldval, iRegI newval) %{
C2_MacroAssembler _masm(&cbuf);
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::uint32,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register,
/*result as bool*/ true);
%}
enc_class riscv_enc_cmpxchg(iRegINoSp res, memory mem, iRegL oldval, iRegL newval) %{
C2_MacroAssembler _masm(&cbuf);
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register,
/*result as bool*/ true);
%}
enc_class riscv_enc_cmpxchgw_acq(iRegINoSp res, memory mem, iRegI oldval, iRegI newval) %{
C2_MacroAssembler _masm(&cbuf);
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int32,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register,
/*result as bool*/ true);
%}
enc_class riscv_enc_cmpxchgn_acq(iRegINoSp res, memory mem, iRegI oldval, iRegI newval) %{
C2_MacroAssembler _masm(&cbuf);
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::uint32,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register,
/*result as bool*/ true);
%}
enc_class riscv_enc_cmpxchg_acq(iRegINoSp res, memory mem, iRegL oldval, iRegL newval) %{
C2_MacroAssembler _masm(&cbuf);
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register,
/*result as bool*/ true);
%}
// compare and branch instruction encodings
enc_class riscv_enc_j(label lbl) %{
C2_MacroAssembler _masm(&cbuf);
Label* L = $lbl$$label;
__ j(*L);
%}
enc_class riscv_enc_far_cmpULtGe_imm0_branch(cmpOpULtGe cmp, iRegIorL op1, label lbl) %{
C2_MacroAssembler _masm(&cbuf);
Label* L = $lbl$$label;
switch ($cmp$$cmpcode) {
case(BoolTest::ge):
__ j(*L);
break;
case(BoolTest::lt):
break;
default:
Unimplemented();
}
%}
// call instruction encodings
enc_class riscv_enc_partial_subtype_check(iRegP sub, iRegP super, iRegP temp, iRegP result) %{
Register sub_reg = as_Register($sub$$reg);
Register super_reg = as_Register($super$$reg);
Register temp_reg = as_Register($temp$$reg);
Register result_reg = as_Register($result$$reg);
Register cr_reg = t1;
Label miss;
Label done;
C2_MacroAssembler _masm(&cbuf);
__ check_klass_subtype_slow_path(sub_reg, super_reg, temp_reg, result_reg,
NULL, &miss);
if ($primary) {
__ mv(result_reg, zr);
} else {
__ mv(cr_reg, zr);
__ j(done);
}
__ bind(miss);
if (!$primary) {
__ mv(cr_reg, 1);
}
__ bind(done);
%}
enc_class riscv_enc_java_static_call(method meth) %{
C2_MacroAssembler _masm(&cbuf);
Assembler::IncompressibleRegion ir(&_masm); // Fixed length: see ret_addr_offset
address addr = (address)$meth$$method;
address call = NULL;
assert_cond(addr != NULL);
if (!_method) {
// A call to a runtime wrapper, e.g. new, new_typeArray_Java, uncommon_trap.
call = __ trampoline_call(Address(addr, relocInfo::runtime_call_type));
if (call == NULL) {
ciEnv::current()->record_failure("CodeCache is full");
return;
}
} else {
int method_index = resolved_method_index(cbuf);
RelocationHolder rspec = _optimized_virtual ? opt_virtual_call_Relocation::spec(method_index)
: static_call_Relocation::spec(method_index);
call = __ trampoline_call(Address(addr, rspec));
if (call == NULL) {
ciEnv::current()->record_failure("CodeCache is full");
return;
}
if (CodeBuffer::supports_shared_stubs() && _method->can_be_statically_bound()) {
// Calls of the same statically bound method can share
// a stub to the interpreter.
cbuf.shared_stub_to_interp_for(_method, call - cbuf.insts_begin());
} else {
// Emit stub for static call
address stub = CompiledStaticCall::emit_to_interp_stub(cbuf, call);
if (stub == NULL) {
ciEnv::current()->record_failure("CodeCache is full");
return;
}
}
}
__ post_call_nop();
%}
enc_class riscv_enc_java_dynamic_call(method meth) %{
C2_MacroAssembler _masm(&cbuf);
Assembler::IncompressibleRegion ir(&_masm); // Fixed length: see ret_addr_offset
int method_index = resolved_method_index(cbuf);
address call = __ ic_call((address)$meth$$method, method_index);
if (call == NULL) {
ciEnv::current()->record_failure("CodeCache is full");
return;
}
__ post_call_nop();
%}
enc_class riscv_enc_call_epilog() %{
C2_MacroAssembler _masm(&cbuf);
if (VerifyStackAtCalls) {
// Check that stack depth is unchanged: find majik cookie on stack
__ call_Unimplemented();
}
%}
enc_class riscv_enc_java_to_runtime(method meth) %{
C2_MacroAssembler _masm(&cbuf);
Assembler::IncompressibleRegion ir(&_masm); // Fixed length: see ret_addr_offset
// some calls to generated routines (arraycopy code) are scheduled
// by C2 as runtime calls. if so we can call them using a jr (they
// will be in a reachable segment) otherwise we have to use a jalr
// which loads the absolute address into a register.
address entry = (address)$meth$$method;
CodeBlob *cb = CodeCache::find_blob(entry);
if (cb != NULL) {
address call = __ trampoline_call(Address(entry, relocInfo::runtime_call_type));
if (call == NULL) {
ciEnv::current()->record_failure("CodeCache is full");
return;
}
__ post_call_nop();
} else {
Label retaddr;
__ la(t1, retaddr);
__ la(t0, RuntimeAddress(entry));
// Leave a breadcrumb for JavaFrameAnchor::capture_last_Java_pc()
__ addi(sp, sp, -2 * wordSize);
__ sd(t1, Address(sp, wordSize));
__ jalr(t0);
__ bind(retaddr);
__ post_call_nop();
__ addi(sp, sp, 2 * wordSize);
}
%}
// using the cr register as the bool result: 0 for success; others failed.
enc_class riscv_enc_fast_lock(iRegP object, iRegP box, iRegPNoSp tmp1, iRegPNoSp tmp2) %{
C2_MacroAssembler _masm(&cbuf);
Register flag = t1;
Register oop = as_Register($object$$reg);
Register box = as_Register($box$$reg);
Register disp_hdr = as_Register($tmp1$$reg);
Register tmp = as_Register($tmp2$$reg);
Label cont;
Label object_has_monitor;
Label no_count;
assert_different_registers(oop, box, tmp, disp_hdr, t0);
// Load markWord from object into displaced_header.
__ ld(disp_hdr, Address(oop, oopDesc::mark_offset_in_bytes()));
if (DiagnoseSyncOnValueBasedClasses != 0) {
__ load_klass(flag, oop);
__ lwu(flag, Address(flag, Klass::access_flags_offset()));
__ andi(flag, flag, JVM_ACC_IS_VALUE_BASED_CLASS, tmp /* tmp */);
__ bnez(flag, cont, true /* is_far */);
}
// Check for existing monitor
__ andi(t0, disp_hdr, markWord::monitor_value);
__ bnez(t0, object_has_monitor);
if (!UseHeavyMonitors) {
// Set tmp to be (markWord of object | UNLOCK_VALUE).
__ ori(tmp, disp_hdr, markWord::unlocked_value);
// Initialize the box. (Must happen before we update the object mark!)
__ sd(tmp, Address(box, BasicLock::displaced_header_offset_in_bytes()));
// Compare object markWord with an unlocked value (tmp) and if
// equal exchange the stack address of our box with object markWord.
// On failure disp_hdr contains the possibly locked markWord.
__ cmpxchg(/*memory address*/oop, /*expected value*/tmp, /*new value*/box, Assembler::int64, Assembler::aq,
Assembler::rl, /*result*/disp_hdr);
__ mv(flag, zr);
__ beq(disp_hdr, tmp, cont); // prepare zero flag and goto cont if we won the cas
assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0");
// If the compare-and-exchange succeeded, then we found an unlocked
// object, will have now locked it will continue at label cont
// We did not see an unlocked object so try the fast recursive case.
// Check if the owner is self by comparing the value in the
// markWord of object (disp_hdr) with the stack pointer.
__ sub(disp_hdr, disp_hdr, sp);
__ mv(tmp, (intptr_t) (~(os::vm_page_size()-1) | (uintptr_t)markWord::lock_mask_in_place));
// If (mark & lock_mask) == 0 and mark - sp < page_size, we are stack-locking and goto cont,
// hence we can store 0 as the displaced header in the box, which indicates that it is a
// recursive lock.
__ andr(tmp/*==0?*/, disp_hdr, tmp);
__ sd(tmp/*==0, perhaps*/, Address(box, BasicLock::displaced_header_offset_in_bytes()));
__ mv(flag, tmp); // we can use the value of tmp as the result here
} else {
__ mv(flag, 1); // Set non-zero flag to indicate 'failure' -> take slow-path
}
__ j(cont);
// Handle existing monitor.
__ bind(object_has_monitor);
// The object's monitor m is unlocked iff m->owner == NULL,
// otherwise m->owner may contain a thread or a stack address.
//
// Try to CAS m->owner from NULL to current thread.
__ add(tmp, disp_hdr, (ObjectMonitor::owner_offset_in_bytes() - markWord::monitor_value));
__ cmpxchg(/*memory address*/tmp, /*expected value*/zr, /*new value*/xthread, Assembler::int64, Assembler::aq,
Assembler::rl, /*result*/flag); // cas succeeds if flag == zr(expected)
// Store a non-null value into the box to avoid looking like a re-entrant
// lock. The fast-path monitor unlock code checks for
// markWord::monitor_value so use markWord::unused_mark which has the
// relevant bit set, and also matches ObjectSynchronizer::slow_enter.
__ mv(tmp, (address)markWord::unused_mark().value());
__ sd(tmp, Address(box, BasicLock::displaced_header_offset_in_bytes()));
__ beqz(flag, cont); // CAS success means locking succeeded
__ bne(flag, xthread, cont); // Check for recursive locking
// Recursive lock case
__ mv(flag, zr);
__ increment(Address(disp_hdr, ObjectMonitor::recursions_offset_in_bytes() - markWord::monitor_value), 1, t0, tmp);
__ bind(cont);
__ bnez(flag, no_count);
__ increment(Address(xthread, JavaThread::held_monitor_count_offset()), 1, t0, tmp);
__ bind(no_count);
%}
// using cr flag to indicate the fast_unlock result: 0 for success; others failed.
enc_class riscv_enc_fast_unlock(iRegP object, iRegP box, iRegPNoSp tmp1, iRegPNoSp tmp2) %{
C2_MacroAssembler _masm(&cbuf);
Register flag = t1;
Register oop = as_Register($object$$reg);
Register box = as_Register($box$$reg);
Register disp_hdr = as_Register($tmp1$$reg);
Register tmp = as_Register($tmp2$$reg);
Label cont;
Label object_has_monitor;
Label no_count;
assert_different_registers(oop, box, tmp, disp_hdr, flag);
if (!UseHeavyMonitors) {
// Find the lock address and load the displaced header from the stack.
__ ld(disp_hdr, Address(box, BasicLock::displaced_header_offset_in_bytes()));
// If the displaced header is 0, we have a recursive unlock.
__ mv(flag, disp_hdr);
__ beqz(disp_hdr, cont);
}
// Handle existing monitor.
__ ld(tmp, Address(oop, oopDesc::mark_offset_in_bytes()));
__ andi(t0, tmp, markWord::monitor_value);
__ bnez(t0, object_has_monitor);
if (!UseHeavyMonitors) {
// Check if it is still a light weight lock, this is true if we
// see the stack address of the basicLock in the markWord of the
// object.
__ cmpxchg(/*memory address*/oop, /*expected value*/box, /*new value*/disp_hdr, Assembler::int64, Assembler::relaxed,
Assembler::rl, /*result*/tmp);
__ xorr(flag, box, tmp); // box == tmp if cas succeeds
} else {
__ mv(flag, 1); // Set non-zero flag to indicate 'failure' -> take slow path
}
__ j(cont);
assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0");
// Handle existing monitor.
__ bind(object_has_monitor);
STATIC_ASSERT(markWord::monitor_value <= INT_MAX);
__ add(tmp, tmp, -(int)markWord::monitor_value); // monitor
__ ld(disp_hdr, Address(tmp, ObjectMonitor::recursions_offset_in_bytes()));
Label notRecursive;
__ beqz(disp_hdr, notRecursive); // Will be 0 if not recursive.
// Recursive lock
__ addi(disp_hdr, disp_hdr, -1);
__ sd(disp_hdr, Address(tmp, ObjectMonitor::recursions_offset_in_bytes()));
__ mv(flag, zr);
__ j(cont);
__ bind(notRecursive);
__ ld(flag, Address(tmp, ObjectMonitor::EntryList_offset_in_bytes()));
__ ld(disp_hdr, Address(tmp, ObjectMonitor::cxq_offset_in_bytes()));
__ orr(flag, flag, disp_hdr); // Will be 0 if both are 0.
__ bnez(flag, cont);
// need a release store here
__ la(tmp, Address(tmp, ObjectMonitor::owner_offset_in_bytes()));
__ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
__ sd(zr, Address(tmp)); // set unowned
__ bind(cont);
__ bnez(flag, no_count);
__ decrement(Address(xthread, JavaThread::held_monitor_count_offset()), 1, t0, tmp);
__ bind(no_count);
%}
// arithmetic encodings
enc_class riscv_enc_divw(iRegI dst, iRegI src1, iRegI src2) %{
C2_MacroAssembler _masm(&cbuf);
Register dst_reg = as_Register($dst$$reg);
Register src1_reg = as_Register($src1$$reg);
Register src2_reg = as_Register($src2$$reg);
__ corrected_idivl(dst_reg, src1_reg, src2_reg, false);
%}
enc_class riscv_enc_div(iRegI dst, iRegI src1, iRegI src2) %{
C2_MacroAssembler _masm(&cbuf);
Register dst_reg = as_Register($dst$$reg);
Register src1_reg = as_Register($src1$$reg);
Register src2_reg = as_Register($src2$$reg);
__ corrected_idivq(dst_reg, src1_reg, src2_reg, false);
%}
enc_class riscv_enc_modw(iRegI dst, iRegI src1, iRegI src2) %{
C2_MacroAssembler _masm(&cbuf);
Register dst_reg = as_Register($dst$$reg);
Register src1_reg = as_Register($src1$$reg);
Register src2_reg = as_Register($src2$$reg);
__ corrected_idivl(dst_reg, src1_reg, src2_reg, true);
%}
enc_class riscv_enc_mod(iRegI dst, iRegI src1, iRegI src2) %{
C2_MacroAssembler _masm(&cbuf);
Register dst_reg = as_Register($dst$$reg);
Register src1_reg = as_Register($src1$$reg);
Register src2_reg = as_Register($src2$$reg);
__ corrected_idivq(dst_reg, src1_reg, src2_reg, true);
%}
enc_class riscv_enc_tail_call(iRegP jump_target) %{
C2_MacroAssembler _masm(&cbuf);
Register target_reg = as_Register($jump_target$$reg);
__ jr(target_reg);
%}
enc_class riscv_enc_tail_jmp(iRegP jump_target) %{
C2_MacroAssembler _masm(&cbuf);
Register target_reg = as_Register($jump_target$$reg);
// exception oop should be in x10
// ret addr has been popped into ra
// callee expects it in x13
__ mv(x13, ra);
__ jr(target_reg);
%}
enc_class riscv_enc_rethrow() %{
C2_MacroAssembler _masm(&cbuf);
__ far_jump(RuntimeAddress(OptoRuntime::rethrow_stub()));
%}
enc_class riscv_enc_ret() %{
C2_MacroAssembler _masm(&cbuf);
__ ret();
%}
%}
//----------FRAME--------------------------------------------------------------
// Definition of frame structure and management information.
//
// S T A C K L A Y O U T Allocators stack-slot number
// | (to get allocators register number
// G Owned by | | v add OptoReg::stack0())
// r CALLER | |
// o | +--------+ pad to even-align allocators stack-slot
// w V | pad0 | numbers; owned by CALLER
// t -----------+--------+----> Matcher::_in_arg_limit, unaligned
// h ^ | in | 5
// | | args | 4 Holes in incoming args owned by SELF
// | | | | 3
// | | +--------+
// V | | old out| Empty on Intel, window on Sparc
// | old |preserve| Must be even aligned.
// | SP-+--------+----> Matcher::_old_SP, even aligned
// | | in | 3 area for Intel ret address
// Owned by |preserve| Empty on Sparc.
// SELF +--------+
// | | pad2 | 2 pad to align old SP
// | +--------+ 1
// | | locks | 0
// | +--------+----> OptoReg::stack0(), even aligned
// | | pad1 | 11 pad to align new SP
// | +--------+
// | | | 10
// | | spills | 9 spills
// V | | 8 (pad0 slot for callee)
// -----------+--------+----> Matcher::_out_arg_limit, unaligned
// ^ | out | 7
// | | args | 6 Holes in outgoing args owned by CALLEE
// Owned by +--------+
// CALLEE | new out| 6 Empty on Intel, window on Sparc
// | new |preserve| Must be even-aligned.
// | SP-+--------+----> Matcher::_new_SP, even aligned
// | | |
//
// Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is
// known from SELF's arguments and the Java calling convention.
// Region 6-7 is determined per call site.
// Note 2: If the calling convention leaves holes in the incoming argument
// area, those holes are owned by SELF. Holes in the outgoing area
// are owned by the CALLEE. Holes should not be necessary in the
// incoming area, as the Java calling convention is completely under
// the control of the AD file. Doubles can be sorted and packed to
// avoid holes. Holes in the outgoing arguments may be necessary for
// varargs C calling conventions.
// Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is
// even aligned with pad0 as needed.
// Region 6 is even aligned. Region 6-7 is NOT even aligned;
// (the latter is true on Intel but is it false on RISCV?)
// region 6-11 is even aligned; it may be padded out more so that
// the region from SP to FP meets the minimum stack alignment.
// Note 4: For I2C adapters, the incoming FP may not meet the minimum stack
// alignment. Region 11, pad1, may be dynamically extended so that
// SP meets the minimum alignment.
frame %{
// These three registers define part of the calling convention
// between compiled code and the interpreter.
// Inline Cache Register or methodOop for I2C.
inline_cache_reg(R31);
// Optional: name the operand used by cisc-spilling to access [stack_pointer + offset]
cisc_spilling_operand_name(indOffset);
// Number of stack slots consumed by locking an object
// generate Compile::sync_stack_slots
// VMRegImpl::slots_per_word = wordSize / stack_slot_size = 8 / 4 = 2
sync_stack_slots(1 * VMRegImpl::slots_per_word);
// Compiled code's Frame Pointer
frame_pointer(R2);
// Interpreter stores its frame pointer in a register which is
// stored to the stack by I2CAdaptors.
// I2CAdaptors convert from interpreted java to compiled java.
interpreter_frame_pointer(R8);
// Stack alignment requirement
stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes)
// Number of outgoing stack slots killed above the out_preserve_stack_slots
// for calls to C. Supports the var-args backing area for register parms.
varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes / BytesPerInt);
// The after-PROLOG location of the return address. Location of
// return address specifies a type (REG or STACK) and a number
// representing the register number (i.e. - use a register name) or
// stack slot.
// Ret Addr is on stack in slot 0 if no locks or verification or alignment.
// Otherwise, it is above the locks and verification slot and alignment word
// TODO this may well be correct but need to check why that - 2 is there
// ppc port uses 0 but we definitely need to allow for fixed_slots
// which folds in the space used for monitors
return_addr(STACK - 2 +
align_up((Compile::current()->in_preserve_stack_slots() +
Compile::current()->fixed_slots()),
stack_alignment_in_slots()));
// Location of compiled Java return values. Same as C for now.
return_value
%{
assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL,
"only return normal values");
static const int lo[Op_RegL + 1] = { // enum name
0, // Op_Node
0, // Op_Set
R10_num, // Op_RegN
R10_num, // Op_RegI
R10_num, // Op_RegP
F10_num, // Op_RegF
F10_num, // Op_RegD
R10_num // Op_RegL
};
static const int hi[Op_RegL + 1] = { // enum name
0, // Op_Node
0, // Op_Set
OptoReg::Bad, // Op_RegN
OptoReg::Bad, // Op_RegI
R10_H_num, // Op_RegP
OptoReg::Bad, // Op_RegF
F10_H_num, // Op_RegD
R10_H_num // Op_RegL
};
return OptoRegPair(hi[ideal_reg], lo[ideal_reg]);
%}
%}
//----------ATTRIBUTES---------------------------------------------------------
//----------Operand Attributes-------------------------------------------------
op_attrib op_cost(1); // Required cost attribute
//----------Instruction Attributes---------------------------------------------
ins_attrib ins_cost(DEFAULT_COST); // Required cost attribute
ins_attrib ins_size(32); // Required size attribute (in bits)
ins_attrib ins_short_branch(0); // Required flag: is this instruction
// a non-matching short branch variant
// of some long branch?
ins_attrib ins_alignment(4); // Required alignment attribute (must
// be a power of 2) specifies the
// alignment that some part of the
// instruction (not necessarily the
// start) requires. If > 1, a
// compute_padding() function must be
// provided for the instruction
//----------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.
//----------Simple Operands----------------------------------------------------
// Integer operands 32 bit
// 32 bit immediate
operand immI()
%{
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 32 bit zero
operand immI0()
%{
predicate(n->get_int() == 0);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 32 bit unit increment
operand immI_1()
%{
predicate(n->get_int() == 1);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 32 bit unit decrement
operand immI_M1()
%{
predicate(n->get_int() == -1);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Unsigned Integer Immediate: 6-bit int, greater than 32
operand uimmI6_ge32() %{
predicate(((unsigned int)(n->get_int()) < 64) && (n->get_int() >= 32));
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
operand immI_le_4()
%{
predicate(n->get_int() <= 4);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
operand immI_16()
%{
predicate(n->get_int() == 16);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
operand immI_24()
%{
predicate(n->get_int() == 24);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
operand immI_31()
%{
predicate(n->get_int() == 31);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
operand immI_63()
%{
predicate(n->get_int() == 63);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 32 bit integer valid for add immediate
operand immIAdd()
%{
predicate(Assembler::operand_valid_for_add_immediate((int64_t)n->get_int()));
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 32 bit integer valid for sub immediate
operand immISub()
%{
predicate(Assembler::operand_valid_for_add_immediate(-(int64_t)n->get_int()));
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 5 bit signed value.
operand immI5()
%{
predicate(n->get_int() <= 15 && n->get_int() >= -16);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 5 bit signed value (simm5)
operand immL5()
%{
predicate(n->get_long() <= 15 && n->get_long() >= -16);
match(ConL);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Integer operands 64 bit
// 64 bit immediate
operand immL()
%{
match(ConL);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 64 bit zero
operand immL0()
%{
predicate(n->get_long() == 0);
match(ConL);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Pointer operands
// Pointer Immediate
operand immP()
%{
match(ConP);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// NULL Pointer Immediate
operand immP0()
%{
predicate(n->get_ptr() == 0);
match(ConP);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Pointer Immediate One
// this is used in object initialization (initial object header)
operand immP_1()
%{
predicate(n->get_ptr() == 1);
match(ConP);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Card Table Byte Map Base
operand immByteMapBase()
%{
// Get base of card map
predicate(BarrierSet::barrier_set()->is_a(BarrierSet::CardTableBarrierSet) &&
(CardTable::CardValue*)n->get_ptr() ==
((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base());
match(ConP);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Int Immediate: low 16-bit mask
operand immI_16bits()
%{
predicate(n->get_int() == 0xFFFF);
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
operand immIpowerOf2() %{
predicate(is_power_of_2((juint)(n->get_int())));
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Long Immediate: low 32-bit mask
operand immL_32bits()
%{
predicate(n->get_long() == 0xFFFFFFFFL);
match(ConL);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 64 bit unit decrement
operand immL_M1()
%{
predicate(n->get_long() == -1);
match(ConL);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 32 bit offset of pc in thread anchor
operand immL_pc_off()
%{
predicate(n->get_long() == in_bytes(JavaThread::frame_anchor_offset()) +
in_bytes(JavaFrameAnchor::last_Java_pc_offset()));
match(ConL);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 64 bit integer valid for add immediate
operand immLAdd()
%{
predicate(Assembler::operand_valid_for_add_immediate(n->get_long()));
match(ConL);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// 64 bit integer valid for sub immediate
operand immLSub()
%{
predicate(Assembler::operand_valid_for_add_immediate(-(n->get_long())));
match(ConL);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Narrow pointer operands
// Narrow Pointer Immediate
operand immN()
%{
match(ConN);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Narrow NULL Pointer Immediate
operand immN0()
%{
predicate(n->get_narrowcon() == 0);
match(ConN);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
operand immNKlass()
%{
match(ConNKlass);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Float and Double operands
// Double Immediate
operand immD()
%{
match(ConD);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Double Immediate: +0.0d
operand immD0()
%{
predicate(jlong_cast(n->getd()) == 0);
match(ConD);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Float Immediate
operand immF()
%{
match(ConF);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Float Immediate: +0.0f.
operand immF0()
%{
predicate(jint_cast(n->getf()) == 0);
match(ConF);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
operand immIOffset()
%{
predicate(is_imm_in_range(n->get_int(), 12, 0));
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
operand immLOffset()
%{
predicate(is_imm_in_range(n->get_long(), 12, 0));
match(ConL);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Scale values
operand immIScale()
%{
predicate(1 <= n->get_int() && (n->get_int() <= 3));
match(ConI);
op_cost(0);
format %{ %}
interface(CONST_INTER);
%}
// Integer 32 bit Register Operands
operand iRegI()
%{
constraint(ALLOC_IN_RC(any_reg32));
match(RegI);
match(iRegINoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Integer 32 bit Register not Special
operand iRegINoSp()
%{
constraint(ALLOC_IN_RC(no_special_reg32));
match(RegI);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Register R10 only
operand iRegI_R10()
%{
constraint(ALLOC_IN_RC(int_r10_reg));
match(RegI);
match(iRegINoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Register R12 only
operand iRegI_R12()
%{
constraint(ALLOC_IN_RC(int_r12_reg));
match(RegI);
match(iRegINoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Register R13 only
operand iRegI_R13()
%{
constraint(ALLOC_IN_RC(int_r13_reg));
match(RegI);
match(iRegINoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Register R14 only
operand iRegI_R14()
%{
constraint(ALLOC_IN_RC(int_r14_reg));
match(RegI);
match(iRegINoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Integer 64 bit Register Operands
operand iRegL()
%{
constraint(ALLOC_IN_RC(any_reg));
match(RegL);
match(iRegLNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Integer 64 bit Register not Special
operand iRegLNoSp()
%{
constraint(ALLOC_IN_RC(no_special_reg));
match(RegL);
match(iRegL_R10);
format %{ %}
interface(REG_INTER);
%}
// Long 64 bit Register R28 only
operand iRegL_R28()
%{
constraint(ALLOC_IN_RC(r28_reg));
match(RegL);
match(iRegLNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Long 64 bit Register R29 only
operand iRegL_R29()
%{
constraint(ALLOC_IN_RC(r29_reg));
match(RegL);
match(iRegLNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Long 64 bit Register R30 only
operand iRegL_R30()
%{
constraint(ALLOC_IN_RC(r30_reg));
match(RegL);
match(iRegLNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Pointer Register Operands
// Pointer Register
operand iRegP()
%{
constraint(ALLOC_IN_RC(ptr_reg));
match(RegP);
match(iRegPNoSp);
match(iRegP_R10);
match(iRegP_R15);
match(javaThread_RegP);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Pointer 64 bit Register not Special
operand iRegPNoSp()
%{
constraint(ALLOC_IN_RC(no_special_ptr_reg));
match(RegP);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand iRegP_R10()
%{
constraint(ALLOC_IN_RC(r10_reg));
match(RegP);
// match(iRegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Pointer 64 bit Register R11 only
operand iRegP_R11()
%{
constraint(ALLOC_IN_RC(r11_reg));
match(RegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand iRegP_R12()
%{
constraint(ALLOC_IN_RC(r12_reg));
match(RegP);
// match(iRegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Pointer 64 bit Register R13 only
operand iRegP_R13()
%{
constraint(ALLOC_IN_RC(r13_reg));
match(RegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand iRegP_R14()
%{
constraint(ALLOC_IN_RC(r14_reg));
match(RegP);
// match(iRegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand iRegP_R15()
%{
constraint(ALLOC_IN_RC(r15_reg));
match(RegP);
// match(iRegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand iRegP_R16()
%{
constraint(ALLOC_IN_RC(r16_reg));
match(RegP);
// match(iRegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Pointer 64 bit Register R28 only
operand iRegP_R28()
%{
constraint(ALLOC_IN_RC(r28_reg));
match(RegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Pointer 64 bit Register R30 only
operand iRegP_R30()
%{
constraint(ALLOC_IN_RC(r30_reg));
match(RegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Pointer 64 bit Register R31 only
operand iRegP_R31()
%{
constraint(ALLOC_IN_RC(r31_reg));
match(RegP);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Pointer Register Operands
// Narrow Pointer Register
operand iRegN()
%{
constraint(ALLOC_IN_RC(any_reg32));
match(RegN);
match(iRegNNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Integer 64 bit Register not Special
operand iRegNNoSp()
%{
constraint(ALLOC_IN_RC(no_special_reg32));
match(RegN);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// heap base register -- used for encoding immN0
operand iRegIHeapbase()
%{
constraint(ALLOC_IN_RC(heapbase_reg));
match(RegI);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Long 64 bit Register R10 only
operand iRegL_R10()
%{
constraint(ALLOC_IN_RC(r10_reg));
match(RegL);
match(iRegLNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Float Register
// Float register operands
operand fRegF()
%{
constraint(ALLOC_IN_RC(float_reg));
match(RegF);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Double Register
// Double register operands
operand fRegD()
%{
constraint(ALLOC_IN_RC(double_reg));
match(RegD);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Generic vector class. This will be used for
// all vector operands.
operand vReg()
%{
constraint(ALLOC_IN_RC(vectora_reg));
match(VecA);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand vReg_V1()
%{
constraint(ALLOC_IN_RC(v1_reg));
match(VecA);
match(vReg);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand vReg_V2()
%{
constraint(ALLOC_IN_RC(v2_reg));
match(VecA);
match(vReg);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand vReg_V3()
%{
constraint(ALLOC_IN_RC(v3_reg));
match(VecA);
match(vReg);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand vReg_V4()
%{
constraint(ALLOC_IN_RC(v4_reg));
match(VecA);
match(vReg);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
operand vReg_V5()
%{
constraint(ALLOC_IN_RC(v5_reg));
match(VecA);
match(vReg);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
// Java Thread Register
operand javaThread_RegP(iRegP reg)
%{
constraint(ALLOC_IN_RC(java_thread_reg)); // java_thread_reg
match(reg);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
//----------Memory Operands----------------------------------------------------
// RISCV has only base_plus_offset and literal address mode, so no need to use
// index and scale. Here set index as 0xffffffff and scale as 0x0.
operand indirect(iRegP reg)
%{
constraint(ALLOC_IN_RC(ptr_reg));
match(reg);
op_cost(0);
format %{ "[$reg]" %}
interface(MEMORY_INTER) %{
base($reg);
index(0xffffffff);
scale(0x0);
disp(0x0);
%}
%}
operand indOffI(iRegP reg, immIOffset off)
%{
constraint(ALLOC_IN_RC(ptr_reg));
match(AddP reg off);
op_cost(0);
format %{ "[$reg, $off]" %}
interface(MEMORY_INTER) %{
base($reg);
index(0xffffffff);
scale(0x0);
disp($off);
%}
%}
operand indOffL(iRegP reg, immLOffset off)
%{
constraint(ALLOC_IN_RC(ptr_reg));
match(AddP reg off);
op_cost(0);
format %{ "[$reg, $off]" %}
interface(MEMORY_INTER) %{
base($reg);
index(0xffffffff);
scale(0x0);
disp($off);
%}
%}
operand indirectN(iRegN reg)
%{
predicate(CompressedOops::shift() == 0);
constraint(ALLOC_IN_RC(ptr_reg));
match(DecodeN reg);
op_cost(0);
format %{ "[$reg]\t# narrow" %}
interface(MEMORY_INTER) %{
base($reg);
index(0xffffffff);
scale(0x0);
disp(0x0);
%}
%}
operand indOffIN(iRegN reg, immIOffset off)
%{
predicate(CompressedOops::shift() == 0);
constraint(ALLOC_IN_RC(ptr_reg));
match(AddP (DecodeN reg) off);
op_cost(0);
format %{ "[$reg, $off]\t# narrow" %}
interface(MEMORY_INTER) %{
base($reg);
index(0xffffffff);
scale(0x0);
disp($off);
%}
%}
operand indOffLN(iRegN reg, immLOffset off)
%{
predicate(CompressedOops::shift() == 0);
constraint(ALLOC_IN_RC(ptr_reg));
match(AddP (DecodeN reg) off);
op_cost(0);
format %{ "[$reg, $off]\t# narrow" %}
interface(MEMORY_INTER) %{
base($reg);
index(0xffffffff);
scale(0x0);
disp($off);
%}
%}
// RISCV opto stubs need to write to the pc slot in the thread anchor
operand thread_anchor_pc(javaThread_RegP reg, immL_pc_off off)
%{
constraint(ALLOC_IN_RC(ptr_reg));
match(AddP reg off);
op_cost(0);
format %{ "[$reg, $off]" %}
interface(MEMORY_INTER) %{
base($reg);
index(0xffffffff);
scale(0x0);
disp($off);
%}
%}
//----------Special Memory Operands--------------------------------------------
// Stack Slot Operand - This operand is used for loading and storing temporary
// values on the stack where a match requires a value to
// flow through memory.
operand stackSlotI(sRegI reg)
%{
constraint(ALLOC_IN_RC(stack_slots));
// No match rule because this operand is only generated in matching
// match(RegI);
format %{ "[$reg]" %}
interface(MEMORY_INTER) %{
base(0x02); // RSP
index(0xffffffff); // No Index
scale(0x0); // No Scale
disp($reg); // Stack Offset
%}
%}
operand stackSlotF(sRegF reg)
%{
constraint(ALLOC_IN_RC(stack_slots));
// No match rule because this operand is only generated in matching
// match(RegF);
format %{ "[$reg]" %}
interface(MEMORY_INTER) %{
base(0x02); // RSP
index(0xffffffff); // No Index
scale(0x0); // No Scale
disp($reg); // Stack Offset
%}
%}
operand stackSlotD(sRegD reg)
%{
constraint(ALLOC_IN_RC(stack_slots));
// No match rule because this operand is only generated in matching
// match(RegD);
format %{ "[$reg]" %}
interface(MEMORY_INTER) %{
base(0x02); // RSP
index(0xffffffff); // No Index
scale(0x0); // No Scale
disp($reg); // Stack Offset
%}
%}
operand stackSlotL(sRegL reg)
%{
constraint(ALLOC_IN_RC(stack_slots));
// No match rule because this operand is only generated in matching
// match(RegL);
format %{ "[$reg]" %}
interface(MEMORY_INTER) %{
base(0x02); // RSP
index(0xffffffff); // No Index
scale(0x0); // No Scale
disp($reg); // Stack Offset
%}
%}
// Special operand allowing long args to int ops to be truncated for free
operand iRegL2I(iRegL reg) %{
op_cost(0);
match(ConvL2I reg);
format %{ "l2i($reg)" %}
interface(REG_INTER)
%}
// Comparison Operands
// NOTE: Label is a predefined operand which should not be redefined in
// the AD file. It is generically handled within the ADLC.
//----------Conditional Branch Operands----------------------------------------
// Comparison Op - This is the operation of the comparison, and is limited to
// the following set of codes:
// L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
//
// Other attributes of the comparison, such as unsignedness, are specified
// by the comparison instruction that sets a condition code flags register.
// That result is represented by a flags operand whose subtype is appropriate
// to the unsignedness (etc.) of the comparison.
//
// Later, the instruction which matches both the Comparison Op (a Bool) and
// the flags (produced by the Cmp) specifies the coding of the comparison op
// by matching a specific subtype of Bool operand below, such as cmpOpU.
// used for signed integral comparisons and fp comparisons
operand cmpOp()
%{
match(Bool);
format %{ "" %}
// the values in interface derives from struct BoolTest::mask
interface(COND_INTER) %{
equal(0x0, "eq");
greater(0x1, "gt");
overflow(0x2, "overflow");
less(0x3, "lt");
not_equal(0x4, "ne");
less_equal(0x5, "le");
no_overflow(0x6, "no_overflow");
greater_equal(0x7, "ge");
%}
%}
// used for unsigned integral comparisons
operand cmpOpU()
%{
match(Bool);
format %{ "" %}
// the values in interface derives from struct BoolTest::mask
interface(COND_INTER) %{
equal(0x0, "eq");
greater(0x1, "gtu");
overflow(0x2, "overflow");
less(0x3, "ltu");
not_equal(0x4, "ne");
less_equal(0x5, "leu");
no_overflow(0x6, "no_overflow");
greater_equal(0x7, "geu");
%}
%}
// used for certain integral comparisons which can be
// converted to bxx instructions
operand cmpOpEqNe()
%{
match(Bool);
op_cost(0);
predicate(n->as_Bool()->_test._test == BoolTest::ne ||
n->as_Bool()->_test._test == BoolTest::eq);
format %{ "" %}
interface(COND_INTER) %{
equal(0x0, "eq");
greater(0x1, "gt");
overflow(0x2, "overflow");
less(0x3, "lt");
not_equal(0x4, "ne");
less_equal(0x5, "le");
no_overflow(0x6, "no_overflow");
greater_equal(0x7, "ge");
%}
%}
operand cmpOpULtGe()
%{
match(Bool);
op_cost(0);
predicate(n->as_Bool()->_test._test == BoolTest::lt ||
n->as_Bool()->_test._test == BoolTest::ge);
format %{ "" %}
interface(COND_INTER) %{
equal(0x0, "eq");
greater(0x1, "gt");
overflow(0x2, "overflow");
less(0x3, "lt");
not_equal(0x4, "ne");
less_equal(0x5, "le");
no_overflow(0x6, "no_overflow");
greater_equal(0x7, "ge");
%}
%}
operand cmpOpUEqNeLeGt()
%{
match(Bool);
op_cost(0);
predicate(n->as_Bool()->_test._test == BoolTest::ne ||
n->as_Bool()->_test._test == BoolTest::eq ||
n->as_Bool()->_test._test == BoolTest::le ||
n->as_Bool()->_test._test == BoolTest::gt);
format %{ "" %}
interface(COND_INTER) %{
equal(0x0, "eq");
greater(0x1, "gt");
overflow(0x2, "overflow");
less(0x3, "lt");
not_equal(0x4, "ne");
less_equal(0x5, "le");
no_overflow(0x6, "no_overflow");
greater_equal(0x7, "ge");
%}
%}
// Flags register, used as output of compare logic
operand rFlagsReg()
%{
constraint(ALLOC_IN_RC(reg_flags));
match(RegFlags);
op_cost(0);
format %{ "RFLAGS" %}
interface(REG_INTER);
%}
// Special Registers
// Method Register
operand inline_cache_RegP(iRegP reg)
%{
constraint(ALLOC_IN_RC(method_reg)); // inline_cache_reg
match(reg);
match(iRegPNoSp);
op_cost(0);
format %{ %}
interface(REG_INTER);
%}
//----------OPERAND CLASSES----------------------------------------------------
// Operand Classes are groups of operands that are used as to simplify
// instruction definitions by not requiring the AD writer to specify
// separate instructions for every form of operand when the
// instruction accepts multiple operand types with the same basic
// encoding and format. The classic case of this is memory operands.
// memory is used to define read/write location for load/store
// instruction defs. we can turn a memory op into an Address
opclass memory(indirect, indOffI, indOffL, indirectN, indOffIN, indOffLN);
// iRegIorL2I is used for src inputs in rules for 32 bit int (I)
// operations. it allows the src to be either an iRegI or a (ConvL2I
// iRegL). in the latter case the l2i normally planted for a ConvL2I
// can be elided because the 32-bit instruction will just employ the
// lower 32 bits anyway.
//
// n.b. this does not elide all L2I conversions. if the truncated
// value is consumed by more than one operation then the ConvL2I
// cannot be bundled into the consuming nodes so an l2i gets planted
// (actually a mvw $dst $src) and the downstream instructions consume
// the result of the l2i as an iRegI input. That's a shame since the
// mvw is actually redundant but its not too costly.
opclass iRegIorL2I(iRegI, iRegL2I);
opclass iRegIorL(iRegI, iRegL);
opclass iRegNorP(iRegN, iRegP);
opclass iRegILNP(iRegI, iRegL, iRegN, iRegP);
opclass iRegILNPNoSp(iRegINoSp, iRegLNoSp, iRegNNoSp, iRegPNoSp);
opclass immIorL(immI, immL);
//----------PIPELINE-----------------------------------------------------------
// Rules which define the behavior of the target architectures pipeline.
// For specific pipelines, e.g. generic RISC-V, define the stages of that pipeline
//pipe_desc(ID, EX, MEM, WR);
#define ID S0
#define EX S1
#define MEM S2
#define WR S3
// Integer ALU reg operation
pipeline %{
attributes %{
// RISC-V instructions are of fixed length
fixed_size_instructions; // Fixed size instructions TODO does
max_instructions_per_bundle = 2; // Generic RISC-V 1, Sifive Series 7 2
// RISC-V instructions come in 32-bit word units
instruction_unit_size = 4; // An instruction is 4 bytes long
instruction_fetch_unit_size = 64; // The processor fetches one line
instruction_fetch_units = 1; // of 64 bytes
// List of nop instructions
nops( MachNop );
%}
// We don't use an actual pipeline model so don't care about resources
// or description. we do use pipeline classes to introduce fixed
// latencies
//----------RESOURCES----------------------------------------------------------
// Resources are the functional units available to the machine
// Generic RISC-V pipeline
// 1 decoder
// 1 instruction decoded per cycle
// 1 load/store ops per cycle, 1 branch, 1 FPU
// 1 mul, 1 div
resources ( DECODE,
ALU,
MUL,
DIV,
BRANCH,
LDST,
FPU);
//----------PIPELINE DESCRIPTION-----------------------------------------------
// Pipeline Description specifies the stages in the machine's pipeline
// Define the pipeline as a generic 6 stage pipeline
pipe_desc(S0, S1, S2, S3, S4, S5);
//----------PIPELINE CLASSES---------------------------------------------------
// Pipeline Classes describe the stages in which input and output are
// referenced by the hardware pipeline.
pipe_class fp_dop_reg_reg_s(fRegF dst, fRegF src1, fRegF src2)
%{
single_instruction;
src1 : S1(read);
src2 : S2(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_dop_reg_reg_d(fRegD dst, fRegD src1, fRegD src2)
%{
src1 : S1(read);
src2 : S2(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_uop_s(fRegF dst, fRegF src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_uop_d(fRegD dst, fRegD src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_d2f(fRegF dst, fRegD src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_f2d(fRegD dst, fRegF src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_f2i(iRegINoSp dst, fRegF src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_f2l(iRegLNoSp dst, fRegF src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_i2f(fRegF dst, iRegIorL2I src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_l2f(fRegF dst, iRegL src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_d2i(iRegINoSp dst, fRegD src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_d2l(iRegLNoSp dst, fRegD src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_i2d(fRegD dst, iRegIorL2I src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_l2d(fRegD dst, iRegIorL2I src)
%{
single_instruction;
src : S1(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_div_s(fRegF dst, fRegF src1, fRegF src2)
%{
single_instruction;
src1 : S1(read);
src2 : S2(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_div_d(fRegD dst, fRegD src1, fRegD src2)
%{
single_instruction;
src1 : S1(read);
src2 : S2(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_sqrt_s(fRegF dst, fRegF src1, fRegF src2)
%{
single_instruction;
src1 : S1(read);
src2 : S2(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_sqrt_d(fRegD dst, fRegD src1, fRegD src2)
%{
single_instruction;
src1 : S1(read);
src2 : S2(read);
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_load_constant_s(fRegF dst)
%{
single_instruction;
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_load_constant_d(fRegD dst)
%{
single_instruction;
dst : S5(write);
DECODE : ID;
FPU : S5;
%}
pipe_class fp_load_mem_s(fRegF dst, memory mem)
%{
single_instruction;
mem : S1(read);
dst : S5(write);
DECODE : ID;
LDST : MEM;
%}
pipe_class fp_load_mem_d(fRegD dst, memory mem)
%{
single_instruction;
mem : S1(read);
dst : S5(write);
DECODE : ID;
LDST : MEM;
%}
pipe_class fp_store_reg_s(fRegF src, memory mem)
%{
single_instruction;
src : S1(read);
mem : S5(write);
DECODE : ID;
LDST : MEM;
%}
pipe_class fp_store_reg_d(fRegD src, memory mem)
%{
single_instruction;
src : S1(read);
mem : S5(write);
DECODE : ID;
LDST : MEM;
%}
//------- Integer ALU operations --------------------------
// Integer ALU reg-reg operation
// Operands needs in ID, result generated in EX
// E.g. ADD Rd, Rs1, Rs2
pipe_class ialu_reg_reg(iRegI dst, iRegI src1, iRegI src2)
%{
single_instruction;
dst : EX(write);
src1 : ID(read);
src2 : ID(read);
DECODE : ID;
ALU : EX;
%}
// Integer ALU reg operation with constant shift
// E.g. SLLI Rd, Rs1, #shift
pipe_class ialu_reg_shift(iRegI dst, iRegI src1)
%{
single_instruction;
dst : EX(write);
src1 : ID(read);
DECODE : ID;
ALU : EX;
%}
// Integer ALU reg-reg operation with variable shift
// both operands must be available in ID
// E.g. SLL Rd, Rs1, Rs2
pipe_class ialu_reg_reg_vshift(iRegI dst, iRegI src1, iRegI src2)
%{
single_instruction;
dst : EX(write);
src1 : ID(read);
src2 : ID(read);
DECODE : ID;
ALU : EX;
%}
// Integer ALU reg operation
// E.g. NEG Rd, Rs2
pipe_class ialu_reg(iRegI dst, iRegI src)
%{
single_instruction;
dst : EX(write);
src : ID(read);
DECODE : ID;
ALU : EX;
%}
// Integer ALU reg immediate operation
// E.g. ADDI Rd, Rs1, #imm
pipe_class ialu_reg_imm(iRegI dst, iRegI src1)
%{
single_instruction;
dst : EX(write);
src1 : ID(read);
DECODE : ID;
ALU : EX;
%}
// Integer ALU immediate operation (no source operands)
// E.g. LI Rd, #imm
pipe_class ialu_imm(iRegI dst)
%{
single_instruction;
dst : EX(write);
DECODE : ID;
ALU : EX;
%}
//------- Multiply pipeline operations --------------------
// Multiply reg-reg
// E.g. MULW Rd, Rs1, Rs2
pipe_class imul_reg_reg(iRegI dst, iRegI src1, iRegI src2)
%{
single_instruction;
dst : WR(write);
src1 : ID(read);
src2 : ID(read);
DECODE : ID;
MUL : WR;
%}
// E.g. MUL RD, Rs1, Rs2
pipe_class lmul_reg_reg(iRegI dst, iRegI src1, iRegI src2)
%{
single_instruction;
fixed_latency(3); // Maximum latency for 64 bit mul
dst : WR(write);
src1 : ID(read);
src2 : ID(read);
DECODE : ID;
MUL : WR;
%}
//------- Divide pipeline operations --------------------
// E.g. DIVW Rd, Rs1, Rs2
pipe_class idiv_reg_reg(iRegI dst, iRegI src1, iRegI src2)
%{
single_instruction;
fixed_latency(8); // Maximum latency for 32 bit divide
dst : WR(write);
src1 : ID(read);
src2 : ID(read);
DECODE : ID;
DIV : WR;
%}
// E.g. DIV RD, Rs1, Rs2
pipe_class ldiv_reg_reg(iRegI dst, iRegI src1, iRegI src2)
%{
single_instruction;
fixed_latency(16); // Maximum latency for 64 bit divide
dst : WR(write);
src1 : ID(read);
src2 : ID(read);
DECODE : ID;
DIV : WR;
%}
//------- Load pipeline operations ------------------------
// Load - prefetch
// Eg. PREFETCH_W mem
pipe_class iload_prefetch(memory mem)
%{
single_instruction;
mem : ID(read);
DECODE : ID;
LDST : MEM;
%}
// Load - reg, mem
// E.g. LA Rd, mem
pipe_class iload_reg_mem(iRegI dst, memory mem)
%{
single_instruction;
dst : WR(write);
mem : ID(read);
DECODE : ID;
LDST : MEM;
%}
// Load - reg, reg
// E.g. LD Rd, Rs
pipe_class iload_reg_reg(iRegI dst, iRegI src)
%{
single_instruction;
dst : WR(write);
src : ID(read);
DECODE : ID;
LDST : MEM;
%}
//------- Store pipeline operations -----------------------
// Store - zr, mem
// E.g. SD zr, mem
pipe_class istore_mem(memory mem)
%{
single_instruction;
mem : ID(read);
DECODE : ID;
LDST : MEM;
%}
// Store - reg, mem
// E.g. SD Rs, mem
pipe_class istore_reg_mem(iRegI src, memory mem)
%{
single_instruction;
mem : ID(read);
src : EX(read);
DECODE : ID;
LDST : MEM;
%}
// Store - reg, reg
// E.g. SD Rs2, Rs1
pipe_class istore_reg_reg(iRegI dst, iRegI src)
%{
single_instruction;
dst : ID(read);
src : EX(read);
DECODE : ID;
LDST : MEM;
%}
//------- Store pipeline operations -----------------------
// Branch
pipe_class pipe_branch()
%{
single_instruction;
DECODE : ID;
BRANCH : EX;
%}
// Branch
pipe_class pipe_branch_reg(iRegI src)
%{
single_instruction;
src : ID(read);
DECODE : ID;
BRANCH : EX;
%}
// Compare & Branch
// E.g. BEQ Rs1, Rs2, L
pipe_class pipe_cmp_branch(iRegI src1, iRegI src2)
%{
single_instruction;
src1 : ID(read);
src2 : ID(read);
DECODE : ID;
BRANCH : EX;
%}
// E.g. BEQZ Rs, L
pipe_class pipe_cmpz_branch(iRegI src)
%{
single_instruction;
src : ID(read);
DECODE : ID;
BRANCH : EX;
%}
//------- Synchronisation operations ----------------------
// Any operation requiring serialization
// E.g. FENCE/Atomic Ops/Load Acquire/Store Release
pipe_class pipe_serial()
%{
single_instruction;
force_serialization;
fixed_latency(16);
DECODE : ID;
LDST : MEM;
%}
pipe_class pipe_slow()
%{
instruction_count(10);
multiple_bundles;
force_serialization;
fixed_latency(16);
DECODE : ID;
LDST : MEM;
%}
// Empty pipeline class
pipe_class pipe_class_empty()
%{
single_instruction;
fixed_latency(0);
%}
// Default pipeline class.
pipe_class pipe_class_default()
%{
single_instruction;
fixed_latency(2);
%}
// Pipeline class for compares.
pipe_class pipe_class_compare()
%{
single_instruction;
fixed_latency(16);
%}
// Pipeline class for memory operations.
pipe_class pipe_class_memory()
%{
single_instruction;
fixed_latency(16);
%}
// Pipeline class for call.
pipe_class pipe_class_call()
%{
single_instruction;
fixed_latency(100);
%}
// Define the class for the Nop node.
define %{
MachNop = pipe_class_empty;
%}
%}
//----------INSTRUCTIONS-------------------------------------------------------
//
// match -- States which machine-independent subtree may be replaced
// by this instruction.
// ins_cost -- The estimated cost of this instruction is used by instruction
// selection to identify a minimum cost tree of machine
// instructions that matches a tree of machine-independent
// instructions.
// format -- A string providing the disassembly for this instruction.
// The value of an instruction's operand may be inserted
// by referring to it with a '$' prefix.
// opcode -- Three instruction opcodes may be provided. These are referred
// to within an encode class as $primary, $secondary, and $tertiary
// rrspectively. The primary opcode is commonly used to
// indicate the type of machine instruction, while secondary
// and tertiary are often used for prefix options or addressing
// modes.
// ins_encode -- A list of encode classes with parameters. The encode class
// name must have been defined in an 'enc_class' specification
// in the encode section of the architecture description.
// ============================================================================
// Memory (Load/Store) Instructions
// Load Instructions
// Load Byte (8 bit signed)
instruct loadB(iRegINoSp dst, memory mem)
%{
match(Set dst (LoadB mem));
ins_cost(LOAD_COST);
format %{ "lb $dst, $mem\t# byte, #@loadB" %}
ins_encode %{
__ lb(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Byte (8 bit signed) into long
instruct loadB2L(iRegLNoSp dst, memory mem)
%{
match(Set dst (ConvI2L (LoadB mem)));
ins_cost(LOAD_COST);
format %{ "lb $dst, $mem\t# byte, #@loadB2L" %}
ins_encode %{
__ lb(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Byte (8 bit unsigned)
instruct loadUB(iRegINoSp dst, memory mem)
%{
match(Set dst (LoadUB mem));
ins_cost(LOAD_COST);
format %{ "lbu $dst, $mem\t# byte, #@loadUB" %}
ins_encode %{
__ lbu(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Byte (8 bit unsigned) into long
instruct loadUB2L(iRegLNoSp dst, memory mem)
%{
match(Set dst (ConvI2L (LoadUB mem)));
ins_cost(LOAD_COST);
format %{ "lbu $dst, $mem\t# byte, #@loadUB2L" %}
ins_encode %{
__ lbu(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Short (16 bit signed)
instruct loadS(iRegINoSp dst, memory mem)
%{
match(Set dst (LoadS mem));
ins_cost(LOAD_COST);
format %{ "lh $dst, $mem\t# short, #@loadS" %}
ins_encode %{
__ lh(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Short (16 bit signed) into long
instruct loadS2L(iRegLNoSp dst, memory mem)
%{
match(Set dst (ConvI2L (LoadS mem)));
ins_cost(LOAD_COST);
format %{ "lh $dst, $mem\t# short, #@loadS2L" %}
ins_encode %{
__ lh(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Char (16 bit unsigned)
instruct loadUS(iRegINoSp dst, memory mem)
%{
match(Set dst (LoadUS mem));
ins_cost(LOAD_COST);
format %{ "lhu $dst, $mem\t# short, #@loadUS" %}
ins_encode %{
__ lhu(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Short/Char (16 bit unsigned) into long
instruct loadUS2L(iRegLNoSp dst, memory mem)
%{
match(Set dst (ConvI2L (LoadUS mem)));
ins_cost(LOAD_COST);
format %{ "lhu $dst, $mem\t# short, #@loadUS2L" %}
ins_encode %{
__ lhu(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Integer (32 bit signed)
instruct loadI(iRegINoSp dst, memory mem)
%{
match(Set dst (LoadI mem));
ins_cost(LOAD_COST);
format %{ "lw $dst, $mem\t# int, #@loadI" %}
ins_encode %{
__ lw(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Integer (32 bit signed) into long
instruct loadI2L(iRegLNoSp dst, memory mem)
%{
match(Set dst (ConvI2L (LoadI mem)));
ins_cost(LOAD_COST);
format %{ "lw $dst, $mem\t# int, #@loadI2L" %}
ins_encode %{
__ lw(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Integer (32 bit unsigned) into long
instruct loadUI2L(iRegLNoSp dst, memory mem, immL_32bits mask)
%{
match(Set dst (AndL (ConvI2L (LoadI mem)) mask));
ins_cost(LOAD_COST);
format %{ "lwu $dst, $mem\t# int, #@loadUI2L" %}
ins_encode %{
__ lwu(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Long (64 bit signed)
instruct loadL(iRegLNoSp dst, memory mem)
%{
match(Set dst (LoadL mem));
ins_cost(LOAD_COST);
format %{ "ld $dst, $mem\t# int, #@loadL" %}
ins_encode %{
__ ld(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Range
instruct loadRange(iRegINoSp dst, memory mem)
%{
match(Set dst (LoadRange mem));
ins_cost(LOAD_COST);
format %{ "lwu $dst, $mem\t# range, #@loadRange" %}
ins_encode %{
__ lwu(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Pointer
instruct loadP(iRegPNoSp dst, memory mem)
%{
match(Set dst (LoadP mem));
predicate(n->as_Load()->barrier_data() == 0);
ins_cost(LOAD_COST);
format %{ "ld $dst, $mem\t# ptr, #@loadP" %}
ins_encode %{
__ ld(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Compressed Pointer
instruct loadN(iRegNNoSp dst, memory mem)
%{
match(Set dst (LoadN mem));
ins_cost(LOAD_COST);
format %{ "lwu $dst, $mem\t# loadN, compressed ptr, #@loadN" %}
ins_encode %{
__ lwu(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Klass Pointer
instruct loadKlass(iRegPNoSp dst, memory mem)
%{
match(Set dst (LoadKlass mem));
ins_cost(LOAD_COST);
format %{ "ld $dst, $mem\t# class, #@loadKlass" %}
ins_encode %{
__ ld(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Narrow Klass Pointer
instruct loadNKlass(iRegNNoSp dst, memory mem)
%{
match(Set dst (LoadNKlass mem));
ins_cost(LOAD_COST);
format %{ "lwu $dst, $mem\t# loadNKlass, compressed class ptr, #@loadNKlass" %}
ins_encode %{
__ lwu(as_Register($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(iload_reg_mem);
%}
// Load Float
instruct loadF(fRegF dst, memory mem)
%{
match(Set dst (LoadF mem));
ins_cost(LOAD_COST);
format %{ "flw $dst, $mem\t# float, #@loadF" %}
ins_encode %{
__ flw(as_FloatRegister($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(fp_load_mem_s);
%}
// Load Double
instruct loadD(fRegD dst, memory mem)
%{
match(Set dst (LoadD mem));
ins_cost(LOAD_COST);
format %{ "fld $dst, $mem\t# double, #@loadD" %}
ins_encode %{
__ fld(as_FloatRegister($dst$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(fp_load_mem_d);
%}
// Load Int Constant
instruct loadConI(iRegINoSp dst, immI src)
%{
match(Set dst src);
ins_cost(ALU_COST);
format %{ "li $dst, $src\t# int, #@loadConI" %}
ins_encode(riscv_enc_li_imm(dst, src));
ins_pipe(ialu_imm);
%}
// Load Long Constant
instruct loadConL(iRegLNoSp dst, immL src)
%{
match(Set dst src);
ins_cost(ALU_COST);
format %{ "li $dst, $src\t# long, #@loadConL" %}
ins_encode(riscv_enc_li_imm(dst, src));
ins_pipe(ialu_imm);
%}
// Load Pointer Constant
instruct loadConP(iRegPNoSp dst, immP con)
%{
match(Set dst con);
ins_cost(ALU_COST);
format %{ "mv $dst, $con\t# ptr, #@loadConP" %}
ins_encode(riscv_enc_mov_p(dst, con));
ins_pipe(ialu_imm);
%}
// Load Null Pointer Constant
instruct loadConP0(iRegPNoSp dst, immP0 con)
%{
match(Set dst con);
ins_cost(ALU_COST);
format %{ "mv $dst, $con\t# NULL ptr, #@loadConP0" %}
ins_encode(riscv_enc_mov_zero(dst));
ins_pipe(ialu_imm);
%}
// Load Pointer Constant One
instruct loadConP1(iRegPNoSp dst, immP_1 con)
%{
match(Set dst con);
ins_cost(ALU_COST);
format %{ "mv $dst, $con\t# load ptr constant one, #@loadConP1" %}
ins_encode(riscv_enc_mov_p1(dst));
ins_pipe(ialu_imm);
%}
// Load Byte Map Base Constant
instruct loadByteMapBase(iRegPNoSp dst, immByteMapBase con)
%{
match(Set dst con);
ins_cost(ALU_COST);
format %{ "mv $dst, $con\t# Byte Map Base, #@loadByteMapBase" %}
ins_encode(riscv_enc_mov_byte_map_base(dst));
ins_pipe(ialu_imm);
%}
// Load Narrow Pointer Constant
instruct loadConN(iRegNNoSp dst, immN con)
%{
match(Set dst con);
ins_cost(ALU_COST * 4);
format %{ "mv $dst, $con\t# compressed ptr, #@loadConN" %}
ins_encode(riscv_enc_mov_n(dst, con));
ins_pipe(ialu_imm);
%}
// Load Narrow Null Pointer Constant
instruct loadConN0(iRegNNoSp dst, immN0 con)
%{
match(Set dst con);
ins_cost(ALU_COST);
format %{ "mv $dst, $con\t# compressed NULL ptr, #@loadConN0" %}
ins_encode(riscv_enc_mov_zero(dst));
ins_pipe(ialu_imm);
%}
// Load Narrow Klass Constant
instruct loadConNKlass(iRegNNoSp dst, immNKlass con)
%{
match(Set dst con);
ins_cost(ALU_COST * 6);
format %{ "mv $dst, $con\t# compressed klass ptr, #@loadConNKlass" %}
ins_encode(riscv_enc_mov_nk(dst, con));
ins_pipe(ialu_imm);
%}
// Load Float Constant
instruct loadConF(fRegF dst, immF con) %{
match(Set dst con);
ins_cost(LOAD_COST);
format %{
"flw $dst, [$constantaddress]\t# load from constant table: float=$con, #@loadConF"
%}
ins_encode %{
__ flw(as_FloatRegister($dst$$reg), $constantaddress($con));
%}
ins_pipe(fp_load_constant_s);
%}
instruct loadConF0(fRegF dst, immF0 con) %{
match(Set dst con);
ins_cost(XFER_COST);
format %{ "fmv.w.x $dst, zr\t# float, #@loadConF0" %}
ins_encode %{
__ fmv_w_x(as_FloatRegister($dst$$reg), zr);
%}
ins_pipe(fp_load_constant_s);
%}
// Load Double Constant
instruct loadConD(fRegD dst, immD con) %{
match(Set dst con);
ins_cost(LOAD_COST);
format %{
"fld $dst, [$constantaddress]\t# load from constant table: double=$con, #@loadConD"
%}
ins_encode %{
__ fld(as_FloatRegister($dst$$reg), $constantaddress($con));
%}
ins_pipe(fp_load_constant_d);
%}
instruct loadConD0(fRegD dst, immD0 con) %{
match(Set dst con);
ins_cost(XFER_COST);
format %{ "fmv.d.x $dst, zr\t# double, #@loadConD0" %}
ins_encode %{
__ fmv_d_x(as_FloatRegister($dst$$reg), zr);
%}
ins_pipe(fp_load_constant_d);
%}
// Store Instructions
// Store CMS card-mark Immediate
instruct storeimmCM0(immI0 zero, memory mem)
%{
match(Set mem (StoreCM mem zero));
ins_cost(STORE_COST);
format %{ "storestore (elided)\n\t"
"sb zr, $mem\t# byte, #@storeimmCM0" %}
ins_encode %{
__ sb(zr, Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_mem);
%}
// Store CMS card-mark Immediate with intervening StoreStore
// needed when using CMS with no conditional card marking
instruct storeimmCM0_ordered(immI0 zero, memory mem)
%{
match(Set mem (StoreCM mem zero));
ins_cost(ALU_COST + STORE_COST);
format %{ "membar(StoreStore)\n\t"
"sb zr, $mem\t# byte, #@storeimmCM0_ordered" %}
ins_encode %{
__ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
__ sb(zr, Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_mem);
%}
// Store Byte
instruct storeB(iRegIorL2I src, memory mem)
%{
match(Set mem (StoreB mem src));
ins_cost(STORE_COST);
format %{ "sb $src, $mem\t# byte, #@storeB" %}
ins_encode %{
__ sb(as_Register($src$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_reg_mem);
%}
instruct storeimmB0(immI0 zero, memory mem)
%{
match(Set mem (StoreB mem zero));
ins_cost(STORE_COST);
format %{ "sb zr, $mem\t# byte, #@storeimmB0" %}
ins_encode %{
__ sb(zr, Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_mem);
%}
// Store Char/Short
instruct storeC(iRegIorL2I src, memory mem)
%{
match(Set mem (StoreC mem src));
ins_cost(STORE_COST);
format %{ "sh $src, $mem\t# short, #@storeC" %}
ins_encode %{
__ sh(as_Register($src$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_reg_mem);
%}
instruct storeimmC0(immI0 zero, memory mem)
%{
match(Set mem (StoreC mem zero));
ins_cost(STORE_COST);
format %{ "sh zr, $mem\t# short, #@storeimmC0" %}
ins_encode %{
__ sh(zr, Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_mem);
%}
// Store Integer
instruct storeI(iRegIorL2I src, memory mem)
%{
match(Set mem(StoreI mem src));
ins_cost(STORE_COST);
format %{ "sw $src, $mem\t# int, #@storeI" %}
ins_encode %{
__ sw(as_Register($src$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_reg_mem);
%}
instruct storeimmI0(immI0 zero, memory mem)
%{
match(Set mem(StoreI mem zero));
ins_cost(STORE_COST);
format %{ "sw zr, $mem\t# int, #@storeimmI0" %}
ins_encode %{
__ sw(zr, Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_mem);
%}
// Store Long (64 bit signed)
instruct storeL(iRegL src, memory mem)
%{
match(Set mem (StoreL mem src));
ins_cost(STORE_COST);
format %{ "sd $src, $mem\t# long, #@storeL" %}
ins_encode %{
__ sd(as_Register($src$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_reg_mem);
%}
// Store Long (64 bit signed)
instruct storeimmL0(immL0 zero, memory mem)
%{
match(Set mem (StoreL mem zero));
ins_cost(STORE_COST);
format %{ "sd zr, $mem\t# long, #@storeimmL0" %}
ins_encode %{
__ sd(zr, Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_mem);
%}
// Store Pointer
instruct storeP(iRegP src, memory mem)
%{
match(Set mem (StoreP mem src));
ins_cost(STORE_COST);
format %{ "sd $src, $mem\t# ptr, #@storeP" %}
ins_encode %{
__ sd(as_Register($src$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_reg_mem);
%}
// Store Pointer
instruct storeimmP0(immP0 zero, memory mem)
%{
match(Set mem (StoreP mem zero));
ins_cost(STORE_COST);
format %{ "sd zr, $mem\t# ptr, #@storeimmP0" %}
ins_encode %{
__ sd(zr, Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_mem);
%}
// Store Compressed Pointer
instruct storeN(iRegN src, memory mem)
%{
match(Set mem (StoreN mem src));
ins_cost(STORE_COST);
format %{ "sw $src, $mem\t# compressed ptr, #@storeN" %}
ins_encode %{
__ sw(as_Register($src$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_reg_mem);
%}
instruct storeImmN0(iRegIHeapbase heapbase, immN0 zero, memory mem)
%{
match(Set mem (StoreN mem zero));
ins_cost(STORE_COST);
format %{ "sw rheapbase, $mem\t# compressed ptr (rheapbase==0), #@storeImmN0" %}
ins_encode %{
__ sw(as_Register($heapbase$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_reg_mem);
%}
// Store Float
instruct storeF(fRegF src, memory mem)
%{
match(Set mem (StoreF mem src));
ins_cost(STORE_COST);
format %{ "fsw $src, $mem\t# float, #@storeF" %}
ins_encode %{
__ fsw(as_FloatRegister($src$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(fp_store_reg_s);
%}
// Store Double
instruct storeD(fRegD src, memory mem)
%{
match(Set mem (StoreD mem src));
ins_cost(STORE_COST);
format %{ "fsd $src, $mem\t# double, #@storeD" %}
ins_encode %{
__ fsd(as_FloatRegister($src$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(fp_store_reg_d);
%}
// Store Compressed Klass Pointer
instruct storeNKlass(iRegN src, memory mem)
%{
match(Set mem (StoreNKlass mem src));
ins_cost(STORE_COST);
format %{ "sw $src, $mem\t# compressed klass ptr, #@storeNKlass" %}
ins_encode %{
__ sw(as_Register($src$$reg), Address(as_Register($mem$$base), $mem$$disp));
%}
ins_pipe(istore_reg_mem);
%}
// ============================================================================
// Prefetch instructions
// Must be safe to execute with invalid address (cannot fault).
instruct prefetchalloc( memory mem ) %{
predicate(UseZicbop);
match(PrefetchAllocation mem);
ins_cost(ALU_COST * 1);
format %{ "prefetch_w $mem\t# Prefetch for write" %}
ins_encode %{
if (is_imm_in_range($mem$$disp, 12, 0)) {
if (($mem$$disp & 0x1f) == 0) {
__ prefetch_w(as_Register($mem$$base), $mem$$disp);
} else {
__ addi(t0, as_Register($mem$$base), $mem$$disp);
__ prefetch_w(t0, 0);
}
} else {
__ mv(t0, $mem$$disp);
__ add(t0, as_Register($mem$$base), t0);
__ prefetch_w(t0, 0);
}
%}
ins_pipe(iload_prefetch);
%}
// ============================================================================
// Atomic operation instructions
//
// standard CompareAndSwapX when we are using barriers
// these have higher priority than the rules selected by a predicate
instruct compareAndSwapB(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
match(Set res (CompareAndSwapB mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 10 + BRANCH_COST * 4);
effect(TEMP_DEF res, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3, KILL cr);
format %{
"cmpxchg $mem, $oldval, $newval\t# (byte) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapB"
%}
ins_encode %{
__ cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int8,
Assembler::relaxed /* acquire */, Assembler::rl /* release */, $res$$Register,
true /* result as bool */, $tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndSwapS(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
match(Set res (CompareAndSwapS mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 11 + BRANCH_COST * 4);
effect(TEMP_DEF res, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3, KILL cr);
format %{
"cmpxchg $mem, $oldval, $newval\t# (short) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapS"
%}
ins_encode %{
__ cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int16,
Assembler::relaxed /* acquire */, Assembler::rl /* release */, $res$$Register,
true /* result as bool */, $tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndSwapI(iRegINoSp res, indirect mem, iRegI oldval, iRegI newval)
%{
match(Set res (CompareAndSwapI mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 6 + BRANCH_COST * 4);
format %{
"cmpxchg $mem, $oldval, $newval\t# (int) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapI"
%}
ins_encode(riscv_enc_cmpxchgw(res, mem, oldval, newval));
ins_pipe(pipe_slow);
%}
instruct compareAndSwapL(iRegINoSp res, indirect mem, iRegL oldval, iRegL newval)
%{
match(Set res (CompareAndSwapL mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 6 + BRANCH_COST * 4);
format %{
"cmpxchg $mem, $oldval, $newval\t# (long) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapL"
%}
ins_encode(riscv_enc_cmpxchg(res, mem, oldval, newval));
ins_pipe(pipe_slow);
%}
instruct compareAndSwapP(iRegINoSp res, indirect mem, iRegP oldval, iRegP newval)
%{
predicate(n->as_LoadStore()->barrier_data() == 0);
match(Set res (CompareAndSwapP mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 6 + BRANCH_COST * 4);
format %{
"cmpxchg $mem, $oldval, $newval\t# (ptr) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapP"
%}
ins_encode(riscv_enc_cmpxchg(res, mem, oldval, newval));
ins_pipe(pipe_slow);
%}
instruct compareAndSwapN(iRegINoSp res, indirect mem, iRegN oldval, iRegN newval)
%{
match(Set res (CompareAndSwapN mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 8 + BRANCH_COST * 4);
format %{
"cmpxchg $mem, $oldval, $newval\t# (narrow oop) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapN"
%}
ins_encode(riscv_enc_cmpxchgn(res, mem, oldval, newval));
ins_pipe(pipe_slow);
%}
// alternative CompareAndSwapX when we are eliding barriers
instruct compareAndSwapBAcq(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndSwapB mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 10 + BRANCH_COST * 4);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg_acq $mem, $oldval, $newval\t# (byte) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapBAcq"
%}
ins_encode %{
__ cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int8,
Assembler::aq /* acquire */, Assembler::rl /* release */, $res$$Register,
true /* result as bool */, $tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndSwapSAcq(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndSwapS mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 11 + BRANCH_COST * 4);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg_acq $mem, $oldval, $newval\t# (short) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapSAcq"
%}
ins_encode %{
__ cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int16,
Assembler::aq /* acquire */, Assembler::rl /* release */, $res$$Register,
true /* result as bool */, $tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndSwapIAcq(iRegINoSp res, indirect mem, iRegI oldval, iRegI newval)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndSwapI mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 6 + BRANCH_COST * 4);
format %{
"cmpxchg_acq $mem, $oldval, $newval\t# (int) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapIAcq"
%}
ins_encode(riscv_enc_cmpxchgw_acq(res, mem, oldval, newval));
ins_pipe(pipe_slow);
%}
instruct compareAndSwapLAcq(iRegINoSp res, indirect mem, iRegL oldval, iRegL newval)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndSwapL mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 6 + BRANCH_COST * 4);
format %{
"cmpxchg_acq $mem, $oldval, $newval\t# (long) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapLAcq"
%}
ins_encode(riscv_enc_cmpxchg_acq(res, mem, oldval, newval));
ins_pipe(pipe_slow);
%}
instruct compareAndSwapPAcq(iRegINoSp res, indirect mem, iRegP oldval, iRegP newval)
%{
predicate(needs_acquiring_load_reserved(n) && (n->as_LoadStore()->barrier_data() == 0));
match(Set res (CompareAndSwapP mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 6 + BRANCH_COST * 4);
format %{
"cmpxchg_acq $mem, $oldval, $newval\t# (ptr) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapPAcq"
%}
ins_encode(riscv_enc_cmpxchg_acq(res, mem, oldval, newval));
ins_pipe(pipe_slow);
%}
instruct compareAndSwapNAcq(iRegINoSp res, indirect mem, iRegN oldval, iRegN newval)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndSwapN mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + ALU_COST * 8 + BRANCH_COST * 4);
format %{
"cmpxchg_acq $mem, $oldval, $newval\t# (narrow oop) if $mem == $oldval then $mem <-- $newval\n\t"
"mv $res, $res == $oldval\t# $res <-- ($res == $oldval ? 1 : 0), #@compareAndSwapNAcq"
%}
ins_encode(riscv_enc_cmpxchgn_acq(res, mem, oldval, newval));
ins_pipe(pipe_slow);
%}
// Sundry CAS operations. Note that release is always true,
// regardless of the memory ordering of the CAS. This is because we
// need the volatile case to be sequentially consistent but there is
// no trailing StoreLoad barrier emitted by C2. Unfortunately we
// can't check the type of memory ordering here, so we always emit a
// sc_d(w) with rl bit set.
instruct compareAndExchangeB(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
match(Set res (CompareAndExchangeB mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST * 5);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg $res = $mem, $oldval, $newval\t# (byte, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeB"
%}
ins_encode %{
__ cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int8,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register,
/*result_as_bool*/ false, $tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeS(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
match(Set res (CompareAndExchangeS mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST * 6);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg $res = $mem, $oldval, $newval\t# (short, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeS"
%}
ins_encode %{
__ cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int16,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register,
/*result_as_bool*/ false, $tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeI(iRegINoSp res, indirect mem, iRegI oldval, iRegI newval)
%{
match(Set res (CompareAndExchangeI mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST);
effect(TEMP_DEF res);
format %{
"cmpxchg $res = $mem, $oldval, $newval\t# (int, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeI"
%}
ins_encode %{
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int32,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeL(iRegLNoSp res, indirect mem, iRegL oldval, iRegL newval)
%{
match(Set res (CompareAndExchangeL mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST);
effect(TEMP_DEF res);
format %{
"cmpxchg $res = $mem, $oldval, $newval\t# (long, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeL"
%}
ins_encode %{
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeN(iRegNNoSp res, indirect mem, iRegN oldval, iRegN newval)
%{
match(Set res (CompareAndExchangeN mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST * 3);
effect(TEMP_DEF res);
format %{
"cmpxchg $res = $mem, $oldval, $newval\t# (narrow oop, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeN"
%}
ins_encode %{
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::uint32,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeP(iRegPNoSp res, indirect mem, iRegP oldval, iRegP newval)
%{
predicate(n->as_LoadStore()->barrier_data() == 0);
match(Set res (CompareAndExchangeP mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST);
effect(TEMP_DEF res);
format %{
"cmpxchg $res = $mem, $oldval, $newval\t# (ptr, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeP"
%}
ins_encode %{
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeBAcq(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndExchangeB mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST * 5);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg_acq $res = $mem, $oldval, $newval\t# (byte, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeBAcq"
%}
ins_encode %{
__ cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int8,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register,
/*result_as_bool*/ false, $tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeSAcq(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndExchangeS mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST * 6);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg_acq $res = $mem, $oldval, $newval\t# (short, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeSAcq"
%}
ins_encode %{
__ cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int16,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register,
/*result_as_bool*/ false, $tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeIAcq(iRegINoSp res, indirect mem, iRegI oldval, iRegI newval)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndExchangeI mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST);
effect(TEMP_DEF res);
format %{
"cmpxchg_acq $res = $mem, $oldval, $newval\t# (int, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeIAcq"
%}
ins_encode %{
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int32,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeLAcq(iRegLNoSp res, indirect mem, iRegL oldval, iRegL newval)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndExchangeL mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST);
effect(TEMP_DEF res);
format %{
"cmpxchg_acq $res = $mem, $oldval, $newval\t# (long, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeLAcq"
%}
ins_encode %{
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangeNAcq(iRegNNoSp res, indirect mem, iRegN oldval, iRegN newval)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (CompareAndExchangeN mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST);
effect(TEMP_DEF res);
format %{
"cmpxchg_acq $res = $mem, $oldval, $newval\t# (narrow oop, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangeNAcq"
%}
ins_encode %{
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::uint32,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct compareAndExchangePAcq(iRegPNoSp res, indirect mem, iRegP oldval, iRegP newval)
%{
predicate(needs_acquiring_load_reserved(n) && (n->as_LoadStore()->barrier_data() == 0));
match(Set res (CompareAndExchangeP mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 3 + ALU_COST);
effect(TEMP_DEF res);
format %{
"cmpxchg_acq $res = $mem, $oldval, $newval\t# (ptr, weak) if $mem == $oldval then $mem <-- $newval, #@compareAndExchangePAcq"
%}
ins_encode %{
__ cmpxchg(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapB(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
match(Set res (WeakCompareAndSwapB mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 6);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg_weak $mem, $oldval, $newval\t# (byte, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapB"
%}
ins_encode %{
__ weak_cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int8,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapS(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
match(Set res (WeakCompareAndSwapS mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 7);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg_weak $mem, $oldval, $newval\t# (short, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapS"
%}
ins_encode %{
__ weak_cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int16,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapI(iRegINoSp res, indirect mem, iRegI oldval, iRegI newval)
%{
match(Set res (WeakCompareAndSwapI mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 2);
format %{
"cmpxchg_weak $mem, $oldval, $newval\t# (int, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapI"
%}
ins_encode %{
__ cmpxchg_weak(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int32,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapL(iRegINoSp res, indirect mem, iRegL oldval, iRegL newval)
%{
match(Set res (WeakCompareAndSwapL mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 2);
format %{
"cmpxchg_weak $mem, $oldval, $newval\t# (long, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapL"
%}
ins_encode %{
__ cmpxchg_weak(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapN(iRegINoSp res, indirect mem, iRegN oldval, iRegN newval)
%{
match(Set res (WeakCompareAndSwapN mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 4);
format %{
"cmpxchg_weak $mem, $oldval, $newval\t# (narrow oop, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapN"
%}
ins_encode %{
__ cmpxchg_weak(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::uint32,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapP(iRegINoSp res, indirect mem, iRegP oldval, iRegP newval)
%{
predicate(n->as_LoadStore()->barrier_data() == 0);
match(Set res (WeakCompareAndSwapP mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 2);
format %{
"cmpxchg_weak $mem, $oldval, $newval\t# (ptr, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapP"
%}
ins_encode %{
__ cmpxchg_weak(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapBAcq(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (WeakCompareAndSwapB mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 6);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg_weak_acq $mem, $oldval, $newval\t# (byte, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapBAcq"
%}
ins_encode %{
__ weak_cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int8,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapSAcq(iRegINoSp res, indirect mem, iRegI_R12 oldval, iRegI_R13 newval,
iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3, rFlagsReg cr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (WeakCompareAndSwapS mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 7);
effect(TEMP_DEF res, KILL cr, USE_KILL oldval, USE_KILL newval, TEMP tmp1, TEMP tmp2, TEMP tmp3);
format %{
"cmpxchg_weak_acq $mem, $oldval, $newval\t# (short, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapSAcq"
%}
ins_encode %{
__ weak_cmpxchg_narrow_value(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int16,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapIAcq(iRegINoSp res, indirect mem, iRegI oldval, iRegI newval)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (WeakCompareAndSwapI mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 2);
format %{
"cmpxchg_weak_acq $mem, $oldval, $newval\t# (int, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapIAcq"
%}
ins_encode %{
__ cmpxchg_weak(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int32,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapLAcq(iRegINoSp res, indirect mem, iRegL oldval, iRegL newval)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (WeakCompareAndSwapL mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 2);
format %{
"cmpxchg_weak_acq $mem, $oldval, $newval\t# (long, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapLAcq"
%}
ins_encode %{
__ cmpxchg_weak(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapNAcq(iRegINoSp res, indirect mem, iRegN oldval, iRegN newval)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set res (WeakCompareAndSwapN mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 4);
format %{
"cmpxchg_weak_acq $mem, $oldval, $newval\t# (narrow oop, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"# $res == 1 when success, #@weakCompareAndSwapNAcq"
%}
ins_encode %{
__ cmpxchg_weak(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::uint32,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct weakCompareAndSwapPAcq(iRegINoSp res, indirect mem, iRegP oldval, iRegP newval)
%{
predicate(needs_acquiring_load_reserved(n) && (n->as_LoadStore()->barrier_data() == 0));
match(Set res (WeakCompareAndSwapP mem (Binary oldval newval)));
ins_cost(LOAD_COST + STORE_COST + BRANCH_COST * 2 + ALU_COST * 2);
format %{
"cmpxchg_weak_acq $mem, $oldval, $newval\t# (ptr, weak) if $mem == $oldval then $mem <-- $newval\n\t"
"\t# $res == 1 when success, #@weakCompareAndSwapPAcq"
%}
ins_encode %{
__ cmpxchg_weak(as_Register($mem$$base), $oldval$$Register, $newval$$Register, Assembler::int64,
/*acquire*/ Assembler::aq, /*release*/ Assembler::rl, $res$$Register);
%}
ins_pipe(pipe_slow);
%}
instruct get_and_setI(indirect mem, iRegI newv, iRegINoSp prev)
%{
match(Set prev (GetAndSetI mem newv));
ins_cost(ALU_COST);
format %{ "atomic_xchgw $prev, $newv, [$mem]\t#@get_and_setI" %}
ins_encode %{
__ atomic_xchgw($prev$$Register, $newv$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_setL(indirect mem, iRegL newv, iRegLNoSp prev)
%{
match(Set prev (GetAndSetL mem newv));
ins_cost(ALU_COST);
format %{ "atomic_xchg $prev, $newv, [$mem]\t#@get_and_setL" %}
ins_encode %{
__ atomic_xchg($prev$$Register, $newv$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_setN(indirect mem, iRegN newv, iRegINoSp prev)
%{
match(Set prev (GetAndSetN mem newv));
ins_cost(ALU_COST);
format %{ "atomic_xchgwu $prev, $newv, [$mem]\t#@get_and_setN" %}
ins_encode %{
__ atomic_xchgwu($prev$$Register, $newv$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_setP(indirect mem, iRegP newv, iRegPNoSp prev)
%{
predicate(n->as_LoadStore()->barrier_data() == 0);
match(Set prev (GetAndSetP mem newv));
ins_cost(ALU_COST);
format %{ "atomic_xchg $prev, $newv, [$mem]\t#@get_and_setP" %}
ins_encode %{
__ atomic_xchg($prev$$Register, $newv$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_setIAcq(indirect mem, iRegI newv, iRegINoSp prev)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set prev (GetAndSetI mem newv));
ins_cost(ALU_COST);
format %{ "atomic_xchgw_acq $prev, $newv, [$mem]\t#@get_and_setIAcq" %}
ins_encode %{
__ atomic_xchgalw($prev$$Register, $newv$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_setLAcq(indirect mem, iRegL newv, iRegLNoSp prev)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set prev (GetAndSetL mem newv));
ins_cost(ALU_COST);
format %{ "atomic_xchg_acq $prev, $newv, [$mem]\t#@get_and_setLAcq" %}
ins_encode %{
__ atomic_xchgal($prev$$Register, $newv$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_setNAcq(indirect mem, iRegN newv, iRegINoSp prev)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set prev (GetAndSetN mem newv));
ins_cost(ALU_COST);
format %{ "atomic_xchgwu_acq $prev, $newv, [$mem]\t#@get_and_setNAcq" %}
ins_encode %{
__ atomic_xchgalwu($prev$$Register, $newv$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_setPAcq(indirect mem, iRegP newv, iRegPNoSp prev)
%{
predicate(needs_acquiring_load_reserved(n) && (n->as_LoadStore()->barrier_data() == 0));
match(Set prev (GetAndSetP mem newv));
ins_cost(ALU_COST);
format %{ "atomic_xchg_acq $prev, $newv, [$mem]\t#@get_and_setPAcq" %}
ins_encode %{
__ atomic_xchgal($prev$$Register, $newv$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addL(indirect mem, iRegLNoSp newval, iRegL incr)
%{
match(Set newval (GetAndAddL mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addL $newval, [$mem], $incr\t#@get_and_addL" %}
ins_encode %{
__ atomic_add($newval$$Register, $incr$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addL_no_res(indirect mem, Universe dummy, iRegL incr)
%{
predicate(n->as_LoadStore()->result_not_used());
match(Set dummy (GetAndAddL mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addL [$mem], $incr\t#@get_and_addL_no_res" %}
ins_encode %{
__ atomic_add(noreg, $incr$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addLi(indirect mem, iRegLNoSp newval, immLAdd incr)
%{
match(Set newval (GetAndAddL mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addL $newval, [$mem], $incr\t#@get_and_addLi" %}
ins_encode %{
__ atomic_add($newval$$Register, $incr$$constant, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addLi_no_res(indirect mem, Universe dummy, immLAdd incr)
%{
predicate(n->as_LoadStore()->result_not_used());
match(Set dummy (GetAndAddL mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addL [$mem], $incr\t#@get_and_addLi_no_res" %}
ins_encode %{
__ atomic_add(noreg, $incr$$constant, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addI(indirect mem, iRegINoSp newval, iRegIorL2I incr)
%{
match(Set newval (GetAndAddI mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addI $newval, [$mem], $incr\t#@get_and_addI" %}
ins_encode %{
__ atomic_addw($newval$$Register, $incr$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addI_no_res(indirect mem, Universe dummy, iRegIorL2I incr)
%{
predicate(n->as_LoadStore()->result_not_used());
match(Set dummy (GetAndAddI mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addI [$mem], $incr\t#@get_and_addI_no_res" %}
ins_encode %{
__ atomic_addw(noreg, $incr$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addIi(indirect mem, iRegINoSp newval, immIAdd incr)
%{
match(Set newval (GetAndAddI mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addI $newval, [$mem], $incr\t#@get_and_addIi" %}
ins_encode %{
__ atomic_addw($newval$$Register, $incr$$constant, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addIi_no_res(indirect mem, Universe dummy, immIAdd incr)
%{
predicate(n->as_LoadStore()->result_not_used());
match(Set dummy (GetAndAddI mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addI [$mem], $incr\t#@get_and_addIi_no_res" %}
ins_encode %{
__ atomic_addw(noreg, $incr$$constant, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addLAcq(indirect mem, iRegLNoSp newval, iRegL incr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set newval (GetAndAddL mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addL_acq $newval, [$mem], $incr\t#@get_and_addLAcq" %}
ins_encode %{
__ atomic_addal($newval$$Register, $incr$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addL_no_resAcq(indirect mem, Universe dummy, iRegL incr) %{
predicate(n->as_LoadStore()->result_not_used() && needs_acquiring_load_reserved(n));
match(Set dummy (GetAndAddL mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addL_acq [$mem], $incr\t#@get_and_addL_no_resAcq" %}
ins_encode %{
__ atomic_addal(noreg, $incr$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addLiAcq(indirect mem, iRegLNoSp newval, immLAdd incr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set newval (GetAndAddL mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addL_acq $newval, [$mem], $incr\t#@get_and_addLiAcq" %}
ins_encode %{
__ atomic_addal($newval$$Register, $incr$$constant, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addLi_no_resAcq(indirect mem, Universe dummy, immLAdd incr)
%{
predicate(n->as_LoadStore()->result_not_used() && needs_acquiring_load_reserved(n));
match(Set dummy (GetAndAddL mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addL_acq [$mem], $incr\t#@get_and_addLi_no_resAcq" %}
ins_encode %{
__ atomic_addal(noreg, $incr$$constant, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addIAcq(indirect mem, iRegINoSp newval, iRegIorL2I incr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set newval (GetAndAddI mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addI_acq $newval, [$mem], $incr\t#@get_and_addIAcq" %}
ins_encode %{
__ atomic_addalw($newval$$Register, $incr$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addI_no_resAcq(indirect mem, Universe dummy, iRegIorL2I incr)
%{
predicate(n->as_LoadStore()->result_not_used() && needs_acquiring_load_reserved(n));
match(Set dummy (GetAndAddI mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addI_acq [$mem], $incr\t#@get_and_addI_no_resAcq" %}
ins_encode %{
__ atomic_addalw(noreg, $incr$$Register, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addIiAcq(indirect mem, iRegINoSp newval, immIAdd incr)
%{
predicate(needs_acquiring_load_reserved(n));
match(Set newval (GetAndAddI mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addI_acq $newval, [$mem], $incr\t#@get_and_addIiAcq" %}
ins_encode %{
__ atomic_addalw($newval$$Register, $incr$$constant, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
instruct get_and_addIi_no_resAcq(indirect mem, Universe dummy, immIAdd incr)
%{
predicate(n->as_LoadStore()->result_not_used() && needs_acquiring_load_reserved(n));
match(Set dummy (GetAndAddI mem incr));
ins_cost(ALU_COST);
format %{ "get_and_addI_acq [$mem], $incr\t#@get_and_addIi_no_resAcq" %}
ins_encode %{
__ atomic_addalw(noreg, $incr$$constant, as_Register($mem$$base));
%}
ins_pipe(pipe_serial);
%}
// ============================================================================
// Arithmetic Instructions
//
// Integer Addition
// TODO
// these currently employ operations which do not set CR and hence are
// not flagged as killing CR but we would like to isolate the cases
// where we want to set flags from those where we don't. need to work
// out how to do that.
instruct addI_reg_reg(iRegINoSp dst, iRegIorL2I src1, iRegIorL2I src2) %{
match(Set dst (AddI src1 src2));
ins_cost(ALU_COST);
format %{ "addw $dst, $src1, $src2\t#@addI_reg_reg" %}
ins_encode %{
__ addw(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
instruct addI_reg_imm(iRegINoSp dst, iRegIorL2I src1, immIAdd src2) %{
match(Set dst (AddI src1 src2));
ins_cost(ALU_COST);
format %{ "addiw $dst, $src1, $src2\t#@addI_reg_imm" %}
ins_encode %{
int32_t con = (int32_t)$src2$$constant;
__ addiw(as_Register($dst$$reg),
as_Register($src1$$reg),
$src2$$constant);
%}
ins_pipe(ialu_reg_imm);
%}
instruct addI_reg_imm_l2i(iRegINoSp dst, iRegL src1, immIAdd src2) %{
match(Set dst (AddI (ConvL2I src1) src2));
ins_cost(ALU_COST);
format %{ "addiw $dst, $src1, $src2\t#@addI_reg_imm_l2i" %}
ins_encode %{
__ addiw(as_Register($dst$$reg),
as_Register($src1$$reg),
$src2$$constant);
%}
ins_pipe(ialu_reg_imm);
%}
// Pointer Addition
instruct addP_reg_reg(iRegPNoSp dst, iRegP src1, iRegL src2) %{
match(Set dst (AddP src1 src2));
ins_cost(ALU_COST);
format %{ "add $dst, $src1, $src2\t# ptr, #@addP_reg_reg" %}
ins_encode %{
__ add(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// If we shift more than 32 bits, we need not convert I2L.
instruct lShiftL_regI_immGE32(iRegLNoSp dst, iRegI src, uimmI6_ge32 scale) %{
match(Set dst (LShiftL (ConvI2L src) scale));
ins_cost(ALU_COST);
format %{ "slli $dst, $src, $scale & 63\t#@lShiftL_regI_immGE32" %}
ins_encode %{
__ slli(as_Register($dst$$reg), as_Register($src$$reg), $scale$$constant & 63);
%}
ins_pipe(ialu_reg_shift);
%}
// Pointer Immediate Addition
// n.b. this needs to be more expensive than using an indirect memory
// operand
instruct addP_reg_imm(iRegPNoSp dst, iRegP src1, immLAdd src2) %{
match(Set dst (AddP src1 src2));
ins_cost(ALU_COST);
format %{ "addi $dst, $src1, $src2\t# ptr, #@addP_reg_imm" %}
ins_encode %{
// src2 is imm, so actually call the addi
__ add(as_Register($dst$$reg),
as_Register($src1$$reg),
$src2$$constant);
%}
ins_pipe(ialu_reg_imm);
%}
// Long Addition
instruct addL_reg_reg(iRegLNoSp dst, iRegL src1, iRegL src2) %{
match(Set dst (AddL src1 src2));
ins_cost(ALU_COST);
format %{ "add $dst, $src1, $src2\t#@addL_reg_reg" %}
ins_encode %{
__ add(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// No constant pool entries requiredLong Immediate Addition.
instruct addL_reg_imm(iRegLNoSp dst, iRegL src1, immLAdd src2) %{
match(Set dst (AddL src1 src2));
ins_cost(ALU_COST);
format %{ "addi $dst, $src1, $src2\t#@addL_reg_imm" %}
ins_encode %{
// src2 is imm, so actually call the addi
__ add(as_Register($dst$$reg),
as_Register($src1$$reg),
$src2$$constant);
%}
ins_pipe(ialu_reg_imm);
%}
// Integer Subtraction
instruct subI_reg_reg(iRegINoSp dst, iRegIorL2I src1, iRegIorL2I src2) %{
match(Set dst (SubI src1 src2));
ins_cost(ALU_COST);
format %{ "subw $dst, $src1, $src2\t#@subI_reg_reg" %}
ins_encode %{
__ subw(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// Immediate Subtraction
instruct subI_reg_imm(iRegINoSp dst, iRegIorL2I src1, immISub src2) %{
match(Set dst (SubI src1 src2));
ins_cost(ALU_COST);
format %{ "addiw $dst, $src1, -$src2\t#@subI_reg_imm" %}
ins_encode %{
// src2 is imm, so actually call the addiw
__ subw(as_Register($dst$$reg),
as_Register($src1$$reg),
$src2$$constant);
%}
ins_pipe(ialu_reg_imm);
%}
// Long Subtraction
instruct subL_reg_reg(iRegLNoSp dst, iRegL src1, iRegL src2) %{
match(Set dst (SubL src1 src2));
ins_cost(ALU_COST);
format %{ "sub $dst, $src1, $src2\t#@subL_reg_reg" %}
ins_encode %{
__ sub(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// No constant pool entries requiredLong Immediate Subtraction.
instruct subL_reg_imm(iRegLNoSp dst, iRegL src1, immLSub src2) %{
match(Set dst (SubL src1 src2));
ins_cost(ALU_COST);
format %{ "addi $dst, $src1, -$src2\t#@subL_reg_imm" %}
ins_encode %{
// src2 is imm, so actually call the addi
__ sub(as_Register($dst$$reg),
as_Register($src1$$reg),
$src2$$constant);
%}
ins_pipe(ialu_reg_imm);
%}
// Integer Negation (special case for sub)
instruct negI_reg(iRegINoSp dst, iRegIorL2I src, immI0 zero) %{
match(Set dst (SubI zero src));
ins_cost(ALU_COST);
format %{ "subw $dst, x0, $src\t# int, #@negI_reg" %}
ins_encode %{
// actually call the subw
__ negw(as_Register($dst$$reg),
as_Register($src$$reg));
%}
ins_pipe(ialu_reg);
%}
// Long Negation
instruct negL_reg(iRegLNoSp dst, iRegL src, immL0 zero) %{
match(Set dst (SubL zero src));
ins_cost(ALU_COST);
format %{ "sub $dst, x0, $src\t# long, #@negL_reg" %}
ins_encode %{
// actually call the sub
__ neg(as_Register($dst$$reg),
as_Register($src$$reg));
%}
ins_pipe(ialu_reg);
%}
// Integer Multiply
instruct mulI(iRegINoSp dst, iRegIorL2I src1, iRegIorL2I src2) %{
match(Set dst (MulI src1 src2));
ins_cost(IMUL_COST);
format %{ "mulw $dst, $src1, $src2\t#@mulI" %}
//this means 2 word multi, and no sign extend to 64 bits
ins_encode %{
// riscv64 mulw will sign-extension to high 32 bits in dst reg
__ mulw(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(imul_reg_reg);
%}
// Long Multiply
instruct mulL(iRegLNoSp dst, iRegL src1, iRegL src2) %{
match(Set dst (MulL src1 src2));
ins_cost(IMUL_COST);
format %{ "mul $dst, $src1, $src2\t#@mulL" %}
ins_encode %{
__ mul(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(lmul_reg_reg);
%}
instruct mulHiL_rReg(iRegLNoSp dst, iRegL src1, iRegL src2)
%{
match(Set dst (MulHiL src1 src2));
ins_cost(IMUL_COST);
format %{ "mulh $dst, $src1, $src2\t# mulhi, #@mulHiL_rReg" %}
ins_encode %{
__ mulh(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(lmul_reg_reg);
%}
// Integer Divide
instruct divI(iRegINoSp dst, iRegIorL2I src1, iRegIorL2I src2) %{
match(Set dst (DivI src1 src2));
ins_cost(IDIVSI_COST);
format %{ "divw $dst, $src1, $src2\t#@divI"%}
ins_encode(riscv_enc_divw(dst, src1, src2));
ins_pipe(idiv_reg_reg);
%}
instruct signExtract(iRegINoSp dst, iRegIorL2I src1, immI_31 div1, immI_31 div2) %{
match(Set dst (URShiftI (RShiftI src1 div1) div2));
ins_cost(ALU_COST);
format %{ "srliw $dst, $src1, $div1\t# int signExtract, #@signExtract" %}
ins_encode %{
__ srliw(as_Register($dst$$reg), as_Register($src1$$reg), 31);
%}
ins_pipe(ialu_reg_shift);
%}
// Long Divide
instruct divL(iRegLNoSp dst, iRegL src1, iRegL src2) %{
match(Set dst (DivL src1 src2));
ins_cost(IDIVDI_COST);
format %{ "div $dst, $src1, $src2\t#@divL" %}
ins_encode(riscv_enc_div(dst, src1, src2));
ins_pipe(ldiv_reg_reg);
%}
instruct signExtractL(iRegLNoSp dst, iRegL src1, immI_63 div1, immI_63 div2) %{
match(Set dst (URShiftL (RShiftL src1 div1) div2));
ins_cost(ALU_COST);
format %{ "srli $dst, $src1, $div1\t# long signExtract, #@signExtractL" %}
ins_encode %{
__ srli(as_Register($dst$$reg), as_Register($src1$$reg), 63);
%}
ins_pipe(ialu_reg_shift);
%}
// Integer Remainder
instruct modI(iRegINoSp dst, iRegIorL2I src1, iRegIorL2I src2) %{
match(Set dst (ModI src1 src2));
ins_cost(IDIVSI_COST);
format %{ "remw $dst, $src1, $src2\t#@modI" %}
ins_encode(riscv_enc_modw(dst, src1, src2));
ins_pipe(ialu_reg_reg);
%}
// Long Remainder
instruct modL(iRegLNoSp dst, iRegL src1, iRegL src2) %{
match(Set dst (ModL src1 src2));
ins_cost(IDIVDI_COST);
format %{ "rem $dst, $src1, $src2\t#@modL" %}
ins_encode(riscv_enc_mod(dst, src1, src2));
ins_pipe(ialu_reg_reg);
%}
// Integer Shifts
// Shift Left Register
// In RV64I, only the low 5 bits of src2 are considered for the shift amount
instruct lShiftI_reg_reg(iRegINoSp dst, iRegIorL2I src1, iRegIorL2I src2) %{
match(Set dst (LShiftI src1 src2));
ins_cost(ALU_COST);
format %{ "sllw $dst, $src1, $src2\t#@lShiftI_reg_reg" %}
ins_encode %{
__ sllw(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg_vshift);
%}
// Shift Left Immediate
instruct lShiftI_reg_imm(iRegINoSp dst, iRegIorL2I src1, immI src2) %{
match(Set dst (LShiftI src1 src2));
ins_cost(ALU_COST);
format %{ "slliw $dst, $src1, ($src2 & 0x1f)\t#@lShiftI_reg_imm" %}
ins_encode %{
// the shift amount is encoded in the lower
// 5 bits of the I-immediate field for RV32I
__ slliw(as_Register($dst$$reg),
as_Register($src1$$reg),
(unsigned) $src2$$constant & 0x1f);
%}
ins_pipe(ialu_reg_shift);
%}
// Shift Right Logical Register
// In RV64I, only the low 5 bits of src2 are considered for the shift amount
instruct urShiftI_reg_reg(iRegINoSp dst, iRegIorL2I src1, iRegIorL2I src2) %{
match(Set dst (URShiftI src1 src2));
ins_cost(ALU_COST);
format %{ "srlw $dst, $src1, $src2\t#@urShiftI_reg_reg" %}
ins_encode %{
__ srlw(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg_vshift);
%}
// Shift Right Logical Immediate
instruct urShiftI_reg_imm(iRegINoSp dst, iRegIorL2I src1, immI src2) %{
match(Set dst (URShiftI src1 src2));
ins_cost(ALU_COST);
format %{ "srliw $dst, $src1, ($src2 & 0x1f)\t#@urShiftI_reg_imm" %}
ins_encode %{
// the shift amount is encoded in the lower
// 6 bits of the I-immediate field for RV64I
__ srliw(as_Register($dst$$reg),
as_Register($src1$$reg),
(unsigned) $src2$$constant & 0x1f);
%}
ins_pipe(ialu_reg_shift);
%}
// Shift Right Arithmetic Register
// In RV64I, only the low 5 bits of src2 are considered for the shift amount
instruct rShiftI_reg_reg(iRegINoSp dst, iRegIorL2I src1, iRegIorL2I src2) %{
match(Set dst (RShiftI src1 src2));
ins_cost(ALU_COST);
format %{ "sraw $dst, $src1, $src2\t#@rShiftI_reg_reg" %}
ins_encode %{
// riscv will sign-ext dst high 32 bits
__ sraw(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg_vshift);
%}
// Shift Right Arithmetic Immediate
instruct rShiftI_reg_imm(iRegINoSp dst, iRegIorL2I src1, immI src2) %{
match(Set dst (RShiftI src1 src2));
ins_cost(ALU_COST);
format %{ "sraiw $dst, $src1, ($src2 & 0x1f)\t#@rShiftI_reg_imm" %}
ins_encode %{
// riscv will sign-ext dst high 32 bits
__ sraiw(as_Register($dst$$reg),
as_Register($src1$$reg),
(unsigned) $src2$$constant & 0x1f);
%}
ins_pipe(ialu_reg_shift);
%}
// Long Shifts
// Shift Left Register
// In RV64I, only the low 6 bits of src2 are considered for the shift amount
instruct lShiftL_reg_reg(iRegLNoSp dst, iRegL src1, iRegIorL2I src2) %{
match(Set dst (LShiftL src1 src2));
ins_cost(ALU_COST);
format %{ "sll $dst, $src1, $src2\t#@lShiftL_reg_reg" %}
ins_encode %{
__ sll(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg_vshift);
%}
// Shift Left Immediate
instruct lShiftL_reg_imm(iRegLNoSp dst, iRegL src1, immI src2) %{
match(Set dst (LShiftL src1 src2));
ins_cost(ALU_COST);
format %{ "slli $dst, $src1, ($src2 & 0x3f)\t#@lShiftL_reg_imm" %}
ins_encode %{
// the shift amount is encoded in the lower
// 6 bits of the I-immediate field for RV64I
__ slli(as_Register($dst$$reg),
as_Register($src1$$reg),
(unsigned) $src2$$constant & 0x3f);
%}
ins_pipe(ialu_reg_shift);
%}
// Shift Right Logical Register
// In RV64I, only the low 6 bits of src2 are considered for the shift amount
instruct urShiftL_reg_reg(iRegLNoSp dst, iRegL src1, iRegIorL2I src2) %{
match(Set dst (URShiftL src1 src2));
ins_cost(ALU_COST);
format %{ "srl $dst, $src1, $src2\t#@urShiftL_reg_reg" %}
ins_encode %{
__ srl(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg_vshift);
%}
// Shift Right Logical Immediate
instruct urShiftL_reg_imm(iRegLNoSp dst, iRegL src1, immI src2) %{
match(Set dst (URShiftL src1 src2));
ins_cost(ALU_COST);
format %{ "srli $dst, $src1, ($src2 & 0x3f)\t#@urShiftL_reg_imm" %}
ins_encode %{
// the shift amount is encoded in the lower
// 6 bits of the I-immediate field for RV64I
__ srli(as_Register($dst$$reg),
as_Register($src1$$reg),
(unsigned) $src2$$constant & 0x3f);
%}
ins_pipe(ialu_reg_shift);
%}
// A special-case pattern for card table stores.
instruct urShiftP_reg_imm(iRegLNoSp dst, iRegP src1, immI src2) %{
match(Set dst (URShiftL (CastP2X src1) src2));
ins_cost(ALU_COST);
format %{ "srli $dst, p2x($src1), ($src2 & 0x3f)\t#@urShiftP_reg_imm" %}
ins_encode %{
// the shift amount is encoded in the lower
// 6 bits of the I-immediate field for RV64I
__ srli(as_Register($dst$$reg),
as_Register($src1$$reg),
(unsigned) $src2$$constant & 0x3f);
%}
ins_pipe(ialu_reg_shift);
%}
// Shift Right Arithmetic Register
// In RV64I, only the low 6 bits of src2 are considered for the shift amount
instruct rShiftL_reg_reg(iRegLNoSp dst, iRegL src1, iRegIorL2I src2) %{
match(Set dst (RShiftL src1 src2));
ins_cost(ALU_COST);
format %{ "sra $dst, $src1, $src2\t#@rShiftL_reg_reg" %}
ins_encode %{
__ sra(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg_vshift);
%}
// Shift Right Arithmetic Immediate
instruct rShiftL_reg_imm(iRegLNoSp dst, iRegL src1, immI src2) %{
match(Set dst (RShiftL src1 src2));
ins_cost(ALU_COST);
format %{ "srai $dst, $src1, ($src2 & 0x3f)\t#@rShiftL_reg_imm" %}
ins_encode %{
// the shift amount is encoded in the lower
// 6 bits of the I-immediate field for RV64I
__ srai(as_Register($dst$$reg),
as_Register($src1$$reg),
(unsigned) $src2$$constant & 0x3f);
%}
ins_pipe(ialu_reg_shift);
%}
instruct regI_not_reg(iRegINoSp dst, iRegI src1, immI_M1 m1) %{
match(Set dst (XorI src1 m1));
ins_cost(ALU_COST);
format %{ "xori $dst, $src1, -1\t#@regI_not_reg" %}
ins_encode %{
__ xori(as_Register($dst$$reg), as_Register($src1$$reg), -1);
%}
ins_pipe(ialu_reg);
%}
instruct regL_not_reg(iRegLNoSp dst, iRegL src1, immL_M1 m1) %{
match(Set dst (XorL src1 m1));
ins_cost(ALU_COST);
format %{ "xori $dst, $src1, -1\t#@regL_not_reg" %}
ins_encode %{
__ xori(as_Register($dst$$reg), as_Register($src1$$reg), -1);
%}
ins_pipe(ialu_reg);
%}
// ============================================================================
// Floating Point Arithmetic Instructions
instruct addF_reg_reg(fRegF dst, fRegF src1, fRegF src2) %{
match(Set dst (AddF src1 src2));
ins_cost(FMUL_SINGLE_COST);
format %{ "fadd.s $dst, $src1, $src2\t#@addF_reg_reg" %}
ins_encode %{
__ fadd_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg));
%}
ins_pipe(fp_dop_reg_reg_s);
%}
instruct addD_reg_reg(fRegD dst, fRegD src1, fRegD src2) %{
match(Set dst (AddD src1 src2));
ins_cost(FMUL_DOUBLE_COST);
format %{ "fadd.d $dst, $src1, $src2\t#@addD_reg_reg" %}
ins_encode %{
__ fadd_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg));
%}
ins_pipe(fp_dop_reg_reg_d);
%}
instruct subF_reg_reg(fRegF dst, fRegF src1, fRegF src2) %{
match(Set dst (SubF src1 src2));
ins_cost(FMUL_SINGLE_COST);
format %{ "fsub.s $dst, $src1, $src2\t#@subF_reg_reg" %}
ins_encode %{
__ fsub_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg));
%}
ins_pipe(fp_dop_reg_reg_s);
%}
instruct subD_reg_reg(fRegD dst, fRegD src1, fRegD src2) %{
match(Set dst (SubD src1 src2));
ins_cost(FMUL_DOUBLE_COST);
format %{ "fsub.d $dst, $src1, $src2\t#@subD_reg_reg" %}
ins_encode %{
__ fsub_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg));
%}
ins_pipe(fp_dop_reg_reg_d);
%}
instruct mulF_reg_reg(fRegF dst, fRegF src1, fRegF src2) %{
match(Set dst (MulF src1 src2));
ins_cost(FMUL_SINGLE_COST);
format %{ "fmul.s $dst, $src1, $src2\t#@mulF_reg_reg" %}
ins_encode %{
__ fmul_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg));
%}
ins_pipe(fp_dop_reg_reg_s);
%}
instruct mulD_reg_reg(fRegD dst, fRegD src1, fRegD src2) %{
match(Set dst (MulD src1 src2));
ins_cost(FMUL_DOUBLE_COST);
format %{ "fmul.d $dst, $src1, $src2\t#@mulD_reg_reg" %}
ins_encode %{
__ fmul_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg));
%}
ins_pipe(fp_dop_reg_reg_d);
%}
// src1 * src2 + src3
instruct maddF_reg_reg(fRegF dst, fRegF src1, fRegF src2, fRegF src3) %{
predicate(UseFMA);
match(Set dst (FmaF src3 (Binary src1 src2)));
ins_cost(FMUL_SINGLE_COST);
format %{ "fmadd.s $dst, $src1, $src2, $src3\t#@maddF_reg_reg" %}
ins_encode %{
__ fmadd_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg),
as_FloatRegister($src3$$reg));
%}
ins_pipe(pipe_class_default);
%}
// src1 * src2 + src3
instruct maddD_reg_reg(fRegD dst, fRegD src1, fRegD src2, fRegD src3) %{
predicate(UseFMA);
match(Set dst (FmaD src3 (Binary src1 src2)));
ins_cost(FMUL_DOUBLE_COST);
format %{ "fmadd.d $dst, $src1, $src2, $src3\t#@maddD_reg_reg" %}
ins_encode %{
__ fmadd_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg),
as_FloatRegister($src3$$reg));
%}
ins_pipe(pipe_class_default);
%}
// src1 * src2 - src3
instruct msubF_reg_reg(fRegF dst, fRegF src1, fRegF src2, fRegF src3) %{
predicate(UseFMA);
match(Set dst (FmaF (NegF src3) (Binary src1 src2)));
ins_cost(FMUL_SINGLE_COST);
format %{ "fmsub.s $dst, $src1, $src2, $src3\t#@msubF_reg_reg" %}
ins_encode %{
__ fmsub_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg),
as_FloatRegister($src3$$reg));
%}
ins_pipe(pipe_class_default);
%}
// src1 * src2 - src3
instruct msubD_reg_reg(fRegD dst, fRegD src1, fRegD src2, fRegD src3) %{
predicate(UseFMA);
match(Set dst (FmaD (NegD src3) (Binary src1 src2)));
ins_cost(FMUL_DOUBLE_COST);
format %{ "fmsub.d $dst, $src1, $src2, $src3\t#@msubD_reg_reg" %}
ins_encode %{
__ fmsub_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg),
as_FloatRegister($src3$$reg));
%}
ins_pipe(pipe_class_default);
%}
// -src1 * src2 + src3
instruct nmsubF_reg_reg(fRegF dst, fRegF src1, fRegF src2, fRegF src3) %{
predicate(UseFMA);
match(Set dst (FmaF src3 (Binary (NegF src1) src2)));
match(Set dst (FmaF src3 (Binary src1 (NegF src2))));
ins_cost(FMUL_SINGLE_COST);
format %{ "fnmsub.s $dst, $src1, $src2, $src3\t#@nmsubF_reg_reg" %}
ins_encode %{
__ fnmsub_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg),
as_FloatRegister($src3$$reg));
%}
ins_pipe(pipe_class_default);
%}
// -src1 * src2 + src3
instruct nmsubD_reg_reg(fRegD dst, fRegD src1, fRegD src2, fRegD src3) %{
predicate(UseFMA);
match(Set dst (FmaD src3 (Binary (NegD src1) src2)));
match(Set dst (FmaD src3 (Binary src1 (NegD src2))));
ins_cost(FMUL_DOUBLE_COST);
format %{ "fnmsub.d $dst, $src1, $src2, $src3\t#@nmsubD_reg_reg" %}
ins_encode %{
__ fnmsub_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg),
as_FloatRegister($src3$$reg));
%}
ins_pipe(pipe_class_default);
%}
// -src1 * src2 - src3
instruct nmaddF_reg_reg(fRegF dst, fRegF src1, fRegF src2, fRegF src3) %{
predicate(UseFMA);
match(Set dst (FmaF (NegF src3) (Binary (NegF src1) src2)));
match(Set dst (FmaF (NegF src3) (Binary src1 (NegF src2))));
ins_cost(FMUL_SINGLE_COST);
format %{ "fnmadd.s $dst, $src1, $src2, $src3\t#@nmaddF_reg_reg" %}
ins_encode %{
__ fnmadd_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg),
as_FloatRegister($src3$$reg));
%}
ins_pipe(pipe_class_default);
%}
// -src1 * src2 - src3
instruct nmaddD_reg_reg(fRegD dst, fRegD src1, fRegD src2, fRegD src3) %{
predicate(UseFMA);
match(Set dst (FmaD (NegD src3) (Binary (NegD src1) src2)));
match(Set dst (FmaD (NegD src3) (Binary src1 (NegD src2))));
ins_cost(FMUL_DOUBLE_COST);
format %{ "fnmadd.d $dst, $src1, $src2, $src3\t#@nmaddD_reg_reg" %}
ins_encode %{
__ fnmadd_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg),
as_FloatRegister($src3$$reg));
%}
ins_pipe(pipe_class_default);
%}
// Math.max(FF)F
instruct maxF_reg_reg(fRegF dst, fRegF src1, fRegF src2, rFlagsReg cr) %{
match(Set dst (MaxF src1 src2));
effect(TEMP_DEF dst, KILL cr);
format %{ "maxF $dst, $src1, $src2" %}
ins_encode %{
__ minmax_FD(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg), as_FloatRegister($src2$$reg),
false /* is_double */, false /* is_min */);
%}
ins_pipe(fp_dop_reg_reg_s);
%}
// Math.min(FF)F
instruct minF_reg_reg(fRegF dst, fRegF src1, fRegF src2, rFlagsReg cr) %{
match(Set dst (MinF src1 src2));
effect(TEMP_DEF dst, KILL cr);
format %{ "minF $dst, $src1, $src2" %}
ins_encode %{
__ minmax_FD(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg), as_FloatRegister($src2$$reg),
false /* is_double */, true /* is_min */);
%}
ins_pipe(fp_dop_reg_reg_s);
%}
// Math.max(DD)D
instruct maxD_reg_reg(fRegD dst, fRegD src1, fRegD src2, rFlagsReg cr) %{
match(Set dst (MaxD src1 src2));
effect(TEMP_DEF dst, KILL cr);
format %{ "maxD $dst, $src1, $src2" %}
ins_encode %{
__ minmax_FD(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg), as_FloatRegister($src2$$reg),
true /* is_double */, false /* is_min */);
%}
ins_pipe(fp_dop_reg_reg_d);
%}
// Math.min(DD)D
instruct minD_reg_reg(fRegD dst, fRegD src1, fRegD src2, rFlagsReg cr) %{
match(Set dst (MinD src1 src2));
effect(TEMP_DEF dst, KILL cr);
format %{ "minD $dst, $src1, $src2" %}
ins_encode %{
__ minmax_FD(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg), as_FloatRegister($src2$$reg),
true /* is_double */, true /* is_min */);
%}
ins_pipe(fp_dop_reg_reg_d);
%}
// Float.isInfinite
instruct isIniniteF_reg_reg(iRegINoSp dst, fRegF src)
%{
match(Set dst (IsInfiniteF src));
format %{ "isInfinite $dst, $src" %}
ins_encode %{
__ fclass_s(as_Register($dst$$reg), as_FloatRegister($src$$reg));
__ andi(as_Register($dst$$reg), as_Register($dst$$reg), 0b10000001);
__ slt(as_Register($dst$$reg), zr, as_Register($dst$$reg));
%}
ins_pipe(fp_dop_reg_reg_s);
%}
// Double.isInfinite
instruct isInfiniteD_reg_reg(iRegINoSp dst, fRegD src)
%{
match(Set dst (IsInfiniteD src));
format %{ "isInfinite $dst, $src" %}
ins_encode %{
__ fclass_d(as_Register($dst$$reg), as_FloatRegister($src$$reg));
__ andi(as_Register($dst$$reg), as_Register($dst$$reg), 0b10000001);
__ slt(as_Register($dst$$reg), zr, as_Register($dst$$reg));
%}
ins_pipe(fp_dop_reg_reg_d);
%}
// Float.isFinite
instruct isFiniteF_reg_reg(iRegINoSp dst, fRegF src)
%{
match(Set dst (IsFiniteF src));
format %{ "isFinite $dst, $src" %}
ins_encode %{
__ fclass_s(as_Register($dst$$reg), as_FloatRegister($src$$reg));
__ andi(as_Register($dst$$reg), as_Register($dst$$reg), 0b0001111110);
__ slt(as_Register($dst$$reg), zr, as_Register($dst$$reg));
%}
ins_pipe(fp_dop_reg_reg_s);
%}
// Double.isFinite
instruct isFiniteD_reg_reg(iRegINoSp dst, fRegD src)
%{
match(Set dst (IsFiniteD src));
format %{ "isFinite $dst, $src" %}
ins_encode %{
__ fclass_d(as_Register($dst$$reg), as_FloatRegister($src$$reg));
__ andi(as_Register($dst$$reg), as_Register($dst$$reg), 0b0001111110);
__ slt(as_Register($dst$$reg), zr, as_Register($dst$$reg));
%}
ins_pipe(fp_dop_reg_reg_d);
%}
instruct divF_reg_reg(fRegF dst, fRegF src1, fRegF src2) %{
match(Set dst (DivF src1 src2));
ins_cost(FDIV_COST);
format %{ "fdiv.s $dst, $src1, $src2\t#@divF_reg_reg" %}
ins_encode %{
__ fdiv_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg));
%}
ins_pipe(fp_div_s);
%}
instruct divD_reg_reg(fRegD dst, fRegD src1, fRegD src2) %{
match(Set dst (DivD src1 src2));
ins_cost(FDIV_COST);
format %{ "fdiv.d $dst, $src1, $src2\t#@divD_reg_reg" %}
ins_encode %{
__ fdiv_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src1$$reg),
as_FloatRegister($src2$$reg));
%}
ins_pipe(fp_div_d);
%}
instruct negF_reg_reg(fRegF dst, fRegF src) %{
match(Set dst (NegF src));
ins_cost(XFER_COST);
format %{ "fsgnjn.s $dst, $src, $src\t#@negF_reg_reg" %}
ins_encode %{
__ fneg_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src$$reg));
%}
ins_pipe(fp_uop_s);
%}
instruct negD_reg_reg(fRegD dst, fRegD src) %{
match(Set dst (NegD src));
ins_cost(XFER_COST);
format %{ "fsgnjn.d $dst, $src, $src\t#@negD_reg_reg" %}
ins_encode %{
__ fneg_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src$$reg));
%}
ins_pipe(fp_uop_d);
%}
instruct absI_reg(iRegINoSp dst, iRegIorL2I src) %{
match(Set dst (AbsI src));
ins_cost(ALU_COST * 3);
format %{
"sraiw t0, $src, 0x1f\n\t"
"addw $dst, $src, t0\n\t"
"xorr $dst, $dst, t0\t#@absI_reg"
%}
ins_encode %{
__ sraiw(t0, as_Register($src$$reg), 0x1f);
__ addw(as_Register($dst$$reg), as_Register($src$$reg), t0);
__ xorr(as_Register($dst$$reg), as_Register($dst$$reg), t0);
%}
ins_pipe(ialu_reg_reg);
%}
instruct absL_reg(iRegLNoSp dst, iRegL src) %{
match(Set dst (AbsL src));
ins_cost(ALU_COST * 3);
format %{
"srai t0, $src, 0x3f\n\t"
"add $dst, $src, t0\n\t"
"xorr $dst, $dst, t0\t#@absL_reg"
%}
ins_encode %{
__ srai(t0, as_Register($src$$reg), 0x3f);
__ add(as_Register($dst$$reg), as_Register($src$$reg), t0);
__ xorr(as_Register($dst$$reg), as_Register($dst$$reg), t0);
%}
ins_pipe(ialu_reg_reg);
%}
instruct absF_reg(fRegF dst, fRegF src) %{
match(Set dst (AbsF src));
ins_cost(XFER_COST);
format %{ "fsgnjx.s $dst, $src, $src\t#@absF_reg" %}
ins_encode %{
__ fabs_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src$$reg));
%}
ins_pipe(fp_uop_s);
%}
instruct absD_reg(fRegD dst, fRegD src) %{
match(Set dst (AbsD src));
ins_cost(XFER_COST);
format %{ "fsgnjx.d $dst, $src, $src\t#@absD_reg" %}
ins_encode %{
__ fabs_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src$$reg));
%}
ins_pipe(fp_uop_d);
%}
instruct sqrtF_reg(fRegF dst, fRegF src) %{
match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
ins_cost(FSQRT_COST);
format %{ "fsqrt.s $dst, $src\t#@sqrtF_reg" %}
ins_encode %{
__ fsqrt_s(as_FloatRegister($dst$$reg),
as_FloatRegister($src$$reg));
%}
ins_pipe(fp_sqrt_s);
%}
instruct sqrtD_reg(fRegD dst, fRegD src) %{
match(Set dst (SqrtD src));
ins_cost(FSQRT_COST);
format %{ "fsqrt.d $dst, $src\t#@sqrtD_reg" %}
ins_encode %{
__ fsqrt_d(as_FloatRegister($dst$$reg),
as_FloatRegister($src$$reg));
%}
ins_pipe(fp_sqrt_d);
%}
// Arithmetic Instructions End
// ============================================================================
// Logical Instructions
// Register And
instruct andI_reg_reg(iRegINoSp dst, iRegI src1, iRegI src2) %{
match(Set dst (AndI src1 src2));
format %{ "andr $dst, $src1, $src2\t#@andI_reg_reg" %}
ins_cost(ALU_COST);
ins_encode %{
__ andr(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// Immediate And
instruct andI_reg_imm(iRegINoSp dst, iRegI src1, immIAdd src2) %{
match(Set dst (AndI src1 src2));
format %{ "andi $dst, $src1, $src2\t#@andI_reg_imm" %}
ins_cost(ALU_COST);
ins_encode %{
__ andi(as_Register($dst$$reg),
as_Register($src1$$reg),
(int32_t)($src2$$constant));
%}
ins_pipe(ialu_reg_imm);
%}
// Register Or
instruct orI_reg_reg(iRegINoSp dst, iRegI src1, iRegI src2) %{
match(Set dst (OrI src1 src2));
format %{ "orr $dst, $src1, $src2\t#@orI_reg_reg" %}
ins_cost(ALU_COST);
ins_encode %{
__ orr(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// Immediate Or
instruct orI_reg_imm(iRegINoSp dst, iRegI src1, immIAdd src2) %{
match(Set dst (OrI src1 src2));
format %{ "ori $dst, $src1, $src2\t#@orI_reg_imm" %}
ins_cost(ALU_COST);
ins_encode %{
__ ori(as_Register($dst$$reg),
as_Register($src1$$reg),
(int32_t)($src2$$constant));
%}
ins_pipe(ialu_reg_imm);
%}
// Register Xor
instruct xorI_reg_reg(iRegINoSp dst, iRegI src1, iRegI src2) %{
match(Set dst (XorI src1 src2));
format %{ "xorr $dst, $src1, $src2\t#@xorI_reg_reg" %}
ins_cost(ALU_COST);
ins_encode %{
__ xorr(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// Immediate Xor
instruct xorI_reg_imm(iRegINoSp dst, iRegI src1, immIAdd src2) %{
match(Set dst (XorI src1 src2));
format %{ "xori $dst, $src1, $src2\t#@xorI_reg_imm" %}
ins_cost(ALU_COST);
ins_encode %{
__ xori(as_Register($dst$$reg),
as_Register($src1$$reg),
(int32_t)($src2$$constant));
%}
ins_pipe(ialu_reg_imm);
%}
// Register And Long
instruct andL_reg_reg(iRegLNoSp dst, iRegL src1, iRegL src2) %{
match(Set dst (AndL src1 src2));
format %{ "andr $dst, $src1, $src2\t#@andL_reg_reg" %}
ins_cost(ALU_COST);
ins_encode %{
__ andr(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// Immediate And Long
instruct andL_reg_imm(iRegLNoSp dst, iRegL src1, immLAdd src2) %{
match(Set dst (AndL src1 src2));
format %{ "andi $dst, $src1, $src2\t#@andL_reg_imm" %}
ins_cost(ALU_COST);
ins_encode %{
__ andi(as_Register($dst$$reg),
as_Register($src1$$reg),
(int32_t)($src2$$constant));
%}
ins_pipe(ialu_reg_imm);
%}
// Register Or Long
instruct orL_reg_reg(iRegLNoSp dst, iRegL src1, iRegL src2) %{
match(Set dst (OrL src1 src2));
format %{ "orr $dst, $src1, $src2\t#@orL_reg_reg" %}
ins_cost(ALU_COST);
ins_encode %{
__ orr(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// Immediate Or Long
instruct orL_reg_imm(iRegLNoSp dst, iRegL src1, immLAdd src2) %{
match(Set dst (OrL src1 src2));
format %{ "ori $dst, $src1, $src2\t#@orL_reg_imm" %}
ins_cost(ALU_COST);
ins_encode %{
__ ori(as_Register($dst$$reg),
as_Register($src1$$reg),
(int32_t)($src2$$constant));
%}
ins_pipe(ialu_reg_imm);
%}
// Register Xor Long
instruct xorL_reg_reg(iRegLNoSp dst, iRegL src1, iRegL src2) %{
match(Set dst (XorL src1 src2));
format %{ "xorr $dst, $src1, $src2\t#@xorL_reg_reg" %}
ins_cost(ALU_COST);
ins_encode %{
__ xorr(as_Register($dst$$reg),
as_Register($src1$$reg),
as_Register($src2$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
// Immediate Xor Long
instruct xorL_reg_imm(iRegLNoSp dst, iRegL src1, immLAdd src2) %{
match(Set dst (XorL src1 src2));
ins_cost(ALU_COST);
format %{ "xori $dst, $src1, $src2\t#@xorL_reg_imm" %}
ins_encode %{
__ xori(as_Register($dst$$reg),
as_Register($src1$$reg),
(int32_t)($src2$$constant));
%}
ins_pipe(ialu_reg_imm);
%}
// ============================================================================
// BSWAP Instructions
instruct bytes_reverse_int(iRegINoSp dst, iRegIorL2I src, rFlagsReg cr) %{
match(Set dst (ReverseBytesI src));
effect(TEMP cr);
ins_cost(ALU_COST * 13);
format %{ "revb_w_w $dst, $src\t#@bytes_reverse_int" %}
ins_encode %{
__ revb_w_w(as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(ialu_reg);
%}
instruct bytes_reverse_long(iRegLNoSp dst, iRegL src, rFlagsReg cr) %{
match(Set dst (ReverseBytesL src));
effect(TEMP cr);
ins_cost(ALU_COST * 29);
format %{ "revb $dst, $src\t#@bytes_reverse_long" %}
ins_encode %{
__ revb(as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(ialu_reg);
%}
instruct bytes_reverse_unsigned_short(iRegINoSp dst, iRegIorL2I src) %{
match(Set dst (ReverseBytesUS src));
ins_cost(ALU_COST * 5);
format %{ "revb_h_h_u $dst, $src\t#@bytes_reverse_unsigned_short" %}
ins_encode %{
__ revb_h_h_u(as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(ialu_reg);
%}
instruct bytes_reverse_short(iRegINoSp dst, iRegIorL2I src) %{
match(Set dst (ReverseBytesS src));
ins_cost(ALU_COST * 5);
format %{ "revb_h_h $dst, $src\t#@bytes_reverse_short" %}
ins_encode %{
__ revb_h_h(as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(ialu_reg);
%}
// ============================================================================
// MemBar Instruction
instruct load_fence() %{
match(LoadFence);
ins_cost(ALU_COST);
format %{ "#@load_fence" %}
ins_encode %{
__ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
%}
ins_pipe(pipe_serial);
%}
instruct membar_acquire() %{
match(MemBarAcquire);
ins_cost(ALU_COST);
format %{ "#@membar_acquire\n\t"
"fence ir iorw" %}
ins_encode %{
__ block_comment("membar_acquire");
__ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
%}
ins_pipe(pipe_serial);
%}
instruct membar_acquire_lock() %{
match(MemBarAcquireLock);
ins_cost(0);
format %{ "#@membar_acquire_lock (elided)" %}
ins_encode %{
__ block_comment("membar_acquire_lock (elided)");
%}
ins_pipe(pipe_serial);
%}
instruct store_fence() %{
match(StoreFence);
ins_cost(ALU_COST);
format %{ "#@store_fence" %}
ins_encode %{
__ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
%}
ins_pipe(pipe_serial);
%}
instruct membar_release() %{
match(MemBarRelease);
ins_cost(ALU_COST);
format %{ "#@membar_release\n\t"
"fence iorw ow" %}
ins_encode %{
__ block_comment("membar_release");
__ membar(MacroAssembler::LoadStore | MacroAssembler::StoreStore);
%}
ins_pipe(pipe_serial);
%}
instruct membar_storestore() %{
match(MemBarStoreStore);
match(StoreStoreFence);
ins_cost(ALU_COST);
format %{ "MEMBAR-store-store\t#@membar_storestore" %}
ins_encode %{
__ membar(MacroAssembler::StoreStore);
%}
ins_pipe(pipe_serial);
%}
instruct membar_release_lock() %{
match(MemBarReleaseLock);
ins_cost(0);
format %{ "#@membar_release_lock (elided)" %}
ins_encode %{
__ block_comment("membar_release_lock (elided)");
%}
ins_pipe(pipe_serial);
%}
instruct membar_volatile() %{
match(MemBarVolatile);
ins_cost(ALU_COST);
format %{ "#@membar_volatile\n\t"
"fence iorw iorw"%}
ins_encode %{
__ block_comment("membar_volatile");
__ membar(MacroAssembler::StoreLoad);
%}
ins_pipe(pipe_serial);
%}
// ============================================================================
// Cast Instructions (Java-level type cast)
instruct castX2P(iRegPNoSp dst, iRegL src) %{
match(Set dst (CastX2P src));
ins_cost(ALU_COST);
format %{ "mv $dst, $src\t# long -> ptr, #@castX2P" %}
ins_encode %{
if ($dst$$reg != $src$$reg) {
__ mv(as_Register($dst$$reg), as_Register($src$$reg));
}
%}
ins_pipe(ialu_reg);
%}
instruct castP2X(iRegLNoSp dst, iRegP src) %{
match(Set dst (CastP2X src));
ins_cost(ALU_COST);
format %{ "mv $dst, $src\t# ptr -> long, #@castP2X" %}
ins_encode %{
if ($dst$$reg != $src$$reg) {
__ mv(as_Register($dst$$reg), as_Register($src$$reg));
}
%}
ins_pipe(ialu_reg);
%}
instruct castPP(iRegPNoSp dst)
%{
match(Set dst (CastPP dst));
ins_cost(0);
size(0);
format %{ "# castPP of $dst, #@castPP" %}
ins_encode(/* empty encoding */);
ins_pipe(pipe_class_empty);
%}
instruct castLL(iRegL dst)
%{
match(Set dst (CastLL dst));
size(0);
format %{ "# castLL of $dst, #@castLL" %}
ins_encode(/* empty encoding */);
ins_cost(0);
ins_pipe(pipe_class_empty);
%}
instruct castII(iRegI dst)
%{
match(Set dst (CastII dst));
size(0);
format %{ "# castII of $dst, #@castII" %}
ins_encode(/* empty encoding */);
ins_cost(0);
ins_pipe(pipe_class_empty);
%}
instruct checkCastPP(iRegPNoSp dst)
%{
match(Set dst (CheckCastPP dst));
size(0);
ins_cost(0);
format %{ "# checkcastPP of $dst, #@checkCastPP" %}
ins_encode(/* empty encoding */);
ins_pipe(pipe_class_empty);
%}
instruct castFF(fRegF dst)
%{
match(Set dst (CastFF dst));
size(0);
format %{ "# castFF of $dst" %}
ins_encode(/* empty encoding */);
ins_cost(0);
ins_pipe(pipe_class_empty);
%}
instruct castDD(fRegD dst)
%{
match(Set dst (CastDD dst));
size(0);
format %{ "# castDD of $dst" %}
ins_encode(/* empty encoding */);
ins_cost(0);
ins_pipe(pipe_class_empty);
%}
instruct castVV(vReg dst)
%{
match(Set dst (CastVV dst));
size(0);
format %{ "# castVV of $dst" %}
ins_encode(/* empty encoding */);
ins_cost(0);
ins_pipe(pipe_class_empty);
%}
// ============================================================================
// Convert Instructions
// int to bool
instruct convI2Bool(iRegINoSp dst, iRegI src)
%{
match(Set dst (Conv2B src));
ins_cost(ALU_COST);
format %{ "snez $dst, $src\t#@convI2Bool" %}
ins_encode %{
__ snez(as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(ialu_reg);
%}
// pointer to bool
instruct convP2Bool(iRegINoSp dst, iRegP src)
%{
match(Set dst (Conv2B src));
ins_cost(ALU_COST);
format %{ "snez $dst, $src\t#@convP2Bool" %}
ins_encode %{
__ snez(as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(ialu_reg);
%}
// int <-> long
instruct convI2L_reg_reg(iRegLNoSp dst, iRegIorL2I src)
%{
match(Set dst (ConvI2L src));
ins_cost(ALU_COST);
format %{ "addw $dst, $src, zr\t#@convI2L_reg_reg" %}
ins_encode %{
__ addw(as_Register($dst$$reg), as_Register($src$$reg), zr);
%}
ins_pipe(ialu_reg);
%}
instruct convL2I_reg(iRegINoSp dst, iRegL src) %{
match(Set dst (ConvL2I src));
ins_cost(ALU_COST);
format %{ "addw $dst, $src, zr\t#@convL2I_reg" %}
ins_encode %{
__ addw(as_Register($dst$$reg), as_Register($src$$reg), zr);
%}
ins_pipe(ialu_reg);
%}
// int to unsigned long (Zero-extend)
instruct convI2UL_reg_reg(iRegLNoSp dst, iRegIorL2I src, immL_32bits mask)
%{
match(Set dst (AndL (ConvI2L src) mask));
ins_cost(ALU_COST * 2);
format %{ "zero_extend $dst, $src, 32\t# i2ul, #@convI2UL_reg_reg" %}
ins_encode %{
__ zero_extend(as_Register($dst$$reg), as_Register($src$$reg), 32);
%}
ins_pipe(ialu_reg_shift);
%}
// float <-> double
instruct convF2D_reg(fRegD dst, fRegF src) %{
match(Set dst (ConvF2D src));
ins_cost(XFER_COST);
format %{ "fcvt.d.s $dst, $src\t#@convF2D_reg" %}
ins_encode %{
__ fcvt_d_s(as_FloatRegister($dst$$reg), as_FloatRegister($src$$reg));
%}
ins_pipe(fp_f2d);
%}
instruct convD2F_reg(fRegF dst, fRegD src) %{
match(Set dst (ConvD2F src));
ins_cost(XFER_COST);
format %{ "fcvt.s.d $dst, $src\t#@convD2F_reg" %}
ins_encode %{
__ fcvt_s_d(as_FloatRegister($dst$$reg), as_FloatRegister($src$$reg));
%}
ins_pipe(fp_d2f);
%}
// float <-> int
instruct convF2I_reg_reg(iRegINoSp dst, fRegF src) %{
match(Set dst (ConvF2I src));
ins_cost(XFER_COST);
format %{ "fcvt.w.s $dst, $src\t#@convF2I_reg_reg" %}
ins_encode %{
__ fcvt_w_s_safe($dst$$Register, $src$$FloatRegister);
%}
ins_pipe(fp_f2i);
%}
instruct convI2F_reg_reg(fRegF dst, iRegIorL2I src) %{
match(Set dst (ConvI2F src));
ins_cost(XFER_COST);
format %{ "fcvt.s.w $dst, $src\t#@convI2F_reg_reg" %}
ins_encode %{
__ fcvt_s_w(as_FloatRegister($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(fp_i2f);
%}
// float <-> long
instruct convF2L_reg_reg(iRegLNoSp dst, fRegF src) %{
match(Set dst (ConvF2L src));
ins_cost(XFER_COST);
format %{ "fcvt.l.s $dst, $src\t#@convF2L_reg_reg" %}
ins_encode %{
__ fcvt_l_s_safe($dst$$Register, $src$$FloatRegister);
%}
ins_pipe(fp_f2l);
%}
instruct convL2F_reg_reg(fRegF dst, iRegL src) %{
match(Set dst (ConvL2F src));
ins_cost(XFER_COST);
format %{ "fcvt.s.l $dst, $src\t#@convL2F_reg_reg" %}
ins_encode %{
__ fcvt_s_l(as_FloatRegister($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(fp_l2f);
%}
// double <-> int
instruct convD2I_reg_reg(iRegINoSp dst, fRegD src) %{
match(Set dst (ConvD2I src));
ins_cost(XFER_COST);
format %{ "fcvt.w.d $dst, $src\t#@convD2I_reg_reg" %}
ins_encode %{
__ fcvt_w_d_safe($dst$$Register, $src$$FloatRegister);
%}
ins_pipe(fp_d2i);
%}
instruct convI2D_reg_reg(fRegD dst, iRegIorL2I src) %{
match(Set dst (ConvI2D src));
ins_cost(XFER_COST);
format %{ "fcvt.d.w $dst, $src\t#@convI2D_reg_reg" %}
ins_encode %{
__ fcvt_d_w(as_FloatRegister($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(fp_i2d);
%}
// double <-> long
instruct convD2L_reg_reg(iRegLNoSp dst, fRegD src) %{
match(Set dst (ConvD2L src));
ins_cost(XFER_COST);
format %{ "fcvt.l.d $dst, $src\t#@convD2L_reg_reg" %}
ins_encode %{
__ fcvt_l_d_safe($dst$$Register, $src$$FloatRegister);
%}
ins_pipe(fp_d2l);
%}
instruct convL2D_reg_reg(fRegD dst, iRegL src) %{
match(Set dst (ConvL2D src));
ins_cost(XFER_COST);
format %{ "fcvt.d.l $dst, $src\t#@convL2D_reg_reg" %}
ins_encode %{
__ fcvt_d_l(as_FloatRegister($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(fp_l2d);
%}
// Convert oop into int for vectors alignment masking
instruct convP2I(iRegINoSp dst, iRegP src) %{
match(Set dst (ConvL2I (CastP2X src)));
ins_cost(ALU_COST * 2);
format %{ "zero_extend $dst, $src, 32\t# ptr -> int, #@convP2I" %}
ins_encode %{
__ zero_extend($dst$$Register, $src$$Register, 32);
%}
ins_pipe(ialu_reg);
%}
// Convert compressed oop into int for vectors alignment masking
// in case of 32bit oops (heap < 4Gb).
instruct convN2I(iRegINoSp dst, iRegN src)
%{
predicate(CompressedOops::shift() == 0);
match(Set dst (ConvL2I (CastP2X (DecodeN src))));
ins_cost(ALU_COST);
format %{ "mv $dst, $src\t# compressed ptr -> int, #@convN2I" %}
ins_encode %{
__ mv($dst$$Register, $src$$Register);
%}
ins_pipe(ialu_reg);
%}
// Convert oop pointer into compressed form
instruct encodeHeapOop(iRegNNoSp dst, iRegP src) %{
match(Set dst (EncodeP src));
ins_cost(ALU_COST);
format %{ "encode_heap_oop $dst, $src\t#@encodeHeapOop" %}
ins_encode %{
Register s = $src$$Register;
Register d = $dst$$Register;
__ encode_heap_oop(d, s);
%}
ins_pipe(ialu_reg);
%}
instruct decodeHeapOop(iRegPNoSp dst, iRegN src) %{
predicate(n->bottom_type()->is_ptr()->ptr() != TypePtr::NotNull &&
n->bottom_type()->is_ptr()->ptr() != TypePtr::Constant);
match(Set dst (DecodeN src));
ins_cost(0);
format %{ "decode_heap_oop $dst, $src\t#@decodeHeapOop" %}
ins_encode %{
Register s = $src$$Register;
Register d = $dst$$Register;
__ decode_heap_oop(d, s);
%}
ins_pipe(ialu_reg);
%}
instruct decodeHeapOop_not_null(iRegPNoSp dst, iRegN src) %{
predicate(n->bottom_type()->is_ptr()->ptr() == TypePtr::NotNull ||
n->bottom_type()->is_ptr()->ptr() == TypePtr::Constant);
match(Set dst (DecodeN src));
ins_cost(0);
format %{ "decode_heap_oop_not_null $dst, $src\t#@decodeHeapOop_not_null" %}
ins_encode %{
Register s = $src$$Register;
Register d = $dst$$Register;
__ decode_heap_oop_not_null(d, s);
%}
ins_pipe(ialu_reg);
%}
// Convert klass pointer into compressed form.
instruct encodeKlass_not_null(iRegNNoSp dst, iRegP src) %{
match(Set dst (EncodePKlass src));
ins_cost(ALU_COST);
format %{ "encode_klass_not_null $dst, $src\t#@encodeKlass_not_null" %}
ins_encode %{
Register src_reg = as_Register($src$$reg);
Register dst_reg = as_Register($dst$$reg);
__ encode_klass_not_null(dst_reg, src_reg, t0);
%}
ins_pipe(ialu_reg);
%}
instruct decodeKlass_not_null(iRegPNoSp dst, iRegN src, iRegPNoSp tmp) %{
match(Set dst (DecodeNKlass src));
effect(TEMP tmp);
ins_cost(ALU_COST);
format %{ "decode_klass_not_null $dst, $src\t#@decodeKlass_not_null" %}
ins_encode %{
Register src_reg = as_Register($src$$reg);
Register dst_reg = as_Register($dst$$reg);
Register tmp_reg = as_Register($tmp$$reg);
__ decode_klass_not_null(dst_reg, src_reg, tmp_reg);
%}
ins_pipe(ialu_reg);
%}
// stack <-> reg and reg <-> reg shuffles with no conversion
instruct MoveF2I_stack_reg(iRegINoSp dst, stackSlotF src) %{
match(Set dst (MoveF2I src));
effect(DEF dst, USE src);
ins_cost(LOAD_COST);
format %{ "lw $dst, $src\t#@MoveF2I_stack_reg" %}
ins_encode %{
__ lw(as_Register($dst$$reg), Address(sp, $src$$disp));
%}
ins_pipe(iload_reg_reg);
%}
instruct MoveI2F_stack_reg(fRegF dst, stackSlotI src) %{
match(Set dst (MoveI2F src));
effect(DEF dst, USE src);
ins_cost(LOAD_COST);
format %{ "flw $dst, $src\t#@MoveI2F_stack_reg" %}
ins_encode %{
__ flw(as_FloatRegister($dst$$reg), Address(sp, $src$$disp));
%}
ins_pipe(pipe_class_memory);
%}
instruct MoveD2L_stack_reg(iRegLNoSp dst, stackSlotD src) %{
match(Set dst (MoveD2L src));
effect(DEF dst, USE src);
ins_cost(LOAD_COST);
format %{ "ld $dst, $src\t#@MoveD2L_stack_reg" %}
ins_encode %{
__ ld(as_Register($dst$$reg), Address(sp, $src$$disp));
%}
ins_pipe(iload_reg_reg);
%}
instruct MoveL2D_stack_reg(fRegD dst, stackSlotL src) %{
match(Set dst (MoveL2D src));
effect(DEF dst, USE src);
ins_cost(LOAD_COST);
format %{ "fld $dst, $src\t#@MoveL2D_stack_reg" %}
ins_encode %{
__ fld(as_FloatRegister($dst$$reg), Address(sp, $src$$disp));
%}
ins_pipe(pipe_class_memory);
%}
instruct MoveF2I_reg_stack(stackSlotI dst, fRegF src) %{
match(Set dst (MoveF2I src));
effect(DEF dst, USE src);
ins_cost(STORE_COST);
format %{ "fsw $src, $dst\t#@MoveF2I_reg_stack" %}
ins_encode %{
__ fsw(as_FloatRegister($src$$reg), Address(sp, $dst$$disp));
%}
ins_pipe(pipe_class_memory);
%}
instruct MoveI2F_reg_stack(stackSlotF dst, iRegI src) %{
match(Set dst (MoveI2F src));
effect(DEF dst, USE src);
ins_cost(STORE_COST);
format %{ "sw $src, $dst\t#@MoveI2F_reg_stack" %}
ins_encode %{
__ sw(as_Register($src$$reg), Address(sp, $dst$$disp));
%}
ins_pipe(istore_reg_reg);
%}
instruct MoveD2L_reg_stack(stackSlotL dst, fRegD src) %{
match(Set dst (MoveD2L src));
effect(DEF dst, USE src);
ins_cost(STORE_COST);
format %{ "fsd $dst, $src\t#@MoveD2L_reg_stack" %}
ins_encode %{
__ fsd(as_FloatRegister($src$$reg), Address(sp, $dst$$disp));
%}
ins_pipe(pipe_class_memory);
%}
instruct MoveL2D_reg_stack(stackSlotD dst, iRegL src) %{
match(Set dst (MoveL2D src));
effect(DEF dst, USE src);
ins_cost(STORE_COST);
format %{ "sd $src, $dst\t#@MoveL2D_reg_stack" %}
ins_encode %{
__ sd(as_Register($src$$reg), Address(sp, $dst$$disp));
%}
ins_pipe(istore_reg_reg);
%}
instruct MoveF2I_reg_reg(iRegINoSp dst, fRegF src) %{
match(Set dst (MoveF2I src));
effect(DEF dst, USE src);
ins_cost(XFER_COST);
format %{ "fmv.x.w $dst, $src\t#@MoveL2D_reg_stack" %}
ins_encode %{
__ fmv_x_w(as_Register($dst$$reg), as_FloatRegister($src$$reg));
%}
ins_pipe(fp_f2i);
%}
instruct MoveI2F_reg_reg(fRegF dst, iRegI src) %{
match(Set dst (MoveI2F src));
effect(DEF dst, USE src);
ins_cost(XFER_COST);
format %{ "fmv.w.x $dst, $src\t#@MoveI2F_reg_reg" %}
ins_encode %{
__ fmv_w_x(as_FloatRegister($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(fp_i2f);
%}
instruct MoveD2L_reg_reg(iRegLNoSp dst, fRegD src) %{
match(Set dst (MoveD2L src));
effect(DEF dst, USE src);
ins_cost(XFER_COST);
format %{ "fmv.x.d $dst, $src\t#@MoveD2L_reg_reg" %}
ins_encode %{
__ fmv_x_d(as_Register($dst$$reg), as_FloatRegister($src$$reg));
%}
ins_pipe(fp_d2l);
%}
instruct MoveL2D_reg_reg(fRegD dst, iRegL src) %{
match(Set dst (MoveL2D src));
effect(DEF dst, USE src);
ins_cost(XFER_COST);
format %{ "fmv.d.x $dst, $src\t#@MoveD2L_reg_reg" %}
ins_encode %{
__ fmv_d_x(as_FloatRegister($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(fp_l2d);
%}
// ============================================================================
// Compare Instructions which set the result float comparisons in dest register.
instruct cmpF3_reg_reg(iRegINoSp dst, fRegF op1, fRegF op2)
%{
match(Set dst (CmpF3 op1 op2));
ins_cost(XFER_COST * 2 + BRANCH_COST + ALU_COST);
format %{ "flt.s $dst, $op2, $op1\t#@cmpF3_reg_reg\n\t"
"bgtz $dst, done\n\t"
"feq.s $dst, $op1, $op2\n\t"
"addi $dst, $dst, -1\t#@cmpF3_reg_reg"
%}
ins_encode %{
// we want -1 for unordered or less than, 0 for equal and 1 for greater than.
__ float_compare(as_Register($dst$$reg), as_FloatRegister($op1$$reg),
as_FloatRegister($op2$$reg), -1 /*unordered_result < 0*/);
%}
ins_pipe(pipe_class_default);
%}
instruct cmpD3_reg_reg(iRegINoSp dst, fRegD op1, fRegD op2)
%{
match(Set dst (CmpD3 op1 op2));
ins_cost(XFER_COST * 2 + BRANCH_COST + ALU_COST);
format %{ "flt.d $dst, $op2, $op1\t#@cmpD3_reg_reg\n\t"
"bgtz $dst, done\n\t"
"feq.d $dst, $op1, $op2\n\t"
"addi $dst, $dst, -1\t#@cmpD3_reg_reg"
%}
ins_encode %{
// we want -1 for unordered or less than, 0 for equal and 1 for greater than.
__ double_compare(as_Register($dst$$reg), as_FloatRegister($op1$$reg), as_FloatRegister($op2$$reg), -1 /*unordered_result < 0*/);
%}
ins_pipe(pipe_class_default);
%}
instruct cmpL3_reg_reg(iRegINoSp dst, iRegL op1, iRegL op2)
%{
match(Set dst (CmpL3 op1 op2));
ins_cost(ALU_COST * 3 + BRANCH_COST);
format %{ "slt $dst, $op2, $op1\t#@cmpL3_reg_reg\n\t"
"bnez $dst, done\n\t"
"slt $dst, $op1, $op2\n\t"
"neg $dst, $dst\t#@cmpL3_reg_reg"
%}
ins_encode %{
__ cmp_l2i(t0, as_Register($op1$$reg), as_Register($op2$$reg));
__ mv(as_Register($dst$$reg), t0);
%}
ins_pipe(pipe_class_default);
%}
instruct cmpLTMask_reg_reg(iRegINoSp dst, iRegI p, iRegI q)
%{
match(Set dst (CmpLTMask p q));
ins_cost(2 * ALU_COST);
format %{ "slt $dst, $p, $q\t#@cmpLTMask_reg_reg\n\t"
"subw $dst, zr, $dst\t#@cmpLTMask_reg_reg"
%}
ins_encode %{
__ slt(as_Register($dst$$reg), as_Register($p$$reg), as_Register($q$$reg));
__ subw(as_Register($dst$$reg), zr, as_Register($dst$$reg));
%}
ins_pipe(ialu_reg_reg);
%}
instruct cmpLTMask_reg_zero(iRegINoSp dst, iRegIorL2I op, immI0 zero)
%{
match(Set dst (CmpLTMask op zero));
ins_cost(ALU_COST);
format %{ "sraiw $dst, $dst, 31\t#@cmpLTMask_reg_reg" %}
ins_encode %{
__ sraiw(as_Register($dst$$reg), as_Register($op$$reg), 31);
%}
ins_pipe(ialu_reg_shift);
%}
// ============================================================================
// Max and Min
instruct minI_rReg(iRegINoSp dst, iRegI src1, iRegI src2)
%{
match(Set dst (MinI src1 src2));
effect(DEF dst, USE src1, USE src2);
ins_cost(BRANCH_COST + ALU_COST * 2);
format %{
"ble $src1, $src2, Lsrc1.\t#@minI_rReg\n\t"
"mv $dst, $src2\n\t"
"j Ldone\n\t"
"bind Lsrc1\n\t"
"mv $dst, $src1\n\t"
"bind\t#@minI_rReg"
%}
ins_encode %{
Label Lsrc1, Ldone;
__ ble(as_Register($src1$$reg), as_Register($src2$$reg), Lsrc1);
__ mv(as_Register($dst$$reg), as_Register($src2$$reg));
__ j(Ldone);
__ bind(Lsrc1);
__ mv(as_Register($dst$$reg), as_Register($src1$$reg));
__ bind(Ldone);
%}
ins_pipe(ialu_reg_reg);
%}
instruct maxI_rReg(iRegINoSp dst, iRegI src1, iRegI src2)
%{
match(Set dst (MaxI src1 src2));
effect(DEF dst, USE src1, USE src2);
ins_cost(BRANCH_COST + ALU_COST * 2);
format %{
"bge $src1, $src2, Lsrc1\t#@maxI_rReg\n\t"
"mv $dst, $src2\n\t"
"j Ldone\n\t"
"bind Lsrc1\n\t"
"mv $dst, $src1\n\t"
"bind\t#@maxI_rReg"
%}
ins_encode %{
Label Lsrc1, Ldone;
__ bge(as_Register($src1$$reg), as_Register($src2$$reg), Lsrc1);
__ mv(as_Register($dst$$reg), as_Register($src2$$reg));
__ j(Ldone);
__ bind(Lsrc1);
__ mv(as_Register($dst$$reg), as_Register($src1$$reg));
__ bind(Ldone);
%}
ins_pipe(ialu_reg_reg);
%}
// ============================================================================
// Branch Instructions
// Direct Branch.
instruct branch(label lbl)
%{
match(Goto);
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "j $lbl\t#@branch" %}
ins_encode(riscv_enc_j(lbl));
ins_pipe(pipe_branch);
%}
// ============================================================================
// Compare and Branch Instructions
// Patterns for short (< 12KiB) variants
// Compare flags and branch near instructions.
instruct cmpFlag_branch(cmpOpEqNe cmp, rFlagsReg cr, label lbl) %{
match(If cmp cr);
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $cr, zr, $lbl\t#@cmpFlag_branch" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($cr$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
// Compare signed int and branch near instructions
instruct cmpI_branch(cmpOp cmp, iRegI op1, iRegI op2, label lbl)
%{
// Same match rule as `far_cmpI_branch'.
match(If cmp (CmpI op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpI_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
instruct cmpI_loop(cmpOp cmp, iRegI op1, iRegI op2, label lbl)
%{
// Same match rule as `far_cmpI_loop'.
match(CountedLoopEnd cmp (CmpI op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpI_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
// Compare unsigned int and branch near instructions
instruct cmpU_branch(cmpOpU cmp, iRegI op1, iRegI op2, label lbl)
%{
// Same match rule as `far_cmpU_branch'.
match(If cmp (CmpU op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpU_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
instruct cmpU_loop(cmpOpU cmp, iRegI op1, iRegI op2, label lbl)
%{
// Same match rule as `far_cmpU_loop'.
match(CountedLoopEnd cmp (CmpU op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpU_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
// Compare signed long and branch near instructions
instruct cmpL_branch(cmpOp cmp, iRegL op1, iRegL op2, label lbl)
%{
// Same match rule as `far_cmpL_branch'.
match(If cmp (CmpL op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpL_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
instruct cmpL_loop(cmpOp cmp, iRegL op1, iRegL op2, label lbl)
%{
// Same match rule as `far_cmpL_loop'.
match(CountedLoopEnd cmp (CmpL op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpL_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
// Compare unsigned long and branch near instructions
instruct cmpUL_branch(cmpOpU cmp, iRegL op1, iRegL op2, label lbl)
%{
// Same match rule as `far_cmpUL_branch'.
match(If cmp (CmpUL op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpUL_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
instruct cmpUL_loop(cmpOpU cmp, iRegL op1, iRegL op2, label lbl)
%{
// Same match rule as `far_cmpUL_loop'.
match(CountedLoopEnd cmp (CmpUL op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpUL_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
// Compare pointer and branch near instructions
instruct cmpP_branch(cmpOpU cmp, iRegP op1, iRegP op2, label lbl)
%{
// Same match rule as `far_cmpP_branch'.
match(If cmp (CmpP op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpP_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
instruct cmpP_loop(cmpOpU cmp, iRegP op1, iRegP op2, label lbl)
%{
// Same match rule as `far_cmpP_loop'.
match(CountedLoopEnd cmp (CmpP op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpP_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
// Compare narrow pointer and branch near instructions
instruct cmpN_branch(cmpOpU cmp, iRegN op1, iRegN op2, label lbl)
%{
// Same match rule as `far_cmpN_branch'.
match(If cmp (CmpN op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpN_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
instruct cmpN_loop(cmpOpU cmp, iRegN op1, iRegN op2, label lbl)
%{
// Same match rule as `far_cmpN_loop'.
match(CountedLoopEnd cmp (CmpN op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, $op2, $lbl\t#@cmpN_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmp_branch);
ins_short_branch(1);
%}
// Compare float and branch near instructions
instruct cmpF_branch(cmpOp cmp, fRegF op1, fRegF op2, label lbl)
%{
// Same match rule as `far_cmpF_branch'.
match(If cmp (CmpF op1 op2));
effect(USE lbl);
ins_cost(XFER_COST + BRANCH_COST);
format %{ "float_b$cmp $op1, $op2, $lbl \t#@cmpF_branch"%}
ins_encode %{
__ float_cmp_branch($cmp$$cmpcode, as_FloatRegister($op1$$reg), as_FloatRegister($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_class_compare);
ins_short_branch(1);
%}
instruct cmpF_loop(cmpOp cmp, fRegF op1, fRegF op2, label lbl)
%{
// Same match rule as `far_cmpF_loop'.
match(CountedLoopEnd cmp (CmpF op1 op2));
effect(USE lbl);
ins_cost(XFER_COST + BRANCH_COST);
format %{ "float_b$cmp $op1, $op2, $lbl\t#@cmpF_loop"%}
ins_encode %{
__ float_cmp_branch($cmp$$cmpcode, as_FloatRegister($op1$$reg), as_FloatRegister($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_class_compare);
ins_short_branch(1);
%}
// Compare double and branch near instructions
instruct cmpD_branch(cmpOp cmp, fRegD op1, fRegD op2, label lbl)
%{
// Same match rule as `far_cmpD_branch'.
match(If cmp (CmpD op1 op2));
effect(USE lbl);
ins_cost(XFER_COST + BRANCH_COST);
format %{ "double_b$cmp $op1, $op2, $lbl\t#@cmpD_branch"%}
ins_encode %{
__ float_cmp_branch($cmp$$cmpcode | C2_MacroAssembler::double_branch_mask, as_FloatRegister($op1$$reg),
as_FloatRegister($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_class_compare);
ins_short_branch(1);
%}
instruct cmpD_loop(cmpOp cmp, fRegD op1, fRegD op2, label lbl)
%{
// Same match rule as `far_cmpD_loop'.
match(CountedLoopEnd cmp (CmpD op1 op2));
effect(USE lbl);
ins_cost(XFER_COST + BRANCH_COST);
format %{ "double_b$cmp $op1, $op2, $lbl\t#@cmpD_loop"%}
ins_encode %{
__ float_cmp_branch($cmp$$cmpcode | C2_MacroAssembler::double_branch_mask, as_FloatRegister($op1$$reg),
as_FloatRegister($op2$$reg), *($lbl$$label));
%}
ins_pipe(pipe_class_compare);
ins_short_branch(1);
%}
// Compare signed int with zero and branch near instructions
instruct cmpI_reg_imm0_branch(cmpOp cmp, iRegI op1, immI0 zero, label lbl)
%{
// Same match rule as `far_cmpI_reg_imm0_branch'.
match(If cmp (CmpI op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpI_reg_imm0_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), zr, *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
instruct cmpI_reg_imm0_loop(cmpOp cmp, iRegI op1, immI0 zero, label lbl)
%{
// Same match rule as `far_cmpI_reg_imm0_loop'.
match(CountedLoopEnd cmp (CmpI op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpI_reg_imm0_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), zr, *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
// Compare unsigned int with zero and branch near instructions
instruct cmpUEqNeLeGt_reg_imm0_branch(cmpOpUEqNeLeGt cmp, iRegI op1, immI0 zero, label lbl)
%{
// Same match rule as `far_cmpUEqNeLeGt_reg_imm0_branch'.
match(If cmp (CmpU op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpUEqNeLeGt_reg_imm0_branch" %}
ins_encode %{
__ enc_cmpUEqNeLeGt_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
instruct cmpUEqNeLeGt_reg_imm0_loop(cmpOpUEqNeLeGt cmp, iRegI op1, immI0 zero, label lbl)
%{
// Same match rule as `far_cmpUEqNeLeGt_reg_imm0_loop'.
match(CountedLoopEnd cmp (CmpU op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpUEqNeLeGt_reg_imm0_loop" %}
ins_encode %{
__ enc_cmpUEqNeLeGt_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
// Compare signed long with zero and branch near instructions
instruct cmpL_reg_imm0_branch(cmpOp cmp, iRegL op1, immL0 zero, label lbl)
%{
// Same match rule as `far_cmpL_reg_imm0_branch'.
match(If cmp (CmpL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpL_reg_imm0_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), zr, *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
instruct cmpL_reg_imm0_loop(cmpOp cmp, iRegL op1, immL0 zero, label lbl)
%{
// Same match rule as `far_cmpL_reg_imm0_loop'.
match(CountedLoopEnd cmp (CmpL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpL_reg_imm0_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), zr, *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
// Compare unsigned long with zero and branch near instructions
instruct cmpULEqNeLeGt_reg_imm0_branch(cmpOpUEqNeLeGt cmp, iRegL op1, immL0 zero, label lbl)
%{
// Same match rule as `far_cmpULEqNeLeGt_reg_imm0_branch'.
match(If cmp (CmpUL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpULEqNeLeGt_reg_imm0_branch" %}
ins_encode %{
__ enc_cmpUEqNeLeGt_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
instruct cmpULEqNeLeGt_reg_imm0_loop(cmpOpUEqNeLeGt cmp, iRegL op1, immL0 zero, label lbl)
%{
// Same match rule as `far_cmpULEqNeLeGt_reg_imm0_loop'.
match(CountedLoopEnd cmp (CmpUL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpULEqNeLeGt_reg_imm0_loop" %}
ins_encode %{
__ enc_cmpUEqNeLeGt_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
// Compare pointer with zero and branch near instructions
instruct cmpP_imm0_branch(cmpOpEqNe cmp, iRegP op1, immP0 zero, label lbl) %{
// Same match rule as `far_cmpP_reg_imm0_branch'.
match(If cmp (CmpP op1 zero));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpP_imm0_branch" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
instruct cmpP_imm0_loop(cmpOpEqNe cmp, iRegP op1, immP0 zero, label lbl) %{
// Same match rule as `far_cmpP_reg_imm0_loop'.
match(CountedLoopEnd cmp (CmpP op1 zero));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpP_imm0_loop" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
// Compare narrow pointer with zero and branch near instructions
instruct cmpN_imm0_branch(cmpOpEqNe cmp, iRegN op1, immN0 zero, label lbl) %{
// Same match rule as `far_cmpN_reg_imm0_branch'.
match(If cmp (CmpN op1 zero));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpN_imm0_branch" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
instruct cmpN_imm0_loop(cmpOpEqNe cmp, iRegN op1, immN0 zero, label lbl) %{
// Same match rule as `far_cmpN_reg_imm0_loop'.
match(CountedLoopEnd cmp (CmpN op1 zero));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpN_imm0_loop" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
// Compare narrow pointer with pointer zero and branch near instructions
instruct cmpP_narrowOop_imm0_branch(cmpOpEqNe cmp, iRegN op1, immP0 zero, label lbl) %{
// Same match rule as `far_cmpP_narrowOop_imm0_branch'.
match(If cmp (CmpP (DecodeN op1) zero));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpP_narrowOop_imm0_branch" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
instruct cmpP_narrowOop_imm0_loop(cmpOpEqNe cmp, iRegN op1, immP0 zero, label lbl) %{
// Same match rule as `far_cmpP_narrowOop_imm0_loop'.
match(CountedLoopEnd cmp (CmpP (DecodeN op1) zero));
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "b$cmp $op1, zr, $lbl\t#@cmpP_narrowOop_imm0_loop" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label));
%}
ins_pipe(pipe_cmpz_branch);
ins_short_branch(1);
%}
// Patterns for far (20KiB) variants
instruct far_cmpFlag_branch(cmpOp cmp, rFlagsReg cr, label lbl) %{
match(If cmp cr);
effect(USE lbl);
ins_cost(BRANCH_COST);
format %{ "far_b$cmp $cr, zr, $lbl\t#@far_cmpFlag_branch"%}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($cr$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
// Compare signed int and branch far instructions
instruct far_cmpI_branch(cmpOp cmp, iRegI op1, iRegI op2, label lbl) %{
match(If cmp (CmpI op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
// the format instruction [far_b$cmp] here is be used as two insructions
// in macroassembler: b$not_cmp(op1, op2, done), j($lbl), bind(done)
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpI_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpI_loop(cmpOp cmp, iRegI op1, iRegI op2, label lbl) %{
match(CountedLoopEnd cmp (CmpI op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpI_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpU_branch(cmpOpU cmp, iRegI op1, iRegI op2, label lbl) %{
match(If cmp (CmpU op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpU_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpU_loop(cmpOpU cmp, iRegI op1, iRegI op2, label lbl) %{
match(CountedLoopEnd cmp (CmpU op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpU_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpL_branch(cmpOp cmp, iRegL op1, iRegL op2, label lbl) %{
match(If cmp (CmpL op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpL_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpLloop(cmpOp cmp, iRegL op1, iRegL op2, label lbl) %{
match(CountedLoopEnd cmp (CmpL op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpL_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpUL_branch(cmpOpU cmp, iRegL op1, iRegL op2, label lbl) %{
match(If cmp (CmpUL op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpUL_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpUL_loop(cmpOpU cmp, iRegL op1, iRegL op2, label lbl) %{
match(CountedLoopEnd cmp (CmpUL op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpUL_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpP_branch(cmpOpU cmp, iRegP op1, iRegP op2, label lbl)
%{
match(If cmp (CmpP op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpP_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpP_loop(cmpOpU cmp, iRegP op1, iRegP op2, label lbl)
%{
match(CountedLoopEnd cmp (CmpP op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpP_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpN_branch(cmpOpU cmp, iRegN op1, iRegN op2, label lbl)
%{
match(If cmp (CmpN op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpN_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
instruct far_cmpN_loop(cmpOpU cmp, iRegN op1, iRegN op2, label lbl)
%{
match(CountedLoopEnd cmp (CmpN op1 op2));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, $op2, $lbl\t#@far_cmpN_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode | C2_MacroAssembler::unsigned_branch_mask, as_Register($op1$$reg),
as_Register($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmp_branch);
%}
// Float compare and branch instructions
instruct far_cmpF_branch(cmpOp cmp, fRegF op1, fRegF op2, label lbl)
%{
match(If cmp (CmpF op1 op2));
effect(USE lbl);
ins_cost(XFER_COST + BRANCH_COST * 2);
format %{ "far_float_b$cmp $op1, $op2, $lbl\t#@far_cmpF_branch"%}
ins_encode %{
__ float_cmp_branch($cmp$$cmpcode, as_FloatRegister($op1$$reg), as_FloatRegister($op2$$reg),
*($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_class_compare);
%}
instruct far_cmpF_loop(cmpOp cmp, fRegF op1, fRegF op2, label lbl)
%{
match(CountedLoopEnd cmp (CmpF op1 op2));
effect(USE lbl);
ins_cost(XFER_COST + BRANCH_COST * 2);
format %{ "far_float_b$cmp $op1, $op2, $lbl\t#@far_cmpF_loop"%}
ins_encode %{
__ float_cmp_branch($cmp$$cmpcode, as_FloatRegister($op1$$reg), as_FloatRegister($op2$$reg),
*($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_class_compare);
%}
// Double compare and branch instructions
instruct far_cmpD_branch(cmpOp cmp, fRegD op1, fRegD op2, label lbl)
%{
match(If cmp (CmpD op1 op2));
effect(USE lbl);
ins_cost(XFER_COST + BRANCH_COST * 2);
format %{ "far_double_b$cmp $op1, $op2, $lbl\t#@far_cmpD_branch"%}
ins_encode %{
__ float_cmp_branch($cmp$$cmpcode | C2_MacroAssembler::double_branch_mask, as_FloatRegister($op1$$reg),
as_FloatRegister($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_class_compare);
%}
instruct far_cmpD_loop(cmpOp cmp, fRegD op1, fRegD op2, label lbl)
%{
match(CountedLoopEnd cmp (CmpD op1 op2));
effect(USE lbl);
ins_cost(XFER_COST + BRANCH_COST * 2);
format %{ "far_double_b$cmp $op1, $op2, $lbl\t#@far_cmpD_loop"%}
ins_encode %{
__ float_cmp_branch($cmp$$cmpcode | C2_MacroAssembler::double_branch_mask, as_FloatRegister($op1$$reg),
as_FloatRegister($op2$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_class_compare);
%}
instruct far_cmpI_reg_imm0_branch(cmpOp cmp, iRegI op1, immI0 zero, label lbl)
%{
match(If cmp (CmpI op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpI_reg_imm0_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), zr, *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpI_reg_imm0_loop(cmpOp cmp, iRegI op1, immI0 zero, label lbl)
%{
match(CountedLoopEnd cmp (CmpI op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpI_reg_imm0_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), zr, *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpUEqNeLeGt_imm0_branch(cmpOpUEqNeLeGt cmp, iRegI op1, immI0 zero, label lbl)
%{
match(If cmp (CmpU op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpUEqNeLeGt_imm0_branch" %}
ins_encode %{
__ enc_cmpUEqNeLeGt_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpUEqNeLeGt_reg_imm0_loop(cmpOpUEqNeLeGt cmp, iRegI op1, immI0 zero, label lbl)
%{
match(CountedLoopEnd cmp (CmpU op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpUEqNeLeGt_reg_imm0_loop" %}
ins_encode %{
__ enc_cmpUEqNeLeGt_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
// compare lt/ge unsigned instructs has no short instruct with same match
instruct far_cmpULtGe_reg_imm0_branch(cmpOpULtGe cmp, iRegI op1, immI0 zero, label lbl)
%{
match(If cmp (CmpU op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "j $lbl if $cmp == ge\t#@far_cmpULtGe_reg_imm0_branch" %}
ins_encode(riscv_enc_far_cmpULtGe_imm0_branch(cmp, op1, lbl));
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpULtGe_reg_imm0_loop(cmpOpULtGe cmp, iRegI op1, immI0 zero, label lbl)
%{
match(CountedLoopEnd cmp (CmpU op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "j $lbl if $cmp == ge\t#@far_cmpULtGe_reg_imm0_loop" %}
ins_encode(riscv_enc_far_cmpULtGe_imm0_branch(cmp, op1, lbl));
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpL_reg_imm0_branch(cmpOp cmp, iRegL op1, immL0 zero, label lbl)
%{
match(If cmp (CmpL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpL_reg_imm0_branch" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), zr, *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpL_reg_imm0_loop(cmpOp cmp, iRegL op1, immL0 zero, label lbl)
%{
match(CountedLoopEnd cmp (CmpL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpL_reg_imm0_loop" %}
ins_encode %{
__ cmp_branch($cmp$$cmpcode, as_Register($op1$$reg), zr, *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpULEqNeLeGt_reg_imm0_branch(cmpOpUEqNeLeGt cmp, iRegL op1, immL0 zero, label lbl)
%{
match(If cmp (CmpUL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpULEqNeLeGt_reg_imm0_branch" %}
ins_encode %{
__ enc_cmpUEqNeLeGt_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpULEqNeLeGt_reg_imm0_loop(cmpOpUEqNeLeGt cmp, iRegL op1, immL0 zero, label lbl)
%{
match(CountedLoopEnd cmp (CmpUL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpULEqNeLeGt_reg_imm0_loop" %}
ins_encode %{
__ enc_cmpUEqNeLeGt_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
// compare lt/ge unsigned instructs has no short instruct with same match
instruct far_cmpULLtGe_reg_imm0_branch(cmpOpULtGe cmp, iRegL op1, immL0 zero, label lbl)
%{
match(If cmp (CmpUL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "j $lbl if $cmp == ge\t#@far_cmpULLtGe_reg_imm0_branch" %}
ins_encode(riscv_enc_far_cmpULtGe_imm0_branch(cmp, op1, lbl));
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpULLtGe_reg_imm0_loop(cmpOpULtGe cmp, iRegL op1, immL0 zero, label lbl)
%{
match(CountedLoopEnd cmp (CmpUL op1 zero));
effect(USE op1, USE lbl);
ins_cost(BRANCH_COST);
format %{ "j $lbl if $cmp == ge\t#@far_cmpULLtGe_reg_imm0_loop" %}
ins_encode(riscv_enc_far_cmpULtGe_imm0_branch(cmp, op1, lbl));
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpP_imm0_branch(cmpOpEqNe cmp, iRegP op1, immP0 zero, label lbl) %{
match(If cmp (CmpP op1 zero));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpP_imm0_branch" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpP_imm0_loop(cmpOpEqNe cmp, iRegP op1, immP0 zero, label lbl) %{
match(CountedLoopEnd cmp (CmpP op1 zero));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpP_imm0_loop" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpN_imm0_branch(cmpOpEqNe cmp, iRegN op1, immN0 zero, label lbl) %{
match(If cmp (CmpN op1 zero));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpN_imm0_branch" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpN_imm0_loop(cmpOpEqNe cmp, iRegN op1, immN0 zero, label lbl) %{
match(CountedLoopEnd cmp (CmpN op1 zero));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpN_imm0_loop" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpP_narrowOop_imm0_branch(cmpOpEqNe cmp, iRegN op1, immP0 zero, label lbl) %{
match(If cmp (CmpP (DecodeN op1) zero));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpP_narrowOop_imm0_branch" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
instruct far_cmpP_narrowOop_imm0_loop(cmpOpEqNe cmp, iRegN op1, immP0 zero, label lbl) %{
match(CountedLoopEnd cmp (CmpP (DecodeN op1) zero));
effect(USE lbl);
ins_cost(BRANCH_COST * 2);
format %{ "far_b$cmp $op1, zr, $lbl\t#@far_cmpP_narrowOop_imm0_loop" %}
ins_encode %{
__ enc_cmpEqNe_imm0_branch($cmp$$cmpcode, as_Register($op1$$reg), *($lbl$$label), /* is_far */ true);
%}
ins_pipe(pipe_cmpz_branch);
%}
// ============================================================================
// Conditional Move Instructions
instruct cmovI_cmpI(iRegINoSp dst, iRegI src, iRegI op1, iRegI op2, cmpOp cop) %{
match(Set dst (CMoveI (Binary cop (CmpI op1 op2)) (Binary dst src)));
ins_cost(ALU_COST + BRANCH_COST);
format %{
"bneg$cop $op1, $op2, skip\t#@cmovI_cmpI\n\t"
"mv $dst, $src\n\t"
"skip:"
%}
ins_encode %{
__ enc_cmove($cop$$cmpcode,
as_Register($op1$$reg), as_Register($op2$$reg),
as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(pipe_slow);
%}
instruct cmovI_cmpU(iRegINoSp dst, iRegI src, iRegI op1, iRegI op2, cmpOpU cop) %{
match(Set dst (CMoveI (Binary cop (CmpU op1 op2)) (Binary dst src)));
ins_cost(ALU_COST + BRANCH_COST);
format %{
"bneg$cop $op1, $op2, skip\t#@cmovI_cmpU\n\t"
"mv $dst, $src\n\t"
"skip:"
%}
ins_encode %{
__ enc_cmove($cop$$cmpcode | C2_MacroAssembler::unsigned_branch_mask,
as_Register($op1$$reg), as_Register($op2$$reg),
as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(pipe_slow);
%}
instruct cmovI_cmpL(iRegINoSp dst, iRegI src, iRegL op1, iRegL op2, cmpOp cop) %{
match(Set dst (CMoveI (Binary cop (CmpL op1 op2)) (Binary dst src)));
ins_cost(ALU_COST + BRANCH_COST);
format %{
"bneg$cop $op1, $op2, skip\t#@cmovI_cmpL\n\t"
"mv $dst, $src\n\t"
"skip:"
%}
ins_encode %{
__ enc_cmove($cop$$cmpcode,
as_Register($op1$$reg), as_Register($op2$$reg),
as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(pipe_slow);
%}
instruct cmovL_cmpL(iRegLNoSp dst, iRegL src, iRegL op1, iRegL op2, cmpOp cop) %{
match(Set dst (CMoveL (Binary cop (CmpL op1 op2)) (Binary dst src)));
ins_cost(ALU_COST + BRANCH_COST);
format %{
"bneg$cop $op1, $op2, skip\t#@cmovL_cmpL\n\t"
"mv $dst, $src\n\t"
"skip:"
%}
ins_encode %{
__ enc_cmove($cop$$cmpcode,
as_Register($op1$$reg), as_Register($op2$$reg),
as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(pipe_slow);
%}
instruct cmovL_cmpUL(iRegLNoSp dst, iRegL src, iRegL op1, iRegL op2, cmpOpU cop) %{
match(Set dst (CMoveL (Binary cop (CmpUL op1 op2)) (Binary dst src)));
ins_cost(ALU_COST + BRANCH_COST);
format %{
"bneg$cop $op1, $op2, skip\t#@cmovL_cmpUL\n\t"
"mv $dst, $src\n\t"
"skip:"
%}
ins_encode %{
__ enc_cmove($cop$$cmpcode | C2_MacroAssembler::unsigned_branch_mask,
as_Register($op1$$reg), as_Register($op2$$reg),
as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(pipe_slow);
%}
instruct cmovI_cmpUL(iRegINoSp dst, iRegI src, iRegL op1, iRegL op2, cmpOpU cop) %{
match(Set dst (CMoveI (Binary cop (CmpUL op1 op2)) (Binary dst src)));
ins_cost(ALU_COST + BRANCH_COST);
format %{
"bneg$cop $op1, $op2\t#@cmovI_cmpUL\n\t"
"mv $dst, $src\n\t"
"skip:"
%}
ins_encode %{
__ enc_cmove($cop$$cmpcode | C2_MacroAssembler::unsigned_branch_mask,
as_Register($op1$$reg), as_Register($op2$$reg),
as_Register($dst$$reg), as_Register($src$$reg));
%}
ins_pipe(pipe_slow);
%}
// ============================================================================
// Procedure Call/Return Instructions
// Call Java Static Instruction
// Note: If this code changes, the corresponding ret_addr_offset() and
// compute_padding() functions will have to be adjusted.
instruct CallStaticJavaDirect(method meth)
%{
match(CallStaticJava);
effect(USE meth);
ins_cost(BRANCH_COST);
format %{ "CALL,static $meth\t#@CallStaticJavaDirect" %}
ins_encode(riscv_enc_java_static_call(meth),
riscv_enc_call_epilog);
ins_pipe(pipe_class_call);
ins_alignment(4);
%}
// TO HERE
// Call Java Dynamic Instruction
// Note: If this code changes, the corresponding ret_addr_offset() and
// compute_padding() functions will have to be adjusted.
instruct CallDynamicJavaDirect(method meth, rFlagsReg cr)
%{
match(CallDynamicJava);
effect(USE meth, KILL cr);
ins_cost(BRANCH_COST + ALU_COST * 6);
format %{ "CALL,dynamic $meth\t#@CallDynamicJavaDirect" %}
ins_encode(riscv_enc_java_dynamic_call(meth),
riscv_enc_call_epilog);
ins_pipe(pipe_class_call);
ins_alignment(4);
%}
// Call Runtime Instruction
instruct CallRuntimeDirect(method meth, rFlagsReg cr)
%{
match(CallRuntime);
effect(USE meth, KILL cr);
ins_cost(BRANCH_COST);
format %{ "CALL, runtime $meth\t#@CallRuntimeDirect" %}
ins_encode(riscv_enc_java_to_runtime(meth));
ins_pipe(pipe_class_call);
%}
// Call Runtime Instruction
instruct CallLeafDirect(method meth, rFlagsReg cr)
%{
match(CallLeaf);
effect(USE meth, KILL cr);
ins_cost(BRANCH_COST);
format %{ "CALL, runtime leaf $meth\t#@CallLeafDirect" %}
ins_encode(riscv_enc_java_to_runtime(meth));
ins_pipe(pipe_class_call);
%}
// Call Runtime Instruction
instruct CallLeafNoFPDirect(method meth, rFlagsReg cr)
%{
match(CallLeafNoFP);
effect(USE meth, KILL cr);
ins_cost(BRANCH_COST);
format %{ "CALL, runtime leaf nofp $meth\t#@CallLeafNoFPDirect" %}
ins_encode(riscv_enc_java_to_runtime(meth));
ins_pipe(pipe_class_call);
%}
// ============================================================================
// Partial Subtype Check
//
// superklass array for an instance of the superklass. Set a hidden
// internal cache on a hit (cache is checked with exposed code in
// gen_subtype_check()). Return zero for a hit. The encoding
// ALSO sets flags.
instruct partialSubtypeCheck(iRegP_R15 result, iRegP_R14 sub, iRegP_R10 super, iRegP_R12 tmp, rFlagsReg cr)
%{
match(Set result (PartialSubtypeCheck sub super));
effect(KILL tmp, KILL cr);
ins_cost(2 * STORE_COST + 3 * LOAD_COST + 4 * ALU_COST + BRANCH_COST * 4);
format %{ "partialSubtypeCheck $result, $sub, $super\t#@partialSubtypeCheck" %}
ins_encode(riscv_enc_partial_subtype_check(sub, super, tmp, result));
opcode(0x1); // Force zero of result reg on hit
ins_pipe(pipe_class_memory);
%}
instruct partialSubtypeCheckVsZero(iRegP_R15 result, iRegP_R14 sub, iRegP_R10 super, iRegP_R12 tmp,
immP0 zero, rFlagsReg cr)
%{
match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
effect(KILL tmp, KILL result);
ins_cost(2 * STORE_COST + 3 * LOAD_COST + 4 * ALU_COST + BRANCH_COST * 4);
format %{ "partialSubtypeCheck $result, $sub, $super == 0\t#@partialSubtypeCheckVsZero" %}
ins_encode(riscv_enc_partial_subtype_check(sub, super, tmp, result));
opcode(0x0); // Don't zero result reg on hit
ins_pipe(pipe_class_memory);
%}
instruct string_compareU(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2, iRegI_R14 cnt2,
iRegI_R10 result, iRegP_R28 tmp1, iRegL_R29 tmp2, iRegL_R30 tmp3, rFlagsReg cr)
%{
predicate(!UseRVV && ((StrCompNode *)n)->encoding() == StrIntrinsicNode::UU);
match(Set result (StrComp(Binary str1 cnt1)(Binary str2 cnt2)));
effect(KILL tmp1, KILL tmp2, KILL tmp3, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
format %{ "String Compare $str1, $cnt1, $str2, $cnt2 -> $result\t#@string_compareU" %}
ins_encode %{
// Count is in 8-bit bytes; non-Compact chars are 16 bits.
__ string_compare($str1$$Register, $str2$$Register,
$cnt1$$Register, $cnt2$$Register, $result$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register,
StrIntrinsicNode::UU);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_compareL(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2, iRegI_R14 cnt2,
iRegI_R10 result, iRegP_R28 tmp1, iRegL_R29 tmp2, iRegL_R30 tmp3, rFlagsReg cr)
%{
predicate(!UseRVV && ((StrCompNode *)n)->encoding() == StrIntrinsicNode::LL);
match(Set result (StrComp(Binary str1 cnt1)(Binary str2 cnt2)));
effect(KILL tmp1, KILL tmp2, KILL tmp3, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
format %{ "String Compare $str1, $cnt1, $str2, $cnt2 -> $result\t#@string_compareL" %}
ins_encode %{
__ string_compare($str1$$Register, $str2$$Register,
$cnt1$$Register, $cnt2$$Register, $result$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register,
StrIntrinsicNode::LL);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_compareUL(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2, iRegI_R14 cnt2,
iRegI_R10 result, iRegP_R28 tmp1, iRegL_R29 tmp2, iRegL_R30 tmp3, rFlagsReg cr)
%{
predicate(!UseRVV && ((StrCompNode *)n)->encoding() == StrIntrinsicNode::UL);
match(Set result (StrComp(Binary str1 cnt1)(Binary str2 cnt2)));
effect(KILL tmp1, KILL tmp2, KILL tmp3, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
format %{"String Compare $str1, $cnt1, $str2, $cnt2 -> $result\t#@string_compareUL" %}
ins_encode %{
__ string_compare($str1$$Register, $str2$$Register,
$cnt1$$Register, $cnt2$$Register, $result$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register,
StrIntrinsicNode::UL);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_compareLU(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2, iRegI_R14 cnt2,
iRegI_R10 result, iRegP_R28 tmp1, iRegL_R29 tmp2, iRegL_R30 tmp3,
rFlagsReg cr)
%{
predicate(!UseRVV && ((StrCompNode *)n)->encoding() == StrIntrinsicNode::LU);
match(Set result (StrComp(Binary str1 cnt1)(Binary str2 cnt2)));
effect(KILL tmp1, KILL tmp2, KILL tmp3, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
format %{ "String Compare $str1, $cnt1, $str2, $cnt2 -> $result\t#@string_compareLU" %}
ins_encode %{
__ string_compare($str1$$Register, $str2$$Register,
$cnt1$$Register, $cnt2$$Register, $result$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register,
StrIntrinsicNode::LU);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_indexofUU(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2, iRegI_R14 cnt2,
iRegI_R10 result, iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3,
iRegINoSp tmp4, iRegINoSp tmp5, iRegINoSp tmp6, rFlagsReg cr)
%{
predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UU);
match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5, TEMP tmp6, KILL cr);
format %{ "String IndexOf $str1,$cnt1,$str2,$cnt2 -> $result (UU)" %}
ins_encode %{
__ string_indexof($str1$$Register, $str2$$Register,
$cnt1$$Register, $cnt2$$Register,
$tmp1$$Register, $tmp2$$Register,
$tmp3$$Register, $tmp4$$Register,
$tmp5$$Register, $tmp6$$Register,
$result$$Register, StrIntrinsicNode::UU);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_indexofLL(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2, iRegI_R14 cnt2,
iRegI_R10 result, iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3,
iRegINoSp tmp4, iRegINoSp tmp5, iRegINoSp tmp6, rFlagsReg cr)
%{
predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::LL);
match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5, TEMP tmp6, KILL cr);
format %{ "String IndexOf $str1,$cnt1,$str2,$cnt2 -> $result (LL)" %}
ins_encode %{
__ string_indexof($str1$$Register, $str2$$Register,
$cnt1$$Register, $cnt2$$Register,
$tmp1$$Register, $tmp2$$Register,
$tmp3$$Register, $tmp4$$Register,
$tmp5$$Register, $tmp6$$Register,
$result$$Register, StrIntrinsicNode::LL);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_indexofUL(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2, iRegI_R14 cnt2,
iRegI_R10 result, iRegINoSp tmp1, iRegINoSp tmp2, iRegINoSp tmp3,
iRegINoSp tmp4, iRegINoSp tmp5, iRegINoSp tmp6, rFlagsReg cr)
%{
predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UL);
match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5, TEMP tmp6, KILL cr);
format %{ "String IndexOf $str1,$cnt1,$str2,$cnt2 -> $result (UL)" %}
ins_encode %{
__ string_indexof($str1$$Register, $str2$$Register,
$cnt1$$Register, $cnt2$$Register,
$tmp1$$Register, $tmp2$$Register,
$tmp3$$Register, $tmp4$$Register,
$tmp5$$Register, $tmp6$$Register,
$result$$Register, StrIntrinsicNode::UL);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_indexof_conUU(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2,
immI_le_4 int_cnt2, iRegI_R10 result, iRegINoSp tmp1, iRegINoSp tmp2,
iRegINoSp tmp3, iRegINoSp tmp4, rFlagsReg cr)
%{
predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UU);
match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 int_cnt2)));
effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr);
format %{ "String IndexOf $str1,$cnt1,$str2,$int_cnt2 -> $result (UU)" %}
ins_encode %{
int icnt2 = (int)$int_cnt2$$constant;
__ string_indexof_linearscan($str1$$Register, $str2$$Register,
$cnt1$$Register, zr,
$tmp1$$Register, $tmp2$$Register,
$tmp3$$Register, $tmp4$$Register,
icnt2, $result$$Register, StrIntrinsicNode::UU);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_indexof_conLL(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2,
immI_le_4 int_cnt2, iRegI_R10 result, iRegINoSp tmp1, iRegINoSp tmp2,
iRegINoSp tmp3, iRegINoSp tmp4, rFlagsReg cr)
%{
predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::LL);
match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 int_cnt2)));
effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr);
format %{ "String IndexOf $str1,$cnt1,$str2,$int_cnt2 -> $result (LL)" %}
ins_encode %{
int icnt2 = (int)$int_cnt2$$constant;
__ string_indexof_linearscan($str1$$Register, $str2$$Register,
$cnt1$$Register, zr,
$tmp1$$Register, $tmp2$$Register,
$tmp3$$Register, $tmp4$$Register,
icnt2, $result$$Register, StrIntrinsicNode::LL);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_indexof_conUL(iRegP_R11 str1, iRegI_R12 cnt1, iRegP_R13 str2,
immI_1 int_cnt2, iRegI_R10 result, iRegINoSp tmp1, iRegINoSp tmp2,
iRegINoSp tmp3, iRegINoSp tmp4, rFlagsReg cr)
%{
predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UL);
match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 int_cnt2)));
effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr);
format %{ "String IndexOf $str1,$cnt1,$str2,$int_cnt2 -> $result (UL)" %}
ins_encode %{
int icnt2 = (int)$int_cnt2$$constant;
__ string_indexof_linearscan($str1$$Register, $str2$$Register,
$cnt1$$Register, zr,
$tmp1$$Register, $tmp2$$Register,
$tmp3$$Register, $tmp4$$Register,
icnt2, $result$$Register, StrIntrinsicNode::UL);
%}
ins_pipe(pipe_class_memory);
%}
instruct stringU_indexof_char(iRegP_R11 str1, iRegI_R12 cnt1, iRegI_R13 ch,
iRegI_R10 result, iRegINoSp tmp1, iRegINoSp tmp2,
iRegINoSp tmp3, iRegINoSp tmp4, rFlagsReg cr)
%{
match(Set result (StrIndexOfChar (Binary str1 cnt1) ch));
predicate(!UseRVV && (((StrIndexOfCharNode*)n)->encoding() == StrIntrinsicNode::U));
effect(USE_KILL str1, USE_KILL cnt1, USE_KILL ch, TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr);
format %{ "StringUTF16 IndexOf char[] $str1,$cnt1,$ch -> $result" %}
ins_encode %{
__ string_indexof_char($str1$$Register, $cnt1$$Register, $ch$$Register,
$result$$Register, $tmp1$$Register, $tmp2$$Register,
$tmp3$$Register, $tmp4$$Register, false /* isU */);
%}
ins_pipe(pipe_class_memory);
%}
instruct stringL_indexof_char(iRegP_R11 str1, iRegI_R12 cnt1, iRegI_R13 ch,
iRegI_R10 result, iRegINoSp tmp1, iRegINoSp tmp2,
iRegINoSp tmp3, iRegINoSp tmp4, rFlagsReg cr)
%{
match(Set result (StrIndexOfChar (Binary str1 cnt1) ch));
predicate(!UseRVV && (((StrIndexOfCharNode*)n)->encoding() == StrIntrinsicNode::L));
effect(USE_KILL str1, USE_KILL cnt1, USE_KILL ch, TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr);
format %{ "StringUTF16 IndexOf char[] $str1,$cnt1,$ch -> $result" %}
ins_encode %{
__ string_indexof_char($str1$$Register, $cnt1$$Register, $ch$$Register,
$result$$Register, $tmp1$$Register, $tmp2$$Register,
$tmp3$$Register, $tmp4$$Register, true /* isL */);
%}
ins_pipe(pipe_class_memory);
%}
// clearing of an array
instruct clearArray_reg_reg(iRegL_R29 cnt, iRegP_R28 base, iRegP_R30 tmp1,
iRegP_R31 tmp2, Universe dummy)
%{
// temp registers must match the one used in StubGenerator::generate_zero_blocks()
predicate(UseBlockZeroing || !UseRVV);
match(Set dummy (ClearArray cnt base));
effect(USE_KILL cnt, USE_KILL base, TEMP tmp1, TEMP tmp2);
ins_cost(4 * DEFAULT_COST);
format %{ "ClearArray $cnt, $base\t#@clearArray_reg_reg" %}
ins_encode %{
address tpc = __ zero_words($base$$Register, $cnt$$Register);
if (tpc == NULL) {
ciEnv::current()->record_failure("CodeCache is full");
return;
}
%}
ins_pipe(pipe_class_memory);
%}
instruct clearArray_imm_reg(immL cnt, iRegP_R28 base, Universe dummy, rFlagsReg cr)
%{
predicate(!UseRVV && (uint64_t)n->in(2)->get_long()
< (uint64_t)(BlockZeroingLowLimit >> LogBytesPerWord));
match(Set dummy (ClearArray cnt base));
effect(USE_KILL base, KILL cr);
ins_cost(4 * DEFAULT_COST);
format %{ "ClearArray $cnt, $base\t#@clearArray_imm_reg" %}
ins_encode %{
__ zero_words($base$$Register, (uint64_t)$cnt$$constant);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_equalsL(iRegP_R11 str1, iRegP_R13 str2, iRegI_R14 cnt,
iRegI_R10 result, rFlagsReg cr)
%{
predicate(!UseRVV && ((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::LL);
match(Set result (StrEquals (Binary str1 str2) cnt));
effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL cr);
format %{ "String Equals $str1, $str2, $cnt -> $result\t#@string_equalsL" %}
ins_encode %{
// Count is in 8-bit bytes; non-Compact chars are 16 bits.
__ string_equals($str1$$Register, $str2$$Register,
$result$$Register, $cnt$$Register, 1);
%}
ins_pipe(pipe_class_memory);
%}
instruct string_equalsU(iRegP_R11 str1, iRegP_R13 str2, iRegI_R14 cnt,
iRegI_R10 result, rFlagsReg cr)
%{
predicate(!UseRVV && ((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::UU);
match(Set result (StrEquals (Binary str1 str2) cnt));
effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL cr);
format %{ "String Equals $str1, $str2, $cnt -> $result\t#@string_equalsU" %}
ins_encode %{
// Count is in 8-bit bytes; non-Compact chars are 16 bits.
__ string_equals($str1$$Register, $str2$$Register,
$result$$Register, $cnt$$Register, 2);
%}
ins_pipe(pipe_class_memory);
%}
instruct array_equalsB(iRegP_R11 ary1, iRegP_R12 ary2, iRegI_R10 result,
iRegP_R13 tmp1, iRegP_R14 tmp2, iRegP_R15 tmp3,
iRegP_R16 tmp4, iRegP_R28 tmp5, rFlagsReg cr)
%{
predicate(!UseRVV && ((AryEqNode*)n)->encoding() == StrIntrinsicNode::LL);
match(Set result (AryEq ary1 ary2));
effect(USE_KILL ary1, USE_KILL ary2, TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL tmp5, KILL cr);
format %{ "Array Equals $ary1, ary2 -> $result\t#@array_equalsB // KILL $tmp5" %}
ins_encode %{
__ arrays_equals($ary1$$Register, $ary2$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register,
$result$$Register, $tmp5$$Register, 1);
%}
ins_pipe(pipe_class_memory);
%}
instruct array_equalsC(iRegP_R11 ary1, iRegP_R12 ary2, iRegI_R10 result,
iRegP_R13 tmp1, iRegP_R14 tmp2, iRegP_R15 tmp3,
iRegP_R16 tmp4, iRegP_R28 tmp5, rFlagsReg cr)
%{
predicate(!UseRVV && ((AryEqNode*)n)->encoding() == StrIntrinsicNode::UU);
match(Set result (AryEq ary1 ary2));
effect(USE_KILL ary1, USE_KILL ary2, TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL tmp5, KILL cr);
format %{ "Array Equals $ary1, ary2 -> $result\t#@array_equalsC // KILL $tmp5" %}
ins_encode %{
__ arrays_equals($ary1$$Register, $ary2$$Register,
$tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register,
$result$$Register, $tmp5$$Register, 2);
%}
ins_pipe(pipe_class_memory);
%}
// ============================================================================
// Safepoint Instructions
instruct safePoint(iRegP poll)
%{
match(SafePoint poll);
ins_cost(2 * LOAD_COST);
format %{
"lwu zr, [$poll]\t# Safepoint: poll for GC, #@safePoint"
%}
ins_encode %{
__ read_polling_page(as_Register($poll$$reg), 0, relocInfo::poll_type);
%}
ins_pipe(pipe_serial); // ins_pipe(iload_reg_mem);
%}
// ============================================================================
// This name is KNOWN by the ADLC and cannot be changed.
// The ADLC forces a 'TypeRawPtr::BOTTOM' output type
// for this guy.
instruct tlsLoadP(javaThread_RegP dst)
%{
match(Set dst (ThreadLocal));
ins_cost(0);
format %{ " -- \t// $dst=Thread::current(), empty, #@tlsLoadP" %}
size(0);
ins_encode( /*empty*/ );
ins_pipe(pipe_class_empty);
%}
// inlined locking and unlocking
// using t1 as the 'flag' register to bridge the BoolNode producers and consumers
instruct cmpFastLock(rFlagsReg cr, iRegP object, iRegP box, iRegPNoSp tmp1, iRegPNoSp tmp2)
%{
match(Set cr (FastLock object box));
effect(TEMP tmp1, TEMP tmp2);
ins_cost(LOAD_COST * 2 + STORE_COST * 3 + ALU_COST * 6 + BRANCH_COST * 3);
format %{ "fastlock $object,$box\t! kills $tmp1,$tmp2, #@cmpFastLock" %}
ins_encode(riscv_enc_fast_lock(object, box, tmp1, tmp2));
ins_pipe(pipe_serial);
%}
// using t1 as the 'flag' register to bridge the BoolNode producers and consumers
instruct cmpFastUnlock(rFlagsReg cr, iRegP object, iRegP box, iRegPNoSp tmp1, iRegPNoSp tmp2)
%{
match(Set cr (FastUnlock object box));
effect(TEMP tmp1, TEMP tmp2);
ins_cost(LOAD_COST * 2 + STORE_COST + ALU_COST * 2 + BRANCH_COST * 4);
format %{ "fastunlock $object,$box\t! kills $tmp1, $tmp2, #@cmpFastUnlock" %}
ins_encode(riscv_enc_fast_unlock(object, box, tmp1, tmp2));
ins_pipe(pipe_serial);
%}
// Tail Call; Jump from runtime stub to Java code.
// Also known as an 'interprocedural jump'.
// Target of jump will eventually return to caller.
// TailJump below removes the return address.
instruct TailCalljmpInd(iRegPNoSp jump_target, inline_cache_RegP method_oop)
%{
match(TailCall jump_target method_oop);
ins_cost(BRANCH_COST);
format %{ "jalr $jump_target\t# $method_oop holds method oop, #@TailCalljmpInd." %}
ins_encode(riscv_enc_tail_call(jump_target));
ins_pipe(pipe_class_call);
%}
instruct TailjmpInd(iRegPNoSp jump_target, iRegP_R10 ex_oop)
%{
match(TailJump jump_target ex_oop);
ins_cost(ALU_COST + BRANCH_COST);
format %{ "jalr $jump_target\t# $ex_oop holds exception oop, #@TailjmpInd." %}
ins_encode(riscv_enc_tail_jmp(jump_target));
ins_pipe(pipe_class_call);
%}
// Create exception oop: created by stack-crawling runtime code.
// Created exception is now available to this handler, and is setup
// just prior to jumping to this handler. No code emitted.
instruct CreateException(iRegP_R10 ex_oop)
%{
match(Set ex_oop (CreateEx));
ins_cost(0);
format %{ " -- \t// exception oop; no code emitted, #@CreateException" %}
size(0);
ins_encode( /*empty*/ );
ins_pipe(pipe_class_empty);
%}
// Rethrow exception: The exception oop will come in the first
// argument position. Then JUMP (not call) to the rethrow stub code.
instruct RethrowException()
%{
match(Rethrow);
ins_cost(BRANCH_COST);
format %{ "j rethrow_stub\t#@RethrowException" %}
ins_encode(riscv_enc_rethrow());
ins_pipe(pipe_class_call);
%}
// Return Instruction
// epilog node loads ret address into ra as part of frame pop
instruct Ret()
%{
match(Return);
ins_cost(BRANCH_COST);
format %{ "ret\t// return register, #@Ret" %}
ins_encode(riscv_enc_ret());
ins_pipe(pipe_branch);
%}
// Die now.
instruct ShouldNotReachHere() %{
match(Halt);
ins_cost(BRANCH_COST);
format %{ "#@ShouldNotReachHere" %}
ins_encode %{
if (is_reachable()) {
__ stop(_halt_reason);
}
%}
ins_pipe(pipe_class_default);
%}
//----------PEEPHOLE RULES-----------------------------------------------------
// These must follow all instruction definitions as they use the names
// defined in the instructions definitions.
//
// peepmatch ( root_instr_name [preceding_instruction]* );
//
// peepconstraint %{
// (instruction_number.operand_name relational_op instruction_number.operand_name
// [, ...] );
// // instruction numbers are zero-based using left to right order in peepmatch
//
// peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
// // provide an instruction_number.operand_name for each operand that appears
// // in the replacement instruction's match rule
//
// ---------VM FLAGS---------------------------------------------------------
//
// All peephole optimizations can be turned off using -XX:-OptoPeephole
//
// Each peephole rule is given an identifying number starting with zero and
// increasing by one in the order seen by the parser. An individual peephole
// can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
// on the command-line.
//
// ---------CURRENT LIMITATIONS----------------------------------------------
//
// Only match adjacent instructions in same basic block
// Only equality constraints
// Only constraints between operands, not (0.dest_reg == RAX_enc)
// Only one replacement instruction
//
//----------SMARTSPILL RULES---------------------------------------------------
// These must follow all instruction definitions as they use the names
// defined in the instructions definitions.
// Local Variables:
// mode: c++
// End:
[0.253QuellennavigatorsProjekt 2026-04-26]
|
2026-05-26
|
|
|
|
|
