[08/13] Use GCC builtins for rint functions if desired.
Commit Message
This patch is using the corresponding GCC builtin for rintf, rint,
rintl and rintf128 if the USE_FUNCTION_BUILTIN macros are defined to one
in math-use-builtins.h.
This is the case for s390 if build with at least --march=z196 --mzarch.
Otherwise the generic implementation is used. The code of the generic
implementation is not changed except changes in code style.
---
sysdeps/generic/math-use-builtins.h | 5 ++
sysdeps/ieee754/dbl-64/s_rint.c | 11 +++-
sysdeps/ieee754/float128/float128_private.h | 3 +
sysdeps/ieee754/flt-32/s_rintf.c | 57 ++++++++++++-------
sysdeps/ieee754/ldbl-128/s_rintl.c | 63 +++++++++++++--------
sysdeps/s390/fpu/math-use-builtins.h | 6 ++
6 files changed, 97 insertions(+), 48 deletions(-)
Comments
On 02/12/2019 11:46, Stefan Liebler wrote:
> This patch is using the corresponding GCC builtin for rintf, rint,
> rintl and rintf128 if the USE_FUNCTION_BUILTIN macros are defined to one
> in math-use-builtins.h.
>
> This is the case for s390 if build with at least --march=z196 --mzarch.
> Otherwise the generic implementation is used. The code of the generic
> implementation is not changed except changes in code style.
LGTM with some changes below.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
> ---
> sysdeps/generic/math-use-builtins.h | 5 ++
> sysdeps/ieee754/dbl-64/s_rint.c | 11 +++-
> sysdeps/ieee754/float128/float128_private.h | 3 +
> sysdeps/ieee754/flt-32/s_rintf.c | 57 ++++++++++++-------
> sysdeps/ieee754/ldbl-128/s_rintl.c | 63 +++++++++++++--------
> sysdeps/s390/fpu/math-use-builtins.h | 6 ++
> 6 files changed, 97 insertions(+), 48 deletions(-)
>
> diff --git a/sysdeps/generic/math-use-builtins.h b/sysdeps/generic/math-use-builtins.h
> index e12490ed41..64b4a4bb5b 100644
> --- a/sysdeps/generic/math-use-builtins.h
> +++ b/sysdeps/generic/math-use-builtins.h
> @@ -26,4 +26,9 @@
> #define USE_NEARBYINTL_BUILTIN 0
> #define USE_NEARBYINTF128_BUILTIN 0
>
> +#define USE_RINT_BUILTIN 0
> +#define USE_RINTF_BUILTIN 0
> +#define USE_RINTL_BUILTIN 0
> +#define USE_RINTF128_BUILTIN 0
> +
> #endif /* math-use-builtins.h */
> diff --git a/sysdeps/ieee754/dbl-64/s_rint.c b/sysdeps/ieee754/dbl-64/s_rint.c
> index f96078c405..b09ed8fc06 100644
> --- a/sysdeps/ieee754/dbl-64/s_rint.c
> +++ b/sysdeps/ieee754/dbl-64/s_rint.c
> @@ -23,16 +23,22 @@
> #include <math.h>
> #include <math_private.h>
> #include <libm-alias-double.h>
> +#include <math-use-builtins.h>
>
> +#if ! USE_RINT_BUILTIN
> static const double
> TWO52[2] = {
> 4.50359962737049600000e+15, /* 0x43300000, 0x00000000 */
> -4.50359962737049600000e+15, /* 0xC3300000, 0x00000000 */
> };
> +#endif
You could just move it inside the !USE_NEARBYINT_BUILTIN within the
function.
>
> double
> __rint (double x)
> {
> +#if USE_RINT_BUILTIN
> + return __builtin_rint (x);
> +#else
> int64_t i0, sx;
> int32_t j0;
> EXTRACT_WORDS64 (i0, x);
> @@ -47,7 +53,7 @@ __rint (double x)
> EXTRACT_WORDS64 (i0, t);
> INSERT_WORDS64 (t, (i0 & UINT64_C (0x7fffffffffffffff))
> | (sx << 63));
> - return t;
> + return t;
> }
> }
> else
> @@ -56,9 +62,10 @@ __rint (double x)
> return x + x; /* inf or NaN */
> else
> return x; /* x is integral */
> - }
> + }
> double w = TWO52[sx] + x;
> return w - TWO52[sx];
> +#endif /* USE_RINT_BUILTIN */
> }
> #ifndef __rint
> libm_alias_double (__rint, rint)
Ok.
> diff --git a/sysdeps/ieee754/float128/float128_private.h b/sysdeps/ieee754/float128/float128_private.h
> index e96986a968..f458e7b85f 100644
> --- a/sysdeps/ieee754/float128/float128_private.h
> +++ b/sysdeps/ieee754/float128/float128_private.h
> @@ -142,6 +142,8 @@
> #include <math-use-builtins.h>
> #undef USE_NEARBYINTL_BUILTIN
> #define USE_NEARBYINTL_BUILTIN USE_NEARBYINTF128_BUILTIN
> +#undef USE_RINTL_BUILTIN
> +#define USE_RINTL_BUILTIN USE_RINTF128_BUILTIN
>
> /* IEEE function renames. */
> #define __ieee754_acoshl __ieee754_acoshf128
> @@ -346,6 +348,7 @@
> #define __builtin_copysignl __builtin_copysignf128
> #define __builtin_signbitl __builtin_signbit
> #define __builtin_nearbyintl __builtin_nearbyintf128
> +#define __builtin_rintl __builtin_rintf128
>
> /* Get the constant suffix from bits/floatn-compat.h. */
> #define L(x) __f128 (x)
Ok.
> diff --git a/sysdeps/ieee754/flt-32/s_rintf.c b/sysdeps/ieee754/flt-32/s_rintf.c
> index 0306dc21f4..7dbf991395 100644
> --- a/sysdeps/ieee754/flt-32/s_rintf.c
> +++ b/sysdeps/ieee754/flt-32/s_rintf.c
> @@ -17,35 +17,48 @@
> #include <math.h>
> #include <math_private.h>
> #include <libm-alias-float.h>
> +#include <math-use-builtins.h>
>
> +#if ! USE_RINTF_BUILTIN
> static const float
> -TWO23[2]={
> - 8.3886080000e+06, /* 0x4b000000 */
> - -8.3886080000e+06, /* 0xcb000000 */
> +TWO23[2] = {
> + 8.3886080000e+06, /* 0x4b000000 */
> + -8.3886080000e+06, /* 0xcb000000 */
> };
> +#endif
>
> float
> -__rintf(float x)
> +__rintf (float x)
> {
> - int32_t i0,j0,sx;
> - float w,t;
> - GET_FLOAT_WORD(i0,x);
> - sx = (i0>>31)&1;
> - j0 = ((i0>>23)&0xff)-0x7f;
> - if(j0<23) {
> - if(j0<0) {
> - w = TWO23[sx]+x;
> - t = w-TWO23[sx];
> - GET_FLOAT_WORD(i0,t);
> - SET_FLOAT_WORD(t,(i0&0x7fffffff)|(sx<<31));
> - return t;
> - }
> - } else {
> - if(j0==0x80) return x+x; /* inf or NaN */
> - else return x; /* x is integral */
> +#if USE_RINTF_BUILTIN
> + return __builtin_rintf (x);
> +#else
> + int32_t i0, j0, sx;
> + float w, t;
> + GET_FLOAT_WORD (i0, x);
> + sx = (i0 >> 31) & 1;
> + j0 = ((i0 >> 23) & 0xff) - 0x7f;
> + if (j0 < 23)
> + {
> + if(j0 < 0)
> + {
> + w = TWO23[sx] + x;
> + t = w - TWO23[sx];
> + GET_FLOAT_WORD (i0, t);
> + SET_FLOAT_WORD (t, (i0 & 0x7fffffff) | (sx << 31));
> + return t;
> }
> - w = TWO23[sx]+x;
> - return w-TWO23[sx];
> + }
> + else
> + {
> + if (j0 == 0x80)
> + return x + x; /* inf or NaN */
> + else
> + return x; /* x is integral */
> + }
> + w = TWO23[sx] + x;
> + return w - TWO23[sx];
> +#endif /* USE_RINTF_BUILTIN */
> }
> #ifndef __rintf
> libm_alias_float (__rint, rint)
Ok, but fix the indentation in a separated patch.
> diff --git a/sysdeps/ieee754/ldbl-128/s_rintl.c b/sysdeps/ieee754/ldbl-128/s_rintl.c
> index b6337e1d8a..1c4eba566d 100644
> --- a/sysdeps/ieee754/ldbl-128/s_rintl.c
> +++ b/sysdeps/ieee754/ldbl-128/s_rintl.c
> @@ -13,7 +13,9 @@
> * ====================================================
> */
>
> -#if defined(LIBM_SCCS) && !defined(lint)
> +#include <math-use-builtins.h>
> +
> +#if ! USE_RINTL_BUILTIN && defined (LIBM_SCCS) && ! defined (lint)
> static char rcsid[] = "$NetBSD: $";
> #endif
>
> @@ -32,33 +34,46 @@ static char rcsid[] = "$NetBSD: $";
> #include <math_private.h>
> #include <libm-alias-ldouble.h>
>
> +#if ! USE_RINTL_BUILTIN
> static const _Float128
> -TWO112[2]={
> - 5.19229685853482762853049632922009600E+33L, /* 0x406F000000000000, 0 */
> - -5.19229685853482762853049632922009600E+33L /* 0xC06F000000000000, 0 */
> +TWO112[2] = {
> + 5.19229685853482762853049632922009600E+33L, /* 0x406F000000000000, 0 */
> + -5.19229685853482762853049632922009600E+33L /* 0xC06F000000000000, 0 */
> };
> +#endif
You could just move it inside the !USE_NEARBYINT_BUILTIN within the
function.
>
> -_Float128 __rintl(_Float128 x)
> +_Float128
> +__rintl (_Float128 x)
> {
> - int64_t i0,j0,sx;
> - uint64_t i1 __attribute__ ((unused));
> - _Float128 w,t;
> - GET_LDOUBLE_WORDS64(i0,i1,x);
> - sx = (((uint64_t)i0)>>63);
> - j0 = ((i0>>48)&0x7fff)-0x3fff;
> - if(j0<112) {
> - if(j0<0) {
> - w = TWO112[sx]+x;
> - t = w-TWO112[sx];
> - GET_LDOUBLE_MSW64(i0,t);
> - SET_LDOUBLE_MSW64(t,(i0&0x7fffffffffffffffLL)|(sx<<63));
> - return t;
> - }
> - } else {
> - if(j0==0x4000) return x+x; /* inf or NaN */
> - else return x; /* x is integral */
> +#if USE_RINTL_BUILTIN
> + return __builtin_rintl (x);
> +#else
> + int64_t i0, j0, sx;
> + uint64_t i1 __attribute__ ((unused));
> + _Float128 w, t;
> + GET_LDOUBLE_WORDS64 (i0, i1, x);
> + sx = (((uint64_t) i0) >> 63);
> + j0 = ((i0 >> 48) & 0x7fff) - 0x3fff;
> + if (j0 < 112)
> + {
> + if (j0 < 0)
> + {
> + w = TWO112[sx] + x;
> + t = w - TWO112[sx];
> + GET_LDOUBLE_MSW64 (i0, t);
> + SET_LDOUBLE_MSW64 (t, (i0 & 0x7fffffffffffffffLL) | (sx << 63));
> + return t;
> }
> - w = TWO112[sx]+x;
> - return w-TWO112[sx];
> + }
> + else
> + {
> + if (j0 == 0x4000)
> + return x + x; /* inf or NaN */
> + else
> + return x; /* x is integral */
> + }
> + w = TWO112[sx] + x;
> + return w - TWO112[sx];
> +#endif /* USE_RINTL_BUILTIN */
> }
> libm_alias_ldouble (__rint, rint)
Ok, but fix the indentation in a separated patch.
> diff --git a/sysdeps/s390/fpu/math-use-builtins.h b/sysdeps/s390/fpu/math-use-builtins.h
> index fd9da8893e..15705fa74a 100644
> --- a/sysdeps/s390/fpu/math-use-builtins.h
> +++ b/sysdeps/s390/fpu/math-use-builtins.h
> @@ -30,10 +30,16 @@
> # define USE_NEARBYINTF_BUILTIN 1
> # define USE_NEARBYINTL_BUILTIN 1
>
> +# define USE_RINT_BUILTIN 1
> +# define USE_RINTF_BUILTIN 1
> +# define USE_RINTL_BUILTIN 1
> +
> # if __GNUC_PREREQ (8, 1)
> # define USE_NEARBYINTF128_BUILTIN 1
> +# define USE_RINTF128_BUILTIN 1
> # else
> # define USE_NEARBYINTF128_BUILTIN 0
> +# define USE_RINTF128_BUILTIN 0
> # endif
>
> #else
>
Ok.
@@ -26,4 +26,9 @@
#define USE_NEARBYINTL_BUILTIN 0
#define USE_NEARBYINTF128_BUILTIN 0
+#define USE_RINT_BUILTIN 0
+#define USE_RINTF_BUILTIN 0
+#define USE_RINTL_BUILTIN 0
+#define USE_RINTF128_BUILTIN 0
+
#endif /* math-use-builtins.h */
@@ -23,16 +23,22 @@
#include <math.h>
#include <math_private.h>
#include <libm-alias-double.h>
+#include <math-use-builtins.h>
+#if ! USE_RINT_BUILTIN
static const double
TWO52[2] = {
4.50359962737049600000e+15, /* 0x43300000, 0x00000000 */
-4.50359962737049600000e+15, /* 0xC3300000, 0x00000000 */
};
+#endif
double
__rint (double x)
{
+#if USE_RINT_BUILTIN
+ return __builtin_rint (x);
+#else
int64_t i0, sx;
int32_t j0;
EXTRACT_WORDS64 (i0, x);
@@ -47,7 +53,7 @@ __rint (double x)
EXTRACT_WORDS64 (i0, t);
INSERT_WORDS64 (t, (i0 & UINT64_C (0x7fffffffffffffff))
| (sx << 63));
- return t;
+ return t;
}
}
else
@@ -56,9 +62,10 @@ __rint (double x)
return x + x; /* inf or NaN */
else
return x; /* x is integral */
- }
+ }
double w = TWO52[sx] + x;
return w - TWO52[sx];
+#endif /* USE_RINT_BUILTIN */
}
#ifndef __rint
libm_alias_double (__rint, rint)
@@ -142,6 +142,8 @@
#include <math-use-builtins.h>
#undef USE_NEARBYINTL_BUILTIN
#define USE_NEARBYINTL_BUILTIN USE_NEARBYINTF128_BUILTIN
+#undef USE_RINTL_BUILTIN
+#define USE_RINTL_BUILTIN USE_RINTF128_BUILTIN
/* IEEE function renames. */
#define __ieee754_acoshl __ieee754_acoshf128
@@ -346,6 +348,7 @@
#define __builtin_copysignl __builtin_copysignf128
#define __builtin_signbitl __builtin_signbit
#define __builtin_nearbyintl __builtin_nearbyintf128
+#define __builtin_rintl __builtin_rintf128
/* Get the constant suffix from bits/floatn-compat.h. */
#define L(x) __f128 (x)
@@ -17,35 +17,48 @@
#include <math.h>
#include <math_private.h>
#include <libm-alias-float.h>
+#include <math-use-builtins.h>
+#if ! USE_RINTF_BUILTIN
static const float
-TWO23[2]={
- 8.3886080000e+06, /* 0x4b000000 */
- -8.3886080000e+06, /* 0xcb000000 */
+TWO23[2] = {
+ 8.3886080000e+06, /* 0x4b000000 */
+ -8.3886080000e+06, /* 0xcb000000 */
};
+#endif
float
-__rintf(float x)
+__rintf (float x)
{
- int32_t i0,j0,sx;
- float w,t;
- GET_FLOAT_WORD(i0,x);
- sx = (i0>>31)&1;
- j0 = ((i0>>23)&0xff)-0x7f;
- if(j0<23) {
- if(j0<0) {
- w = TWO23[sx]+x;
- t = w-TWO23[sx];
- GET_FLOAT_WORD(i0,t);
- SET_FLOAT_WORD(t,(i0&0x7fffffff)|(sx<<31));
- return t;
- }
- } else {
- if(j0==0x80) return x+x; /* inf or NaN */
- else return x; /* x is integral */
+#if USE_RINTF_BUILTIN
+ return __builtin_rintf (x);
+#else
+ int32_t i0, j0, sx;
+ float w, t;
+ GET_FLOAT_WORD (i0, x);
+ sx = (i0 >> 31) & 1;
+ j0 = ((i0 >> 23) & 0xff) - 0x7f;
+ if (j0 < 23)
+ {
+ if(j0 < 0)
+ {
+ w = TWO23[sx] + x;
+ t = w - TWO23[sx];
+ GET_FLOAT_WORD (i0, t);
+ SET_FLOAT_WORD (t, (i0 & 0x7fffffff) | (sx << 31));
+ return t;
}
- w = TWO23[sx]+x;
- return w-TWO23[sx];
+ }
+ else
+ {
+ if (j0 == 0x80)
+ return x + x; /* inf or NaN */
+ else
+ return x; /* x is integral */
+ }
+ w = TWO23[sx] + x;
+ return w - TWO23[sx];
+#endif /* USE_RINTF_BUILTIN */
}
#ifndef __rintf
libm_alias_float (__rint, rint)
@@ -13,7 +13,9 @@
* ====================================================
*/
-#if defined(LIBM_SCCS) && !defined(lint)
+#include <math-use-builtins.h>
+
+#if ! USE_RINTL_BUILTIN && defined (LIBM_SCCS) && ! defined (lint)
static char rcsid[] = "$NetBSD: $";
#endif
@@ -32,33 +34,46 @@ static char rcsid[] = "$NetBSD: $";
#include <math_private.h>
#include <libm-alias-ldouble.h>
+#if ! USE_RINTL_BUILTIN
static const _Float128
-TWO112[2]={
- 5.19229685853482762853049632922009600E+33L, /* 0x406F000000000000, 0 */
- -5.19229685853482762853049632922009600E+33L /* 0xC06F000000000000, 0 */
+TWO112[2] = {
+ 5.19229685853482762853049632922009600E+33L, /* 0x406F000000000000, 0 */
+ -5.19229685853482762853049632922009600E+33L /* 0xC06F000000000000, 0 */
};
+#endif
-_Float128 __rintl(_Float128 x)
+_Float128
+__rintl (_Float128 x)
{
- int64_t i0,j0,sx;
- uint64_t i1 __attribute__ ((unused));
- _Float128 w,t;
- GET_LDOUBLE_WORDS64(i0,i1,x);
- sx = (((uint64_t)i0)>>63);
- j0 = ((i0>>48)&0x7fff)-0x3fff;
- if(j0<112) {
- if(j0<0) {
- w = TWO112[sx]+x;
- t = w-TWO112[sx];
- GET_LDOUBLE_MSW64(i0,t);
- SET_LDOUBLE_MSW64(t,(i0&0x7fffffffffffffffLL)|(sx<<63));
- return t;
- }
- } else {
- if(j0==0x4000) return x+x; /* inf or NaN */
- else return x; /* x is integral */
+#if USE_RINTL_BUILTIN
+ return __builtin_rintl (x);
+#else
+ int64_t i0, j0, sx;
+ uint64_t i1 __attribute__ ((unused));
+ _Float128 w, t;
+ GET_LDOUBLE_WORDS64 (i0, i1, x);
+ sx = (((uint64_t) i0) >> 63);
+ j0 = ((i0 >> 48) & 0x7fff) - 0x3fff;
+ if (j0 < 112)
+ {
+ if (j0 < 0)
+ {
+ w = TWO112[sx] + x;
+ t = w - TWO112[sx];
+ GET_LDOUBLE_MSW64 (i0, t);
+ SET_LDOUBLE_MSW64 (t, (i0 & 0x7fffffffffffffffLL) | (sx << 63));
+ return t;
}
- w = TWO112[sx]+x;
- return w-TWO112[sx];
+ }
+ else
+ {
+ if (j0 == 0x4000)
+ return x + x; /* inf or NaN */
+ else
+ return x; /* x is integral */
+ }
+ w = TWO112[sx] + x;
+ return w - TWO112[sx];
+#endif /* USE_RINTL_BUILTIN */
}
libm_alias_ldouble (__rint, rint)
@@ -30,10 +30,16 @@
# define USE_NEARBYINTF_BUILTIN 1
# define USE_NEARBYINTL_BUILTIN 1
+# define USE_RINT_BUILTIN 1
+# define USE_RINTF_BUILTIN 1
+# define USE_RINTL_BUILTIN 1
+
# if __GNUC_PREREQ (8, 1)
# define USE_NEARBYINTF128_BUILTIN 1
+# define USE_RINTF128_BUILTIN 1
# else
# define USE_NEARBYINTF128_BUILTIN 0
+# define USE_RINTF128_BUILTIN 0
# endif
#else