// 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 ppc64 import ( "cmd/internal/obj" "cmd/internal/objabi" "encoding/binary" "fmt" "internal/buildcfg" "log" "math" "math/bits" "sort" ) // ctxt9 holds state while assembling a single function. // Each function gets a fresh ctxt9. // This allows for multiple functions to be safely concurrently assembled. type ctxt9 struct { ctxt *obj.Link newprog obj.ProgAlloc cursym *obj.LSym autosize int32 instoffset int64 pc int64 } // Instruction layout. const ( r0iszero = 1 ) const ( // R bit option in prefixed load/store/add D-form operations PFX_R_ABS = 0 // Offset is absolute PFX_R_PCREL = 1 // Offset is relative to PC, RA should be 0 ) const ( // The preferred hardware nop instruction. NOP = 0x60000000 ) type Optab struct { as obj.As // Opcode a1 uint8 // p.From argument (obj.Addr). p is of type obj.Prog. a2 uint8 // p.Reg argument (int16 Register) a3 uint8 // p.RestArgs[0] (obj.AddrPos) a4 uint8 // p.RestArgs[1] a5 uint8 // p.RestARgs[2] a6 uint8 // p.To (obj.Addr) type_ int8 // cases in asmout below. E.g., 44 = st r,(ra+rb); 45 = ld (ra+rb), r size int8 // Text space in bytes to lay operation // A prefixed instruction is generated by this opcode. This cannot be placed // across a 64B PC address. Opcodes should not translate to more than one // prefixed instruction. The prefixed instruction should be written first // (e.g when Optab.size > 8). ispfx bool asmout func(*ctxt9, *obj.Prog, *Optab, *[5]uint32) } // optab contains an array to be sliced of accepted operand combinations for an // instruction. Unused arguments and fields are not explicitly enumerated, and // should not be listed for clarity. Unused arguments and values should always // assume the default value for the given type. // // optab does not list every valid ppc64 opcode, it enumerates representative // operand combinations for a class of instruction. The variable oprange indexes // all valid ppc64 opcodes. // // oprange is initialized to point a slice within optab which contains the valid // operand combinations for a given instruction. This is initialized from buildop. // // Likewise, each slice of optab is dynamically sorted using the ocmp Sort interface // to arrange entries to minimize text size of each opcode. // // optab is the sorted result of combining optabBase, optabGen, and prefixableOptab. var optab []Optab var optabBase = []Optab{ {as: obj.ATEXT, a1: C_LOREG, a6: C_TEXTSIZE, type_: 0, size: 0}, {as: obj.ATEXT, a1: C_LOREG, a3: C_32CON, a6: C_TEXTSIZE, type_: 0, size: 0}, {as: obj.ATEXT, a1: C_ADDR, a6: C_TEXTSIZE, type_: 0, size: 0}, {as: obj.ATEXT, a1: C_ADDR, a3: C_32CON, a6: C_TEXTSIZE, type_: 0, size: 0}, /* move register */ {as: AADD, a1: C_REG, a2: C_REG, a6: C_REG, type_: 2, size: 4}, {as: AADD, a1: C_REG, a6: C_REG, type_: 2, size: 4}, {as: AADD, a1: C_S16CON, a2: C_REG, a6: C_REG, type_: 4, size: 4}, {as: AADD, a1: C_S16CON, a6: C_REG, type_: 4, size: 4}, {as: AADD, a1: C_U16CON, a2: C_REG, a6: C_REG, type_: 22, size: 8}, {as: AADD, a1: C_U16CON, a6: C_REG, type_: 22, size: 8}, {as: AADDIS, a1: C_S16CON, a2: C_REG, a6: C_REG, type_: 20, size: 4}, {as: AADDIS, a1: C_S16CON, a6: C_REG, type_: 20, size: 4}, {as: AADDC, a1: C_REG, a2: C_REG, a6: C_REG, type_: 2, size: 4}, {as: AADDC, a1: C_REG, a6: C_REG, type_: 2, size: 4}, {as: AADDC, a1: C_S16CON, a2: C_REG, a6: C_REG, type_: 4, size: 4}, {as: AADDC, a1: C_S16CON, a6: C_REG, type_: 4, size: 4}, {as: AADDC, a1: C_32CON, a2: C_REG, a6: C_REG, type_: 22, size: 12}, {as: AADDC, a1: C_32CON, a6: C_REG, type_: 22, size: 12}, {as: AAND, a1: C_REG, a2: C_REG, a6: C_REG, type_: 6, size: 4}, /* logical, no literal */ {as: AAND, a1: C_REG, a6: C_REG, type_: 6, size: 4}, {as: AANDCC, a1: C_REG, a2: C_REG, a6: C_REG, type_: 6, size: 4}, {as: AANDCC, a1: C_REG, a6: C_REG, type_: 6, size: 4}, {as: AANDCC, a1: C_U16CON, a6: C_REG, type_: 58, size: 4}, {as: AANDCC, a1: C_U16CON, a2: C_REG, a6: C_REG, type_: 58, size: 4}, {as: AANDCC, a1: C_S16CON, a6: C_REG, type_: 23, size: 8}, {as: AANDCC, a1: C_S16CON, a2: C_REG, a6: C_REG, type_: 23, size: 8}, {as: AANDCC, a1: C_32CON, a6: C_REG, type_: 23, size: 12}, {as: AANDCC, a1: C_32CON, a2: C_REG, a6: C_REG, type_: 23, size: 12}, {as: AANDISCC, a1: C_U16CON, a6: C_REG, type_: 58, size: 4}, {as: AANDISCC, a1: C_U16CON, a2: C_REG, a6: C_REG, type_: 58, size: 4}, {as: AMULLW, a1: C_REG, a2: C_REG, a6: C_REG, type_: 2, size: 4}, {as: AMULLW, a1: C_REG, a6: C_REG, type_: 2, size: 4}, {as: AMULLW, a1: C_S16CON, a2: C_REG, a6: C_REG, type_: 4, size: 4}, {as: AMULLW, a1: C_S16CON, a6: C_REG, type_: 4, size: 4}, {as: AMULLW, a1: C_32CON, a2: C_REG, a6: C_REG, type_: 22, size: 12}, {as: AMULLW, a1: C_32CON, a6: C_REG, type_: 22, size: 12}, {as: ASUBC, a1: C_REG, a2: C_REG, a6: C_REG, type_: 10, size: 4}, {as: ASUBC, a1: C_REG, a6: C_REG, type_: 10, size: 4}, {as: ASUBC, a1: C_REG, a3: C_S16CON, a6: C_REG, type_: 27, size: 4}, {as: ASUBC, a1: C_REG, a3: C_32CON, a6: C_REG, type_: 28, size: 12}, {as: AOR, a1: C_REG, a2: C_REG, a6: C_REG, type_: 6, size: 4}, /* logical, literal not cc (or/xor) */ {as: AOR, a1: C_REG, a6: C_REG, type_: 6, size: 4}, {as: AOR, a1: C_U16CON, a6: C_REG, type_: 58, size: 4}, {as: AOR, a1: C_U16CON, a2: C_REG, a6: C_REG, type_: 58, size: 4}, {as: AOR, a1: C_S16CON, a6: C_REG, type_: 23, size: 8}, {as: AOR, a1: C_S16CON, a2: C_REG, a6: C_REG, type_: 23, size: 8}, {as: AOR, a1: C_U32CON, a2: C_REG, a6: C_REG, type_: 21, size: 8}, {as: AOR, a1: C_U32CON, a6: C_REG, type_: 21, size: 8}, {as: AOR, a1: C_32CON, a6: C_REG, type_: 23, size: 12}, {as: AOR, a1: C_32CON, a2: C_REG, a6: C_REG, type_: 23, size: 12}, {as: AORIS, a1: C_U16CON, a6: C_REG, type_: 58, size: 4}, {as: AORIS, a1: C_U16CON, a2: C_REG, a6: C_REG, type_: 58, size: 4}, {as: ADIVW, a1: C_REG, a2: C_REG, a6: C_REG, type_: 2, size: 4}, /* op r1[,r2],r3 */ {as: ADIVW, a1: C_REG, a6: C_REG, type_: 2, size: 4}, {as: ASUB, a1: C_REG, a2: C_REG, a6: C_REG, type_: 10, size: 4}, /* op r2[,r1],r3 */ {as: ASUB, a1: C_REG, a6: C_REG, type_: 10, size: 4}, {as: ASLW, a1: C_REG, a6: C_REG, type_: 6, size: 4}, {as: ASLW, a1: C_REG, a2: C_REG, a6: C_REG, type_: 6, size: 4}, {as: ASLD, a1: C_REG, a6: C_REG, type_: 6, size: 4}, {as: ASLD, a1: C_REG, a2: C_REG, a6: C_REG, type_: 6, size: 4}, {as: ASLD, a1: C_U15CON, a2: C_REG, a6: C_REG, type_: 25, size: 4}, {as: ASLD, a1: C_U15CON, a6: C_REG, type_: 25, size: 4}, {as: AEXTSWSLI, a1: C_U15CON, a6: C_REG, type_: 25, size: 4}, {as: AEXTSWSLI, a1: C_U15CON, a2: C_REG, a6: C_REG, type_: 25, size: 4}, {as: ASLW, a1: C_U15CON, a2: C_REG, a6: C_REG, type_: 57, size: 4}, {as: ASLW, a1: C_U15CON, a6: C_REG, type_: 57, size: 4}, {as: ASRAW, a1: C_REG, a6: C_REG, type_: 6, size: 4}, {as: ASRAW, a1: C_REG, a2: C_REG, a6: C_REG, type_: 6, size: 4}, {as: ASRAW, a1: C_U15CON, a2: C_REG, a6: C_REG, type_: 56, size: 4}, {as: ASRAW, a1: C_U15CON, a6: C_REG, type_: 56, size: 4}, {as: ASRAD, a1: C_REG, a6: C_REG, type_: 6, size: 4}, {as: ASRAD, a1: C_REG, a2: C_REG, a6: C_REG, type_: 6, size: 4}, {as: ASRAD, a1: C_U15CON, a2: C_REG, a6: C_REG, type_: 56, size: 4}, {as: ASRAD, a1: C_U15CON, a6: C_REG, type_: 56, size: 4}, {as: ARLWNM, a1: C_U15CON, a2: C_REG, a3: C_32CON, a6: C_REG, type_: 63, size: 4}, {as: ARLWNM, a1: C_U15CON, a2: C_REG, a3: C_U15CON, a4: C_U15CON, a6: C_REG, type_: 63, size: 4}, {as: ARLWNM, a1: C_REG, a2: C_REG, a3: C_32CON, a6: C_REG, type_: 63, size: 4}, {as: ARLWNM, a1: C_REG, a2: C_REG, a3: C_U15CON, a4: C_U15CON, a6: C_REG, type_: 63, size: 4}, {as: ACLRLSLWI, a1: C_U15CON, a2: C_REG, a3: C_32CON, a6: C_REG, type_: 62, size: 4}, {as: ARLDMI, a1: C_U15CON, a2: C_REG, a3: C_32CON, a6: C_REG, type_: 30, size: 4}, {as: ARLDC, a1: C_U15CON, a2: C_REG, a3: C_32CON, a6: C_REG, type_: 29, size: 4}, {as: ARLDC, a1: C_REG, a3: C_U8CON, a4: C_U8CON, a6: C_REG, type_: 9, size: 4}, {as: ARLDCL, a1: C_U15CON, a2: C_REG, a3: C_32CON, a6: C_REG, type_: 29, size: 4}, {as: ARLDCL, a1: C_REG, a2: C_REG, a3: C_32CON, a6: C_REG, type_: 14, size: 4}, {as: ARLDICL, a1: C_REG, a2: C_REG, a3: C_32CON, a6: C_REG, type_: 14, size: 4}, {as: ARLDICL, a1: C_U15CON, a2: C_REG, a3: C_32CON, a6: C_REG, type_: 14, size: 4}, {as: ARLDCL, a1: C_REG, a3: C_32CON, a6: C_REG, type_: 14, size: 4}, {as: AFADD, a1: C_FREG, a6: C_FREG, type_: 2, size: 4}, {as: AFADD, a1: C_FREG, a2: C_FREG, a6: C_FREG, type_: 2, size: 4}, {as: AFABS, a1: C_FREG, a6: C_FREG, type_: 33, size: 4}, {as: AFABS, a6: C_FREG, type_: 33, size: 4}, {as: AFMADD, a1: C_FREG, a2: C_FREG, a3: C_FREG, a6: C_FREG, type_: 34, size: 4}, {as: AFMUL, a1: C_FREG, a6: C_FREG, type_: 32, size: 4}, {as: AFMUL, a1: C_FREG, a2: C_FREG, a6: C_FREG, type_: 32, size: 4}, {as: AMOVBU, a1: C_REG, a6: C_SOREG, type_: 7, size: 4}, {as: AMOVBU, a1: C_REG, a6: C_XOREG, type_: 108, size: 4}, {as: AMOVBU, a1: C_SOREG, a6: C_REG, type_: 8, size: 8}, {as: AMOVBU, a1: C_XOREG, a6: C_REG, type_: 109, size: 8}, {as: AMOVBZU, a1: C_REG, a6: C_SOREG, type_: 7, size: 4}, {as: AMOVBZU, a1: C_REG, a6: C_XOREG, type_: 108, size: 4}, {as: AMOVBZU, a1: C_SOREG, a6: C_REG, type_: 8, size: 4}, {as: AMOVBZU, a1: C_XOREG, a6: C_REG, type_: 109, size: 4}, {as: AMOVHBR, a1: C_REG, a6: C_XOREG, type_: 44, size: 4}, {as: AMOVHBR, a1: C_XOREG, a6: C_REG, type_: 45, size: 4}, {as: AMOVB, a1: C_SOREG, a6: C_REG, type_: 8, size: 8}, {as: AMOVB, a1: C_XOREG, a6: C_REG, type_: 109, size: 8}, {as: AMOVB, a1: C_REG, a6: C_SOREG, type_: 7, size: 4}, {as: AMOVB, a1: C_REG, a6: C_XOREG, type_: 108, size: 4}, {as: AMOVB, a1: C_REG, a6: C_REG, type_: 13, size: 4}, {as: AMOVBZ, a1: C_SOREG, a6: C_REG, type_: 8, size: 4}, {as: AMOVBZ, a1: C_XOREG, a6: C_REG, type_: 109, size: 4}, {as: AMOVBZ, a1: C_REG, a6: C_SOREG, type_: 7, size: 4}, {as: AMOVBZ, a1: C_REG, a6: C_XOREG, type_: 108, size: 4}, {as: AMOVBZ, a1: C_REG, a6: C_REG, type_: 13, size: 4}, {as: AMOVD, a1: C_16CON, a6: C_REG, type_: 3, size: 4}, {as: AMOVD, a1: C_SACON, a6: C_REG, type_: 3, size: 4}, {as: AMOVD, a1: C_SOREG, a6: C_REG, type_: 8, size: 4}, {as: AMOVD, a1: C_XOREG, a6: C_REG, type_: 109, size: 4}, {as: AMOVD, a1: C_SOREG, a6: C_SPR, type_: 107, size: 8}, {as: AMOVD, a1: C_SPR, a6: C_REG, type_: 66, size: 4}, {as: AMOVD, a1: C_REG, a6: C_SOREG, type_: 7, size: 4}, {as: AMOVD, a1: C_REG, a6: C_XOREG, type_: 108, size: 4}, {as: AMOVD, a1: C_SPR, a6: C_SOREG, type_: 106, size: 8}, {as: AMOVD, a1: C_REG, a6: C_SPR, type_: 66, size: 4}, {as: AMOVD, a1: C_REG, a6: C_REG, type_: 13, size: 4}, {as: AMOVW, a1: C_16CON, a6: C_REG, type_: 3, size: 4}, {as: AMOVW, a1: C_SACON, a6: C_REG, type_: 3, size: 4}, {as: AMOVW, a1: C_CREG, a6: C_REG, type_: 68, size: 4}, {as: AMOVW, a1: C_SOREG, a6: C_REG, type_: 8, size: 4}, {as: AMOVW, a1: C_XOREG, a6: C_REG, type_: 109, size: 4}, {as: AMOVW, a1: C_SPR, a6: C_REG, type_: 66, size: 4}, {as: AMOVW, a1: C_REG, a6: C_CREG, type_: 69, size: 4}, {as: AMOVW, a1: C_REG, a6: C_SOREG, type_: 7, size: 4}, {as: AMOVW, a1: C_REG, a6: C_XOREG, type_: 108, size: 4}, {as: AMOVW, a1: C_REG, a6: C_SPR, type_: 66, size: 4}, {as: AMOVW, a1: C_REG, a6: C_REG, type_: 13, size: 4}, {as: AFMOVD, a1: C_S16CON, a6: C_FREG, type_: 24, size: 8}, {as: AFMOVD, a1: C_SOREG, a6: C_FREG, type_: 8, size: 4}, {as: AFMOVD, a1: C_XOREG, a6: C_FREG, type_: 109, size: 4}, {as: AFMOVD, a1: C_ZCON, a6: C_FREG, type_: 24, size: 4}, {as: AFMOVD, a1: C_FREG, a6: C_FREG, type_: 33, size: 4}, {as: AFMOVD, a1: C_FREG, a6: C_SOREG, type_: 7, size: 4}, {as: AFMOVD, a1: C_FREG, a6: C_XOREG, type_: 108, size: 4}, {as: AFMOVSX, a1: C_XOREG, a6: C_FREG, type_: 45, size: 4}, {as: AFMOVSX, a1: C_FREG, a6: C_XOREG, type_: 44, size: 4}, {as: AFMOVSZ, a1: C_ZOREG, a6: C_FREG, type_: 45, size: 4}, {as: AFMOVSZ, a1: C_XOREG, a6: C_FREG, type_: 45, size: 4}, {as: AMOVFL, a1: C_CREG, a6: C_CREG, type_: 67, size: 4}, {as: AMOVFL, a1: C_FPSCR, a6: C_CREG, type_: 73, size: 4}, {as: AMOVFL, a1: C_FPSCR, a6: C_FREG, type_: 53, size: 4}, {as: AMOVFL, a1: C_FREG, a3: C_32CON, a6: C_FPSCR, type_: 64, size: 4}, {as: AMOVFL, a1: C_FREG, a6: C_FPSCR, type_: 64, size: 4}, {as: AMOVFL, a1: C_32CON, a6: C_FPSCR, type_: 65, size: 4}, {as: AMOVFL, a1: C_REG, a6: C_CREG, type_: 69, size: 4}, {as: AMOVFL, a1: C_REG, a6: C_32CON, type_: 69, size: 4}, {as: ASYSCALL, type_: 5, size: 4}, {as: ASYSCALL, a1: C_REG, type_: 77, size: 12}, {as: ASYSCALL, a1: C_U15CON, type_: 77, size: 12}, {as: ABEQ, a6: C_BRA, type_: 16, size: 4}, {as: ABEQ, a1: C_CREG, a6: C_BRA, type_: 16, size: 4}, {as: ABR, a6: C_BRA, type_: 11, size: 4}, // b label {as: ABR, a6: C_BRAPIC, type_: 11, size: 8}, // b label; nop {as: ABR, a6: C_LR, type_: 18, size: 4}, // blr {as: ABR, a6: C_CTR, type_: 18, size: 4}, // bctr {as: ABC, a1: C_U15CON, a2: C_CRBIT, a6: C_BRA, type_: 16, size: 4}, // bc bo, bi, label {as: ABC, a1: C_U15CON, a2: C_CRBIT, a6: C_LR, type_: 18, size: 4}, // bclr bo, bi {as: ABC, a1: C_U15CON, a2: C_CRBIT, a3: C_U15CON, a6: C_LR, type_: 18, size: 4}, // bclr bo, bi, bh {as: ABC, a1: C_U15CON, a2: C_CRBIT, a6: C_CTR, type_: 18, size: 4}, // bcctr bo, bi {as: ABDNZ, a6: C_BRA, type_: 16, size: 4}, {as: ASYNC, type_: 46, size: 4}, {as: AWORD, a1: C_32CON, type_: 40, size: 4}, {as: ADWORD, a1: C_64CON, type_: 31, size: 8}, {as: ADWORD, a1: C_LACON, type_: 31, size: 8}, {as: AADDME, a1: C_REG, a6: C_REG, type_: 47, size: 4}, {as: AEXTSB, a1: C_REG, a6: C_REG, type_: 48, size: 4}, {as: AEXTSB, a6: C_REG, type_: 48, size: 4}, {as: AISEL, a1: C_U5CON, a2: C_REG, a3: C_REG, a6: C_REG, type_: 84, size: 4}, {as: AISEL, a1: C_CRBIT, a2: C_REG, a3: C_REG, a6: C_REG, type_: 84, size: 4}, {as: ANEG, a1: C_REG, a6: C_REG, type_: 47, size: 4}, {as: ANEG, a6: C_REG, type_: 47, size: 4}, {as: AREM, a1: C_REG, a6: C_REG, type_: 50, size: 12}, {as: AREM, a1: C_REG, a2: C_REG, a6: C_REG, type_: 50, size: 12}, {as: AREMU, a1: C_REG, a6: C_REG, type_: 50, size: 16}, {as: AREMU, a1: C_REG, a2: C_REG, a6: C_REG, type_: 50, size: 16}, {as: AREMD, a1: C_REG, a6: C_REG, type_: 51, size: 12}, {as: AREMD, a1: C_REG, a2: C_REG, a6: C_REG, type_: 51, size: 12}, {as: AMTFSB0, a1: C_U15CON, type_: 52, size: 4}, /* Other ISA 2.05+ instructions */ {as: APOPCNTD, a1: C_REG, a6: C_REG, type_: 93, size: 4}, /* population count, x-form */ {as: ACMPB, a1: C_REG, a2: C_REG, a6: C_REG, type_: 92, size: 4}, /* compare byte, x-form */ {as: ACMPEQB, a1: C_REG, a2: C_REG, a6: C_CREG, type_: 92, size: 4}, /* compare equal byte, x-form, ISA 3.0 */ {as: ACMPEQB, a1: C_REG, a6: C_REG, type_: 70, size: 4}, {as: AFTDIV, a1: C_FREG, a2: C_FREG, a6: C_U15CON, type_: 92, size: 4}, /* floating test for sw divide, x-form */ {as: AFTSQRT, a1: C_FREG, a6: C_U15CON, type_: 93, size: 4}, /* floating test for sw square root, x-form */ {as: ACOPY, a1: C_REG, a6: C_REG, type_: 92, size: 4}, /* copy/paste facility, x-form */ {as: ADARN, a1: C_U15CON, a6: C_REG, type_: 92, size: 4}, /* deliver random number, x-form */ {as: AMADDHD, a1: C_REG, a2: C_REG, a3: C_REG, a6: C_REG, type_: 83, size: 4}, /* multiply-add high/low doubleword, va-form */ {as: AADDEX, a1: C_REG, a2: C_REG, a3: C_U15CON, a6: C_REG, type_: 94, size: 4}, /* add extended using alternate carry, z23-form */ {as: ACRAND, a1: C_CRBIT, a2: C_CRBIT, a6: C_CRBIT, type_: 2, size: 4}, /* logical ops for condition register bits xl-form */ /* Misc ISA 3.0 instructions */ {as: ASETB, a1: C_CREG, a6: C_REG, type_: 110, size: 4}, {as: AVCLZLSBB, a1: C_VREG, a6: C_REG, type_: 85, size: 4}, /* Vector instructions */ /* Vector load */ {as: ALVEBX, a1: C_XOREG, a6: C_VREG, type_: 45, size: 4}, /* vector load, x-form */ /* Vector store */ {as: ASTVEBX, a1: C_VREG, a6: C_XOREG, type_: 44, size: 4}, /* vector store, x-form */ /* Vector logical */ {as: AVAND, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector and, vx-form */ {as: AVOR, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector or, vx-form */ /* Vector add */ {as: AVADDUM, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector add unsigned modulo, vx-form */ {as: AVADDCU, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector add & write carry unsigned, vx-form */ {as: AVADDUS, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector add unsigned saturate, vx-form */ {as: AVADDSS, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector add signed saturate, vx-form */ {as: AVADDE, a1: C_VREG, a2: C_VREG, a3: C_VREG, a6: C_VREG, type_: 83, size: 4}, /* vector add extended, va-form */ /* Vector subtract */ {as: AVSUBUM, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector subtract unsigned modulo, vx-form */ {as: AVSUBCU, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector subtract & write carry unsigned, vx-form */ {as: AVSUBUS, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector subtract unsigned saturate, vx-form */ {as: AVSUBSS, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector subtract signed saturate, vx-form */ {as: AVSUBE, a1: C_VREG, a2: C_VREG, a3: C_VREG, a6: C_VREG, type_: 83, size: 4}, /* vector subtract extended, va-form */ /* Vector multiply */ {as: AVMULESB, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector multiply, vx-form */ {as: AVPMSUM, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector polynomial multiply & sum, vx-form */ {as: AVMSUMUDM, a1: C_VREG, a2: C_VREG, a3: C_VREG, a6: C_VREG, type_: 83, size: 4}, /* vector multiply-sum, va-form */ /* Vector rotate */ {as: AVR, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector rotate, vx-form */ /* Vector shift */ {as: AVS, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector shift, vx-form */ {as: AVSA, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector shift algebraic, vx-form */ {as: AVSOI, a1: C_U16CON, a2: C_VREG, a3: C_VREG, a6: C_VREG, type_: 83, size: 4}, /* vector shift by octet immediate, va-form */ /* Vector count */ {as: AVCLZ, a1: C_VREG, a6: C_VREG, type_: 85, size: 4}, /* vector count leading zeros, vx-form */ {as: AVPOPCNT, a1: C_VREG, a6: C_VREG, type_: 85, size: 4}, /* vector population count, vx-form */ /* Vector compare */ {as: AVCMPEQ, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector compare equal, vc-form */ {as: AVCMPGT, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector compare greater than, vc-form */ {as: AVCMPNEZB, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector compare not equal, vx-form */ /* Vector merge */ {as: AVMRGOW, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector merge odd word, vx-form */ /* Vector permute */ {as: AVPERM, a1: C_VREG, a2: C_VREG, a3: C_VREG, a6: C_VREG, type_: 83, size: 4}, /* vector permute, va-form */ /* Vector bit permute */ {as: AVBPERMQ, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector bit permute, vx-form */ /* Vector select */ {as: AVSEL, a1: C_VREG, a2: C_VREG, a3: C_VREG, a6: C_VREG, type_: 83, size: 4}, /* vector select, va-form */ /* Vector splat */ {as: AVSPLTB, a1: C_S16CON, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, {as: AVSPLTISB, a1: C_S16CON, a6: C_VREG, type_: 82, size: 4}, /* Vector AES */ {as: AVCIPH, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector AES cipher, vx-form */ {as: AVNCIPH, a1: C_VREG, a2: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector AES inverse cipher, vx-form */ {as: AVSBOX, a1: C_VREG, a6: C_VREG, type_: 82, size: 4}, /* vector AES subbytes, vx-form */ /* Vector SHA */ {as: AVSHASIGMA, a1: C_U16CON, a2: C_VREG, a3: C_U16CON, a6: C_VREG, type_: 82, size: 4}, /* vector SHA sigma, vx-form */ /* VSX vector load */ {as: ALXVD2X, a1: C_XOREG, a6: C_VSREG, type_: 87, size: 4}, /* vsx vector load, xx1-form */ {as: ALXV, a1: C_SOREG, a6: C_VSREG, type_: 96, size: 4}, /* vsx vector load, dq-form */ {as: ALXVL, a1: C_REG, a2: C_REG, a6: C_VSREG, type_: 98, size: 4}, /* vsx vector load length */ /* VSX vector store */ {as: ASTXVD2X, a1: C_VSREG, a6: C_XOREG, type_: 86, size: 4}, /* vsx vector store, xx1-form */ {as: ASTXV, a1: C_VSREG, a6: C_SOREG, type_: 97, size: 4}, /* vsx vector store, dq-form */ {as: ASTXVL, a1: C_VSREG, a2: C_REG, a6: C_REG, type_: 99, size: 4}, /* vsx vector store with length x-form */ /* VSX scalar load */ {as: ALXSDX, a1: C_XOREG, a6: C_VSREG, type_: 87, size: 4}, /* vsx scalar load, xx1-form */ /* VSX scalar store */ {as: ASTXSDX, a1: C_VSREG, a6: C_XOREG, type_: 86, size: 4}, /* vsx scalar store, xx1-form */ /* VSX scalar as integer load */ {as: ALXSIWAX, a1: C_XOREG, a6: C_VSREG, type_: 87, size: 4}, /* vsx scalar as integer load, xx1-form */ /* VSX scalar store as integer */ {as: ASTXSIWX, a1: C_VSREG, a6: C_XOREG, type_: 86, size: 4}, /* vsx scalar as integer store, xx1-form */ /* VSX move from VSR */ {as: AMFVSRD, a1: C_VSREG, a6: C_REG, type_: 88, size: 4}, {as: AMFVSRD, a1: C_FREG, a6: C_REG, type_: 88, size: 4}, /* VSX move to VSR */ {as: AMTVSRD, a1: C_REG, a6: C_VSREG, type_: 104, size: 4}, {as: AMTVSRD, a1: C_REG, a6: C_FREG, type_: 104, size: 4}, {as: AMTVSRDD, a1: C_REG, a2: C_REG, a6: C_VSREG, type_: 104, size: 4}, /* VSX xx3-form */ {as: AXXLAND, a1: C_FREG, a2: C_FREG, a6: C_FREG, type_: 90, size: 4}, /* vsx xx3-form (FPR usage) */ {as: AXXLAND, a1: C_VSREG, a2: C_VSREG, a6: C_VSREG, type_: 90, size: 4}, /* vsx xx3-form */ /* VSX select */ {as: AXXSEL, a1: C_VSREG, a2: C_VSREG, a3: C_VSREG, a6: C_VSREG, type_: 91, size: 4}, /* vsx select, xx4-form */ /* VSX merge */ {as: AXXMRGHW, a1: C_VSREG, a2: C_VSREG, a6: C_VSREG, type_: 90, size: 4}, /* vsx merge, xx3-form */ /* VSX splat */ {as: AXXSPLTW, a1: C_VSREG, a3: C_U15CON, a6: C_VSREG, type_: 89, size: 4}, /* vsx splat, xx2-form */ {as: AXXSPLTIB, a1: C_U15CON, a6: C_VSREG, type_: 100, size: 4}, /* vsx splat, xx2-form */ /* VSX permute */ {as: AXXPERM, a1: C_VSREG, a2: C_VSREG, a6: C_VSREG, type_: 90, size: 4}, /* vsx permute, xx3-form */ /* VSX shift */ {as: AXXSLDWI, a1: C_VSREG, a2: C_VSREG, a3: C_U15CON, a6: C_VSREG, type_: 90, size: 4}, /* vsx shift immediate, xx3-form */ /* VSX reverse bytes */ {as: AXXBRQ, a1: C_VSREG, a6: C_VSREG, type_: 101, size: 4}, /* vsx reverse bytes */ /* VSX scalar FP-FP conversion */ {as: AXSCVDPSP, a1: C_VSREG, a6: C_VSREG, type_: 89, size: 4}, /* vsx scalar fp-fp conversion, xx2-form */ /* VSX vector FP-FP conversion */ {as: AXVCVDPSP, a1: C_VSREG, a6: C_VSREG, type_: 89, size: 4}, /* vsx vector fp-fp conversion, xx2-form */ /* VSX scalar FP-integer conversion */ {as: AXSCVDPSXDS, a1: C_VSREG, a6: C_VSREG, type_: 89, size: 4}, /* vsx scalar fp-integer conversion, xx2-form */ /* VSX scalar integer-FP conversion */ {as: AXSCVSXDDP, a1: C_VSREG, a6: C_VSREG, type_: 89, size: 4}, /* vsx scalar integer-fp conversion, xx2-form */ /* VSX vector FP-integer conversion */ {as: AXVCVDPSXDS, a1: C_VSREG, a6: C_VSREG, type_: 89, size: 4}, /* vsx vector fp-integer conversion, xx2-form */ /* VSX vector integer-FP conversion */ {as: AXVCVSXDDP, a1: C_VSREG, a6: C_VSREG, type_: 89, size: 4}, /* vsx vector integer-fp conversion, xx2-form */ {as: ACMP, a1: C_REG, a6: C_REG, type_: 70, size: 4}, {as: ACMP, a1: C_REG, a2: C_CREG, a6: C_REG, type_: 70, size: 4}, {as: ACMP, a1: C_REG, a6: C_S16CON, type_: 70, size: 4}, {as: ACMP, a1: C_REG, a2: C_CREG, a6: C_S16CON, type_: 70, size: 4}, {as: ACMPU, a1: C_REG, a6: C_REG, type_: 70, size: 4}, {as: ACMPU, a1: C_REG, a2: C_CREG, a6: C_REG, type_: 70, size: 4}, {as: ACMPU, a1: C_REG, a6: C_U16CON, type_: 70, size: 4}, {as: ACMPU, a1: C_REG, a2: C_CREG, a6: C_U16CON, type_: 70, size: 4}, {as: AFCMPO, a1: C_FREG, a6: C_FREG, type_: 70, size: 4}, {as: AFCMPO, a1: C_FREG, a2: C_CREG, a6: C_FREG, type_: 70, size: 4}, {as: ATW, a1: C_32CON, a2: C_REG, a6: C_REG, type_: 60, size: 4}, {as: ATW, a1: C_32CON, a2: C_REG, a6: C_S16CON, type_: 61, size: 4}, {as: ADCBF, a1: C_SOREG, type_: 43, size: 4}, {as: ADCBF, a1: C_XOREG, type_: 43, size: 4}, {as: ADCBF, a1: C_XOREG, a2: C_REG, a6: C_U15CON, type_: 43, size: 4}, {as: ADCBF, a1: C_SOREG, a6: C_U15CON, type_: 43, size: 4}, {as: ADCBF, a1: C_XOREG, a6: C_U15CON, type_: 43, size: 4}, {as: ASTDCCC, a1: C_REG, a2: C_REG, a6: C_XOREG, type_: 44, size: 4}, {as: ASTDCCC, a1: C_REG, a6: C_XOREG, type_: 44, size: 4}, {as: ALDAR, a1: C_XOREG, a6: C_REG, type_: 45, size: 4}, {as: ALDAR, a1: C_XOREG, a3: C_U16CON, a6: C_REG, type_: 45, size: 4}, {as: AEIEIO, type_: 46, size: 4}, {as: ATLBIE, a1: C_REG, type_: 49, size: 4}, {as: ATLBIE, a1: C_U15CON, a6: C_REG, type_: 49, size: 4}, {as: ASLBMFEE, a1: C_REG, a6: C_REG, type_: 55, size: 4}, {as: ASLBMTE, a1: C_REG, a6: C_REG, type_: 55, size: 4}, {as: ASTSW, a1: C_REG, a6: C_XOREG, type_: 44, size: 4}, {as: ASTSW, a1: C_REG, a3: C_32CON, a6: C_ZOREG, type_: 41, size: 4}, {as: ALSW, a1: C_XOREG, a6: C_REG, type_: 45, size: 4}, {as: ALSW, a1: C_ZOREG, a3: C_32CON, a6: C_REG, type_: 42, size: 4}, {as: obj.AUNDEF, type_: 78, size: 4}, {as: obj.APCDATA, a1: C_32CON, a6: C_32CON, type_: 0, size: 0}, {as: obj.AFUNCDATA, a1: C_U15CON, a6: C_ADDR, type_: 0, size: 0}, {as: obj.ANOP, type_: 0, size: 0}, {as: obj.ANOP, a1: C_32CON, type_: 0, size: 0}, // NOP operand variations added for #40689 {as: obj.ANOP, a1: C_REG, type_: 0, size: 0}, // to preserve previous behavior {as: obj.ANOP, a1: C_FREG, type_: 0, size: 0}, {as: obj.ADUFFZERO, a6: C_BRA, type_: 11, size: 4}, // same as ABR/ABL {as: obj.ADUFFCOPY, a6: C_BRA, type_: 11, size: 4}, // same as ABR/ABL {as: obj.APCALIGN, a1: C_32CON, type_: 0, size: 0}, // align code } // These are opcodes above which may generate different sequences depending on whether prefix opcode support // is available type PrefixableOptab struct { Optab minGOPPC64 int // Minimum GOPPC64 required to support this. pfxsize int8 // Instruction sequence size when prefixed opcodes are used } // The prefixable optab entry contains the pseudo-opcodes which generate relocations, or may generate // a more efficient sequence of instructions if a prefixed version exists (ex. paddi instead of oris/ori/add). // // This table is meant to transform all sequences which might be TOC-relative into an equivalent PC-relative // sequence. It also encompasses several transformations which do not involve relocations, those could be // separated and applied to AIX and other non-ELF targets. Likewise, the prefixed forms do not have encoding // restrictions on the offset, so they are also used for static binary to allow better code generation. e.x // // MOVD something-byte-aligned(Rx), Ry // MOVD 3(Rx), Ry // // is allowed when the prefixed forms are used. // // This requires an ISA 3.1 compatible cpu (e.g Power10), and when linking externally an ELFv2 1.5 compliant. var prefixableOptab = []PrefixableOptab{ {Optab: Optab{as: AMOVD, a1: C_S34CON, a6: C_REG, type_: 19, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVD, a1: C_ADDR, a6: C_REG, type_: 75, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVD, a1: C_TLS_LE, a6: C_REG, type_: 79, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVD, a1: C_TLS_IE, a6: C_REG, type_: 80, size: 12}, minGOPPC64: 10, pfxsize: 12}, {Optab: Optab{as: AMOVD, a1: C_LACON, a6: C_REG, type_: 26, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVD, a1: C_LOREG, a6: C_REG, type_: 36, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVD, a1: C_REG, a6: C_LOREG, type_: 35, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVD, a1: C_REG, a6: C_ADDR, type_: 74, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVW, a1: C_32CON, a6: C_REG, type_: 19, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVW, a1: C_LACON, a6: C_REG, type_: 26, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVW, a1: C_LOREG, a6: C_REG, type_: 36, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVW, a1: C_ADDR, a6: C_REG, type_: 75, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVW, a1: C_REG, a6: C_LOREG, type_: 35, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVW, a1: C_REG, a6: C_ADDR, type_: 74, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVB, a1: C_REG, a6: C_LOREG, type_: 35, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVB, a1: C_LOREG, a6: C_REG, type_: 36, size: 12}, minGOPPC64: 10, pfxsize: 12}, {Optab: Optab{as: AMOVB, a1: C_ADDR, a6: C_REG, type_: 75, size: 12}, minGOPPC64: 10, pfxsize: 12}, {Optab: Optab{as: AMOVB, a1: C_REG, a6: C_ADDR, type_: 74, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVBZ, a1: C_LOREG, a6: C_REG, type_: 36, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVBZ, a1: C_ADDR, a6: C_REG, type_: 75, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVBZ, a1: C_REG, a6: C_LOREG, type_: 35, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AMOVBZ, a1: C_REG, a6: C_ADDR, type_: 74, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AFMOVD, a1: C_LOREG, a6: C_FREG, type_: 36, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AFMOVD, a1: C_ADDR, a6: C_FREG, type_: 75, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AFMOVD, a1: C_FREG, a6: C_LOREG, type_: 35, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AFMOVD, a1: C_FREG, a6: C_ADDR, type_: 74, size: 8}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AADD, a1: C_32CON, a2: C_REG, a6: C_REG, type_: 22, size: 12}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AADD, a1: C_32CON, a6: C_REG, type_: 22, size: 12}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AADD, a1: C_S34CON, a2: C_REG, a6: C_REG, type_: 22, size: 20}, minGOPPC64: 10, pfxsize: 8}, {Optab: Optab{as: AADD, a1: C_S34CON, a6: C_REG, type_: 22, size: 20}, minGOPPC64: 10, pfxsize: 8}, } var oprange [ALAST & obj.AMask][]Optab var xcmp [C_NCLASS][C_NCLASS]bool var pfxEnabled = false // ISA 3.1 prefixed instructions are supported. var buildOpCfg = "" // Save the os/cpu/arch tuple used to configure the assembler in buildop // padding bytes to add to align code as requested. func addpad(pc, a int64, ctxt *obj.Link, cursym *obj.LSym) int { switch a { case 8, 16, 32, 64: // By default function alignment is 16. If an alignment > 16 is // requested then the function alignment must also be promoted. // The function alignment is not promoted on AIX at this time. // TODO: Investigate AIX function alignment. if ctxt.Headtype != objabi.Haix && cursym.Func().Align < int32(a) { cursym.Func().Align = int32(a) } if pc&(a-1) != 0 { return int(a - (pc & (a - 1))) } default: ctxt.Diag("Unexpected alignment: %d for PCALIGN directive\n", a) } return 0 } func span9(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { p := cursym.Func().Text if p == nil || p.Link == nil { // handle external functions and ELF section symbols return } if oprange[AANDN&obj.AMask] == nil { ctxt.Diag("ppc64 ops not initialized, call ppc64.buildop first") } c := ctxt9{ctxt: ctxt, newprog: newprog, cursym: cursym, autosize: int32(p.To.Offset)} 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.APCALIGN { a := c.vregoff(&p.From) m = addpad(pc, a, ctxt, cursym) } else { if p.As != obj.ANOP && p.As != obj.AFUNCDATA && p.As != obj.APCDATA { 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 var out [5]uint32 var falign int32 // Track increased alignment requirements for prefix. for bflag != 0 { bflag = 0 pc = 0 falign = 0 // Note, linker bumps function symbols to funcAlign. 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_ == 16 || o.type_ == 17) && p.To.Target() != nil { otxt = p.To.Target().Pc - pc if otxt < -(1<<15)+10 || otxt >= (1<<15)-10 { // Assemble the instruction with a target not too far to figure out BI and BO fields. // If only the CTR or BI (the CR bit) are tested, the conditional branch can be inverted, // and only one extra branch is needed to reach the target. tgt := p.To.Target() p.To.SetTarget(p.Link) o.asmout(&c, p, o, &out) p.To.SetTarget(tgt) bo := int64(out[0]>>21) & 31 bi := int16((out[0] >> 16) & 31) invertible := false if bo&0x14 == 0x14 { // A conditional branch that is unconditionally taken. This cannot be inverted. } else if bo&0x10 == 0x10 { // A branch based on the value of CTR. Invert the CTR comparison against zero bit. bo ^= 0x2 invertible = true } else if bo&0x04 == 0x04 { // A branch based on CR bit. Invert the BI comparison bit. bo ^= 0x8 invertible = true } if invertible { // Rewrite // BC bo,...,far_away_target // NEXT_INSN // to: // BC invert(bo),next_insn // JMP far_away_target // next_insn: // NEXT_INSN p.As = ABC p.From = obj.Addr{Type: obj.TYPE_CONST, Name: obj.NAME_NONE, Offset: bo} q = c.newprog() q.As = ABR q.To.Type = obj.TYPE_BRANCH q.To.SetTarget(p.To.Target()) q.Link = p.Link p.To.SetTarget(p.Link) p.Link = q p.Reg = REG_CRBIT0 + bi } else { // Rewrite // BC ...,far_away_target // NEXT_INSN // to // BC ...,tmp // JMP next_insn // tmp: // JMP far_away_target // next_insn: // NEXT_INSN q = c.newprog() q.Link = p.Link p.Link = q q.As = ABR 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 = ABR q.To.Type = obj.TYPE_BRANCH q.To.SetTarget(q.Link.Link) } bflag = 1 } } m = int(o.size) if m == 0 { if p.As == obj.APCALIGN { a := c.vregoff(&p.From) m = addpad(pc, a, ctxt, cursym) } else { if p.As != obj.ANOP && p.As != obj.AFUNCDATA && p.As != obj.APCDATA { ctxt.Diag("zero-width instruction\n%v", p) } continue } } // Prefixed instructions cannot be placed across a 64B boundary. // Mark and adjust the PC of those which do. A nop will be // inserted during final assembly. if o.ispfx { mark := p.Mark &^ PFX_X64B if pc&63 == 60 { p.Pc += 4 m += 4 mark |= PFX_X64B } // Marks may be adjusted if a too-far conditional branch is // fixed up above. Likewise, inserting a NOP may cause a // branch target to become too far away. We need to run // another iteration and verify no additional changes // are needed. if mark != p.Mark { bflag = 1 p.Mark = mark } // Check for 16 or 32B crossing of this prefixed insn. // These do no require padding, but do require increasing // the function alignment to prevent them from potentially // crossing a 64B boundary when the linker assigns the final // PC. switch p.Pc & 31 { case 28: // 32B crossing falign = 64 case 12: // 16B crossing if falign < 64 { falign = 32 } } } pc += int64(m) } c.cursym.Size = pc } c.cursym.Size = pc c.cursym.Func().Align = falign c.cursym.Grow(c.cursym.Size) // lay out the code, emitting code and data relocations. bp := c.cursym.P var i int32 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 span9 is too small, need at least %d for %v", o.size/4, p) } // asmout is not set up to add large amounts of padding if o.type_ == 0 && p.As == obj.APCALIGN { aln := c.vregoff(&p.From) v := addpad(p.Pc, aln, c.ctxt, c.cursym) if v > 0 { // Same padding instruction for all for i = 0; i < int32(v/4); i++ { c.ctxt.Arch.ByteOrder.PutUint32(bp, NOP) bp = bp[4:] } } } else { if p.Mark&PFX_X64B != 0 { c.ctxt.Arch.ByteOrder.PutUint32(bp, NOP) bp = bp[4:] } o.asmout(&c, p, o, &out) for i = 0; i < int32(o.size/4); i++ { c.ctxt.Arch.ByteOrder.PutUint32(bp, out[i]) bp = bp[4:] } } } } func isint32(v int64) bool { return int64(int32(v)) == v } func isuint32(v uint64) bool { return uint64(uint32(v)) == v } func (c *ctxt9) aclassreg(reg int16) int { if REG_R0 <= reg && reg <= REG_R31 { return C_REGP + int(reg&1) } if REG_F0 <= reg && reg <= REG_F31 { return C_FREGP + int(reg&1) } if REG_V0 <= reg && reg <= REG_V31 { return C_VREG } if REG_VS0 <= reg && reg <= REG_VS63 { return C_VSREGP + int(reg&1) } if REG_CR0 <= reg && reg <= REG_CR7 || reg == REG_CR { return C_CREG } if REG_CR0LT <= reg && reg <= REG_CR7SO { return C_CRBIT } if REG_SPR0 <= reg && reg <= REG_SPR0+1023 { switch reg { case REG_LR: return C_LR case REG_CTR: return C_CTR } return C_SPR } if REG_A0 <= reg && reg <= REG_A7 { return C_AREG } if reg == REG_FPSCR { return C_FPSCR } return C_GOK } func (c *ctxt9) aclass(a *obj.Addr) int { switch a.Type { case obj.TYPE_NONE: return C_NONE case obj.TYPE_REG: return c.aclassreg(a.Reg) case obj.TYPE_MEM: if a.Index != 0 { if a.Name != obj.NAME_NONE || a.Offset != 0 { c.ctxt.Logf("Unexpected Instruction operand index %d offset %d class %d \n", a.Index, a.Offset, a.Class) } return C_XOREG } switch a.Name { case obj.NAME_GOTREF, obj.NAME_TOCREF: return C_ADDR case obj.NAME_EXTERN, obj.NAME_STATIC: c.instoffset = a.Offset if a.Sym == nil { break } else if a.Sym.Type == objabi.STLSBSS { // For PIC builds, use 12 byte got initial-exec TLS accesses. if c.ctxt.Flag_shared { return C_TLS_IE } // Otherwise, use 8 byte local-exec TLS accesses. return C_TLS_LE } else { return C_ADDR } case obj.NAME_AUTO: a.Reg = REGSP c.instoffset = int64(c.autosize) + a.Offset if c.instoffset >= -BIG && c.instoffset < BIG { return C_SOREG } return C_LOREG case obj.NAME_PARAM: a.Reg = REGSP c.instoffset = int64(c.autosize) + a.Offset + c.ctxt.Arch.FixedFrameSize if c.instoffset >= -BIG && c.instoffset < BIG { return C_SOREG } return C_LOREG case obj.NAME_NONE: c.instoffset = a.Offset if a.Offset == 0 && a.Index == 0 { return C_ZOREG } else if c.instoffset >= -BIG && c.instoffset < BIG { return C_SOREG } else { return C_LOREG } } return C_GOK case obj.TYPE_TEXTSIZE: return C_TEXTSIZE case obj.TYPE_FCONST: // The only cases where FCONST will occur are with float64 +/- 0. // All other float constants are generated in memory. f64 := a.Val.(float64) if f64 == 0 { if math.Signbit(f64) { return C_S16CON } return C_ZCON } log.Fatalf("Unexpected nonzero FCONST operand %v", a) 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 return C_LACON case obj.NAME_AUTO: a.Reg = REGSP c.instoffset = int64(c.autosize) + a.Offset if c.instoffset >= -BIG && c.instoffset < BIG { return C_SACON } return C_LACON case obj.NAME_PARAM: a.Reg = REGSP c.instoffset = int64(c.autosize) + a.Offset + c.ctxt.Arch.FixedFrameSize if c.instoffset >= -BIG && c.instoffset < BIG { return C_SACON } return C_LACON default: return C_GOK } if c.instoffset >= 0 { sbits := bits.Len64(uint64(c.instoffset)) switch { case sbits <= 5: return C_ZCON + sbits case sbits <= 8: return C_U8CON case sbits <= 15: return C_U15CON case sbits <= 16: return C_U16CON case sbits <= 31: return C_U31CON case sbits <= 32: return C_U32CON case sbits <= 33: return C_S34CON default: return C_64CON } } else { sbits := bits.Len64(uint64(^c.instoffset)) switch { case sbits <= 15: return C_S16CON case sbits <= 31: return C_S32CON case sbits <= 33: return C_S34CON default: return C_64CON } } case obj.TYPE_BRANCH: if a.Sym != nil && c.ctxt.Flag_dynlink && !pfxEnabled { return C_BRAPIC } return C_BRA } return C_GOK } func prasm(p *obj.Prog) { fmt.Printf("%v\n", p) } func (c *ctxt9) oplook(p *obj.Prog) *Optab { 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-- argsv := [3]int{C_NONE + 1, C_NONE + 1, C_NONE + 1} for i, ap := range p.RestArgs { argsv[i] = int(ap.Addr.Class) if argsv[i] == 0 { argsv[i] = c.aclass(&ap.Addr) + 1 ap.Addr.Class = int8(argsv[i]) } } a3 := argsv[0] - 1 a4 := argsv[1] - 1 a5 := argsv[2] - 1 a6 := int(p.To.Class) if a6 == 0 { a6 = c.aclass(&p.To) + 1 p.To.Class = int8(a6) } a6-- a2 := C_NONE if p.Reg != 0 { a2 = c.aclassreg(p.Reg) } // c.ctxt.Logf("oplook %v %d %d %d %d\n", p, a1, a2, a3, a4, a5, a6) ops := oprange[p.As&obj.AMask] c1 := &xcmp[a1] c2 := &xcmp[a2] c3 := &xcmp[a3] c4 := &xcmp[a4] c5 := &xcmp[a5] c6 := &xcmp[a6] for i := range ops { op := &ops[i] if c1[op.a1] && c2[op.a2] && c3[op.a3] && c4[op.a4] && c5[op.a5] && c6[op.a6] { p.Optab = uint16(cap(optab) - cap(ops) + i + 1) return op } } c.ctxt.Diag("illegal combination %v %v %v %v %v %v %v", p.As, DRconv(a1), DRconv(a2), DRconv(a3), DRconv(a4), DRconv(a5), DRconv(a6)) prasm(p) if ops == nil { ops = optab } return &ops[0] } // Compare two operand types (ex C_REG, or C_U15CON) // and return true if b is compatible with a. // // Argument comparison isn't reflexitive, so care must be taken. // a is the argument type as found in optab, b is the argument as // fitted by aclass. func cmp(a int, b int) bool { if a == b { return true } switch a { case C_SPR: if b == C_LR || b == C_CTR { return true } case C_U1CON: return cmp(C_ZCON, b) case C_U2CON: return cmp(C_U1CON, b) case C_U3CON: return cmp(C_U2CON, b) case C_U4CON: return cmp(C_U3CON, b) case C_U5CON: return cmp(C_U4CON, b) case C_U8CON: return cmp(C_U5CON, b) case C_U15CON: return cmp(C_U8CON, b) case C_S16CON: return cmp(C_U15CON, b) case C_U16CON: return cmp(C_U15CON, b) case C_16CON: return cmp(C_S16CON, b) || cmp(C_U16CON, b) case C_U31CON: return cmp(C_U16CON, b) case C_U32CON: return cmp(C_U31CON, b) case C_S32CON: return cmp(C_U31CON, b) || cmp(C_S16CON, b) case C_32CON: return cmp(C_S32CON, b) || cmp(C_U32CON, b) case C_S34CON: return cmp(C_32CON, b) case C_64CON: return cmp(C_S34CON, b) case C_LACON: return cmp(C_SACON, b) case C_SOREG: return cmp(C_ZOREG, b) case C_LOREG: return cmp(C_SOREG, b) case C_XOREG: return cmp(C_REG, b) || cmp(C_ZOREG, b) // An even/odd register input always matches the regular register types. case C_REG: return cmp(C_REGP, b) || (b == C_ZCON && r0iszero != 0) case C_FREG: return cmp(C_FREGP, b) case C_VSREG: /* Allow any VR argument as a VSR operand. */ return cmp(C_VSREGP, b) || cmp(C_VREG, b) case C_ANY: return true } return false } // Used when sorting the optab. Sorting is // done in a way so that the best choice of // opcode/operand combination is considered first. func optabLess(i, j int) bool { p1 := &optab[i] p2 := &optab[j] n := int(p1.as) - int(p2.as) // same opcode if n != 0 { return n < 0 } // Consider those that generate fewer // instructions first. n = int(p1.size) - int(p2.size) if n != 0 { return n < 0 } // operand order should match // better choices first 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 } n = int(p1.a4) - int(p2.a4) if n != 0 { return n < 0 } n = int(p1.a5) - int(p2.a5) if n != 0 { return n < 0 } n = int(p1.a6) - int(p2.a6) if n != 0 { return n < 0 } return false } // Add an entry to the opcode table for // a new opcode b0 with the same operand combinations // as opcode a. func opset(a, b0 obj.As) { oprange[a&obj.AMask] = oprange[b0] } // Determine if the build configuration requires a TOC pointer. // It is assumed this always called after buildop. func NeedTOCpointer(ctxt *obj.Link) bool { return !pfxEnabled && ctxt.Flag_shared } // Build the opcode table func buildop(ctxt *obj.Link) { // Limit PC-relative prefix instruction usage to supported and tested targets. pfxEnabled = buildcfg.GOPPC64 >= 10 && buildcfg.GOOS == "linux" cfg := fmt.Sprintf("power%d/%s/%s", buildcfg.GOPPC64, buildcfg.GOARCH, buildcfg.GOOS) if cfg == buildOpCfg { // Already initialized to correct OS/cpu; stop now. // This happens in the cmd/asm tests, // each of which re-initializes the arch. return } buildOpCfg = cfg // Configure the optab entries which may generate prefix opcodes. prefixOptab := make([]Optab, 0, len(prefixableOptab)) for _, entry := range prefixableOptab { entry := entry if pfxEnabled && buildcfg.GOPPC64 >= entry.minGOPPC64 { // Enable prefix opcode generation and resize. entry.ispfx = true entry.size = entry.pfxsize } prefixOptab = append(prefixOptab, entry.Optab) } for i := 0; i < C_NCLASS; i++ { for n := 0; n < C_NCLASS; n++ { if cmp(n, i) { xcmp[i][n] = true } } } // Append the generated entries, sort, and fill out oprange. optab = make([]Optab, 0, len(optabBase)+len(optabGen)+len(prefixOptab)) optab = append(optab, optabBase...) optab = append(optab, optabGen...) optab = append(optab, prefixOptab...) sort.Slice(optab, optabLess) for i := range optab { // Use the legacy assembler function if none provided. if optab[i].asmout == nil { optab[i].asmout = asmout } } for i := 0; i < len(optab); { r := optab[i].as r0 := r & obj.AMask start := i for i < len(optab) && optab[i].as == r { i++ } oprange[r0] = optab[start:i] switch r { default: if !opsetGen(r) { ctxt.Diag("unknown op in build: %v", r) log.Fatalf("instruction missing from switch in asm9.go:buildop: %v", r) } case ADCBF: /* unary indexed: op (b+a); op (b) */ opset(ADCBI, r0) opset(ADCBST, r0) opset(ADCBT, r0) opset(ADCBTST, r0) opset(ADCBZ, r0) opset(AICBI, r0) case ASTDCCC: /* indexed store: op s,(b+a); op s,(b) */ opset(ASTWCCC, r0) opset(ASTHCCC, r0) opset(ASTBCCC, r0) case AREM: /* macro */ opset(AREM, r0) case AREMU: opset(AREMU, r0) case AREMD: opset(AREMDU, r0) case AMULLW: opset(AMULLD, r0) case ADIVW: /* op Rb[,Ra],Rd */ opset(AMULHW, r0) opset(AMULHWCC, r0) opset(AMULHWU, r0) opset(AMULHWUCC, r0) opset(AMULLWCC, r0) opset(AMULLWVCC, r0) opset(AMULLWV, r0) opset(ADIVWCC, r0) opset(ADIVWV, r0) opset(ADIVWVCC, r0) opset(ADIVWU, r0) opset(ADIVWUCC, r0) opset(ADIVWUV, r0) opset(ADIVWUVCC, r0) opset(AMODUD, r0) opset(AMODUW, r0) opset(AMODSD, r0) opset(AMODSW, r0) opset(AADDCC, r0) opset(AADDCV, r0) opset(AADDCVCC, r0) opset(AADDV, r0) opset(AADDVCC, r0) opset(AADDE, r0) opset(AADDECC, r0) opset(AADDEV, r0) opset(AADDEVCC, r0) opset(AMULHD, r0) opset(AMULHDCC, r0) opset(AMULHDU, r0) opset(AMULHDUCC, r0) opset(AMULLDCC, r0) opset(AMULLDVCC, r0) opset(AMULLDV, r0) opset(ADIVD, r0) opset(ADIVDCC, r0) opset(ADIVDE, r0) opset(ADIVDEU, r0) opset(ADIVDECC, r0) opset(ADIVDEUCC, r0) opset(ADIVDVCC, r0) opset(ADIVDV, r0) opset(ADIVDU, r0) opset(ADIVDUV, r0) opset(ADIVDUVCC, r0) opset(ADIVDUCC, r0) case ACRAND: opset(ACRANDN, r0) opset(ACREQV, r0) opset(ACRNAND, r0) opset(ACRNOR, r0) opset(ACROR, r0) opset(ACRORN, r0) opset(ACRXOR, r0) case APOPCNTD: /* popcntd, popcntw, popcntb, cnttzw, cnttzd */ opset(APOPCNTW, r0) opset(APOPCNTB, r0) opset(ACNTTZW, r0) opset(ACNTTZWCC, r0) opset(ACNTTZD, r0) opset(ACNTTZDCC, r0) case ACOPY: /* copy, paste. */ opset(APASTECC, r0) case AMADDHD: /* maddhd, maddhdu, maddld */ opset(AMADDHDU, r0) opset(AMADDLD, r0) case AMOVBZ: /* lbz, stz, rlwm(r/r), lhz, lha, stz, and x variants */ opset(AMOVH, r0) opset(AMOVHZ, r0) case AMOVBZU: /* lbz[x]u, stb[x]u, lhz[x]u, lha[x]u, sth[u]x, ld[x]u, std[u]x */ opset(AMOVHU, r0) opset(AMOVHZU, r0) opset(AMOVWU, r0) opset(AMOVWZU, r0) opset(AMOVDU, r0) opset(AMOVMW, r0) case ALVEBX: /* lvebx, lvehx, lvewx, lvx, lvxl, lvsl, lvsr */ opset(ALVEHX, r0) opset(ALVEWX, r0) opset(ALVX, r0) opset(ALVXL, r0) opset(ALVSL, r0) opset(ALVSR, r0) case ASTVEBX: /* stvebx, stvehx, stvewx, stvx, stvxl */ opset(ASTVEHX, r0) opset(ASTVEWX, r0) opset(ASTVX, r0) opset(ASTVXL, r0) case AVAND: /* vand, vandc, vnand */ opset(AVAND, r0) opset(AVANDC, r0) opset(AVNAND, r0) case AVMRGOW: /* vmrgew, vmrgow */ opset(AVMRGEW, r0) case AVOR: /* vor, vorc, vxor, vnor, veqv */ opset(AVOR, r0) opset(AVORC, r0) opset(AVXOR, r0) opset(AVNOR, r0) opset(AVEQV, r0) case AVADDUM: /* vaddubm, vadduhm, vadduwm, vaddudm, vadduqm */ opset(AVADDUBM, r0) opset(AVADDUHM, r0) opset(AVADDUWM, r0) opset(AVADDUDM, r0) opset(AVADDUQM, r0) case AVADDCU: /* vaddcuq, vaddcuw */ opset(AVADDCUQ, r0) opset(AVADDCUW, r0) case AVADDUS: /* vaddubs, vadduhs, vadduws */ opset(AVADDUBS, r0) opset(AVADDUHS, r0) opset(AVADDUWS, r0) case AVADDSS: /* vaddsbs, vaddshs, vaddsws */ opset(AVADDSBS, r0) opset(AVADDSHS, r0) opset(AVADDSWS, r0) case AVADDE: /* vaddeuqm, vaddecuq */ opset(AVADDEUQM, r0) opset(AVADDECUQ, r0) case AVSUBUM: /* vsububm, vsubuhm, vsubuwm, vsubudm, vsubuqm */ opset(AVSUBUBM, r0) opset(AVSUBUHM, r0) opset(AVSUBUWM, r0) opset(AVSUBUDM, r0) opset(AVSUBUQM, r0) case AVSUBCU: /* vsubcuq, vsubcuw */ opset(AVSUBCUQ, r0) opset(AVSUBCUW, r0) case AVSUBUS: /* vsububs, vsubuhs, vsubuws */ opset(AVSUBUBS, r0) opset(AVSUBUHS, r0) opset(AVSUBUWS, r0) case AVSUBSS: /* vsubsbs, vsubshs, vsubsws */ opset(AVSUBSBS, r0) opset(AVSUBSHS, r0) opset(AVSUBSWS, r0) case AVSUBE: /* vsubeuqm, vsubecuq */ opset(AVSUBEUQM, r0) opset(AVSUBECUQ, r0) case AVMULESB: /* vmulesb, vmulosb, vmuleub, vmuloub, vmulosh, vmulouh, vmulesw, vmulosw, vmuleuw, vmulouw, vmuluwm */ opset(AVMULOSB, r0) opset(AVMULEUB, r0) opset(AVMULOUB, r0) opset(AVMULESH, r0) opset(AVMULOSH, r0) opset(AVMULEUH, r0) opset(AVMULOUH, r0) opset(AVMULESW, r0) opset(AVMULOSW, r0) opset(AVMULEUW, r0) opset(AVMULOUW, r0) opset(AVMULUWM, r0) case AVPMSUM: /* vpmsumb, vpmsumh, vpmsumw, vpmsumd */ opset(AVPMSUMB, r0) opset(AVPMSUMH, r0) opset(AVPMSUMW, r0) opset(AVPMSUMD, r0) case AVR: /* vrlb, vrlh, vrlw, vrld */ opset(AVRLB, r0) opset(AVRLH, r0) opset(AVRLW, r0) opset(AVRLD, r0) case AVS: /* vs[l,r], vs[l,r]o, vs[l,r]b, vs[l,r]h, vs[l,r]w, vs[l,r]d */ opset(AVSLB, r0) opset(AVSLH, r0) opset(AVSLW, r0) opset(AVSL, r0) opset(AVSLO, r0) opset(AVSRB, r0) opset(AVSRH, r0) opset(AVSRW, r0) opset(AVSR, r0) opset(AVSRO, r0) opset(AVSLD, r0) opset(AVSRD, r0) case AVSA: /* vsrab, vsrah, vsraw, vsrad */ opset(AVSRAB, r0) opset(AVSRAH, r0) opset(AVSRAW, r0) opset(AVSRAD, r0) case AVSOI: /* vsldoi */ opset(AVSLDOI, r0) case AVCLZ: /* vclzb, vclzh, vclzw, vclzd */ opset(AVCLZB, r0) opset(AVCLZH, r0) opset(AVCLZW, r0) opset(AVCLZD, r0) case AVPOPCNT: /* vpopcntb, vpopcnth, vpopcntw, vpopcntd */ opset(AVPOPCNTB, r0) opset(AVPOPCNTH, r0) opset(AVPOPCNTW, r0) opset(AVPOPCNTD, r0) case AVCMPEQ: /* vcmpequb[.], vcmpequh[.], vcmpequw[.], vcmpequd[.] */ opset(AVCMPEQUB, r0) opset(AVCMPEQUBCC, r0) opset(AVCMPEQUH, r0) opset(AVCMPEQUHCC, r0) opset(AVCMPEQUW, r0) opset(AVCMPEQUWCC, r0) opset(AVCMPEQUD, r0) opset(AVCMPEQUDCC, r0) case AVCMPGT: /* vcmpgt[u,s]b[.], vcmpgt[u,s]h[.], vcmpgt[u,s]w[.], vcmpgt[u,s]d[.] */ opset(AVCMPGTUB, r0) opset(AVCMPGTUBCC, r0) opset(AVCMPGTUH, r0) opset(AVCMPGTUHCC, r0) opset(AVCMPGTUW, r0) opset(AVCMPGTUWCC, r0) opset(AVCMPGTUD, r0) opset(AVCMPGTUDCC, r0) opset(AVCMPGTSB, r0) opset(AVCMPGTSBCC, r0) opset(AVCMPGTSH, r0) opset(AVCMPGTSHCC, r0) opset(AVCMPGTSW, r0) opset(AVCMPGTSWCC, r0) opset(AVCMPGTSD, r0) opset(AVCMPGTSDCC, r0) case AVCMPNEZB: /* vcmpnezb[.] */ opset(AVCMPNEZBCC, r0) opset(AVCMPNEB, r0) opset(AVCMPNEBCC, r0) opset(AVCMPNEH, r0) opset(AVCMPNEHCC, r0) opset(AVCMPNEW, r0) opset(AVCMPNEWCC, r0) case AVPERM: /* vperm */ opset(AVPERMXOR, r0) opset(AVPERMR, r0) case AVBPERMQ: /* vbpermq, vbpermd */ opset(AVBPERMD, r0) case AVSEL: /* vsel */ opset(AVSEL, r0) case AVSPLTB: /* vspltb, vsplth, vspltw */ opset(AVSPLTH, r0) opset(AVSPLTW, r0) case AVSPLTISB: /* vspltisb, vspltish, vspltisw */ opset(AVSPLTISH, r0) opset(AVSPLTISW, r0) case AVCIPH: /* vcipher, vcipherlast */ opset(AVCIPHER, r0) opset(AVCIPHERLAST, r0) case AVNCIPH: /* vncipher, vncipherlast */ opset(AVNCIPHER, r0) opset(AVNCIPHERLAST, r0) case AVSBOX: /* vsbox */ opset(AVSBOX, r0) case AVSHASIGMA: /* vshasigmaw, vshasigmad */ opset(AVSHASIGMAW, r0) opset(AVSHASIGMAD, r0) case ALXVD2X: /* lxvd2x, lxvdsx, lxvw4x, lxvh8x, lxvb16x */ opset(ALXVDSX, r0) opset(ALXVW4X, r0) opset(ALXVH8X, r0) opset(ALXVB16X, r0) case ALXV: /* lxv */ opset(ALXV, r0) case ALXVL: /* lxvl, lxvll, lxvx */ opset(ALXVLL, r0) opset(ALXVX, r0) case ASTXVD2X: /* stxvd2x, stxvdsx, stxvw4x, stxvh8x, stxvb16x */ opset(ASTXVW4X, r0) opset(ASTXVH8X, r0) opset(ASTXVB16X, r0) case ASTXV: /* stxv */ opset(ASTXV, r0) case ASTXVL: /* stxvl, stxvll, stvx */ opset(ASTXVLL, r0) opset(ASTXVX, r0) case ALXSDX: /* lxsdx */ opset(ALXSDX, r0) case ASTXSDX: /* stxsdx */ opset(ASTXSDX, r0) case ALXSIWAX: /* lxsiwax, lxsiwzx */ opset(ALXSIWZX, r0) case ASTXSIWX: /* stxsiwx */ opset(ASTXSIWX, r0) case AMFVSRD: /* mfvsrd, mfvsrwz (and extended mnemonics), mfvsrld */ opset(AMFFPRD, r0) opset(AMFVRD, r0) opset(AMFVSRWZ, r0) opset(AMFVSRLD, r0) case AMTVSRD: /* mtvsrd, mtvsrwa, mtvsrwz (and extended mnemonics), mtvsrdd, mtvsrws */ opset(AMTFPRD, r0) opset(AMTVRD, r0) opset(AMTVSRWA, r0) opset(AMTVSRWZ, r0) opset(AMTVSRWS, r0) case AXXLAND: opset(AXXLANDC, r0) opset(AXXLEQV, r0) opset(AXXLNAND, r0) opset(AXXLORC, r0) opset(AXXLNOR, r0) opset(AXXLORQ, r0) opset(AXXLXOR, r0) opset(AXXLOR, r0) opset(AXSMAXJDP, r0) opset(AXSMINJDP, r0) case AXXSEL: /* xxsel */ opset(AXXSEL, r0) case AXXMRGHW: /* xxmrghw, xxmrglw */ opset(AXXMRGLW, r0) case AXXSPLTW: /* xxspltw */ opset(AXXSPLTW, r0) case AXXSPLTIB: /* xxspltib */ opset(AXXSPLTIB, r0) case AXXPERM: /* xxpermdi */ opset(AXXPERM, r0) case AXXSLDWI: /* xxsldwi */ opset(AXXPERMDI, r0) opset(AXXSLDWI, r0) case AXXBRQ: /* xxbrq, xxbrd, xxbrw, xxbrh */ opset(AXXBRD, r0) opset(AXXBRW, r0) opset(AXXBRH, r0) case AXSCVDPSP: /* xscvdpsp, xscvspdp, xscvdpspn, xscvspdpn */ opset(AXSCVSPDP, r0) opset(AXSCVDPSPN, r0) opset(AXSCVSPDPN, r0) case AXVCVDPSP: /* xvcvdpsp, xvcvspdp */ opset(AXVCVSPDP, r0) case AXSCVDPSXDS: /* xscvdpsxds, xscvdpsxws, xscvdpuxds, xscvdpuxws */ opset(AXSCVDPSXWS, r0) opset(AXSCVDPUXDS, r0) opset(AXSCVDPUXWS, r0) case AXSCVSXDDP: /* xscvsxddp, xscvuxddp, xscvsxdsp, xscvuxdsp */ opset(AXSCVUXDDP, r0) opset(AXSCVSXDSP, r0) opset(AXSCVUXDSP, r0) case AXVCVDPSXDS: /* xvcvdpsxds, xvcvdpsxws, xvcvdpuxds, xvcvdpuxws, xvcvspsxds, xvcvspsxws, xvcvspuxds, xvcvspuxws */ opset(AXVCVDPSXDS, r0) opset(AXVCVDPSXWS, r0) opset(AXVCVDPUXDS, r0) opset(AXVCVDPUXWS, r0) opset(AXVCVSPSXDS, r0) opset(AXVCVSPSXWS, r0) opset(AXVCVSPUXDS, r0) opset(AXVCVSPUXWS, r0) case AXVCVSXDDP: /* xvcvsxddp, xvcvsxwdp, xvcvuxddp, xvcvuxwdp, xvcvsxdsp, xvcvsxwsp, xvcvuxdsp, xvcvuxwsp */ opset(AXVCVSXWDP, r0) opset(AXVCVUXDDP, r0) opset(AXVCVUXWDP, r0) opset(AXVCVSXDSP, r0) opset(AXVCVSXWSP, r0) opset(AXVCVUXDSP, r0) opset(AXVCVUXWSP, r0) case AAND: /* logical op Rb,Rs,Ra; no literal */ opset(AANDN, r0) opset(AANDNCC, r0) opset(AEQV, r0) opset(AEQVCC, r0) opset(ANAND, r0) opset(ANANDCC, r0) opset(ANOR, r0) opset(ANORCC, r0) opset(AORCC, r0) opset(AORN, r0) opset(AORNCC, r0) opset(AXORCC, r0) case AADDME: /* op Ra, Rd */ opset(AADDMECC, r0) opset(AADDMEV, r0) opset(AADDMEVCC, r0) opset(AADDZE, r0) opset(AADDZECC, r0) opset(AADDZEV, r0) opset(AADDZEVCC, r0) opset(ASUBME, r0) opset(ASUBMECC, r0) opset(ASUBMEV, r0) opset(ASUBMEVCC, r0) opset(ASUBZE, r0) opset(ASUBZECC, r0) opset(ASUBZEV, r0) opset(ASUBZEVCC, r0) case AADDC: opset(AADDCCC, r0) case ABEQ: opset(ABGE, r0) opset(ABGT, r0) opset(ABLE, r0) opset(ABLT, r0) opset(ABNE, r0) opset(ABVC, r0) opset(ABVS, r0) case ABR: opset(ABL, r0) case ABC: opset(ABCL, r0) case ABDNZ: opset(ABDZ, r0) case AEXTSB: /* op Rs, Ra */ opset(AEXTSBCC, r0) opset(AEXTSH, r0) opset(AEXTSHCC, r0) opset(ACNTLZW, r0) opset(ACNTLZWCC, r0) opset(ACNTLZD, r0) opset(AEXTSW, r0) opset(AEXTSWCC, r0) opset(ACNTLZDCC, r0) case AFABS: /* fop [s,]d */ opset(AFABSCC, r0) opset(AFNABS, r0) opset(AFNABSCC, r0) opset(AFNEG, r0) opset(AFNEGCC, r0) opset(AFRSP, r0) opset(AFRSPCC, r0) opset(AFCTIW, r0) opset(AFCTIWCC, r0) opset(AFCTIWZ, r0) opset(AFCTIWZCC, r0) opset(AFCTID, r0) opset(AFCTIDCC, r0) opset(AFCTIDZ, r0) opset(AFCTIDZCC, r0) opset(AFCFID, r0) opset(AFCFIDCC, r0) opset(AFCFIDU, r0) opset(AFCFIDUCC, r0) opset(AFCFIDS, r0) opset(AFCFIDSCC, r0) opset(AFRES, r0) opset(AFRESCC, r0) opset(AFRIM, r0) opset(AFRIMCC, r0) opset(AFRIP, r0) opset(AFRIPCC, r0) opset(AFRIZ, r0) opset(AFRIZCC, r0) opset(AFRIN, r0) opset(AFRINCC, r0) opset(AFRSQRTE, r0) opset(AFRSQRTECC, r0) opset(AFSQRT, r0) opset(AFSQRTCC, r0) opset(AFSQRTS, r0) opset(AFSQRTSCC, r0) case AFADD: opset(AFADDS, r0) opset(AFADDCC, r0) opset(AFADDSCC, r0) opset(AFCPSGN, r0) opset(AFCPSGNCC, r0) opset(AFDIV, r0) opset(AFDIVS, r0) opset(AFDIVCC, r0) opset(AFDIVSCC, r0) opset(AFSUB, r0) opset(AFSUBS, r0) opset(AFSUBCC, r0) opset(AFSUBSCC, r0) case AFMADD: opset(AFMADDCC, r0) opset(AFMADDS, r0) opset(AFMADDSCC, r0) opset(AFMSUB, r0) opset(AFMSUBCC, r0) opset(AFMSUBS, r0) opset(AFMSUBSCC, r0) opset(AFNMADD, r0) opset(AFNMADDCC, r0) opset(AFNMADDS, r0) opset(AFNMADDSCC, r0) opset(AFNMSUB, r0) opset(AFNMSUBCC, r0) opset(AFNMSUBS, r0) opset(AFNMSUBSCC, r0) opset(AFSEL, r0) opset(AFSELCC, r0) case AFMUL: opset(AFMULS, r0) opset(AFMULCC, r0) opset(AFMULSCC, r0) case AFCMPO: opset(AFCMPU, r0) case AMTFSB0: opset(AMTFSB0CC, r0) opset(AMTFSB1, r0) opset(AMTFSB1CC, r0) case ANEG: /* op [Ra,] Rd */ opset(ANEGCC, r0) opset(ANEGV, r0) opset(ANEGVCC, r0) case AOR: /* or/xor Rb,Rs,Ra; ori/xori $uimm,Rs,R */ opset(AXOR, r0) case AORIS: /* oris/xoris $uimm,Rs,Ra */ opset(AXORIS, r0) case ASLW: opset(ASLWCC, r0) opset(ASRW, r0) opset(ASRWCC, r0) opset(AROTLW, r0) case ASLD: opset(ASLDCC, r0) opset(ASRD, r0) opset(ASRDCC, r0) opset(AROTL, r0) case ASRAW: /* sraw Rb,Rs,Ra; srawi sh,Rs,Ra */ opset(ASRAWCC, r0) case AEXTSWSLI: opset(AEXTSWSLICC, r0) case ASRAD: /* sraw Rb,Rs,Ra; srawi sh,Rs,Ra */ opset(ASRADCC, r0) case ASUB: /* SUB Ra,Rb,Rd => subf Rd,ra,rb */ opset(ASUB, r0) opset(ASUBCC, r0) opset(ASUBV, r0) opset(ASUBVCC, r0) opset(ASUBCCC, r0) opset(ASUBCV, r0) opset(ASUBCVCC, r0) opset(ASUBE, r0) opset(ASUBECC, r0) opset(ASUBEV, r0) opset(ASUBEVCC, r0) case ASYNC: opset(AISYNC, r0) opset(ALWSYNC, r0) opset(APTESYNC, r0) opset(ATLBSYNC, r0) case ARLWNM: opset(ARLWNMCC, r0) opset(ARLWMI, r0) opset(ARLWMICC, r0) case ARLDMI: opset(ARLDMICC, r0) opset(ARLDIMI, r0) opset(ARLDIMICC, r0) case ARLDC: opset(ARLDCCC, r0) case ARLDCL: opset(ARLDCR, r0) opset(ARLDCLCC, r0) opset(ARLDCRCC, r0) case ARLDICL: opset(ARLDICLCC, r0) opset(ARLDICR, r0) opset(ARLDICRCC, r0) opset(ARLDIC, r0) opset(ARLDICCC, r0) opset(ACLRLSLDI, r0) case AFMOVD: opset(AFMOVDCC, r0) opset(AFMOVDU, r0) opset(AFMOVS, r0) opset(AFMOVSU, r0) case ALDAR: opset(ALBAR, r0) opset(ALHAR, r0) opset(ALWAR, r0) case ASYSCALL: /* just the op; flow of control */ opset(ARFI, r0) opset(ARFCI, r0) opset(ARFID, r0) opset(AHRFID, r0) case AMOVHBR: opset(AMOVWBR, r0) opset(AMOVDBR, r0) case ASLBMFEE: opset(ASLBMFEV, r0) case ATW: opset(ATD, r0) case ATLBIE: opset(ASLBIE, r0) opset(ATLBIEL, r0) case AEIEIO: opset(ASLBIA, r0) case ACMP: opset(ACMPW, r0) case ACMPU: opset(ACMPWU, r0) case ACMPB: opset(ACMPB, r0) case AFTDIV: opset(AFTDIV, r0) case AFTSQRT: opset(AFTSQRT, r0) case AMOVW: /* load/store/move word with sign extension; move 32-bit literals */ opset(AMOVWZ, r0) /* Same as above, but zero extended */ case AVCLZLSBB: opset(AVCTZLSBB, r0) case AADD, AADDIS, AANDCC, /* and. Rb,Rs,Ra; andi. $uimm,Rs,Ra */ AANDISCC, AFMOVSX, AFMOVSZ, ALSW, AMOVD, /* load/store/move 64-bit values, including 32-bit literals with/without sign-extension */ AMOVB, /* macro: move byte with sign extension */ AMOVBU, /* macro: move byte with sign extension & update */ AMOVFL, /* op $s[,r2],r3; op r1[,r2],r3; no cc/v */ ASUBC, /* op r1,$s,r3; op r1[,r2],r3 */ ASTSW, ASLBMTE, AWORD, ADWORD, ADARN, AVMSUMUDM, AADDEX, ACMPEQB, ACLRLSLWI, AMTVSRDD, APNOP, AISEL, ASETB, obj.ANOP, obj.ATEXT, obj.AUNDEF, obj.AFUNCDATA, obj.APCALIGN, obj.APCDATA, obj.ADUFFZERO, obj.ADUFFCOPY: break } } } func OPVXX1(o uint32, xo uint32, oe uint32) uint32 { return o<<26 | xo<<1 | oe<<11 } func OPVXX2(o uint32, xo uint32, oe uint32) uint32 { return o<<26 | xo<<2 | oe<<11 } func OPVXX2VA(o uint32, xo uint32, oe uint32) uint32 { return o<<26 | xo<<2 | oe<<16 } func OPVXX3(o uint32, xo uint32, oe uint32) uint32 { return o<<26 | xo<<3 | oe<<11 } func OPVXX4(o uint32, xo uint32, oe uint32) uint32 { return o<<26 | xo<<4 | oe<<11 } func OPDQ(o uint32, xo uint32, oe uint32) uint32 { return o<<26 | xo | oe<<4 } func OPVX(o uint32, xo uint32, oe uint32, rc uint32) uint32 { return o<<26 | xo | oe<<11 | rc&1 } func OPVC(o uint32, xo uint32, oe uint32, rc uint32) uint32 { return o<<26 | xo | oe<<11 | (rc&1)<<10 } func OPVCC(o uint32, xo uint32, oe uint32, rc uint32) uint32 { return o<<26 | xo<<1 | oe<<10 | rc&1 } func OPCC(o uint32, xo uint32, rc uint32) uint32 { return OPVCC(o, xo, 0, rc) } /* Generate MD-form opcode */ func OPMD(o, xo, rc uint32) uint32 { return o<<26 | xo<<2 | rc&1 } /* the order is dest, a/s, b/imm for both arithmetic and logical operations. */ func AOP_RRR(op uint32, d uint32, a uint32, b uint32) uint32 { return op | (d&31)<<21 | (a&31)<<16 | (b&31)<<11 } /* VX-form 2-register operands, r/none/r */ func AOP_RR(op uint32, d uint32, a uint32) uint32 { return op | (d&31)<<21 | (a&31)<<11 } /* VA-form 4-register operands */ func AOP_RRRR(op uint32, d uint32, a uint32, b uint32, c uint32) uint32 { return op | (d&31)<<21 | (a&31)<<16 | (b&31)<<11 | (c&31)<<6 } func AOP_IRR(op uint32, d uint32, a uint32, simm uint32) uint32 { return op | (d&31)<<21 | (a&31)<<16 | simm&0xFFFF } /* VX-form 2-register + UIM operands */ func AOP_VIRR(op uint32, d uint32, a uint32, simm uint32) uint32 { return op | (d&31)<<21 | (simm&0xFFFF)<<16 | (a&31)<<11 } /* VX-form 2-register + ST + SIX operands */ func AOP_IIRR(op uint32, d uint32, a uint32, sbit uint32, simm uint32) uint32 { return op | (d&31)<<21 | (a&31)<<16 | (sbit&1)<<15 | (simm&0xF)<<11 } /* VA-form 3-register + SHB operands */ func AOP_IRRR(op uint32, d uint32, a uint32, b uint32, simm uint32) uint32 { return op | (d&31)<<21 | (a&31)<<16 | (b&31)<<11 | (simm&0xF)<<6 } /* VX-form 1-register + SIM operands */ func AOP_IR(op uint32, d uint32, simm uint32) uint32 { return op | (d&31)<<21 | (simm&31)<<16 } /* XX1-form 3-register operands, 1 VSR operand */ func AOP_XX1(op uint32, r uint32, a uint32, b uint32) uint32 { return op | (r&31)<<21 | (a&31)<<16 | (b&31)<<11 | (r&32)>>5 } /* XX2-form 3-register operands, 2 VSR operands */ func AOP_XX2(op uint32, xt uint32, a uint32, xb uint32) uint32 { return op | (xt&31)<<21 | (a&3)<<16 | (xb&31)<<11 | (xb&32)>>4 | (xt&32)>>5 } /* XX3-form 3 VSR operands */ func AOP_XX3(op uint32, xt uint32, xa uint32, xb uint32) uint32 { return op | (xt&31)<<21 | (xa&31)<<16 | (xb&31)<<11 | (xa&32)>>3 | (xb&32)>>4 | (xt&32)>>5 } /* XX3-form 3 VSR operands + immediate */ func AOP_XX3I(op uint32, xt uint32, xa uint32, xb uint32, c uint32) uint32 { return op | (xt&31)<<21 | (xa&31)<<16 | (xb&31)<<11 | (c&3)<<8 | (xa&32)>>3 | (xb&32)>>4 | (xt&32)>>5 } /* XX4-form, 4 VSR operands */ func AOP_XX4(op uint32, xt uint32, xa uint32, xb uint32, xc uint32) uint32 { return op | (xt&31)<<21 | (xa&31)<<16 | (xb&31)<<11 | (xc&31)<<6 | (xc&32)>>2 | (xa&32)>>3 | (xb&32)>>4 | (xt&32)>>5 } /* DQ-form, VSR register, register + offset operands */ func AOP_DQ(op uint32, xt uint32, a uint32, b uint32) uint32 { /* The EA for this instruction form is (RA) + DQ << 4, where DQ is a 12-bit signed integer. */ /* In order to match the output of the GNU objdump (and make the usage in Go asm easier), the */ /* instruction is called using the sign extended value (i.e. a valid offset would be -32752 or 32752, */ /* not -2047 or 2047), so 'b' needs to be adjusted to the expected 12-bit DQ value. Bear in mind that */ /* bits 0 to 3 in 'dq' need to be zero, otherwise this will generate an illegal instruction. */ /* If in doubt how this instruction form is encoded, refer to ISA 3.0b, pages 492 and 507. */ dq := b >> 4 return op | (xt&31)<<21 | (a&31)<<16 | (dq&4095)<<4 | (xt&32)>>2 } /* Z23-form, 3-register operands + CY field */ func AOP_Z23I(op uint32, d uint32, a uint32, b uint32, c uint32) uint32 { return op | (d&31)<<21 | (a&31)<<16 | (b&31)<<11 | (c&3)<<9 } /* X-form, 3-register operands + EH field */ func AOP_RRRI(op uint32, d uint32, a uint32, b uint32, c uint32) uint32 { return op | (d&31)<<21 | (a&31)<<16 | (b&31)<<11 | (c & 1) } func LOP_RRR(op uint32, a uint32, s uint32, b uint32) uint32 { return op | (s&31)<<21 | (a&31)<<16 | (b&31)<<11 } func LOP_IRR(op uint32, a uint32, s uint32, uimm uint32) uint32 { return op | (s&31)<<21 | (a&31)<<16 | uimm&0xFFFF } func OP_BR(op uint32, li uint32, aa uint32) uint32 { return op | li&0x03FFFFFC | aa<<1 } func OP_BC(op uint32, bo uint32, bi uint32, bd uint32, aa uint32) uint32 { return op | (bo&0x1F)<<21 | (bi&0x1F)<<16 | bd&0xFFFC | aa<<1 } func OP_BCR(op uint32, bo uint32, bi uint32) uint32 { return op | (bo&0x1F)<<21 | (bi&0x1F)<<16 } func OP_RLW(op uint32, a uint32, s uint32, sh uint32, mb uint32, me uint32) uint32 { return op | (s&31)<<21 | (a&31)<<16 | (sh&31)<<11 | (mb&31)<<6 | (me&31)<<1 } func AOP_EXTSWSLI(op uint32, a uint32, s uint32, sh uint32) uint32 { return op | (a&31)<<21 | (s&31)<<16 | (sh&31)<<11 | ((sh&32)>>5)<<1 } func AOP_ISEL(op uint32, t uint32, a uint32, b uint32, bc uint32) uint32 { return op | (t&31)<<21 | (a&31)<<16 | (b&31)<<11 | (bc&0x1F)<<6 } /* MD-form 2-register, 2 6-bit immediate operands */ func AOP_MD(op uint32, a uint32, s uint32, sh uint32, m uint32) uint32 { return op | (s&31)<<21 | (a&31)<<16 | (sh&31)<<11 | ((sh&32)>>5)<<1 | (m&31)<<6 | ((m&32)>>5)<<5 } /* MDS-form 3-register, 1 6-bit immediate operands. rsh argument is a register. */ func AOP_MDS(op, to, from, rsh, m uint32) uint32 { return AOP_MD(op, to, from, rsh&31, m) } func AOP_PFX_00_8LS(r, ie uint32) uint32 { return 1<<26 | 0<<24 | 0<<23 | (r&1)<<20 | (ie & 0x3FFFF) } func AOP_PFX_10_MLS(r, ie uint32) uint32 { return 1<<26 | 2<<24 | 0<<23 | (r&1)<<20 | (ie & 0x3FFFF) } const ( /* each rhs is OPVCC(_, _, _, _) */ OP_ADD = 31<<26 | 266<<1 | 0<<10 | 0 OP_ADDI = 14<<26 | 0<<1 | 0<<10 | 0 OP_ADDIS = 15<<26 | 0<<1 | 0<<10 | 0 OP_ANDI = 28<<26 | 0<<1 | 0<<10 | 0 OP_EXTSB = 31<<26 | 954<<1 | 0<<10 | 0 OP_EXTSH = 31<<26 | 922<<1 | 0<<10 | 0 OP_EXTSW = 31<<26 | 986<<1 | 0<<10 | 0 OP_ISEL = 31<<26 | 15<<1 | 0<<10 | 0 OP_MCRF = 19<<26 | 0<<1 | 0<<10 | 0 OP_MCRFS = 63<<26 | 64<<1 | 0<<10 | 0 OP_MCRXR = 31<<26 | 512<<1 | 0<<10 | 0 OP_MFCR = 31<<26 | 19<<1 | 0<<10 | 0 OP_MFFS = 63<<26 | 583<<1 | 0<<10 | 0 OP_MFSPR = 31<<26 | 339<<1 | 0<<10 | 0 OP_MFSR = 31<<26 | 595<<1 | 0<<10 | 0 OP_MFSRIN = 31<<26 | 659<<1 | 0<<10 | 0 OP_MTCRF = 31<<26 | 144<<1 | 0<<10 | 0 OP_MTFSF = 63<<26 | 711<<1 | 0<<10 | 0 OP_MTFSFI = 63<<26 | 134<<1 | 0<<10 | 0 OP_MTSPR = 31<<26 | 467<<1 | 0<<10 | 0 OP_MTSR = 31<<26 | 210<<1 | 0<<10 | 0 OP_MTSRIN = 31<<26 | 242<<1 | 0<<10 | 0 OP_MULLW = 31<<26 | 235<<1 | 0<<10 | 0 OP_MULLD = 31<<26 | 233<<1 | 0<<10 | 0 OP_OR = 31<<26 | 444<<1 | 0<<10 | 0 OP_ORI = 24<<26 | 0<<1 | 0<<10 | 0 OP_ORIS = 25<<26 | 0<<1 | 0<<10 | 0 OP_XORI = 26<<26 | 0<<1 | 0<<10 | 0 OP_XORIS = 27<<26 | 0<<1 | 0<<10 | 0 OP_RLWINM = 21<<26 | 0<<1 | 0<<10 | 0 OP_RLWNM = 23<<26 | 0<<1 | 0<<10 | 0 OP_SUBF = 31<<26 | 40<<1 | 0<<10 | 0 OP_RLDIC = 30<<26 | 4<<1 | 0<<10 | 0 OP_RLDICR = 30<<26 | 2<<1 | 0<<10 | 0 OP_RLDICL = 30<<26 | 0<<1 | 0<<10 | 0 OP_RLDCL = 30<<26 | 8<<1 | 0<<10 | 0 OP_EXTSWSLI = 31<<26 | 445<<2 OP_SETB = 31<<26 | 128<<1 ) func pfxadd(rt, ra int16, r uint32, imm32 int64) (uint32, uint32) { return AOP_PFX_10_MLS(r, uint32(imm32>>16)), AOP_IRR(14<<26, uint32(rt), uint32(ra), uint32(imm32)) } func pfxload(a obj.As, reg int16, base int16, r uint32) (uint32, uint32) { switch a { case AMOVH: return AOP_PFX_10_MLS(r, 0), AOP_IRR(42<<26, uint32(reg), uint32(base), 0) case AMOVW: return AOP_PFX_00_8LS(r, 0), AOP_IRR(41<<26, uint32(reg), uint32(base), 0) case AMOVD: return AOP_PFX_00_8LS(r, 0), AOP_IRR(57<<26, uint32(reg), uint32(base), 0) case AMOVBZ, AMOVB: return AOP_PFX_10_MLS(r, 0), AOP_IRR(34<<26, uint32(reg), uint32(base), 0) case AMOVHZ: return AOP_PFX_10_MLS(r, 0), AOP_IRR(40<<26, uint32(reg), uint32(base), 0) case AMOVWZ: return AOP_PFX_10_MLS(r, 0), AOP_IRR(32<<26, uint32(reg), uint32(base), 0) case AFMOVS: return AOP_PFX_10_MLS(r, 0), AOP_IRR(48<<26, uint32(reg), uint32(base), 0) case AFMOVD: return AOP_PFX_10_MLS(r, 0), AOP_IRR(50<<26, uint32(reg), uint32(base), 0) } log.Fatalf("Error no pfxload for %v\n", a) return 0, 0 } func pfxstore(a obj.As, reg int16, base int16, r uint32) (uint32, uint32) { switch a { case AMOVD: return AOP_PFX_00_8LS(r, 0), AOP_IRR(61<<26, uint32(reg), uint32(base), 0) case AMOVBZ, AMOVB: return AOP_PFX_10_MLS(r, 0), AOP_IRR(38<<26, uint32(reg), uint32(base), 0) case AMOVHZ, AMOVH: return AOP_PFX_10_MLS(r, 0), AOP_IRR(44<<26, uint32(reg), uint32(base), 0) case AMOVWZ, AMOVW: return AOP_PFX_10_MLS(r, 0), AOP_IRR(36<<26, uint32(reg), uint32(base), 0) case AFMOVS: return AOP_PFX_10_MLS(r, 0), AOP_IRR(52<<26, uint32(reg), uint32(base), 0) case AFMOVD: return AOP_PFX_10_MLS(r, 0), AOP_IRR(54<<26, uint32(reg), uint32(base), 0) } log.Fatalf("Error no pfxstore for %v\n", a) return 0, 0 } func oclass(a *obj.Addr) int { return int(a.Class) - 1 } const ( D_FORM = iota DS_FORM ) // This function determines when a non-indexed load or store is D or // DS form for use in finding the size of the offset field in the instruction. // The size is needed when setting the offset value in the instruction // and when generating relocation for that field. // DS form instructions include: ld, ldu, lwa, std, stdu. All other // loads and stores with an offset field are D form. This function should // only be called with the same opcodes as are handled by opstore and opload. func (c *ctxt9) opform(insn uint32) int { switch insn { default: c.ctxt.Diag("bad insn in loadform: %x", insn) case OPVCC(58, 0, 0, 0), // ld OPVCC(58, 0, 0, 1), // ldu OPVCC(58, 0, 0, 0) | 1<<1, // lwa OPVCC(62, 0, 0, 0), // std OPVCC(62, 0, 0, 1): //stdu return DS_FORM case OP_ADDI, // add OPVCC(32, 0, 0, 0), // lwz OPVCC(33, 0, 0, 0), // lwzu OPVCC(34, 0, 0, 0), // lbz OPVCC(35, 0, 0, 0), // lbzu OPVCC(40, 0, 0, 0), // lhz OPVCC(41, 0, 0, 0), // lhzu OPVCC(42, 0, 0, 0), // lha OPVCC(43, 0, 0, 0), // lhau OPVCC(46, 0, 0, 0), // lmw OPVCC(48, 0, 0, 0), // lfs OPVCC(49, 0, 0, 0), // lfsu OPVCC(50, 0, 0, 0), // lfd OPVCC(51, 0, 0, 0), // lfdu OPVCC(36, 0, 0, 0), // stw OPVCC(37, 0, 0, 0), // stwu OPVCC(38, 0, 0, 0), // stb OPVCC(39, 0, 0, 0), // stbu OPVCC(44, 0, 0, 0), // sth OPVCC(45, 0, 0, 0), // sthu OPVCC(47, 0, 0, 0), // stmw OPVCC(52, 0, 0, 0), // stfs OPVCC(53, 0, 0, 0), // stfsu OPVCC(54, 0, 0, 0), // stfd OPVCC(55, 0, 0, 0): // stfdu return D_FORM } return 0 } // Encode instructions and create relocation for accessing s+d according to the // instruction op with source or destination (as appropriate) register reg. func (c *ctxt9) symbolAccess(s *obj.LSym, d int64, reg int16, op uint32, reuse bool) (o1, o2 uint32, rel *obj.Reloc) { if c.ctxt.Headtype == objabi.Haix { // Every symbol access must be made via a TOC anchor. c.ctxt.Diag("symbolAccess called for %s", s.Name) } var base uint32 form := c.opform(op) if c.ctxt.Flag_shared { base = REG_R2 } else { base = REG_R0 } // If reg can be reused when computing the symbol address, // use it instead of REGTMP. if !reuse { o1 = AOP_IRR(OP_ADDIS, REGTMP, base, 0) o2 = AOP_IRR(op, uint32(reg), REGTMP, 0) } else { o1 = AOP_IRR(OP_ADDIS, uint32(reg), base, 0) o2 = AOP_IRR(op, uint32(reg), uint32(reg), 0) } rel = obj.Addrel(c.cursym) rel.Off = int32(c.pc) rel.Siz = 8 rel.Sym = s rel.Add = d if c.ctxt.Flag_shared { switch form { case D_FORM: rel.Type = objabi.R_ADDRPOWER_TOCREL case DS_FORM: rel.Type = objabi.R_ADDRPOWER_TOCREL_DS } } else { switch form { case D_FORM: rel.Type = objabi.R_ADDRPOWER case DS_FORM: rel.Type = objabi.R_ADDRPOWER_DS } } return } // Determine the mask begin (mb) and mask end (me) values // for a valid word rotate mask. A valid 32 bit mask is of // the form 1+0*1+ or 0*1+0*. // // Note, me is inclusive. func decodeMask32(mask uint32) (mb, me uint32, valid bool) { mb = uint32(bits.LeadingZeros32(mask)) me = uint32(32 - bits.TrailingZeros32(mask)) mbn := uint32(bits.LeadingZeros32(^mask)) men := uint32(32 - bits.TrailingZeros32(^mask)) // Check for a wrapping mask (e.g bits at 0 and 31) if mb == 0 && me == 32 { // swap the inverted values mb, me = men, mbn } // Validate mask is of the binary form 1+0*1+ or 0*1+0* // Isolate rightmost 1 (if none 0) and add. v := mask vp := (v & -v) + v // Likewise, check for the wrapping (inverted) case. vn := ^v vpn := (vn & -vn) + vn return mb, (me - 1) & 31, (v&vp == 0 || vn&vpn == 0) && v != 0 } // Decompose a mask of contiguous bits into a begin (mb) and // end (me) value. // // 64b mask values cannot wrap on any valid PPC64 instruction. // Only masks of the form 0*1+0* are valid. // // Note, me is inclusive. func decodeMask64(mask int64) (mb, me uint32, valid bool) { m := uint64(mask) mb = uint32(bits.LeadingZeros64(m)) me = uint32(64 - bits.TrailingZeros64(m)) valid = ((m&-m)+m)&m == 0 && m != 0 return mb, (me - 1) & 63, valid } // Load the lower 16 bits of a constant into register r. func loadl16(r int, d int64) uint32 { v := uint16(d) if v == 0 { // Avoid generating "ori r,r,0", r != 0. Instead, generate the architectually preferred nop. // For example, "ori r31,r31,0" is a special execution serializing nop on Power10 called "exser". return NOP } return LOP_IRR(OP_ORI, uint32(r), uint32(r), uint32(v)) } // Load the upper 16 bits of a 32b constant into register r. func loadu32(r int, d int64) uint32 { v := int32(d >> 16) if isuint32(uint64(d)) { return LOP_IRR(OP_ORIS, uint32(r), REGZERO, uint32(v)) } return AOP_IRR(OP_ADDIS, uint32(r), REGZERO, uint32(v)) } func high16adjusted(d int32) uint16 { if d&0x8000 != 0 { return uint16((d >> 16) + 1) } return uint16(d >> 16) } func asmout(c *ctxt9, p *obj.Prog, o *Optab, out *[5]uint32) { o1 := uint32(0) o2 := uint32(0) o3 := uint32(0) o4 := uint32(0) o5 := uint32(0) //print("%v => case %d\n", p, o->type); switch o.type_ { default: c.ctxt.Diag("unknown type %d", o.type_) prasm(p) case 0: /* pseudo ops */ break case 2: /* int/cr/fp op Rb,[Ra],Rd */ r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } o1 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(r), uint32(p.From.Reg)) case 3: /* mov $soreg/16con, r ==> addi/ori $i,reg',r */ d := c.vregoff(&p.From) v := int32(d) r := int(p.From.Reg) if r0iszero != 0 /*TypeKind(100016)*/ && p.To.Reg == 0 && (r != 0 || v != 0) { c.ctxt.Diag("literal operation on R0\n%v", p) } if int64(int16(d)) == d { // MOVD $int16, Ry or MOVD $offset(Rx), Ry o1 = AOP_IRR(uint32(OP_ADDI), uint32(p.To.Reg), uint32(r), uint32(v)) } else { // MOVD $uint16, Ry if int64(uint16(d)) != d || (r != 0 && r != REGZERO) { c.ctxt.Diag("Rule expects a uint16 constant load. got:\n%v", p) } o1 = LOP_IRR(uint32(OP_ORI), uint32(p.To.Reg), uint32(0), uint32(v)) } case 4: /* add/mul $scon,[r1],r2 */ v := c.regoff(&p.From) r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } if r0iszero != 0 /*TypeKind(100016)*/ && p.To.Reg == 0 { c.ctxt.Diag("literal operation on R0\n%v", p) } if int32(int16(v)) != v { log.Fatalf("mishandled instruction %v", p) } o1 = AOP_IRR(c.opirr(p.As), uint32(p.To.Reg), uint32(r), uint32(v)) case 5: /* syscall */ o1 = c.oprrr(p.As) case 6: /* logical op Rb,[Rs,]Ra; no literal */ r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } // AROTL and AROTLW are extended mnemonics, which map to RLDCL and RLWNM. switch p.As { case AROTL: o1 = AOP_MD(OP_RLDCL, uint32(p.To.Reg), uint32(r), uint32(p.From.Reg), uint32(0)) case AROTLW: o1 = OP_RLW(OP_RLWNM, uint32(p.To.Reg), uint32(r), uint32(p.From.Reg), 0, 31) default: if p.As == AOR && p.From.Type == obj.TYPE_CONST && p.From.Offset == 0 { // Compile "OR $0, Rx, Ry" into ori. If Rx == Ry == 0, this is the preferred // hardware no-op. This happens because $0 matches C_REG before C_ZCON. o1 = LOP_IRR(OP_ORI, uint32(p.To.Reg), uint32(r), 0) } else { o1 = LOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(r), uint32(p.From.Reg)) } } case 7: /* mov r, soreg ==> stw o(r) */ r := int(p.To.Reg) v := c.regoff(&p.To) if int32(int16(v)) != v { log.Fatalf("mishandled instruction %v", p) } // Offsets in DS form stores must be a multiple of 4 inst := c.opstore(p.As) if c.opform(inst) == DS_FORM && v&0x3 != 0 { log.Fatalf("invalid offset for DS form load/store %v", p) } o1 = AOP_IRR(inst, uint32(p.From.Reg), uint32(r), uint32(v)) case 8: /* mov soreg, r ==> lbz/lhz/lwz o(r), lbz o(r) + extsb r,r */ r := int(p.From.Reg) v := c.regoff(&p.From) if int32(int16(v)) != v { log.Fatalf("mishandled instruction %v", p) } // Offsets in DS form loads must be a multiple of 4 inst := c.opload(p.As) if c.opform(inst) == DS_FORM && v&0x3 != 0 { log.Fatalf("invalid offset for DS form load/store %v", p) } o1 = AOP_IRR(inst, uint32(p.To.Reg), uint32(r), uint32(v)) // Sign extend MOVB operations. This is ignored for other cases (o.size == 4). o2 = LOP_RRR(OP_EXTSB, uint32(p.To.Reg), uint32(p.To.Reg), 0) case 9: /* RLDC Ra, $sh, $mb, Rb */ sh := uint32(p.RestArgs[0].Addr.Offset) & 0x3F mb := uint32(p.RestArgs[1].Addr.Offset) & 0x3F o1 = AOP_RRR(c.opirr(p.As), uint32(p.From.Reg), uint32(p.To.Reg), (uint32(sh) & 0x1F)) o1 |= (sh & 0x20) >> 4 // sh[5] is placed in bit 1. o1 |= (mb & 0x1F) << 6 // mb[0:4] is placed in bits 6-10. o1 |= (mb & 0x20) // mb[5] is placed in bit 5 case 10: /* sub Ra,[Rb],Rd => subf Rd,Ra,Rb */ r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } o1 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(r)) case 11: /* br/bl bra */ v := int32(0) if p.To.Target() != nil { v = int32(p.To.Target().Pc - p.Pc) if v&03 != 0 { c.ctxt.Diag("odd branch target address\n%v", p) v &^= 03 } if v < -(1<<25) || v >= 1<<24 { c.ctxt.Diag("branch too far\n%v", p) } } o1 = OP_BR(c.opirr(p.As), uint32(v), 0) if p.To.Sym != nil { rel := obj.Addrel(c.cursym) rel.Off = int32(c.pc) rel.Siz = 4 rel.Sym = p.To.Sym v += int32(p.To.Offset) if v&03 != 0 { c.ctxt.Diag("odd branch target address\n%v", p) v &^= 03 } rel.Add = int64(v) rel.Type = objabi.R_CALLPOWER } o2 = NOP // nop, sometimes overwritten by ld r2, 24(r1) when dynamic linking case 13: /* mov[bhwd]{z,} r,r */ // This needs to handle "MOV* $0, Rx". This shows up because $0 also // matches C_REG if r0iszero. This happens because C_REG sorts before C_U16CON // TODO: fix the above behavior and cleanup this exception. if p.From.Type == obj.TYPE_CONST { o1 = LOP_IRR(OP_ADDI, REGZERO, uint32(p.To.Reg), 0) break } if p.To.Type == obj.TYPE_CONST { c.ctxt.Diag("cannot move into constant 0\n%v", p) } switch p.As { case AMOVB: o1 = LOP_RRR(OP_EXTSB, uint32(p.To.Reg), uint32(p.From.Reg), 0) case AMOVBZ: o1 = OP_RLW(OP_RLWINM, uint32(p.To.Reg), uint32(p.From.Reg), 0, 24, 31) case AMOVH: o1 = LOP_RRR(OP_EXTSH, uint32(p.To.Reg), uint32(p.From.Reg), 0) case AMOVHZ: o1 = OP_RLW(OP_RLWINM, uint32(p.To.Reg), uint32(p.From.Reg), 0, 16, 31) case AMOVW: o1 = LOP_RRR(OP_EXTSW, uint32(p.To.Reg), uint32(p.From.Reg), 0) case AMOVWZ: o1 = OP_RLW(OP_RLDIC, uint32(p.To.Reg), uint32(p.From.Reg), 0, 0, 0) | 1<<5 /* MB=32 */ case AMOVD: o1 = LOP_RRR(OP_OR, uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.From.Reg)) default: c.ctxt.Diag("internal: bad register move/truncation\n%v", p) } case 14: /* rldc[lr] Rb,Rs,$mask,Ra -- left, right give different masks */ r := uint32(p.Reg) if r == 0 { r = uint32(p.To.Reg) } d := c.vregoff(p.GetFrom3()) switch p.As { // These opcodes expect a mask operand that has to be converted into the // appropriate operand. The way these were defined, not all valid masks are possible. // Left here for compatibility in case they were used or generated. case ARLDCL, ARLDCLCC: mb, me, valid := decodeMask64(d) if me != 63 || !valid { c.ctxt.Diag("invalid mask for rotate: %x (end != bit 63)\n%v", uint64(d), p) } o1 = AOP_MDS(c.oprrr(p.As), uint32(p.To.Reg), r, uint32(p.From.Reg), mb) case ARLDCR, ARLDCRCC: mb, me, valid := decodeMask64(d) if mb != 0 || !valid { c.ctxt.Diag("invalid mask for rotate: %x (start != 0)\n%v", uint64(d), p) } o1 = AOP_MDS(c.oprrr(p.As), uint32(p.To.Reg), r, uint32(p.From.Reg), me) // These opcodes use a shift count like the ppc64 asm, no mask conversion done case ARLDICR, ARLDICRCC: me := uint32(d) sh := c.regoff(&p.From) if me < 0 || me > 63 || sh > 63 { c.ctxt.Diag("Invalid me or sh for RLDICR: %x %x\n%v", int(d), sh, p) } o1 = AOP_MD(c.oprrr(p.As), uint32(p.To.Reg), r, uint32(sh), me) case ARLDICL, ARLDICLCC, ARLDIC, ARLDICCC: mb := uint32(d) sh := c.regoff(&p.From) if mb < 0 || mb > 63 || sh > 63 { c.ctxt.Diag("Invalid mb or sh for RLDIC, RLDICL: %x %x\n%v", mb, sh, p) } o1 = AOP_MD(c.oprrr(p.As), uint32(p.To.Reg), r, uint32(sh), mb) case ACLRLSLDI: // This is an extended mnemonic defined in the ISA section C.8.1 // clrlsldi ra,rs,b,n --> rldic ra,rs,n,b-n // It maps onto RLDIC so is directly generated here based on the operands from // the clrlsldi. n := int32(d) b := c.regoff(&p.From) if n > b || b > 63 { c.ctxt.Diag("Invalid n or b for CLRLSLDI: %x %x\n%v", n, b, p) } o1 = AOP_MD(OP_RLDIC, uint32(p.To.Reg), uint32(r), uint32(n), uint32(b)-uint32(n)) default: c.ctxt.Diag("unexpected op in rldc case\n%v", p) } case 16: /* bc bo,bi,bra */ a := 0 r := int(p.Reg) if p.From.Type == obj.TYPE_CONST { a = int(c.regoff(&p.From)) } else if p.From.Type == obj.TYPE_REG { if r != 0 { c.ctxt.Diag("unexpected register setting for branch with CR: %d\n", r) } // BI values for the CR switch p.From.Reg { case REG_CR0: r = BI_CR0 case REG_CR1: r = BI_CR1 case REG_CR2: r = BI_CR2 case REG_CR3: r = BI_CR3 case REG_CR4: r = BI_CR4 case REG_CR5: r = BI_CR5 case REG_CR6: r = BI_CR6 case REG_CR7: r = BI_CR7 default: c.ctxt.Diag("unrecognized register: expecting CR\n") } } v := int32(0) if p.To.Target() != nil { v = int32(p.To.Target().Pc - p.Pc) } if v&03 != 0 { c.ctxt.Diag("odd branch target address\n%v", p) v &^= 03 } if v < -(1<<16) || v >= 1<<15 { c.ctxt.Diag("branch too far\n%v", p) } o1 = OP_BC(c.opirr(p.As), uint32(a), uint32(r), uint32(v), 0) case 18: /* br/bl (lr/ctr); bc/bcl bo,bi,(lr/ctr) */ var v int32 var bh uint32 = 0 if p.As == ABC || p.As == ABCL { v = c.regoff(&p.From) & 31 } else { v = 20 /* unconditional */ } r := int(p.Reg) if r == 0 { r = 0 } switch oclass(&p.To) { case C_CTR: o1 = OPVCC(19, 528, 0, 0) case C_LR: o1 = OPVCC(19, 16, 0, 0) default: c.ctxt.Diag("bad optab entry (18): %d\n%v", p.To.Class, p) v = 0 } // Insert optional branch hint for bclr[l]/bcctr[l] if p.From3Type() != obj.TYPE_NONE { bh = uint32(p.GetFrom3().Offset) if bh == 2 || bh > 3 { log.Fatalf("BH must be 0,1,3 for %v", p) } o1 |= bh << 11 } if p.As == ABL || p.As == ABCL { o1 |= 1 } o1 = OP_BCR(o1, uint32(v), uint32(r)) case 19: /* mov $lcon,r ==> cau+or */ d := c.vregoff(&p.From) if o.ispfx { o1, o2 = pfxadd(p.To.Reg, REG_R0, PFX_R_ABS, d) } else { o1 = loadu32(int(p.To.Reg), d) o2 = LOP_IRR(OP_ORI, uint32(p.To.Reg), uint32(p.To.Reg), uint32(int32(d))) } case 20: /* add $ucon,,r | addis $addcon,r,r */ v := c.regoff(&p.From) r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } o1 = AOP_IRR(c.opirr(p.As), uint32(p.To.Reg), uint32(r), uint32(v)) case 21: /* or $u32con,rx[,ry] => oris + ori (similar for xor) */ var opu, opl uint32 r := uint32(p.Reg) if r == 0 { r = uint32(p.To.Reg) } switch p.As { case AOR: opu, opl = OP_ORIS, OP_ORI case AXOR: opu, opl = OP_XORIS, OP_XORI default: c.ctxt.Diag("unhandled opcode.\n%v", p) } o1 = LOP_IRR(opu, uint32(p.To.Reg), r, uint32(p.From.Offset>>16)) o2 = LOP_IRR(opl, uint32(p.To.Reg), uint32(p.To.Reg), uint32(p.From.Offset)&0xFFFF) case 22: /* add $lcon/$andcon,r1,r2 ==> oris+ori+add/ori+add, add $s34con,r1 ==> addis+ori+slw+ori+add */ if p.To.Reg == REGTMP || p.Reg == REGTMP { c.ctxt.Diag("can't synthesize large constant\n%v", p) } d := c.vregoff(&p.From) r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } if p.From.Sym != nil { c.ctxt.Diag("%v is not supported", p) } if o.ispfx { o1, o2 = pfxadd(int16(p.To.Reg), int16(r), PFX_R_ABS, d) } else if o.size == 8 { o1 = LOP_IRR(OP_ORI, REGTMP, REGZERO, uint32(int32(d))) // tmp = uint16(d) o2 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), REGTMP, uint32(r)) // to = tmp + from } else if o.size == 12 { // Note, o1 is ADDIS if d is negative, ORIS otherwise. o1 = loadu32(REGTMP, d) // tmp = d & 0xFFFF0000 o2 = loadl16(REGTMP, d) // tmp |= d & 0xFFFF o3 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), REGTMP, uint32(r)) // to = from + tmp } else { // For backwards compatibility with GOPPC64 < 10, generate 34b constants in register. o1 = LOP_IRR(OP_ADDIS, REGZERO, REGTMP, uint32(d>>32)) // tmp = sign_extend((d>>32)&0xFFFF0000) o2 = loadl16(REGTMP, int64(d>>16)) // tmp |= (d>>16)&0xFFFF o3 = AOP_MD(OP_RLDICR, REGTMP, REGTMP, 16, 63-16) // tmp <<= 16 o4 = loadl16(REGTMP, int64(uint16(d))) // tmp |= d&0xFFFF o5 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), REGTMP, uint32(r)) } case 23: /* and $lcon/$addcon,r1,r2 ==> oris+ori+and/addi+and */ if p.To.Reg == REGTMP || p.Reg == REGTMP { c.ctxt.Diag("can't synthesize large constant\n%v", p) } d := c.vregoff(&p.From) r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } // With S16CON operand, generate 2 instructions using ADDI for signed value, // with 32CON operand generate 3 instructions. if o.size == 8 { o1 = LOP_IRR(OP_ADDI, REGZERO, REGTMP, uint32(int32(d))) o2 = LOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), REGTMP, uint32(r)) } else { o1 = loadu32(REGTMP, d) o2 = loadl16(REGTMP, d) o3 = LOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), REGTMP, uint32(r)) } if p.From.Sym != nil { c.ctxt.Diag("%v is not supported", p) } case 24: /* lfd fA,float64(0) -> xxlxor xsA,xsaA,xsaA + fneg for -0 */ o1 = AOP_XX3I(c.oprrr(AXXLXOR), uint32(p.To.Reg), uint32(p.To.Reg), uint32(p.To.Reg), uint32(0)) // This is needed for -0. if o.size == 8 { o2 = AOP_RRR(c.oprrr(AFNEG), uint32(p.To.Reg), 0, uint32(p.To.Reg)) } case 25: /* sld[.] $sh,rS,rA -> rldicr[.] $sh,rS,mask(0,63-sh),rA; srd[.] -> rldicl */ v := c.regoff(&p.From) if v < 0 { v = 0 } else if v > 63 { v = 63 } r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } var a int op := uint32(0) switch p.As { case ASLD, ASLDCC: a = int(63 - v) op = OP_RLDICR case ASRD, ASRDCC: a = int(v) v = 64 - v op = OP_RLDICL case AROTL: a = int(0) op = OP_RLDICL case AEXTSWSLI, AEXTSWSLICC: a = int(v) default: c.ctxt.Diag("unexpected op in sldi case\n%v", p) a = 0 o1 = 0 } if p.As == AEXTSWSLI || p.As == AEXTSWSLICC { o1 = AOP_EXTSWSLI(OP_EXTSWSLI, uint32(r), uint32(p.To.Reg), uint32(v)) } else { o1 = AOP_MD(op, uint32(p.To.Reg), uint32(r), uint32(v), uint32(a)) } if p.As == ASLDCC || p.As == ASRDCC || p.As == AEXTSWSLICC { o1 |= 1 // Set the condition code bit } case 26: /* mov $lsext/auto/oreg,,r2 ==> addis+addi */ v := c.vregoff(&p.From) r := int(p.From.Reg) var rel *obj.Reloc switch p.From.Name { case obj.NAME_EXTERN, obj.NAME_STATIC: // Load a 32 bit constant, or relocation depending on if a symbol is attached o1, o2, rel = c.symbolAccess(p.From.Sym, v, p.To.Reg, OP_ADDI, true) default: // Add a 32 bit offset to a register. o1 = AOP_IRR(OP_ADDIS, uint32(p.To.Reg), uint32(r), uint32(high16adjusted(int32(v)))) o2 = AOP_IRR(OP_ADDI, uint32(p.To.Reg), uint32(p.To.Reg), uint32(v)) } if o.ispfx { if rel == nil { o1, o2 = pfxadd(int16(p.To.Reg), int16(r), PFX_R_ABS, v) } else { o1, o2 = pfxadd(int16(p.To.Reg), REG_R0, PFX_R_PCREL, 0) rel.Type = objabi.R_ADDRPOWER_PCREL34 } } case 27: /* subc ra,$simm,rd => subfic rd,ra,$simm */ v := c.regoff(p.GetFrom3()) r := int(p.From.Reg) o1 = AOP_IRR(c.opirr(p.As), uint32(p.To.Reg), uint32(r), uint32(v)) case 28: /* subc r1,$lcon,r2 ==> cau+or+subfc */ if p.To.Reg == REGTMP || p.From.Reg == REGTMP { c.ctxt.Diag("can't synthesize large constant\n%v", p) } v := c.vregoff(p.GetFrom3()) o1 = AOP_IRR(OP_ADDIS, REGTMP, REGZERO, uint32(v)>>16) o2 = loadl16(REGTMP, v) o3 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), REGTMP) if p.From.Sym != nil { c.ctxt.Diag("%v is not supported", p) } case 29: /* rldic[lr]? $sh,s,$mask,a -- left, right, plain give different masks */ sh := uint32(c.regoff(&p.From)) d := c.vregoff(p.GetFrom3()) mb, me, valid := decodeMask64(d) var a uint32 switch p.As { case ARLDC, ARLDCCC: a = mb if me != (63-sh) || !valid { c.ctxt.Diag("invalid mask for shift: %016x (mb=%d,me=%d) (shift %d)\n%v", uint64(d), mb, me, sh, p) } case ARLDCL, ARLDCLCC: a = mb if mb != 63 || !valid { c.ctxt.Diag("invalid mask for shift: %016x (mb=%d,me=%d) (shift %d)\n%v", uint64(d), mb, me, sh, p) } case ARLDCR, ARLDCRCC: a = me if mb != 0 || !valid { c.ctxt.Diag("invalid mask for shift: %016x (mb=%d,me=%d) (shift %d)\n%v", uint64(d), mb, me, sh, p) } default: c.ctxt.Diag("unexpected op in rldic case\n%v", p) } o1 = AOP_MD(c.opirr(p.As), uint32(p.To.Reg), uint32(p.Reg), sh, a) case 30: /* rldimi $sh,s,$mask,a */ sh := uint32(c.regoff(&p.From)) d := c.vregoff(p.GetFrom3()) // Original opcodes had mask operands which had to be converted to a shift count as expected by // the ppc64 asm. switch p.As { case ARLDMI, ARLDMICC: mb, me, valid := decodeMask64(d) if me != (63-sh) || !valid { c.ctxt.Diag("invalid mask for shift: %x %x (shift %d)\n%v", uint64(d), me, sh, p) } o1 = AOP_MD(c.opirr(p.As), uint32(p.To.Reg), uint32(p.Reg), sh, mb) // Opcodes with shift count operands. case ARLDIMI, ARLDIMICC: o1 = AOP_MD(c.opirr(p.As), uint32(p.To.Reg), uint32(p.Reg), sh, uint32(d)) } case 31: /* dword */ d := c.vregoff(&p.From) if c.ctxt.Arch.ByteOrder == binary.BigEndian { o1 = uint32(d >> 32) o2 = uint32(d) } else { o1 = uint32(d) o2 = uint32(d >> 32) } if p.From.Sym != nil { rel := obj.Addrel(c.cursym) rel.Off = int32(c.pc) rel.Siz = 8 rel.Sym = p.From.Sym rel.Add = p.From.Offset rel.Type = objabi.R_ADDR o2 = 0 o1 = o2 } case 32: /* fmul frc,fra,frd */ r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } o1 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(r), 0) | (uint32(p.From.Reg)&31)<<6 case 33: /* fabs [frb,]frd; fmr. frb,frd */ r := int(p.From.Reg) if oclass(&p.From) == C_NONE { r = int(p.To.Reg) } o1 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), 0, uint32(r)) case 34: /* FMADDx fra,frb,frc,frt (t=a*c±b) */ o1 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg)) | (uint32(p.GetFrom3().Reg)&31)<<6 case 35: /* mov r,lext/lauto/loreg ==> cau $(v>>16),sb,r'; store o(r') */ v := c.regoff(&p.To) r := int(p.To.Reg) // Offsets in DS form stores must be a multiple of 4 if o.ispfx { o1, o2 = pfxstore(p.As, p.From.Reg, int16(r), PFX_R_ABS) o1 |= uint32((v >> 16) & 0x3FFFF) o2 |= uint32(v & 0xFFFF) } else { inst := c.opstore(p.As) if c.opform(inst) == DS_FORM && v&0x3 != 0 { log.Fatalf("invalid offset for DS form load/store %v", p) } o1 = AOP_IRR(OP_ADDIS, REGTMP, uint32(r), uint32(high16adjusted(v))) o2 = AOP_IRR(inst, uint32(p.From.Reg), REGTMP, uint32(v)) } case 36: /* mov b/bz/h/hz lext/lauto/lreg,r ==> lbz+extsb/lbz/lha/lhz etc */ v := c.regoff(&p.From) r := int(p.From.Reg) if o.ispfx { o1, o2 = pfxload(p.As, p.To.Reg, int16(r), PFX_R_ABS) o1 |= uint32((v >> 16) & 0x3FFFF) o2 |= uint32(v & 0xFFFF) } else { if o.a6 == C_REG { // Reuse the base register when loading a GPR (C_REG) to avoid // using REGTMP (R31) when possible. o1 = AOP_IRR(OP_ADDIS, uint32(p.To.Reg), uint32(r), uint32(high16adjusted(v))) o2 = AOP_IRR(c.opload(p.As), uint32(p.To.Reg), uint32(p.To.Reg), uint32(v)) } else { o1 = AOP_IRR(OP_ADDIS, uint32(REGTMP), uint32(r), uint32(high16adjusted(v))) o2 = AOP_IRR(c.opload(p.As), uint32(p.To.Reg), uint32(REGTMP), uint32(v)) } } // Sign extend MOVB if needed o3 = LOP_RRR(OP_EXTSB, uint32(p.To.Reg), uint32(p.To.Reg), 0) case 40: /* word */ o1 = uint32(c.regoff(&p.From)) case 41: /* stswi */ if p.To.Type == obj.TYPE_MEM && p.To.Index == 0 && p.To.Offset != 0 { c.ctxt.Diag("Invalid addressing mode used in index type instruction: %v", p.As) } o1 = AOP_RRR(c.opirr(p.As), uint32(p.From.Reg), uint32(p.To.Reg), 0) | (uint32(c.regoff(p.GetFrom3()))&0x7F)<<11 case 42: /* lswi */ if p.From.Type == obj.TYPE_MEM && p.From.Index == 0 && p.From.Offset != 0 { c.ctxt.Diag("Invalid addressing mode used in index type instruction: %v", p.As) } o1 = AOP_RRR(c.opirr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), 0) | (uint32(c.regoff(p.GetFrom3()))&0x7F)<<11 case 43: /* data cache instructions: op (Ra+[Rb]), [th|l] */ /* TH field for dcbt/dcbtst: */ /* 0 = Block access - program will soon access EA. */ /* 8-15 = Stream access - sequence of access (data stream). See section 4.3.2 of the ISA for details. */ /* 16 = Block access - program will soon make a transient access to EA. */ /* 17 = Block access - program will not access EA for a long time. */ /* L field for dcbf: */ /* 0 = invalidates the block containing EA in all processors. */ /* 1 = same as 0, but with limited scope (i.e. block in the current processor will not be reused soon). */ /* 3 = same as 1, but with even more limited scope (i.e. block in the current processor primary cache will not be reused soon). */ if p.To.Type == obj.TYPE_NONE { o1 = AOP_RRR(c.oprrr(p.As), 0, uint32(p.From.Index), uint32(p.From.Reg)) } else { th := c.regoff(&p.To) o1 = AOP_RRR(c.oprrr(p.As), uint32(th), uint32(p.From.Index), uint32(p.From.Reg)) } case 44: /* indexed store */ o1 = AOP_RRR(c.opstorex(p.As), uint32(p.From.Reg), uint32(p.To.Index), uint32(p.To.Reg)) case 45: /* indexed load */ switch p.As { /* The assembler accepts a 4-operand l*arx instruction. The fourth operand is an Exclusive Access Hint (EH) */ /* The EH field can be used as a lock acquire/release hint as follows: */ /* 0 = Atomic Update (fetch-and-operate or similar algorithm) */ /* 1 = Exclusive Access (lock acquire and release) */ case ALBAR, ALHAR, ALWAR, ALDAR: if p.From3Type() != obj.TYPE_NONE { eh := int(c.regoff(p.GetFrom3())) if eh > 1 { c.ctxt.Diag("illegal EH field\n%v", p) } o1 = AOP_RRRI(c.oploadx(p.As), uint32(p.To.Reg), uint32(p.From.Index), uint32(p.From.Reg), uint32(eh)) } else { o1 = AOP_RRR(c.oploadx(p.As), uint32(p.To.Reg), uint32(p.From.Index), uint32(p.From.Reg)) } default: o1 = AOP_RRR(c.oploadx(p.As), uint32(p.To.Reg), uint32(p.From.Index), uint32(p.From.Reg)) } case 46: /* plain op */ o1 = c.oprrr(p.As) case 47: /* op Ra, Rd; also op [Ra,] Rd */ r := int(p.From.Reg) if r == 0 { r = int(p.To.Reg) } o1 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(r), 0) case 48: /* op Rs, Ra */ r := int(p.From.Reg) if r == 0 { r = int(p.To.Reg) } o1 = LOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(r), 0) case 49: /* op Rb; op $n, Rb */ if p.From.Type != obj.TYPE_REG { /* tlbie $L, rB */ v := c.regoff(&p.From) & 1 o1 = AOP_RRR(c.oprrr(p.As), 0, 0, uint32(p.To.Reg)) | uint32(v)<<21 } else { o1 = AOP_RRR(c.oprrr(p.As), 0, 0, uint32(p.From.Reg)) } case 50: /* rem[u] r1[,r2],r3 */ r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } v := c.oprrr(p.As) t := v & (1<<10 | 1) /* OE|Rc */ o1 = AOP_RRR(v&^t, REGTMP, uint32(r), uint32(p.From.Reg)) o2 = AOP_RRR(OP_MULLW, REGTMP, REGTMP, uint32(p.From.Reg)) o3 = AOP_RRR(OP_SUBF|t, uint32(p.To.Reg), REGTMP, uint32(r)) if p.As == AREMU { o4 = o3 /* Clear top 32 bits */ o3 = OP_RLW(OP_RLDIC, REGTMP, REGTMP, 0, 0, 0) | 1<<5 } case 51: /* remd[u] r1[,r2],r3 */ r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } v := c.oprrr(p.As) t := v & (1<<10 | 1) /* OE|Rc */ o1 = AOP_RRR(v&^t, REGTMP, uint32(r), uint32(p.From.Reg)) o2 = AOP_RRR(OP_MULLD, REGTMP, REGTMP, uint32(p.From.Reg)) o3 = AOP_RRR(OP_SUBF|t, uint32(p.To.Reg), REGTMP, uint32(r)) /* cases 50,51: removed; can be reused. */ /* cases 50,51: removed; can be reused. */ case 52: /* mtfsbNx cr(n) */ v := c.regoff(&p.From) & 31 o1 = AOP_RRR(c.oprrr(p.As), uint32(v), 0, 0) case 53: /* mffsX ,fr1 */ o1 = AOP_RRR(OP_MFFS, uint32(p.To.Reg), 0, 0) case 55: /* op Rb, Rd */ o1 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), 0, uint32(p.From.Reg)) case 56: /* sra $sh,[s,]a; srd $sh,[s,]a */ v := c.regoff(&p.From) r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } o1 = AOP_RRR(c.opirr(p.As), uint32(r), uint32(p.To.Reg), uint32(v)&31) if (p.As == ASRAD || p.As == ASRADCC) && (v&0x20 != 0) { o1 |= 1 << 1 /* mb[5] */ } case 57: /* slw $sh,[s,]a -> rlwinm ... */ v := c.regoff(&p.From) r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } /* * Let user (gs) shoot himself in the foot. * qc has already complained. * if(v < 0 || v > 31) ctxt->diag("illegal shift %ld\n%v", v, p); */ if v < 0 { v = 0 } else if v > 32 { v = 32 } var mask [2]uint8 switch p.As { case AROTLW: mask[0], mask[1] = 0, 31 case ASRW, ASRWCC: mask[0], mask[1] = uint8(v), 31 v = 32 - v default: mask[0], mask[1] = 0, uint8(31-v) } o1 = OP_RLW(OP_RLWINM, uint32(p.To.Reg), uint32(r), uint32(v), uint32(mask[0]), uint32(mask[1])) if p.As == ASLWCC || p.As == ASRWCC { o1 |= 1 // set the condition code } case 58: /* logical $andcon,[s],a */ v := c.regoff(&p.From) r := int(p.Reg) if r == 0 { r = int(p.To.Reg) } o1 = LOP_IRR(c.opirr(p.As), uint32(p.To.Reg), uint32(r), uint32(v)) case 60: /* tw to,a,b */ r := int(c.regoff(&p.From) & 31) o1 = AOP_RRR(c.oprrr(p.As), uint32(r), uint32(p.Reg), uint32(p.To.Reg)) case 61: /* tw to,a,$simm */ r := int(c.regoff(&p.From) & 31) v := c.regoff(&p.To) o1 = AOP_IRR(c.opirr(p.As), uint32(r), uint32(p.Reg), uint32(v)) case 62: /* clrlslwi $sh,s,$mask,a */ v := c.regoff(&p.From) n := c.regoff(p.GetFrom3()) // This is an extended mnemonic described in the ISA C.8.2 // clrlslwi ra,rs,b,n -> rlwinm ra,rs,n,b-n,31-n // It maps onto rlwinm which is directly generated here. if n > v || v >= 32 { c.ctxt.Diag("Invalid n or b for CLRLSLWI: %x %x\n%v", v, n, p) } o1 = OP_RLW(OP_RLWINM, uint32(p.To.Reg), uint32(p.Reg), uint32(n), uint32(v-n), uint32(31-n)) case 63: /* rlwimi/rlwnm/rlwinm [$sh,b],s,[$mask or mb,me],a*/ var mb, me uint32 if len(p.RestArgs) == 1 { // Mask needs decomposed into mb and me. var valid bool // Note, optab rules ensure $mask is a 32b constant. mb, me, valid = decodeMask32(uint32(p.RestArgs[0].Addr.Offset)) if !valid { c.ctxt.Diag("cannot generate mask #%x\n%v", uint64(p.RestArgs[0].Addr.Offset), p) } } else { // Otherwise, mask is already passed as mb and me in RestArgs. mb, me = uint32(p.RestArgs[0].Addr.Offset), uint32(p.RestArgs[1].Addr.Offset) } if p.From.Type == obj.TYPE_CONST { o1 = OP_RLW(c.opirr(p.As), uint32(p.To.Reg), uint32(p.Reg), uint32(p.From.Offset), mb, me) } else { o1 = OP_RLW(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.Reg), uint32(p.From.Reg), mb, me) } case 64: /* mtfsf fr[, $m] {,fpcsr} */ var v int32 if p.From3Type() != obj.TYPE_NONE { v = c.regoff(p.GetFrom3()) & 255 } else { v = 255 } o1 = OP_MTFSF | uint32(v)<<17 | uint32(p.From.Reg)<<11 case 65: /* MOVFL $imm,FPSCR(n) => mtfsfi crfd,imm */ if p.To.Reg == 0 { c.ctxt.Diag("must specify FPSCR(n)\n%v", p) } o1 = OP_MTFSFI | (uint32(p.To.Reg)&15)<<23 | (uint32(c.regoff(&p.From))&31)<<12 case 66: /* mov spr,r1; mov r1,spr */ var r int var v int32 if REG_R0 <= p.From.Reg && p.From.Reg <= REG_R31 { r = int(p.From.Reg) v = int32(p.To.Reg) o1 = OPVCC(31, 467, 0, 0) /* mtspr */ } else { r = int(p.To.Reg) v = int32(p.From.Reg) o1 = OPVCC(31, 339, 0, 0) /* mfspr */ } o1 = AOP_RRR(o1, uint32(r), 0, 0) | (uint32(v)&0x1f)<<16 | ((uint32(v)>>5)&0x1f)<<11 case 67: /* mcrf crfD,crfS */ if p.From.Reg == REG_CR || p.To.Reg == REG_CR { c.ctxt.Diag("CR argument must be a conditional register field (CR0-CR7)\n%v", p) } o1 = AOP_RRR(OP_MCRF, ((uint32(p.To.Reg) & 7) << 2), ((uint32(p.From.Reg) & 7) << 2), 0) case 68: /* mfcr rD; mfocrf CRM,rD */ o1 = AOP_RRR(OP_MFCR, uint32(p.To.Reg), 0, 0) /* form, whole register */ if p.From.Reg != REG_CR { v := uint32(1) << uint(7-(p.From.Reg&7)) /* CR(n) */ o1 |= 1<<20 | v<<12 /* new form, mfocrf */ } case 69: /* mtcrf CRM,rS, mtocrf CRx,rS */ var v uint32 if p.To.Reg == REG_CR { v = 0xff } else if p.To.Offset != 0 { // MOVFL gpr, constant v = uint32(p.To.Offset) } else { // p.To.Reg == REG_CRx v = 1 << uint(7-(p.To.Reg&7)) } // Use mtocrf form if only one CR field moved. if bits.OnesCount32(v) == 1 { v |= 1 << 8 } o1 = AOP_RRR(OP_MTCRF, uint32(p.From.Reg), 0, 0) | uint32(v)<<12 case 70: /* cmp* r,r,cr or cmp*i r,i,cr or fcmp f,f,cr or cmpeqb r,r */ r := uint32(p.Reg&7) << 2 if p.To.Type == obj.TYPE_CONST { o1 = AOP_IRR(c.opirr(p.As), r, uint32(p.From.Reg), uint32(uint16(p.To.Offset))) } else { o1 = AOP_RRR(c.oprrr(p.As), r, uint32(p.From.Reg), uint32(p.To.Reg)) } case 72: /* slbmte (Rb+Rs -> slb[Rb]) -> Rs, Rb */ o1 = AOP_RRR(c.oprrr(p.As), uint32(p.From.Reg), 0, uint32(p.To.Reg)) case 73: /* mcrfs crfD,crfS */ if p.From.Type != obj.TYPE_REG || p.From.Reg != REG_FPSCR || p.To.Type != obj.TYPE_REG || p.To.Reg < REG_CR0 || REG_CR7 < p.To.Reg { c.ctxt.Diag("illegal FPSCR/CR field number\n%v", p) } o1 = AOP_RRR(OP_MCRFS, ((uint32(p.To.Reg) & 7) << 2), ((0 & 7) << 2), 0) case 77: /* syscall $scon, syscall Rx */ if p.From.Type == obj.TYPE_CONST { if p.From.Offset > BIG || p.From.Offset < -BIG { c.ctxt.Diag("illegal syscall, sysnum too large: %v", p) } o1 = AOP_IRR(OP_ADDI, REGZERO, REGZERO, uint32(p.From.Offset)) } else if p.From.Type == obj.TYPE_REG { o1 = LOP_RRR(OP_OR, REGZERO, uint32(p.From.Reg), uint32(p.From.Reg)) } else { c.ctxt.Diag("illegal syscall: %v", p) o1 = 0x7fe00008 // trap always } o2 = c.oprrr(p.As) o3 = AOP_RRR(c.oprrr(AXOR), REGZERO, REGZERO, REGZERO) // XOR R0, R0 case 78: /* undef */ o1 = 0 /* "An instruction consisting entirely of binary 0s is guaranteed always to be an illegal instruction." */ /* relocation operations */ case 74: var rel *obj.Reloc v := c.vregoff(&p.To) // Offsets in DS form stores must be a multiple of 4 inst := c.opstore(p.As) // Can't reuse base for store instructions. o1, o2, rel = c.symbolAccess(p.To.Sym, v, p.From.Reg, inst, false) // Rewrite as a prefixed store if supported. if o.ispfx { o1, o2 = pfxstore(p.As, p.From.Reg, REG_R0, PFX_R_PCREL) rel.Type = objabi.R_ADDRPOWER_PCREL34 } else if c.opform(inst) == DS_FORM && v&0x3 != 0 { log.Fatalf("invalid offset for DS form load/store %v", p) } case 75: // 32 bit offset symbol loads (got/toc/addr) var rel *obj.Reloc v := p.From.Offset // Offsets in DS form loads must be a multiple of 4 inst := c.opload(p.As) switch p.From.Name { case obj.NAME_GOTREF, obj.NAME_TOCREF: if v != 0 { c.ctxt.Diag("invalid offset for GOT/TOC access %v", p) } o1 = AOP_IRR(OP_ADDIS, uint32(p.To.Reg), REG_R2, 0) o2 = AOP_IRR(inst, uint32(p.To.Reg), uint32(p.To.Reg), 0) rel = obj.Addrel(c.cursym) rel.Off = int32(c.pc) rel.Siz = 8 rel.Sym = p.From.Sym switch p.From.Name { case obj.NAME_GOTREF: rel.Type = objabi.R_ADDRPOWER_GOT case obj.NAME_TOCREF: rel.Type = objabi.R_ADDRPOWER_TOCREL_DS } default: reuseBaseReg := o.a6 == C_REG // Reuse To.Reg as base register if it is a GPR. o1, o2, rel = c.symbolAccess(p.From.Sym, v, p.To.Reg, inst, reuseBaseReg) } // Convert to prefixed forms if supported. if o.ispfx { switch rel.Type { case objabi.R_ADDRPOWER, objabi.R_ADDRPOWER_DS, objabi.R_ADDRPOWER_TOCREL, objabi.R_ADDRPOWER_TOCREL_DS: o1, o2 = pfxload(p.As, p.To.Reg, REG_R0, PFX_R_PCREL) rel.Type = objabi.R_ADDRPOWER_PCREL34 case objabi.R_POWER_TLS_IE: o1, o2 = pfxload(p.As, p.To.Reg, REG_R0, PFX_R_PCREL) rel.Type = objabi.R_POWER_TLS_IE_PCREL34 case objabi.R_ADDRPOWER_GOT: o1, o2 = pfxload(p.As, p.To.Reg, REG_R0, PFX_R_PCREL) rel.Type = objabi.R_ADDRPOWER_GOT_PCREL34 default: // We've failed to convert a TOC-relative relocation to a PC-relative one. log.Fatalf("Unable convert TOC-relative relocation %v to PC-relative", rel.Type) } } else if c.opform(inst) == DS_FORM && v&0x3 != 0 { log.Fatalf("invalid offset for DS form load/store %v", p) } o3 = LOP_RRR(OP_EXTSB, uint32(p.To.Reg), uint32(p.To.Reg), 0) case 79: if p.From.Offset != 0 { c.ctxt.Diag("invalid offset against tls var %v", p) } rel := obj.Addrel(c.cursym) rel.Off = int32(c.pc) rel.Siz = 8 rel.Sym = p.From.Sym if !o.ispfx { o1 = AOP_IRR(OP_ADDIS, uint32(p.To.Reg), REG_R13, 0) o2 = AOP_IRR(OP_ADDI, uint32(p.To.Reg), uint32(p.To.Reg), 0) rel.Type = objabi.R_POWER_TLS_LE } else { o1, o2 = pfxadd(p.To.Reg, REG_R13, PFX_R_ABS, 0) rel.Type = objabi.R_POWER_TLS_LE_TPREL34 } case 80: if p.From.Offset != 0 { c.ctxt.Diag("invalid offset against tls var %v", p) } rel := obj.Addrel(c.cursym) rel.Off = int32(c.pc) rel.Siz = 8 rel.Sym = p.From.Sym rel.Type = objabi.R_POWER_TLS_IE if !o.ispfx { o1 = AOP_IRR(OP_ADDIS, uint32(p.To.Reg), REG_R2, 0) o2 = AOP_IRR(c.opload(AMOVD), uint32(p.To.Reg), uint32(p.To.Reg), 0) } else { o1, o2 = pfxload(p.As, p.To.Reg, REG_R0, PFX_R_PCREL) rel.Type = objabi.R_POWER_TLS_IE_PCREL34 } o3 = AOP_RRR(OP_ADD, uint32(p.To.Reg), uint32(p.To.Reg), REG_R13) rel = obj.Addrel(c.cursym) rel.Off = int32(c.pc) + 8 rel.Siz = 4 rel.Sym = p.From.Sym rel.Type = objabi.R_POWER_TLS case 82: /* vector instructions, VX-form and VC-form */ if p.From.Type == obj.TYPE_REG { /* reg reg none OR reg reg reg */ /* 3-register operand order: VRA, VRB, VRT */ /* 2-register operand order: VRA, VRT */ o1 = AOP_RRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg)) } else if p.From3Type() == obj.TYPE_CONST { /* imm imm reg reg */ /* operand order: SIX, VRA, ST, VRT */ six := int(c.regoff(&p.From)) st := int(c.regoff(p.GetFrom3())) o1 = AOP_IIRR(c.opiirr(p.As), uint32(p.To.Reg), uint32(p.Reg), uint32(st), uint32(six)) } else if p.From3Type() == obj.TYPE_NONE && p.Reg != 0 { /* imm reg reg */ /* operand order: UIM, VRB, VRT */ uim := int(c.regoff(&p.From)) o1 = AOP_VIRR(c.opirr(p.As), uint32(p.To.Reg), uint32(p.Reg), uint32(uim)) } else { /* imm reg */ /* operand order: SIM, VRT */ sim := int(c.regoff(&p.From)) o1 = AOP_IR(c.opirr(p.As), uint32(p.To.Reg), uint32(sim)) } case 83: /* vector instructions, VA-form */ if p.From.Type == obj.TYPE_REG { /* reg reg reg reg */ /* 4-register operand order: VRA, VRB, VRC, VRT */ o1 = AOP_RRRR(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg), uint32(p.GetFrom3().Reg)) } else if p.From.Type == obj.TYPE_CONST { /* imm reg reg reg */ /* operand order: SHB, VRA, VRB, VRT */ shb := int(c.regoff(&p.From)) o1 = AOP_IRRR(c.opirrr(p.As), uint32(p.To.Reg), uint32(p.Reg), uint32(p.GetFrom3().Reg), uint32(shb)) } case 84: // ISEL BC,RA,RB,RT -> isel rt,ra,rb,bc bc := c.vregoff(&p.From) if o.a1 == C_CRBIT { // CR bit is encoded as a register, not a constant. bc = int64(p.From.Reg) } // rt = To.Reg, ra = p.Reg, rb = p.From3.Reg o1 = AOP_ISEL(OP_ISEL, uint32(p.To.Reg), uint32(p.Reg), uint32(p.GetFrom3().Reg), uint32(bc)) case 85: /* vector instructions, VX-form */ /* reg none reg */ /* 2-register operand order: VRB, VRT */ o1 = AOP_RR(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg)) case 86: /* VSX indexed store, XX1-form */ /* reg reg reg */ /* 3-register operand order: XT, (RB)(RA*1) */ o1 = AOP_XX1(c.opstorex(p.As), uint32(p.From.Reg), uint32(p.To.Index), uint32(p.To.Reg)) case 87: /* VSX indexed load, XX1-form */ /* reg reg reg */ /* 3-register operand order: (RB)(RA*1), XT */ o1 = AOP_XX1(c.oploadx(p.As), uint32(p.To.Reg), uint32(p.From.Index), uint32(p.From.Reg)) case 88: /* VSX mfvsr* instructions, XX1-form XS,RA */ o1 = AOP_XX1(c.oprrr(p.As), uint32(p.From.Reg), uint32(p.To.Reg), uint32(p.Reg)) case 89: /* VSX instructions, XX2-form */ /* reg none reg OR reg imm reg */ /* 2-register operand order: XB, XT or XB, UIM, XT*/ uim := int(c.regoff(p.GetFrom3())) o1 = AOP_XX2(c.oprrr(p.As), uint32(p.To.Reg), uint32(uim), uint32(p.From.Reg)) case 90: /* VSX instructions, XX3-form */ if p.From3Type() == obj.TYPE_NONE { /* reg reg reg */ /* 3-register operand order: XA, XB, XT */ o1 = AOP_XX3(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg)) } else if p.From3Type() == obj.TYPE_CONST { /* reg reg reg imm */ /* operand order: XA, XB, DM, XT */ dm := int(c.regoff(p.GetFrom3())) o1 = AOP_XX3I(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg), uint32(dm)) } case 91: /* VSX instructions, XX4-form */ /* reg reg reg reg */ /* 3-register operand order: XA, XB, XC, XT */ o1 = AOP_XX4(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg), uint32(p.GetFrom3().Reg)) case 92: /* X-form instructions, 3-operands */ if p.To.Type == obj.TYPE_CONST { /* imm reg reg */ xf := int32(p.From.Reg) if REG_F0 <= xf && xf <= REG_F31 { /* operand order: FRA, FRB, BF */ bf := int(c.regoff(&p.To)) << 2 o1 = AOP_RRR(c.opirr(p.As), uint32(bf), uint32(p.From.Reg), uint32(p.Reg)) } else { /* operand order: RA, RB, L */ l := int(c.regoff(&p.To)) o1 = AOP_RRR(c.opirr(p.As), uint32(l), uint32(p.From.Reg), uint32(p.Reg)) } } else if p.From3Type() == obj.TYPE_CONST { /* reg reg imm */ /* operand order: RB, L, RA */ l := int(c.regoff(p.GetFrom3())) o1 = AOP_RRR(c.opirr(p.As), uint32(l), uint32(p.To.Reg), uint32(p.From.Reg)) } else if p.To.Type == obj.TYPE_REG { cr := int32(p.To.Reg) if REG_CR0 <= cr && cr <= REG_CR7 { /* cr reg reg */ /* operand order: RA, RB, BF */ bf := (int(p.To.Reg) & 7) << 2 o1 = AOP_RRR(c.opirr(p.As), uint32(bf), uint32(p.From.Reg), uint32(p.Reg)) } else if p.From.Type == obj.TYPE_CONST { /* reg imm */ /* operand order: L, RT */ l := int(c.regoff(&p.From)) o1 = AOP_RRR(c.opirr(p.As), uint32(p.To.Reg), uint32(l), uint32(p.Reg)) } else { switch p.As { case ACOPY, APASTECC: o1 = AOP_RRR(c.opirr(p.As), uint32(1), uint32(p.From.Reg), uint32(p.To.Reg)) default: /* reg reg reg */ /* operand order: RS, RB, RA */ o1 = AOP_RRR(c.oprrr(p.As), uint32(p.From.Reg), uint32(p.To.Reg), uint32(p.Reg)) } } } case 93: /* X-form instructions, 2-operands */ if p.To.Type == obj.TYPE_CONST { /* imm reg */ /* operand order: FRB, BF */ bf := int(c.regoff(&p.To)) << 2 o1 = AOP_RR(c.opirr(p.As), uint32(bf), uint32(p.From.Reg)) } else if p.Reg == 0 { /* popcnt* r,r, X-form */ /* operand order: RS, RA */ o1 = AOP_RRR(c.oprrr(p.As), uint32(p.From.Reg), uint32(p.To.Reg), uint32(p.Reg)) } case 94: /* Z23-form instructions, 4-operands */ /* reg reg reg imm */ /* operand order: RA, RB, CY, RT */ cy := int(c.regoff(p.GetFrom3())) o1 = AOP_Z23I(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg), uint32(cy)) case 96: /* VSX load, DQ-form */ /* reg imm reg */ /* operand order: (RA)(DQ), XT */ dq := int16(c.regoff(&p.From)) if (dq & 15) != 0 { c.ctxt.Diag("invalid offset for DQ form load/store %v", dq) } o1 = AOP_DQ(c.opload(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(dq)) case 97: /* VSX store, DQ-form */ /* reg imm reg */ /* operand order: XT, (RA)(DQ) */ dq := int16(c.regoff(&p.To)) if (dq & 15) != 0 { c.ctxt.Diag("invalid offset for DQ form load/store %v", dq) } o1 = AOP_DQ(c.opstore(p.As), uint32(p.From.Reg), uint32(p.To.Reg), uint32(dq)) case 98: /* VSX indexed load or load with length (also left-justified), x-form */ /* vsreg, reg, reg */ o1 = AOP_XX1(c.opload(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg)) case 99: /* VSX store with length (also left-justified) x-form */ /* reg, reg, vsreg */ o1 = AOP_XX1(c.opstore(p.As), uint32(p.From.Reg), uint32(p.Reg), uint32(p.To.Reg)) case 100: /* VSX X-form XXSPLTIB */ if p.From.Type == obj.TYPE_CONST { /* imm reg */ uim := int(c.regoff(&p.From)) /* imm reg */ /* Use AOP_XX1 form with 0 for one of the registers. */ o1 = AOP_XX1(c.oprrr(p.As), uint32(p.To.Reg), uint32(0), uint32(uim)) } else { c.ctxt.Diag("invalid ops for %v", p.As) } case 101: o1 = AOP_XX2(c.oprrr(p.As), uint32(p.To.Reg), uint32(0), uint32(p.From.Reg)) case 104: /* VSX mtvsr* instructions, XX1-form RA,RB,XT */ o1 = AOP_XX1(c.oprrr(p.As), uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg)) case 106: /* MOVD spr, soreg */ v := int32(p.From.Reg) o1 = OPVCC(31, 339, 0, 0) /* mfspr */ o1 = AOP_RRR(o1, uint32(REGTMP), 0, 0) | (uint32(v)&0x1f)<<16 | ((uint32(v)>>5)&0x1f)<<11 so := c.regoff(&p.To) o2 = AOP_IRR(c.opstore(AMOVD), uint32(REGTMP), uint32(p.To.Reg), uint32(so)) if so&0x3 != 0 { log.Fatalf("invalid offset for DS form load/store %v", p) } if p.To.Reg == REGTMP { log.Fatalf("SPR move to memory will clobber R31 %v", p) } case 107: /* MOVD soreg, spr */ v := int32(p.From.Reg) so := c.regoff(&p.From) o1 = AOP_IRR(c.opload(AMOVD), uint32(REGTMP), uint32(v), uint32(so)) o2 = OPVCC(31, 467, 0, 0) /* mtspr */ v = int32(p.To.Reg) o2 = AOP_RRR(o2, uint32(REGTMP), 0, 0) | (uint32(v)&0x1f)<<16 | ((uint32(v)>>5)&0x1f)<<11 if so&0x3 != 0 { log.Fatalf("invalid offset for DS form load/store %v", p) } case 108: /* mov r, xoreg ==> stwx rx,ry */ r := int(p.To.Reg) o1 = AOP_RRR(c.opstorex(p.As), uint32(p.From.Reg), uint32(p.To.Index), uint32(r)) case 109: /* mov xoreg, r ==> lbzx/lhzx/lwzx rx,ry, lbzx rx,ry + extsb r,r */ r := int(p.From.Reg) o1 = AOP_RRR(c.oploadx(p.As), uint32(p.To.Reg), uint32(p.From.Index), uint32(r)) // Sign extend MOVB operations. This is ignored for other cases (o.size == 4). o2 = LOP_RRR(OP_EXTSB, uint32(p.To.Reg), uint32(p.To.Reg), 0) case 110: /* SETB creg, rt */ bfa := uint32(p.From.Reg) << 2 rt := uint32(p.To.Reg) o1 = LOP_RRR(OP_SETB, bfa, rt, 0) } out[0] = o1 out[1] = o2 out[2] = o3 out[3] = o4 out[4] = o5 } func (c *ctxt9) vregoff(a *obj.Addr) int64 { c.instoffset = 0 if a != nil { c.aclass(a) } return c.instoffset } func (c *ctxt9) regoff(a *obj.Addr) int32 { return int32(c.vregoff(a)) } func (c *ctxt9) oprrr(a obj.As) uint32 { switch a { case AADD: return OPVCC(31, 266, 0, 0) case AADDCC: return OPVCC(31, 266, 0, 1) case AADDV: return OPVCC(31, 266, 1, 0) case AADDVCC: return OPVCC(31, 266, 1, 1) case AADDC: return OPVCC(31, 10, 0, 0) case AADDCCC: return OPVCC(31, 10, 0, 1) case AADDCV: return OPVCC(31, 10, 1, 0) case AADDCVCC: return OPVCC(31, 10, 1, 1) case AADDE: return OPVCC(31, 138, 0, 0) case AADDECC: return OPVCC(31, 138, 0, 1) case AADDEV: return OPVCC(31, 138, 1, 0) case AADDEVCC: return OPVCC(31, 138, 1, 1) case AADDME: return OPVCC(31, 234, 0, 0) case AADDMECC: return OPVCC(31, 234, 0, 1) case AADDMEV: return OPVCC(31, 234, 1, 0) case AADDMEVCC: return OPVCC(31, 234, 1, 1) case AADDZE: return OPVCC(31, 202, 0, 0) case AADDZECC: return OPVCC(31, 202, 0, 1) case AADDZEV: return OPVCC(31, 202, 1, 0) case AADDZEVCC: return OPVCC(31, 202, 1, 1) case AADDEX: return OPVCC(31, 170, 0, 0) /* addex - v3.0b */ case AAND: return OPVCC(31, 28, 0, 0) case AANDCC: return OPVCC(31, 28, 0, 1) case AANDN: return OPVCC(31, 60, 0, 0) case AANDNCC: return OPVCC(31, 60, 0, 1) case ACMP: return OPVCC(31, 0, 0, 0) | 1<<21 /* L=1 */ case ACMPU: return OPVCC(31, 32, 0, 0) | 1<<21 case ACMPW: return OPVCC(31, 0, 0, 0) /* L=0 */ case ACMPWU: return OPVCC(31, 32, 0, 0) case ACMPB: return OPVCC(31, 508, 0, 0) /* cmpb - v2.05 */ case ACMPEQB: return OPVCC(31, 224, 0, 0) /* cmpeqb - v3.00 */ case ACNTLZW: return OPVCC(31, 26, 0, 0) case ACNTLZWCC: return OPVCC(31, 26, 0, 1) case ACNTLZD: return OPVCC(31, 58, 0, 0) case ACNTLZDCC: return OPVCC(31, 58, 0, 1) case ACRAND: return OPVCC(19, 257, 0, 0) case ACRANDN: return OPVCC(19, 129, 0, 0) case ACREQV: return OPVCC(19, 289, 0, 0) case ACRNAND: return OPVCC(19, 225, 0, 0) case ACRNOR: return OPVCC(19, 33, 0, 0) case ACROR: return OPVCC(19, 449, 0, 0) case ACRORN: return OPVCC(19, 417, 0, 0) case ACRXOR: return OPVCC(19, 193, 0, 0) case ADCBF: return OPVCC(31, 86, 0, 0) case ADCBI: return OPVCC(31, 470, 0, 0) case ADCBST: return OPVCC(31, 54, 0, 0) case ADCBT: return OPVCC(31, 278, 0, 0) case ADCBTST: return OPVCC(31, 246, 0, 0) case ADCBZ: return OPVCC(31, 1014, 0, 0) case AMODUD: return OPVCC(31, 265, 0, 0) /* modud - v3.0 */ case AMODUW: return OPVCC(31, 267, 0, 0) /* moduw - v3.0 */ case AMODSD: return OPVCC(31, 777, 0, 0) /* modsd - v3.0 */ case AMODSW: return OPVCC(31, 779, 0, 0) /* modsw - v3.0 */ case ADIVW, AREM: return OPVCC(31, 491, 0, 0) case ADIVWCC: return OPVCC(31, 491, 0, 1) case ADIVWV: return OPVCC(31, 491, 1, 0) case ADIVWVCC: return OPVCC(31, 491, 1, 1) case ADIVWU, AREMU: return OPVCC(31, 459, 0, 0) case ADIVWUCC: return OPVCC(31, 459, 0, 1) case ADIVWUV: return OPVCC(31, 459, 1, 0) case ADIVWUVCC: return OPVCC(31, 459, 1, 1) case ADIVD, AREMD: return OPVCC(31, 489, 0, 0) case ADIVDCC: return OPVCC(31, 489, 0, 1) case ADIVDE: return OPVCC(31, 425, 0, 0) case ADIVDECC: return OPVCC(31, 425, 0, 1) case ADIVDEU: return OPVCC(31, 393, 0, 0) case ADIVDEUCC: return OPVCC(31, 393, 0, 1) case ADIVDV: return OPVCC(31, 489, 1, 0) case ADIVDVCC: return OPVCC(31, 489, 1, 1) case ADIVDU, AREMDU: return OPVCC(31, 457, 0, 0) case ADIVDUCC: return OPVCC(31, 457, 0, 1) case ADIVDUV: return OPVCC(31, 457, 1, 0) case ADIVDUVCC: return OPVCC(31, 457, 1, 1) case AEIEIO: return OPVCC(31, 854, 0, 0) case AEQV: return OPVCC(31, 284, 0, 0) case AEQVCC: return OPVCC(31, 284, 0, 1) case AEXTSB: return OPVCC(31, 954, 0, 0) case AEXTSBCC: return OPVCC(31, 954, 0, 1) case AEXTSH: return OPVCC(31, 922, 0, 0) case AEXTSHCC: return OPVCC(31, 922, 0, 1) case AEXTSW: return OPVCC(31, 986, 0, 0) case AEXTSWCC: return OPVCC(31, 986, 0, 1) case AFABS: return OPVCC(63, 264, 0, 0) case AFABSCC: return OPVCC(63, 264, 0, 1) case AFADD: return OPVCC(63, 21, 0, 0) case AFADDCC: return OPVCC(63, 21, 0, 1) case AFADDS: return OPVCC(59, 21, 0, 0) case AFADDSCC: return OPVCC(59, 21, 0, 1) case AFCMPO: return OPVCC(63, 32, 0, 0) case AFCMPU: return OPVCC(63, 0, 0, 0) case AFCFID: return OPVCC(63, 846, 0, 0) case AFCFIDCC: return OPVCC(63, 846, 0, 1) case AFCFIDU: return OPVCC(63, 974, 0, 0) case AFCFIDUCC: return OPVCC(63, 974, 0, 1) case AFCFIDS: return OPVCC(59, 846, 0, 0) case AFCFIDSCC: return OPVCC(59, 846, 0, 1) case AFCTIW: return OPVCC(63, 14, 0, 0) case AFCTIWCC: return OPVCC(63, 14, 0, 1) case AFCTIWZ: return OPVCC(63, 15, 0, 0) case AFCTIWZCC: return OPVCC(63, 15, 0, 1) case AFCTID: return OPVCC(63, 814, 0, 0) case AFCTIDCC: return OPVCC(63, 814, 0, 1) case AFCTIDZ: return OPVCC(63, 815, 0, 0) case AFCTIDZCC: return OPVCC(63, 815, 0, 1) case AFDIV: return OPVCC(63, 18, 0, 0) case AFDIVCC: return OPVCC(63, 18, 0, 1) case AFDIVS: return OPVCC(59, 18, 0, 0) case AFDIVSCC: return OPVCC(59, 18, 0, 1) case AFMADD: return OPVCC(63, 29, 0, 0) case AFMADDCC: return OPVCC(63, 29, 0, 1) case AFMADDS: return OPVCC(59, 29, 0, 0) case AFMADDSCC: return OPVCC(59, 29, 0, 1) case AFMOVS, AFMOVD: return OPVCC(63, 72, 0, 0) /* load */ case AFMOVDCC: return OPVCC(63, 72, 0, 1) case AFMSUB: return OPVCC(63, 28, 0, 0) case AFMSUBCC: return OPVCC(63, 28, 0, 1) case AFMSUBS: return OPVCC(59, 28, 0, 0) case AFMSUBSCC: return OPVCC(59, 28, 0, 1) case AFMUL: return OPVCC(63, 25, 0, 0) case AFMULCC: return OPVCC(63, 25, 0, 1) case AFMULS: return OPVCC(59, 25, 0, 0) case AFMULSCC: return OPVCC(59, 25, 0, 1) case AFNABS: return OPVCC(63, 136, 0, 0) case AFNABSCC: return OPVCC(63, 136, 0, 1) case AFNEG: return OPVCC(63, 40, 0, 0) case AFNEGCC: return OPVCC(63, 40, 0, 1) case AFNMADD: return OPVCC(63, 31, 0, 0) case AFNMADDCC: return OPVCC(63, 31, 0, 1) case AFNMADDS: return OPVCC(59, 31, 0, 0) case AFNMADDSCC: return OPVCC(59, 31, 0, 1) case AFNMSUB: return OPVCC(63, 30, 0, 0) case AFNMSUBCC: return OPVCC(63, 30, 0, 1) case AFNMSUBS: return OPVCC(59, 30, 0, 0) case AFNMSUBSCC: return OPVCC(59, 30, 0, 1) case AFCPSGN: return OPVCC(63, 8, 0, 0) case AFCPSGNCC: return OPVCC(63, 8, 0, 1) case AFRES: return OPVCC(59, 24, 0, 0) case AFRESCC: return OPVCC(59, 24, 0, 1) case AFRIM: return OPVCC(63, 488, 0, 0) case AFRIMCC: return OPVCC(63, 488, 0, 1) case AFRIP: return OPVCC(63, 456, 0, 0) case AFRIPCC: return OPVCC(63, 456, 0, 1) case AFRIZ: return OPVCC(63, 424, 0, 0) case AFRIZCC: return OPVCC(63, 424, 0, 1) case AFRIN: return OPVCC(63, 392, 0, 0) case AFRINCC: return OPVCC(63, 392, 0, 1) case AFRSP: return OPVCC(63, 12, 0, 0) case AFRSPCC: return OPVCC(63, 12, 0, 1) case AFRSQRTE: return OPVCC(63, 26, 0, 0) case AFRSQRTECC: return OPVCC(63, 26, 0, 1) case AFSEL: return OPVCC(63, 23, 0, 0) case AFSELCC: return OPVCC(63, 23, 0, 1) case AFSQRT: return OPVCC(63, 22, 0, 0) case AFSQRTCC: return OPVCC(63, 22, 0, 1) case AFSQRTS: return OPVCC(59, 22, 0, 0) case AFSQRTSCC: return OPVCC(59, 22, 0, 1) case AFSUB: return OPVCC(63, 20, 0, 0) case AFSUBCC: return OPVCC(63, 20, 0, 1) case AFSUBS: return OPVCC(59, 20, 0, 0) case AFSUBSCC: return OPVCC(59, 20, 0, 1) case AICBI: return OPVCC(31, 982, 0, 0) case AISYNC: return OPVCC(19, 150, 0, 0) case AMTFSB0: return OPVCC(63, 70, 0, 0) case AMTFSB0CC: return OPVCC(63, 70, 0, 1) case AMTFSB1: return OPVCC(63, 38, 0, 0) case AMTFSB1CC: return OPVCC(63, 38, 0, 1) case AMULHW: return OPVCC(31, 75, 0, 0) case AMULHWCC: return OPVCC(31, 75, 0, 1) case AMULHWU: return OPVCC(31, 11, 0, 0) case AMULHWUCC: return OPVCC(31, 11, 0, 1) case AMULLW: return OPVCC(31, 235, 0, 0) case AMULLWCC: return OPVCC(31, 235, 0, 1) case AMULLWV: return OPVCC(31, 235, 1, 0) case AMULLWVCC: return OPVCC(31, 235, 1, 1) case AMULHD: return OPVCC(31, 73, 0, 0) case AMULHDCC: return OPVCC(31, 73, 0, 1) case AMULHDU: return OPVCC(31, 9, 0, 0) case AMULHDUCC: return OPVCC(31, 9, 0, 1) case AMULLD: return OPVCC(31, 233, 0, 0) case AMULLDCC: return OPVCC(31, 233, 0, 1) case AMULLDV: return OPVCC(31, 233, 1, 0) case AMULLDVCC: return OPVCC(31, 233, 1, 1) case ANAND: return OPVCC(31, 476, 0, 0) case ANANDCC: return OPVCC(31, 476, 0, 1) case ANEG: return OPVCC(31, 104, 0, 0) case ANEGCC: return OPVCC(31, 104, 0, 1) case ANEGV: return OPVCC(31, 104, 1, 0) case ANEGVCC: return OPVCC(31, 104, 1, 1) case ANOR: return OPVCC(31, 124, 0, 0) case ANORCC: return OPVCC(31, 124, 0, 1) case AOR: return OPVCC(31, 444, 0, 0) case AORCC: return OPVCC(31, 444, 0, 1) case AORN: return OPVCC(31, 412, 0, 0) case AORNCC: return OPVCC(31, 412, 0, 1) case APOPCNTD: return OPVCC(31, 506, 0, 0) /* popcntd - v2.06 */ case APOPCNTW: return OPVCC(31, 378, 0, 0) /* popcntw - v2.06 */ case APOPCNTB: return OPVCC(31, 122, 0, 0) /* popcntb - v2.02 */ case ACNTTZW: return OPVCC(31, 538, 0, 0) /* cnttzw - v3.00 */ case ACNTTZWCC: return OPVCC(31, 538, 0, 1) /* cnttzw. - v3.00 */ case ACNTTZD: return OPVCC(31, 570, 0, 0) /* cnttzd - v3.00 */ case ACNTTZDCC: return OPVCC(31, 570, 0, 1) /* cnttzd. - v3.00 */ case ARFI: return OPVCC(19, 50, 0, 0) case ARFCI: return OPVCC(19, 51, 0, 0) case ARFID: return OPVCC(19, 18, 0, 0) case AHRFID: return OPVCC(19, 274, 0, 0) case ARLWNM: return OPVCC(23, 0, 0, 0) case ARLWNMCC: return OPVCC(23, 0, 0, 1) case ARLDCL: return OPVCC(30, 8, 0, 0) case ARLDCLCC: return OPVCC(30, 0, 0, 1) case ARLDCR: return OPVCC(30, 9, 0, 0) case ARLDCRCC: return OPVCC(30, 9, 0, 1) case ARLDICL: return OPVCC(30, 0, 0, 0) case ARLDICLCC: return OPVCC(30, 0, 0, 1) case ARLDICR: return OPMD(30, 1, 0) // rldicr case ARLDICRCC: return OPMD(30, 1, 1) // rldicr. case ARLDIC: return OPMD(30, 2, 0) // rldic case ARLDICCC: return OPMD(30, 2, 1) // rldic. case ASYSCALL: return OPVCC(17, 1, 0, 0) case ASLW: return OPVCC(31, 24, 0, 0) case ASLWCC: return OPVCC(31, 24, 0, 1) case ASLD: return OPVCC(31, 27, 0, 0) case ASLDCC: return OPVCC(31, 27, 0, 1) case ASRAW: return OPVCC(31, 792, 0, 0) case ASRAWCC: return OPVCC(31, 792, 0, 1) case ASRAD: return OPVCC(31, 794, 0, 0) case ASRADCC: return OPVCC(31, 794, 0, 1) case AEXTSWSLI: return OPVCC(31, 445, 0, 0) case AEXTSWSLICC: return OPVCC(31, 445, 0, 1) case ASRW: return OPVCC(31, 536, 0, 0) case ASRWCC: return OPVCC(31, 536, 0, 1) case ASRD: return OPVCC(31, 539, 0, 0) case ASRDCC: return OPVCC(31, 539, 0, 1) case ASUB: return OPVCC(31, 40, 0, 0) case ASUBCC: return OPVCC(31, 40, 0, 1) case ASUBV: return OPVCC(31, 40, 1, 0) case ASUBVCC: return OPVCC(31, 40, 1, 1) case ASUBC: return OPVCC(31, 8, 0, 0) case ASUBCCC: return OPVCC(31, 8, 0, 1) case ASUBCV: return OPVCC(31, 8, 1, 0) case ASUBCVCC: return OPVCC(31, 8, 1, 1) case ASUBE: return OPVCC(31, 136, 0, 0) case ASUBECC: return OPVCC(31, 136, 0, 1) case ASUBEV: return OPVCC(31, 136, 1, 0) case ASUBEVCC: return OPVCC(31, 136, 1, 1) case ASUBME: return OPVCC(31, 232, 0, 0) case ASUBMECC: return OPVCC(31, 232, 0, 1) case ASUBMEV: return OPVCC(31, 232, 1, 0) case ASUBMEVCC: return OPVCC(31, 232, 1, 1) case ASUBZE: return OPVCC(31, 200, 0, 0) case ASUBZECC: return OPVCC(31, 200, 0, 1) case ASUBZEV: return OPVCC(31, 200, 1, 0) case ASUBZEVCC: return OPVCC(31, 200, 1, 1) case ASYNC: return OPVCC(31, 598, 0, 0) case ALWSYNC: return OPVCC(31, 598, 0, 0) | 1<<21 case APTESYNC: return OPVCC(31, 598, 0, 0) | 2<<21 case ATLBIE: return OPVCC(31, 306, 0, 0) case ATLBIEL: return OPVCC(31, 274, 0, 0) case ATLBSYNC: return OPVCC(31, 566, 0, 0) case ASLBIA: return OPVCC(31, 498, 0, 0) case ASLBIE: return OPVCC(31, 434, 0, 0) case ASLBMFEE: return OPVCC(31, 915, 0, 0) case ASLBMFEV: return OPVCC(31, 851, 0, 0) case ASLBMTE: return OPVCC(31, 402, 0, 0) case ATW: return OPVCC(31, 4, 0, 0) case ATD: return OPVCC(31, 68, 0, 0) /* Vector (VMX/Altivec) instructions */ /* ISA 2.03 enables these for PPC970. For POWERx processors, these */ /* are enabled starting at POWER6 (ISA 2.05). */ case AVAND: return OPVX(4, 1028, 0, 0) /* vand - v2.03 */ case AVANDC: return OPVX(4, 1092, 0, 0) /* vandc - v2.03 */ case AVNAND: return OPVX(4, 1412, 0, 0) /* vnand - v2.07 */ case AVOR: return OPVX(4, 1156, 0, 0) /* vor - v2.03 */ case AVORC: return OPVX(4, 1348, 0, 0) /* vorc - v2.07 */ case AVNOR: return OPVX(4, 1284, 0, 0) /* vnor - v2.03 */ case AVXOR: return OPVX(4, 1220, 0, 0) /* vxor - v2.03 */ case AVEQV: return OPVX(4, 1668, 0, 0) /* veqv - v2.07 */ case AVADDUBM: return OPVX(4, 0, 0, 0) /* vaddubm - v2.03 */ case AVADDUHM: return OPVX(4, 64, 0, 0) /* vadduhm - v2.03 */ case AVADDUWM: return OPVX(4, 128, 0, 0) /* vadduwm - v2.03 */ case AVADDUDM: return OPVX(4, 192, 0, 0) /* vaddudm - v2.07 */ case AVADDUQM: return OPVX(4, 256, 0, 0) /* vadduqm - v2.07 */ case AVADDCUQ: return OPVX(4, 320, 0, 0) /* vaddcuq - v2.07 */ case AVADDCUW: return OPVX(4, 384, 0, 0) /* vaddcuw - v2.03 */ case AVADDUBS: return OPVX(4, 512, 0, 0) /* vaddubs - v2.03 */ case AVADDUHS: return OPVX(4, 576, 0, 0) /* vadduhs - v2.03 */ case AVADDUWS: return OPVX(4, 640, 0, 0) /* vadduws - v2.03 */ case AVADDSBS: return OPVX(4, 768, 0, 0) /* vaddsbs - v2.03 */ case AVADDSHS: return OPVX(4, 832, 0, 0) /* vaddshs - v2.03 */ case AVADDSWS: return OPVX(4, 896, 0, 0) /* vaddsws - v2.03 */ case AVADDEUQM: return OPVX(4, 60, 0, 0) /* vaddeuqm - v2.07 */ case AVADDECUQ: return OPVX(4, 61, 0, 0) /* vaddecuq - v2.07 */ case AVMULESB: return OPVX(4, 776, 0, 0) /* vmulesb - v2.03 */ case AVMULOSB: return OPVX(4, 264, 0, 0) /* vmulosb - v2.03 */ case AVMULEUB: return OPVX(4, 520, 0, 0) /* vmuleub - v2.03 */ case AVMULOUB: return OPVX(4, 8, 0, 0) /* vmuloub - v2.03 */ case AVMULESH: return OPVX(4, 840, 0, 0) /* vmulesh - v2.03 */ case AVMULOSH: return OPVX(4, 328, 0, 0) /* vmulosh - v2.03 */ case AVMULEUH: return OPVX(4, 584, 0, 0) /* vmuleuh - v2.03 */ case AVMULOUH: return OPVX(4, 72, 0, 0) /* vmulouh - v2.03 */ case AVMULESW: return OPVX(4, 904, 0, 0) /* vmulesw - v2.07 */ case AVMULOSW: return OPVX(4, 392, 0, 0) /* vmulosw - v2.07 */ case AVMULEUW: return OPVX(4, 648, 0, 0) /* vmuleuw - v2.07 */ case AVMULOUW: return OPVX(4, 136, 0, 0) /* vmulouw - v2.07 */ case AVMULUWM: return OPVX(4, 137, 0, 0) /* vmuluwm - v2.07 */ case AVPMSUMB: return OPVX(4, 1032, 0, 0) /* vpmsumb - v2.07 */ case AVPMSUMH: return OPVX(4, 1096, 0, 0) /* vpmsumh - v2.07 */ case AVPMSUMW: return OPVX(4, 1160, 0, 0) /* vpmsumw - v2.07 */ case AVPMSUMD: return OPVX(4, 1224, 0, 0) /* vpmsumd - v2.07 */ case AVMSUMUDM: return OPVX(4, 35, 0, 0) /* vmsumudm - v3.00b */ case AVSUBUBM: return OPVX(4, 1024, 0, 0) /* vsububm - v2.03 */ case AVSUBUHM: return OPVX(4, 1088, 0, 0) /* vsubuhm - v2.03 */ case AVSUBUWM: return OPVX(4, 1152, 0, 0) /* vsubuwm - v2.03 */ case AVSUBUDM: return OPVX(4, 1216, 0, 0) /* vsubudm - v2.07 */ case AVSUBUQM: return OPVX(4, 1280, 0, 0) /* vsubuqm - v2.07 */ case AVSUBCUQ: return OPVX(4, 1344, 0, 0) /* vsubcuq - v2.07 */ case AVSUBCUW: return OPVX(4, 1408, 0, 0) /* vsubcuw - v2.03 */ case AVSUBUBS: return OPVX(4, 1536, 0, 0) /* vsububs - v2.03 */ case AVSUBUHS: return OPVX(4, 1600, 0, 0) /* vsubuhs - v2.03 */ case AVSUBUWS: return OPVX(4, 1664, 0, 0) /* vsubuws - v2.03 */ case AVSUBSBS: return OPVX(4, 1792, 0, 0) /* vsubsbs - v2.03 */ case AVSUBSHS: return OPVX(4, 1856, 0, 0) /* vsubshs - v2.03 */ case AVSUBSWS: return OPVX(4, 1920, 0, 0) /* vsubsws - v2.03 */ case AVSUBEUQM: return OPVX(4, 62, 0, 0) /* vsubeuqm - v2.07 */ case AVSUBECUQ: return OPVX(4, 63, 0, 0) /* vsubecuq - v2.07 */ case AVRLB: return OPVX(4, 4, 0, 0) /* vrlb - v2.03 */ case AVRLH: return OPVX(4, 68, 0, 0) /* vrlh - v2.03 */ case AVRLW: return OPVX(4, 132, 0, 0) /* vrlw - v2.03 */ case AVRLD: return OPVX(4, 196, 0, 0) /* vrld - v2.07 */ case AVMRGOW: return OPVX(4, 1676, 0, 0) /* vmrgow - v2.07 */ case AVMRGEW: return OPVX(4, 1932, 0, 0) /* vmrgew - v2.07 */ case AVSLB: return OPVX(4, 260, 0, 0) /* vslh - v2.03 */ case AVSLH: return OPVX(4, 324, 0, 0) /* vslh - v2.03 */ case AVSLW: return OPVX(4, 388, 0, 0) /* vslw - v2.03 */ case AVSL: return OPVX(4, 452, 0, 0) /* vsl - v2.03 */ case AVSLO: return OPVX(4, 1036, 0, 0) /* vsl - v2.03 */ case AVSRB: return OPVX(4, 516, 0, 0) /* vsrb - v2.03 */ case AVSRH: return OPVX(4, 580, 0, 0) /* vsrh - v2.03 */ case AVSRW: return OPVX(4, 644, 0, 0) /* vsrw - v2.03 */ case AVSR: return OPVX(4, 708, 0, 0) /* vsr - v2.03 */ case AVSRO: return OPVX(4, 1100, 0, 0) /* vsro - v2.03 */ case AVSLD: return OPVX(4, 1476, 0, 0) /* vsld - v2.07 */ case AVSRD: return OPVX(4, 1732, 0, 0) /* vsrd - v2.07 */ case AVSRAB: return OPVX(4, 772, 0, 0) /* vsrab - v2.03 */ case AVSRAH: return OPVX(4, 836, 0, 0) /* vsrah - v2.03 */ case AVSRAW: return OPVX(4, 900, 0, 0) /* vsraw - v2.03 */ case AVSRAD: return OPVX(4, 964, 0, 0) /* vsrad - v2.07 */ case AVBPERMQ: return OPVC(4, 1356, 0, 0) /* vbpermq - v2.07 */ case AVBPERMD: return OPVC(4, 1484, 0, 0) /* vbpermd - v3.00 */ case AVCLZB: return OPVX(4, 1794, 0, 0) /* vclzb - v2.07 */ case AVCLZH: return OPVX(4, 1858, 0, 0) /* vclzh - v2.07 */ case AVCLZW: return OPVX(4, 1922, 0, 0) /* vclzw - v2.07 */ case AVCLZD: return OPVX(4, 1986, 0, 0) /* vclzd - v2.07 */ case AVCLZLSBB: return OPVX(4, 1538, 0, 0) /* vclzlsbb - v3.0 */ case AVCTZLSBB: return OPVX(4, 1538, 0, 0) | 1<<16 /* vctzlsbb - v3.0 */ case AVPOPCNTB: return OPVX(4, 1795, 0, 0) /* vpopcntb - v2.07 */ case AVPOPCNTH: return OPVX(4, 1859, 0, 0) /* vpopcnth - v2.07 */ case AVPOPCNTW: return OPVX(4, 1923, 0, 0) /* vpopcntw - v2.07 */ case AVPOPCNTD: return OPVX(4, 1987, 0, 0) /* vpopcntd - v2.07 */ case AVCMPEQUB: return OPVC(4, 6, 0, 0) /* vcmpequb - v2.03 */ case AVCMPEQUBCC: return OPVC(4, 6, 0, 1) /* vcmpequb. - v2.03 */ case AVCMPEQUH: return OPVC(4, 70, 0, 0) /* vcmpequh - v2.03 */ case AVCMPEQUHCC: return OPVC(4, 70, 0, 1) /* vcmpequh. - v2.03 */ case AVCMPEQUW: return OPVC(4, 134, 0, 0) /* vcmpequw - v2.03 */ case AVCMPEQUWCC: return OPVC(4, 134, 0, 1) /* vcmpequw. - v2.03 */ case AVCMPEQUD: return OPVC(4, 199, 0, 0) /* vcmpequd - v2.07 */ case AVCMPEQUDCC: return OPVC(4, 199, 0, 1) /* vcmpequd. - v2.07 */ case AVCMPGTUB: return OPVC(4, 518, 0, 0) /* vcmpgtub - v2.03 */ case AVCMPGTUBCC: return OPVC(4, 518, 0, 1) /* vcmpgtub. - v2.03 */ case AVCMPGTUH: return OPVC(4, 582, 0, 0) /* vcmpgtuh - v2.03 */ case AVCMPGTUHCC: return OPVC(4, 582, 0, 1) /* vcmpgtuh. - v2.03 */ case AVCMPGTUW: return OPVC(4, 646, 0, 0) /* vcmpgtuw - v2.03 */ case AVCMPGTUWCC: return OPVC(4, 646, 0, 1) /* vcmpgtuw. - v2.03 */ case AVCMPGTUD: return OPVC(4, 711, 0, 0) /* vcmpgtud - v2.07 */ case AVCMPGTUDCC: return OPVC(4, 711, 0, 1) /* vcmpgtud. v2.07 */ case AVCMPGTSB: return OPVC(4, 774, 0, 0) /* vcmpgtsb - v2.03 */ case AVCMPGTSBCC: return OPVC(4, 774, 0, 1) /* vcmpgtsb. - v2.03 */ case AVCMPGTSH: return OPVC(4, 838, 0, 0) /* vcmpgtsh - v2.03 */ case AVCMPGTSHCC: return OPVC(4, 838, 0, 1) /* vcmpgtsh. - v2.03 */ case AVCMPGTSW: return OPVC(4, 902, 0, 0) /* vcmpgtsw - v2.03 */ case AVCMPGTSWCC: return OPVC(4, 902, 0, 1) /* vcmpgtsw. - v2.03 */ case AVCMPGTSD: return OPVC(4, 967, 0, 0) /* vcmpgtsd - v2.07 */ case AVCMPGTSDCC: return OPVC(4, 967, 0, 1) /* vcmpgtsd. - v2.07 */ case AVCMPNEZB: return OPVC(4, 263, 0, 0) /* vcmpnezb - v3.00 */ case AVCMPNEZBCC: return OPVC(4, 263, 0, 1) /* vcmpnezb. - v3.00 */ case AVCMPNEB: return OPVC(4, 7, 0, 0) /* vcmpneb - v3.00 */ case AVCMPNEBCC: return OPVC(4, 7, 0, 1) /* vcmpneb. - v3.00 */ case AVCMPNEH: return OPVC(4, 71, 0, 0) /* vcmpneh - v3.00 */ case AVCMPNEHCC: return OPVC(4, 71, 0, 1) /* vcmpneh. - v3.00 */ case AVCMPNEW: return OPVC(4, 135, 0, 0) /* vcmpnew - v3.00 */ case AVCMPNEWCC: return OPVC(4, 135, 0, 1) /* vcmpnew. - v3.00 */ case AVPERM: return OPVX(4, 43, 0, 0) /* vperm - v2.03 */ case AVPERMXOR: return OPVX(4, 45, 0, 0) /* vpermxor - v2.03 */ case AVPERMR: return OPVX(4, 59, 0, 0) /* vpermr - v3.0 */ case AVSEL: return OPVX(4, 42, 0, 0) /* vsel - v2.03 */ case AVCIPHER: return OPVX(4, 1288, 0, 0) /* vcipher - v2.07 */ case AVCIPHERLAST: return OPVX(4, 1289, 0, 0) /* vcipherlast - v2.07 */ case AVNCIPHER: return OPVX(4, 1352, 0, 0) /* vncipher - v2.07 */ case AVNCIPHERLAST: return OPVX(4, 1353, 0, 0) /* vncipherlast - v2.07 */ case AVSBOX: return OPVX(4, 1480, 0, 0) /* vsbox - v2.07 */ /* End of vector instructions */ /* Vector scalar (VSX) instructions */ /* ISA 2.06 enables these for POWER7. */ case AMFVSRD, AMFVRD, AMFFPRD: return OPVXX1(31, 51, 0) /* mfvsrd - v2.07 */ case AMFVSRWZ: return OPVXX1(31, 115, 0) /* mfvsrwz - v2.07 */ case AMFVSRLD: return OPVXX1(31, 307, 0) /* mfvsrld - v3.00 */ case AMTVSRD, AMTFPRD, AMTVRD: return OPVXX1(31, 179, 0) /* mtvsrd - v2.07 */ case AMTVSRWA: return OPVXX1(31, 211, 0) /* mtvsrwa - v2.07 */ case AMTVSRWZ: return OPVXX1(31, 243, 0) /* mtvsrwz - v2.07 */ case AMTVSRDD: return OPVXX1(31, 435, 0) /* mtvsrdd - v3.00 */ case AMTVSRWS: return OPVXX1(31, 403, 0) /* mtvsrws - v3.00 */ case AXXLAND: return OPVXX3(60, 130, 0) /* xxland - v2.06 */ case AXXLANDC: return OPVXX3(60, 138, 0) /* xxlandc - v2.06 */ case AXXLEQV: return OPVXX3(60, 186, 0) /* xxleqv - v2.07 */ case AXXLNAND: return OPVXX3(60, 178, 0) /* xxlnand - v2.07 */ case AXXLORC: return OPVXX3(60, 170, 0) /* xxlorc - v2.07 */ case AXXLNOR: return OPVXX3(60, 162, 0) /* xxlnor - v2.06 */ case AXXLOR, AXXLORQ: return OPVXX3(60, 146, 0) /* xxlor - v2.06 */ case AXXLXOR: return OPVXX3(60, 154, 0) /* xxlxor - v2.06 */ case AXSMINJDP: return OPVXX3(60, 152, 0) /* xsminjdp - v3.0 */ case AXSMAXJDP: return OPVXX3(60, 144, 0) /* xsmaxjdp - v3.0 */ case AXXSEL: return OPVXX4(60, 3, 0) /* xxsel - v2.06 */ case AXXMRGHW: return OPVXX3(60, 18, 0) /* xxmrghw - v2.06 */ case AXXMRGLW: return OPVXX3(60, 50, 0) /* xxmrglw - v2.06 */ case AXXSPLTW: return OPVXX2(60, 164, 0) /* xxspltw - v2.06 */ case AXXSPLTIB: return OPVCC(60, 360, 0, 0) /* xxspltib - v3.0 */ case AXXPERM: return OPVXX3(60, 26, 0) /* xxperm - v2.06 */ case AXXPERMDI: return OPVXX3(60, 10, 0) /* xxpermdi - v2.06 */ case AXXSLDWI: return OPVXX3(60, 2, 0) /* xxsldwi - v2.06 */ case AXXBRQ: return OPVXX2VA(60, 475, 31) /* xxbrq - v3.0 */ case AXXBRD: return OPVXX2VA(60, 475, 23) /* xxbrd - v3.0 */ case AXXBRW: return OPVXX2VA(60, 475, 15) /* xxbrw - v3.0 */ case AXXBRH: return OPVXX2VA(60, 475, 7) /* xxbrh - v3.0 */ case AXSCVDPSP: return OPVXX2(60, 265, 0) /* xscvdpsp - v2.06 */ case AXSCVSPDP: return OPVXX2(60, 329, 0) /* xscvspdp - v2.06 */ case AXSCVDPSPN: return OPVXX2(60, 267, 0) /* xscvdpspn - v2.07 */ case AXSCVSPDPN: return OPVXX2(60, 331, 0) /* xscvspdpn - v2.07 */ case AXVCVDPSP: return OPVXX2(60, 393, 0) /* xvcvdpsp - v2.06 */ case AXVCVSPDP: return OPVXX2(60, 457, 0) /* xvcvspdp - v2.06 */ case AXSCVDPSXDS: return OPVXX2(60, 344, 0) /* xscvdpsxds - v2.06 */ case AXSCVDPSXWS: return OPVXX2(60, 88, 0) /* xscvdpsxws - v2.06 */ case AXSCVDPUXDS: return OPVXX2(60, 328, 0) /* xscvdpuxds - v2.06 */ case AXSCVDPUXWS: return OPVXX2(60, 72, 0) /* xscvdpuxws - v2.06 */ case AXSCVSXDDP: return OPVXX2(60, 376, 0) /* xscvsxddp - v2.06 */ case AXSCVUXDDP: return OPVXX2(60, 360, 0) /* xscvuxddp - v2.06 */ case AXSCVSXDSP: return OPVXX2(60, 312, 0) /* xscvsxdsp - v2.06 */ case AXSCVUXDSP: return OPVXX2(60, 296, 0) /* xscvuxdsp - v2.06 */ case AXVCVDPSXDS: return OPVXX2(60, 472, 0) /* xvcvdpsxds - v2.06 */ case AXVCVDPSXWS: return OPVXX2(60, 216, 0) /* xvcvdpsxws - v2.06 */ case AXVCVDPUXDS: return OPVXX2(60, 456, 0) /* xvcvdpuxds - v2.06 */ case AXVCVDPUXWS: return OPVXX2(60, 200, 0) /* xvcvdpuxws - v2.06 */ case AXVCVSPSXDS: return OPVXX2(60, 408, 0) /* xvcvspsxds - v2.07 */ case AXVCVSPSXWS: return OPVXX2(60, 152, 0) /* xvcvspsxws - v2.07 */ case AXVCVSPUXDS: return OPVXX2(60, 392, 0) /* xvcvspuxds - v2.07 */ case AXVCVSPUXWS: return OPVXX2(60, 136, 0) /* xvcvspuxws - v2.07 */ case AXVCVSXDDP: return OPVXX2(60, 504, 0) /* xvcvsxddp - v2.06 */ case AXVCVSXWDP: return OPVXX2(60, 248, 0) /* xvcvsxwdp - v2.06 */ case AXVCVUXDDP: return OPVXX2(60, 488, 0) /* xvcvuxddp - v2.06 */ case AXVCVUXWDP: return OPVXX2(60, 232, 0) /* xvcvuxwdp - v2.06 */ case AXVCVSXDSP: return OPVXX2(60, 440, 0) /* xvcvsxdsp - v2.06 */ case AXVCVSXWSP: return OPVXX2(60, 184, 0) /* xvcvsxwsp - v2.06 */ case AXVCVUXDSP: return OPVXX2(60, 424, 0) /* xvcvuxdsp - v2.06 */ case AXVCVUXWSP: return OPVXX2(60, 168, 0) /* xvcvuxwsp - v2.06 */ /* End of VSX instructions */ case AMADDHD: return OPVX(4, 48, 0, 0) /* maddhd - v3.00 */ case AMADDHDU: return OPVX(4, 49, 0, 0) /* maddhdu - v3.00 */ case AMADDLD: return OPVX(4, 51, 0, 0) /* maddld - v3.00 */ case AXOR: return OPVCC(31, 316, 0, 0) case AXORCC: return OPVCC(31, 316, 0, 1) } c.ctxt.Diag("bad r/r, r/r/r or r/r/r/r opcode %v", a) return 0 } func (c *ctxt9) opirrr(a obj.As) uint32 { switch a { /* Vector (VMX/Altivec) instructions */ /* ISA 2.03 enables these for PPC970. For POWERx processors, these */ /* are enabled starting at POWER6 (ISA 2.05). */ case AVSLDOI: return OPVX(4, 44, 0, 0) /* vsldoi - v2.03 */ } c.ctxt.Diag("bad i/r/r/r opcode %v", a) return 0 } func (c *ctxt9) opiirr(a obj.As) uint32 { switch a { /* Vector (VMX/Altivec) instructions */ /* ISA 2.07 enables these for POWER8 and beyond. */ case AVSHASIGMAW: return OPVX(4, 1666, 0, 0) /* vshasigmaw - v2.07 */ case AVSHASIGMAD: return OPVX(4, 1730, 0, 0) /* vshasigmad - v2.07 */ } c.ctxt.Diag("bad i/i/r/r opcode %v", a) return 0 } func (c *ctxt9) opirr(a obj.As) uint32 { switch a { case AADD: return OPVCC(14, 0, 0, 0) case AADDC: return OPVCC(12, 0, 0, 0) case AADDCCC: return OPVCC(13, 0, 0, 0) case AADDIS: return OPVCC(15, 0, 0, 0) /* ADDIS */ case AANDCC: return OPVCC(28, 0, 0, 0) case AANDISCC: return OPVCC(29, 0, 0, 0) /* ANDIS. */ case ABR: return OPVCC(18, 0, 0, 0) case ABL: return OPVCC(18, 0, 0, 0) | 1 case obj.ADUFFZERO: return OPVCC(18, 0, 0, 0) | 1 case obj.ADUFFCOPY: return OPVCC(18, 0, 0, 0) | 1 case ABC: return OPVCC(16, 0, 0, 0) case ABCL: return OPVCC(16, 0, 0, 0) | 1 case ABEQ: return AOP_RRR(16<<26, BO_BCR, BI_EQ, 0) case ABGE: return AOP_RRR(16<<26, BO_NOTBCR, BI_LT, 0) case ABGT: return AOP_RRR(16<<26, BO_BCR, BI_GT, 0) case ABLE: return AOP_RRR(16<<26, BO_NOTBCR, BI_GT, 0) case ABLT: return AOP_RRR(16<<26, BO_BCR, BI_LT, 0) case ABNE: return AOP_RRR(16<<26, BO_NOTBCR, BI_EQ, 0) case ABVC: return AOP_RRR(16<<26, BO_NOTBCR, BI_FU, 0) case ABVS: return AOP_RRR(16<<26, BO_BCR, BI_FU, 0) case ABDZ: return AOP_RRR(16<<26, BO_NOTBCTR, 0, 0) case ABDNZ: return AOP_RRR(16<<26, BO_BCTR, 0, 0) case ACMP: return OPVCC(11, 0, 0, 0) | 1<<21 /* L=1 */ case ACMPU: return OPVCC(10, 0, 0, 0) | 1<<21 case ACMPW: return OPVCC(11, 0, 0, 0) /* L=0 */ case ACMPWU: return OPVCC(10, 0, 0, 0) case ACMPEQB: return OPVCC(31, 224, 0, 0) /* cmpeqb - v3.00 */ case ALSW: return OPVCC(31, 597, 0, 0) case ACOPY: return OPVCC(31, 774, 0, 0) /* copy - v3.00 */ case APASTECC: return OPVCC(31, 902, 0, 1) /* paste. - v3.00 */ case ADARN: return OPVCC(31, 755, 0, 0) /* darn - v3.00 */ case AMULLW, AMULLD: return OPVCC(7, 0, 0, 0) /* mulli works with MULLW or MULLD */ case AOR: return OPVCC(24, 0, 0, 0) case AORIS: return OPVCC(25, 0, 0, 0) /* ORIS */ case ARLWMI: return OPVCC(20, 0, 0, 0) /* rlwimi */ case ARLWMICC: return OPVCC(20, 0, 0, 1) case ARLDMI: return OPMD(30, 3, 0) /* rldimi */ case ARLDMICC: return OPMD(30, 3, 1) /* rldimi. */ case ARLDIMI: return OPMD(30, 3, 0) /* rldimi */ case ARLDIMICC: return OPMD(30, 3, 1) /* rldimi. */ case ARLWNM: return OPVCC(21, 0, 0, 0) /* rlwinm */ case ARLWNMCC: return OPVCC(21, 0, 0, 1) case ARLDCL: return OPMD(30, 0, 0) /* rldicl */ case ARLDCLCC: return OPMD(30, 0, 1) /* rldicl. */ case ARLDCR: return OPMD(30, 1, 0) /* rldicr */ case ARLDCRCC: return OPMD(30, 1, 1) /* rldicr. */ case ARLDC: return OPMD(30, 2, 0) /* rldic */ case ARLDCCC: return OPMD(30, 2, 1) /* rldic. */ case ASRAW: return OPVCC(31, 824, 0, 0) case ASRAWCC: return OPVCC(31, 824, 0, 1) case ASRAD: return OPVCC(31, (413 << 1), 0, 0) case ASRADCC: return OPVCC(31, (413 << 1), 0, 1) case AEXTSWSLI: return OPVCC(31, 445, 0, 0) case AEXTSWSLICC: return OPVCC(31, 445, 0, 1) case ASTSW: return OPVCC(31, 725, 0, 0) case ASUBC: return OPVCC(8, 0, 0, 0) case ATW: return OPVCC(3, 0, 0, 0) case ATD: return OPVCC(2, 0, 0, 0) /* Vector (VMX/Altivec) instructions */ /* ISA 2.03 enables these for PPC970. For POWERx processors, these */ /* are enabled starting at POWER6 (ISA 2.05). */ case AVSPLTB: return OPVX(4, 524, 0, 0) /* vspltb - v2.03 */ case AVSPLTH: return OPVX(4, 588, 0, 0) /* vsplth - v2.03 */ case AVSPLTW: return OPVX(4, 652, 0, 0) /* vspltw - v2.03 */ case AVSPLTISB: return OPVX(4, 780, 0, 0) /* vspltisb - v2.03 */ case AVSPLTISH: return OPVX(4, 844, 0, 0) /* vspltish - v2.03 */ case AVSPLTISW: return OPVX(4, 908, 0, 0) /* vspltisw - v2.03 */ /* End of vector instructions */ case AFTDIV: return OPVCC(63, 128, 0, 0) /* ftdiv - v2.06 */ case AFTSQRT: return OPVCC(63, 160, 0, 0) /* ftsqrt - v2.06 */ case AXOR: return OPVCC(26, 0, 0, 0) /* XORIL */ case AXORIS: return OPVCC(27, 0, 0, 0) /* XORIS */ } c.ctxt.Diag("bad opcode i/r or i/r/r %v", a) return 0 } /* * load o(a),d */ func (c *ctxt9) opload(a obj.As) uint32 { switch a { case AMOVD: return OPVCC(58, 0, 0, 0) /* ld */ case AMOVDU: return OPVCC(58, 0, 0, 1) /* ldu */ case AMOVWZ: return OPVCC(32, 0, 0, 0) /* lwz */ case AMOVWZU: return OPVCC(33, 0, 0, 0) /* lwzu */ case AMOVW: return OPVCC(58, 0, 0, 0) | 1<<1 /* lwa */ case ALXV: return OPDQ(61, 1, 0) /* lxv - ISA v3.0 */ case ALXVL: return OPVXX1(31, 269, 0) /* lxvl - ISA v3.0 */ case ALXVLL: return OPVXX1(31, 301, 0) /* lxvll - ISA v3.0 */ case ALXVX: return OPVXX1(31, 268, 0) /* lxvx - ISA v3.0 */ /* no AMOVWU */ case AMOVB, AMOVBZ: return OPVCC(34, 0, 0, 0) /* load */ case AMOVBU, AMOVBZU: return OPVCC(35, 0, 0, 0) case AFMOVD: return OPVCC(50, 0, 0, 0) case AFMOVDU: return OPVCC(51, 0, 0, 0) case AFMOVS: return OPVCC(48, 0, 0, 0) case AFMOVSU: return OPVCC(49, 0, 0, 0) case AMOVH: return OPVCC(42, 0, 0, 0) case AMOVHU: return OPVCC(43, 0, 0, 0) case AMOVHZ: return OPVCC(40, 0, 0, 0) case AMOVHZU: return OPVCC(41, 0, 0, 0) case AMOVMW: return OPVCC(46, 0, 0, 0) /* lmw */ } c.ctxt.Diag("bad load opcode %v", a) return 0 } /* * indexed load a(b),d */ func (c *ctxt9) oploadx(a obj.As) uint32 { switch a { case AMOVWZ: return OPVCC(31, 23, 0, 0) /* lwzx */ case AMOVWZU: return OPVCC(31, 55, 0, 0) /* lwzux */ case AMOVW: return OPVCC(31, 341, 0, 0) /* lwax */ case AMOVWU: return OPVCC(31, 373, 0, 0) /* lwaux */ case AMOVB, AMOVBZ: return OPVCC(31, 87, 0, 0) /* lbzx */ case AMOVBU, AMOVBZU: return OPVCC(31, 119, 0, 0) /* lbzux */ case AFMOVD: return OPVCC(31, 599, 0, 0) /* lfdx */ case AFMOVDU: return OPVCC(31, 631, 0, 0) /* lfdux */ case AFMOVS: return OPVCC(31, 535, 0, 0) /* lfsx */ case AFMOVSU: return OPVCC(31, 567, 0, 0) /* lfsux */ case AFMOVSX: return OPVCC(31, 855, 0, 0) /* lfiwax - power6, isa 2.05 */ case AFMOVSZ: return OPVCC(31, 887, 0, 0) /* lfiwzx - power7, isa 2.06 */ case AMOVH: return OPVCC(31, 343, 0, 0) /* lhax */ case AMOVHU: return OPVCC(31, 375, 0, 0) /* lhaux */ case AMOVHBR: return OPVCC(31, 790, 0, 0) /* lhbrx */ case AMOVWBR: return OPVCC(31, 534, 0, 0) /* lwbrx */ case AMOVDBR: return OPVCC(31, 532, 0, 0) /* ldbrx */ case AMOVHZ: return OPVCC(31, 279, 0, 0) /* lhzx */ case AMOVHZU: return OPVCC(31, 311, 0, 0) /* lhzux */ case ALBAR: return OPVCC(31, 52, 0, 0) /* lbarx */ case ALHAR: return OPVCC(31, 116, 0, 0) /* lharx */ case ALWAR: return OPVCC(31, 20, 0, 0) /* lwarx */ case ALDAR: return OPVCC(31, 84, 0, 0) /* ldarx */ case ALSW: return OPVCC(31, 533, 0, 0) /* lswx */ case AMOVD: return OPVCC(31, 21, 0, 0) /* ldx */ case AMOVDU: return OPVCC(31, 53, 0, 0) /* ldux */ /* Vector (VMX/Altivec) instructions */ case ALVEBX: return OPVCC(31, 7, 0, 0) /* lvebx - v2.03 */ case ALVEHX: return OPVCC(31, 39, 0, 0) /* lvehx - v2.03 */ case ALVEWX: return OPVCC(31, 71, 0, 0) /* lvewx - v2.03 */ case ALVX: return OPVCC(31, 103, 0, 0) /* lvx - v2.03 */ case ALVXL: return OPVCC(31, 359, 0, 0) /* lvxl - v2.03 */ case ALVSL: return OPVCC(31, 6, 0, 0) /* lvsl - v2.03 */ case ALVSR: return OPVCC(31, 38, 0, 0) /* lvsr - v2.03 */ /* End of vector instructions */ /* Vector scalar (VSX) instructions */ case ALXVX: return OPVXX1(31, 268, 0) /* lxvx - ISA v3.0 */ case ALXVD2X: return OPVXX1(31, 844, 0) /* lxvd2x - v2.06 */ case ALXVW4X: return OPVXX1(31, 780, 0) /* lxvw4x - v2.06 */ case ALXVH8X: return OPVXX1(31, 812, 0) /* lxvh8x - v3.00 */ case ALXVB16X: return OPVXX1(31, 876, 0) /* lxvb16x - v3.00 */ case ALXVDSX: return OPVXX1(31, 332, 0) /* lxvdsx - v2.06 */ case ALXSDX: return OPVXX1(31, 588, 0) /* lxsdx - v2.06 */ case ALXSIWAX: return OPVXX1(31, 76, 0) /* lxsiwax - v2.07 */ case ALXSIWZX: return OPVXX1(31, 12, 0) /* lxsiwzx - v2.07 */ } c.ctxt.Diag("bad loadx opcode %v", a) return 0 } /* * store s,o(d) */ func (c *ctxt9) opstore(a obj.As) uint32 { switch a { case AMOVB, AMOVBZ: return OPVCC(38, 0, 0, 0) /* stb */ case AMOVBU, AMOVBZU: return OPVCC(39, 0, 0, 0) /* stbu */ case AFMOVD: return OPVCC(54, 0, 0, 0) /* stfd */ case AFMOVDU: return OPVCC(55, 0, 0, 0) /* stfdu */ case AFMOVS: return OPVCC(52, 0, 0, 0) /* stfs */ case AFMOVSU: return OPVCC(53, 0, 0, 0) /* stfsu */ case AMOVHZ, AMOVH: return OPVCC(44, 0, 0, 0) /* sth */ case AMOVHZU, AMOVHU: return OPVCC(45, 0, 0, 0) /* sthu */ case AMOVMW: return OPVCC(47, 0, 0, 0) /* stmw */ case ASTSW: return OPVCC(31, 725, 0, 0) /* stswi */ case AMOVWZ, AMOVW: return OPVCC(36, 0, 0, 0) /* stw */ case AMOVWZU, AMOVWU: return OPVCC(37, 0, 0, 0) /* stwu */ case AMOVD: return OPVCC(62, 0, 0, 0) /* std */ case AMOVDU: return OPVCC(62, 0, 0, 1) /* stdu */ case ASTXV: return OPDQ(61, 5, 0) /* stxv ISA 3.0 */ case ASTXVL: return OPVXX1(31, 397, 0) /* stxvl ISA 3.0 */ case ASTXVLL: return OPVXX1(31, 429, 0) /* stxvll ISA 3.0 */ case ASTXVX: return OPVXX1(31, 396, 0) /* stxvx - ISA v3.0 */ } c.ctxt.Diag("unknown store opcode %v", a) return 0 } /* * indexed store s,a(b) */ func (c *ctxt9) opstorex(a obj.As) uint32 { switch a { case AMOVB, AMOVBZ: return OPVCC(31, 215, 0, 0) /* stbx */ case AMOVBU, AMOVBZU: return OPVCC(31, 247, 0, 0) /* stbux */ case AFMOVD: return OPVCC(31, 727, 0, 0) /* stfdx */ case AFMOVDU: return OPVCC(31, 759, 0, 0) /* stfdux */ case AFMOVS: return OPVCC(31, 663, 0, 0) /* stfsx */ case AFMOVSU: return OPVCC(31, 695, 0, 0) /* stfsux */ case AFMOVSX: return OPVCC(31, 983, 0, 0) /* stfiwx */ case AMOVHZ, AMOVH: return OPVCC(31, 407, 0, 0) /* sthx */ case AMOVHBR: return OPVCC(31, 918, 0, 0) /* sthbrx */ case AMOVHZU, AMOVHU: return OPVCC(31, 439, 0, 0) /* sthux */ case AMOVWZ, AMOVW: return OPVCC(31, 151, 0, 0) /* stwx */ case AMOVWZU, AMOVWU: return OPVCC(31, 183, 0, 0) /* stwux */ case ASTSW: return OPVCC(31, 661, 0, 0) /* stswx */ case AMOVWBR: return OPVCC(31, 662, 0, 0) /* stwbrx */ case AMOVDBR: return OPVCC(31, 660, 0, 0) /* stdbrx */ case ASTBCCC: return OPVCC(31, 694, 0, 1) /* stbcx. */ case ASTHCCC: return OPVCC(31, 726, 0, 1) /* sthcx. */ case ASTWCCC: return OPVCC(31, 150, 0, 1) /* stwcx. */ case ASTDCCC: return OPVCC(31, 214, 0, 1) /* stwdx. */ case AMOVD: return OPVCC(31, 149, 0, 0) /* stdx */ case AMOVDU: return OPVCC(31, 181, 0, 0) /* stdux */ /* Vector (VMX/Altivec) instructions */ case ASTVEBX: return OPVCC(31, 135, 0, 0) /* stvebx - v2.03 */ case ASTVEHX: return OPVCC(31, 167, 0, 0) /* stvehx - v2.03 */ case ASTVEWX: return OPVCC(31, 199, 0, 0) /* stvewx - v2.03 */ case ASTVX: return OPVCC(31, 231, 0, 0) /* stvx - v2.03 */ case ASTVXL: return OPVCC(31, 487, 0, 0) /* stvxl - v2.03 */ /* End of vector instructions */ /* Vector scalar (VSX) instructions */ case ASTXVX: return OPVXX1(31, 396, 0) /* stxvx - v3.0 */ case ASTXVD2X: return OPVXX1(31, 972, 0) /* stxvd2x - v2.06 */ case ASTXVW4X: return OPVXX1(31, 908, 0) /* stxvw4x - v2.06 */ case ASTXVH8X: return OPVXX1(31, 940, 0) /* stxvh8x - v3.0 */ case ASTXVB16X: return OPVXX1(31, 1004, 0) /* stxvb16x - v3.0 */ case ASTXSDX: return OPVXX1(31, 716, 0) /* stxsdx - v2.06 */ case ASTXSIWX: return OPVXX1(31, 140, 0) /* stxsiwx - v2.07 */ /* End of vector scalar instructions */ } c.ctxt.Diag("unknown storex opcode %v", a) return 0 }