Adding upstream version 1.18.10.upstream/1.18.10upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

6 files changed, 4391 insertions, 0 deletions

diff --git a/src/cmd/internal/obj/mips/a.out.go b/src/cmd/internal/obj/mips/a.out.go
new file mode 100644
index 0000000..c6ce53a
--- /dev/null
+++ b/src/cmd/internal/obj/mips/a.out.go
@@ -0,0 +1,483 @@
+// cmd/9c/9.out.h from Vita Nuova.
+//
+//	Copyright © 1994-1999 Lucent Technologies Inc.  All rights reserved.
+//	Portions Copyright © 1995-1997 C H Forsyth (forsyth@terzarima.net)
+//	Portions Copyright © 1997-1999 Vita Nuova Limited
+//	Portions Copyright © 2000-2008 Vita Nuova Holdings Limited (www.vitanuova.com)
+//	Portions Copyright © 2004,2006 Bruce Ellis
+//	Portions Copyright © 2005-2007 C H Forsyth (forsyth@terzarima.net)
+//	Revisions Copyright © 2000-2008 Lucent Technologies Inc. and others
+//	Portions Copyright © 2009 The Go Authors. All rights reserved.
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+// THE SOFTWARE.
+
+package mips
+
+import (
+	"cmd/internal/obj"
+)
+
+//go:generate go run ../stringer.go -i $GOFILE -o anames.go -p mips
+
+/*
+ * mips 64
+ */
+const (
+	NSNAME = 8
+	NSYM   = 50
+	NREG   = 32 /* number of general registers */
+	NFREG  = 32 /* number of floating point registers */
+	NWREG  = 32 /* number of MSA registers */
+)
+
+const (
+	REG_R0 = obj.RBaseMIPS + iota // must be a multiple of 32
+	REG_R1
+	REG_R2
+	REG_R3
+	REG_R4
+	REG_R5
+	REG_R6
+	REG_R7
+	REG_R8
+	REG_R9
+	REG_R10
+	REG_R11
+	REG_R12
+	REG_R13
+	REG_R14
+	REG_R15
+	REG_R16
+	REG_R17
+	REG_R18
+	REG_R19
+	REG_R20
+	REG_R21
+	REG_R22
+	REG_R23
+	REG_R24
+	REG_R25
+	REG_R26
+	REG_R27
+	REG_R28
+	REG_R29
+	REG_R30
+	REG_R31
+
+	REG_F0 // must be a multiple of 32
+	REG_F1
+	REG_F2
+	REG_F3
+	REG_F4
+	REG_F5
+	REG_F6
+	REG_F7
+	REG_F8
+	REG_F9
+	REG_F10
+	REG_F11
+	REG_F12
+	REG_F13
+	REG_F14
+	REG_F15
+	REG_F16
+	REG_F17
+	REG_F18
+	REG_F19
+	REG_F20
+	REG_F21
+	REG_F22
+	REG_F23
+	REG_F24
+	REG_F25
+	REG_F26
+	REG_F27
+	REG_F28
+	REG_F29
+	REG_F30
+	REG_F31
+
+	// co-processor 0 control registers
+	REG_M0 // must be a multiple of 32
+	REG_M1
+	REG_M2
+	REG_M3
+	REG_M4
+	REG_M5
+	REG_M6
+	REG_M7
+	REG_M8
+	REG_M9
+	REG_M10
+	REG_M11
+	REG_M12
+	REG_M13
+	REG_M14
+	REG_M15
+	REG_M16
+	REG_M17
+	REG_M18
+	REG_M19
+	REG_M20
+	REG_M21
+	REG_M22
+	REG_M23
+	REG_M24
+	REG_M25
+	REG_M26
+	REG_M27
+	REG_M28
+	REG_M29
+	REG_M30
+	REG_M31
+
+	// FPU control registers
+	REG_FCR0 // must be a multiple of 32
+	REG_FCR1
+	REG_FCR2
+	REG_FCR3
+	REG_FCR4
+	REG_FCR5
+	REG_FCR6
+	REG_FCR7
+	REG_FCR8
+	REG_FCR9
+	REG_FCR10
+	REG_FCR11
+	REG_FCR12
+	REG_FCR13
+	REG_FCR14
+	REG_FCR15
+	REG_FCR16
+	REG_FCR17
+	REG_FCR18
+	REG_FCR19
+	REG_FCR20
+	REG_FCR21
+	REG_FCR22
+	REG_FCR23
+	REG_FCR24
+	REG_FCR25
+	REG_FCR26
+	REG_FCR27
+	REG_FCR28
+	REG_FCR29
+	REG_FCR30
+	REG_FCR31
+
+	// MSA registers
+	// The lower bits of W registers are alias to F registers
+	REG_W0 // must be a multiple of 32
+	REG_W1
+	REG_W2
+	REG_W3
+	REG_W4
+	REG_W5
+	REG_W6
+	REG_W7
+	REG_W8
+	REG_W9
+	REG_W10
+	REG_W11
+	REG_W12
+	REG_W13
+	REG_W14
+	REG_W15
+	REG_W16
+	REG_W17
+	REG_W18
+	REG_W19
+	REG_W20
+	REG_W21
+	REG_W22
+	REG_W23
+	REG_W24
+	REG_W25
+	REG_W26
+	REG_W27
+	REG_W28
+	REG_W29
+	REG_W30
+	REG_W31
+
+	REG_HI
+	REG_LO
+
+	REG_LAST = REG_LO // the last defined register
+
+	REG_SPECIAL = REG_M0
+
+	REGZERO = REG_R0 /* set to zero */
+	REGSP   = REG_R29
+	REGSB   = REG_R28
+	REGLINK = REG_R31
+	REGRET  = REG_R1
+	REGARG  = -1      /* -1 disables passing the first argument in register */
+	REGRT1  = REG_R1  /* reserved for runtime, duffzero and duffcopy */
+	REGRT2  = REG_R2  /* reserved for runtime, duffcopy */
+	REGCTXT = REG_R22 /* context for closures */
+	REGG    = REG_R30 /* G */
+	REGTMP  = REG_R23 /* used by the linker */
+	FREGRET = REG_F0
+)
+
+// https://llvm.org/svn/llvm-project/llvm/trunk/lib/Target/Mips/MipsRegisterInfo.td search for DwarfRegNum
+// https://gcc.gnu.org/viewcvs/gcc/trunk/gcc/config/mips/mips.c?view=co&revision=258099&content-type=text%2Fplain search for mips_dwarf_regno
+// For now, this is adequate for both 32 and 64 bit.
+var MIPSDWARFRegisters = map[int16]int16{}
+
+func init() {
+	// f assigns dwarfregisters[from:to] = (base):(to-from+base)
+	f := func(from, to, base int16) {
+		for r := int16(from); r <= to; r++ {
+			MIPSDWARFRegisters[r] = (r - from) + base
+		}
+	}
+	f(REG_R0, REG_R31, 0)
+	f(REG_F0, REG_F31, 32) // For 32-bit MIPS, compiler only uses even numbered registers --  see cmd/compile/internal/ssa/gen/MIPSOps.go
+	MIPSDWARFRegisters[REG_HI] = 64
+	MIPSDWARFRegisters[REG_LO] = 65
+	// The lower bits of W registers are alias to F registers
+	f(REG_W0, REG_W31, 32)
+}
+
+const (
+	BIG = 32766
+)
+
+const (
+	/* mark flags */
+	FOLL    = 1 << 0
+	LABEL   = 1 << 1
+	LEAF    = 1 << 2
+	SYNC    = 1 << 3
+	BRANCH  = 1 << 4
+	LOAD    = 1 << 5
+	FCMP    = 1 << 6
+	NOSCHED = 1 << 7
+
+	NSCHED = 20
+)
+
+const (
+	C_NONE = iota
+	C_REG
+	C_FREG
+	C_FCREG
+	C_MREG /* special processor register */
+	C_WREG /* MSA registers */
+	C_HI
+	C_LO
+	C_ZCON
+	C_SCON /* 16 bit signed */
+	C_UCON /* 32 bit signed, low 16 bits 0 */
+	C_ADD0CON
+	C_AND0CON
+	C_ADDCON /* -0x8000 <= v < 0 */
+	C_ANDCON /* 0 < v <= 0xFFFF */
+	C_LCON   /* other 32 */
+	C_DCON   /* other 64 (could subdivide further) */
+	C_SACON  /* $n(REG) where n <= int16 */
+	C_SECON
+	C_LACON /* $n(REG) where int16 < n <= int32 */
+	C_LECON
+	C_DACON /* $n(REG) where int32 < n */
+	C_STCON /* $tlsvar */
+	C_SBRA
+	C_LBRA
+	C_SAUTO
+	C_LAUTO
+	C_SEXT
+	C_LEXT
+	C_ZOREG
+	C_SOREG
+	C_LOREG
+	C_GOK
+	C_ADDR
+	C_TLS
+	C_TEXTSIZE
+
+	C_NCLASS /* must be the last */
+)
+
+const (
+	AABSD = obj.ABaseMIPS + obj.A_ARCHSPECIFIC + iota
+	AABSF
+	AABSW
+	AADD
+	AADDD
+	AADDF
+	AADDU
+	AADDW
+	AAND
+	ABEQ
+	ABFPF
+	ABFPT
+	ABGEZ
+	ABGEZAL
+	ABGTZ
+	ABLEZ
+	ABLTZ
+	ABLTZAL
+	ABNE
+	ABREAK
+	ACLO
+	ACLZ
+	ACMOVF
+	ACMOVN
+	ACMOVT
+	ACMOVZ
+	ACMPEQD
+	ACMPEQF
+	ACMPGED
+	ACMPGEF
+	ACMPGTD
+	ACMPGTF
+	ADIV
+	ADIVD
+	ADIVF
+	ADIVU
+	ADIVW
+	AGOK
+	ALL
+	ALLV
+	ALUI
+	AMADD
+	AMOVB
+	AMOVBU
+	AMOVD
+	AMOVDF
+	AMOVDW
+	AMOVF
+	AMOVFD
+	AMOVFW
+	AMOVH
+	AMOVHU
+	AMOVW
+	AMOVWD
+	AMOVWF
+	AMOVWL
+	AMOVWR
+	AMSUB
+	AMUL
+	AMULD
+	AMULF
+	AMULU
+	AMULW
+	ANEGD
+	ANEGF
+	ANEGW
+	ANEGV
+	ANOOP // hardware nop
+	ANOR
+	AOR
+	AREM
+	AREMU
+	ARFE
+	AROTR
+	AROTRV
+	ASC
+	ASCV
+	ASGT
+	ASGTU
+	ASLL
+	ASQRTD
+	ASQRTF
+	ASRA
+	ASRL
+	ASUB
+	ASUBD
+	ASUBF
+	ASUBU
+	ASUBW
+	ASYNC
+	ASYSCALL
+	ATEQ
+	ATLBP
+	ATLBR
+	ATLBWI
+	ATLBWR
+	ATNE
+	AWORD
+	AXOR
+
+	/* 64-bit */
+	AMOVV
+	AMOVVL
+	AMOVVR
+	ASLLV
+	ASRAV
+	ASRLV
+	ADIVV
+	ADIVVU
+	AREMV
+	AREMVU
+	AMULV
+	AMULVU
+	AADDV
+	AADDVU
+	ASUBV
+	ASUBVU
+
+	/* 64-bit FP */
+	ATRUNCFV
+	ATRUNCDV
+	ATRUNCFW
+	ATRUNCDW
+	AMOVWU
+	AMOVFV
+	AMOVDV
+	AMOVVF
+	AMOVVD
+
+	/* MSA */
+	AVMOVB
+	AVMOVH
+	AVMOVW
+	AVMOVD
+
+	ALAST
+
+	// aliases
+	AJMP = obj.AJMP
+	AJAL = obj.ACALL
+	ARET = obj.ARET
+)
+
+func init() {
+	// The asm encoder generally assumes that the lowest 5 bits of the
+	// REG_XX constants match the machine instruction encoding, i.e.
+	// the lowest 5 bits is the register number.
+	// Check this here.
+	if REG_R0%32 != 0 {
+		panic("REG_R0 is not a multiple of 32")
+	}
+	if REG_F0%32 != 0 {
+		panic("REG_F0 is not a multiple of 32")
+	}
+	if REG_M0%32 != 0 {
+		panic("REG_M0 is not a multiple of 32")
+	}
+	if REG_FCR0%32 != 0 {
+		panic("REG_FCR0 is not a multiple of 32")
+	}
+	if REG_W0%32 != 0 {
+		panic("REG_W0 is not a multiple of 32")
+	}
+}
diff --git a/src/cmd/internal/obj/mips/anames.go b/src/cmd/internal/obj/mips/anames.go
new file mode 100644
index 0000000..ca2ad5a
--- /dev/null
+++ b/src/cmd/internal/obj/mips/anames.go
@@ -0,0 +1,137 @@
+// Code generated by stringer -i a.out.go -o anames.go -p mips; DO NOT EDIT.
+
+package mips
+
+import "cmd/internal/obj"
+
+var Anames = []string{
+	obj.A_ARCHSPECIFIC: "ABSD",
+	"ABSF",
+	"ABSW",
+	"ADD",
+	"ADDD",
+	"ADDF",
+	"ADDU",
+	"ADDW",
+	"AND",
+	"BEQ",
+	"BFPF",
+	"BFPT",
+	"BGEZ",
+	"BGEZAL",
+	"BGTZ",
+	"BLEZ",
+	"BLTZ",
+	"BLTZAL",
+	"BNE",
+	"BREAK",
+	"CLO",
+	"CLZ",
+	"CMOVF",
+	"CMOVN",
+	"CMOVT",
+	"CMOVZ",
+	"CMPEQD",
+	"CMPEQF",
+	"CMPGED",
+	"CMPGEF",
+	"CMPGTD",
+	"CMPGTF",
+	"DIV",
+	"DIVD",
+	"DIVF",
+	"DIVU",
+	"DIVW",
+	"GOK",
+	"LL",
+	"LLV",
+	"LUI",
+	"MADD",
+	"MOVB",
+	"MOVBU",
+	"MOVD",
+	"MOVDF",
+	"MOVDW",
+	"MOVF",
+	"MOVFD",
+	"MOVFW",
+	"MOVH",
+	"MOVHU",
+	"MOVW",
+	"MOVWD",
+	"MOVWF",
+	"MOVWL",
+	"MOVWR",
+	"MSUB",
+	"MUL",
+	"MULD",
+	"MULF",
+	"MULU",
+	"MULW",
+	"NEGD",
+	"NEGF",
+	"NEGW",
+	"NEGV",
+	"NOOP",
+	"NOR",
+	"OR",
+	"REM",
+	"REMU",
+	"RFE",
+	"ROTR",
+	"ROTRV",
+	"SC",
+	"SCV",
+	"SGT",
+	"SGTU",
+	"SLL",
+	"SQRTD",
+	"SQRTF",
+	"SRA",
+	"SRL",
+	"SUB",
+	"SUBD",
+	"SUBF",
+	"SUBU",
+	"SUBW",
+	"SYNC",
+	"SYSCALL",
+	"TEQ",
+	"TLBP",
+	"TLBR",
+	"TLBWI",
+	"TLBWR",
+	"TNE",
+	"WORD",
+	"XOR",
+	"MOVV",
+	"MOVVL",
+	"MOVVR",
+	"SLLV",
+	"SRAV",
+	"SRLV",
+	"DIVV",
+	"DIVVU",
+	"REMV",
+	"REMVU",
+	"MULV",
+	"MULVU",
+	"ADDV",
+	"ADDVU",
+	"SUBV",
+	"SUBVU",
+	"TRUNCFV",
+	"TRUNCDV",
+	"TRUNCFW",
+	"TRUNCDW",
+	"MOVWU",
+	"MOVFV",
+	"MOVDV",
+	"MOVVF",
+	"MOVVD",
+	"VMOVB",
+	"VMOVH",
+	"VMOVW",
+	"VMOVD",
+	"LAST",
+}
diff --git a/src/cmd/internal/obj/mips/anames0.go b/src/cmd/internal/obj/mips/anames0.go
new file mode 100644
index 0000000..c300696
--- /dev/null
+++ b/src/cmd/internal/obj/mips/anames0.go
@@ -0,0 +1,45 @@
+// Copyright 2015 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package mips
+
+var cnames0 = []string{
+	"NONE",
+	"REG",
+	"FREG",
+	"FCREG",
+	"MREG",
+	"WREG",
+	"HI",
+	"LO",
+	"ZCON",
+	"SCON",
+	"UCON",
+	"ADD0CON",
+	"AND0CON",
+	"ADDCON",
+	"ANDCON",
+	"LCON",
+	"DCON",
+	"SACON",
+	"SECON",
+	"LACON",
+	"LECON",
+	"DACON",
+	"STCON",
+	"SBRA",
+	"LBRA",
+	"SAUTO",
+	"LAUTO",
+	"SEXT",
+	"LEXT",
+	"ZOREG",
+	"SOREG",
+	"LOREG",
+	"GOK",
+	"ADDR",
+	"TLS",
+	"TEXTSIZE",
+	"NCLASS",
+}
diff --git a/src/cmd/internal/obj/mips/asm0.go b/src/cmd/internal/obj/mips/asm0.go
new file mode 100644
index 0000000..e475ffd
--- /dev/null
+++ b/src/cmd/internal/obj/mips/asm0.go
@@ -0,0 +1,2121 @@
+// cmd/9l/optab.c, cmd/9l/asmout.c from Vita Nuova.
+//
+//	Copyright © 1994-1999 Lucent Technologies Inc.  All rights reserved.
+//	Portions Copyright © 1995-1997 C H Forsyth (forsyth@terzarima.net)
+//	Portions Copyright © 1997-1999 Vita Nuova Limited
+//	Portions Copyright © 2000-2008 Vita Nuova Holdings Limited (www.vitanuova.com)
+//	Portions Copyright © 2004,2006 Bruce Ellis
+//	Portions Copyright © 2005-2007 C H Forsyth (forsyth@terzarima.net)
+//	Revisions Copyright © 2000-2008 Lucent Technologies Inc. and others
+//	Portions Copyright © 2009 The Go Authors. All rights reserved.
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+// THE SOFTWARE.
+
+package mips
+
+import (
+	"cmd/internal/obj"
+	"cmd/internal/objabi"
+	"cmd/internal/sys"
+	"fmt"
+	"log"
+	"sort"
+)
+
+// ctxt0 holds state while assembling a single function.
+// Each function gets a fresh ctxt0.
+// This allows for multiple functions to be safely concurrently assembled.
+type ctxt0 struct {
+	ctxt       *obj.Link
+	newprog    obj.ProgAlloc
+	cursym     *obj.LSym
+	autosize   int32
+	instoffset int64
+	pc         int64
+}
+
+// Instruction layout.
+
+const (
+	mips64FuncAlign = 8
+)
+
+const (
+	r0iszero = 1
+)
+
+type Optab struct {
+	as     obj.As
+	a1     uint8
+	a2     uint8
+	a3     uint8
+	type_  int8
+	size   int8
+	param  int16
+	family sys.ArchFamily // 0 means both sys.MIPS and sys.MIPS64
+	flag   uint8
+}
+
+const (
+	// Optab.flag
+	NOTUSETMP = 1 << iota // p expands to multiple instructions, but does NOT use REGTMP
+)
+
+var optab = []Optab{
+	{obj.ATEXT, C_LEXT, C_NONE, C_TEXTSIZE, 0, 0, 0, sys.MIPS64, 0},
+	{obj.ATEXT, C_ADDR, C_NONE, C_TEXTSIZE, 0, 0, 0, 0, 0},
+
+	{AMOVW, C_REG, C_NONE, C_REG, 1, 4, 0, 0, 0},
+	{AMOVV, C_REG, C_NONE, C_REG, 1, 4, 0, sys.MIPS64, 0},
+	{AMOVB, C_REG, C_NONE, C_REG, 12, 8, 0, 0, NOTUSETMP},
+	{AMOVBU, C_REG, C_NONE, C_REG, 13, 4, 0, 0, 0},
+	{AMOVWU, C_REG, C_NONE, C_REG, 14, 8, 0, sys.MIPS64, NOTUSETMP},
+
+	{ASUB, C_REG, C_REG, C_REG, 2, 4, 0, 0, 0},
+	{ASUBV, C_REG, C_REG, C_REG, 2, 4, 0, sys.MIPS64, 0},
+	{AADD, C_REG, C_REG, C_REG, 2, 4, 0, 0, 0},
+	{AADDV, C_REG, C_REG, C_REG, 2, 4, 0, sys.MIPS64, 0},
+	{AAND, C_REG, C_REG, C_REG, 2, 4, 0, 0, 0},
+	{ASUB, C_REG, C_NONE, C_REG, 2, 4, 0, 0, 0},
+	{ASUBV, C_REG, C_NONE, C_REG, 2, 4, 0, sys.MIPS64, 0},
+	{AADD, C_REG, C_NONE, C_REG, 2, 4, 0, 0, 0},
+	{AADDV, C_REG, C_NONE, C_REG, 2, 4, 0, sys.MIPS64, 0},
+	{AAND, C_REG, C_NONE, C_REG, 2, 4, 0, 0, 0},
+	{ACMOVN, C_REG, C_REG, C_REG, 2, 4, 0, 0, 0},
+	{ANEGW, C_REG, C_NONE, C_REG, 2, 4, 0, 0, 0},
+	{ANEGV, C_REG, C_NONE, C_REG, 2, 4, 0, sys.MIPS64, 0},
+
+	{ASLL, C_REG, C_NONE, C_REG, 9, 4, 0, 0, 0},
+	{ASLL, C_REG, C_REG, C_REG, 9, 4, 0, 0, 0},
+	{ASLLV, C_REG, C_NONE, C_REG, 9, 4, 0, sys.MIPS64, 0},
+	{ASLLV, C_REG, C_REG, C_REG, 9, 4, 0, sys.MIPS64, 0},
+	{ACLO, C_REG, C_NONE, C_REG, 9, 4, 0, 0, 0},
+
+	{AADDF, C_FREG, C_NONE, C_FREG, 32, 4, 0, 0, 0},
+	{AADDF, C_FREG, C_REG, C_FREG, 32, 4, 0, 0, 0},
+	{ACMPEQF, C_FREG, C_REG, C_NONE, 32, 4, 0, 0, 0},
+	{AABSF, C_FREG, C_NONE, C_FREG, 33, 4, 0, 0, 0},
+	{AMOVVF, C_FREG, C_NONE, C_FREG, 33, 4, 0, sys.MIPS64, 0},
+	{AMOVF, C_FREG, C_NONE, C_FREG, 33, 4, 0, 0, 0},
+	{AMOVD, C_FREG, C_NONE, C_FREG, 33, 4, 0, 0, 0},
+
+	{AMOVW, C_REG, C_NONE, C_SEXT, 7, 4, REGSB, sys.MIPS64, 0},
+	{AMOVWU, C_REG, C_NONE, C_SEXT, 7, 4, REGSB, sys.MIPS64, 0},
+	{AMOVV, C_REG, C_NONE, C_SEXT, 7, 4, REGSB, sys.MIPS64, 0},
+	{AMOVB, C_REG, C_NONE, C_SEXT, 7, 4, REGSB, sys.MIPS64, 0},
+	{AMOVBU, C_REG, C_NONE, C_SEXT, 7, 4, REGSB, sys.MIPS64, 0},
+	{AMOVWL, C_REG, C_NONE, C_SEXT, 7, 4, REGSB, sys.MIPS64, 0},
+	{AMOVVL, C_REG, C_NONE, C_SEXT, 7, 4, REGSB, sys.MIPS64, 0},
+	{AMOVW, C_REG, C_NONE, C_SAUTO, 7, 4, REGSP, 0, 0},
+	{AMOVWU, C_REG, C_NONE, C_SAUTO, 7, 4, REGSP, sys.MIPS64, 0},
+	{AMOVV, C_REG, C_NONE, C_SAUTO, 7, 4, REGSP, sys.MIPS64, 0},
+	{AMOVB, C_REG, C_NONE, C_SAUTO, 7, 4, REGSP, 0, 0},
+	{AMOVBU, C_REG, C_NONE, C_SAUTO, 7, 4, REGSP, 0, 0},
+	{AMOVWL, C_REG, C_NONE, C_SAUTO, 7, 4, REGSP, 0, 0},
+	{AMOVVL, C_REG, C_NONE, C_SAUTO, 7, 4, REGSP, sys.MIPS64, 0},
+	{AMOVW, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, 0, 0},
+	{AMOVWU, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, sys.MIPS64, 0},
+	{AMOVV, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, sys.MIPS64, 0},
+	{AMOVB, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, 0, 0},
+	{AMOVBU, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, 0, 0},
+	{AMOVWL, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, 0, 0},
+	{AMOVVL, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, sys.MIPS64, 0},
+	{ASC, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, 0, 0},
+	{ASCV, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, sys.MIPS64, 0},
+
+	{AMOVW, C_SEXT, C_NONE, C_REG, 8, 4, REGSB, sys.MIPS64, 0},
+	{AMOVWU, C_SEXT, C_NONE, C_REG, 8, 4, REGSB, sys.MIPS64, 0},
+	{AMOVV, C_SEXT, C_NONE, C_REG, 8, 4, REGSB, sys.MIPS64, 0},
+	{AMOVB, C_SEXT, C_NONE, C_REG, 8, 4, REGSB, sys.MIPS64, 0},
+	{AMOVBU, C_SEXT, C_NONE, C_REG, 8, 4, REGSB, sys.MIPS64, 0},
+	{AMOVWL, C_SEXT, C_NONE, C_REG, 8, 4, REGSB, sys.MIPS64, 0},
+	{AMOVVL, C_SEXT, C_NONE, C_REG, 8, 4, REGSB, sys.MIPS64, 0},
+	{AMOVW, C_SAUTO, C_NONE, C_REG, 8, 4, REGSP, 0, 0},
+	{AMOVWU, C_SAUTO, C_NONE, C_REG, 8, 4, REGSP, sys.MIPS64, 0},
+	{AMOVV, C_SAUTO, C_NONE, C_REG, 8, 4, REGSP, sys.MIPS64, 0},
+	{AMOVB, C_SAUTO, C_NONE, C_REG, 8, 4, REGSP, 0, 0},
+	{AMOVBU, C_SAUTO, C_NONE, C_REG, 8, 4, REGSP, 0, 0},
+	{AMOVWL, C_SAUTO, C_NONE, C_REG, 8, 4, REGSP, 0, 0},
+	{AMOVVL, C_SAUTO, C_NONE, C_REG, 8, 4, REGSP, sys.MIPS64, 0},
+	{AMOVW, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, 0, 0},
+	{AMOVWU, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, sys.MIPS64, 0},
+	{AMOVV, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, sys.MIPS64, 0},
+	{AMOVB, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, 0, 0},
+	{AMOVBU, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, 0, 0},
+	{AMOVWL, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, 0, 0},
+	{AMOVVL, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, sys.MIPS64, 0},
+	{ALL, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, 0, 0},
+	{ALLV, C_SOREG, C_NONE, C_REG, 8, 4, REGZERO, sys.MIPS64, 0},
+
+	{AMOVW, C_REG, C_NONE, C_LEXT, 35, 12, REGSB, sys.MIPS64, 0},
+	{AMOVWU, C_REG, C_NONE, C_LEXT, 35, 12, REGSB, sys.MIPS64, 0},
+	{AMOVV, C_REG, C_NONE, C_LEXT, 35, 12, REGSB, sys.MIPS64, 0},
+	{AMOVB, C_REG, C_NONE, C_LEXT, 35, 12, REGSB, sys.MIPS64, 0},
+	{AMOVBU, C_REG, C_NONE, C_LEXT, 35, 12, REGSB, sys.MIPS64, 0},
+	{AMOVW, C_REG, C_NONE, C_LAUTO, 35, 12, REGSP, 0, 0},
+	{AMOVWU, C_REG, C_NONE, C_LAUTO, 35, 12, REGSP, sys.MIPS64, 0},
+	{AMOVV, C_REG, C_NONE, C_LAUTO, 35, 12, REGSP, sys.MIPS64, 0},
+	{AMOVB, C_REG, C_NONE, C_LAUTO, 35, 12, REGSP, 0, 0},
+	{AMOVBU, C_REG, C_NONE, C_LAUTO, 35, 12, REGSP, 0, 0},
+	{AMOVW, C_REG, C_NONE, C_LOREG, 35, 12, REGZERO, 0, 0},
+	{AMOVWU, C_REG, C_NONE, C_LOREG, 35, 12, REGZERO, sys.MIPS64, 0},
+	{AMOVV, C_REG, C_NONE, C_LOREG, 35, 12, REGZERO, sys.MIPS64, 0},
+	{AMOVB, C_REG, C_NONE, C_LOREG, 35, 12, REGZERO, 0, 0},
+	{AMOVBU, C_REG, C_NONE, C_LOREG, 35, 12, REGZERO, 0, 0},
+	{ASC, C_REG, C_NONE, C_LOREG, 35, 12, REGZERO, 0, 0},
+	{AMOVW, C_REG, C_NONE, C_ADDR, 50, 8, 0, sys.MIPS, 0},
+	{AMOVW, C_REG, C_NONE, C_ADDR, 50, 12, 0, sys.MIPS64, 0},
+	{AMOVWU, C_REG, C_NONE, C_ADDR, 50, 12, 0, sys.MIPS64, 0},
+	{AMOVV, C_REG, C_NONE, C_ADDR, 50, 12, 0, sys.MIPS64, 0},
+	{AMOVB, C_REG, C_NONE, C_ADDR, 50, 8, 0, sys.MIPS, 0},
+	{AMOVB, C_REG, C_NONE, C_ADDR, 50, 12, 0, sys.MIPS64, 0},
+	{AMOVBU, C_REG, C_NONE, C_ADDR, 50, 8, 0, sys.MIPS, 0},
+	{AMOVBU, C_REG, C_NONE, C_ADDR, 50, 12, 0, sys.MIPS64, 0},
+	{AMOVW, C_REG, C_NONE, C_TLS, 53, 8, 0, 0, NOTUSETMP},
+	{AMOVWU, C_REG, C_NONE, C_TLS, 53, 8, 0, sys.MIPS64, NOTUSETMP},
+	{AMOVV, C_REG, C_NONE, C_TLS, 53, 8, 0, sys.MIPS64, NOTUSETMP},
+	{AMOVB, C_REG, C_NONE, C_TLS, 53, 8, 0, 0, NOTUSETMP},
+	{AMOVBU, C_REG, C_NONE, C_TLS, 53, 8, 0, 0, NOTUSETMP},
+
+	{AMOVW, C_LEXT, C_NONE, C_REG, 36, 12, REGSB, sys.MIPS64, 0},
+	{AMOVWU, C_LEXT, C_NONE, C_REG, 36, 12, REGSB, sys.MIPS64, 0},
+	{AMOVV, C_LEXT, C_NONE, C_REG, 36, 12, REGSB, sys.MIPS64, 0},
+	{AMOVB, C_LEXT, C_NONE, C_REG, 36, 12, REGSB, sys.MIPS64, 0},
+	{AMOVBU, C_LEXT, C_NONE, C_REG, 36, 12, REGSB, sys.MIPS64, 0},
+	{AMOVW, C_LAUTO, C_NONE, C_REG, 36, 12, REGSP, 0, 0},
+	{AMOVWU, C_LAUTO, C_NONE, C_REG, 36, 12, REGSP, sys.MIPS64, 0},
+	{AMOVV, C_LAUTO, C_NONE, C_REG, 36, 12, REGSP, sys.MIPS64, 0},
+	{AMOVB, C_LAUTO, C_NONE, C_REG, 36, 12, REGSP, 0, 0},
+	{AMOVBU, C_LAUTO, C_NONE, C_REG, 36, 12, REGSP, 0, 0},
+	{AMOVW, C_LOREG, C_NONE, C_REG, 36, 12, REGZERO, 0, 0},
+	{AMOVWU, C_LOREG, C_NONE, C_REG, 36, 12, REGZERO, sys.MIPS64, 0},
+	{AMOVV, C_LOREG, C_NONE, C_REG, 36, 12, REGZERO, sys.MIPS64, 0},
+	{AMOVB, C_LOREG, C_NONE, C_REG, 36, 12, REGZERO, 0, 0},
+	{AMOVBU, C_LOREG, C_NONE, C_REG, 36, 12, REGZERO, 0, 0},
+	{AMOVW, C_ADDR, C_NONE, C_REG, 51, 8, 0, sys.MIPS, 0},
+	{AMOVW, C_ADDR, C_NONE, C_REG, 51, 12, 0, sys.MIPS64, 0},
+	{AMOVWU, C_ADDR, C_NONE, C_REG, 51, 12, 0, sys.MIPS64, 0},
+	{AMOVV, C_ADDR, C_NONE, C_REG, 51, 12, 0, sys.MIPS64, 0},
+	{AMOVB, C_ADDR, C_NONE, C_REG, 51, 8, 0, sys.MIPS, 0},
+	{AMOVB, C_ADDR, C_NONE, C_REG, 51, 12, 0, sys.MIPS64, 0},
+	{AMOVBU, C_ADDR, C_NONE, C_REG, 51, 8, 0, sys.MIPS, 0},
+	{AMOVBU, C_ADDR, C_NONE, C_REG, 51, 12, 0, sys.MIPS64, 0},
+	{AMOVW, C_TLS, C_NONE, C_REG, 54, 8, 0, 0, NOTUSETMP},
+	{AMOVWU, C_TLS, C_NONE, C_REG, 54, 8, 0, sys.MIPS64, NOTUSETMP},
+	{AMOVV, C_TLS, C_NONE, C_REG, 54, 8, 0, sys.MIPS64, NOTUSETMP},
+	{AMOVB, C_TLS, C_NONE, C_REG, 54, 8, 0, 0, NOTUSETMP},
+	{AMOVBU, C_TLS, C_NONE, C_REG, 54, 8, 0, 0, NOTUSETMP},
+
+	{AMOVW, C_SECON, C_NONE, C_REG, 3, 4, REGSB, sys.MIPS64, 0},
+	{AMOVV, C_SECON, C_NONE, C_REG, 3, 4, REGSB, sys.MIPS64, 0},
+	{AMOVW, C_SACON, C_NONE, C_REG, 3, 4, REGSP, 0, 0},
+	{AMOVV, C_SACON, C_NONE, C_REG, 3, 4, REGSP, sys.MIPS64, 0},
+	{AMOVW, C_LECON, C_NONE, C_REG, 52, 8, REGSB, sys.MIPS, NOTUSETMP},
+	{AMOVW, C_LECON, C_NONE, C_REG, 52, 12, REGSB, sys.MIPS64, NOTUSETMP},
+	{AMOVV, C_LECON, C_NONE, C_REG, 52, 12, REGSB, sys.MIPS64, NOTUSETMP},
+
+	{AMOVW, C_LACON, C_NONE, C_REG, 26, 12, REGSP, 0, 0},
+	{AMOVV, C_LACON, C_NONE, C_REG, 26, 12, REGSP, sys.MIPS64, 0},
+	{AMOVW, C_ADDCON, C_NONE, C_REG, 3, 4, REGZERO, 0, 0},
+	{AMOVV, C_ADDCON, C_NONE, C_REG, 3, 4, REGZERO, sys.MIPS64, 0},
+	{AMOVW, C_ANDCON, C_NONE, C_REG, 3, 4, REGZERO, 0, 0},
+	{AMOVV, C_ANDCON, C_NONE, C_REG, 3, 4, REGZERO, sys.MIPS64, 0},
+	{AMOVW, C_STCON, C_NONE, C_REG, 55, 8, 0, 0, NOTUSETMP},
+	{AMOVV, C_STCON, C_NONE, C_REG, 55, 8, 0, sys.MIPS64, NOTUSETMP},
+
+	{AMOVW, C_UCON, C_NONE, C_REG, 24, 4, 0, 0, 0},
+	{AMOVV, C_UCON, C_NONE, C_REG, 24, 4, 0, sys.MIPS64, 0},
+	{AMOVW, C_LCON, C_NONE, C_REG, 19, 8, 0, 0, NOTUSETMP},
+	{AMOVV, C_LCON, C_NONE, C_REG, 19, 8, 0, sys.MIPS64, NOTUSETMP},
+
+	{AMOVW, C_HI, C_NONE, C_REG, 20, 4, 0, 0, 0},
+	{AMOVV, C_HI, C_NONE, C_REG, 20, 4, 0, sys.MIPS64, 0},
+	{AMOVW, C_LO, C_NONE, C_REG, 20, 4, 0, 0, 0},
+	{AMOVV, C_LO, C_NONE, C_REG, 20, 4, 0, sys.MIPS64, 0},
+	{AMOVW, C_REG, C_NONE, C_HI, 21, 4, 0, 0, 0},
+	{AMOVV, C_REG, C_NONE, C_HI, 21, 4, 0, sys.MIPS64, 0},
+	{AMOVW, C_REG, C_NONE, C_LO, 21, 4, 0, 0, 0},
+	{AMOVV, C_REG, C_NONE, C_LO, 21, 4, 0, sys.MIPS64, 0},
+
+	{AMUL, C_REG, C_REG, C_NONE, 22, 4, 0, 0, 0},
+	{AMUL, C_REG, C_REG, C_REG, 22, 4, 0, 0, 0},
+	{AMULV, C_REG, C_REG, C_NONE, 22, 4, 0, sys.MIPS64, 0},
+
+	{AADD, C_ADD0CON, C_REG, C_REG, 4, 4, 0, 0, 0},
+	{AADD, C_ADD0CON, C_NONE, C_REG, 4, 4, 0, 0, 0},
+	{AADD, C_ANDCON, C_REG, C_REG, 10, 8, 0, 0, 0},
+	{AADD, C_ANDCON, C_NONE, C_REG, 10, 8, 0, 0, 0},
+
+	{AADDV, C_ADD0CON, C_REG, C_REG, 4, 4, 0, sys.MIPS64, 0},
+	{AADDV, C_ADD0CON, C_NONE, C_REG, 4, 4, 0, sys.MIPS64, 0},
+	{AADDV, C_ANDCON, C_REG, C_REG, 10, 8, 0, sys.MIPS64, 0},
+	{AADDV, C_ANDCON, C_NONE, C_REG, 10, 8, 0, sys.MIPS64, 0},
+
+	{AAND, C_AND0CON, C_REG, C_REG, 4, 4, 0, 0, 0},
+	{AAND, C_AND0CON, C_NONE, C_REG, 4, 4, 0, 0, 0},
+	{AAND, C_ADDCON, C_REG, C_REG, 10, 8, 0, 0, 0},
+	{AAND, C_ADDCON, C_NONE, C_REG, 10, 8, 0, 0, 0},
+
+	{AADD, C_UCON, C_REG, C_REG, 25, 8, 0, 0, 0},
+	{AADD, C_UCON, C_NONE, C_REG, 25, 8, 0, 0, 0},
+	{AADDV, C_UCON, C_REG, C_REG, 25, 8, 0, sys.MIPS64, 0},
+	{AADDV, C_UCON, C_NONE, C_REG, 25, 8, 0, sys.MIPS64, 0},
+	{AAND, C_UCON, C_REG, C_REG, 25, 8, 0, 0, 0},
+	{AAND, C_UCON, C_NONE, C_REG, 25, 8, 0, 0, 0},
+
+	{AADD, C_LCON, C_NONE, C_REG, 23, 12, 0, 0, 0},
+	{AADDV, C_LCON, C_NONE, C_REG, 23, 12, 0, sys.MIPS64, 0},
+	{AAND, C_LCON, C_NONE, C_REG, 23, 12, 0, 0, 0},
+	{AADD, C_LCON, C_REG, C_REG, 23, 12, 0, 0, 0},
+	{AADDV, C_LCON, C_REG, C_REG, 23, 12, 0, sys.MIPS64, 0},
+	{AAND, C_LCON, C_REG, C_REG, 23, 12, 0, 0, 0},
+
+	{ASLL, C_SCON, C_REG, C_REG, 16, 4, 0, 0, 0},
+	{ASLL, C_SCON, C_NONE, C_REG, 16, 4, 0, 0, 0},
+
+	{ASLLV, C_SCON, C_REG, C_REG, 16, 4, 0, sys.MIPS64, 0},
+	{ASLLV, C_SCON, C_NONE, C_REG, 16, 4, 0, sys.MIPS64, 0},
+
+	{ASYSCALL, C_NONE, C_NONE, C_NONE, 5, 4, 0, 0, 0},
+
+	{ABEQ, C_REG, C_REG, C_SBRA, 6, 4, 0, 0, 0},
+	{ABEQ, C_REG, C_NONE, C_SBRA, 6, 4, 0, 0, 0},
+	{ABLEZ, C_REG, C_NONE, C_SBRA, 6, 4, 0, 0, 0},
+	{ABFPT, C_NONE, C_NONE, C_SBRA, 6, 8, 0, 0, NOTUSETMP},
+
+	{AJMP, C_NONE, C_NONE, C_LBRA, 11, 4, 0, 0, 0},
+	{AJAL, C_NONE, C_NONE, C_LBRA, 11, 4, 0, 0, 0},
+
+	{AJMP, C_NONE, C_NONE, C_ZOREG, 18, 4, REGZERO, 0, 0},
+	{AJAL, C_NONE, C_NONE, C_ZOREG, 18, 4, REGLINK, 0, 0},
+
+	{AMOVW, C_SEXT, C_NONE, C_FREG, 27, 4, REGSB, sys.MIPS64, 0},
+	{AMOVF, C_SEXT, C_NONE, C_FREG, 27, 4, REGSB, sys.MIPS64, 0},
+	{AMOVD, C_SEXT, C_NONE, C_FREG, 27, 4, REGSB, sys.MIPS64, 0},
+	{AMOVW, C_SAUTO, C_NONE, C_FREG, 27, 4, REGSP, sys.MIPS64, 0},
+	{AMOVF, C_SAUTO, C_NONE, C_FREG, 27, 4, REGSP, 0, 0},
+	{AMOVD, C_SAUTO, C_NONE, C_FREG, 27, 4, REGSP, 0, 0},
+	{AMOVW, C_SOREG, C_NONE, C_FREG, 27, 4, REGZERO, sys.MIPS64, 0},
+	{AMOVF, C_SOREG, C_NONE, C_FREG, 27, 4, REGZERO, 0, 0},
+	{AMOVD, C_SOREG, C_NONE, C_FREG, 27, 4, REGZERO, 0, 0},
+
+	{AMOVW, C_LEXT, C_NONE, C_FREG, 27, 12, REGSB, sys.MIPS64, 0},
+	{AMOVF, C_LEXT, C_NONE, C_FREG, 27, 12, REGSB, sys.MIPS64, 0},
+	{AMOVD, C_LEXT, C_NONE, C_FREG, 27, 12, REGSB, sys.MIPS64, 0},
+	{AMOVW, C_LAUTO, C_NONE, C_FREG, 27, 12, REGSP, sys.MIPS64, 0},
+	{AMOVF, C_LAUTO, C_NONE, C_FREG, 27, 12, REGSP, 0, 0},
+	{AMOVD, C_LAUTO, C_NONE, C_FREG, 27, 12, REGSP, 0, 0},
+	{AMOVW, C_LOREG, C_NONE, C_FREG, 27, 12, REGZERO, sys.MIPS64, 0},
+	{AMOVF, C_LOREG, C_NONE, C_FREG, 27, 12, REGZERO, 0, 0},
+	{AMOVD, C_LOREG, C_NONE, C_FREG, 27, 12, REGZERO, 0, 0},
+	{AMOVF, C_ADDR, C_NONE, C_FREG, 51, 8, 0, sys.MIPS, 0},
+	{AMOVF, C_ADDR, C_NONE, C_FREG, 51, 12, 0, sys.MIPS64, 0},
+	{AMOVD, C_ADDR, C_NONE, C_FREG, 51, 8, 0, sys.MIPS, 0},
+	{AMOVD, C_ADDR, C_NONE, C_FREG, 51, 12, 0, sys.MIPS64, 0},
+
+	{AMOVW, C_FREG, C_NONE, C_SEXT, 28, 4, REGSB, sys.MIPS64, 0},
+	{AMOVF, C_FREG, C_NONE, C_SEXT, 28, 4, REGSB, sys.MIPS64, 0},
+	{AMOVD, C_FREG, C_NONE, C_SEXT, 28, 4, REGSB, sys.MIPS64, 0},
+	{AMOVW, C_FREG, C_NONE, C_SAUTO, 28, 4, REGSP, sys.MIPS64, 0},
+	{AMOVF, C_FREG, C_NONE, C_SAUTO, 28, 4, REGSP, 0, 0},
+	{AMOVD, C_FREG, C_NONE, C_SAUTO, 28, 4, REGSP, 0, 0},
+	{AMOVW, C_FREG, C_NONE, C_SOREG, 28, 4, REGZERO, sys.MIPS64, 0},
+	{AMOVF, C_FREG, C_NONE, C_SOREG, 28, 4, REGZERO, 0, 0},
+	{AMOVD, C_FREG, C_NONE, C_SOREG, 28, 4, REGZERO, 0, 0},
+
+	{AMOVW, C_FREG, C_NONE, C_LEXT, 28, 12, REGSB, sys.MIPS64, 0},
+	{AMOVF, C_FREG, C_NONE, C_LEXT, 28, 12, REGSB, sys.MIPS64, 0},
+	{AMOVD, C_FREG, C_NONE, C_LEXT, 28, 12, REGSB, sys.MIPS64, 0},
+	{AMOVW, C_FREG, C_NONE, C_LAUTO, 28, 12, REGSP, sys.MIPS64, 0},
+	{AMOVF, C_FREG, C_NONE, C_LAUTO, 28, 12, REGSP, 0, 0},
+	{AMOVD, C_FREG, C_NONE, C_LAUTO, 28, 12, REGSP, 0, 0},
+	{AMOVW, C_FREG, C_NONE, C_LOREG, 28, 12, REGZERO, sys.MIPS64, 0},
+	{AMOVF, C_FREG, C_NONE, C_LOREG, 28, 12, REGZERO, 0, 0},
+	{AMOVD, C_FREG, C_NONE, C_LOREG, 28, 12, REGZERO, 0, 0},
+	{AMOVF, C_FREG, C_NONE, C_ADDR, 50, 8, 0, sys.MIPS, 0},
+	{AMOVF, C_FREG, C_NONE, C_ADDR, 50, 12, 0, sys.MIPS64, 0},
+	{AMOVD, C_FREG, C_NONE, C_ADDR, 50, 8, 0, sys.MIPS, 0},
+	{AMOVD, C_FREG, C_NONE, C_ADDR, 50, 12, 0, sys.MIPS64, 0},
+
+	{AMOVW, C_REG, C_NONE, C_FREG, 30, 4, 0, 0, 0},
+	{AMOVW, C_FREG, C_NONE, C_REG, 31, 4, 0, 0, 0},
+	{AMOVV, C_REG, C_NONE, C_FREG, 47, 4, 0, sys.MIPS64, 0},
+	{AMOVV, C_FREG, C_NONE, C_REG, 48, 4, 0, sys.MIPS64, 0},
+
+	{AMOVW, C_ADDCON, C_NONE, C_FREG, 34, 8, 0, sys.MIPS64, 0},
+	{AMOVW, C_ANDCON, C_NONE, C_FREG, 34, 8, 0, sys.MIPS64, 0},
+
+	{AMOVW, C_REG, C_NONE, C_MREG, 37, 4, 0, 0, 0},
+	{AMOVV, C_REG, C_NONE, C_MREG, 37, 4, 0, sys.MIPS64, 0},
+	{AMOVW, C_MREG, C_NONE, C_REG, 38, 4, 0, 0, 0},
+	{AMOVV, C_MREG, C_NONE, C_REG, 38, 4, 0, sys.MIPS64, 0},
+
+	{AWORD, C_LCON, C_NONE, C_NONE, 40, 4, 0, 0, 0},
+
+	{AMOVW, C_REG, C_NONE, C_FCREG, 41, 4, 0, 0, 0},
+	{AMOVV, C_REG, C_NONE, C_FCREG, 41, 4, 0, sys.MIPS64, 0},
+	{AMOVW, C_FCREG, C_NONE, C_REG, 42, 4, 0, 0, 0},
+	{AMOVV, C_FCREG, C_NONE, C_REG, 42, 4, 0, sys.MIPS64, 0},
+
+	{ATEQ, C_SCON, C_REG, C_REG, 15, 4, 0, 0, 0},
+	{ATEQ, C_SCON, C_NONE, C_REG, 15, 4, 0, 0, 0},
+	{ACMOVT, C_REG, C_NONE, C_REG, 17, 4, 0, 0, 0},
+
+	{AVMOVB, C_SCON, C_NONE, C_WREG, 56, 4, 0, sys.MIPS64, 0},
+	{AVMOVB, C_ADDCON, C_NONE, C_WREG, 56, 4, 0, sys.MIPS64, 0},
+	{AVMOVB, C_SOREG, C_NONE, C_WREG, 57, 4, 0, sys.MIPS64, 0},
+	{AVMOVB, C_WREG, C_NONE, C_SOREG, 58, 4, 0, sys.MIPS64, 0},
+
+	{ABREAK, C_REG, C_NONE, C_SEXT, 7, 4, REGSB, sys.MIPS64, 0}, /* really CACHE instruction */
+	{ABREAK, C_REG, C_NONE, C_SAUTO, 7, 4, REGSP, sys.MIPS64, 0},
+	{ABREAK, C_REG, C_NONE, C_SOREG, 7, 4, REGZERO, sys.MIPS64, 0},
+	{ABREAK, C_NONE, C_NONE, C_NONE, 5, 4, 0, 0, 0},
+
+	{obj.AUNDEF, C_NONE, C_NONE, C_NONE, 49, 4, 0, 0, 0},
+	{obj.APCDATA, C_LCON, C_NONE, C_LCON, 0, 0, 0, 0, 0},
+	{obj.AFUNCDATA, C_SCON, C_NONE, C_ADDR, 0, 0, 0, 0, 0},
+	{obj.ANOP, C_NONE, C_NONE, C_NONE, 0, 0, 0, 0, 0},
+	{obj.ANOP, C_LCON, C_NONE, C_NONE, 0, 0, 0, 0, 0}, // nop variants, see #40689
+	{obj.ANOP, C_REG, C_NONE, C_NONE, 0, 0, 0, 0, 0},
+	{obj.ANOP, C_FREG, C_NONE, C_NONE, 0, 0, 0, 0, 0},
+	{obj.ADUFFZERO, C_NONE, C_NONE, C_LBRA, 11, 4, 0, 0, 0}, // same as AJMP
+	{obj.ADUFFCOPY, C_NONE, C_NONE, C_LBRA, 11, 4, 0, 0, 0}, // same as AJMP
+
+	{obj.AXXX, C_NONE, C_NONE, C_NONE, 0, 4, 0, 0, 0},
+}
+
+var oprange [ALAST & obj.AMask][]Optab
+
+var xcmp [C_NCLASS][C_NCLASS]bool
+
+func span0(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
+	if ctxt.Retpoline {
+		ctxt.Diag("-spectre=ret not supported on mips")
+		ctxt.Retpoline = false // don't keep printing
+	}
+
+	p := cursym.Func().Text
+	if p == nil || p.Link == nil { // handle external functions and ELF section symbols
+		return
+	}
+
+	c := ctxt0{ctxt: ctxt, newprog: newprog, cursym: cursym, autosize: int32(p.To.Offset + ctxt.FixedFrameSize())}
+
+	if oprange[AOR&obj.AMask] == nil {
+		c.ctxt.Diag("mips ops not initialized, call mips.buildop first")
+	}
+
+	pc := int64(0)
+	p.Pc = pc
+
+	var m int
+	var o *Optab
+	for p = p.Link; p != nil; p = p.Link {
+		p.Pc = pc
+		o = c.oplook(p)
+		m = int(o.size)
+		if m == 0 {
+			if p.As != obj.ANOP && p.As != obj.AFUNCDATA && p.As != obj.APCDATA {
+				c.ctxt.Diag("zero-width instruction\n%v", p)
+			}
+			continue
+		}
+
+		pc += int64(m)
+	}
+
+	c.cursym.Size = pc
+
+	/*
+	 * if any procedure is large enough to
+	 * generate a large SBRA branch, then
+	 * generate extra passes putting branches
+	 * around jmps to fix. this is rare.
+	 */
+	bflag := 1
+
+	var otxt int64
+	var q *obj.Prog
+	for bflag != 0 {
+		bflag = 0
+		pc = 0
+		for p = c.cursym.Func().Text.Link; p != nil; p = p.Link {
+			p.Pc = pc
+			o = c.oplook(p)
+
+			// very large conditional branches
+			if o.type_ == 6 && p.To.Target() != nil {
+				otxt = p.To.Target().Pc - pc
+				if otxt < -(1<<17)+10 || otxt >= (1<<17)-10 {
+					q = c.newprog()
+					q.Link = p.Link
+					p.Link = q
+					q.As = AJMP
+					q.Pos = p.Pos
+					q.To.Type = obj.TYPE_BRANCH
+					q.To.SetTarget(p.To.Target())
+					p.To.SetTarget(q)
+					q = c.newprog()
+					q.Link = p.Link
+					p.Link = q
+					q.As = AJMP
+					q.Pos = p.Pos
+					q.To.Type = obj.TYPE_BRANCH
+					q.To.SetTarget(q.Link.Link)
+
+					c.addnop(p.Link)
+					c.addnop(p)
+					bflag = 1
+				}
+			}
+
+			m = int(o.size)
+			if m == 0 {
+				if p.As != obj.ANOP && p.As != obj.AFUNCDATA && p.As != obj.APCDATA {
+					c.ctxt.Diag("zero-width instruction\n%v", p)
+				}
+				continue
+			}
+
+			pc += int64(m)
+		}
+
+		c.cursym.Size = pc
+	}
+	if c.ctxt.Arch.Family == sys.MIPS64 {
+		pc += -pc & (mips64FuncAlign - 1)
+	}
+	c.cursym.Size = pc
+
+	/*
+	 * lay out the code, emitting code and data relocations.
+	 */
+
+	c.cursym.Grow(c.cursym.Size)
+
+	bp := c.cursym.P
+	var i int32
+	var out [4]uint32
+	for p := c.cursym.Func().Text.Link; p != nil; p = p.Link {
+		c.pc = p.Pc
+		o = c.oplook(p)
+		if int(o.size) > 4*len(out) {
+			log.Fatalf("out array in span0 is too small, need at least %d for %v", o.size/4, p)
+		}
+		c.asmout(p, o, out[:])
+		for i = 0; i < int32(o.size/4); i++ {
+			c.ctxt.Arch.ByteOrder.PutUint32(bp, out[i])
+			bp = bp[4:]
+		}
+	}
+
+	// Mark nonpreemptible instruction sequences.
+	// We use REGTMP as a scratch register during call injection,
+	// so instruction sequences that use REGTMP are unsafe to
+	// preempt asynchronously.
+	obj.MarkUnsafePoints(c.ctxt, c.cursym.Func().Text, c.newprog, c.isUnsafePoint, c.isRestartable)
+}
+
+// isUnsafePoint returns whether p is an unsafe point.
+func (c *ctxt0) isUnsafePoint(p *obj.Prog) bool {
+	// If p explicitly uses REGTMP, it's unsafe to preempt, because the
+	// preemption sequence clobbers REGTMP.
+	return p.From.Reg == REGTMP || p.To.Reg == REGTMP || p.Reg == REGTMP
+}
+
+// isRestartable returns whether p is a multi-instruction sequence that,
+// if preempted, can be restarted.
+func (c *ctxt0) isRestartable(p *obj.Prog) bool {
+	if c.isUnsafePoint(p) {
+		return false
+	}
+	// If p is a multi-instruction sequence with uses REGTMP inserted by
+	// the assembler in order to materialize a large constant/offset, we
+	// can restart p (at the start of the instruction sequence), recompute
+	// the content of REGTMP, upon async preemption. Currently, all cases
+	// of assembler-inserted REGTMP fall into this category.
+	// If p doesn't use REGTMP, it can be simply preempted, so we don't
+	// mark it.
+	o := c.oplook(p)
+	return o.size > 4 && o.flag&NOTUSETMP == 0
+}
+
+func isint32(v int64) bool {
+	return int64(int32(v)) == v
+}
+
+func isuint32(v uint64) bool {
+	return uint64(uint32(v)) == v
+}
+
+func (c *ctxt0) aclass(a *obj.Addr) int {
+	switch a.Type {
+	case obj.TYPE_NONE:
+		return C_NONE
+
+	case obj.TYPE_REG:
+		if REG_R0 <= a.Reg && a.Reg <= REG_R31 {
+			return C_REG
+		}
+		if REG_F0 <= a.Reg && a.Reg <= REG_F31 {
+			return C_FREG
+		}
+		if REG_M0 <= a.Reg && a.Reg <= REG_M31 {
+			return C_MREG
+		}
+		if REG_FCR0 <= a.Reg && a.Reg <= REG_FCR31 {
+			return C_FCREG
+		}
+		if REG_W0 <= a.Reg && a.Reg <= REG_W31 {
+			return C_WREG
+		}
+		if a.Reg == REG_LO {
+			return C_LO
+		}
+		if a.Reg == REG_HI {
+			return C_HI
+		}
+		return C_GOK
+
+	case obj.TYPE_MEM:
+		switch a.Name {
+		case obj.NAME_EXTERN,
+			obj.NAME_STATIC:
+			if a.Sym == nil {
+				break
+			}
+			c.instoffset = a.Offset
+			if a.Sym != nil { // use relocation
+				if a.Sym.Type == objabi.STLSBSS {
+					return C_TLS
+				}
+				return C_ADDR
+			}
+			return C_LEXT
+
+		case obj.NAME_AUTO:
+			if a.Reg == REGSP {
+				// unset base register for better printing, since
+				// a.Offset is still relative to pseudo-SP.
+				a.Reg = obj.REG_NONE
+			}
+			c.instoffset = int64(c.autosize) + a.Offset
+			if c.instoffset >= -BIG && c.instoffset < BIG {
+				return C_SAUTO
+			}
+			return C_LAUTO
+
+		case obj.NAME_PARAM:
+			if a.Reg == REGSP {
+				// unset base register for better printing, since
+				// a.Offset is still relative to pseudo-FP.
+				a.Reg = obj.REG_NONE
+			}
+			c.instoffset = int64(c.autosize) + a.Offset + c.ctxt.FixedFrameSize()
+			if c.instoffset >= -BIG && c.instoffset < BIG {
+				return C_SAUTO
+			}
+			return C_LAUTO
+
+		case obj.NAME_NONE:
+			c.instoffset = a.Offset
+			if c.instoffset == 0 {
+				return C_ZOREG
+			}
+			if c.instoffset >= -BIG && c.instoffset < BIG {
+				return C_SOREG
+			}
+			return C_LOREG
+		}
+
+		return C_GOK
+
+	case obj.TYPE_TEXTSIZE:
+		return C_TEXTSIZE
+
+	case obj.TYPE_CONST,
+		obj.TYPE_ADDR:
+		switch a.Name {
+		case obj.NAME_NONE:
+			c.instoffset = a.Offset
+			if a.Reg != 0 {
+				if -BIG <= c.instoffset && c.instoffset <= BIG {
+					return C_SACON
+				}
+				if isint32(c.instoffset) {
+					return C_LACON
+				}
+				return C_DACON
+			}
+
+		case obj.NAME_EXTERN,
+			obj.NAME_STATIC:
+			s := a.Sym
+			if s == nil {
+				return C_GOK
+			}
+
+			c.instoffset = a.Offset
+			if s.Type == objabi.STLSBSS {
+				return C_STCON // address of TLS variable
+			}
+			return C_LECON
+
+		case obj.NAME_AUTO:
+			if a.Reg == REGSP {
+				// unset base register for better printing, since
+				// a.Offset is still relative to pseudo-SP.
+				a.Reg = obj.REG_NONE
+			}
+			c.instoffset = int64(c.autosize) + a.Offset
+			if c.instoffset >= -BIG && c.instoffset < BIG {
+				return C_SACON
+			}
+			return C_LACON
+
+		case obj.NAME_PARAM:
+			if a.Reg == REGSP {
+				// unset base register for better printing, since
+				// a.Offset is still relative to pseudo-FP.
+				a.Reg = obj.REG_NONE
+			}
+			c.instoffset = int64(c.autosize) + a.Offset + c.ctxt.FixedFrameSize()
+			if c.instoffset >= -BIG && c.instoffset < BIG {
+				return C_SACON
+			}
+			return C_LACON
+
+		default:
+			return C_GOK
+		}
+
+		if c.instoffset >= 0 {
+			if c.instoffset == 0 {
+				return C_ZCON
+			}
+			if c.instoffset <= 0x7fff {
+				return C_SCON
+			}
+			if c.instoffset <= 0xffff {
+				return C_ANDCON
+			}
+			if c.instoffset&0xffff == 0 && isuint32(uint64(c.instoffset)) { /* && (instoffset & (1<<31)) == 0) */
+				return C_UCON
+			}
+			if isint32(c.instoffset) || isuint32(uint64(c.instoffset)) {
+				return C_LCON
+			}
+			return C_LCON // C_DCON
+		}
+
+		if c.instoffset >= -0x8000 {
+			return C_ADDCON
+		}
+		if c.instoffset&0xffff == 0 && isint32(c.instoffset) {
+			return C_UCON
+		}
+		if isint32(c.instoffset) {
+			return C_LCON
+		}
+		return C_LCON // C_DCON
+
+	case obj.TYPE_BRANCH:
+		return C_SBRA
+	}
+
+	return C_GOK
+}
+
+func prasm(p *obj.Prog) {
+	fmt.Printf("%v\n", p)
+}
+
+func (c *ctxt0) oplook(p *obj.Prog) *Optab {
+	if oprange[AOR&obj.AMask] == nil {
+		c.ctxt.Diag("mips ops not initialized, call mips.buildop first")
+	}
+
+	a1 := int(p.Optab)
+	if a1 != 0 {
+		return &optab[a1-1]
+	}
+	a1 = int(p.From.Class)
+	if a1 == 0 {
+		a1 = c.aclass(&p.From) + 1
+		p.From.Class = int8(a1)
+	}
+
+	a1--
+	a3 := int(p.To.Class)
+	if a3 == 0 {
+		a3 = c.aclass(&p.To) + 1
+		p.To.Class = int8(a3)
+	}
+
+	a3--
+	a2 := C_NONE
+	if p.Reg != 0 {
+		a2 = C_REG
+	}
+
+	ops := oprange[p.As&obj.AMask]
+	c1 := &xcmp[a1]
+	c3 := &xcmp[a3]
+	for i := range ops {
+		op := &ops[i]
+		if int(op.a2) == a2 && c1[op.a1] && c3[op.a3] && (op.family == 0 || c.ctxt.Arch.Family == op.family) {
+			p.Optab = uint16(cap(optab) - cap(ops) + i + 1)
+			return op
+		}
+	}
+
+	c.ctxt.Diag("illegal combination %v %v %v %v", p.As, DRconv(a1), DRconv(a2), DRconv(a3))
+	prasm(p)
+	// Turn illegal instruction into an UNDEF, avoid crashing in asmout.
+	return &Optab{obj.AUNDEF, C_NONE, C_NONE, C_NONE, 49, 4, 0, 0, 0}
+}
+
+func cmp(a int, b int) bool {
+	if a == b {
+		return true
+	}
+	switch a {
+	case C_LCON:
+		if b == C_ZCON || b == C_SCON || b == C_UCON || b == C_ADDCON || b == C_ANDCON {
+			return true
+		}
+
+	case C_ADD0CON:
+		if b == C_ADDCON {
+			return true
+		}
+		fallthrough
+
+	case C_ADDCON:
+		if b == C_ZCON || b == C_SCON {
+			return true
+		}
+
+	case C_AND0CON:
+		if b == C_ANDCON {
+			return true
+		}
+		fallthrough
+
+	case C_ANDCON:
+		if b == C_ZCON || b == C_SCON {
+			return true
+		}
+
+	case C_UCON:
+		if b == C_ZCON {
+			return true
+		}
+
+	case C_SCON:
+		if b == C_ZCON {
+			return true
+		}
+
+	case C_LACON:
+		if b == C_SACON {
+			return true
+		}
+
+	case C_LBRA:
+		if b == C_SBRA {
+			return true
+		}
+
+	case C_LEXT:
+		if b == C_SEXT {
+			return true
+		}
+
+	case C_LAUTO:
+		if b == C_SAUTO {
+			return true
+		}
+
+	case C_REG:
+		if b == C_ZCON {
+			return r0iszero != 0 /*TypeKind(100016)*/
+		}
+
+	case C_LOREG:
+		if b == C_ZOREG || b == C_SOREG {
+			return true
+		}
+
+	case C_SOREG:
+		if b == C_ZOREG {
+			return true
+		}
+	}
+
+	return false
+}
+
+type ocmp []Optab
+
+func (x ocmp) Len() int {
+	return len(x)
+}
+
+func (x ocmp) Swap(i, j int) {
+	x[i], x[j] = x[j], x[i]
+}
+
+func (x ocmp) Less(i, j int) bool {
+	p1 := &x[i]
+	p2 := &x[j]
+	n := int(p1.as) - int(p2.as)
+	if n != 0 {
+		return n < 0
+	}
+	n = int(p1.a1) - int(p2.a1)
+	if n != 0 {
+		return n < 0
+	}
+	n = int(p1.a2) - int(p2.a2)
+	if n != 0 {
+		return n < 0
+	}
+	n = int(p1.a3) - int(p2.a3)
+	if n != 0 {
+		return n < 0
+	}
+	return false
+}
+
+func opset(a, b0 obj.As) {
+	oprange[a&obj.AMask] = oprange[b0]
+}
+
+func buildop(ctxt *obj.Link) {
+	if oprange[AOR&obj.AMask] != nil {
+		// Already initialized; stop now.
+		// This happens in the cmd/asm tests,
+		// each of which re-initializes the arch.
+		return
+	}
+
+	var n int
+
+	for i := 0; i < C_NCLASS; i++ {
+		for n = 0; n < C_NCLASS; n++ {
+			if cmp(n, i) {
+				xcmp[i][n] = true
+			}
+		}
+	}
+	for n = 0; optab[n].as != obj.AXXX; n++ {
+	}
+	sort.Sort(ocmp(optab[:n]))
+	for i := 0; i < n; i++ {
+		r := optab[i].as
+		r0 := r & obj.AMask
+		start := i
+		for optab[i].as == r {
+			i++
+		}
+		oprange[r0] = optab[start:i]
+		i--
+
+		switch r {
+		default:
+			ctxt.Diag("unknown op in build: %v", r)
+			ctxt.DiagFlush()
+			log.Fatalf("bad code")
+
+		case AABSF:
+			opset(AMOVFD, r0)
+			opset(AMOVDF, r0)
+			opset(AMOVWF, r0)
+			opset(AMOVFW, r0)
+			opset(AMOVWD, r0)
+			opset(AMOVDW, r0)
+			opset(ANEGF, r0)
+			opset(ANEGD, r0)
+			opset(AABSD, r0)
+			opset(ATRUNCDW, r0)
+			opset(ATRUNCFW, r0)
+			opset(ASQRTF, r0)
+			opset(ASQRTD, r0)
+
+		case AMOVVF:
+			opset(AMOVVD, r0)
+			opset(AMOVFV, r0)
+			opset(AMOVDV, r0)
+			opset(ATRUNCDV, r0)
+			opset(ATRUNCFV, r0)
+
+		case AADD:
+			opset(ASGT, r0)
+			opset(ASGTU, r0)
+			opset(AADDU, r0)
+
+		case AADDV:
+			opset(AADDVU, r0)
+
+		case AADDF:
+			opset(ADIVF, r0)
+			opset(ADIVD, r0)
+			opset(AMULF, r0)
+			opset(AMULD, r0)
+			opset(ASUBF, r0)
+			opset(ASUBD, r0)
+			opset(AADDD, r0)
+
+		case AAND:
+			opset(AOR, r0)
+			opset(AXOR, r0)
+
+		case ABEQ:
+			opset(ABNE, r0)
+
+		case ABLEZ:
+			opset(ABGEZ, r0)
+			opset(ABGEZAL, r0)
+			opset(ABLTZ, r0)
+			opset(ABLTZAL, r0)
+			opset(ABGTZ, r0)
+
+		case AMOVB:
+			opset(AMOVH, r0)
+
+		case AMOVBU:
+			opset(AMOVHU, r0)
+
+		case AMUL:
+			opset(AREM, r0)
+			opset(AREMU, r0)
+			opset(ADIVU, r0)
+			opset(AMULU, r0)
+			opset(ADIV, r0)
+			opset(AMADD, r0)
+			opset(AMSUB, r0)
+
+		case AMULV:
+			opset(ADIVV, r0)
+			opset(ADIVVU, r0)
+			opset(AMULVU, r0)
+			opset(AREMV, r0)
+			opset(AREMVU, r0)
+
+		case ASLL:
+			opset(ASRL, r0)
+			opset(ASRA, r0)
+			opset(AROTR, r0)
+
+		case ASLLV:
+			opset(ASRAV, r0)
+			opset(ASRLV, r0)
+			opset(AROTRV, r0)
+
+		case ASUB:
+			opset(ASUBU, r0)
+			opset(ANOR, r0)
+
+		case ASUBV:
+			opset(ASUBVU, r0)
+
+		case ASYSCALL:
+			opset(ASYNC, r0)
+			opset(ANOOP, r0)
+			opset(ATLBP, r0)
+			opset(ATLBR, r0)
+			opset(ATLBWI, r0)
+			opset(ATLBWR, r0)
+
+		case ACMPEQF:
+			opset(ACMPGTF, r0)
+			opset(ACMPGTD, r0)
+			opset(ACMPGEF, r0)
+			opset(ACMPGED, r0)
+			opset(ACMPEQD, r0)
+
+		case ABFPT:
+			opset(ABFPF, r0)
+
+		case AMOVWL:
+			opset(AMOVWR, r0)
+
+		case AMOVVL:
+			opset(AMOVVR, r0)
+
+		case AVMOVB:
+			opset(AVMOVH, r0)
+			opset(AVMOVW, r0)
+			opset(AVMOVD, r0)
+
+		case AMOVW,
+			AMOVD,
+			AMOVF,
+			AMOVV,
+			ABREAK,
+			ARFE,
+			AJAL,
+			AJMP,
+			AMOVWU,
+			ALL,
+			ALLV,
+			ASC,
+			ASCV,
+			ANEGW,
+			ANEGV,
+			AWORD,
+			obj.ANOP,
+			obj.ATEXT,
+			obj.AUNDEF,
+			obj.AFUNCDATA,
+			obj.APCDATA,
+			obj.ADUFFZERO,
+			obj.ADUFFCOPY:
+			break
+
+		case ACMOVN:
+			opset(ACMOVZ, r0)
+
+		case ACMOVT:
+			opset(ACMOVF, r0)
+
+		case ACLO:
+			opset(ACLZ, r0)
+
+		case ATEQ:
+			opset(ATNE, r0)
+		}
+	}
+}
+
+func OP(x uint32, y uint32) uint32 {
+	return x<<3 | y<<0
+}
+
+func SP(x uint32, y uint32) uint32 {
+	return x<<29 | y<<26
+}
+
+func BCOND(x uint32, y uint32) uint32 {
+	return x<<19 | y<<16
+}
+
+func MMU(x uint32, y uint32) uint32 {
+	return SP(2, 0) | 16<<21 | x<<3 | y<<0
+}
+
+func FPF(x uint32, y uint32) uint32 {
+	return SP(2, 1) | 16<<21 | x<<3 | y<<0
+}
+
+func FPD(x uint32, y uint32) uint32 {
+	return SP(2, 1) | 17<<21 | x<<3 | y<<0
+}
+
+func FPW(x uint32, y uint32) uint32 {
+	return SP(2, 1) | 20<<21 | x<<3 | y<<0
+}
+
+func FPV(x uint32, y uint32) uint32 {
+	return SP(2, 1) | 21<<21 | x<<3 | y<<0
+}
+
+func OP_RRR(op uint32, r1 uint32, r2 uint32, r3 uint32) uint32 {
+	return op | (r1&31)<<16 | (r2&31)<<21 | (r3&31)<<11
+}
+
+func OP_IRR(op uint32, i uint32, r2 uint32, r3 uint32) uint32 {
+	return op | i&0xFFFF | (r2&31)<<21 | (r3&31)<<16
+}
+
+func OP_SRR(op uint32, s uint32, r2 uint32, r3 uint32) uint32 {
+	return op | (s&31)<<6 | (r2&31)<<16 | (r3&31)<<11
+}
+
+func OP_FRRR(op uint32, r1 uint32, r2 uint32, r3 uint32) uint32 {
+	return op | (r1&31)<<16 | (r2&31)<<11 | (r3&31)<<6
+}
+
+func OP_JMP(op uint32, i uint32) uint32 {
+	return op | i&0x3FFFFFF
+}
+
+func OP_VI10(op uint32, df uint32, s10 int32, wd uint32, minor uint32) uint32 {
+	return 0x1e<<26 | (op&7)<<23 | (df&3)<<21 | uint32(s10&0x3FF)<<11 | (wd&31)<<6 | minor&0x3F
+}
+
+func OP_VMI10(s10 int32, rs uint32, wd uint32, minor uint32, df uint32) uint32 {
+	return 0x1e<<26 | uint32(s10&0x3FF)<<16 | (rs&31)<<11 | (wd&31)<<6 | (minor&15)<<2 | df&3
+}
+
+func (c *ctxt0) asmout(p *obj.Prog, o *Optab, out []uint32) {
+	o1 := uint32(0)
+	o2 := uint32(0)
+	o3 := uint32(0)
+	o4 := uint32(0)
+
+	add := AADDU
+
+	if c.ctxt.Arch.Family == sys.MIPS64 {
+		add = AADDVU
+	}
+	switch o.type_ {
+	default:
+		c.ctxt.Diag("unknown type %d %v", o.type_)
+		prasm(p)
+
+	case 0: /* pseudo ops */
+		break
+
+	case 1: /* mov r1,r2 ==> OR r1,r0,r2 */
+		a := AOR
+		if p.As == AMOVW && c.ctxt.Arch.Family == sys.MIPS64 {
+			// on MIPS64, most of the 32-bit instructions have unpredictable behavior,
+			// but SLL is special that the result is always sign-extended to 64-bit.
+			a = ASLL
+		}
+		o1 = OP_RRR(c.oprrr(a), uint32(p.From.Reg), uint32(REGZERO), uint32(p.To.Reg))
+
+	case 2: /* add/sub r1,[r2],r3 */
+		r := int(p.Reg)
+		if p.As == ANEGW || p.As == ANEGV {
+			r = REGZERO
+		}
+		if r == 0 {
+			r = int(p.To.Reg)
+		}
+		o1 = OP_RRR(c.oprrr(p.As), uint32(p.From.Reg), uint32(r), uint32(p.To.Reg))
+
+	case 3: /* mov $soreg, r ==> or/add $i,o,r */
+		v := c.regoff(&p.From)
+
+		r := int(p.From.Reg)
+		if r == 0 {
+			r = int(o.param)
+		}
+		a := add
+		if o.a1 == C_ANDCON {
+			a = AOR
+		}
+
+		o1 = OP_IRR(c.opirr(a), uint32(v), uint32(r), uint32(p.To.Reg))
+
+	case 4: /* add $scon,[r1],r2 */
+		v := c.regoff(&p.From)
+
+		r := int(p.Reg)
+		if r == 0 {
+			r = int(p.To.Reg)
+		}
+
+		o1 = OP_IRR(c.opirr(p.As), uint32(v), uint32(r), uint32(p.To.Reg))
+
+	case 5: /* syscall */
+		o1 = c.oprrr(p.As)
+
+	case 6: /* beq r1,[r2],sbra */
+		v := int32(0)
+		if p.To.Target() == nil {
+			v = int32(-4) >> 2
+		} else {
+			v = int32(p.To.Target().Pc-p.Pc-4) >> 2
+		}
+		if (v<<16)>>16 != v {
+			c.ctxt.Diag("short branch too far\n%v", p)
+		}
+		o1 = OP_IRR(c.opirr(p.As), uint32(v), uint32(p.From.Reg), uint32(p.Reg))
+		// for ABFPT and ABFPF only: always fill delay slot with 0
+		// see comments in func preprocess for details.
+		o2 = 0
+
+	case 7: /* mov r, soreg ==> sw o(r) */
+		r := int(p.To.Reg)
+		if r == 0 {
+			r = int(o.param)
+		}
+		v := c.regoff(&p.To)
+		o1 = OP_IRR(c.opirr(p.As), uint32(v), uint32(r), uint32(p.From.Reg))
+
+	case 8: /* mov soreg, r ==> lw o(r) */
+		r := int(p.From.Reg)
+		if r == 0 {
+			r = int(o.param)
+		}
+		v := c.regoff(&p.From)
+		o1 = OP_IRR(c.opirr(-p.As), uint32(v), uint32(r), uint32(p.To.Reg))
+
+	case 9: /* sll r1,[r2],r3 */
+		r := int(p.Reg)
+
+		if r == 0 {
+			r = int(p.To.Reg)
+		}
+		o1 = OP_RRR(c.oprrr(p.As), uint32(r), uint32(p.From.Reg), uint32(p.To.Reg))
+
+	case 10: /* add $con,[r1],r2 ==> mov $con, t; add t,[r1],r2 */
+		v := c.regoff(&p.From)
+		a := AOR
+		if v < 0 {
+			a = AADDU
+		}
+		o1 = OP_IRR(c.opirr(a), uint32(v), uint32(0), uint32(REGTMP))
+		r := int(p.Reg)
+		if r == 0 {
+			r = int(p.To.Reg)
+		}
+		o2 = OP_RRR(c.oprrr(p.As), uint32(REGTMP), uint32(r), uint32(p.To.Reg))
+
+	case 11: /* jmp lbra */
+		v := int32(0)
+		if c.aclass(&p.To) == C_SBRA && p.To.Sym == nil && p.As == AJMP {
+			// use PC-relative branch for short branches
+			// BEQ	R0, R0, sbra
+			if p.To.Target() == nil {
+				v = int32(-4) >> 2
+			} else {
+				v = int32(p.To.Target().Pc-p.Pc-4) >> 2
+			}
+			if (v<<16)>>16 == v {
+				o1 = OP_IRR(c.opirr(ABEQ), uint32(v), uint32(REGZERO), uint32(REGZERO))
+				break
+			}
+		}
+		if p.To.Target() == nil {
+			v = int32(p.Pc) >> 2
+		} else {
+			v = int32(p.To.Target().Pc) >> 2
+		}
+		o1 = OP_JMP(c.opirr(p.As), uint32(v))
+		if p.To.Sym == nil {
+			p.To.Sym = c.cursym.Func().Text.From.Sym
+			p.To.Offset = p.To.Target().Pc
+		}
+		rel := obj.Addrel(c.cursym)
+		rel.Off = int32(c.pc)
+		rel.Siz = 4
+		rel.Sym = p.To.Sym
+		rel.Add = p.To.Offset
+		if p.As == AJAL {
+			rel.Type = objabi.R_CALLMIPS
+		} else {
+			rel.Type = objabi.R_JMPMIPS
+		}
+
+	case 12: /* movbs r,r */
+		// NOTE: this case does not use REGTMP. If it ever does,
+		// remove the NOTUSETMP flag in optab.
+		v := 16
+		if p.As == AMOVB {
+			v = 24
+		}
+		o1 = OP_SRR(c.opirr(ASLL), uint32(v), uint32(p.From.Reg), uint32(p.To.Reg))
+		o2 = OP_SRR(c.opirr(ASRA), uint32(v), uint32(p.To.Reg), uint32(p.To.Reg))
+
+	case 13: /* movbu r,r */
+		if p.As == AMOVBU {
+			o1 = OP_IRR(c.opirr(AAND), uint32(0xff), uint32(p.From.Reg), uint32(p.To.Reg))
+		} else {
+			o1 = OP_IRR(c.opirr(AAND), uint32(0xffff), uint32(p.From.Reg), uint32(p.To.Reg))
+		}
+
+	case 14: /* movwu r,r */
+		// NOTE: this case does not use REGTMP. If it ever does,
+		// remove the NOTUSETMP flag in optab.
+		o1 = OP_SRR(c.opirr(-ASLLV), uint32(0), uint32(p.From.Reg), uint32(p.To.Reg))
+		o2 = OP_SRR(c.opirr(-ASRLV), uint32(0), uint32(p.To.Reg), uint32(p.To.Reg))
+
+	case 15: /* teq $c r,r */
+		v := c.regoff(&p.From)
+		r := int(p.Reg)
+		if r == 0 {
+			r = REGZERO
+		}
+		/* only use 10 bits of trap code */
+		o1 = OP_IRR(c.opirr(p.As), (uint32(v)&0x3FF)<<6, uint32(r), uint32(p.To.Reg))
+
+	case 16: /* sll $c,[r1],r2 */
+		v := c.regoff(&p.From)
+		r := int(p.Reg)
+		if r == 0 {
+			r = int(p.To.Reg)
+		}
+
+		/* OP_SRR will use only the low 5 bits of the shift value */
+		if v >= 32 && vshift(p.As) {
+			o1 = OP_SRR(c.opirr(-p.As), uint32(v-32), uint32(r), uint32(p.To.Reg))
+		} else {
+			o1 = OP_SRR(c.opirr(p.As), uint32(v), uint32(r), uint32(p.To.Reg))
+		}
+
+	case 17:
+		o1 = OP_RRR(c.oprrr(p.As), uint32(REGZERO), uint32(p.From.Reg), uint32(p.To.Reg))
+
+	case 18: /* jmp [r1],0(r2) */
+		r := int(p.Reg)
+		if r == 0 {
+			r = int(o.param)
+		}
+		o1 = OP_RRR(c.oprrr(p.As), uint32(0), uint32(p.To.Reg), uint32(r))
+		if p.As == obj.ACALL {
+			rel := obj.Addrel(c.cursym)
+			rel.Off = int32(c.pc)
+			rel.Siz = 0
+			rel.Type = objabi.R_CALLIND
+		}
+
+	case 19: /* mov $lcon,r ==> lu+or */
+		// NOTE: this case does not use REGTMP. If it ever does,
+		// remove the NOTUSETMP flag in optab.
+		v := c.regoff(&p.From)
+		o1 = OP_IRR(c.opirr(ALUI), uint32(v>>16), uint32(REGZERO), uint32(p.To.Reg))
+		o2 = OP_IRR(c.opirr(AOR), uint32(v), uint32(p.To.Reg), uint32(p.To.Reg))
+
+	case 20: /* mov lo/hi,r */
+		a := OP(2, 0) /* mfhi */
+		if p.From.Reg == REG_LO {
+			a = OP(2, 2) /* mflo */
+		}
+		o1 = OP_RRR(a, uint32(REGZERO), uint32(REGZERO), uint32(p.To.Reg))
+
+	case 21: /* mov r,lo/hi */
+		a := OP(2, 1) /* mthi */
+		if p.To.Reg == REG_LO {
+			a = OP(2, 3) /* mtlo */
+		}
+		o1 = OP_RRR(a, uint32(REGZERO), uint32(p.From.Reg), uint32(REGZERO))
+
+	case 22: /* mul r1,r2 [r3]*/
+		if p.To.Reg != 0 {
+			r := int(p.Reg)
+			if r == 0 {
+				r = int(p.To.Reg)
+			}
+			a := SP(3, 4) | 2 /* mul */
+			o1 = OP_RRR(a, uint32(p.From.Reg), uint32(r), uint32(p.To.Reg))
+		} else {
+			o1 = OP_RRR(c.oprrr(p.As), uint32(p.From.Reg), uint32(p.Reg), uint32(REGZERO))
+		}
+
+	case 23: /* add $lcon,r1,r2 ==> lu+or+add */
+		v := c.regoff(&p.From)
+		o1 = OP_IRR(c.opirr(ALUI), uint32(v>>16), uint32(REGZERO), uint32(REGTMP))
+		o2 = OP_IRR(c.opirr(AOR), uint32(v), uint32(REGTMP), uint32(REGTMP))
+		r := int(p.Reg)
+		if r == 0 {
+			r = int(p.To.Reg)
+		}
+		o3 = OP_RRR(c.oprrr(p.As), uint32(REGTMP), uint32(r), uint32(p.To.Reg))
+
+	case 24: /* mov $ucon,r ==> lu r */
+		v := c.regoff(&p.From)
+		o1 = OP_IRR(c.opirr(ALUI), uint32(v>>16), uint32(REGZERO), uint32(p.To.Reg))
+
+	case 25: /* add/and $ucon,[r1],r2 ==> lu $con,t; add t,[r1],r2 */
+		v := c.regoff(&p.From)
+		o1 = OP_IRR(c.opirr(ALUI), uint32(v>>16), uint32(REGZERO), uint32(REGTMP))
+		r := int(p.Reg)
+		if r == 0 {
+			r = int(p.To.Reg)
+		}
+		o2 = OP_RRR(c.oprrr(p.As), uint32(REGTMP), uint32(r), uint32(p.To.Reg))
+
+	case 26: /* mov $lsext/auto/oreg,r ==> lu+or+add */
+		v := c.regoff(&p.From)
+		o1 = OP_IRR(c.opirr(ALUI), uint32(v>>16), uint32(REGZERO), uint32(REGTMP))
+		o2 = OP_IRR(c.opirr(AOR), uint32(v), uint32(REGTMP), uint32(REGTMP))
+		r := int(p.From.Reg)
+		if r == 0 {
+			r = int(o.param)
+		}
+		o3 = OP_RRR(c.oprrr(add), uint32(REGTMP), uint32(r), uint32(p.To.Reg))
+
+	case 27: /* mov [sl]ext/auto/oreg,fr ==> lwc1 o(r) */
+		v := c.regoff(&p.From)
+		r := int(p.From.Reg)
+		if r == 0 {
+			r = int(o.param)
+		}
+		a := -AMOVF
+		if p.As == AMOVD {
+			a = -AMOVD
+		}
+		switch o.size {
+		case 12:
+			o1 = OP_IRR(c.opirr(ALUI), uint32((v+1<<15)>>16), uint32(REGZERO), uint32(REGTMP))
+			o2 = OP_RRR(c.oprrr(add), uint32(r), uint32(REGTMP), uint32(REGTMP))
+			o3 = OP_IRR(c.opirr(a), uint32(v), uint32(REGTMP), uint32(p.To.Reg))
+
+		case 4:
+			o1 = OP_IRR(c.opirr(a), uint32(v), uint32(r), uint32(p.To.Reg))
+		}
+
+	case 28: /* mov fr,[sl]ext/auto/oreg ==> swc1 o(r) */
+		v := c.regoff(&p.To)
+		r := int(p.To.Reg)
+		if r == 0 {
+			r = int(o.param)
+		}
+		a := AMOVF
+		if p.As == AMOVD {
+			a = AMOVD
+		}
+		switch o.size {
+		case 12:
+			o1 = OP_IRR(c.opirr(ALUI), uint32((v+1<<15)>>16), uint32(REGZERO), uint32(REGTMP))
+			o2 = OP_RRR(c.oprrr(add), uint32(r), uint32(REGTMP), uint32(REGTMP))
+			o3 = OP_IRR(c.opirr(a), uint32(v), uint32(REGTMP), uint32(p.From.Reg))
+
+		case 4:
+			o1 = OP_IRR(c.opirr(a), uint32(v), uint32(r), uint32(p.From.Reg))
+		}
+
+	case 30: /* movw r,fr */
+		a := SP(2, 1) | (4 << 21) /* mtc1 */
+		o1 = OP_RRR(a, uint32(p.From.Reg), uint32(0), uint32(p.To.Reg))
+
+	case 31: /* movw fr,r */
+		a := SP(2, 1) | (0 << 21) /* mtc1 */
+		o1 = OP_RRR(a, uint32(p.To.Reg), uint32(0), uint32(p.From.Reg))
+
+	case 32: /* fadd fr1,[fr2],fr3 */
+		r := int(p.Reg)
+		if r == 0 {
+			r = int(p.To.Reg)
+		}
+		o1 = OP_FRRR(c.oprrr(p.As), uint32(p.From.Reg), uint32(r), uint32(p.To.Reg))
+
+	case 33: /* fabs fr1, fr3 */
+		o1 = OP_FRRR(c.oprrr(p.As), uint32(0), uint32(p.From.Reg), uint32(p.To.Reg))
+
+	case 34: /* mov $con,fr ==> or/add $i,t; mov t,fr */
+		v := c.regoff(&p.From)
+		a := AADDU
+		if o.a1 == C_ANDCON {
+			a = AOR
+		}
+		o1 = OP_IRR(c.opirr(a), uint32(v), uint32(0), uint32(REGTMP))
+		o2 = OP_RRR(SP(2, 1)|(4<<21), uint32(REGTMP), uint32(0), uint32(p.To.Reg)) /* mtc1 */
+
+	case 35: /* mov r,lext/auto/oreg ==> sw o(REGTMP) */
+		v := c.regoff(&p.To)
+		r := int(p.To.Reg)
+		if r == 0 {
+			r = int(o.param)
+		}
+		o1 = OP_IRR(c.opirr(ALUI), uint32((v+1<<15)>>16), uint32(REGZERO), uint32(REGTMP))
+		o2 = OP_RRR(c.oprrr(add), uint32(r), uint32(REGTMP), uint32(REGTMP))
+		o3 = OP_IRR(c.opirr(p.As), uint32(v), uint32(REGTMP), uint32(p.From.Reg))
+
+	case 36: /* mov lext/auto/oreg,r ==> lw o(REGTMP) */
+		v := c.regoff(&p.From)
+		r := int(p.From.Reg)
+		if r == 0 {
+			r = int(o.param)
+		}
+		o1 = OP_IRR(c.opirr(ALUI), uint32((v+1<<15)>>16), uint32(REGZERO), uint32(REGTMP))
+		o2 = OP_RRR(c.oprrr(add), uint32(r), uint32(REGTMP), uint32(REGTMP))
+		o3 = OP_IRR(c.opirr(-p.As), uint32(v), uint32(REGTMP), uint32(p.To.Reg))
+
+	case 37: /* movw r,mr */
+		a := SP(2, 0) | (4 << 21) /* mtc0 */
+		if p.As == AMOVV {
+			a = SP(2, 0) | (5 << 21) /* dmtc0 */
+		}
+		o1 = OP_RRR(a, uint32(p.From.Reg), uint32(0), uint32(p.To.Reg))
+
+	case 38: /* movw mr,r */
+		a := SP(2, 0) | (0 << 21) /* mfc0 */
+		if p.As == AMOVV {
+			a = SP(2, 0) | (1 << 21) /* dmfc0 */
+		}
+		o1 = OP_RRR(a, uint32(p.To.Reg), uint32(0), uint32(p.From.Reg))
+
+	case 40: /* word */
+		o1 = uint32(c.regoff(&p.From))
+
+	case 41: /* movw f,fcr */
+		o1 = OP_RRR(SP(2, 1)|(6<<21), uint32(p.From.Reg), uint32(0), uint32(p.To.Reg)) /* mtcc1 */
+
+	case 42: /* movw fcr,r */
+		o1 = OP_RRR(SP(2, 1)|(2<<21), uint32(p.To.Reg), uint32(0), uint32(p.From.Reg)) /* mfcc1 */
+
+	case 47: /* movv r,fr */
+		a := SP(2, 1) | (5 << 21) /* dmtc1 */
+		o1 = OP_RRR(a, uint32(p.From.Reg), uint32(0), uint32(p.To.Reg))
+
+	case 48: /* movv fr,r */
+		a := SP(2, 1) | (1 << 21) /* dmtc1 */
+		o1 = OP_RRR(a, uint32(p.To.Reg), uint32(0), uint32(p.From.Reg))
+
+	case 49: /* undef */
+		o1 = 52 /* trap -- teq r0, r0 */
+
+	/* relocation operations */
+	case 50: /* mov r,addr ==> lu + add REGSB, REGTMP + sw o(REGTMP) */
+		o1 = OP_IRR(c.opirr(ALUI), uint32(0), uint32(REGZERO), uint32(REGTMP))
+		rel := obj.Addrel(c.cursym)
+		rel.Off = int32(c.pc)
+		rel.Siz = 4
+		rel.Sym = p.To.Sym
+		rel.Add = p.To.Offset
+		rel.Type = objabi.R_ADDRMIPSU
+		o2 = OP_IRR(c.opirr(p.As), uint32(0), uint32(REGTMP), uint32(p.From.Reg))
+		rel2 := obj.Addrel(c.cursym)
+		rel2.Off = int32(c.pc + 4)
+		rel2.Siz = 4
+		rel2.Sym = p.To.Sym
+		rel2.Add = p.To.Offset
+		rel2.Type = objabi.R_ADDRMIPS
+
+		if o.size == 12 {
+			o3 = o2
+			o2 = OP_RRR(c.oprrr(AADDVU), uint32(REGSB), uint32(REGTMP), uint32(REGTMP))
+			rel2.Off += 4
+		}
+
+	case 51: /* mov addr,r ==> lu + add REGSB, REGTMP + lw o(REGTMP) */
+		o1 = OP_IRR(c.opirr(ALUI), uint32(0), uint32(REGZERO), uint32(REGTMP))
+		rel := obj.Addrel(c.cursym)
+		rel.Off = int32(c.pc)
+		rel.Siz = 4
+		rel.Sym = p.From.Sym
+		rel.Add = p.From.Offset
+		rel.Type = objabi.R_ADDRMIPSU
+		o2 = OP_IRR(c.opirr(-p.As), uint32(0), uint32(REGTMP), uint32(p.To.Reg))
+		rel2 := obj.Addrel(c.cursym)
+		rel2.Off = int32(c.pc + 4)
+		rel2.Siz = 4
+		rel2.Sym = p.From.Sym
+		rel2.Add = p.From.Offset
+		rel2.Type = objabi.R_ADDRMIPS
+
+		if o.size == 12 {
+			o3 = o2
+			o2 = OP_RRR(c.oprrr(AADDVU), uint32(REGSB), uint32(REGTMP), uint32(REGTMP))
+			rel2.Off += 4
+		}
+
+	case 52: /* mov $lext, r ==> lu + add REGSB, r + add */
+		// NOTE: this case does not use REGTMP. If it ever does,
+		// remove the NOTUSETMP flag in optab.
+		o1 = OP_IRR(c.opirr(ALUI), uint32(0), uint32(REGZERO), uint32(p.To.Reg))
+		rel := obj.Addrel(c.cursym)
+		rel.Off = int32(c.pc)
+		rel.Siz = 4
+		rel.Sym = p.From.Sym
+		rel.Add = p.From.Offset
+		rel.Type = objabi.R_ADDRMIPSU
+		o2 = OP_IRR(c.opirr(add), uint32(0), uint32(p.To.Reg), uint32(p.To.Reg))
+		rel2 := obj.Addrel(c.cursym)
+		rel2.Off = int32(c.pc + 4)
+		rel2.Siz = 4
+		rel2.Sym = p.From.Sym
+		rel2.Add = p.From.Offset
+		rel2.Type = objabi.R_ADDRMIPS
+
+		if o.size == 12 {
+			o3 = o2
+			o2 = OP_RRR(c.oprrr(AADDVU), uint32(REGSB), uint32(p.To.Reg), uint32(p.To.Reg))
+			rel2.Off += 4
+		}
+
+	case 53: /* mov r, tlsvar ==> rdhwr + sw o(r3) */
+		// clobbers R3 !
+		// load thread pointer with RDHWR, R3 is used for fast kernel emulation on Linux
+		// NOTE: this case does not use REGTMP. If it ever does,
+		// remove the NOTUSETMP flag in optab.
+		o1 = (037<<26 + 073) | (29 << 11) | (3 << 16) // rdhwr $29, r3
+		o2 = OP_IRR(c.opirr(p.As), uint32(0), uint32(REG_R3), uint32(p.From.Reg))
+		rel := obj.Addrel(c.cursym)
+		rel.Off = int32(c.pc + 4)
+		rel.Siz = 4
+		rel.Sym = p.To.Sym
+		rel.Add = p.To.Offset
+		rel.Type = objabi.R_ADDRMIPSTLS
+
+	case 54: /* mov tlsvar, r ==> rdhwr + lw o(r3) */
+		// clobbers R3 !
+		// NOTE: this case does not use REGTMP. If it ever does,
+		// remove the NOTUSETMP flag in optab.
+		o1 = (037<<26 + 073) | (29 << 11) | (3 << 16) // rdhwr $29, r3
+		o2 = OP_IRR(c.opirr(-p.As), uint32(0), uint32(REG_R3), uint32(p.To.Reg))
+		rel := obj.Addrel(c.cursym)
+		rel.Off = int32(c.pc + 4)
+		rel.Siz = 4
+		rel.Sym = p.From.Sym
+		rel.Add = p.From.Offset
+		rel.Type = objabi.R_ADDRMIPSTLS
+
+	case 55: /* mov $tlsvar, r ==> rdhwr + add */
+		// clobbers R3 !
+		// NOTE: this case does not use REGTMP. If it ever does,
+		// remove the NOTUSETMP flag in optab.
+		o1 = (037<<26 + 073) | (29 << 11) | (3 << 16) // rdhwr $29, r3
+		o2 = OP_IRR(c.opirr(add), uint32(0), uint32(REG_R3), uint32(p.To.Reg))
+		rel := obj.Addrel(c.cursym)
+		rel.Off = int32(c.pc + 4)
+		rel.Siz = 4
+		rel.Sym = p.From.Sym
+		rel.Add = p.From.Offset
+		rel.Type = objabi.R_ADDRMIPSTLS
+
+	case 56: /* vmov{b,h,w,d} $scon, wr */
+
+		v := c.regoff(&p.From)
+		o1 = OP_VI10(110, c.twobitdf(p.As), v, uint32(p.To.Reg), 7)
+
+	case 57: /* vld $soreg, wr */
+		v := c.lsoffset(p.As, c.regoff(&p.From))
+		o1 = OP_VMI10(v, uint32(p.From.Reg), uint32(p.To.Reg), 8, c.twobitdf(p.As))
+
+	case 58: /* vst wr, $soreg */
+		v := c.lsoffset(p.As, c.regoff(&p.To))
+		o1 = OP_VMI10(v, uint32(p.To.Reg), uint32(p.From.Reg), 9, c.twobitdf(p.As))
+	}
+
+	out[0] = o1
+	out[1] = o2
+	out[2] = o3
+	out[3] = o4
+}
+
+func (c *ctxt0) vregoff(a *obj.Addr) int64 {
+	c.instoffset = 0
+	c.aclass(a)
+	return c.instoffset
+}
+
+func (c *ctxt0) regoff(a *obj.Addr) int32 {
+	return int32(c.vregoff(a))
+}
+
+func (c *ctxt0) oprrr(a obj.As) uint32 {
+	switch a {
+	case AADD:
+		return OP(4, 0)
+	case AADDU:
+		return OP(4, 1)
+	case ASGT:
+		return OP(5, 2)
+	case ASGTU:
+		return OP(5, 3)
+	case AAND:
+		return OP(4, 4)
+	case AOR:
+		return OP(4, 5)
+	case AXOR:
+		return OP(4, 6)
+	case ASUB:
+		return OP(4, 2)
+	case ASUBU, ANEGW:
+		return OP(4, 3)
+	case ANOR:
+		return OP(4, 7)
+	case ASLL:
+		return OP(0, 4)
+	case ASRL:
+		return OP(0, 6)
+	case ASRA:
+		return OP(0, 7)
+	case AROTR:
+		return OP(8, 6)
+	case ASLLV:
+		return OP(2, 4)
+	case ASRLV:
+		return OP(2, 6)
+	case ASRAV:
+		return OP(2, 7)
+	case AROTRV:
+		return OP(10, 6)
+	case AADDV:
+		return OP(5, 4)
+	case AADDVU:
+		return OP(5, 5)
+	case ASUBV:
+		return OP(5, 6)
+	case ASUBVU, ANEGV:
+		return OP(5, 7)
+	case AREM,
+		ADIV:
+		return OP(3, 2)
+	case AREMU,
+		ADIVU:
+		return OP(3, 3)
+	case AMUL:
+		return OP(3, 0)
+	case AMULU:
+		return OP(3, 1)
+	case AREMV,
+		ADIVV:
+		return OP(3, 6)
+	case AREMVU,
+		ADIVVU:
+		return OP(3, 7)
+	case AMULV:
+		return OP(3, 4)
+	case AMULVU:
+		return OP(3, 5)
+
+	case AJMP:
+		return OP(1, 0)
+	case AJAL:
+		return OP(1, 1)
+
+	case ABREAK:
+		return OP(1, 5)
+	case ASYSCALL:
+		return OP(1, 4)
+	case ATLBP:
+		return MMU(1, 0)
+	case ATLBR:
+		return MMU(0, 1)
+	case ATLBWI:
+		return MMU(0, 2)
+	case ATLBWR:
+		return MMU(0, 6)
+	case ARFE:
+		return MMU(2, 0)
+
+	case ADIVF:
+		return FPF(0, 3)
+	case ADIVD:
+		return FPD(0, 3)
+	case AMULF:
+		return FPF(0, 2)
+	case AMULD:
+		return FPD(0, 2)
+	case ASUBF:
+		return FPF(0, 1)
+	case ASUBD:
+		return FPD(0, 1)
+	case AADDF:
+		return FPF(0, 0)
+	case AADDD:
+		return FPD(0, 0)
+	case ATRUNCFV:
+		return FPF(1, 1)
+	case ATRUNCDV:
+		return FPD(1, 1)
+	case ATRUNCFW:
+		return FPF(1, 5)
+	case ATRUNCDW:
+		return FPD(1, 5)
+	case AMOVFV:
+		return FPF(4, 5)
+	case AMOVDV:
+		return FPD(4, 5)
+	case AMOVVF:
+		return FPV(4, 0)
+	case AMOVVD:
+		return FPV(4, 1)
+	case AMOVFW:
+		return FPF(4, 4)
+	case AMOVDW:
+		return FPD(4, 4)
+	case AMOVWF:
+		return FPW(4, 0)
+	case AMOVDF:
+		return FPD(4, 0)
+	case AMOVWD:
+		return FPW(4, 1)
+	case AMOVFD:
+		return FPF(4, 1)
+	case AABSF:
+		return FPF(0, 5)
+	case AABSD:
+		return FPD(0, 5)
+	case AMOVF:
+		return FPF(0, 6)
+	case AMOVD:
+		return FPD(0, 6)
+	case ANEGF:
+		return FPF(0, 7)
+	case ANEGD:
+		return FPD(0, 7)
+	case ACMPEQF:
+		return FPF(6, 2)
+	case ACMPEQD:
+		return FPD(6, 2)
+	case ACMPGTF:
+		return FPF(7, 4)
+	case ACMPGTD:
+		return FPD(7, 4)
+	case ACMPGEF:
+		return FPF(7, 6)
+	case ACMPGED:
+		return FPD(7, 6)
+
+	case ASQRTF:
+		return FPF(0, 4)
+	case ASQRTD:
+		return FPD(0, 4)
+
+	case ASYNC:
+		return OP(1, 7)
+	case ANOOP:
+		return 0
+
+	case ACMOVN:
+		return OP(1, 3)
+	case ACMOVZ:
+		return OP(1, 2)
+	case ACMOVT:
+		return OP(0, 1) | (1 << 16)
+	case ACMOVF:
+		return OP(0, 1) | (0 << 16)
+	case ACLO:
+		return SP(3, 4) | OP(4, 1)
+	case ACLZ:
+		return SP(3, 4) | OP(4, 0)
+	case AMADD:
+		return SP(3, 4) | OP(0, 0)
+	case AMSUB:
+		return SP(3, 4) | OP(0, 4)
+	}
+
+	if a < 0 {
+		c.ctxt.Diag("bad rrr opcode -%v", -a)
+	} else {
+		c.ctxt.Diag("bad rrr opcode %v", a)
+	}
+	return 0
+}
+
+func (c *ctxt0) opirr(a obj.As) uint32 {
+	switch a {
+	case AADD:
+		return SP(1, 0)
+	case AADDU:
+		return SP(1, 1)
+	case ASGT:
+		return SP(1, 2)
+	case ASGTU:
+		return SP(1, 3)
+	case AAND:
+		return SP(1, 4)
+	case AOR:
+		return SP(1, 5)
+	case AXOR:
+		return SP(1, 6)
+	case ALUI:
+		return SP(1, 7)
+	case ASLL:
+		return OP(0, 0)
+	case ASRL:
+		return OP(0, 2)
+	case ASRA:
+		return OP(0, 3)
+	case AROTR:
+		return OP(0, 2) | 1<<21
+	case AADDV:
+		return SP(3, 0)
+	case AADDVU:
+		return SP(3, 1)
+
+	case AJMP:
+		return SP(0, 2)
+	case AJAL,
+		obj.ADUFFZERO,
+		obj.ADUFFCOPY:
+		return SP(0, 3)
+	case ABEQ:
+		return SP(0, 4)
+	case -ABEQ:
+		return SP(2, 4) /* likely */
+	case ABNE:
+		return SP(0, 5)
+	case -ABNE:
+		return SP(2, 5) /* likely */
+	case ABGEZ:
+		return SP(0, 1) | BCOND(0, 1)
+	case -ABGEZ:
+		return SP(0, 1) | BCOND(0, 3) /* likely */
+	case ABGEZAL:
+		return SP(0, 1) | BCOND(2, 1)
+	case -ABGEZAL:
+		return SP(0, 1) | BCOND(2, 3) /* likely */
+	case ABGTZ:
+		return SP(0, 7)
+	case -ABGTZ:
+		return SP(2, 7) /* likely */
+	case ABLEZ:
+		return SP(0, 6)
+	case -ABLEZ:
+		return SP(2, 6) /* likely */
+	case ABLTZ:
+		return SP(0, 1) | BCOND(0, 0)
+	case -ABLTZ:
+		return SP(0, 1) | BCOND(0, 2) /* likely */
+	case ABLTZAL:
+		return SP(0, 1) | BCOND(2, 0)
+	case -ABLTZAL:
+		return SP(0, 1) | BCOND(2, 2) /* likely */
+	case ABFPT:
+		return SP(2, 1) | (257 << 16)
+	case -ABFPT:
+		return SP(2, 1) | (259 << 16) /* likely */
+	case ABFPF:
+		return SP(2, 1) | (256 << 16)
+	case -ABFPF:
+		return SP(2, 1) | (258 << 16) /* likely */
+
+	case AMOVB,
+		AMOVBU:
+		return SP(5, 0)
+	case AMOVH,
+		AMOVHU:
+		return SP(5, 1)
+	case AMOVW,
+		AMOVWU:
+		return SP(5, 3)
+	case AMOVV:
+		return SP(7, 7)
+	case AMOVF:
+		return SP(7, 1)
+	case AMOVD:
+		return SP(7, 5)
+	case AMOVWL:
+		return SP(5, 2)
+	case AMOVWR:
+		return SP(5, 6)
+	case AMOVVL:
+		return SP(5, 4)
+	case AMOVVR:
+		return SP(5, 5)
+
+	case ABREAK:
+		return SP(5, 7)
+
+	case -AMOVWL:
+		return SP(4, 2)
+	case -AMOVWR:
+		return SP(4, 6)
+	case -AMOVVL:
+		return SP(3, 2)
+	case -AMOVVR:
+		return SP(3, 3)
+	case -AMOVB:
+		return SP(4, 0)
+	case -AMOVBU:
+		return SP(4, 4)
+	case -AMOVH:
+		return SP(4, 1)
+	case -AMOVHU:
+		return SP(4, 5)
+	case -AMOVW:
+		return SP(4, 3)
+	case -AMOVWU:
+		return SP(4, 7)
+	case -AMOVV:
+		return SP(6, 7)
+	case -AMOVF:
+		return SP(6, 1)
+	case -AMOVD:
+		return SP(6, 5)
+
+	case ASLLV:
+		return OP(7, 0)
+	case ASRLV:
+		return OP(7, 2)
+	case ASRAV:
+		return OP(7, 3)
+	case AROTRV:
+		return OP(7, 2) | 1<<21
+	case -ASLLV:
+		return OP(7, 4)
+	case -ASRLV:
+		return OP(7, 6)
+	case -ASRAV:
+		return OP(7, 7)
+	case -AROTRV:
+		return OP(7, 6) | 1<<21
+
+	case ATEQ:
+		return OP(6, 4)
+	case ATNE:
+		return OP(6, 6)
+	case -ALL:
+		return SP(6, 0)
+	case -ALLV:
+		return SP(6, 4)
+	case ASC:
+		return SP(7, 0)
+	case ASCV:
+		return SP(7, 4)
+	}
+
+	if a < 0 {
+		c.ctxt.Diag("bad irr opcode -%v", -a)
+	} else {
+		c.ctxt.Diag("bad irr opcode %v", a)
+	}
+	return 0
+}
+
+func vshift(a obj.As) bool {
+	switch a {
+	case ASLLV,
+		ASRLV,
+		ASRAV,
+		AROTRV:
+		return true
+	}
+	return false
+}
+
+// MSA Two-bit Data Format Field Encoding
+func (c *ctxt0) twobitdf(a obj.As) uint32 {
+	switch a {
+	case AVMOVB:
+		return 0
+	case AVMOVH:
+		return 1
+	case AVMOVW:
+		return 2
+	case AVMOVD:
+		return 3
+	default:
+		c.ctxt.Diag("unsupported data format %v", a)
+	}
+	return 0
+}
+
+// MSA Load/Store offset have to be multiple of size of data format
+func (c *ctxt0) lsoffset(a obj.As, o int32) int32 {
+	var mod int32
+	switch a {
+	case AVMOVB:
+		mod = 1
+	case AVMOVH:
+		mod = 2
+	case AVMOVW:
+		mod = 4
+	case AVMOVD:
+		mod = 8
+	default:
+		c.ctxt.Diag("unsupported instruction:%v", a)
+	}
+
+	if o%mod != 0 {
+		c.ctxt.Diag("invalid offset for %v: %d is not a multiple of %d", a, o, mod)
+	}
+
+	return o / mod
+}
diff --git a/src/cmd/internal/obj/mips/list0.go b/src/cmd/internal/obj/mips/list0.go
new file mode 100644
index 0000000..f734e21
--- /dev/null
+++ b/src/cmd/internal/obj/mips/list0.go
@@ -0,0 +1,83 @@
+// cmd/9l/list.c from Vita Nuova.
+//
+//	Copyright © 1994-1999 Lucent Technologies Inc.  All rights reserved.
+//	Portions Copyright © 1995-1997 C H Forsyth (forsyth@terzarima.net)
+//	Portions Copyright © 1997-1999 Vita Nuova Limited
+//	Portions Copyright © 2000-2008 Vita Nuova Holdings Limited (www.vitanuova.com)
+//	Portions Copyright © 2004,2006 Bruce Ellis
+//	Portions Copyright © 2005-2007 C H Forsyth (forsyth@terzarima.net)
+//	Revisions Copyright © 2000-2008 Lucent Technologies Inc. and others
+//	Portions Copyright © 2009 The Go Authors. All rights reserved.
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+// THE SOFTWARE.
+
+package mips
+
+import (
+	"cmd/internal/obj"
+	"fmt"
+)
+
+func init() {
+	obj.RegisterRegister(obj.RBaseMIPS, REG_LAST+1, rconv)
+	obj.RegisterOpcode(obj.ABaseMIPS, Anames)
+}
+
+func rconv(r int) string {
+	if r == 0 {
+		return "NONE"
+	}
+	if r == REGG {
+		// Special case.
+		return "g"
+	}
+	if REG_R0 <= r && r <= REG_R31 {
+		return fmt.Sprintf("R%d", r-REG_R0)
+	}
+	if REG_F0 <= r && r <= REG_F31 {
+		return fmt.Sprintf("F%d", r-REG_F0)
+	}
+	if REG_M0 <= r && r <= REG_M31 {
+		return fmt.Sprintf("M%d", r-REG_M0)
+	}
+	if REG_FCR0 <= r && r <= REG_FCR31 {
+		return fmt.Sprintf("FCR%d", r-REG_FCR0)
+	}
+	if REG_W0 <= r && r <= REG_W31 {
+		return fmt.Sprintf("W%d", r-REG_W0)
+	}
+	if r == REG_HI {
+		return "HI"
+	}
+	if r == REG_LO {
+		return "LO"
+	}
+
+	return fmt.Sprintf("Rgok(%d)", r-obj.RBaseMIPS)
+}
+
+func DRconv(a int) string {
+	s := "C_??"
+	if a >= C_NONE && a <= C_NCLASS {
+		s = cnames0[a]
+	}
+	var fp string
+	fp += s
+	return fp
+}
diff --git a/src/cmd/internal/obj/mips/obj0.go b/src/cmd/internal/obj/mips/obj0.go
new file mode 100644
index 0000000..b96a28a
--- /dev/null
+++ b/src/cmd/internal/obj/mips/obj0.go
@@ -0,0 +1,1522 @@
+// cmd/9l/noop.c, cmd/9l/pass.c, cmd/9l/span.c from Vita Nuova.
+//
+//	Copyright © 1994-1999 Lucent Technologies Inc.  All rights reserved.
+//	Portions Copyright © 1995-1997 C H Forsyth (forsyth@terzarima.net)
+//	Portions Copyright © 1997-1999 Vita Nuova Limited
+//	Portions Copyright © 2000-2008 Vita Nuova Holdings Limited (www.vitanuova.com)
+//	Portions Copyright © 2004,2006 Bruce Ellis
+//	Portions Copyright © 2005-2007 C H Forsyth (forsyth@terzarima.net)
+//	Revisions Copyright © 2000-2008 Lucent Technologies Inc. and others
+//	Portions Copyright © 2009 The Go Authors. All rights reserved.
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+// THE SOFTWARE.
+
+package mips
+
+import (
+	"cmd/internal/obj"
+	"cmd/internal/objabi"
+	"cmd/internal/sys"
+	"encoding/binary"
+	"fmt"
+	"log"
+	"math"
+)
+
+func progedit(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) {
+	c := ctxt0{ctxt: ctxt, newprog: newprog}
+
+	p.From.Class = 0
+	p.To.Class = 0
+
+	// Rewrite JMP/JAL to symbol as TYPE_BRANCH.
+	switch p.As {
+	case AJMP,
+		AJAL,
+		ARET,
+		obj.ADUFFZERO,
+		obj.ADUFFCOPY:
+		if p.To.Sym != nil {
+			p.To.Type = obj.TYPE_BRANCH
+		}
+	}
+
+	// Rewrite float constants to values stored in memory.
+	switch p.As {
+	case AMOVF:
+		if p.From.Type == obj.TYPE_FCONST {
+			f32 := float32(p.From.Val.(float64))
+			if math.Float32bits(f32) == 0 {
+				p.As = AMOVW
+				p.From.Type = obj.TYPE_REG
+				p.From.Reg = REGZERO
+				break
+			}
+			p.From.Type = obj.TYPE_MEM
+			p.From.Sym = ctxt.Float32Sym(f32)
+			p.From.Name = obj.NAME_EXTERN
+			p.From.Offset = 0
+		}
+
+	case AMOVD:
+		if p.From.Type == obj.TYPE_FCONST {
+			f64 := p.From.Val.(float64)
+			if math.Float64bits(f64) == 0 && c.ctxt.Arch.Family == sys.MIPS64 {
+				p.As = AMOVV
+				p.From.Type = obj.TYPE_REG
+				p.From.Reg = REGZERO
+				break
+			}
+			p.From.Type = obj.TYPE_MEM
+			p.From.Sym = ctxt.Float64Sym(f64)
+			p.From.Name = obj.NAME_EXTERN
+			p.From.Offset = 0
+		}
+
+		// Put >32-bit constants in memory and load them
+	case AMOVV:
+		if p.From.Type == obj.TYPE_CONST && p.From.Name == obj.NAME_NONE && p.From.Reg == 0 && int64(int32(p.From.Offset)) != p.From.Offset {
+			p.From.Type = obj.TYPE_MEM
+			p.From.Sym = ctxt.Int64Sym(p.From.Offset)
+			p.From.Name = obj.NAME_EXTERN
+			p.From.Offset = 0
+		}
+	}
+
+	// Rewrite SUB constants into ADD.
+	switch p.As {
+	case ASUB:
+		if p.From.Type == obj.TYPE_CONST {
+			p.From.Offset = -p.From.Offset
+			p.As = AADD
+		}
+
+	case ASUBU:
+		if p.From.Type == obj.TYPE_CONST {
+			p.From.Offset = -p.From.Offset
+			p.As = AADDU
+		}
+
+	case ASUBV:
+		if p.From.Type == obj.TYPE_CONST {
+			p.From.Offset = -p.From.Offset
+			p.As = AADDV
+		}
+
+	case ASUBVU:
+		if p.From.Type == obj.TYPE_CONST {
+			p.From.Offset = -p.From.Offset
+			p.As = AADDVU
+		}
+	}
+}
+
+func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
+	// TODO(minux): add morestack short-cuts with small fixed frame-size.
+	c := ctxt0{ctxt: ctxt, newprog: newprog, cursym: cursym}
+
+	// a switch for enabling/disabling instruction scheduling
+	nosched := true
+
+	if c.cursym.Func().Text == nil || c.cursym.Func().Text.Link == nil {
+		return
+	}
+
+	p := c.cursym.Func().Text
+	textstksiz := p.To.Offset
+	if textstksiz == -ctxt.FixedFrameSize() {
+		// Historical way to mark NOFRAME.
+		p.From.Sym.Set(obj.AttrNoFrame, true)
+		textstksiz = 0
+	}
+	if textstksiz < 0 {
+		c.ctxt.Diag("negative frame size %d - did you mean NOFRAME?", textstksiz)
+	}
+	if p.From.Sym.NoFrame() {
+		if textstksiz != 0 {
+			c.ctxt.Diag("NOFRAME functions must have a frame size of 0, not %d", textstksiz)
+		}
+	}
+
+	c.cursym.Func().Args = p.To.Val.(int32)
+	c.cursym.Func().Locals = int32(textstksiz)
+
+	/*
+	 * find leaf subroutines
+	 * expand RET
+	 * expand BECOME pseudo
+	 */
+
+	for p := c.cursym.Func().Text; p != nil; p = p.Link {
+		switch p.As {
+		/* too hard, just leave alone */
+		case obj.ATEXT:
+			p.Mark |= LABEL | LEAF | SYNC
+			if p.Link != nil {
+				p.Link.Mark |= LABEL
+			}
+
+		/* too hard, just leave alone */
+		case AMOVW,
+			AMOVV:
+			if p.To.Type == obj.TYPE_REG && p.To.Reg >= REG_SPECIAL {
+				p.Mark |= LABEL | SYNC
+				break
+			}
+			if p.From.Type == obj.TYPE_REG && p.From.Reg >= REG_SPECIAL {
+				p.Mark |= LABEL | SYNC
+			}
+
+		/* too hard, just leave alone */
+		case ASYSCALL,
+			AWORD,
+			ATLBWR,
+			ATLBWI,
+			ATLBP,
+			ATLBR:
+			p.Mark |= LABEL | SYNC
+
+		case ANOR:
+			if p.To.Type == obj.TYPE_REG {
+				if p.To.Reg == REGZERO {
+					p.Mark |= LABEL | SYNC
+				}
+			}
+
+		case ABGEZAL,
+			ABLTZAL,
+			AJAL,
+			obj.ADUFFZERO,
+			obj.ADUFFCOPY:
+			c.cursym.Func().Text.Mark &^= LEAF
+			fallthrough
+
+		case AJMP,
+			ABEQ,
+			ABGEZ,
+			ABGTZ,
+			ABLEZ,
+			ABLTZ,
+			ABNE,
+			ABFPT, ABFPF:
+			if p.As == ABFPT || p.As == ABFPF {
+				// We don't treat ABFPT and ABFPF as branches here,
+				// so that we will always fill nop (0x0) in their
+				// delay slot during assembly.
+				// This is to workaround a kernel FPU emulator bug
+				// where it uses the user stack to simulate the
+				// instruction in the delay slot if it's not 0x0,
+				// and somehow that leads to SIGSEGV when the kernel
+				// jump to the stack.
+				p.Mark |= SYNC
+			} else {
+				p.Mark |= BRANCH
+			}
+			q1 := p.To.Target()
+			if q1 != nil {
+				for q1.As == obj.ANOP {
+					q1 = q1.Link
+					p.To.SetTarget(q1)
+				}
+
+				if q1.Mark&LEAF == 0 {
+					q1.Mark |= LABEL
+				}
+			}
+			//else {
+			//	p.Mark |= LABEL
+			//}
+			q1 = p.Link
+			if q1 != nil {
+				q1.Mark |= LABEL
+			}
+
+		case ARET:
+			if p.Link != nil {
+				p.Link.Mark |= LABEL
+			}
+		}
+	}
+
+	var mov, add obj.As
+	if c.ctxt.Arch.Family == sys.MIPS64 {
+		add = AADDV
+		mov = AMOVV
+	} else {
+		add = AADDU
+		mov = AMOVW
+	}
+
+	var q *obj.Prog
+	var q1 *obj.Prog
+	autosize := int32(0)
+	var p1 *obj.Prog
+	var p2 *obj.Prog
+	for p := c.cursym.Func().Text; p != nil; p = p.Link {
+		o := p.As
+		switch o {
+		case obj.ATEXT:
+			autosize = int32(textstksiz)
+
+			if p.Mark&LEAF != 0 && autosize == 0 {
+				// A leaf function with no locals has no frame.
+				p.From.Sym.Set(obj.AttrNoFrame, true)
+			}
+
+			if !p.From.Sym.NoFrame() {
+				// If there is a stack frame at all, it includes
+				// space to save the LR.
+				autosize += int32(c.ctxt.FixedFrameSize())
+			}
+
+			if autosize&4 != 0 && c.ctxt.Arch.Family == sys.MIPS64 {
+				autosize += 4
+			}
+
+			if autosize == 0 && c.cursym.Func().Text.Mark&LEAF == 0 {
+				if c.cursym.Func().Text.From.Sym.NoSplit() {
+					if ctxt.Debugvlog {
+						ctxt.Logf("save suppressed in: %s\n", c.cursym.Name)
+					}
+
+					c.cursym.Func().Text.Mark |= LEAF
+				}
+			}
+
+			p.To.Offset = int64(autosize) - ctxt.FixedFrameSize()
+
+			if c.cursym.Func().Text.Mark&LEAF != 0 {
+				c.cursym.Set(obj.AttrLeaf, true)
+				if p.From.Sym.NoFrame() {
+					break
+				}
+			}
+
+			if !p.From.Sym.NoSplit() {
+				p = c.stacksplit(p, autosize) // emit split check
+			}
+
+			q = p
+
+			if autosize != 0 {
+				// Make sure to save link register for non-empty frame, even if
+				// it is a leaf function, so that traceback works.
+				// Store link register before decrement SP, so if a signal comes
+				// during the execution of the function prologue, the traceback
+				// code will not see a half-updated stack frame.
+				// This sequence is not async preemptible, as if we open a frame
+				// at the current SP, it will clobber the saved LR.
+				q = c.ctxt.StartUnsafePoint(q, c.newprog)
+
+				q = obj.Appendp(q, newprog)
+				q.As = mov
+				q.Pos = p.Pos
+				q.From.Type = obj.TYPE_REG
+				q.From.Reg = REGLINK
+				q.To.Type = obj.TYPE_MEM
+				q.To.Offset = int64(-autosize)
+				q.To.Reg = REGSP
+
+				q = obj.Appendp(q, newprog)
+				q.As = add
+				q.Pos = p.Pos
+				q.From.Type = obj.TYPE_CONST
+				q.From.Offset = int64(-autosize)
+				q.To.Type = obj.TYPE_REG
+				q.To.Reg = REGSP
+				q.Spadj = +autosize
+
+				q = c.ctxt.EndUnsafePoint(q, c.newprog, -1)
+			}
+
+			if c.cursym.Func().Text.From.Sym.Wrapper() && c.cursym.Func().Text.Mark&LEAF == 0 {
+				// if(g->panic != nil && g->panic->argp == FP) g->panic->argp = bottom-of-frame
+				//
+				//	MOV	g_panic(g), R1
+				//	BEQ	R1, end
+				//	MOV	panic_argp(R1), R2
+				//	ADD	$(autosize+FIXED_FRAME), R29, R3
+				//	BNE	R2, R3, end
+				//	ADD	$FIXED_FRAME, R29, R2
+				//	MOV	R2, panic_argp(R1)
+				// end:
+				//	NOP
+				//
+				// The NOP is needed to give the jumps somewhere to land.
+				// It is a liblink NOP, not an mips NOP: it encodes to 0 instruction bytes.
+				//
+				// We don't generate this for leafs because that means the wrapped
+				// function was inlined into the wrapper.
+
+				q = obj.Appendp(q, newprog)
+
+				q.As = mov
+				q.From.Type = obj.TYPE_MEM
+				q.From.Reg = REGG
+				q.From.Offset = 4 * int64(c.ctxt.Arch.PtrSize) // G.panic
+				q.To.Type = obj.TYPE_REG
+				q.To.Reg = REG_R1
+
+				q = obj.Appendp(q, newprog)
+				q.As = ABEQ
+				q.From.Type = obj.TYPE_REG
+				q.From.Reg = REG_R1
+				q.To.Type = obj.TYPE_BRANCH
+				q.Mark |= BRANCH
+				p1 = q
+
+				q = obj.Appendp(q, newprog)
+				q.As = mov
+				q.From.Type = obj.TYPE_MEM
+				q.From.Reg = REG_R1
+				q.From.Offset = 0 // Panic.argp
+				q.To.Type = obj.TYPE_REG
+				q.To.Reg = REG_R2
+
+				q = obj.Appendp(q, newprog)
+				q.As = add
+				q.From.Type = obj.TYPE_CONST
+				q.From.Offset = int64(autosize) + ctxt.FixedFrameSize()
+				q.Reg = REGSP
+				q.To.Type = obj.TYPE_REG
+				q.To.Reg = REG_R3
+
+				q = obj.Appendp(q, newprog)
+				q.As = ABNE
+				q.From.Type = obj.TYPE_REG
+				q.From.Reg = REG_R2
+				q.Reg = REG_R3
+				q.To.Type = obj.TYPE_BRANCH
+				q.Mark |= BRANCH
+				p2 = q
+
+				q = obj.Appendp(q, newprog)
+				q.As = add
+				q.From.Type = obj.TYPE_CONST
+				q.From.Offset = ctxt.FixedFrameSize()
+				q.Reg = REGSP
+				q.To.Type = obj.TYPE_REG
+				q.To.Reg = REG_R2
+
+				q = obj.Appendp(q, newprog)
+				q.As = mov
+				q.From.Type = obj.TYPE_REG
+				q.From.Reg = REG_R2
+				q.To.Type = obj.TYPE_MEM
+				q.To.Reg = REG_R1
+				q.To.Offset = 0 // Panic.argp
+
+				q = obj.Appendp(q, newprog)
+
+				q.As = obj.ANOP
+				p1.To.SetTarget(q)
+				p2.To.SetTarget(q)
+			}
+
+		case ARET:
+			if p.From.Type == obj.TYPE_CONST {
+				ctxt.Diag("using BECOME (%v) is not supported!", p)
+				break
+			}
+
+			retSym := p.To.Sym
+			p.To.Name = obj.NAME_NONE // clear fields as we may modify p to other instruction
+			p.To.Sym = nil
+
+			if c.cursym.Func().Text.Mark&LEAF != 0 {
+				if autosize == 0 {
+					p.As = AJMP
+					p.From = obj.Addr{}
+					if retSym != nil { // retjmp
+						p.To.Type = obj.TYPE_BRANCH
+						p.To.Name = obj.NAME_EXTERN
+						p.To.Sym = retSym
+					} else {
+						p.To.Type = obj.TYPE_MEM
+						p.To.Reg = REGLINK
+						p.To.Offset = 0
+					}
+					p.Mark |= BRANCH
+					break
+				}
+
+				p.As = add
+				p.From.Type = obj.TYPE_CONST
+				p.From.Offset = int64(autosize)
+				p.To.Type = obj.TYPE_REG
+				p.To.Reg = REGSP
+				p.Spadj = -autosize
+
+				q = c.newprog()
+				q.As = AJMP
+				q.Pos = p.Pos
+				if retSym != nil { // retjmp
+					q.To.Type = obj.TYPE_BRANCH
+					q.To.Name = obj.NAME_EXTERN
+					q.To.Sym = retSym
+				} else {
+					q.To.Type = obj.TYPE_MEM
+					q.To.Reg = REGLINK
+					q.To.Offset = 0
+				}
+				q.Mark |= BRANCH
+				q.Spadj = +autosize
+
+				q.Link = p.Link
+				p.Link = q
+				break
+			}
+
+			p.As = mov
+			p.From.Type = obj.TYPE_MEM
+			p.From.Offset = 0
+			p.From.Reg = REGSP
+			p.To.Type = obj.TYPE_REG
+			p.To.Reg = REGLINK
+
+			if autosize != 0 {
+				q = c.newprog()
+				q.As = add
+				q.Pos = p.Pos
+				q.From.Type = obj.TYPE_CONST
+				q.From.Offset = int64(autosize)
+				q.To.Type = obj.TYPE_REG
+				q.To.Reg = REGSP
+				q.Spadj = -autosize
+
+				q.Link = p.Link
+				p.Link = q
+			}
+
+			q1 = c.newprog()
+			q1.As = AJMP
+			q1.Pos = p.Pos
+			if retSym != nil { // retjmp
+				q1.To.Type = obj.TYPE_BRANCH
+				q1.To.Name = obj.NAME_EXTERN
+				q1.To.Sym = retSym
+			} else {
+				q1.To.Type = obj.TYPE_MEM
+				q1.To.Offset = 0
+				q1.To.Reg = REGLINK
+			}
+			q1.Mark |= BRANCH
+			q1.Spadj = +autosize
+
+			q1.Link = q.Link
+			q.Link = q1
+
+		case AADD,
+			AADDU,
+			AADDV,
+			AADDVU:
+			if p.To.Type == obj.TYPE_REG && p.To.Reg == REGSP && p.From.Type == obj.TYPE_CONST {
+				p.Spadj = int32(-p.From.Offset)
+			}
+
+		case obj.AGETCALLERPC:
+			if cursym.Leaf() {
+				/* MOV LR, Rd */
+				p.As = mov
+				p.From.Type = obj.TYPE_REG
+				p.From.Reg = REGLINK
+			} else {
+				/* MOV (RSP), Rd */
+				p.As = mov
+				p.From.Type = obj.TYPE_MEM
+				p.From.Reg = REGSP
+			}
+		}
+
+		if p.To.Type == obj.TYPE_REG && p.To.Reg == REGSP && p.Spadj == 0 {
+			f := c.cursym.Func()
+			if f.FuncFlag&objabi.FuncFlag_SPWRITE == 0 {
+				c.cursym.Func().FuncFlag |= objabi.FuncFlag_SPWRITE
+				if ctxt.Debugvlog || !ctxt.IsAsm {
+					ctxt.Logf("auto-SPWRITE: %s %v\n", c.cursym.Name, p)
+					if !ctxt.IsAsm {
+						ctxt.Diag("invalid auto-SPWRITE in non-assembly")
+						ctxt.DiagFlush()
+						log.Fatalf("bad SPWRITE")
+					}
+				}
+			}
+		}
+	}
+
+	if c.ctxt.Arch.Family == sys.MIPS {
+		// rewrite MOVD into two MOVF in 32-bit mode to avoid unaligned memory access
+		for p = c.cursym.Func().Text; p != nil; p = p1 {
+			p1 = p.Link
+
+			if p.As != AMOVD {
+				continue
+			}
+			if p.From.Type != obj.TYPE_MEM && p.To.Type != obj.TYPE_MEM {
+				continue
+			}
+
+			p.As = AMOVF
+			q = c.newprog()
+			*q = *p
+			q.Link = p.Link
+			p.Link = q
+			p1 = q.Link
+
+			var addrOff int64
+			if c.ctxt.Arch.ByteOrder == binary.BigEndian {
+				addrOff = 4 // swap load/save order
+			}
+			if p.From.Type == obj.TYPE_MEM {
+				reg := REG_F0 + (p.To.Reg-REG_F0)&^1
+				p.To.Reg = reg
+				q.To.Reg = reg + 1
+				p.From.Offset += addrOff
+				q.From.Offset += 4 - addrOff
+			} else if p.To.Type == obj.TYPE_MEM {
+				reg := REG_F0 + (p.From.Reg-REG_F0)&^1
+				p.From.Reg = reg
+				q.From.Reg = reg + 1
+				p.To.Offset += addrOff
+				q.To.Offset += 4 - addrOff
+			}
+		}
+	}
+
+	if nosched {
+		// if we don't do instruction scheduling, simply add
+		// NOP after each branch instruction.
+		for p = c.cursym.Func().Text; p != nil; p = p.Link {
+			if p.Mark&BRANCH != 0 {
+				c.addnop(p)
+			}
+		}
+		return
+	}
+
+	// instruction scheduling
+	q = nil                   // p - 1
+	q1 = c.cursym.Func().Text // top of block
+	o := 0                    // count of instructions
+	for p = c.cursym.Func().Text; p != nil; p = p1 {
+		p1 = p.Link
+		o++
+		if p.Mark&NOSCHED != 0 {
+			if q1 != p {
+				c.sched(q1, q)
+			}
+			for ; p != nil; p = p.Link {
+				if p.Mark&NOSCHED == 0 {
+					break
+				}
+				q = p
+			}
+			p1 = p
+			q1 = p
+			o = 0
+			continue
+		}
+		if p.Mark&(LABEL|SYNC) != 0 {
+			if q1 != p {
+				c.sched(q1, q)
+			}
+			q1 = p
+			o = 1
+		}
+		if p.Mark&(BRANCH|SYNC) != 0 {
+			c.sched(q1, p)
+			q1 = p1
+			o = 0
+		}
+		if o >= NSCHED {
+			c.sched(q1, p)
+			q1 = p1
+			o = 0
+		}
+		q = p
+	}
+}
+
+func (c *ctxt0) stacksplit(p *obj.Prog, framesize int32) *obj.Prog {
+	var mov, add obj.As
+
+	if c.ctxt.Arch.Family == sys.MIPS64 {
+		add = AADDV
+		mov = AMOVV
+	} else {
+		add = AADDU
+		mov = AMOVW
+	}
+
+	if c.ctxt.Flag_maymorestack != "" {
+		// Save LR and REGCTXT.
+		frameSize := 2 * c.ctxt.Arch.PtrSize
+
+		p = c.ctxt.StartUnsafePoint(p, c.newprog)
+
+		// MOV	REGLINK, -8/-16(SP)
+		p = obj.Appendp(p, c.newprog)
+		p.As = mov
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = REGLINK
+		p.To.Type = obj.TYPE_MEM
+		p.To.Offset = int64(-frameSize)
+		p.To.Reg = REGSP
+
+		// MOV	REGCTXT, -4/-8(SP)
+		p = obj.Appendp(p, c.newprog)
+		p.As = mov
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = REGCTXT
+		p.To.Type = obj.TYPE_MEM
+		p.To.Offset = -int64(c.ctxt.Arch.PtrSize)
+		p.To.Reg = REGSP
+
+		// ADD	$-8/$-16, SP
+		p = obj.Appendp(p, c.newprog)
+		p.As = add
+		p.From.Type = obj.TYPE_CONST
+		p.From.Offset = int64(-frameSize)
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = REGSP
+		p.Spadj = int32(frameSize)
+
+		// JAL	maymorestack
+		p = obj.Appendp(p, c.newprog)
+		p.As = AJAL
+		p.To.Type = obj.TYPE_BRANCH
+		// See ../x86/obj6.go
+		p.To.Sym = c.ctxt.LookupABI(c.ctxt.Flag_maymorestack, c.cursym.ABI())
+		p.Mark |= BRANCH
+
+		// Restore LR and REGCTXT.
+
+		// MOV	0(SP), REGLINK
+		p = obj.Appendp(p, c.newprog)
+		p.As = mov
+		p.From.Type = obj.TYPE_MEM
+		p.From.Offset = 0
+		p.From.Reg = REGSP
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = REGLINK
+
+		// MOV	4/8(SP), REGCTXT
+		p = obj.Appendp(p, c.newprog)
+		p.As = mov
+		p.From.Type = obj.TYPE_MEM
+		p.From.Offset = int64(c.ctxt.Arch.PtrSize)
+		p.From.Reg = REGSP
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = REGCTXT
+
+		// ADD	$8/$16, SP
+		p = obj.Appendp(p, c.newprog)
+		p.As = add
+		p.From.Type = obj.TYPE_CONST
+		p.From.Offset = int64(frameSize)
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = REGSP
+		p.Spadj = int32(-frameSize)
+
+		p = c.ctxt.EndUnsafePoint(p, c.newprog, -1)
+	}
+
+	// Jump back to here after morestack returns.
+	startPred := p
+
+	// MOV	g_stackguard(g), R1
+	p = obj.Appendp(p, c.newprog)
+
+	p.As = mov
+	p.From.Type = obj.TYPE_MEM
+	p.From.Reg = REGG
+	p.From.Offset = 2 * int64(c.ctxt.Arch.PtrSize) // G.stackguard0
+	if c.cursym.CFunc() {
+		p.From.Offset = 3 * int64(c.ctxt.Arch.PtrSize) // G.stackguard1
+	}
+	p.To.Type = obj.TYPE_REG
+	p.To.Reg = REG_R1
+
+	// Mark the stack bound check and morestack call async nonpreemptible.
+	// If we get preempted here, when resumed the preemption request is
+	// cleared, but we'll still call morestack, which will double the stack
+	// unnecessarily. See issue #35470.
+	p = c.ctxt.StartUnsafePoint(p, c.newprog)
+
+	var q *obj.Prog
+	if framesize <= objabi.StackSmall {
+		// small stack: SP < stackguard
+		//	AGTU	SP, stackguard, R1
+		p = obj.Appendp(p, c.newprog)
+
+		p.As = ASGTU
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = REGSP
+		p.Reg = REG_R1
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = REG_R1
+	} else {
+		// large stack: SP-framesize < stackguard-StackSmall
+		offset := int64(framesize) - objabi.StackSmall
+		if framesize > objabi.StackBig {
+			// Such a large stack we need to protect against underflow.
+			// The runtime guarantees SP > objabi.StackBig, but
+			// framesize is large enough that SP-framesize may
+			// underflow, causing a direct comparison with the
+			// stack guard to incorrectly succeed. We explicitly
+			// guard against underflow.
+			//
+			//	SGTU	$(framesize-StackSmall), SP, R2
+			//	BNE	R2, label-of-call-to-morestack
+
+			p = obj.Appendp(p, c.newprog)
+			p.As = ASGTU
+			p.From.Type = obj.TYPE_CONST
+			p.From.Offset = offset
+			p.Reg = REGSP
+			p.To.Type = obj.TYPE_REG
+			p.To.Reg = REG_R2
+
+			p = obj.Appendp(p, c.newprog)
+			q = p
+			p.As = ABNE
+			p.From.Type = obj.TYPE_REG
+			p.From.Reg = REG_R2
+			p.To.Type = obj.TYPE_BRANCH
+			p.Mark |= BRANCH
+		}
+
+		// Check against the stack guard. We've ensured this won't underflow.
+		//	ADD	$-(framesize-StackSmall), SP, R2
+		//	SGTU	R2, stackguard, R1
+		p = obj.Appendp(p, c.newprog)
+
+		p.As = add
+		p.From.Type = obj.TYPE_CONST
+		p.From.Offset = -offset
+		p.Reg = REGSP
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = REG_R2
+
+		p = obj.Appendp(p, c.newprog)
+		p.As = ASGTU
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = REG_R2
+		p.Reg = REG_R1
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = REG_R1
+	}
+
+	// q1: BNE	R1, done
+	p = obj.Appendp(p, c.newprog)
+	q1 := p
+
+	p.As = ABNE
+	p.From.Type = obj.TYPE_REG
+	p.From.Reg = REG_R1
+	p.To.Type = obj.TYPE_BRANCH
+	p.Mark |= BRANCH
+
+	// MOV	LINK, R3
+	p = obj.Appendp(p, c.newprog)
+
+	p.As = mov
+	p.From.Type = obj.TYPE_REG
+	p.From.Reg = REGLINK
+	p.To.Type = obj.TYPE_REG
+	p.To.Reg = REG_R3
+	if q != nil {
+		q.To.SetTarget(p)
+		p.Mark |= LABEL
+	}
+
+	p = c.ctxt.EmitEntryStackMap(c.cursym, p, c.newprog)
+
+	// JAL	runtime.morestack(SB)
+	p = obj.Appendp(p, c.newprog)
+
+	p.As = AJAL
+	p.To.Type = obj.TYPE_BRANCH
+	if c.cursym.CFunc() {
+		p.To.Sym = c.ctxt.Lookup("runtime.morestackc")
+	} else if !c.cursym.Func().Text.From.Sym.NeedCtxt() {
+		p.To.Sym = c.ctxt.Lookup("runtime.morestack_noctxt")
+	} else {
+		p.To.Sym = c.ctxt.Lookup("runtime.morestack")
+	}
+	p.Mark |= BRANCH
+
+	p = c.ctxt.EndUnsafePoint(p, c.newprog, -1)
+
+	// JMP	start
+	p = obj.Appendp(p, c.newprog)
+
+	p.As = AJMP
+	p.To.Type = obj.TYPE_BRANCH
+	p.To.SetTarget(startPred.Link)
+	startPred.Link.Mark |= LABEL
+	p.Mark |= BRANCH
+
+	// placeholder for q1's jump target
+	p = obj.Appendp(p, c.newprog)
+
+	p.As = obj.ANOP // zero-width place holder
+	q1.To.SetTarget(p)
+
+	return p
+}
+
+func (c *ctxt0) addnop(p *obj.Prog) {
+	q := c.newprog()
+	q.As = ANOOP
+	q.Pos = p.Pos
+	q.Link = p.Link
+	p.Link = q
+}
+
+const (
+	E_HILO  = 1 << 0
+	E_FCR   = 1 << 1
+	E_MCR   = 1 << 2
+	E_MEM   = 1 << 3
+	E_MEMSP = 1 << 4 /* uses offset and size */
+	E_MEMSB = 1 << 5 /* uses offset and size */
+	ANYMEM  = E_MEM | E_MEMSP | E_MEMSB
+	//DELAY = LOAD|BRANCH|FCMP
+	DELAY = BRANCH /* only schedule branch */
+)
+
+type Dep struct {
+	ireg uint32
+	freg uint32
+	cc   uint32
+}
+
+type Sch struct {
+	p       obj.Prog
+	set     Dep
+	used    Dep
+	soffset int32
+	size    uint8
+	nop     uint8
+	comp    bool
+}
+
+func (c *ctxt0) sched(p0, pe *obj.Prog) {
+	var sch [NSCHED]Sch
+
+	/*
+	 * build side structure
+	 */
+	s := sch[:]
+	for p := p0; ; p = p.Link {
+		s[0].p = *p
+		c.markregused(&s[0])
+		if p == pe {
+			break
+		}
+		s = s[1:]
+	}
+	se := s
+
+	for i := cap(sch) - cap(se); i >= 0; i-- {
+		s = sch[i:]
+		if s[0].p.Mark&DELAY == 0 {
+			continue
+		}
+		if -cap(s) < -cap(se) {
+			if !conflict(&s[0], &s[1]) {
+				continue
+			}
+		}
+
+		var t []Sch
+		var j int
+		for j = cap(sch) - cap(s) - 1; j >= 0; j-- {
+			t = sch[j:]
+			if t[0].comp {
+				if s[0].p.Mark&BRANCH != 0 {
+					continue
+				}
+			}
+			if t[0].p.Mark&DELAY != 0 {
+				if -cap(s) >= -cap(se) || conflict(&t[0], &s[1]) {
+					continue
+				}
+			}
+			for u := t[1:]; -cap(u) <= -cap(s); u = u[1:] {
+				if c.depend(&u[0], &t[0]) {
+					continue
+				}
+			}
+			goto out2
+		}
+
+		if s[0].p.Mark&BRANCH != 0 {
+			s[0].nop = 1
+		}
+		continue
+
+	out2:
+		// t[0] is the instruction being moved to fill the delay
+		stmp := t[0]
+		copy(t[:i-j], t[1:i-j+1])
+		s[0] = stmp
+
+		if t[i-j-1].p.Mark&BRANCH != 0 {
+			// t[i-j] is being put into a branch delay slot
+			// combine its Spadj with the branch instruction
+			t[i-j-1].p.Spadj += t[i-j].p.Spadj
+			t[i-j].p.Spadj = 0
+		}
+
+		i--
+	}
+
+	/*
+	 * put it all back
+	 */
+	var p *obj.Prog
+	var q *obj.Prog
+	for s, p = sch[:], p0; -cap(s) <= -cap(se); s, p = s[1:], q {
+		q = p.Link
+		if q != s[0].p.Link {
+			*p = s[0].p
+			p.Link = q
+		}
+		for s[0].nop != 0 {
+			s[0].nop--
+			c.addnop(p)
+		}
+	}
+}
+
+func (c *ctxt0) markregused(s *Sch) {
+	p := &s.p
+	s.comp = c.compound(p)
+	s.nop = 0
+	if s.comp {
+		s.set.ireg |= 1 << (REGTMP - REG_R0)
+		s.used.ireg |= 1 << (REGTMP - REG_R0)
+	}
+
+	ar := 0  /* dest is really reference */
+	ad := 0  /* source/dest is really address */
+	ld := 0  /* opcode is load instruction */
+	sz := 20 /* size of load/store for overlap computation */
+
+	/*
+	 * flags based on opcode
+	 */
+	switch p.As {
+	case obj.ATEXT:
+		c.autosize = int32(p.To.Offset + 8)
+		ad = 1
+
+	case AJAL:
+		r := p.Reg
+		if r == 0 {
+			r = REGLINK
+		}
+		s.set.ireg |= 1 << uint(r-REG_R0)
+		ar = 1
+		ad = 1
+
+	case ABGEZAL,
+		ABLTZAL:
+		s.set.ireg |= 1 << (REGLINK - REG_R0)
+		fallthrough
+	case ABEQ,
+		ABGEZ,
+		ABGTZ,
+		ABLEZ,
+		ABLTZ,
+		ABNE:
+		ar = 1
+		ad = 1
+
+	case ABFPT,
+		ABFPF:
+		ad = 1
+		s.used.cc |= E_FCR
+
+	case ACMPEQD,
+		ACMPEQF,
+		ACMPGED,
+		ACMPGEF,
+		ACMPGTD,
+		ACMPGTF:
+		ar = 1
+		s.set.cc |= E_FCR
+		p.Mark |= FCMP
+
+	case AJMP:
+		ar = 1
+		ad = 1
+
+	case AMOVB,
+		AMOVBU:
+		sz = 1
+		ld = 1
+
+	case AMOVH,
+		AMOVHU:
+		sz = 2
+		ld = 1
+
+	case AMOVF,
+		AMOVW,
+		AMOVWL,
+		AMOVWR:
+		sz = 4
+		ld = 1
+
+	case AMOVD,
+		AMOVV,
+		AMOVVL,
+		AMOVVR:
+		sz = 8
+		ld = 1
+
+	case ADIV,
+		ADIVU,
+		AMUL,
+		AMULU,
+		AREM,
+		AREMU,
+		ADIVV,
+		ADIVVU,
+		AMULV,
+		AMULVU,
+		AREMV,
+		AREMVU:
+		s.set.cc = E_HILO
+		fallthrough
+	case AADD,
+		AADDU,
+		AADDV,
+		AADDVU,
+		AAND,
+		ANOR,
+		AOR,
+		ASGT,
+		ASGTU,
+		ASLL,
+		ASRA,
+		ASRL,
+		ASLLV,
+		ASRAV,
+		ASRLV,
+		ASUB,
+		ASUBU,
+		ASUBV,
+		ASUBVU,
+		AXOR,
+
+		AADDD,
+		AADDF,
+		AADDW,
+		ASUBD,
+		ASUBF,
+		ASUBW,
+		AMULF,
+		AMULD,
+		AMULW,
+		ADIVF,
+		ADIVD,
+		ADIVW:
+		if p.Reg == 0 {
+			if p.To.Type == obj.TYPE_REG {
+				p.Reg = p.To.Reg
+			}
+			//if(p->reg == NREG)
+			//	print("botch %P\n", p);
+		}
+	}
+
+	/*
+	 * flags based on 'to' field
+	 */
+	cls := int(p.To.Class)
+	if cls == 0 {
+		cls = c.aclass(&p.To) + 1
+		p.To.Class = int8(cls)
+	}
+	cls--
+	switch cls {
+	default:
+		fmt.Printf("unknown class %d %v\n", cls, p)
+
+	case C_ZCON,
+		C_SCON,
+		C_ADD0CON,
+		C_AND0CON,
+		C_ADDCON,
+		C_ANDCON,
+		C_UCON,
+		C_LCON,
+		C_NONE,
+		C_SBRA,
+		C_LBRA,
+		C_ADDR,
+		C_TEXTSIZE:
+		break
+
+	case C_HI,
+		C_LO:
+		s.set.cc |= E_HILO
+
+	case C_FCREG:
+		s.set.cc |= E_FCR
+
+	case C_MREG:
+		s.set.cc |= E_MCR
+
+	case C_ZOREG,
+		C_SOREG,
+		C_LOREG:
+		cls = int(p.To.Reg)
+		s.used.ireg |= 1 << uint(cls-REG_R0)
+		if ad != 0 {
+			break
+		}
+		s.size = uint8(sz)
+		s.soffset = c.regoff(&p.To)
+
+		m := uint32(ANYMEM)
+		if cls == REGSB {
+			m = E_MEMSB
+		}
+		if cls == REGSP {
+			m = E_MEMSP
+		}
+
+		if ar != 0 {
+			s.used.cc |= m
+		} else {
+			s.set.cc |= m
+		}
+
+	case C_SACON,
+		C_LACON:
+		s.used.ireg |= 1 << (REGSP - REG_R0)
+
+	case C_SECON,
+		C_LECON:
+		s.used.ireg |= 1 << (REGSB - REG_R0)
+
+	case C_REG:
+		if ar != 0 {
+			s.used.ireg |= 1 << uint(p.To.Reg-REG_R0)
+		} else {
+			s.set.ireg |= 1 << uint(p.To.Reg-REG_R0)
+		}
+
+	case C_FREG:
+		if ar != 0 {
+			s.used.freg |= 1 << uint(p.To.Reg-REG_F0)
+		} else {
+			s.set.freg |= 1 << uint(p.To.Reg-REG_F0)
+		}
+		if ld != 0 && p.From.Type == obj.TYPE_REG {
+			p.Mark |= LOAD
+		}
+
+	case C_SAUTO,
+		C_LAUTO:
+		s.used.ireg |= 1 << (REGSP - REG_R0)
+		if ad != 0 {
+			break
+		}
+		s.size = uint8(sz)
+		s.soffset = c.regoff(&p.To)
+
+		if ar != 0 {
+			s.used.cc |= E_MEMSP
+		} else {
+			s.set.cc |= E_MEMSP
+		}
+
+	case C_SEXT,
+		C_LEXT:
+		s.used.ireg |= 1 << (REGSB - REG_R0)
+		if ad != 0 {
+			break
+		}
+		s.size = uint8(sz)
+		s.soffset = c.regoff(&p.To)
+
+		if ar != 0 {
+			s.used.cc |= E_MEMSB
+		} else {
+			s.set.cc |= E_MEMSB
+		}
+	}
+
+	/*
+	 * flags based on 'from' field
+	 */
+	cls = int(p.From.Class)
+	if cls == 0 {
+		cls = c.aclass(&p.From) + 1
+		p.From.Class = int8(cls)
+	}
+	cls--
+	switch cls {
+	default:
+		fmt.Printf("unknown class %d %v\n", cls, p)
+
+	case C_ZCON,
+		C_SCON,
+		C_ADD0CON,
+		C_AND0CON,
+		C_ADDCON,
+		C_ANDCON,
+		C_UCON,
+		C_LCON,
+		C_NONE,
+		C_SBRA,
+		C_LBRA,
+		C_ADDR,
+		C_TEXTSIZE:
+		break
+
+	case C_HI,
+		C_LO:
+		s.used.cc |= E_HILO
+
+	case C_FCREG:
+		s.used.cc |= E_FCR
+
+	case C_MREG:
+		s.used.cc |= E_MCR
+
+	case C_ZOREG,
+		C_SOREG,
+		C_LOREG:
+		cls = int(p.From.Reg)
+		s.used.ireg |= 1 << uint(cls-REG_R0)
+		if ld != 0 {
+			p.Mark |= LOAD
+		}
+		s.size = uint8(sz)
+		s.soffset = c.regoff(&p.From)
+
+		m := uint32(ANYMEM)
+		if cls == REGSB {
+			m = E_MEMSB
+		}
+		if cls == REGSP {
+			m = E_MEMSP
+		}
+
+		s.used.cc |= m
+
+	case C_SACON,
+		C_LACON:
+		cls = int(p.From.Reg)
+		if cls == 0 {
+			cls = REGSP
+		}
+		s.used.ireg |= 1 << uint(cls-REG_R0)
+
+	case C_SECON,
+		C_LECON:
+		s.used.ireg |= 1 << (REGSB - REG_R0)
+
+	case C_REG:
+		s.used.ireg |= 1 << uint(p.From.Reg-REG_R0)
+
+	case C_FREG:
+		s.used.freg |= 1 << uint(p.From.Reg-REG_F0)
+		if ld != 0 && p.To.Type == obj.TYPE_REG {
+			p.Mark |= LOAD
+		}
+
+	case C_SAUTO,
+		C_LAUTO:
+		s.used.ireg |= 1 << (REGSP - REG_R0)
+		if ld != 0 {
+			p.Mark |= LOAD
+		}
+		if ad != 0 {
+			break
+		}
+		s.size = uint8(sz)
+		s.soffset = c.regoff(&p.From)
+
+		s.used.cc |= E_MEMSP
+
+	case C_SEXT:
+	case C_LEXT:
+		s.used.ireg |= 1 << (REGSB - REG_R0)
+		if ld != 0 {
+			p.Mark |= LOAD
+		}
+		if ad != 0 {
+			break
+		}
+		s.size = uint8(sz)
+		s.soffset = c.regoff(&p.From)
+
+		s.used.cc |= E_MEMSB
+	}
+
+	cls = int(p.Reg)
+	if cls != 0 {
+		if REG_F0 <= cls && cls <= REG_F31 {
+			s.used.freg |= 1 << uint(cls-REG_F0)
+		} else {
+			s.used.ireg |= 1 << uint(cls-REG_R0)
+		}
+	}
+	s.set.ireg &^= (1 << (REGZERO - REG_R0)) /* R0 can't be set */
+}
+
+/*
+ * test to see if two instructions can be
+ * interchanged without changing semantics
+ */
+func (c *ctxt0) depend(sa, sb *Sch) bool {
+	if sa.set.ireg&(sb.set.ireg|sb.used.ireg) != 0 {
+		return true
+	}
+	if sb.set.ireg&sa.used.ireg != 0 {
+		return true
+	}
+
+	if sa.set.freg&(sb.set.freg|sb.used.freg) != 0 {
+		return true
+	}
+	if sb.set.freg&sa.used.freg != 0 {
+		return true
+	}
+
+	/*
+	 * special case.
+	 * loads from same address cannot pass.
+	 * this is for hardware fifo's and the like
+	 */
+	if sa.used.cc&sb.used.cc&E_MEM != 0 {
+		if sa.p.Reg == sb.p.Reg {
+			if c.regoff(&sa.p.From) == c.regoff(&sb.p.From) {
+				return true
+			}
+		}
+	}
+
+	x := (sa.set.cc & (sb.set.cc | sb.used.cc)) | (sb.set.cc & sa.used.cc)
+	if x != 0 {
+		/*
+		 * allow SB and SP to pass each other.
+		 * allow SB to pass SB iff doffsets are ok
+		 * anything else conflicts
+		 */
+		if x != E_MEMSP && x != E_MEMSB {
+			return true
+		}
+		x = sa.set.cc | sb.set.cc | sa.used.cc | sb.used.cc
+		if x&E_MEM != 0 {
+			return true
+		}
+		if offoverlap(sa, sb) {
+			return true
+		}
+	}
+
+	return false
+}
+
+func offoverlap(sa, sb *Sch) bool {
+	if sa.soffset < sb.soffset {
+		if sa.soffset+int32(sa.size) > sb.soffset {
+			return true
+		}
+		return false
+	}
+	if sb.soffset+int32(sb.size) > sa.soffset {
+		return true
+	}
+	return false
+}
+
+/*
+ * test 2 adjacent instructions
+ * and find out if inserted instructions
+ * are desired to prevent stalls.
+ */
+func conflict(sa, sb *Sch) bool {
+	if sa.set.ireg&sb.used.ireg != 0 {
+		return true
+	}
+	if sa.set.freg&sb.used.freg != 0 {
+		return true
+	}
+	if sa.set.cc&sb.used.cc != 0 {
+		return true
+	}
+	return false
+}
+
+func (c *ctxt0) compound(p *obj.Prog) bool {
+	o := c.oplook(p)
+	if o.size != 4 {
+		return true
+	}
+	if p.To.Type == obj.TYPE_REG && p.To.Reg == REGSB {
+		return true
+	}
+	return false
+}
+
+var Linkmips64 = obj.LinkArch{
+	Arch:           sys.ArchMIPS64,
+	Init:           buildop,
+	Preprocess:     preprocess,
+	Assemble:       span0,
+	Progedit:       progedit,
+	DWARFRegisters: MIPSDWARFRegisters,
+}
+
+var Linkmips64le = obj.LinkArch{
+	Arch:           sys.ArchMIPS64LE,
+	Init:           buildop,
+	Preprocess:     preprocess,
+	Assemble:       span0,
+	Progedit:       progedit,
+	DWARFRegisters: MIPSDWARFRegisters,
+}
+
+var Linkmips = obj.LinkArch{
+	Arch:           sys.ArchMIPS,
+	Init:           buildop,
+	Preprocess:     preprocess,
+	Assemble:       span0,
+	Progedit:       progedit,
+	DWARFRegisters: MIPSDWARFRegisters,
+}
+
+var Linkmipsle = obj.LinkArch{
+	Arch:           sys.ArchMIPSLE,
+	Init:           buildop,
+	Preprocess:     preprocess,
+	Assemble:       span0,
+	Progedit:       progedit,
+	DWARFRegisters: MIPSDWARFRegisters,
+}