bdc9d6f5bf
(CONST): Don't define. * convex.h: Replace CONST with const. (CONST): Don't define. * dlx.h: Replace CONST with const. * or32.h (CONST): Don't define.
282 lines
9.9 KiB
C
282 lines
9.9 KiB
C
/* Table of opcodes for the AMD 29000 family.
|
|
Copyright 1990, 1991, 1993, 1994, 2002 Free Software Foundation, Inc.
|
|
|
|
This file is part of GDB and GAS.
|
|
|
|
This program is free software; you can redistribute it and/or modify
|
|
it under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation; either version 2 of the License, or
|
|
(at your option) any later version.
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
GNU General Public License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with this program; if not, write to the Free Software
|
|
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
|
|
|
|
struct a29k_opcode {
|
|
/* Name of the instruction. */
|
|
char *name;
|
|
|
|
/* Opcode word */
|
|
unsigned long opcode;
|
|
|
|
/* A string of characters which describe the operands.
|
|
Valid characters are:
|
|
, Itself. The character appears in the assembly code.
|
|
a RA. The register number is in bits 8-15 of the instruction.
|
|
b RB. The register number is in bits 0-7 of the instruction.
|
|
c RC. The register number is in bits 16-23 of the instruction.
|
|
i An immediate operand is in bits 0-7 of the instruction.
|
|
x Bits 0-7 and 16-23 of the instruction are bits 0-7 and 8-15
|
|
(respectively) of the immediate operand.
|
|
h Same as x but the instruction contains bits 16-31 of the
|
|
immediate operand.
|
|
X Same as x but bits 16-31 of the signed immediate operand
|
|
are set to 1 (thus the operand is always negative).
|
|
P,A Bits 0-7 and 16-23 of the instruction are bits 2-9 and 10-17
|
|
(respectively) of the immediate operand.
|
|
P=PC-relative, sign-extended to 32 bits.
|
|
A=Absolute, zero-extended to 32 bits.
|
|
e CE bit (bit 23) for a load/store instruction.
|
|
n Control field (bits 16-22) for a load/store instruction.
|
|
v Immediate operand in bits 16-23 of the instruction.
|
|
(used for trap numbers).
|
|
s SA. Special-purpose register number in bits 8-15
|
|
of the instruction.
|
|
u UI--bit 7 of the instruction.
|
|
r RND--bits 4-6 of the instruction.
|
|
d FD--bits 2-3 of the instruction.
|
|
f FS--bits 0-1 of the instruction.
|
|
I ID--bits 16-17 of the instruction.
|
|
|
|
Extensions for 29050:
|
|
|
|
d FMT--bits 2-3 of the instruction (not really new).
|
|
f ACN--bits 0-1 of the instruction (not really new).
|
|
F FUNC--Special function in bits 18-21 of the instruction.
|
|
C ACN--bits 16-17 specifying the accumlator register. */
|
|
char *args;
|
|
};
|
|
|
|
static const struct a29k_opcode a29k_opcodes[] =
|
|
{
|
|
|
|
{ "add", 0x14000000, "c,a,b" },
|
|
{ "add", 0x15000000, "c,a,i" },
|
|
{ "addc", 0x1c000000, "c,a,b" },
|
|
{ "addc", 0x1d000000, "c,a,i" },
|
|
{ "addcs", 0x18000000, "c,a,b" },
|
|
{ "addcs", 0x19000000, "c,a,i" },
|
|
{ "addcu", 0x1a000000, "c,a,b" },
|
|
{ "addcu", 0x1b000000, "c,a,i" },
|
|
{ "adds", 0x10000000, "c,a,b" },
|
|
{ "adds", 0x11000000, "c,a,i" },
|
|
{ "addu", 0x12000000, "c,a,b" },
|
|
{ "addu", 0x13000000, "c,a,i" },
|
|
{ "and", 0x90000000, "c,a,b" },
|
|
{ "and", 0x91000000, "c,a,i" },
|
|
{ "andn", 0x9c000000, "c,a,b" },
|
|
{ "andn", 0x9d000000, "c,a,i" },
|
|
{ "aseq", 0x70000000, "v,a,b" },
|
|
{ "aseq", 0x71000000, "v,a,i" },
|
|
{ "asge", 0x5c000000, "v,a,b" },
|
|
{ "asge", 0x5d000000, "v,a,i" },
|
|
{ "asgeu", 0x5e000000, "v,a,b" },
|
|
{ "asgeu", 0x5f000000, "v,a,i" },
|
|
{ "asgt", 0x58000000, "v,a,b" },
|
|
{ "asgt", 0x59000000, "v,a,i" },
|
|
{ "asgtu", 0x5a000000, "v,a,b" },
|
|
{ "asgtu", 0x5b000000, "v,a,i" },
|
|
{ "asle", 0x54000000, "v,a,b" },
|
|
{ "asle", 0x55000000, "v,a,i" },
|
|
{ "asleu", 0x56000000, "v,a,b" },
|
|
{ "asleu", 0x57000000, "v,a,i" },
|
|
{ "aslt", 0x50000000, "v,a,b" },
|
|
{ "aslt", 0x51000000, "v,a,i" },
|
|
{ "asltu", 0x52000000, "v,a,b" },
|
|
{ "asltu", 0x53000000, "v,a,i" },
|
|
{ "asneq", 0x72000000, "v,a,b" },
|
|
{ "asneq", 0x73000000, "v,a,i" },
|
|
{ "call", 0xa8000000, "a,P" },
|
|
{ "call", 0xa9000000, "a,A" },
|
|
{ "calli", 0xc8000000, "a,b" },
|
|
{ "class", 0xe6000000, "c,a,f" },
|
|
{ "clz", 0x08000000, "c,b" },
|
|
{ "clz", 0x09000000, "c,i" },
|
|
{ "const", 0x03000000, "a,x" },
|
|
{ "consth", 0x02000000, "a,h" },
|
|
{ "consthz", 0x05000000, "a,h" },
|
|
{ "constn", 0x01000000, "a,X" },
|
|
{ "convert", 0xe4000000, "c,a,u,r,d,f" },
|
|
{ "cpbyte", 0x2e000000, "c,a,b" },
|
|
{ "cpbyte", 0x2f000000, "c,a,i" },
|
|
{ "cpeq", 0x60000000, "c,a,b" },
|
|
{ "cpeq", 0x61000000, "c,a,i" },
|
|
{ "cpge", 0x4c000000, "c,a,b" },
|
|
{ "cpge", 0x4d000000, "c,a,i" },
|
|
{ "cpgeu", 0x4e000000, "c,a,b" },
|
|
{ "cpgeu", 0x4f000000, "c,a,i" },
|
|
{ "cpgt", 0x48000000, "c,a,b" },
|
|
{ "cpgt", 0x49000000, "c,a,i" },
|
|
{ "cpgtu", 0x4a000000, "c,a,b" },
|
|
{ "cpgtu", 0x4b000000, "c,a,i" },
|
|
{ "cple", 0x44000000, "c,a,b" },
|
|
{ "cple", 0x45000000, "c,a,i" },
|
|
{ "cpleu", 0x46000000, "c,a,b" },
|
|
{ "cpleu", 0x47000000, "c,a,i" },
|
|
{ "cplt", 0x40000000, "c,a,b" },
|
|
{ "cplt", 0x41000000, "c,a,i" },
|
|
{ "cpltu", 0x42000000, "c,a,b" },
|
|
{ "cpltu", 0x43000000, "c,a,i" },
|
|
{ "cpneq", 0x62000000, "c,a,b" },
|
|
{ "cpneq", 0x63000000, "c,a,i" },
|
|
{ "dadd", 0xf1000000, "c,a,b" },
|
|
{ "ddiv", 0xf7000000, "c,a,b" },
|
|
{ "deq", 0xeb000000, "c,a,b" },
|
|
{ "dge", 0xef000000, "c,a,b" },
|
|
{ "dgt", 0xed000000, "c,a,b" },
|
|
{ "div", 0x6a000000, "c,a,b" },
|
|
{ "div", 0x6b000000, "c,a,i" },
|
|
{ "div0", 0x68000000, "c,b" },
|
|
{ "div0", 0x69000000, "c,i" },
|
|
{ "divide", 0xe1000000, "c,a,b" },
|
|
{ "dividu", 0xe3000000, "c,a,b" },
|
|
{ "divl", 0x6c000000, "c,a,b" },
|
|
{ "divl", 0x6d000000, "c,a,i" },
|
|
{ "divrem", 0x6e000000, "c,a,b" },
|
|
{ "divrem", 0x6f000000, "c,a,i" },
|
|
{ "dmac", 0xd9000000, "F,C,a,b" },
|
|
{ "dmsm", 0xdb000000, "c,a,b" },
|
|
{ "dmul", 0xf5000000, "c,a,b" },
|
|
{ "dsub", 0xf3000000, "c,a,b" },
|
|
{ "emulate", 0xd7000000, "v,a,b" },
|
|
{ "exbyte", 0x0a000000, "c,a,b" },
|
|
{ "exbyte", 0x0b000000, "c,a,i" },
|
|
{ "exhw", 0x7c000000, "c,a,b" },
|
|
{ "exhw", 0x7d000000, "c,a,i" },
|
|
{ "exhws", 0x7e000000, "c,a" },
|
|
{ "extract", 0x7a000000, "c,a,b" },
|
|
{ "extract", 0x7b000000, "c,a,i" },
|
|
{ "fadd", 0xf0000000, "c,a,b" },
|
|
{ "fdiv", 0xf6000000, "c,a,b" },
|
|
{ "fdmul", 0xf9000000, "c,a,b" },
|
|
{ "feq", 0xea000000, "c,a,b" },
|
|
{ "fge", 0xee000000, "c,a,b" },
|
|
{ "fgt", 0xec000000, "c,a,b" },
|
|
{ "fmac", 0xd8000000, "F,C,a,b" },
|
|
{ "fmsm", 0xda000000, "c,a,b" },
|
|
{ "fmul", 0xf4000000, "c,a,b" },
|
|
{ "fsub", 0xf2000000, "c,a,b" },
|
|
{ "halt", 0x89000000, "" },
|
|
{ "inbyte", 0x0c000000, "c,a,b" },
|
|
{ "inbyte", 0x0d000000, "c,a,i" },
|
|
{ "inhw", 0x78000000, "c,a,b" },
|
|
{ "inhw", 0x79000000, "c,a,i" },
|
|
{ "inv", 0x9f000000, "I" },
|
|
{ "iret", 0x88000000, "" },
|
|
{ "iretinv", 0x8c000000, "I" },
|
|
{ "jmp", 0xa0000000, "P" },
|
|
{ "jmp", 0xa1000000, "A" },
|
|
{ "jmpf", 0xa4000000, "a,P" },
|
|
{ "jmpf", 0xa5000000, "a,A" },
|
|
{ "jmpfdec", 0xb4000000, "a,P" },
|
|
{ "jmpfdec", 0xb5000000, "a,A" },
|
|
{ "jmpfi", 0xc4000000, "a,b" },
|
|
{ "jmpi", 0xc0000000, "b" },
|
|
{ "jmpt", 0xac000000, "a,P" },
|
|
{ "jmpt", 0xad000000, "a,A" },
|
|
{ "jmpti", 0xcc000000, "a,b" },
|
|
{ "load", 0x16000000, "e,n,a,b" },
|
|
{ "load", 0x17000000, "e,n,a,i" },
|
|
{ "loadl", 0x06000000, "e,n,a,b" },
|
|
{ "loadl", 0x07000000, "e,n,a,i" },
|
|
{ "loadm", 0x36000000, "e,n,a,b" },
|
|
{ "loadm", 0x37000000, "e,n,a,i" },
|
|
{ "loadset", 0x26000000, "e,n,a,b" },
|
|
{ "loadset", 0x27000000, "e,n,a,i" },
|
|
{ "mfacc", 0xe9000100, "c,d,f" },
|
|
{ "mfsr", 0xc6000000, "c,s" },
|
|
{ "mftlb", 0xb6000000, "c,a" },
|
|
{ "mtacc", 0xe8010000, "a,d,f" },
|
|
{ "mtsr", 0xce000000, "s,b" },
|
|
{ "mtsrim", 0x04000000, "s,x" },
|
|
{ "mttlb", 0xbe000000, "a,b" },
|
|
{ "mul", 0x64000000, "c,a,b" },
|
|
{ "mul", 0x65000000, "c,a,i" },
|
|
{ "mull", 0x66000000, "c,a,b" },
|
|
{ "mull", 0x67000000, "c,a,i" },
|
|
{ "multiplu", 0xe2000000, "c,a,b" },
|
|
{ "multiply", 0xe0000000, "c,a,b" },
|
|
{ "multm", 0xde000000, "c,a,b" },
|
|
{ "multmu", 0xdf000000, "c,a,b" },
|
|
{ "mulu", 0x74000000, "c,a,b" },
|
|
{ "mulu", 0x75000000, "c,a,i" },
|
|
{ "nand", 0x9a000000, "c,a,b" },
|
|
{ "nand", 0x9b000000, "c,a,i" },
|
|
{ "nop", 0x70400101, "" },
|
|
{ "nor", 0x98000000, "c,a,b" },
|
|
{ "nor", 0x99000000, "c,a,i" },
|
|
{ "or", 0x92000000, "c,a,b" },
|
|
{ "or", 0x93000000, "c,a,i" },
|
|
{ "orn", 0xaa000000, "c,a,b" },
|
|
{ "orn", 0xab000000, "c,a,i" },
|
|
|
|
/* The description of "setip" in Chapter 8 ("instruction set") of the user's
|
|
manual claims that these are absolute register numbers. But section
|
|
7.2.1 explains that they are not. The latter is correct, so print
|
|
these normally ("lr0", "lr5", etc.). */
|
|
{ "setip", 0x9e000000, "c,a,b" },
|
|
|
|
{ "sll", 0x80000000, "c,a,b" },
|
|
{ "sll", 0x81000000, "c,a,i" },
|
|
{ "sqrt", 0xe5000000, "c,a,f" },
|
|
{ "sra", 0x86000000, "c,a,b" },
|
|
{ "sra", 0x87000000, "c,a,i" },
|
|
{ "srl", 0x82000000, "c,a,b" },
|
|
{ "srl", 0x83000000, "c,a,i" },
|
|
{ "store", 0x1e000000, "e,n,a,b" },
|
|
{ "store", 0x1f000000, "e,n,a,i" },
|
|
{ "storel", 0x0e000000, "e,n,a,b" },
|
|
{ "storel", 0x0f000000, "e,n,a,i" },
|
|
{ "storem", 0x3e000000, "e,n,a,b" },
|
|
{ "storem", 0x3f000000, "e,n,a,i" },
|
|
{ "sub", 0x24000000, "c,a,b" },
|
|
{ "sub", 0x25000000, "c,a,i" },
|
|
{ "subc", 0x2c000000, "c,a,b" },
|
|
{ "subc", 0x2d000000, "c,a,i" },
|
|
{ "subcs", 0x28000000, "c,a,b" },
|
|
{ "subcs", 0x29000000, "c,a,i" },
|
|
{ "subcu", 0x2a000000, "c,a,b" },
|
|
{ "subcu", 0x2b000000, "c,a,i" },
|
|
{ "subr", 0x34000000, "c,a,b" },
|
|
{ "subr", 0x35000000, "c,a,i" },
|
|
{ "subrc", 0x3c000000, "c,a,b" },
|
|
{ "subrc", 0x3d000000, "c,a,i" },
|
|
{ "subrcs", 0x38000000, "c,a,b" },
|
|
{ "subrcs", 0x39000000, "c,a,i" },
|
|
{ "subrcu", 0x3a000000, "c,a,b" },
|
|
{ "subrcu", 0x3b000000, "c,a,i" },
|
|
{ "subrs", 0x30000000, "c,a,b" },
|
|
{ "subrs", 0x31000000, "c,a,i" },
|
|
{ "subru", 0x32000000, "c,a,b" },
|
|
{ "subru", 0x33000000, "c,a,i" },
|
|
{ "subs", 0x20000000, "c,a,b" },
|
|
{ "subs", 0x21000000, "c,a,i" },
|
|
{ "subu", 0x22000000, "c,a,b" },
|
|
{ "subu", 0x23000000, "c,a,i" },
|
|
{ "xnor", 0x96000000, "c,a,b" },
|
|
{ "xnor", 0x97000000, "c,a,i" },
|
|
{ "xor", 0x94000000, "c,a,b" },
|
|
{ "xor", 0x95000000, "c,a,i" },
|
|
|
|
{ "", 0x0, "" } /* Dummy entry, not included in NUM_OPCODES. This
|
|
lets code examine entry i+1 without checking
|
|
if we've run off the end of the table. */
|
|
};
|
|
|
|
const unsigned int num_opcodes = (((sizeof a29k_opcodes) / (sizeof a29k_opcodes[0])) - 1);
|