42a4f53d2b
This commit applies all changes made after running the gdb/copyright.py script. Note that one file was flagged by the script, due to an invalid copyright header (gdb/unittests/basic_string_view/element_access/char/empty.cc). As the file was copied from GCC's libstdc++-v3 testsuite, this commit leaves this file untouched for the time being; a patch to fix the header was sent to gcc-patches first. gdb/ChangeLog: Update copyright year range in all GDB files.
2080 lines
45 KiB
C
2080 lines
45 KiB
C
/* This file is part of SIS (SPARC instruction simulator)
|
|
|
|
Copyright (C) 1995-2019 Free Software Foundation, Inc.
|
|
Contributed by Jiri Gaisler, European Space Agency
|
|
|
|
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 3 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, see <http://www.gnu.org/licenses/>. */
|
|
|
|
#include "config.h"
|
|
#include "sis.h"
|
|
#include <math.h>
|
|
#include <stdio.h>
|
|
|
|
extern int32 sis_verbose, sparclite;
|
|
int ext_irl = 0;
|
|
|
|
/* Load/store interlock delay */
|
|
#define FLSTHOLD 1
|
|
|
|
/* Load delay (delete if unwanted - speeds up simulation) */
|
|
#define LOAD_DEL 1
|
|
|
|
#define T_LD 2
|
|
#define T_LDD 3
|
|
#define T_ST 3
|
|
#define T_STD 4
|
|
#define T_LDST 4
|
|
#define T_JMPL 2
|
|
#define T_RETT 2
|
|
|
|
#define FSR_QNE 0x2000
|
|
#define FP_EXE_MODE 0
|
|
#define FP_EXC_PE 1
|
|
#define FP_EXC_MODE 2
|
|
|
|
#define FBA 8
|
|
#define FBN 0
|
|
#define FBNE 1
|
|
#define FBLG 2
|
|
#define FBUL 3
|
|
#define FBL 4
|
|
#define FBUG 5
|
|
#define FBG 6
|
|
#define FBU 7
|
|
#define FBA 8
|
|
#define FBE 9
|
|
#define FBUE 10
|
|
#define FBGE 11
|
|
#define FBUGE 12
|
|
#define FBLE 13
|
|
#define FBULE 14
|
|
#define FBO 15
|
|
|
|
#define FCC_E 0
|
|
#define FCC_L 1
|
|
#define FCC_G 2
|
|
#define FCC_U 3
|
|
|
|
#define PSR_ET 0x20
|
|
#define PSR_EF 0x1000
|
|
#define PSR_PS 0x40
|
|
#define PSR_S 0x80
|
|
#define PSR_N 0x0800000
|
|
#define PSR_Z 0x0400000
|
|
#define PSR_V 0x0200000
|
|
#define PSR_C 0x0100000
|
|
#define PSR_CC 0x0F00000
|
|
#define PSR_CWP 0x7
|
|
#define PSR_PIL 0x0f00
|
|
|
|
#define ICC_N (icc >> 3)
|
|
#define ICC_Z (icc >> 2)
|
|
#define ICC_V (icc >> 1)
|
|
#define ICC_C (icc)
|
|
|
|
#define FP_PRES (sregs->fpu_pres)
|
|
|
|
#define TRAP_IEXC 1
|
|
#define TRAP_UNIMP 2
|
|
#define TRAP_PRIVI 3
|
|
#define TRAP_FPDIS 4
|
|
#define TRAP_WOFL 5
|
|
#define TRAP_WUFL 6
|
|
#define TRAP_UNALI 7
|
|
#define TRAP_FPEXC 8
|
|
#define TRAP_DEXC 9
|
|
#define TRAP_TAG 10
|
|
#define TRAP_DIV0 0x2a
|
|
|
|
#define FSR_TT 0x1C000
|
|
#define FP_IEEE 0x04000
|
|
#define FP_UNIMP 0x0C000
|
|
#define FP_SEQ_ERR 0x10000
|
|
|
|
#define BICC_BN 0
|
|
#define BICC_BE 1
|
|
#define BICC_BLE 2
|
|
#define BICC_BL 3
|
|
#define BICC_BLEU 4
|
|
#define BICC_BCS 5
|
|
#define BICC_NEG 6
|
|
#define BICC_BVS 7
|
|
#define BICC_BA 8
|
|
#define BICC_BNE 9
|
|
#define BICC_BG 10
|
|
#define BICC_BGE 11
|
|
#define BICC_BGU 12
|
|
#define BICC_BCC 13
|
|
#define BICC_POS 14
|
|
#define BICC_BVC 15
|
|
|
|
#define INST_SIMM13 0x1fff
|
|
#define INST_RS2 0x1f
|
|
#define INST_I 0x2000
|
|
#define ADD 0x00
|
|
#define ADDCC 0x10
|
|
#define ADDX 0x08
|
|
#define ADDXCC 0x18
|
|
#define TADDCC 0x20
|
|
#define TSUBCC 0x21
|
|
#define TADDCCTV 0x22
|
|
#define TSUBCCTV 0x23
|
|
#define IAND 0x01
|
|
#define IANDCC 0x11
|
|
#define IANDN 0x05
|
|
#define IANDNCC 0x15
|
|
#define MULScc 0x24
|
|
#define DIVScc 0x1D
|
|
#define SMUL 0x0B
|
|
#define SMULCC 0x1B
|
|
#define UMUL 0x0A
|
|
#define UMULCC 0x1A
|
|
#define SDIV 0x0F
|
|
#define SDIVCC 0x1F
|
|
#define UDIV 0x0E
|
|
#define UDIVCC 0x1E
|
|
#define IOR 0x02
|
|
#define IORCC 0x12
|
|
#define IORN 0x06
|
|
#define IORNCC 0x16
|
|
#define SLL 0x25
|
|
#define SRA 0x27
|
|
#define SRL 0x26
|
|
#define SUB 0x04
|
|
#define SUBCC 0x14
|
|
#define SUBX 0x0C
|
|
#define SUBXCC 0x1C
|
|
#define IXNOR 0x07
|
|
#define IXNORCC 0x17
|
|
#define IXOR 0x03
|
|
#define IXORCC 0x13
|
|
#define SETHI 0x04
|
|
#define BICC 0x02
|
|
#define FPBCC 0x06
|
|
#define RDY 0x28
|
|
#define RDPSR 0x29
|
|
#define RDWIM 0x2A
|
|
#define RDTBR 0x2B
|
|
#define SCAN 0x2C
|
|
#define WRY 0x30
|
|
#define WRPSR 0x31
|
|
#define WRWIM 0x32
|
|
#define WRTBR 0x33
|
|
#define JMPL 0x38
|
|
#define RETT 0x39
|
|
#define TICC 0x3A
|
|
#define SAVE 0x3C
|
|
#define RESTORE 0x3D
|
|
#define LDD 0x03
|
|
#define LDDA 0x13
|
|
#define LD 0x00
|
|
#define LDA 0x10
|
|
#define LDF 0x20
|
|
#define LDDF 0x23
|
|
#define LDSTUB 0x0D
|
|
#define LDSTUBA 0x1D
|
|
#define LDUB 0x01
|
|
#define LDUBA 0x11
|
|
#define LDSB 0x09
|
|
#define LDSBA 0x19
|
|
#define LDUH 0x02
|
|
#define LDUHA 0x12
|
|
#define LDSH 0x0A
|
|
#define LDSHA 0x1A
|
|
#define LDFSR 0x21
|
|
#define ST 0x04
|
|
#define STA 0x14
|
|
#define STB 0x05
|
|
#define STBA 0x15
|
|
#define STD 0x07
|
|
#define STDA 0x17
|
|
#define STF 0x24
|
|
#define STDFQ 0x26
|
|
#define STDF 0x27
|
|
#define STFSR 0x25
|
|
#define STH 0x06
|
|
#define STHA 0x16
|
|
#define SWAP 0x0F
|
|
#define SWAPA 0x1F
|
|
#define FLUSH 0x3B
|
|
|
|
#define SIGN_BIT 0x80000000
|
|
|
|
/* # of cycles overhead when a trap is taken */
|
|
#define TRAP_C 3
|
|
|
|
/* Forward declarations */
|
|
|
|
static uint32 sub_cc (uint32 psr, int32 operand1, int32 operand2,
|
|
int32 result);
|
|
static uint32 add_cc (uint32 psr, int32 operand1, int32 operand2,
|
|
int32 result);
|
|
static void log_cc (int32 result, struct pstate *sregs);
|
|
static int fpexec (uint32 op3, uint32 rd, uint32 rs1, uint32 rs2,
|
|
struct pstate *sregs);
|
|
static int chk_asi (struct pstate *sregs, uint32 *asi, uint32 op3);
|
|
|
|
|
|
extern struct estate ebase;
|
|
extern int32 nfp,ift;
|
|
|
|
#ifdef ERRINJ
|
|
extern uint32 errtt, errftt;
|
|
#endif
|
|
|
|
static uint32
|
|
sub_cc(psr, operand1, operand2, result)
|
|
uint32 psr;
|
|
int32 operand1;
|
|
int32 operand2;
|
|
int32 result;
|
|
{
|
|
psr = ((psr & ~PSR_N) | ((result >> 8) & PSR_N));
|
|
if (result)
|
|
psr &= ~PSR_Z;
|
|
else
|
|
psr |= PSR_Z;
|
|
psr = (psr & ~PSR_V) | ((((operand1 & ~operand2 & ~result) |
|
|
(~operand1 & operand2 & result)) >> 10) & PSR_V);
|
|
psr = (psr & ~PSR_C) | ((((~operand1 & operand2) |
|
|
((~operand1 | operand2) & result)) >> 11) & PSR_C);
|
|
return psr;
|
|
}
|
|
|
|
uint32
|
|
add_cc(psr, operand1, operand2, result)
|
|
uint32 psr;
|
|
int32 operand1;
|
|
int32 operand2;
|
|
int32 result;
|
|
{
|
|
psr = ((psr & ~PSR_N) | ((result >> 8) & PSR_N));
|
|
if (result)
|
|
psr &= ~PSR_Z;
|
|
else
|
|
psr |= PSR_Z;
|
|
psr = (psr & ~PSR_V) | ((((operand1 & operand2 & ~result) |
|
|
(~operand1 & ~operand2 & result)) >> 10) & PSR_V);
|
|
psr = (psr & ~PSR_C) | ((((operand1 & operand2) |
|
|
((operand1 | operand2) & ~result)) >> 11) & PSR_C);
|
|
return psr;
|
|
}
|
|
|
|
static void
|
|
log_cc(result, sregs)
|
|
int32 result;
|
|
struct pstate *sregs;
|
|
{
|
|
sregs->psr &= ~(PSR_CC); /* Zero CC bits */
|
|
sregs->psr = (sregs->psr | ((result >> 8) & PSR_N));
|
|
if (result == 0)
|
|
sregs->psr |= PSR_Z;
|
|
}
|
|
|
|
/* Add two unsigned 32-bit integers, and calculate the carry out. */
|
|
|
|
static uint32
|
|
add32 (uint32 n1, uint32 n2, int *carry)
|
|
{
|
|
uint32 result = n1 + n2;
|
|
|
|
*carry = result < n1 || result < n2;
|
|
return result;
|
|
}
|
|
|
|
/* Multiply two 32-bit integers. */
|
|
|
|
static void
|
|
mul64 (uint32 n1, uint32 n2, uint32 *result_hi, uint32 *result_lo, int msigned)
|
|
{
|
|
uint32 lo, mid1, mid2, hi, reg_lo, reg_hi;
|
|
int carry;
|
|
int sign = 0;
|
|
|
|
/* If this is a signed multiply, calculate the sign of the result
|
|
and make the operands positive. */
|
|
if (msigned)
|
|
{
|
|
sign = (n1 ^ n2) & SIGN_BIT;
|
|
if (n1 & SIGN_BIT)
|
|
n1 = -n1;
|
|
if (n2 & SIGN_BIT)
|
|
n2 = -n2;
|
|
|
|
}
|
|
|
|
/* We can split the 32x32 into four 16x16 operations. This ensures
|
|
that we do not lose precision on 32bit only hosts: */
|
|
lo = ((n1 & 0xFFFF) * (n2 & 0xFFFF));
|
|
mid1 = ((n1 & 0xFFFF) * ((n2 >> 16) & 0xFFFF));
|
|
mid2 = (((n1 >> 16) & 0xFFFF) * (n2 & 0xFFFF));
|
|
hi = (((n1 >> 16) & 0xFFFF) * ((n2 >> 16) & 0xFFFF));
|
|
|
|
/* We now need to add all of these results together, taking care
|
|
to propogate the carries from the additions: */
|
|
reg_lo = add32 (lo, (mid1 << 16), &carry);
|
|
reg_hi = carry;
|
|
reg_lo = add32 (reg_lo, (mid2 << 16), &carry);
|
|
reg_hi += (carry + ((mid1 >> 16) & 0xFFFF) + ((mid2 >> 16) & 0xFFFF) + hi);
|
|
|
|
/* Negate result if necessary. */
|
|
if (sign)
|
|
{
|
|
reg_hi = ~ reg_hi;
|
|
reg_lo = - reg_lo;
|
|
if (reg_lo == 0)
|
|
reg_hi++;
|
|
}
|
|
|
|
*result_lo = reg_lo;
|
|
*result_hi = reg_hi;
|
|
}
|
|
|
|
|
|
/* Divide a 64-bit integer by a 32-bit integer. We cheat and assume
|
|
that the host compiler supports long long operations. */
|
|
|
|
static void
|
|
div64 (uint32 n1_hi, uint32 n1_low, uint32 n2, uint32 *result, int msigned)
|
|
{
|
|
uint64 n1;
|
|
|
|
n1 = ((uint64) n1_hi) << 32;
|
|
n1 |= ((uint64) n1_low) & 0xffffffff;
|
|
|
|
if (msigned)
|
|
{
|
|
int64 n1_s = (int64) n1;
|
|
int32 n2_s = (int32) n2;
|
|
n1_s = n1_s / n2_s;
|
|
n1 = (uint64) n1_s;
|
|
}
|
|
else
|
|
n1 = n1 / n2;
|
|
|
|
*result = (uint32) (n1 & 0xffffffff);
|
|
}
|
|
|
|
|
|
static int
|
|
extract_short (uint32 data, uint32 address)
|
|
{
|
|
return ((data >> ((2 - (address & 2)) * 8)) & 0xffff);
|
|
}
|
|
|
|
static int
|
|
extract_short_signed (uint32 data, uint32 address)
|
|
{
|
|
uint32 tmp = ((data >> ((2 - (address & 2)) * 8)) & 0xffff);
|
|
if (tmp & 0x8000)
|
|
tmp |= 0xffff0000;
|
|
return tmp;
|
|
}
|
|
|
|
static int
|
|
extract_byte (uint32 data, uint32 address)
|
|
{
|
|
return ((data >> ((3 - (address & 3)) * 8)) & 0xff);
|
|
}
|
|
|
|
static int
|
|
extract_byte_signed (uint32 data, uint32 address)
|
|
{
|
|
uint32 tmp = ((data >> ((3 - (address & 3)) * 8)) & 0xff);
|
|
if (tmp & 0x80)
|
|
tmp |= 0xffffff00;
|
|
return tmp;
|
|
}
|
|
|
|
int
|
|
dispatch_instruction(sregs)
|
|
struct pstate *sregs;
|
|
{
|
|
|
|
uint32 cwp, op, op2, op3, asi, rd, cond, rs1,
|
|
rs2;
|
|
uint32 ldep, icc;
|
|
int32 operand1, operand2, *rdd, result, eicc,
|
|
new_cwp;
|
|
int32 pc, npc, data, address, ws, mexc, fcc;
|
|
int32 ddata[2];
|
|
|
|
sregs->ninst++;
|
|
cwp = ((sregs->psr & PSR_CWP) << 4);
|
|
op = sregs->inst >> 30;
|
|
pc = sregs->npc;
|
|
npc = sregs->npc + 4;
|
|
op3 = rd = rs1 = operand2 = eicc = 0;
|
|
rdd = 0;
|
|
if (op & 2) {
|
|
|
|
op3 = (sregs->inst >> 19) & 0x3f;
|
|
rs1 = (sregs->inst >> 14) & 0x1f;
|
|
rd = (sregs->inst >> 25) & 0x1f;
|
|
|
|
#ifdef LOAD_DEL
|
|
|
|
/* Check if load dependecy is possible */
|
|
if (ebase.simtime <= sregs->ildtime)
|
|
ldep = (((op3 & 0x38) != 0x28) && ((op3 & 0x3e) != 0x34) && (sregs->ildreg != 0));
|
|
else
|
|
ldep = 0;
|
|
if (sregs->inst & INST_I) {
|
|
if (ldep && (sregs->ildreg == rs1))
|
|
sregs->hold++;
|
|
operand2 = sregs->inst;
|
|
operand2 = ((operand2 << 19) >> 19); /* sign extend */
|
|
} else {
|
|
rs2 = sregs->inst & INST_RS2;
|
|
if (rs2 > 7)
|
|
operand2 = sregs->r[(cwp + rs2) & 0x7f];
|
|
else
|
|
operand2 = sregs->g[rs2];
|
|
if (ldep && ((sregs->ildreg == rs1) || (sregs->ildreg == rs2)))
|
|
sregs->hold++;
|
|
}
|
|
#else
|
|
if (sregs->inst & INST_I) {
|
|
operand2 = sregs->inst;
|
|
operand2 = ((operand2 << 19) >> 19); /* sign extend */
|
|
} else {
|
|
rs2 = sregs->inst & INST_RS2;
|
|
if (rs2 > 7)
|
|
operand2 = sregs->r[(cwp + rs2) & 0x7f];
|
|
else
|
|
operand2 = sregs->g[rs2];
|
|
}
|
|
#endif
|
|
|
|
if (rd > 7)
|
|
rdd = &(sregs->r[(cwp + rd) & 0x7f]);
|
|
else
|
|
rdd = &(sregs->g[rd]);
|
|
if (rs1 > 7)
|
|
rs1 = sregs->r[(cwp + rs1) & 0x7f];
|
|
else
|
|
rs1 = sregs->g[rs1];
|
|
}
|
|
switch (op) {
|
|
case 0:
|
|
op2 = (sregs->inst >> 22) & 0x7;
|
|
switch (op2) {
|
|
case SETHI:
|
|
rd = (sregs->inst >> 25) & 0x1f;
|
|
if (rd > 7)
|
|
rdd = &(sregs->r[(cwp + rd) & 0x7f]);
|
|
else
|
|
rdd = &(sregs->g[rd]);
|
|
*rdd = sregs->inst << 10;
|
|
break;
|
|
case BICC:
|
|
#ifdef STAT
|
|
sregs->nbranch++;
|
|
#endif
|
|
icc = sregs->psr >> 20;
|
|
cond = ((sregs->inst >> 25) & 0x0f);
|
|
switch (cond) {
|
|
case BICC_BN:
|
|
eicc = 0;
|
|
break;
|
|
case BICC_BE:
|
|
eicc = ICC_Z;
|
|
break;
|
|
case BICC_BLE:
|
|
eicc = ICC_Z | (ICC_N ^ ICC_V);
|
|
break;
|
|
case BICC_BL:
|
|
eicc = (ICC_N ^ ICC_V);
|
|
break;
|
|
case BICC_BLEU:
|
|
eicc = ICC_C | ICC_Z;
|
|
break;
|
|
case BICC_BCS:
|
|
eicc = ICC_C;
|
|
break;
|
|
case BICC_NEG:
|
|
eicc = ICC_N;
|
|
break;
|
|
case BICC_BVS:
|
|
eicc = ICC_V;
|
|
break;
|
|
case BICC_BA:
|
|
eicc = 1;
|
|
if (sregs->inst & 0x20000000)
|
|
sregs->annul = 1;
|
|
break;
|
|
case BICC_BNE:
|
|
eicc = ~(ICC_Z);
|
|
break;
|
|
case BICC_BG:
|
|
eicc = ~(ICC_Z | (ICC_N ^ ICC_V));
|
|
break;
|
|
case BICC_BGE:
|
|
eicc = ~(ICC_N ^ ICC_V);
|
|
break;
|
|
case BICC_BGU:
|
|
eicc = ~(ICC_C | ICC_Z);
|
|
break;
|
|
case BICC_BCC:
|
|
eicc = ~(ICC_C);
|
|
break;
|
|
case BICC_POS:
|
|
eicc = ~(ICC_N);
|
|
break;
|
|
case BICC_BVC:
|
|
eicc = ~(ICC_V);
|
|
break;
|
|
}
|
|
if (eicc & 1) {
|
|
operand1 = sregs->inst;
|
|
operand1 = ((operand1 << 10) >> 8); /* sign extend */
|
|
npc = sregs->pc + operand1;
|
|
} else {
|
|
if (sregs->inst & 0x20000000)
|
|
sregs->annul = 1;
|
|
}
|
|
break;
|
|
case FPBCC:
|
|
#ifdef STAT
|
|
sregs->nbranch++;
|
|
#endif
|
|
if (!((sregs->psr & PSR_EF) && FP_PRES)) {
|
|
sregs->trap = TRAP_FPDIS;
|
|
break;
|
|
}
|
|
if (ebase.simtime < sregs->ftime) {
|
|
sregs->ftime = ebase.simtime + sregs->hold;
|
|
}
|
|
cond = ((sregs->inst >> 25) & 0x0f);
|
|
fcc = (sregs->fsr >> 10) & 0x3;
|
|
switch (cond) {
|
|
case FBN:
|
|
eicc = 0;
|
|
break;
|
|
case FBNE:
|
|
eicc = (fcc != FCC_E);
|
|
break;
|
|
case FBLG:
|
|
eicc = (fcc == FCC_L) || (fcc == FCC_G);
|
|
break;
|
|
case FBUL:
|
|
eicc = (fcc == FCC_L) || (fcc == FCC_U);
|
|
break;
|
|
case FBL:
|
|
eicc = (fcc == FCC_L);
|
|
break;
|
|
case FBUG:
|
|
eicc = (fcc == FCC_G) || (fcc == FCC_U);
|
|
break;
|
|
case FBG:
|
|
eicc = (fcc == FCC_G);
|
|
break;
|
|
case FBU:
|
|
eicc = (fcc == FCC_U);
|
|
break;
|
|
case FBA:
|
|
eicc = 1;
|
|
if (sregs->inst & 0x20000000)
|
|
sregs->annul = 1;
|
|
break;
|
|
case FBE:
|
|
eicc = !(fcc != FCC_E);
|
|
break;
|
|
case FBUE:
|
|
eicc = !((fcc == FCC_L) || (fcc == FCC_G));
|
|
break;
|
|
case FBGE:
|
|
eicc = !((fcc == FCC_L) || (fcc == FCC_U));
|
|
break;
|
|
case FBUGE:
|
|
eicc = !(fcc == FCC_L);
|
|
break;
|
|
case FBLE:
|
|
eicc = !((fcc == FCC_G) || (fcc == FCC_U));
|
|
break;
|
|
case FBULE:
|
|
eicc = !(fcc == FCC_G);
|
|
break;
|
|
case FBO:
|
|
eicc = !(fcc == FCC_U);
|
|
break;
|
|
}
|
|
if (eicc) {
|
|
operand1 = sregs->inst;
|
|
operand1 = ((operand1 << 10) >> 8); /* sign extend */
|
|
npc = sregs->pc + operand1;
|
|
} else {
|
|
if (sregs->inst & 0x20000000)
|
|
sregs->annul = 1;
|
|
}
|
|
break;
|
|
|
|
default:
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
break;
|
|
case 1: /* CALL */
|
|
#ifdef STAT
|
|
sregs->nbranch++;
|
|
#endif
|
|
sregs->r[(cwp + 15) & 0x7f] = sregs->pc;
|
|
npc = sregs->pc + (sregs->inst << 2);
|
|
break;
|
|
|
|
case 2:
|
|
if ((op3 >> 1) == 0x1a) {
|
|
if (!((sregs->psr & PSR_EF) && FP_PRES)) {
|
|
sregs->trap = TRAP_FPDIS;
|
|
} else {
|
|
rs1 = (sregs->inst >> 14) & 0x1f;
|
|
rs2 = sregs->inst & 0x1f;
|
|
sregs->trap = fpexec(op3, rd, rs1, rs2, sregs);
|
|
}
|
|
} else {
|
|
|
|
switch (op3) {
|
|
case TICC:
|
|
icc = sregs->psr >> 20;
|
|
cond = ((sregs->inst >> 25) & 0x0f);
|
|
switch (cond) {
|
|
case BICC_BN:
|
|
eicc = 0;
|
|
break;
|
|
case BICC_BE:
|
|
eicc = ICC_Z;
|
|
break;
|
|
case BICC_BLE:
|
|
eicc = ICC_Z | (ICC_N ^ ICC_V);
|
|
break;
|
|
case BICC_BL:
|
|
eicc = (ICC_N ^ ICC_V);
|
|
break;
|
|
case BICC_BLEU:
|
|
eicc = ICC_C | ICC_Z;
|
|
break;
|
|
case BICC_BCS:
|
|
eicc = ICC_C;
|
|
break;
|
|
case BICC_NEG:
|
|
eicc = ICC_N;
|
|
break;
|
|
case BICC_BVS:
|
|
eicc = ICC_V;
|
|
break;
|
|
case BICC_BA:
|
|
eicc = 1;
|
|
break;
|
|
case BICC_BNE:
|
|
eicc = ~(ICC_Z);
|
|
break;
|
|
case BICC_BG:
|
|
eicc = ~(ICC_Z | (ICC_N ^ ICC_V));
|
|
break;
|
|
case BICC_BGE:
|
|
eicc = ~(ICC_N ^ ICC_V);
|
|
break;
|
|
case BICC_BGU:
|
|
eicc = ~(ICC_C | ICC_Z);
|
|
break;
|
|
case BICC_BCC:
|
|
eicc = ~(ICC_C);
|
|
break;
|
|
case BICC_POS:
|
|
eicc = ~(ICC_N);
|
|
break;
|
|
case BICC_BVC:
|
|
eicc = ~(ICC_V);
|
|
break;
|
|
}
|
|
if (eicc & 1) {
|
|
sregs->trap = (0x80 | ((rs1 + operand2) & 0x7f));
|
|
}
|
|
break;
|
|
|
|
case MULScc:
|
|
operand1 =
|
|
(((sregs->psr & PSR_V) ^ ((sregs->psr & PSR_N) >> 2))
|
|
<< 10) | (rs1 >> 1);
|
|
if ((sregs->y & 1) == 0)
|
|
operand2 = 0;
|
|
*rdd = operand1 + operand2;
|
|
sregs->y = (rs1 << 31) | (sregs->y >> 1);
|
|
sregs->psr = add_cc(sregs->psr, operand1, operand2, *rdd);
|
|
break;
|
|
case DIVScc:
|
|
{
|
|
int sign;
|
|
uint32 result, remainder;
|
|
int c0, y31;
|
|
|
|
if (!sparclite) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
|
|
sign = ((sregs->psr & PSR_V) != 0) ^ ((sregs->psr & PSR_N) != 0);
|
|
|
|
remainder = (sregs->y << 1) | (rs1 >> 31);
|
|
|
|
/* If true sign is positive, calculate remainder - divisor.
|
|
Otherwise, calculate remainder + divisor. */
|
|
if (sign == 0)
|
|
operand2 = ~operand2 + 1;
|
|
result = remainder + operand2;
|
|
|
|
/* The SPARClite User's Manual is not clear on how
|
|
the "carry out" of the above ALU operation is to
|
|
be calculated. From trial and error tests
|
|
on the the chip itself, it appears that it is
|
|
a normal addition carry, and not a subtraction borrow,
|
|
even in cases where the divisor is subtracted
|
|
from the remainder. FIXME: get the true story
|
|
from Fujitsu. */
|
|
c0 = result < (uint32) remainder
|
|
|| result < (uint32) operand2;
|
|
|
|
if (result & 0x80000000)
|
|
sregs->psr |= PSR_N;
|
|
else
|
|
sregs->psr &= ~PSR_N;
|
|
|
|
y31 = (sregs->y & 0x80000000) == 0x80000000;
|
|
|
|
if (result == 0 && sign == y31)
|
|
sregs->psr |= PSR_Z;
|
|
else
|
|
sregs->psr &= ~PSR_Z;
|
|
|
|
sign = (sign && !y31) || (!c0 && (sign || !y31));
|
|
|
|
if (sign ^ (result >> 31))
|
|
sregs->psr |= PSR_V;
|
|
else
|
|
sregs->psr &= ~PSR_V;
|
|
|
|
if (!sign)
|
|
sregs->psr |= PSR_C;
|
|
else
|
|
sregs->psr &= ~PSR_C;
|
|
|
|
sregs->y = result;
|
|
|
|
if (rd != 0)
|
|
*rdd = (rs1 << 1) | !sign;
|
|
}
|
|
break;
|
|
case SMUL:
|
|
{
|
|
mul64 (rs1, operand2, &sregs->y, rdd, 1);
|
|
}
|
|
break;
|
|
case SMULCC:
|
|
{
|
|
uint32 result;
|
|
|
|
mul64 (rs1, operand2, &sregs->y, &result, 1);
|
|
|
|
if (result & 0x80000000)
|
|
sregs->psr |= PSR_N;
|
|
else
|
|
sregs->psr &= ~PSR_N;
|
|
|
|
if (result == 0)
|
|
sregs->psr |= PSR_Z;
|
|
else
|
|
sregs->psr &= ~PSR_Z;
|
|
|
|
*rdd = result;
|
|
}
|
|
break;
|
|
case UMUL:
|
|
{
|
|
mul64 (rs1, operand2, &sregs->y, rdd, 0);
|
|
}
|
|
break;
|
|
case UMULCC:
|
|
{
|
|
uint32 result;
|
|
|
|
mul64 (rs1, operand2, &sregs->y, &result, 0);
|
|
|
|
if (result & 0x80000000)
|
|
sregs->psr |= PSR_N;
|
|
else
|
|
sregs->psr &= ~PSR_N;
|
|
|
|
if (result == 0)
|
|
sregs->psr |= PSR_Z;
|
|
else
|
|
sregs->psr &= ~PSR_Z;
|
|
|
|
*rdd = result;
|
|
}
|
|
break;
|
|
case SDIV:
|
|
{
|
|
if (sparclite) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
|
|
if (operand2 == 0) {
|
|
sregs->trap = TRAP_DIV0;
|
|
break;
|
|
}
|
|
|
|
div64 (sregs->y, rs1, operand2, rdd, 1);
|
|
}
|
|
break;
|
|
case SDIVCC:
|
|
{
|
|
uint32 result;
|
|
|
|
if (sparclite) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
|
|
if (operand2 == 0) {
|
|
sregs->trap = TRAP_DIV0;
|
|
break;
|
|
}
|
|
|
|
div64 (sregs->y, rs1, operand2, &result, 1);
|
|
|
|
if (result & 0x80000000)
|
|
sregs->psr |= PSR_N;
|
|
else
|
|
sregs->psr &= ~PSR_N;
|
|
|
|
if (result == 0)
|
|
sregs->psr |= PSR_Z;
|
|
else
|
|
sregs->psr &= ~PSR_Z;
|
|
|
|
/* FIXME: should set overflow flag correctly. */
|
|
sregs->psr &= ~(PSR_C | PSR_V);
|
|
|
|
*rdd = result;
|
|
}
|
|
break;
|
|
case UDIV:
|
|
{
|
|
if (sparclite) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
|
|
if (operand2 == 0) {
|
|
sregs->trap = TRAP_DIV0;
|
|
break;
|
|
}
|
|
|
|
div64 (sregs->y, rs1, operand2, rdd, 0);
|
|
}
|
|
break;
|
|
case UDIVCC:
|
|
{
|
|
uint32 result;
|
|
|
|
if (sparclite) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
|
|
if (operand2 == 0) {
|
|
sregs->trap = TRAP_DIV0;
|
|
break;
|
|
}
|
|
|
|
div64 (sregs->y, rs1, operand2, &result, 0);
|
|
|
|
if (result & 0x80000000)
|
|
sregs->psr |= PSR_N;
|
|
else
|
|
sregs->psr &= ~PSR_N;
|
|
|
|
if (result == 0)
|
|
sregs->psr |= PSR_Z;
|
|
else
|
|
sregs->psr &= ~PSR_Z;
|
|
|
|
/* FIXME: should set overflow flag correctly. */
|
|
sregs->psr &= ~(PSR_C | PSR_V);
|
|
|
|
*rdd = result;
|
|
}
|
|
break;
|
|
case IXNOR:
|
|
*rdd = rs1 ^ ~operand2;
|
|
break;
|
|
case IXNORCC:
|
|
*rdd = rs1 ^ ~operand2;
|
|
log_cc(*rdd, sregs);
|
|
break;
|
|
case IXOR:
|
|
*rdd = rs1 ^ operand2;
|
|
break;
|
|
case IXORCC:
|
|
*rdd = rs1 ^ operand2;
|
|
log_cc(*rdd, sregs);
|
|
break;
|
|
case IOR:
|
|
*rdd = rs1 | operand2;
|
|
break;
|
|
case IORCC:
|
|
*rdd = rs1 | operand2;
|
|
log_cc(*rdd, sregs);
|
|
break;
|
|
case IORN:
|
|
*rdd = rs1 | ~operand2;
|
|
break;
|
|
case IORNCC:
|
|
*rdd = rs1 | ~operand2;
|
|
log_cc(*rdd, sregs);
|
|
break;
|
|
case IANDNCC:
|
|
*rdd = rs1 & ~operand2;
|
|
log_cc(*rdd, sregs);
|
|
break;
|
|
case IANDN:
|
|
*rdd = rs1 & ~operand2;
|
|
break;
|
|
case IAND:
|
|
*rdd = rs1 & operand2;
|
|
break;
|
|
case IANDCC:
|
|
*rdd = rs1 & operand2;
|
|
log_cc(*rdd, sregs);
|
|
break;
|
|
case SUB:
|
|
*rdd = rs1 - operand2;
|
|
break;
|
|
case SUBCC:
|
|
*rdd = rs1 - operand2;
|
|
sregs->psr = sub_cc(sregs->psr, rs1, operand2, *rdd);
|
|
break;
|
|
case SUBX:
|
|
*rdd = rs1 - operand2 - ((sregs->psr >> 20) & 1);
|
|
break;
|
|
case SUBXCC:
|
|
*rdd = rs1 - operand2 - ((sregs->psr >> 20) & 1);
|
|
sregs->psr = sub_cc(sregs->psr, rs1, operand2, *rdd);
|
|
break;
|
|
case ADD:
|
|
*rdd = rs1 + operand2;
|
|
break;
|
|
case ADDCC:
|
|
*rdd = rs1 + operand2;
|
|
sregs->psr = add_cc(sregs->psr, rs1, operand2, *rdd);
|
|
break;
|
|
case ADDX:
|
|
*rdd = rs1 + operand2 + ((sregs->psr >> 20) & 1);
|
|
break;
|
|
case ADDXCC:
|
|
*rdd = rs1 + operand2 + ((sregs->psr >> 20) & 1);
|
|
sregs->psr = add_cc(sregs->psr, rs1, operand2, *rdd);
|
|
break;
|
|
case TADDCC:
|
|
*rdd = rs1 + operand2;
|
|
sregs->psr = add_cc(sregs->psr, rs1, operand2, *rdd);
|
|
if ((rs1 | operand2) & 0x3)
|
|
sregs->psr |= PSR_V;
|
|
break;
|
|
case TSUBCC:
|
|
*rdd = rs1 - operand2;
|
|
sregs->psr = sub_cc (sregs->psr, rs1, operand2, *rdd);
|
|
if ((rs1 | operand2) & 0x3)
|
|
sregs->psr |= PSR_V;
|
|
break;
|
|
case TADDCCTV:
|
|
*rdd = rs1 + operand2;
|
|
result = add_cc(0, rs1, operand2, *rdd);
|
|
if ((rs1 | operand2) & 0x3)
|
|
result |= PSR_V;
|
|
if (result & PSR_V) {
|
|
sregs->trap = TRAP_TAG;
|
|
} else {
|
|
sregs->psr = (sregs->psr & ~PSR_CC) | result;
|
|
}
|
|
break;
|
|
case TSUBCCTV:
|
|
*rdd = rs1 - operand2;
|
|
result = add_cc (0, rs1, operand2, *rdd);
|
|
if ((rs1 | operand2) & 0x3)
|
|
result |= PSR_V;
|
|
if (result & PSR_V)
|
|
{
|
|
sregs->trap = TRAP_TAG;
|
|
}
|
|
else
|
|
{
|
|
sregs->psr = (sregs->psr & ~PSR_CC) | result;
|
|
}
|
|
break;
|
|
case SLL:
|
|
*rdd = rs1 << (operand2 & 0x1f);
|
|
break;
|
|
case SRL:
|
|
*rdd = rs1 >> (operand2 & 0x1f);
|
|
break;
|
|
case SRA:
|
|
*rdd = ((int) rs1) >> (operand2 & 0x1f);
|
|
break;
|
|
case FLUSH:
|
|
if (ift) sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
case SAVE:
|
|
new_cwp = ((sregs->psr & PSR_CWP) - 1) & PSR_CWP;
|
|
if (sregs->wim & (1 << new_cwp)) {
|
|
sregs->trap = TRAP_WOFL;
|
|
break;
|
|
}
|
|
if (rd > 7)
|
|
rdd = &(sregs->r[((new_cwp << 4) + rd) & 0x7f]);
|
|
*rdd = rs1 + operand2;
|
|
sregs->psr = (sregs->psr & ~PSR_CWP) | new_cwp;
|
|
break;
|
|
case RESTORE:
|
|
|
|
new_cwp = ((sregs->psr & PSR_CWP) + 1) & PSR_CWP;
|
|
if (sregs->wim & (1 << new_cwp)) {
|
|
sregs->trap = TRAP_WUFL;
|
|
break;
|
|
}
|
|
if (rd > 7)
|
|
rdd = &(sregs->r[((new_cwp << 4) + rd) & 0x7f]);
|
|
*rdd = rs1 + operand2;
|
|
sregs->psr = (sregs->psr & ~PSR_CWP) | new_cwp;
|
|
break;
|
|
case RDPSR:
|
|
if (!(sregs->psr & PSR_S)) {
|
|
sregs->trap = TRAP_PRIVI;
|
|
break;
|
|
}
|
|
*rdd = sregs->psr;
|
|
break;
|
|
case RDY:
|
|
if (!sparclite)
|
|
*rdd = sregs->y;
|
|
else {
|
|
int rs1_is_asr = (sregs->inst >> 14) & 0x1f;
|
|
if ( 0 == rs1_is_asr )
|
|
*rdd = sregs->y;
|
|
else if ( 17 == rs1_is_asr )
|
|
*rdd = sregs->asr17;
|
|
else {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
}
|
|
break;
|
|
case RDWIM:
|
|
if (!(sregs->psr & PSR_S)) {
|
|
sregs->trap = TRAP_PRIVI;
|
|
break;
|
|
}
|
|
*rdd = sregs->wim;
|
|
break;
|
|
case RDTBR:
|
|
if (!(sregs->psr & PSR_S)) {
|
|
sregs->trap = TRAP_PRIVI;
|
|
break;
|
|
}
|
|
*rdd = sregs->tbr;
|
|
break;
|
|
case WRPSR:
|
|
if ((sregs->psr & 0x1f) > 7) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
if (!(sregs->psr & PSR_S)) {
|
|
sregs->trap = TRAP_PRIVI;
|
|
break;
|
|
}
|
|
sregs->psr = (sregs->psr & 0xff000000) |
|
|
(rs1 ^ operand2) & 0x00f03fff;
|
|
break;
|
|
case WRWIM:
|
|
if (!(sregs->psr & PSR_S)) {
|
|
sregs->trap = TRAP_PRIVI;
|
|
break;
|
|
}
|
|
sregs->wim = (rs1 ^ operand2) & 0x0ff;
|
|
break;
|
|
case WRTBR:
|
|
if (!(sregs->psr & PSR_S)) {
|
|
sregs->trap = TRAP_PRIVI;
|
|
break;
|
|
}
|
|
sregs->tbr = (sregs->tbr & 0x00000ff0) |
|
|
((rs1 ^ operand2) & 0xfffff000);
|
|
break;
|
|
case WRY:
|
|
if (!sparclite)
|
|
sregs->y = (rs1 ^ operand2);
|
|
else {
|
|
if ( 0 == rd )
|
|
sregs->y = (rs1 ^ operand2);
|
|
else if ( 17 == rd )
|
|
sregs->asr17 = (rs1 ^ operand2);
|
|
else {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
}
|
|
break;
|
|
case JMPL:
|
|
|
|
#ifdef STAT
|
|
sregs->nbranch++;
|
|
#endif
|
|
sregs->icnt = T_JMPL; /* JMPL takes two cycles */
|
|
if (rs1 & 0x3) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
*rdd = sregs->pc;
|
|
npc = rs1 + operand2;
|
|
break;
|
|
case RETT:
|
|
address = rs1 + operand2;
|
|
new_cwp = ((sregs->psr & PSR_CWP) + 1) & PSR_CWP;
|
|
sregs->icnt = T_RETT; /* RETT takes two cycles */
|
|
if (sregs->psr & PSR_ET) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
if (!(sregs->psr & PSR_S)) {
|
|
sregs->trap = TRAP_PRIVI;
|
|
break;
|
|
}
|
|
if (sregs->wim & (1 << new_cwp)) {
|
|
sregs->trap = TRAP_WUFL;
|
|
break;
|
|
}
|
|
if (address & 0x3) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
sregs->psr = (sregs->psr & ~PSR_CWP) | new_cwp | PSR_ET;
|
|
sregs->psr =
|
|
(sregs->psr & ~PSR_S) | ((sregs->psr & PSR_PS) << 1);
|
|
npc = address;
|
|
break;
|
|
|
|
case SCAN:
|
|
{
|
|
uint32 result, mask;
|
|
int i;
|
|
|
|
if (!sparclite) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
mask = (operand2 & 0x80000000) | (operand2 >> 1);
|
|
result = rs1 ^ mask;
|
|
|
|
for (i = 0; i < 32; i++) {
|
|
if (result & 0x80000000)
|
|
break;
|
|
result <<= 1;
|
|
}
|
|
|
|
*rdd = i == 32 ? 63 : i;
|
|
}
|
|
break;
|
|
|
|
default:
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
}
|
|
break;
|
|
case 3: /* Load/store instructions */
|
|
|
|
address = rs1 + operand2;
|
|
|
|
if (sregs->psr & PSR_S)
|
|
asi = 11;
|
|
else
|
|
asi = 10;
|
|
|
|
if (op3 & 4) {
|
|
sregs->icnt = T_ST; /* Set store instruction count */
|
|
#ifdef STAT
|
|
sregs->nstore++;
|
|
#endif
|
|
} else {
|
|
sregs->icnt = T_LD; /* Set load instruction count */
|
|
#ifdef STAT
|
|
sregs->nload++;
|
|
#endif
|
|
}
|
|
|
|
/* Decode load/store instructions */
|
|
|
|
switch (op3) {
|
|
case LDDA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case LDD:
|
|
if (address & 0x7) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
if (rd & 1) {
|
|
rd &= 0x1e;
|
|
if (rd > 7)
|
|
rdd = &(sregs->r[(cwp + rd) & 0x7f]);
|
|
else
|
|
rdd = &(sregs->g[rd]);
|
|
}
|
|
mexc = memory_read (asi, address, ddata, 2, &ws);
|
|
sregs->hold += ws;
|
|
mexc |= memory_read (asi, address+4, &ddata[1], 2, &ws);
|
|
sregs->hold += ws;
|
|
sregs->icnt = T_LDD;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
} else {
|
|
rdd[0] = ddata[0];
|
|
rdd[1] = ddata[1];
|
|
#ifdef STAT
|
|
sregs->nload++; /* Double load counts twice */
|
|
#endif
|
|
}
|
|
break;
|
|
|
|
case LDA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case LD:
|
|
if (address & 0x3) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
mexc = memory_read(asi, address, &data, 2, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
} else {
|
|
*rdd = data;
|
|
}
|
|
break;
|
|
case LDSTUBA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case LDSTUB:
|
|
mexc = memory_read(asi, address, &data, 0, &ws);
|
|
sregs->hold += ws;
|
|
sregs->icnt = T_LDST;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
break;
|
|
}
|
|
data = extract_byte (data, address);
|
|
*rdd = data;
|
|
data = 0x0ff;
|
|
mexc = memory_write(asi, address, &data, 0, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
}
|
|
#ifdef STAT
|
|
sregs->nload++;
|
|
#endif
|
|
break;
|
|
case LDSBA:
|
|
case LDUBA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case LDSB:
|
|
case LDUB:
|
|
mexc = memory_read(asi, address, &data, 0, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
break;
|
|
}
|
|
if (op3 == LDSB)
|
|
data = extract_byte_signed (data, address);
|
|
else
|
|
data = extract_byte (data, address);
|
|
*rdd = data;
|
|
break;
|
|
case LDSHA:
|
|
case LDUHA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case LDSH:
|
|
case LDUH:
|
|
if (address & 0x1) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
mexc = memory_read(asi, address, &data, 1, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
break;
|
|
}
|
|
if (op3 == LDSH)
|
|
data = extract_short_signed (data, address);
|
|
else
|
|
data = extract_short (data, address);
|
|
*rdd = data;
|
|
break;
|
|
case LDF:
|
|
if (!((sregs->psr & PSR_EF) && FP_PRES)) {
|
|
sregs->trap = TRAP_FPDIS;
|
|
break;
|
|
}
|
|
if (address & 0x3) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
if (ebase.simtime < sregs->ftime) {
|
|
if ((sregs->frd == rd) || (sregs->frs1 == rd) ||
|
|
(sregs->frs2 == rd))
|
|
sregs->fhold += (sregs->ftime - ebase.simtime);
|
|
}
|
|
mexc = memory_read(asi, address, &data, 2, &ws);
|
|
sregs->hold += ws;
|
|
sregs->flrd = rd;
|
|
sregs->ltime = ebase.simtime + sregs->icnt + FLSTHOLD +
|
|
sregs->hold + sregs->fhold;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
} else {
|
|
sregs->fs[rd] = *((float32 *) & data);
|
|
}
|
|
break;
|
|
case LDDF:
|
|
if (!((sregs->psr & PSR_EF) && FP_PRES)) {
|
|
sregs->trap = TRAP_FPDIS;
|
|
break;
|
|
}
|
|
if (address & 0x7) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
if (ebase.simtime < sregs->ftime) {
|
|
if (((sregs->frd >> 1) == (rd >> 1)) ||
|
|
((sregs->frs1 >> 1) == (rd >> 1)) ||
|
|
((sregs->frs2 >> 1) == (rd >> 1)))
|
|
sregs->fhold += (sregs->ftime - ebase.simtime);
|
|
}
|
|
mexc = memory_read (asi, address, ddata, 2, &ws);
|
|
sregs->hold += ws;
|
|
mexc |= memory_read (asi, address+4, &ddata[1], 2, &ws);
|
|
sregs->hold += ws;
|
|
sregs->icnt = T_LDD;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
} else {
|
|
rd &= 0x1E;
|
|
sregs->flrd = rd;
|
|
sregs->fs[rd] = *((float32 *) & ddata[0]);
|
|
#ifdef STAT
|
|
sregs->nload++; /* Double load counts twice */
|
|
#endif
|
|
sregs->fs[rd + 1] = *((float32 *) & ddata[1]);
|
|
sregs->ltime = ebase.simtime + sregs->icnt + FLSTHOLD +
|
|
sregs->hold + sregs->fhold;
|
|
}
|
|
break;
|
|
case LDFSR:
|
|
if (ebase.simtime < sregs->ftime) {
|
|
sregs->fhold += (sregs->ftime - ebase.simtime);
|
|
}
|
|
if (!((sregs->psr & PSR_EF) && FP_PRES)) {
|
|
sregs->trap = TRAP_FPDIS;
|
|
break;
|
|
}
|
|
if (address & 0x3) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
mexc = memory_read(asi, address, &data, 2, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
} else {
|
|
sregs->fsr =
|
|
(sregs->fsr & 0x7FF000) | (data & ~0x7FF000);
|
|
set_fsr(sregs->fsr);
|
|
}
|
|
break;
|
|
case STFSR:
|
|
if (!((sregs->psr & PSR_EF) && FP_PRES)) {
|
|
sregs->trap = TRAP_FPDIS;
|
|
break;
|
|
}
|
|
if (address & 0x3) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
if (ebase.simtime < sregs->ftime) {
|
|
sregs->fhold += (sregs->ftime - ebase.simtime);
|
|
}
|
|
mexc = memory_write(asi, address, &sregs->fsr, 2, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
}
|
|
break;
|
|
|
|
case STA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case ST:
|
|
if (address & 0x3) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
mexc = memory_write(asi, address, rdd, 2, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
}
|
|
break;
|
|
case STBA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case STB:
|
|
mexc = memory_write(asi, address, rdd, 0, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
}
|
|
break;
|
|
case STDA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case STD:
|
|
if (address & 0x7) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
if (rd & 1) {
|
|
rd &= 0x1e;
|
|
if (rd > 7)
|
|
rdd = &(sregs->r[(cwp + rd) & 0x7f]);
|
|
else
|
|
rdd = &(sregs->g[rd]);
|
|
}
|
|
mexc = memory_write(asi, address, rdd, 3, &ws);
|
|
sregs->hold += ws;
|
|
sregs->icnt = T_STD;
|
|
#ifdef STAT
|
|
sregs->nstore++; /* Double store counts twice */
|
|
#endif
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
break;
|
|
}
|
|
break;
|
|
case STDFQ:
|
|
if ((sregs->psr & 0x1f) > 7) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
if (!((sregs->psr & PSR_EF) && FP_PRES)) {
|
|
sregs->trap = TRAP_FPDIS;
|
|
break;
|
|
}
|
|
if (address & 0x7) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
if (!(sregs->fsr & FSR_QNE)) {
|
|
sregs->fsr = (sregs->fsr & ~FSR_TT) | FP_SEQ_ERR;
|
|
break;
|
|
}
|
|
rdd = &(sregs->fpq[0]);
|
|
mexc = memory_write(asi, address, rdd, 3, &ws);
|
|
sregs->hold += ws;
|
|
sregs->icnt = T_STD;
|
|
#ifdef STAT
|
|
sregs->nstore++; /* Double store counts twice */
|
|
#endif
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
break;
|
|
} else {
|
|
sregs->fsr &= ~FSR_QNE;
|
|
sregs->fpstate = FP_EXE_MODE;
|
|
}
|
|
break;
|
|
case STHA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case STH:
|
|
if (address & 0x1) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
mexc = memory_write(asi, address, rdd, 1, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
}
|
|
break;
|
|
case STF:
|
|
if (!((sregs->psr & PSR_EF) && FP_PRES)) {
|
|
sregs->trap = TRAP_FPDIS;
|
|
break;
|
|
}
|
|
if (address & 0x3) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
if (ebase.simtime < sregs->ftime) {
|
|
if (sregs->frd == rd)
|
|
sregs->fhold += (sregs->ftime - ebase.simtime);
|
|
}
|
|
mexc = memory_write(asi, address, &sregs->fsi[rd], 2, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
}
|
|
break;
|
|
case STDF:
|
|
if (!((sregs->psr & PSR_EF) && FP_PRES)) {
|
|
sregs->trap = TRAP_FPDIS;
|
|
break;
|
|
}
|
|
if (address & 0x7) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
rd &= 0x1E;
|
|
if (ebase.simtime < sregs->ftime) {
|
|
if ((sregs->frd == rd) || (sregs->frd + 1 == rd))
|
|
sregs->fhold += (sregs->ftime - ebase.simtime);
|
|
}
|
|
mexc = memory_write(asi, address, &sregs->fsi[rd], 3, &ws);
|
|
sregs->hold += ws;
|
|
sregs->icnt = T_STD;
|
|
#ifdef STAT
|
|
sregs->nstore++; /* Double store counts twice */
|
|
#endif
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
}
|
|
break;
|
|
case SWAPA:
|
|
if (!chk_asi(sregs, &asi, op3)) break;
|
|
case SWAP:
|
|
if (address & 0x3) {
|
|
sregs->trap = TRAP_UNALI;
|
|
break;
|
|
}
|
|
mexc = memory_read(asi, address, &data, 2, &ws);
|
|
sregs->hold += ws;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
break;
|
|
}
|
|
mexc = memory_write(asi, address, rdd, 2, &ws);
|
|
sregs->hold += ws;
|
|
sregs->icnt = T_LDST;
|
|
if (mexc) {
|
|
sregs->trap = TRAP_DEXC;
|
|
break;
|
|
} else
|
|
*rdd = data;
|
|
#ifdef STAT
|
|
sregs->nload++;
|
|
#endif
|
|
break;
|
|
|
|
|
|
default:
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
|
|
#ifdef LOAD_DEL
|
|
|
|
if (!(op3 & 4)) {
|
|
sregs->ildtime = ebase.simtime + sregs->hold + sregs->icnt;
|
|
sregs->ildreg = rd;
|
|
if ((op3 | 0x10) == 0x13)
|
|
sregs->ildreg |= 1; /* Double load, odd register loaded
|
|
* last */
|
|
}
|
|
#endif
|
|
break;
|
|
|
|
default:
|
|
sregs->trap = TRAP_UNIMP;
|
|
break;
|
|
}
|
|
sregs->g[0] = 0;
|
|
if (!sregs->trap) {
|
|
sregs->pc = pc;
|
|
sregs->npc = npc;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
#define T_FABSs 2
|
|
#define T_FADDs 4
|
|
#define T_FADDd 4
|
|
#define T_FCMPs 4
|
|
#define T_FCMPd 4
|
|
#define T_FDIVs 20
|
|
#define T_FDIVd 35
|
|
#define T_FMOVs 2
|
|
#define T_FMULs 5
|
|
#define T_FMULd 9
|
|
#define T_FNEGs 2
|
|
#define T_FSQRTs 37
|
|
#define T_FSQRTd 65
|
|
#define T_FSUBs 4
|
|
#define T_FSUBd 4
|
|
#define T_FdTOi 7
|
|
#define T_FdTOs 3
|
|
#define T_FiTOs 6
|
|
#define T_FiTOd 6
|
|
#define T_FsTOi 6
|
|
#define T_FsTOd 2
|
|
|
|
#define FABSs 0x09
|
|
#define FADDs 0x41
|
|
#define FADDd 0x42
|
|
#define FCMPs 0x51
|
|
#define FCMPd 0x52
|
|
#define FCMPEs 0x55
|
|
#define FCMPEd 0x56
|
|
#define FDIVs 0x4D
|
|
#define FDIVd 0x4E
|
|
#define FMOVs 0x01
|
|
#define FMULs 0x49
|
|
#define FMULd 0x4A
|
|
#define FNEGs 0x05
|
|
#define FSQRTs 0x29
|
|
#define FSQRTd 0x2A
|
|
#define FSUBs 0x45
|
|
#define FSUBd 0x46
|
|
#define FdTOi 0xD2
|
|
#define FdTOs 0xC6
|
|
#define FiTOs 0xC4
|
|
#define FiTOd 0xC8
|
|
#define FsTOi 0xD1
|
|
#define FsTOd 0xC9
|
|
|
|
|
|
static int
|
|
fpexec(op3, rd, rs1, rs2, sregs)
|
|
uint32 op3, rd, rs1, rs2;
|
|
struct pstate *sregs;
|
|
{
|
|
uint32 opf, tem, accex;
|
|
int32 fcc;
|
|
uint32 ldadj;
|
|
|
|
if (sregs->fpstate == FP_EXC_MODE) {
|
|
sregs->fsr = (sregs->fsr & ~FSR_TT) | FP_SEQ_ERR;
|
|
sregs->fpstate = FP_EXC_PE;
|
|
return 0;
|
|
}
|
|
if (sregs->fpstate == FP_EXC_PE) {
|
|
sregs->fpstate = FP_EXC_MODE;
|
|
return TRAP_FPEXC;
|
|
}
|
|
opf = (sregs->inst >> 5) & 0x1ff;
|
|
|
|
/*
|
|
* Check if we already have an FPop in the pipe. If so, halt until it is
|
|
* finished by incrementing fhold with the remaining execution time
|
|
*/
|
|
|
|
if (ebase.simtime < sregs->ftime) {
|
|
sregs->fhold = (sregs->ftime - ebase.simtime);
|
|
} else {
|
|
sregs->fhold = 0;
|
|
|
|
/* Check load dependencies. */
|
|
|
|
if (ebase.simtime < sregs->ltime) {
|
|
|
|
/* Don't check rs1 if single operand instructions */
|
|
|
|
if (((opf >> 6) == 0) || ((opf >> 6) == 3))
|
|
rs1 = 32;
|
|
|
|
/* Adjust for double floats */
|
|
|
|
ldadj = opf & 1;
|
|
if (!(((sregs->flrd - rs1) >> ldadj) && ((sregs->flrd - rs2) >> ldadj)))
|
|
sregs->fhold++;
|
|
}
|
|
}
|
|
|
|
sregs->finst++;
|
|
|
|
sregs->frs1 = rs1; /* Store src and dst for dependecy check */
|
|
sregs->frs2 = rs2;
|
|
sregs->frd = rd;
|
|
|
|
sregs->ftime = ebase.simtime + sregs->hold + sregs->fhold;
|
|
|
|
/* SPARC is big-endian - swap double floats if host is little-endian */
|
|
/* This is ugly - I know ... */
|
|
|
|
/* FIXME: should use (HOST_BYTE_ORDER == CURRENT_TARGET_BYTE_ORDER)
|
|
but what about machines where float values are different endianness
|
|
from integer values? */
|
|
|
|
#ifdef HOST_LITTLE_ENDIAN
|
|
rs1 &= 0x1f;
|
|
switch (opf) {
|
|
case FADDd:
|
|
case FDIVd:
|
|
case FMULd:
|
|
case FSQRTd:
|
|
case FSUBd:
|
|
case FCMPd:
|
|
case FCMPEd:
|
|
case FdTOi:
|
|
case FdTOs:
|
|
sregs->fdp[rs1 | 1] = sregs->fs[rs1 & ~1];
|
|
sregs->fdp[rs1 & ~1] = sregs->fs[rs1 | 1];
|
|
sregs->fdp[rs2 | 1] = sregs->fs[rs2 & ~1];
|
|
sregs->fdp[rs2 & ~1] = sregs->fs[rs2 | 1];
|
|
default:
|
|
break;
|
|
}
|
|
#endif
|
|
|
|
clear_accex();
|
|
|
|
switch (opf) {
|
|
case FABSs:
|
|
sregs->fs[rd] = fabs(sregs->fs[rs2]);
|
|
sregs->ftime += T_FABSs;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FADDs:
|
|
sregs->fs[rd] = sregs->fs[rs1] + sregs->fs[rs2];
|
|
sregs->ftime += T_FADDs;
|
|
break;
|
|
case FADDd:
|
|
sregs->fd[rd >> 1] = sregs->fd[rs1 >> 1] + sregs->fd[rs2 >> 1];
|
|
sregs->ftime += T_FADDd;
|
|
break;
|
|
case FCMPs:
|
|
case FCMPEs:
|
|
if (sregs->fs[rs1] == sregs->fs[rs2])
|
|
fcc = 3;
|
|
else if (sregs->fs[rs1] < sregs->fs[rs2])
|
|
fcc = 2;
|
|
else if (sregs->fs[rs1] > sregs->fs[rs2])
|
|
fcc = 1;
|
|
else
|
|
fcc = 0;
|
|
sregs->fsr |= 0x0C00;
|
|
sregs->fsr &= ~(fcc << 10);
|
|
sregs->ftime += T_FCMPs;
|
|
sregs->frd = 32; /* rd ignored */
|
|
if ((fcc == 0) && (opf == FCMPEs)) {
|
|
sregs->fpstate = FP_EXC_PE;
|
|
sregs->fsr = (sregs->fsr & ~0x1C000) | (1 << 14);
|
|
}
|
|
break;
|
|
case FCMPd:
|
|
case FCMPEd:
|
|
if (sregs->fd[rs1 >> 1] == sregs->fd[rs2 >> 1])
|
|
fcc = 3;
|
|
else if (sregs->fd[rs1 >> 1] < sregs->fd[rs2 >> 1])
|
|
fcc = 2;
|
|
else if (sregs->fd[rs1 >> 1] > sregs->fd[rs2 >> 1])
|
|
fcc = 1;
|
|
else
|
|
fcc = 0;
|
|
sregs->fsr |= 0x0C00;
|
|
sregs->fsr &= ~(fcc << 10);
|
|
sregs->ftime += T_FCMPd;
|
|
sregs->frd = 32; /* rd ignored */
|
|
if ((fcc == 0) && (opf == FCMPEd)) {
|
|
sregs->fpstate = FP_EXC_PE;
|
|
sregs->fsr = (sregs->fsr & ~FSR_TT) | FP_IEEE;
|
|
}
|
|
break;
|
|
case FDIVs:
|
|
sregs->fs[rd] = sregs->fs[rs1] / sregs->fs[rs2];
|
|
sregs->ftime += T_FDIVs;
|
|
break;
|
|
case FDIVd:
|
|
sregs->fd[rd >> 1] = sregs->fd[rs1 >> 1] / sregs->fd[rs2 >> 1];
|
|
sregs->ftime += T_FDIVd;
|
|
break;
|
|
case FMOVs:
|
|
sregs->fs[rd] = sregs->fs[rs2];
|
|
sregs->ftime += T_FMOVs;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FMULs:
|
|
sregs->fs[rd] = sregs->fs[rs1] * sregs->fs[rs2];
|
|
sregs->ftime += T_FMULs;
|
|
break;
|
|
case FMULd:
|
|
sregs->fd[rd >> 1] = sregs->fd[rs1 >> 1] * sregs->fd[rs2 >> 1];
|
|
sregs->ftime += T_FMULd;
|
|
break;
|
|
case FNEGs:
|
|
sregs->fs[rd] = -sregs->fs[rs2];
|
|
sregs->ftime += T_FNEGs;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FSQRTs:
|
|
if (sregs->fs[rs2] < 0.0) {
|
|
sregs->fpstate = FP_EXC_PE;
|
|
sregs->fsr = (sregs->fsr & ~FSR_TT) | FP_IEEE;
|
|
sregs->fsr = (sregs->fsr & 0x1f) | 0x10;
|
|
break;
|
|
}
|
|
sregs->fs[rd] = sqrt(sregs->fs[rs2]);
|
|
sregs->ftime += T_FSQRTs;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FSQRTd:
|
|
if (sregs->fd[rs2 >> 1] < 0.0) {
|
|
sregs->fpstate = FP_EXC_PE;
|
|
sregs->fsr = (sregs->fsr & ~FSR_TT) | FP_IEEE;
|
|
sregs->fsr = (sregs->fsr & 0x1f) | 0x10;
|
|
break;
|
|
}
|
|
sregs->fd[rd >> 1] = sqrt(sregs->fd[rs2 >> 1]);
|
|
sregs->ftime += T_FSQRTd;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FSUBs:
|
|
sregs->fs[rd] = sregs->fs[rs1] - sregs->fs[rs2];
|
|
sregs->ftime += T_FSUBs;
|
|
break;
|
|
case FSUBd:
|
|
sregs->fd[rd >> 1] = sregs->fd[rs1 >> 1] - sregs->fd[rs2 >> 1];
|
|
sregs->ftime += T_FSUBd;
|
|
break;
|
|
case FdTOi:
|
|
sregs->fsi[rd] = (int) sregs->fd[rs2 >> 1];
|
|
sregs->ftime += T_FdTOi;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FdTOs:
|
|
sregs->fs[rd] = (float32) sregs->fd[rs2 >> 1];
|
|
sregs->ftime += T_FdTOs;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FiTOs:
|
|
sregs->fs[rd] = (float32) sregs->fsi[rs2];
|
|
sregs->ftime += T_FiTOs;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FiTOd:
|
|
sregs->fd[rd >> 1] = (float64) sregs->fsi[rs2];
|
|
sregs->ftime += T_FiTOd;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FsTOi:
|
|
sregs->fsi[rd] = (int) sregs->fs[rs2];
|
|
sregs->ftime += T_FsTOi;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
case FsTOd:
|
|
sregs->fd[rd >> 1] = sregs->fs[rs2];
|
|
sregs->ftime += T_FsTOd;
|
|
sregs->frs1 = 32; /* rs1 ignored */
|
|
break;
|
|
|
|
default:
|
|
sregs->fsr = (sregs->fsr & ~FSR_TT) | FP_UNIMP;
|
|
sregs->fpstate = FP_EXC_PE;
|
|
}
|
|
|
|
#ifdef ERRINJ
|
|
if (errftt) {
|
|
sregs->fsr = (sregs->fsr & ~FSR_TT) | (errftt << 14);
|
|
sregs->fpstate = FP_EXC_PE;
|
|
if (sis_verbose) printf("Inserted fpu error %X\n",errftt);
|
|
errftt = 0;
|
|
}
|
|
#endif
|
|
|
|
accex = get_accex();
|
|
|
|
#ifdef HOST_LITTLE_ENDIAN
|
|
switch (opf) {
|
|
case FADDd:
|
|
case FDIVd:
|
|
case FMULd:
|
|
case FSQRTd:
|
|
case FSUBd:
|
|
case FiTOd:
|
|
case FsTOd:
|
|
sregs->fs[rd & ~1] = sregs->fdp[rd | 1];
|
|
sregs->fs[rd | 1] = sregs->fdp[rd & ~1];
|
|
default:
|
|
break;
|
|
}
|
|
#endif
|
|
if (sregs->fpstate == FP_EXC_PE) {
|
|
sregs->fpq[0] = sregs->pc;
|
|
sregs->fpq[1] = sregs->inst;
|
|
sregs->fsr |= FSR_QNE;
|
|
} else {
|
|
tem = (sregs->fsr >> 23) & 0x1f;
|
|
if (tem & accex) {
|
|
sregs->fpstate = FP_EXC_PE;
|
|
sregs->fsr = (sregs->fsr & ~FSR_TT) | FP_IEEE;
|
|
sregs->fsr = ((sregs->fsr & ~0x1f) | accex);
|
|
} else {
|
|
sregs->fsr = ((((sregs->fsr >> 5) | accex) << 5) | accex);
|
|
}
|
|
if (sregs->fpstate == FP_EXC_PE) {
|
|
sregs->fpq[0] = sregs->pc;
|
|
sregs->fpq[1] = sregs->inst;
|
|
sregs->fsr |= FSR_QNE;
|
|
}
|
|
}
|
|
clear_accex();
|
|
|
|
return 0;
|
|
|
|
|
|
}
|
|
|
|
static int
|
|
chk_asi(sregs, asi, op3)
|
|
struct pstate *sregs;
|
|
uint32 *asi, op3;
|
|
|
|
{
|
|
if (!(sregs->psr & PSR_S)) {
|
|
sregs->trap = TRAP_PRIVI;
|
|
return 0;
|
|
} else if (sregs->inst & INST_I) {
|
|
sregs->trap = TRAP_UNIMP;
|
|
return 0;
|
|
} else
|
|
*asi = (sregs->inst >> 5) & 0x0ff;
|
|
return 1;
|
|
}
|
|
|
|
int
|
|
execute_trap(sregs)
|
|
struct pstate *sregs;
|
|
{
|
|
int32 cwp;
|
|
|
|
if (sregs->trap == 256) {
|
|
sregs->pc = 0;
|
|
sregs->npc = 4;
|
|
sregs->trap = 0;
|
|
} else if (sregs->trap == 257) {
|
|
return ERROR;
|
|
} else {
|
|
|
|
if ((sregs->psr & PSR_ET) == 0)
|
|
return ERROR;
|
|
|
|
sregs->tbr = (sregs->tbr & 0xfffff000) | (sregs->trap << 4);
|
|
sregs->trap = 0;
|
|
sregs->psr &= ~PSR_ET;
|
|
sregs->psr |= ((sregs->psr & PSR_S) >> 1);
|
|
sregs->annul = 0;
|
|
sregs->psr = (((sregs->psr & PSR_CWP) - 1) & 0x7) | (sregs->psr & ~PSR_CWP);
|
|
cwp = ((sregs->psr & PSR_CWP) << 4);
|
|
sregs->r[(cwp + 17) & 0x7f] = sregs->pc;
|
|
sregs->r[(cwp + 18) & 0x7f] = sregs->npc;
|
|
sregs->psr |= PSR_S;
|
|
sregs->pc = sregs->tbr;
|
|
sregs->npc = sregs->tbr + 4;
|
|
|
|
if ( 0 != (1 & sregs->asr17) ) {
|
|
/* single vector trapping! */
|
|
sregs->pc = sregs->tbr & 0xfffff000;
|
|
sregs->npc = sregs->pc + 4;
|
|
}
|
|
|
|
/* Increase simulator time */
|
|
sregs->icnt = TRAP_C;
|
|
|
|
}
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
extern struct irqcell irqarr[16];
|
|
|
|
int
|
|
check_interrupts(sregs)
|
|
struct pstate *sregs;
|
|
{
|
|
#ifdef ERRINJ
|
|
if (errtt) {
|
|
sregs->trap = errtt;
|
|
if (sis_verbose) printf("Inserted error trap 0x%02X\n",errtt);
|
|
errtt = 0;
|
|
}
|
|
#endif
|
|
|
|
if ((ext_irl) && (sregs->psr & PSR_ET) &&
|
|
((ext_irl == 15) || (ext_irl > (int) ((sregs->psr & PSR_PIL) >> 8)))) {
|
|
if (sregs->trap == 0) {
|
|
sregs->trap = 16 + ext_irl;
|
|
irqarr[ext_irl & 0x0f].callback(irqarr[ext_irl & 0x0f].arg);
|
|
return 1;
|
|
}
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
void
|
|
init_regs(sregs)
|
|
struct pstate *sregs;
|
|
{
|
|
sregs->pc = 0;
|
|
sregs->npc = 4;
|
|
sregs->trap = 0;
|
|
sregs->psr &= 0x00f03fdf;
|
|
sregs->psr |= 0x11000080; /* Set supervisor bit */
|
|
sregs->breakpoint = 0;
|
|
sregs->annul = 0;
|
|
sregs->fpstate = FP_EXE_MODE;
|
|
sregs->fpqn = 0;
|
|
sregs->ftime = 0;
|
|
sregs->ltime = 0;
|
|
sregs->err_mode = 0;
|
|
ext_irl = 0;
|
|
sregs->g[0] = 0;
|
|
#ifdef HOST_LITTLE_ENDIAN
|
|
sregs->fdp = (float32 *) sregs->fd;
|
|
sregs->fsi = (int32 *) sregs->fs;
|
|
#else
|
|
sregs->fs = (float32 *) sregs->fd;
|
|
sregs->fsi = (int32 *) sregs->fd;
|
|
#endif
|
|
sregs->fsr = 0;
|
|
sregs->fpu_pres = !nfp;
|
|
set_fsr(sregs->fsr);
|
|
sregs->bphit = 0;
|
|
sregs->ildreg = 0;
|
|
sregs->ildtime = 0;
|
|
|
|
sregs->y = 0;
|
|
sregs->asr17 = 0;
|
|
|
|
sregs->rett_err = 0;
|
|
sregs->jmpltime = 0;
|
|
}
|