softfloat: Move minmax_flags to softfloat-parts.c.inc
Rename to parts$N_minmax. Combine 3 bool arguments to a bitmask. Introduce ftype_minmax functions as a common optimization point. Fold bfloat16 expansions into the same macro as the other types. Reviewed-by: David Hildenbrand <david@redhat.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This commit is contained in:
parent
37c954a1b9
commit
e1c4667a9b
@ -938,3 +938,83 @@ static void partsN(uint_to_float)(FloatPartsN *p, uint64_t a,
|
||||
p->frac_hi = a << shift;
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* Float min/max.
|
||||
*/
|
||||
static FloatPartsN *partsN(minmax)(FloatPartsN *a, FloatPartsN *b,
|
||||
float_status *s, int flags)
|
||||
{
|
||||
int ab_mask = float_cmask(a->cls) | float_cmask(b->cls);
|
||||
int a_exp, b_exp, cmp;
|
||||
|
||||
if (unlikely(ab_mask & float_cmask_anynan)) {
|
||||
/*
|
||||
* For minnum/maxnum, if one operand is a QNaN, and the other
|
||||
* operand is numerical, then return numerical argument.
|
||||
*/
|
||||
if ((flags & minmax_isnum)
|
||||
&& !(ab_mask & float_cmask_snan)
|
||||
&& (ab_mask & ~float_cmask_qnan)) {
|
||||
return is_nan(a->cls) ? b : a;
|
||||
}
|
||||
return parts_pick_nan(a, b, s);
|
||||
}
|
||||
|
||||
a_exp = a->exp;
|
||||
b_exp = b->exp;
|
||||
|
||||
if (unlikely(ab_mask != float_cmask_normal)) {
|
||||
switch (a->cls) {
|
||||
case float_class_normal:
|
||||
break;
|
||||
case float_class_inf:
|
||||
a_exp = INT16_MAX;
|
||||
break;
|
||||
case float_class_zero:
|
||||
a_exp = INT16_MIN;
|
||||
break;
|
||||
default:
|
||||
g_assert_not_reached();
|
||||
break;
|
||||
}
|
||||
switch (b->cls) {
|
||||
case float_class_normal:
|
||||
break;
|
||||
case float_class_inf:
|
||||
b_exp = INT16_MAX;
|
||||
break;
|
||||
case float_class_zero:
|
||||
b_exp = INT16_MIN;
|
||||
break;
|
||||
default:
|
||||
g_assert_not_reached();
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
/* Compare magnitudes. */
|
||||
cmp = a_exp - b_exp;
|
||||
if (cmp == 0) {
|
||||
cmp = frac_cmp(a, b);
|
||||
}
|
||||
|
||||
/*
|
||||
* Take the sign into account.
|
||||
* For ismag, only do this if the magnitudes are equal.
|
||||
*/
|
||||
if (!(flags & minmax_ismag) || cmp == 0) {
|
||||
if (a->sign != b->sign) {
|
||||
/* For differing signs, the negative operand is less. */
|
||||
cmp = a->sign ? -1 : 1;
|
||||
} else if (a->sign) {
|
||||
/* For two negative operands, invert the magnitude comparison. */
|
||||
cmp = -cmp;
|
||||
}
|
||||
}
|
||||
|
||||
if (flags & minmax_ismin) {
|
||||
cmp = -cmp;
|
||||
}
|
||||
return cmp < 0 ? b : a;
|
||||
}
|
||||
|
198
fpu/softfloat.c
198
fpu/softfloat.c
@ -482,6 +482,15 @@ enum {
|
||||
float_cmask_anynan = float_cmask_qnan | float_cmask_snan,
|
||||
};
|
||||
|
||||
/* Flags for parts_minmax. */
|
||||
enum {
|
||||
/* Set for minimum; clear for maximum. */
|
||||
minmax_ismin = 1,
|
||||
/* Set for the IEEE 754-2008 minNum() and maxNum() operations. */
|
||||
minmax_isnum = 2,
|
||||
/* Set for the IEEE 754-2008 minNumMag() and minNumMag() operations. */
|
||||
minmax_ismag = 4,
|
||||
};
|
||||
|
||||
/* Simple helpers for checking if, or what kind of, NaN we have */
|
||||
static inline __attribute__((unused)) bool is_nan(FloatClass c)
|
||||
@ -865,6 +874,14 @@ static void parts128_uint_to_float(FloatParts128 *p, uint64_t a,
|
||||
#define parts_uint_to_float(P, I, Z, S) \
|
||||
PARTS_GENERIC_64_128(uint_to_float, P)(P, I, Z, S)
|
||||
|
||||
static FloatParts64 *parts64_minmax(FloatParts64 *a, FloatParts64 *b,
|
||||
float_status *s, int flags);
|
||||
static FloatParts128 *parts128_minmax(FloatParts128 *a, FloatParts128 *b,
|
||||
float_status *s, int flags);
|
||||
|
||||
#define parts_minmax(A, B, S, F) \
|
||||
PARTS_GENERIC_64_128(minmax, A)(A, B, S, F)
|
||||
|
||||
/*
|
||||
* Helper functions for softfloat-parts.c.inc, per-size operations.
|
||||
*/
|
||||
@ -3258,145 +3275,74 @@ float128 uint64_to_float128(uint64_t a, float_status *status)
|
||||
return float128_round_pack_canonical(&p, status);
|
||||
}
|
||||
|
||||
/* Float Min/Max */
|
||||
/* min() and max() functions. These can't be implemented as
|
||||
* 'compare and pick one input' because that would mishandle
|
||||
* NaNs and +0 vs -0.
|
||||
*
|
||||
* minnum() and maxnum() functions. These are similar to the min()
|
||||
* and max() functions but if one of the arguments is a QNaN and
|
||||
* the other is numerical then the numerical argument is returned.
|
||||
* SNaNs will get quietened before being returned.
|
||||
* minnum() and maxnum correspond to the IEEE 754-2008 minNum()
|
||||
* and maxNum() operations. min() and max() are the typical min/max
|
||||
* semantics provided by many CPUs which predate that specification.
|
||||
*
|
||||
* minnummag() and maxnummag() functions correspond to minNumMag()
|
||||
* and minNumMag() from the IEEE-754 2008.
|
||||
/*
|
||||
* Minimum and maximum
|
||||
*/
|
||||
static FloatParts64 minmax_floats(FloatParts64 a, FloatParts64 b, bool ismin,
|
||||
bool ieee, bool ismag, float_status *s)
|
||||
|
||||
static float16 float16_minmax(float16 a, float16 b, float_status *s, int flags)
|
||||
{
|
||||
if (unlikely(is_nan(a.cls) || is_nan(b.cls))) {
|
||||
if (ieee) {
|
||||
/* Takes two floating-point values `a' and `b', one of
|
||||
* which is a NaN, and returns the appropriate NaN
|
||||
* result. If either `a' or `b' is a signaling NaN,
|
||||
* the invalid exception is raised.
|
||||
*/
|
||||
if (is_snan(a.cls) || is_snan(b.cls)) {
|
||||
return *parts_pick_nan(&a, &b, s);
|
||||
} else if (is_nan(a.cls) && !is_nan(b.cls)) {
|
||||
return b;
|
||||
} else if (is_nan(b.cls) && !is_nan(a.cls)) {
|
||||
return a;
|
||||
}
|
||||
}
|
||||
return *parts_pick_nan(&a, &b, s);
|
||||
} else {
|
||||
int a_exp, b_exp;
|
||||
FloatParts64 pa, pb, *pr;
|
||||
|
||||
switch (a.cls) {
|
||||
case float_class_normal:
|
||||
a_exp = a.exp;
|
||||
break;
|
||||
case float_class_inf:
|
||||
a_exp = INT_MAX;
|
||||
break;
|
||||
case float_class_zero:
|
||||
a_exp = INT_MIN;
|
||||
break;
|
||||
default:
|
||||
g_assert_not_reached();
|
||||
break;
|
||||
}
|
||||
switch (b.cls) {
|
||||
case float_class_normal:
|
||||
b_exp = b.exp;
|
||||
break;
|
||||
case float_class_inf:
|
||||
b_exp = INT_MAX;
|
||||
break;
|
||||
case float_class_zero:
|
||||
b_exp = INT_MIN;
|
||||
break;
|
||||
default:
|
||||
g_assert_not_reached();
|
||||
break;
|
||||
}
|
||||
float16_unpack_canonical(&pa, a, s);
|
||||
float16_unpack_canonical(&pb, b, s);
|
||||
pr = parts_minmax(&pa, &pb, s, flags);
|
||||
|
||||
if (ismag && (a_exp != b_exp || a.frac != b.frac)) {
|
||||
bool a_less = a_exp < b_exp;
|
||||
if (a_exp == b_exp) {
|
||||
a_less = a.frac < b.frac;
|
||||
}
|
||||
return a_less ^ ismin ? b : a;
|
||||
}
|
||||
|
||||
if (a.sign == b.sign) {
|
||||
bool a_less = a_exp < b_exp;
|
||||
if (a_exp == b_exp) {
|
||||
a_less = a.frac < b.frac;
|
||||
}
|
||||
return a.sign ^ a_less ^ ismin ? b : a;
|
||||
} else {
|
||||
return a.sign ^ ismin ? b : a;
|
||||
}
|
||||
}
|
||||
return float16_round_pack_canonical(pr, s);
|
||||
}
|
||||
|
||||
#define MINMAX(sz, name, ismin, isiee, ismag) \
|
||||
float ## sz float ## sz ## _ ## name(float ## sz a, float ## sz b, \
|
||||
float_status *s) \
|
||||
{ \
|
||||
FloatParts64 pa, pb, pr; \
|
||||
float ## sz ## _unpack_canonical(&pa, a, s); \
|
||||
float ## sz ## _unpack_canonical(&pb, b, s); \
|
||||
pr = minmax_floats(pa, pb, ismin, isiee, ismag, s); \
|
||||
return float ## sz ## _round_pack_canonical(&pr, s); \
|
||||
static bfloat16 bfloat16_minmax(bfloat16 a, bfloat16 b,
|
||||
float_status *s, int flags)
|
||||
{
|
||||
FloatParts64 pa, pb, *pr;
|
||||
|
||||
bfloat16_unpack_canonical(&pa, a, s);
|
||||
bfloat16_unpack_canonical(&pb, b, s);
|
||||
pr = parts_minmax(&pa, &pb, s, flags);
|
||||
|
||||
return bfloat16_round_pack_canonical(pr, s);
|
||||
}
|
||||
|
||||
MINMAX(16, min, true, false, false)
|
||||
MINMAX(16, minnum, true, true, false)
|
||||
MINMAX(16, minnummag, true, true, true)
|
||||
MINMAX(16, max, false, false, false)
|
||||
MINMAX(16, maxnum, false, true, false)
|
||||
MINMAX(16, maxnummag, false, true, true)
|
||||
static float32 float32_minmax(float32 a, float32 b, float_status *s, int flags)
|
||||
{
|
||||
FloatParts64 pa, pb, *pr;
|
||||
|
||||
MINMAX(32, min, true, false, false)
|
||||
MINMAX(32, minnum, true, true, false)
|
||||
MINMAX(32, minnummag, true, true, true)
|
||||
MINMAX(32, max, false, false, false)
|
||||
MINMAX(32, maxnum, false, true, false)
|
||||
MINMAX(32, maxnummag, false, true, true)
|
||||
float32_unpack_canonical(&pa, a, s);
|
||||
float32_unpack_canonical(&pb, b, s);
|
||||
pr = parts_minmax(&pa, &pb, s, flags);
|
||||
|
||||
MINMAX(64, min, true, false, false)
|
||||
MINMAX(64, minnum, true, true, false)
|
||||
MINMAX(64, minnummag, true, true, true)
|
||||
MINMAX(64, max, false, false, false)
|
||||
MINMAX(64, maxnum, false, true, false)
|
||||
MINMAX(64, maxnummag, false, true, true)
|
||||
|
||||
#undef MINMAX
|
||||
|
||||
#define BF16_MINMAX(name, ismin, isiee, ismag) \
|
||||
bfloat16 bfloat16_ ## name(bfloat16 a, bfloat16 b, float_status *s) \
|
||||
{ \
|
||||
FloatParts64 pa, pb, pr; \
|
||||
bfloat16_unpack_canonical(&pa, a, s); \
|
||||
bfloat16_unpack_canonical(&pb, b, s); \
|
||||
pr = minmax_floats(pa, pb, ismin, isiee, ismag, s); \
|
||||
return bfloat16_round_pack_canonical(&pr, s); \
|
||||
return float32_round_pack_canonical(pr, s);
|
||||
}
|
||||
|
||||
BF16_MINMAX(min, true, false, false)
|
||||
BF16_MINMAX(minnum, true, true, false)
|
||||
BF16_MINMAX(minnummag, true, true, true)
|
||||
BF16_MINMAX(max, false, false, false)
|
||||
BF16_MINMAX(maxnum, false, true, false)
|
||||
BF16_MINMAX(maxnummag, false, true, true)
|
||||
static float64 float64_minmax(float64 a, float64 b, float_status *s, int flags)
|
||||
{
|
||||
FloatParts64 pa, pb, *pr;
|
||||
|
||||
#undef BF16_MINMAX
|
||||
float64_unpack_canonical(&pa, a, s);
|
||||
float64_unpack_canonical(&pb, b, s);
|
||||
pr = parts_minmax(&pa, &pb, s, flags);
|
||||
|
||||
return float64_round_pack_canonical(pr, s);
|
||||
}
|
||||
|
||||
#define MINMAX_1(type, name, flags) \
|
||||
type type##_##name(type a, type b, float_status *s) \
|
||||
{ return type##_minmax(a, b, s, flags); }
|
||||
|
||||
#define MINMAX_2(type) \
|
||||
MINMAX_1(type, max, 0) \
|
||||
MINMAX_1(type, maxnum, minmax_isnum) \
|
||||
MINMAX_1(type, maxnummag, minmax_isnum | minmax_ismag) \
|
||||
MINMAX_1(type, min, minmax_ismin) \
|
||||
MINMAX_1(type, minnum, minmax_ismin | minmax_isnum) \
|
||||
MINMAX_1(type, minnummag, minmax_ismin | minmax_isnum | minmax_ismag)
|
||||
|
||||
MINMAX_2(float16)
|
||||
MINMAX_2(bfloat16)
|
||||
MINMAX_2(float32)
|
||||
MINMAX_2(float64)
|
||||
|
||||
#undef MINMAX_1
|
||||
#undef MINMAX_2
|
||||
|
||||
/* Floating point compare */
|
||||
static FloatRelation compare_floats(FloatParts64 a, FloatParts64 b, bool is_quiet,
|
||||
|
Loading…
Reference in New Issue
Block a user