X-Git-Url: http://drtracing.org/?a=blobdiff_plain;f=sim%2Fmips%2Fcp1.c;h=82fdea2bcb58f3257f09e0f095a1267c6114da6f;hb=cbd7581f343d85b4216db2eefdf601f6d988062d;hp=e4bfba2b028a75d5fdda7618c25dc0e9502f6332;hpb=196496eda07d1faa8d71fbbf000fa09a05c8e908;p=deliverable%2Fbinutils-gdb.git
diff --git a/sim/mips/cp1.c b/sim/mips/cp1.c
index e4bfba2b02..82fdea2bcb 100644
--- a/sim/mips/cp1.c
+++ b/sim/mips/cp1.c
@@ -1,24 +1,25 @@
/*> cp1.c <*/
/* MIPS Simulator FPU (CoProcessor 1) support.
- Copyright (C) 2002 Free Software Foundation, Inc.
- Originally created by Cygnus Solutions, modified substially
- by Broadcom Corporation (SiByte).
+ Copyright (C) 2002-2020 Free Software Foundation, Inc.
+ Originally created by Cygnus Solutions. Extensive modifications,
+ including paired-single operation support and MIPS-3D support
+ contributed by Ed Satterthwaite and Chris Demetriou, of Broadcom
+ Corporation (SiByte).
This file is part of GDB, the GNU debugger.
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, or (at your option)
-any later version.
+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, write to the Free Software Foundation, Inc.,
-59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
+You should have received a copy of the GNU General Public License
+along with this program. If not, see . */
/* XXX: The following notice should be removed as soon as is practical: */
/* Floating Point Support for gdb MIPS simulators
@@ -71,13 +72,28 @@ with this program; if not, write to the Free Software Foundation, Inc.,
siiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
s = 1bit = sign
i = 63bits = integer
+
+ PAIRED SINGLE precision floating:
+ seeeeeeeefffffffffffffffffffffffseeeeeeeefffffffffffffffffffffff
+ | upper || lower |
+ s = 1bit = sign
+ e = 8bits = exponent
+ f = 23bits = fraction
+ Note: upper = [63..32], lower = [31..0]
*/
-/* Explicit QNaN values used when value required: */
+/* Extract packed single values: */
+#define FP_PS_upper(v) (((v) >> 32) & (unsigned)0xFFFFFFFF)
+#define FP_PS_lower(v) ((v) & (unsigned)0xFFFFFFFF)
+#define FP_PS_cat(u,l) (((unsigned64)((u) & (unsigned)0xFFFFFFFF) << 32) \
+ | (unsigned64)((l) & 0xFFFFFFFF))
+
+/* Explicit QNaN values. */
#define FPQNaN_SINGLE (0x7FBFFFFF)
#define FPQNaN_WORD (0x7FFFFFFF)
#define FPQNaN_DOUBLE (UNSIGNED64 (0x7FF7FFFFFFFFFFFF))
#define FPQNaN_LONG (UNSIGNED64 (0x7FFFFFFFFFFFFFFF))
+#define FPQNaN_PS (FP_PS_cat (FPQNaN_SINGLE, FPQNaN_SINGLE))
static const char *fpu_format_name (FP_formats fmt);
#ifdef DEBUG
@@ -93,11 +109,11 @@ value_fpr (sim_cpu *cpu,
uword64 value = 0;
int err = 0;
- /* Treat unused register values, as fixed-point 64bit values: */
- if ((fmt == fmt_uninterpreted) || (fmt == fmt_unknown))
+ /* Treat unused register values, as fixed-point 64bit values. */
+ if (fmt == fmt_unknown)
{
#if 1
- /* If request to read data as "uninterpreted", then use the current
+ /* If request to read data as "unknown", then use the current
encoding: */
fmt = FPR_STATE[fpr];
#else
@@ -105,21 +121,24 @@ value_fpr (sim_cpu *cpu,
#endif
}
- /* For values not yet accessed, set to the desired format: */
- if (FPR_STATE[fpr] == fmt_uninterpreted)
+ /* For values not yet accessed, set to the desired format. */
+ if (fmt < fmt_uninterpreted)
{
- FPR_STATE[fpr] = fmt;
+ if (FPR_STATE[fpr] == fmt_uninterpreted)
+ {
+ FPR_STATE[fpr] = fmt;
#ifdef DEBUG
- printf ("DBG: Register %d was fmt_uninterpreted. Now %s\n", fpr,
- fpu_format_name (fmt));
+ printf ("DBG: Register %d was fmt_uninterpreted. Now %s\n", fpr,
+ fpu_format_name (fmt));
#endif /* DEBUG */
- }
- if (fmt != FPR_STATE[fpr])
- {
- sim_io_eprintf (SD, "FPR %d (format %s) being accessed with format %s - setting to unknown (PC = 0x%s)\n",
- fpr, fpu_format_name (FPR_STATE[fpr]),
- fpu_format_name (fmt), pr_addr (cia));
- FPR_STATE[fpr] = fmt_unknown;
+ }
+ else if (fmt != FPR_STATE[fpr])
+ {
+ sim_io_eprintf (SD, "FPR %d (format %s) being accessed with format %s - setting to unknown (PC = 0x%s)\n",
+ fpr, fpu_format_name (FPR_STATE[fpr]),
+ fpu_format_name (fmt), pr_addr (cia));
+ FPR_STATE[fpr] = fmt_unknown;
+ }
}
if (FPR_STATE[fpr] == fmt_unknown)
@@ -131,6 +150,7 @@ value_fpr (sim_cpu *cpu,
case fmt_double: value = FPQNaN_DOUBLE; break;
case fmt_word: value = FPQNaN_WORD; break;
case fmt_long: value = FPQNaN_LONG; break;
+ case fmt_ps: value = FPQNaN_PS; break;
default: err = -1; break;
}
}
@@ -138,14 +158,17 @@ value_fpr (sim_cpu *cpu,
{
switch (fmt)
{
+ case fmt_uninterpreted_32:
case fmt_single:
case fmt_word:
value = (FGR[fpr] & 0xFFFFFFFF);
break;
+ case fmt_uninterpreted_64:
case fmt_uninterpreted:
case fmt_double:
case fmt_long:
+ case fmt_ps:
value = FGR[fpr];
break;
@@ -158,17 +181,19 @@ value_fpr (sim_cpu *cpu,
{
switch (fmt)
{
+ case fmt_uninterpreted_32:
case fmt_single:
case fmt_word:
value = (FGR[fpr] & 0xFFFFFFFF);
break;
+ case fmt_uninterpreted_64:
case fmt_uninterpreted:
case fmt_double:
case fmt_long:
if ((fpr & 1) == 0)
{
- /* Even registers numbers only. */
+ /* Even register numbers only. */
#ifdef DEBUG
printf ("DBG: ValueFPR: FGR[%d] = %s, FGR[%d] = %s\n",
fpr + 1, pr_uword64 ((uword64) FGR[fpr+1]),
@@ -183,6 +208,10 @@ value_fpr (sim_cpu *cpu,
}
break;
+ case fmt_ps:
+ SignalException (ReservedInstruction, 0);
+ break;
+
default:
err = -1;
break;
@@ -237,6 +266,7 @@ store_fpr (sim_cpu *cpu,
case fmt_uninterpreted:
case fmt_double:
case fmt_long:
+ case fmt_ps:
FGR[fpr] = value;
FPR_STATE[fpr] = fmt;
break;
@@ -275,11 +305,16 @@ store_fpr (sim_cpu *cpu,
else
{
FPR_STATE[fpr] = fmt_unknown;
- FPR_STATE[fpr + 1] = fmt_unknown;
+ FPR_STATE[fpr ^ 1] = fmt_unknown;
SignalException (ReservedInstruction, 0);
}
break;
+ case fmt_ps:
+ FPR_STATE[fpr] = fmt_unknown;
+ SignalException (ReservedInstruction, 0);
+ break;
+
default:
FPR_STATE[fpr] = fmt_unknown;
err = -1;
@@ -299,7 +334,7 @@ store_fpr (sim_cpu *cpu,
}
-/* CP1 control/status registers */
+/* CP1 control/status register access functions. */
void
test_fcsr (sim_cpu *cpu,
@@ -324,22 +359,22 @@ value_fcr(sim_cpu *cpu,
switch (fcr)
{
- case 0: /* FP Implementation and Revision Register */
+ case 0: /* FP Implementation and Revision Register. */
value = FCR0;
break;
- case 25: /* FP Condition Codes Register */
+ case 25: /* FP Condition Codes Register (derived from FCSR). */
value = (FCR31 & fcsr_FCC_mask) >> fcsr_FCC_shift;
- value = (value & 0x1) | (value >> 1); /* close FCC gap */
+ value = (value & 0x1) | (value >> 1); /* Close FCC gap. */
break;
- case 26: /* FP Exceptions Register */
+ case 26: /* FP Exceptions Register (derived from FCSR). */
value = FCR31 & (fcsr_CAUSE_mask | fcsr_FLAGS_mask);
break;
- case 28: /* FP Enables Register */
+ case 28: /* FP Enables Register (derived from FCSR). */
value = FCR31 & (fcsr_ENABLES_mask | fcsr_RM_mask);
- if (FCR31 & fcsr_FS)
- value |= 0x4; /* nonstandard FS bit */
+ if ((FCR31 & fcsr_FS) != 0)
+ value |= fenr_FS;
break;
- case 31: /* FP Control/Status Register */
+ case 31: /* FP Control/Status Register (FCSR). */
value = FCR31 & ~fcsr_ZERO_mask;
break;
}
@@ -358,18 +393,18 @@ store_fcr(sim_cpu *cpu,
v = VL4_8(value);
switch (fcr)
{
- case 25: /* FP Condition Codes Register */
- v = (v << 1) | (v & 0x1); /* adjust for FCC gap */
+ case 25: /* FP Condition Codes Register (stored into FCSR). */
+ v = (v << 1) | (v & 0x1); /* Adjust for FCC gap. */
FCR31 &= ~fcsr_FCC_mask;
FCR31 |= ((v << fcsr_FCC_shift) & fcsr_FCC_mask);
break;
- case 26: /* FP Exceptions Register */
+ case 26: /* FP Exceptions Register (stored into FCSR). */
FCR31 &= ~(fcsr_CAUSE_mask | fcsr_FLAGS_mask);
FCR31 |= (v & (fcsr_CAUSE_mask | fcsr_FLAGS_mask));
test_fcsr(cpu, cia);
break;
- case 28: /* FP Enables Register */
- if (v & 0x4) /* nonstandard FS bit */
+ case 28: /* FP Enables Register (stored into FCSR). */
+ if ((v & fenr_FS) != 0)
v |= fcsr_FS;
else
v &= ~fcsr_FS;
@@ -377,7 +412,7 @@ store_fcr(sim_cpu *cpu,
FCR31 |= (v & (fcsr_FS | fcsr_ENABLES_mask | fcsr_RM_mask));
test_fcsr(cpu, cia);
break;
- case 31: /* FP Control/Status Register */
+ case 31: /* FP Control/Status Register (FCSR). */
FCR31 = v & ~fcsr_ZERO_mask;
test_fcsr(cpu, cia);
break;
@@ -414,7 +449,7 @@ update_fcsr (sim_cpu *cpu,
if (status & sim_fpu_status_inexact)
cause |= (1 << IR);
#if 0 /* Not yet. */
- /* Implicit clearing of other bits by unimplemented done by callers. */
+ /* Implicit clearing of other bits by unimplemented done by callers. */
if (status & sim_fpu_status_unimplemented)
cause |= (1 << UO);
#endif
@@ -426,6 +461,59 @@ update_fcsr (sim_cpu *cpu,
return;
}
+static sim_fpu_round
+rounding_mode(int rm)
+{
+ sim_fpu_round round;
+
+ switch (rm)
+ {
+ case FP_RM_NEAREST:
+ /* Round result to nearest representable value. When two
+ representable values are equally near, round to the value
+ that has a least significant bit of zero (i.e. is even). */
+ round = sim_fpu_round_near;
+ break;
+ case FP_RM_TOZERO:
+ /* Round result to the value closest to, and not greater in
+ magnitude than, the result. */
+ round = sim_fpu_round_zero;
+ break;
+ case FP_RM_TOPINF:
+ /* Round result to the value closest to, and not less than,
+ the result. */
+ round = sim_fpu_round_up;
+ break;
+ case FP_RM_TOMINF:
+ /* Round result to the value closest to, and not greater than,
+ the result. */
+ round = sim_fpu_round_down;
+ break;
+ default:
+ round = 0;
+ fprintf (stderr, "Bad switch\n");
+ abort ();
+ }
+ return round;
+}
+
+/* When the FS bit is set, MIPS processors return zero for
+ denormalized results and optionally replace denormalized inputs
+ with zero. When FS is clear, some implementation trap on input
+ and/or output, while other perform the operation in hardware. */
+static sim_fpu_denorm
+denorm_mode(sim_cpu *cpu)
+{
+ sim_fpu_denorm denorm;
+
+ /* XXX: FIXME: Eventually should be CPU model dependent. */
+ if (GETFS())
+ denorm = sim_fpu_denorm_zero;
+ else
+ denorm = 0;
+ return denorm;
+}
+
/* Comparison operations. */
@@ -500,7 +588,9 @@ fp_cmp(sim_cpu *cpu,
{
sim_fpu_status status = 0;
- /* The format type should already have been checked: */
+ /* The format type should already have been checked. The FCSR is
+ updated before the condition codes so that any exceptions will
+ be signalled before the condition codes are changed. */
switch (fmt)
{
case fmt_single:
@@ -512,6 +602,18 @@ fp_cmp(sim_cpu *cpu,
SETFCC (cc, result);
break;
}
+ case fmt_ps:
+ {
+ int result0, result1;
+ status = fp_test(FP_PS_lower (op1), FP_PS_lower (op2), fmt_single,
+ abs, cond, &result0);
+ status |= fp_test(FP_PS_upper (op1), FP_PS_upper (op2), fmt_single,
+ abs, cond, &result1);
+ update_fcsr (cpu, cia, status);
+ SETFCC (cc, result0);
+ SETFCC (cc+1, result1);
+ break;
+ }
default:
sim_io_eprintf (SD, "Bad switch\n");
abort ();
@@ -530,6 +632,9 @@ fp_unary(sim_cpu *cpu,
{
sim_fpu wop;
sim_fpu ans;
+ sim_fpu_round round = rounding_mode (GETRM());
+ sim_fpu_denorm denorm = denorm_mode (cpu);
+ sim_fpu_status status = 0;
unsigned64 result = 0;
/* The format type has already been checked: */
@@ -539,7 +644,8 @@ fp_unary(sim_cpu *cpu,
{
unsigned32 res;
sim_fpu_32to (&wop, op);
- (*sim_fpu_op) (&ans, &wop);
+ status |= (*sim_fpu_op) (&ans, &wop);
+ status |= sim_fpu_round_32 (&ans, round, denorm);
sim_fpu_to32 (&res, &ans);
result = res;
break;
@@ -548,16 +654,32 @@ fp_unary(sim_cpu *cpu,
{
unsigned64 res;
sim_fpu_64to (&wop, op);
- (*sim_fpu_op) (&ans, &wop);
+ status |= (*sim_fpu_op) (&ans, &wop);
+ status |= sim_fpu_round_64 (&ans, round, denorm);
sim_fpu_to64 (&res, &ans);
result = res;
break;
}
+ case fmt_ps:
+ {
+ int status_u = 0, status_l = 0;
+ unsigned32 res_u, res_l;
+ sim_fpu_32to (&wop, FP_PS_upper(op));
+ status_u |= (*sim_fpu_op) (&ans, &wop);
+ sim_fpu_to32 (&res_u, &ans);
+ sim_fpu_32to (&wop, FP_PS_lower(op));
+ status_l |= (*sim_fpu_op) (&ans, &wop);
+ sim_fpu_to32 (&res_l, &ans);
+ result = FP_PS_cat(res_u, res_l);
+ status = status_u | status_l;
+ break;
+ }
default:
sim_io_eprintf (SD, "Bad switch\n");
abort ();
}
+ update_fcsr (cpu, cia, status);
return result;
}
@@ -572,6 +694,9 @@ fp_binary(sim_cpu *cpu,
sim_fpu wop1;
sim_fpu wop2;
sim_fpu ans;
+ sim_fpu_round round = rounding_mode (GETRM());
+ sim_fpu_denorm denorm = denorm_mode (cpu);
+ sim_fpu_status status = 0;
unsigned64 result = 0;
/* The format type has already been checked: */
@@ -582,7 +707,8 @@ fp_binary(sim_cpu *cpu,
unsigned32 res;
sim_fpu_32to (&wop1, op1);
sim_fpu_32to (&wop2, op2);
- (*sim_fpu_op) (&ans, &wop1, &wop2);
+ status |= (*sim_fpu_op) (&ans, &wop1, &wop2);
+ status |= sim_fpu_round_32 (&ans, round, denorm);
sim_fpu_to32 (&res, &ans);
result = res;
break;
@@ -592,16 +718,260 @@ fp_binary(sim_cpu *cpu,
unsigned64 res;
sim_fpu_64to (&wop1, op1);
sim_fpu_64to (&wop2, op2);
- (*sim_fpu_op) (&ans, &wop1, &wop2);
+ status |= (*sim_fpu_op) (&ans, &wop1, &wop2);
+ status |= sim_fpu_round_64 (&ans, round, denorm);
sim_fpu_to64 (&res, &ans);
result = res;
break;
}
+ case fmt_ps:
+ {
+ int status_u = 0, status_l = 0;
+ unsigned32 res_u, res_l;
+ sim_fpu_32to (&wop1, FP_PS_upper(op1));
+ sim_fpu_32to (&wop2, FP_PS_upper(op2));
+ status_u |= (*sim_fpu_op) (&ans, &wop1, &wop2);
+ sim_fpu_to32 (&res_u, &ans);
+ sim_fpu_32to (&wop1, FP_PS_lower(op1));
+ sim_fpu_32to (&wop2, FP_PS_lower(op2));
+ status_l |= (*sim_fpu_op) (&ans, &wop1, &wop2);
+ sim_fpu_to32 (&res_l, &ans);
+ result = FP_PS_cat(res_u, res_l);
+ status = status_u | status_l;
+ break;
+ }
+ default:
+ sim_io_eprintf (SD, "Bad switch\n");
+ abort ();
+ }
+
+ update_fcsr (cpu, cia, status);
+ return result;
+}
+
+/* Common MAC code for single operands (.s or .d), defers setting FCSR. */
+static sim_fpu_status
+inner_mac(int (*sim_fpu_op)(sim_fpu *, const sim_fpu *, const sim_fpu *),
+ unsigned64 op1,
+ unsigned64 op2,
+ unsigned64 op3,
+ int scale,
+ int negate,
+ FP_formats fmt,
+ sim_fpu_round round,
+ sim_fpu_denorm denorm,
+ unsigned64 *result)
+{
+ sim_fpu wop1;
+ sim_fpu wop2;
+ sim_fpu ans;
+ sim_fpu_status status = 0;
+ sim_fpu_status op_status;
+ unsigned64 temp = 0;
+
+ switch (fmt)
+ {
+ case fmt_single:
+ {
+ unsigned32 res;
+ sim_fpu_32to (&wop1, op1);
+ sim_fpu_32to (&wop2, op2);
+ status |= sim_fpu_mul (&ans, &wop1, &wop2);
+ if (scale != 0 && sim_fpu_is_number (&ans)) /* number or denorm */
+ ans.normal_exp += scale;
+ status |= sim_fpu_round_32 (&ans, round, denorm);
+ wop1 = ans;
+ op_status = 0;
+ sim_fpu_32to (&wop2, op3);
+ op_status |= (*sim_fpu_op) (&ans, &wop1, &wop2);
+ op_status |= sim_fpu_round_32 (&ans, round, denorm);
+ status |= op_status;
+ if (negate)
+ {
+ wop1 = ans;
+ op_status = sim_fpu_neg (&ans, &wop1);
+ op_status |= sim_fpu_round_32 (&ans, round, denorm);
+ status |= op_status;
+ }
+ sim_fpu_to32 (&res, &ans);
+ temp = res;
+ break;
+ }
+ case fmt_double:
+ {
+ unsigned64 res;
+ sim_fpu_64to (&wop1, op1);
+ sim_fpu_64to (&wop2, op2);
+ status |= sim_fpu_mul (&ans, &wop1, &wop2);
+ if (scale != 0 && sim_fpu_is_number (&ans)) /* number or denorm */
+ ans.normal_exp += scale;
+ status |= sim_fpu_round_64 (&ans, round, denorm);
+ wop1 = ans;
+ op_status = 0;
+ sim_fpu_64to (&wop2, op3);
+ op_status |= (*sim_fpu_op) (&ans, &wop1, &wop2);
+ op_status |= sim_fpu_round_64 (&ans, round, denorm);
+ status |= op_status;
+ if (negate)
+ {
+ wop1 = ans;
+ op_status = sim_fpu_neg (&ans, &wop1);
+ op_status |= sim_fpu_round_64 (&ans, round, denorm);
+ status |= op_status;
+ }
+ sim_fpu_to64 (&res, &ans);
+ temp = res;
+ break;
+ }
+ default:
+ fprintf (stderr, "Bad switch\n");
+ abort ();
+ }
+ *result = temp;
+ return status;
+}
+
+/* Common implementation of madd, nmadd, msub, nmsub that does
+ intermediate rounding per spec. Also used for recip2 and rsqrt2,
+ which are transformed into equivalent nmsub operations. The scale
+ argument is an adjustment to the exponent of the intermediate
+ product op1*op2. It is currently non-zero for rsqrt2 (-1), which
+ requires an effective division by 2. */
+static unsigned64
+fp_mac(sim_cpu *cpu,
+ address_word cia,
+ int (*sim_fpu_op)(sim_fpu *, const sim_fpu *, const sim_fpu *),
+ unsigned64 op1,
+ unsigned64 op2,
+ unsigned64 op3,
+ int scale,
+ int negate,
+ FP_formats fmt)
+{
+ sim_fpu_round round = rounding_mode (GETRM());
+ sim_fpu_denorm denorm = denorm_mode (cpu);
+ sim_fpu_status status = 0;
+ unsigned64 result = 0;
+
+ /* The format type has already been checked: */
+ switch (fmt)
+ {
+ case fmt_single:
+ case fmt_double:
+ status = inner_mac(sim_fpu_op, op1, op2, op3, scale,
+ negate, fmt, round, denorm, &result);
+ break;
+ case fmt_ps:
+ {
+ int status_u, status_l;
+ unsigned64 result_u, result_l;
+ status_u = inner_mac(sim_fpu_op, FP_PS_upper(op1), FP_PS_upper(op2),
+ FP_PS_upper(op3), scale, negate, fmt_single,
+ round, denorm, &result_u);
+ status_l = inner_mac(sim_fpu_op, FP_PS_lower(op1), FP_PS_lower(op2),
+ FP_PS_lower(op3), scale, negate, fmt_single,
+ round, denorm, &result_l);
+ result = FP_PS_cat(result_u, result_l);
+ status = status_u | status_l;
+ break;
+ }
+ default:
+ sim_io_eprintf (SD, "Bad switch\n");
+ abort ();
+ }
+
+ update_fcsr (cpu, cia, status);
+ return result;
+}
+
+/* Common rsqrt code for single operands (.s or .d), intermediate rounding. */
+static sim_fpu_status
+inner_rsqrt(unsigned64 op1,
+ FP_formats fmt,
+ sim_fpu_round round,
+ sim_fpu_denorm denorm,
+ unsigned64 *result)
+{
+ sim_fpu wop1;
+ sim_fpu ans;
+ sim_fpu_status status = 0;
+ sim_fpu_status op_status;
+ unsigned64 temp = 0;
+
+ switch (fmt)
+ {
+ case fmt_single:
+ {
+ unsigned32 res;
+ sim_fpu_32to (&wop1, op1);
+ status |= sim_fpu_sqrt (&ans, &wop1);
+ status |= sim_fpu_round_32 (&ans, status, round);
+ wop1 = ans;
+ op_status = sim_fpu_inv (&ans, &wop1);
+ op_status |= sim_fpu_round_32 (&ans, round, denorm);
+ sim_fpu_to32 (&res, &ans);
+ temp = res;
+ status |= op_status;
+ break;
+ }
+ case fmt_double:
+ {
+ unsigned64 res;
+ sim_fpu_64to (&wop1, op1);
+ status |= sim_fpu_sqrt (&ans, &wop1);
+ status |= sim_fpu_round_64 (&ans, round, denorm);
+ wop1 = ans;
+ op_status = sim_fpu_inv (&ans, &wop1);
+ op_status |= sim_fpu_round_64 (&ans, round, denorm);
+ sim_fpu_to64 (&res, &ans);
+ temp = res;
+ status |= op_status;
+ break;
+ }
+ default:
+ fprintf (stderr, "Bad switch\n");
+ abort ();
+ }
+ *result = temp;
+ return status;
+}
+
+static unsigned64
+fp_inv_sqrt(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ FP_formats fmt)
+{
+ sim_fpu_round round = rounding_mode (GETRM());
+ sim_fpu_round denorm = denorm_mode (cpu);
+ sim_fpu_status status = 0;
+ unsigned64 result = 0;
+
+ /* The format type has already been checked: */
+ switch (fmt)
+ {
+ case fmt_single:
+ case fmt_double:
+ status = inner_rsqrt (op1, fmt, round, denorm, &result);
+ break;
+ case fmt_ps:
+ {
+ int status_u, status_l;
+ unsigned64 result_u, result_l;
+ status_u = inner_rsqrt (FP_PS_upper(op1), fmt_single, round, denorm,
+ &result_u);
+ status_l = inner_rsqrt (FP_PS_lower(op1), fmt_single, round, denorm,
+ &result_l);
+ result = FP_PS_cat(result_u, result_l);
+ status = status_u | status_l;
+ break;
+ }
default:
sim_io_eprintf (SD, "Bad switch\n");
abort ();
}
+ update_fcsr (cpu, cia, status);
return result;
}
@@ -682,6 +1052,309 @@ fp_sqrt(sim_cpu *cpu,
return fp_unary(cpu, cia, &sim_fpu_sqrt, op, fmt);
}
+unsigned64
+fp_rsqrt(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op,
+ FP_formats fmt)
+{
+ return fp_inv_sqrt(cpu, cia, op, fmt);
+}
+
+unsigned64
+fp_madd(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ unsigned64 op2,
+ unsigned64 op3,
+ FP_formats fmt)
+{
+ return fp_mac(cpu, cia, &sim_fpu_add, op1, op2, op3, 0, 0, fmt);
+}
+
+unsigned64
+fp_msub(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ unsigned64 op2,
+ unsigned64 op3,
+ FP_formats fmt)
+{
+ return fp_mac(cpu, cia, &sim_fpu_sub, op1, op2, op3, 0, 0, fmt);
+}
+
+unsigned64
+fp_nmadd(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ unsigned64 op2,
+ unsigned64 op3,
+ FP_formats fmt)
+{
+ return fp_mac(cpu, cia, &sim_fpu_add, op1, op2, op3, 0, 1, fmt);
+}
+
+unsigned64
+fp_nmsub(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ unsigned64 op2,
+ unsigned64 op3,
+ FP_formats fmt)
+{
+ return fp_mac(cpu, cia, &sim_fpu_sub, op1, op2, op3, 0, 1, fmt);
+}
+
+
+/* MIPS-3D ASE operations. */
+
+/* Variant of fp_binary for *r.ps MIPS-3D operations. */
+static unsigned64
+fp_binary_r(sim_cpu *cpu,
+ address_word cia,
+ int (*sim_fpu_op)(sim_fpu *, const sim_fpu *, const sim_fpu *),
+ unsigned64 op1,
+ unsigned64 op2)
+{
+ sim_fpu wop1;
+ sim_fpu wop2;
+ sim_fpu ans;
+ sim_fpu_round round = rounding_mode (GETRM ());
+ sim_fpu_denorm denorm = denorm_mode (cpu);
+ sim_fpu_status status_u, status_l;
+ unsigned64 result;
+ unsigned32 res_u, res_l;
+
+ /* The format must be fmt_ps. */
+ status_u = 0;
+ sim_fpu_32to (&wop1, FP_PS_upper (op1));
+ sim_fpu_32to (&wop2, FP_PS_lower (op1));
+ status_u |= (*sim_fpu_op) (&ans, &wop1, &wop2);
+ status_u |= sim_fpu_round_32 (&ans, round, denorm);
+ sim_fpu_to32 (&res_u, &ans);
+ status_l = 0;
+ sim_fpu_32to (&wop1, FP_PS_upper (op2));
+ sim_fpu_32to (&wop2, FP_PS_lower (op2));
+ status_l |= (*sim_fpu_op) (&ans, &wop1, &wop2);
+ status_l |= sim_fpu_round_32 (&ans, round, denorm);
+ sim_fpu_to32 (&res_l, &ans);
+ result = FP_PS_cat (res_u, res_l);
+
+ update_fcsr (cpu, cia, status_u | status_l);
+ return result;
+}
+
+unsigned64
+fp_add_r(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ unsigned64 op2,
+ FP_formats fmt)
+{
+ return fp_binary_r (cpu, cia, &sim_fpu_add, op1, op2);
+}
+
+unsigned64
+fp_mul_r(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ unsigned64 op2,
+ FP_formats fmt)
+{
+ return fp_binary_r (cpu, cia, &sim_fpu_mul, op1, op2);
+}
+
+#define NR_FRAC_GUARD (60)
+#define IMPLICIT_1 LSBIT64 (NR_FRAC_GUARD)
+
+static int
+fpu_inv1(sim_fpu *f, const sim_fpu *l)
+{
+ static const sim_fpu sim_fpu_one = {
+ sim_fpu_class_number, 0, IMPLICIT_1, 0
+ };
+ int status = 0;
+ sim_fpu t;
+
+ if (sim_fpu_is_zero (l))
+ {
+ *f = sim_fpu_maxfp;
+ f->sign = l->sign;
+ return sim_fpu_status_invalid_div0;
+ }
+ if (sim_fpu_is_infinity (l))
+ {
+ *f = sim_fpu_zero;
+ f->sign = l->sign;
+ return status;
+ }
+ status |= sim_fpu_div (f, &sim_fpu_one, l);
+ return status;
+}
+
+static int
+fpu_inv1_32(sim_fpu *f, const sim_fpu *l)
+{
+ if (sim_fpu_is_zero (l))
+ {
+ *f = sim_fpu_max32;
+ f->sign = l->sign;
+ return sim_fpu_status_invalid_div0;
+ }
+ return fpu_inv1 (f, l);
+}
+
+static int
+fpu_inv1_64(sim_fpu *f, const sim_fpu *l)
+{
+ if (sim_fpu_is_zero (l))
+ {
+ *f = sim_fpu_max64;
+ f->sign = l->sign;
+ return sim_fpu_status_invalid_div0;
+ }
+ return fpu_inv1 (f, l);
+}
+
+unsigned64
+fp_recip1(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op,
+ FP_formats fmt)
+{
+ switch (fmt)
+ {
+ case fmt_single:
+ case fmt_ps:
+ return fp_unary (cpu, cia, &fpu_inv1_32, op, fmt);
+ case fmt_double:
+ return fp_unary (cpu, cia, &fpu_inv1_64, op, fmt);
+ }
+ return 0;
+}
+
+unsigned64
+fp_recip2(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ unsigned64 op2,
+ FP_formats fmt)
+{
+ static const unsigned64 one_single = UNSIGNED64 (0x3F800000);
+ static const unsigned64 one_double = UNSIGNED64 (0x3FF0000000000000);
+ static const unsigned64 one_ps = (UNSIGNED64 (0x3F800000) << 32 | UNSIGNED64 (0x3F800000));
+ unsigned64 one;
+
+ /* Implemented as nmsub fd, 1, fs, ft. */
+ switch (fmt)
+ {
+ case fmt_single: one = one_single; break;
+ case fmt_double: one = one_double; break;
+ case fmt_ps: one = one_ps; break;
+ default: one = 0; abort ();
+ }
+ return fp_mac (cpu, cia, &sim_fpu_sub, op1, op2, one, 0, 1, fmt);
+}
+
+static int
+fpu_inv_sqrt1(sim_fpu *f, const sim_fpu *l)
+{
+ static const sim_fpu sim_fpu_one = {
+ sim_fpu_class_number, 0, IMPLICIT_1, 0
+ };
+ int status = 0;
+ sim_fpu t;
+
+ if (sim_fpu_is_zero (l))
+ {
+ *f = sim_fpu_maxfp;
+ f->sign = l->sign;
+ return sim_fpu_status_invalid_div0;
+ }
+ if (sim_fpu_is_infinity (l))
+ {
+ if (!l->sign)
+ {
+ f->class = sim_fpu_class_zero;
+ f->sign = 0;
+ }
+ else
+ {
+ *f = sim_fpu_qnan;
+ status = sim_fpu_status_invalid_sqrt;
+ }
+ return status;
+ }
+ status |= sim_fpu_sqrt (&t, l);
+ status |= sim_fpu_div (f, &sim_fpu_one, &t);
+ return status;
+}
+
+static int
+fpu_inv_sqrt1_32(sim_fpu *f, const sim_fpu *l)
+{
+ if (sim_fpu_is_zero (l))
+ {
+ *f = sim_fpu_max32;
+ f->sign = l->sign;
+ return sim_fpu_status_invalid_div0;
+ }
+ return fpu_inv_sqrt1 (f, l);
+}
+
+static int
+fpu_inv_sqrt1_64(sim_fpu *f, const sim_fpu *l)
+{
+ if (sim_fpu_is_zero (l))
+ {
+ *f = sim_fpu_max64;
+ f->sign = l->sign;
+ return sim_fpu_status_invalid_div0;
+ }
+ return fpu_inv_sqrt1 (f, l);
+}
+
+unsigned64
+fp_rsqrt1(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op,
+ FP_formats fmt)
+{
+ switch (fmt)
+ {
+ case fmt_single:
+ case fmt_ps:
+ return fp_unary (cpu, cia, &fpu_inv_sqrt1_32, op, fmt);
+ case fmt_double:
+ return fp_unary (cpu, cia, &fpu_inv_sqrt1_64, op, fmt);
+ }
+ return 0;
+}
+
+unsigned64
+fp_rsqrt2(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ unsigned64 op2,
+ FP_formats fmt)
+{
+ static const unsigned64 half_single = UNSIGNED64 (0x3F000000);
+ static const unsigned64 half_double = UNSIGNED64 (0x3FE0000000000000);
+ static const unsigned64 half_ps = (UNSIGNED64 (0x3F000000) << 32 | UNSIGNED64 (0x3F000000));
+ unsigned64 half;
+
+ /* Implemented as (nmsub fd, 0.5, fs, ft)/2, where the divide is
+ done by scaling the exponent during multiply. */
+ switch (fmt)
+ {
+ case fmt_single: half = half_single; break;
+ case fmt_double: half = half_double; break;
+ case fmt_ps: half = half_ps; break;
+ default: half = 0; abort ();
+ }
+ return fp_mac (cpu, cia, &sim_fpu_sub, op1, op2, half, -1, 1, fmt);
+}
+
/* Conversion operations. */
@@ -694,47 +1367,11 @@ convert (sim_cpu *cpu,
FP_formats to)
{
sim_fpu wop;
- sim_fpu_round round;
+ sim_fpu_round round = rounding_mode (rm);
+ sim_fpu_denorm denorm = denorm_mode (cpu);
unsigned32 result32;
unsigned64 result64;
-
-#ifdef DEBUG
-#if 0 /* FIXME: doesn't compile */
- printf ("DBG: Convert: mode %s : op 0x%s : from %s : to %s : (PC = 0x%s)\n",
- fpu_rounding_mode_name (rm), pr_addr (op), fpu_format_name (from),
- fpu_format_name (to), pr_addr (IPC));
-#endif
-#endif /* DEBUG */
-
- switch (rm)
- {
- case FP_RM_NEAREST:
- /* Round result to nearest representable value. When two
- representable values are equally near, round to the value
- that has a least significant bit of zero (i.e. is even). */
- round = sim_fpu_round_near;
- break;
- case FP_RM_TOZERO:
- /* Round result to the value closest to, and not greater in
- magnitude than, the result. */
- round = sim_fpu_round_zero;
- break;
- case FP_RM_TOPINF:
- /* Round result to the value closest to, and not less than,
- the result. */
- round = sim_fpu_round_up;
- break;
-
- case FP_RM_TOMINF:
- /* Round result to the value closest to, and not greater than,
- the result. */
- round = sim_fpu_round_down;
- break;
- default:
- round = 0;
- fprintf (stderr, "Bad switch\n");
- abort ();
- }
+ sim_fpu_status status = 0;
/* Convert the input to sim_fpu internal format */
switch (from)
@@ -746,13 +1383,13 @@ convert (sim_cpu *cpu,
sim_fpu_32to (&wop, op);
break;
case fmt_word:
- sim_fpu_i32to (&wop, op, round);
+ status = sim_fpu_i32to (&wop, op, round);
break;
case fmt_long:
- sim_fpu_i64to (&wop, op, round);
+ status = sim_fpu_i64to (&wop, op, round);
break;
default:
- fprintf (stderr, "Bad switch\n");
+ sim_io_eprintf (SD, "Bad switch\n");
abort ();
}
@@ -760,38 +1397,148 @@ convert (sim_cpu *cpu,
/* The value WOP is converted to the destination format, rounding
using mode RM. When the destination is a fixed-point format, then
a source value of Infinity, NaN or one which would round to an
- integer outside the fixed point range then an IEEE Invalid
- Operation condition is raised. */
+ integer outside the fixed point range then an IEEE Invalid Operation
+ condition is raised. Not used if destination format is PS. */
switch (to)
{
case fmt_single:
- sim_fpu_round_32 (&wop, round, 0);
+ status |= sim_fpu_round_32 (&wop, round, denorm);
+ /* For a NaN, normalize mantissa bits (cvt.s.d can't preserve them) */
+ if (sim_fpu_is_qnan (&wop))
+ wop = sim_fpu_qnan;
sim_fpu_to32 (&result32, &wop);
result64 = result32;
break;
case fmt_double:
- sim_fpu_round_64 (&wop, round, 0);
+ status |= sim_fpu_round_64 (&wop, round, denorm);
+ /* For a NaN, normalize mantissa bits (make cvt.d.s consistent) */
+ if (sim_fpu_is_qnan (&wop))
+ wop = sim_fpu_qnan;
sim_fpu_to64 (&result64, &wop);
break;
case fmt_word:
- sim_fpu_to32i (&result32, &wop, round);
+ status |= sim_fpu_to32i (&result32, &wop, round);
result64 = result32;
break;
case fmt_long:
- sim_fpu_to64i (&result64, &wop, round);
+ status |= sim_fpu_to64i (&result64, &wop, round);
break;
default:
result64 = 0;
- fprintf (stderr, "Bad switch\n");
+ sim_io_eprintf (SD, "Bad switch\n");
abort ();
}
-#ifdef DEBUG
- printf ("DBG: Convert: returning 0x%s (to format = %s)\n",
- pr_addr (result64), fpu_format_name (to));
-#endif /* DEBUG */
+ update_fcsr (cpu, cia, status);
+ return result64;
+}
+
+unsigned64
+ps_lower(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op)
+{
+ return FP_PS_lower (op);
+}
- return (result64);
+unsigned64
+ps_upper(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op)
+{
+ return FP_PS_upper(op);
+}
+
+unsigned64
+pack_ps(sim_cpu *cpu,
+ address_word cia,
+ unsigned64 op1,
+ unsigned64 op2,
+ FP_formats fmt)
+{
+ unsigned64 result = 0;
+
+ /* The registers must specify FPRs valid for operands of type
+ "fmt". If they are not valid, the result is undefined. */
+
+ /* The format type should already have been checked: */
+ switch (fmt)
+ {
+ case fmt_single:
+ {
+ sim_fpu wop;
+ unsigned32 res_u, res_l;
+ sim_fpu_32to (&wop, op1);
+ sim_fpu_to32 (&res_u, &wop);
+ sim_fpu_32to (&wop, op2);
+ sim_fpu_to32 (&res_l, &wop);
+ result = FP_PS_cat(res_u, res_l);
+ break;
+ }
+ default:
+ sim_io_eprintf (SD, "Bad switch\n");
+ abort ();
+ }
+
+ return result;
+}
+
+unsigned64
+convert_ps (sim_cpu *cpu,
+ address_word cia,
+ int rm,
+ unsigned64 op,
+ FP_formats from,
+ FP_formats to)
+{
+ sim_fpu wop_u, wop_l;
+ sim_fpu_round round = rounding_mode (rm);
+ sim_fpu_denorm denorm = denorm_mode (cpu);
+ unsigned32 res_u, res_l;
+ unsigned64 result;
+ sim_fpu_status status_u = 0, status_l = 0;
+
+ /* As convert, but used only for paired values (formats PS, PW) */
+
+ /* Convert the input to sim_fpu internal format */
+ switch (from)
+ {
+ case fmt_word: /* fmt_pw */
+ sim_fpu_i32to (&wop_u, (op >> 32) & (unsigned)0xFFFFFFFF, round);
+ sim_fpu_i32to (&wop_l, op & (unsigned)0xFFFFFFFF, round);
+ break;
+ case fmt_ps:
+ sim_fpu_32to (&wop_u, FP_PS_upper(op));
+ sim_fpu_32to (&wop_l, FP_PS_lower(op));
+ break;
+ default:
+ sim_io_eprintf (SD, "Bad switch\n");
+ abort ();
+ }
+
+ /* Convert sim_fpu format into the output */
+ switch (to)
+ {
+ case fmt_word: /* fmt_pw */
+ status_u |= sim_fpu_to32i (&res_u, &wop_u, round);
+ status_l |= sim_fpu_to32i (&res_l, &wop_l, round);
+ result = (((unsigned64)res_u) << 32) | (unsigned64)res_l;
+ break;
+ case fmt_ps:
+ status_u |= sim_fpu_round_32 (&wop_u, 0, round);
+ status_l |= sim_fpu_round_32 (&wop_l, 0, round);
+ sim_fpu_to32 (&res_u, &wop_u);
+ sim_fpu_to32 (&res_l, &wop_l);
+ result = FP_PS_cat(res_u, res_l);
+ break;
+ default:
+ result = 0;
+ sim_io_eprintf (SD, "Bad switch\n");
+ abort ();
+ }
+
+ update_fcsr (cpu, cia, status_u | status_l);
+ return result;
}
static const char *
@@ -807,6 +1554,8 @@ fpu_format_name (FP_formats fmt)
return "word";
case fmt_long:
return "long";
+ case fmt_ps:
+ return "ps";
case fmt_unknown:
return "";
case fmt_uninterpreted: