MIPS: math-emu: Switch to using the MIPS rounding modes.

Previously math-emu was using the IEEE-754 constants internally.  These
were differing by having the constants for rounding to +/- infinity
switched, so a conversion was necessary.  This would be entirely
avoidable if the MIPS constants were used throughout, so get rid of
the bloat.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
This commit is contained in:
Ralf Baechle 2014-04-30 11:21:55 +02:00
parent aef3fb76aa
commit 56a6473339
15 changed files with 63 additions and 91 deletions

View file

@ -67,21 +67,6 @@ static int fpux_emu(struct pt_regs *,
/* Determine rounding mode from the RM bits of the FCSR */
#define modeindex(v) ((v) & FPU_CSR_RM)
/* Convert MIPS rounding mode (0..3) to IEEE library modes. */
static const unsigned char ieee_rm[4] = {
[FPU_CSR_RN] = IEEE754_RN,
[FPU_CSR_RZ] = IEEE754_RZ,
[FPU_CSR_RU] = IEEE754_RU,
[FPU_CSR_RD] = IEEE754_RD,
};
/* Convert IEEE library modes to MIPS rounding mode (0..3). */
static const unsigned char mips_rm[4] = {
[IEEE754_RN] = FPU_CSR_RN,
[IEEE754_RZ] = FPU_CSR_RZ,
[IEEE754_RD] = FPU_CSR_RD,
[IEEE754_RU] = FPU_CSR_RU,
};
/* convert condition code register number to csr bit */
static const unsigned int fpucondbit[8] = {
FPU_CSR_COND0,
@ -907,8 +892,7 @@ static int cop1Emulate(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
/* cop control register rd -> gpr[rt] */
if (MIPSInst_RD(ir) == FPCREG_CSR) {
value = ctx->fcr31;
value = (value & ~FPU_CSR_RM) |
mips_rm[modeindex(value)];
value = (value & ~FPU_CSR_RM) | modeindex(value);
pr_debug("%p gpr[%d]<-csr=%08x\n",
(void *) (xcp->cp0_epc),
MIPSInst_RT(ir), value);
@ -939,9 +923,8 @@ static int cop1Emulate(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
* Don't write reserved bits,
* and convert to ieee library modes
*/
ctx->fcr31 = (value &
~(FPU_CSR_RSVD | FPU_CSR_RM)) |
ieee_rm[modeindex(value)];
ctx->fcr31 = (value & ~(FPU_CSR_RSVD | FPU_CSR_RM)) |
modeindex(value);
}
if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
return SIGFPE;
@ -1515,7 +1498,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
oldrm = ieee754_csr.rm;
SPFROMREG(fs, MIPSInst_FS(ir));
ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
ieee754_csr.rm = modeindex(MIPSInst_FUNC(ir));
rv.w = ieee754sp_tint(fs);
ieee754_csr.rm = oldrm;
rfmt = w_fmt;
@ -1539,7 +1522,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
oldrm = ieee754_csr.rm;
SPFROMREG(fs, MIPSInst_FS(ir));
ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
ieee754_csr.rm = modeindex(MIPSInst_FUNC(ir));
rv.l = ieee754sp_tlong(fs);
ieee754_csr.rm = oldrm;
rfmt = l_fmt;
@ -1692,7 +1675,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
oldrm = ieee754_csr.rm;
DPFROMREG(fs, MIPSInst_FS(ir));
ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
ieee754_csr.rm = modeindex(MIPSInst_FUNC(ir));
rv.w = ieee754dp_tint(fs);
ieee754_csr.rm = oldrm;
rfmt = w_fmt;
@ -1716,7 +1699,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
oldrm = ieee754_csr.rm;
DPFROMREG(fs, MIPSInst_FS(ir));
ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
ieee754_csr.rm = modeindex(MIPSInst_FUNC(ir));
rv.l = ieee754dp_tlong(fs);
ieee754_csr.rm = oldrm;
rfmt = l_fmt;
@ -1926,11 +1909,7 @@ int fpu_emulator_cop1Handler(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
* ieee754_csr. But ieee754_csr.rm is ieee
* library modes. (not mips rounding mode)
*/
/* convert to ieee library modes */
ieee754_csr.rm = ieee_rm[ieee754_csr.rm];
sig = cop1Emulate(xcp, ctx, dec_insn, fault_addr);
/* revert to mips rounding mode */
ieee754_csr.rm = mips_rm[ieee754_csr.rm];
}
if (has_fpu)

View file

@ -91,7 +91,7 @@ union ieee754dp ieee754dp_add(union ieee754dp x, union ieee754dp y)
if (xs == ys)
return x;
else
return ieee754dp_zero(ieee754_csr.rm == IEEE754_RD);
return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD);
case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
@ -168,7 +168,7 @@ union ieee754dp ieee754dp_add(union ieee754dp x, union ieee754dp y)
xs = ys;
}
if (xm == 0)
return ieee754dp_zero(ieee754_csr.rm == IEEE754_RD);
return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD);
/*
* Normalize to rounding precision.

View file

@ -80,7 +80,7 @@ union ieee754dp ieee754dp_sqrt(union ieee754dp x)
oldcsr = ieee754_csr;
ieee754_csr.mx &= ~IEEE754_INEXACT;
ieee754_csr.sx &= ~IEEE754_INEXACT;
ieee754_csr.rm = IEEE754_RN;
ieee754_csr.rm = FPU_CSR_RN;
/* adjust exponent to prevent overflow */
scalx = 0;
@ -122,7 +122,7 @@ union ieee754dp ieee754dp_sqrt(union ieee754dp x)
/* twiddle last bit to force y correctly rounded */
/* set RZ, clear INEX flag */
ieee754_csr.rm = IEEE754_RZ;
ieee754_csr.rm = FPU_CSR_RZ;
ieee754_csr.sx &= ~IEEE754_INEXACT;
/* t=x/y; ...chopped quotient, possibly inexact */
@ -139,10 +139,10 @@ union ieee754dp ieee754dp_sqrt(union ieee754dp x)
oldcsr.sx |= IEEE754_INEXACT;
switch (oldcsr.rm) {
case IEEE754_RU:
case FPU_CSR_RU:
y.bits += 1;
/* drop through */
case IEEE754_RN:
case FPU_CSR_RN:
t.bits += 1;
break;
}

View file

@ -91,7 +91,7 @@ union ieee754dp ieee754dp_sub(union ieee754dp x, union ieee754dp y)
if (xs != ys)
return x;
else
return ieee754dp_zero(ieee754_csr.rm == IEEE754_RD);
return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD);
case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
@ -171,7 +171,7 @@ union ieee754dp ieee754dp_sub(union ieee754dp x, union ieee754dp y)
xs = ys;
}
if (xm == 0) {
if (ieee754_csr.rm == IEEE754_RD)
if (ieee754_csr.rm == FPU_CSR_RD)
return ieee754dp_zero(1); /* round negative inf. => sign = -1 */
else
return ieee754dp_zero(0); /* other round modes => sign = 1 */

View file

@ -74,17 +74,17 @@ int ieee754dp_tint(union ieee754dp x)
to be zero */
odd = (xm & 0x1) != 0x0;
switch (ieee754_csr.rm) {
case IEEE754_RN:
case FPU_CSR_RN:
if (round && (sticky || odd))
xm++;
break;
case IEEE754_RZ:
case FPU_CSR_RZ:
break;
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if ((round || sticky) && !xs)
xm++;
break;
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if ((round || sticky) && xs)
xm++;
break;

View file

@ -79,17 +79,17 @@ s64 ieee754dp_tlong(union ieee754dp x)
}
odd = (xm & 0x1) != 0x0;
switch (ieee754_csr.rm) {
case IEEE754_RN:
case FPU_CSR_RN:
if (round && (sticky || odd))
xm++;
break;
case IEEE754_RZ:
case FPU_CSR_RZ:
break;
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if ((round || sticky) && !xs)
xm++;
break;
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if ((round || sticky) && xs)
xm++;
break;

View file

@ -126,13 +126,6 @@ enum {
#define IEEE754_CGT 0x04
#define IEEE754_CUN 0x08
/* rounding mode
*/
#define IEEE754_RN 0 /* round to nearest */
#define IEEE754_RZ 1 /* round toward zero */
#define IEEE754_RD 2 /* round toward -Infinity */
#define IEEE754_RU 3 /* round toward +Infinity */
/* "normal" comparisons
*/
static inline int ieee754sp_eq(union ieee754sp x, union ieee754sp y)

View file

@ -67,17 +67,17 @@ static u64 ieee754dp_get_rounding(int sn, u64 xm)
*/
if (xm & (DP_MBIT(3) - 1)) {
switch (ieee754_csr.rm) {
case IEEE754_RZ:
case FPU_CSR_RZ:
break;
case IEEE754_RN:
case FPU_CSR_RN:
xm += 0x3 + ((xm >> 3) & 1);
/* xm += (xm&0x8)?0x4:0x3 */
break;
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if (!sn) /* ?? */
xm += 0x8;
break;
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if (sn) /* ?? */
xm += 0x8;
break;
@ -108,15 +108,15 @@ union ieee754dp ieee754dp_format(int sn, int xe, u64 xm)
ieee754_setcx(IEEE754_INEXACT);
switch(ieee754_csr.rm) {
case IEEE754_RN:
case IEEE754_RZ:
case FPU_CSR_RN:
case FPU_CSR_RZ:
return ieee754dp_zero(sn);
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if (sn == 0)
return ieee754dp_min(0);
else
return ieee754dp_zero(1);
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if (sn == 0)
return ieee754dp_zero(0);
else
@ -172,16 +172,16 @@ union ieee754dp ieee754dp_format(int sn, int xe, u64 xm)
ieee754_setcx(IEEE754_INEXACT);
/* -O can be table indexed by (rm,sn) */
switch (ieee754_csr.rm) {
case IEEE754_RN:
case FPU_CSR_RN:
return ieee754dp_inf(sn);
case IEEE754_RZ:
case FPU_CSR_RZ:
return ieee754dp_max(sn);
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if (sn == 0)
return ieee754dp_inf(0);
else
return ieee754dp_max(1);
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if (sn == 0)
return ieee754dp_max(0);
else

View file

@ -67,17 +67,17 @@ static unsigned ieee754sp_get_rounding(int sn, unsigned xm)
*/
if (xm & (SP_MBIT(3) - 1)) {
switch (ieee754_csr.rm) {
case IEEE754_RZ:
case FPU_CSR_RZ:
break;
case IEEE754_RN:
case FPU_CSR_RN:
xm += 0x3 + ((xm >> 3) & 1);
/* xm += (xm&0x8)?0x4:0x3 */
break;
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if (!sn) /* ?? */
xm += 0x8;
break;
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if (sn) /* ?? */
xm += 0x8;
break;
@ -108,15 +108,15 @@ union ieee754sp ieee754sp_format(int sn, int xe, unsigned xm)
ieee754_setcx(IEEE754_INEXACT);
switch(ieee754_csr.rm) {
case IEEE754_RN:
case IEEE754_RZ:
case FPU_CSR_RN:
case FPU_CSR_RZ:
return ieee754sp_zero(sn);
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if (sn == 0)
return ieee754sp_min(0);
else
return ieee754sp_zero(1);
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if (sn == 0)
return ieee754sp_zero(0);
else
@ -170,16 +170,16 @@ union ieee754sp ieee754sp_format(int sn, int xe, unsigned xm)
ieee754_setcx(IEEE754_INEXACT);
/* -O can be table indexed by (rm,sn) */
switch (ieee754_csr.rm) {
case IEEE754_RN:
case FPU_CSR_RN:
return ieee754sp_inf(sn);
case IEEE754_RZ:
case FPU_CSR_RZ:
return ieee754sp_max(sn);
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if (sn == 0)
return ieee754sp_inf(0);
else
return ieee754sp_max(1);
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if (sn == 0)
return ieee754sp_max(0);
else

View file

@ -91,7 +91,7 @@ union ieee754sp ieee754sp_add(union ieee754sp x, union ieee754sp y)
if (xs == ys)
return x;
else
return ieee754sp_zero(ieee754_csr.rm == IEEE754_RD);
return ieee754sp_zero(ieee754_csr.rm == FPU_CSR_RD);
case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
@ -165,7 +165,7 @@ union ieee754sp ieee754sp_add(union ieee754sp x, union ieee754sp y)
xs = ys;
}
if (xm == 0)
return ieee754sp_zero(ieee754_csr.rm == IEEE754_RD);
return ieee754sp_zero(ieee754_csr.rm == FPU_CSR_RD);
/*
* Normalize in extended single precision

View file

@ -57,8 +57,8 @@ union ieee754sp ieee754sp_fdp(union ieee754dp x)
/* can't possibly be sp representable */
ieee754_setcx(IEEE754_UNDERFLOW);
ieee754_setcx(IEEE754_INEXACT);
if ((ieee754_csr.rm == IEEE754_RU && !xs) ||
(ieee754_csr.rm == IEEE754_RD && xs))
if ((ieee754_csr.rm == FPU_CSR_RU && !xs) ||
(ieee754_csr.rm == FPU_CSR_RD && xs))
return ieee754sp_mind(xs);
return ieee754sp_zero(xs);

View file

@ -100,10 +100,10 @@ union ieee754sp ieee754sp_sqrt(union ieee754sp x)
if (ix != 0) {
ieee754_setcx(IEEE754_INEXACT);
switch (ieee754_csr.rm) {
case IEEE754_RU:
case FPU_CSR_RU:
q += 2;
break;
case IEEE754_RN:
case FPU_CSR_RN:
q += (q & 1);
break;
}

View file

@ -91,7 +91,7 @@ union ieee754sp ieee754sp_sub(union ieee754sp x, union ieee754sp y)
if (xs != ys)
return x;
else
return ieee754sp_zero(ieee754_csr.rm == IEEE754_RD);
return ieee754sp_zero(ieee754_csr.rm == FPU_CSR_RD);
case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
@ -165,7 +165,7 @@ union ieee754sp ieee754sp_sub(union ieee754sp x, union ieee754sp y)
xs = ys;
}
if (xm == 0) {
if (ieee754_csr.rm == IEEE754_RD)
if (ieee754_csr.rm == FPU_CSR_RD)
return ieee754sp_zero(1); /* round negative inf. => sign = -1 */
else
return ieee754sp_zero(0); /* other round modes => sign = 1 */

View file

@ -79,17 +79,17 @@ int ieee754sp_tint(union ieee754sp x)
}
odd = (xm & 0x1) != 0x0;
switch (ieee754_csr.rm) {
case IEEE754_RN:
case FPU_CSR_RN:
if (round && (sticky || odd))
xm++;
break;
case IEEE754_RZ:
case FPU_CSR_RZ:
break;
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if ((round || sticky) && !xs)
xm++;
break;
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if ((round || sticky) && xs)
xm++;
break;

View file

@ -76,17 +76,17 @@ s64 ieee754sp_tlong(union ieee754sp x)
}
odd = (xm & 0x1) != 0x0;
switch (ieee754_csr.rm) {
case IEEE754_RN:
case FPU_CSR_RN:
if (round && (sticky || odd))
xm++;
break;
case IEEE754_RZ:
case FPU_CSR_RZ:
break;
case IEEE754_RU: /* toward +Infinity */
case FPU_CSR_RU: /* toward +Infinity */
if ((round || sticky) && !xs)
xm++;
break;
case IEEE754_RD: /* toward -Infinity */
case FPU_CSR_RD: /* toward -Infinity */
if ((round || sticky) && xs)
xm++;
break;