* [PATCH v6 1/3] math: redirect roundeven function
@ 2021-05-22 0:22 Shen-Ta Hsieh
2021-05-22 0:22 ` [PATCH v6 2/3] x86_64: roundeven with sse4.1 support Shen-Ta Hsieh
` (2 more replies)
0 siblings, 3 replies; 12+ messages in thread
From: Shen-Ta Hsieh @ 2021-05-22 0:22 UTC (permalink / raw)
To: libc-alpha; +Cc: H . J . Lu, Shen-Ta Hsieh
This patch redirect roundeven function for futhermore changes.
Signed-off-by: Shen-Ta Hsieh <ibmibmibm.tw@gmail.com>
---
include/math.h | 2 +-
sysdeps/ieee754/dbl-64/s_roundeven.c | 4 +++-
sysdeps/ieee754/float128/s_roundevenf128.c | 1 +
sysdeps/ieee754/flt-32/s_roundevenf.c | 3 +++
sysdeps/ieee754/ldbl-128/s_roundevenl.c | 1 +
sysdeps/ieee754/ldbl-96/s_roundevenl.c | 1 +
6 files changed, 10 insertions(+), 2 deletions(-)
diff --git a/include/math.h b/include/math.h
index e1c2a4eb64..b4772d3d3b 100644
--- a/include/math.h
+++ b/include/math.h
@@ -38,7 +38,6 @@ libm_hidden_proto (__issignaling)
libm_hidden_proto (__issignalingf)
libm_hidden_proto (__exp)
libm_hidden_proto (__expf)
-libm_hidden_proto (__roundeven)
# if !defined __NO_LONG_DOUBLE_MATH \
&& __LDOUBLE_REDIRECTS_TO_FLOAT128_ABI == 0
@@ -159,6 +158,7 @@ fabsf128 (_Float128 x)
MATH_REDIRECT (sqrt, "__ieee754_", MATH_REDIRECT_UNARY_ARGS)
MATH_REDIRECT (ceil, "__", MATH_REDIRECT_UNARY_ARGS)
MATH_REDIRECT (floor, "__", MATH_REDIRECT_UNARY_ARGS)
+MATH_REDIRECT (roundeven, "__", MATH_REDIRECT_UNARY_ARGS)
MATH_REDIRECT (rint, "__", MATH_REDIRECT_UNARY_ARGS)
MATH_REDIRECT (trunc, "__", MATH_REDIRECT_UNARY_ARGS)
MATH_REDIRECT (round, "__", MATH_REDIRECT_UNARY_ARGS)
diff --git a/sysdeps/ieee754/dbl-64/s_roundeven.c b/sysdeps/ieee754/dbl-64/s_roundeven.c
index 943b2c634c..5deff3bb8a 100644
--- a/sysdeps/ieee754/dbl-64/s_roundeven.c
+++ b/sysdeps/ieee754/dbl-64/s_roundeven.c
@@ -16,6 +16,7 @@
License along with the GNU C Library; if not, see
<https://www.gnu.org/licenses/>. */
+#define NO_MATH_REDIRECT
#include <math.h>
#include <math_private.h>
#include <libm-alias-double.h>
@@ -66,5 +67,6 @@ __roundeven (double x)
INSERT_WORDS64 (x, ix);
return x;
}
-hidden_def (__roundeven)
+#ifndef __roundeven
libm_alias_double (__roundeven, roundeven)
+#endif
diff --git a/sysdeps/ieee754/float128/s_roundevenf128.c b/sysdeps/ieee754/float128/s_roundevenf128.c
index 5a9b3f395f..e0faf727f6 100644
--- a/sysdeps/ieee754/float128/s_roundevenf128.c
+++ b/sysdeps/ieee754/float128/s_roundevenf128.c
@@ -1,2 +1,3 @@
+#define NO_MATH_REDIRECT
#include <float128_private.h>
#include "../ldbl-128/s_roundevenl.c"
diff --git a/sysdeps/ieee754/flt-32/s_roundevenf.c b/sysdeps/ieee754/flt-32/s_roundevenf.c
index ae54365bbe..2cc0d7b001 100644
--- a/sysdeps/ieee754/flt-32/s_roundevenf.c
+++ b/sysdeps/ieee754/flt-32/s_roundevenf.c
@@ -17,6 +17,7 @@
License along with the GNU C Library; if not, see
<https://www.gnu.org/licenses/>. */
+#define NO_MATH_REDIRECT
#include <math.h>
#include <math_private.h>
#include <libm-alias-float.h>
@@ -67,4 +68,6 @@ __roundevenf (float x)
SET_FLOAT_WORD (x, ix);
return x;
}
+#ifndef __roundevenf
libm_alias_float (__roundeven, roundeven)
+#endif
diff --git a/sysdeps/ieee754/ldbl-128/s_roundevenl.c b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
index 02765be0e0..7c9c11fad2 100644
--- a/sysdeps/ieee754/ldbl-128/s_roundevenl.c
+++ b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
@@ -17,6 +17,7 @@
License along with the GNU C Library; if not, see
<https://www.gnu.org/licenses/>. */
+#define NO_MATH_REDIRECT
#include <math.h>
#include <math_private.h>
#include <libm-alias-ldouble.h>
diff --git a/sysdeps/ieee754/ldbl-96/s_roundevenl.c b/sysdeps/ieee754/ldbl-96/s_roundevenl.c
index 8340116ed8..f826829c31 100644
--- a/sysdeps/ieee754/ldbl-96/s_roundevenl.c
+++ b/sysdeps/ieee754/ldbl-96/s_roundevenl.c
@@ -17,6 +17,7 @@
License along with the GNU C Library; if not, see
<https://www.gnu.org/licenses/>. */
+#define NO_MATH_REDIRECT
#include <math.h>
#include <math_private.h>
#include <libm-alias-ldouble.h>
--
2.31.1
^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH v6 2/3] x86_64: roundeven with sse4.1 support
2021-05-22 0:22 [PATCH v6 1/3] math: redirect roundeven function Shen-Ta Hsieh
@ 2021-05-22 0:22 ` Shen-Ta Hsieh
2021-05-22 4:52 ` Paul Zimmermann
` (2 more replies)
2021-05-22 0:22 ` [PATCH v6 3/3] Use GCC builtins for roundeven functions if desired Shen-Ta Hsieh
2021-05-23 18:54 ` [PATCH v6 1/3] math: redirect roundeven function H.J. Lu
2 siblings, 3 replies; 12+ messages in thread
From: Shen-Ta Hsieh @ 2021-05-22 0:22 UTC (permalink / raw)
To: libc-alpha; +Cc: H . J . Lu, Shen-Ta Hsieh
This patch adds support for the sse4.1 hardware floating point
roundeven.
Here is a benchmark result on my AMD Ryzen 9 3900X system:
* benchmark result before this commit
| | roundeven | roundevenf |
|------------|--------------|--------------|
| duration | 3.77659e+09 | 3.77504e+09 |
| iterations | 3.97043e+08 | 4.36752e+08 |
| max | 83.714 | 58.861 |
| min | 7.144 | 6.27 |
| mean | 9.51179 | 8.64345 |
* benchmark result after this commit
| | roundeven | roundevenf |
|------------|--------------|--------------|
| duration | 3.76913e+09 | 3.76923e+09 |
| iterations | 5.55921e+08 | 5.64822e+08 |
| max | 211.698 | 439.09 |
| min | 6.498 | 6.422 |
| mean | 6.77998 | 6.6733 |
Signed-off-by: Shen-Ta Hsieh <ibmibmibm.tw@gmail.com>
---
sysdeps/x86_64/fpu/multiarch/Makefile | 5 +--
sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c | 2 ++
.../x86_64/fpu/multiarch/s_roundeven-sse4_1.S | 24 ++++++++++++++
sysdeps/x86_64/fpu/multiarch/s_roundeven.c | 31 +++++++++++++++++++
sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c | 3 ++
.../fpu/multiarch/s_roundevenf-sse4_1.S | 24 ++++++++++++++
sysdeps/x86_64/fpu/multiarch/s_roundevenf.c | 31 +++++++++++++++++++
7 files changed, 118 insertions(+), 2 deletions(-)
create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S
create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundeven.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c
create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S
create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundevenf.c
diff --git a/sysdeps/x86_64/fpu/multiarch/Makefile b/sysdeps/x86_64/fpu/multiarch/Makefile
index 57892c56bb..d425ffd6d3 100644
--- a/sysdeps/x86_64/fpu/multiarch/Makefile
+++ b/sysdeps/x86_64/fpu/multiarch/Makefile
@@ -1,11 +1,12 @@
ifeq ($(subdir),math)
libm-sysdep_routines += s_floor-c s_ceil-c s_floorf-c s_ceilf-c \
s_rint-c s_rintf-c s_nearbyint-c s_nearbyintf-c \
- s_trunc-c s_truncf-c
+ s_roundeven-c s_roundevenf-c s_trunc-c s_truncf-c
libm-sysdep_routines += s_ceil-sse4_1 s_ceilf-sse4_1 s_floor-sse4_1 \
s_floorf-sse4_1 s_nearbyint-sse4_1 \
- s_nearbyintf-sse4_1 s_rint-sse4_1 s_rintf-sse4_1 \
+ s_nearbyintf-sse4_1 s_roundeven-sse4_1 \
+ s_roundevenf-sse4_1 s_rint-sse4_1 s_rintf-sse4_1 \
s_trunc-sse4_1 s_truncf-sse4_1
libm-sysdep_routines += e_exp-fma e_log-fma e_pow-fma s_atan-fma \
diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c b/sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c
new file mode 100644
index 0000000000..c7be43cb22
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c
@@ -0,0 +1,2 @@
+#define __roundeven __roundeven_c
+#include <sysdeps/ieee754/dbl-64/s_roundeven.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S b/sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S
new file mode 100644
index 0000000000..6ae8f6b1d3
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S
@@ -0,0 +1,24 @@
+/* Copyright (C) 2021 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <https://www.gnu.org/licenses/>. */
+
+#include <sysdep.h>
+
+ .section .text.sse4.1,"ax",@progbits
+ENTRY(__roundeven_sse41)
+ roundsd $8, %xmm0, %xmm0
+ ret
+END(__roundeven_sse41)
diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundeven.c b/sysdeps/x86_64/fpu/multiarch/s_roundeven.c
new file mode 100644
index 0000000000..d92eda652a
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_roundeven.c
@@ -0,0 +1,31 @@
+/* Multiple versions of __roundeven.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <https://www.gnu.org/licenses/>. */
+
+#include <libm-alias-double.h>
+
+#define roundeven __redirect_roundeven
+#define __roundeven __redirect___roundeven
+#include <math.h>
+#undef roundeven
+#undef __roundeven
+
+#define SYMBOL_NAME roundeven
+#include "ifunc-sse4_1.h"
+
+libc_ifunc_redirected (__redirect_roundeven, __roundeven, IFUNC_SELECTOR ());
+libm_alias_double (__roundeven, roundeven)
diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c
new file mode 100644
index 0000000000..72a6e7d1fb
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c
@@ -0,0 +1,3 @@
+#undef __roundevenf
+#define __roundevenf __roundevenf_c
+#include <sysdeps/ieee754/flt-32/s_roundevenf.c>
diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S
new file mode 100644
index 0000000000..a76e10807e
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S
@@ -0,0 +1,24 @@
+/* Copyright (C) 2021 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <https://www.gnu.org/licenses/>. */
+
+#include <sysdep.h>
+
+ .section .text.sse4.1,"ax",@progbits
+ENTRY(__roundevenf_sse41)
+ roundss $8, %xmm0, %xmm0
+ ret
+END(__roundevenf_sse41)
diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundevenf.c b/sysdeps/x86_64/fpu/multiarch/s_roundevenf.c
new file mode 100644
index 0000000000..2ee196e68f
--- /dev/null
+++ b/sysdeps/x86_64/fpu/multiarch/s_roundevenf.c
@@ -0,0 +1,31 @@
+/* Multiple versions of __roundevenf.
+ Copyright (C) 2021 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <https://www.gnu.org/licenses/>. */
+
+#include <libm-alias-float.h>
+
+#define roundevenf __redirect_roundevenf
+#define __roundevenf __redirect___roundevenf
+#include <math.h>
+#undef roundevenf
+#undef __roundevenf
+
+#define SYMBOL_NAME roundevenf
+#include "ifunc-sse4_1.h"
+
+libc_ifunc_redirected (__redirect_roundevenf, __roundevenf, IFUNC_SELECTOR ());
+libm_alias_float (__roundeven, roundeven)
--
2.31.1
^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH v6 3/3] Use GCC builtins for roundeven functions if desired.
2021-05-22 0:22 [PATCH v6 1/3] math: redirect roundeven function Shen-Ta Hsieh
2021-05-22 0:22 ` [PATCH v6 2/3] x86_64: roundeven with sse4.1 support Shen-Ta Hsieh
@ 2021-05-22 0:22 ` Shen-Ta Hsieh
2021-05-23 19:03 ` H.J. Lu
2021-05-23 18:54 ` [PATCH v6 1/3] math: redirect roundeven function H.J. Lu
2 siblings, 1 reply; 12+ messages in thread
From: Shen-Ta Hsieh @ 2021-05-22 0:22 UTC (permalink / raw)
To: libc-alpha; +Cc: H . J . Lu, Shen-Ta Hsieh
This patch is using the corresponding GCC builtin for roundevenf,
roundeven and roundevenl if the USE_FUNCTION_BUILTIN macros are defined
to one in math-use-builtins.h.
These builtin functions is supported since GCC 10.
The code of the generic implementation is not changed.
Signed-off-by: Shen-Ta Hsieh <ibmibmibm.tw@gmail.com>
---
sysdeps/generic/math-use-builtins-roundeven.h | 4 ++++
sysdeps/generic/math-use-builtins.h | 1 +
sysdeps/ieee754/dbl-64/s_roundeven.c | 5 +++++
sysdeps/ieee754/flt-32/s_roundevenf.c | 5 +++++
sysdeps/ieee754/ldbl-128/s_roundevenl.c | 5 +++++
5 files changed, 20 insertions(+)
create mode 100644 sysdeps/generic/math-use-builtins-roundeven.h
diff --git a/sysdeps/generic/math-use-builtins-roundeven.h b/sysdeps/generic/math-use-builtins-roundeven.h
new file mode 100644
index 0000000000..bd2e84c93f
--- /dev/null
+++ b/sysdeps/generic/math-use-builtins-roundeven.h
@@ -0,0 +1,4 @@
+#define USE_ROUNDEVEN_BUILTIN 0
+#define USE_ROUNDEVENF_BUILTIN 0
+#define USE_ROUNDEVENL_BUILTIN 0
+#define USE_ROUNDEVENF128_BUILTIN 0
diff --git a/sysdeps/generic/math-use-builtins.h b/sysdeps/generic/math-use-builtins.h
index cc3e9f2201..19d2d1cf3c 100644
--- a/sysdeps/generic/math-use-builtins.h
+++ b/sysdeps/generic/math-use-builtins.h
@@ -30,6 +30,7 @@
#include <math-use-builtins-ceil.h>
#include <math-use-builtins-trunc.h>
#include <math-use-builtins-round.h>
+#include <math-use-builtins-roundeven.h>
#include <math-use-builtins-copysign.h>
#include <math-use-builtins-sqrt.h>
#include <math-use-builtins-fma.h>
diff --git a/sysdeps/ieee754/dbl-64/s_roundeven.c b/sysdeps/ieee754/dbl-64/s_roundeven.c
index 5deff3bb8a..88c051e9fe 100644
--- a/sysdeps/ieee754/dbl-64/s_roundeven.c
+++ b/sysdeps/ieee754/dbl-64/s_roundeven.c
@@ -21,6 +21,7 @@
#include <math_private.h>
#include <libm-alias-double.h>
#include <stdint.h>
+#include <math-use-builtins.h>
#define BIAS 0x3ff
#define MANT_DIG 53
@@ -29,6 +30,9 @@
double
__roundeven (double x)
{
+#if USE_ROUNDEVEN_BUILTIN
+ return __builtin_roundeven (x);
+#else
uint64_t ix, ux;
EXTRACT_WORDS64 (ix, x);
ux = ix & 0x7fffffffffffffffULL;
@@ -66,6 +70,7 @@ __roundeven (double x)
ix &= 0x8000000000000000ULL;
INSERT_WORDS64 (x, ix);
return x;
+#endif /* ! USE_ROUNDEVEN_BUILTIN */
}
#ifndef __roundeven
libm_alias_double (__roundeven, roundeven)
diff --git a/sysdeps/ieee754/flt-32/s_roundevenf.c b/sysdeps/ieee754/flt-32/s_roundevenf.c
index 2cc0d7b001..4f568579d8 100644
--- a/sysdeps/ieee754/flt-32/s_roundevenf.c
+++ b/sysdeps/ieee754/flt-32/s_roundevenf.c
@@ -21,6 +21,7 @@
#include <math.h>
#include <math_private.h>
#include <libm-alias-float.h>
+#include <math-use-builtins.h>
#include <stdint.h>
#define BIAS 0x7f
@@ -30,6 +31,9 @@
float
__roundevenf (float x)
{
+#if USE_ROUNDEVENF_BUILTIN
+ return __builtin_roundevenf (x);
+#else
uint32_t ix, ux;
GET_FLOAT_WORD (ix, x);
ux = ix & 0x7fffffff;
@@ -67,6 +71,7 @@ __roundevenf (float x)
ix &= 0x80000000;
SET_FLOAT_WORD (x, ix);
return x;
+#endif /* ! USE_ROUNDEVENF_BUILTIN */
}
#ifndef __roundevenf
libm_alias_float (__roundeven, roundeven)
diff --git a/sysdeps/ieee754/ldbl-128/s_roundevenl.c b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
index 7c9c11fad2..afbf13d4f2 100644
--- a/sysdeps/ieee754/ldbl-128/s_roundevenl.c
+++ b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
@@ -21,6 +21,7 @@
#include <math.h>
#include <math_private.h>
#include <libm-alias-ldouble.h>
+#include <math-use-builtins.h>
#include <stdint.h>
#define BIAS 0x3fff
@@ -30,6 +31,9 @@
_Float128
__roundevenl (_Float128 x)
{
+#if USE_ROUNDEVENL_BUILTIN
+ return __builtin_roundevenl (x);
+#else
uint64_t hx, lx, uhx;
GET_LDOUBLE_WORDS64 (hx, lx, x);
uhx = hx & 0x7fffffffffffffffULL;
@@ -101,5 +105,6 @@ __roundevenl (_Float128 x)
}
SET_LDOUBLE_WORDS64 (x, hx, lx);
return x;
+#endif /* ! USE_ROUNDEVENL_BUILTIN */
}
libm_alias_ldouble (__roundeven, roundeven)
--
2.31.1
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v6 2/3] x86_64: roundeven with sse4.1 support
2021-05-22 0:22 ` [PATCH v6 2/3] x86_64: roundeven with sse4.1 support Shen-Ta Hsieh
@ 2021-05-22 4:52 ` Paul Zimmermann
2021-05-23 0:41 ` 謝昇達(Shen-Ta Hsieh)
2021-05-23 17:10 ` H.J. Lu
2021-05-23 17:12 ` Noah Goldstein
2 siblings, 1 reply; 12+ messages in thread
From: Paul Zimmermann @ 2021-05-22 4:52 UTC (permalink / raw)
To: Shen-Ta Hsieh; +Cc: libc-alpha, ibmibmibm.tw
Dear Shen-Ta,
> Here is a benchmark result on my AMD Ryzen 9 3900X system:
>
> * benchmark result before this commit
> | | roundeven | roundevenf |
> |------------|--------------|--------------|
> | duration | 3.77659e+09 | 3.77504e+09 |
> | iterations | 3.97043e+08 | 4.36752e+08 |
> | max | 83.714 | 58.861 |
> | min | 7.144 | 6.27 |
> | mean | 9.51179 | 8.64345 |
>
> * benchmark result after this commit
> | | roundeven | roundevenf |
> |------------|--------------|--------------|
> | duration | 3.76913e+09 | 3.76923e+09 |
> | iterations | 5.55921e+08 | 5.64822e+08 |
> | max | 211.698 | 439.09 |
> | min | 6.498 | 6.422 |
> | mean | 6.77998 | 6.6733 |
I wonder why the max times have increased by a factor 2.5 and 7.5.
In my experiments I noticed that the "mean" time was quite stable,
while the "max" time could vary a lot between different runs. Thus
I usually run 5 times "make bench" and keep the smallest times:
$ ./testrun.sh benchtests/bench-roundeven
You can also try make USE_RDTSCP=1 bench (cf benchtests/README).
And it would be nice to have figures on another hardware (for example Intel).
Best regards,
Paul Zimmermann
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v6 2/3] x86_64: roundeven with sse4.1 support
2021-05-22 4:52 ` Paul Zimmermann
@ 2021-05-23 0:41 ` 謝昇達(Shen-Ta Hsieh)
2021-05-23 16:14 ` Paul Zimmermann
0 siblings, 1 reply; 12+ messages in thread
From: 謝昇達(Shen-Ta Hsieh) @ 2021-05-23 0:41 UTC (permalink / raw)
To: Paul Zimmermann; +Cc: libc-alpha
Dear Paul,
Paul Zimmermann <Paul.Zimmermann@inria.fr> 於 2021年5月22日 週六 下午12:52寫道:
>
> Dear Shen-Ta,
>
> > Here is a benchmark result on my AMD Ryzen 9 3900X system:
> >
> > * benchmark result before this commit
> > | | roundeven | roundevenf |
> > |------------|--------------|--------------|
> > | duration | 3.77659e+09 | 3.77504e+09 |
> > | iterations | 3.97043e+08 | 4.36752e+08 |
> > | max | 83.714 | 58.861 |
> > | min | 7.144 | 6.27 |
> > | mean | 9.51179 | 8.64345 |
> >
> > * benchmark result after this commit
> > | | roundeven | roundevenf |
> > |------------|--------------|--------------|
> > | duration | 3.76913e+09 | 3.76923e+09 |
> > | iterations | 5.55921e+08 | 5.64822e+08 |
> > | max | 211.698 | 439.09 |
> > | min | 6.498 | 6.422 |
> > | mean | 6.77998 | 6.6733 |
>
> I wonder why the max times have increased by a factor 2.5 and 7.5.
> In my experiments I noticed that the "mean" time was quite stable,
> while the "max" time could vary a lot between different runs. Thus
> I usually run 5 times "make bench" and keep the smallest times:
>
> $ ./testrun.sh benchtests/bench-roundeven
>
> You can also try make USE_RDTSCP=1 bench (cf benchtests/README).
>
> And it would be nice to have figures on another hardware (for example Intel).
>
> Best regards,
> Paul Zimmermann
>
I've tried to run another benchmark following your advises, and here
is the result:
# AMD Ryzen 9 3900X 12-Core Processor
* benchmark result before this commit
| | roundeven | roundevenf |
|------------|--------------|--------------|
| duration | 3.75587e+09 | 3.75114e+09 |
| iterations | 3.93053e+08 | 4.35402e+08 |
| max | 52.592 | 58.71 |
| min | 7.98 | 7.22 |
| mean | 9.55563 | 8.61535 |
* benchmark result after this commit
| | roundeven | roundevenf |
|------------|---------------|--------------|
| duration | 3.73815e+09 | 3.73738e+09 |
| iterations | 5.82692e+08 | 5.91498e+08 |
| max | 56.468 | 51.642 |
| min | 6.27 | 6.156 |
| mean | 6.41532 | 6.3185 |
# Intel(R) Pentium(R) CPU D1508 @ 2.20GHz
* benchmark result before this commit
| | roundeven | roundevenf |
|------------|--------------|--------------|
| duration | 2.18208e+09 | 2.18258e+09 |
| iterations | 2.39932e+08 | 2.46924e+08 |
| max | 96.378 | 98.035 |
| min | 6.776 | 5.94 |
| mean | 9.09456 | 8.83907 |
* benchmark result after this commit
| | roundeven | roundevenf |
|------------|--------------|--------------|
| duration | 2.17415e+09 | 2.17005e+09 |
| iterations | 3.56193e+08 | 4.09824e+08 |
| max | 51.693 | 97.192 |
| min | 5.926 | 5.093 |
| mean | 6.10385 | 5.29507 |
I ran each tests about 20 times, and pick a stable result out.
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v6 2/3] x86_64: roundeven with sse4.1 support
2021-05-23 0:41 ` 謝昇達(Shen-Ta Hsieh)
@ 2021-05-23 16:14 ` Paul Zimmermann
0 siblings, 0 replies; 12+ messages in thread
From: Paul Zimmermann @ 2021-05-23 16:14 UTC (permalink / raw)
To: ibmibmibm.tw; +Cc: libc-alpha
Dear Shen-Ta,
thank you, this is a clear improvement! Sorry I cannot comment on the patch
itself since I'm not familiar with the redirect mechanism nor sse 4.1.
Paul
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v6 2/3] x86_64: roundeven with sse4.1 support
2021-05-22 0:22 ` [PATCH v6 2/3] x86_64: roundeven with sse4.1 support Shen-Ta Hsieh
2021-05-22 4:52 ` Paul Zimmermann
@ 2021-05-23 17:10 ` H.J. Lu
2021-05-23 17:12 ` Noah Goldstein
2 siblings, 0 replies; 12+ messages in thread
From: H.J. Lu @ 2021-05-23 17:10 UTC (permalink / raw)
To: Shen-Ta Hsieh; +Cc: GNU C Library
On Fri, May 21, 2021 at 5:23 PM Shen-Ta Hsieh <ibmibmibm.tw@gmail.com> wrote:
>
> This patch adds support for the sse4.1 hardware floating point
> roundeven.
>
> Here is a benchmark result on my AMD Ryzen 9 3900X system:
>
> * benchmark result before this commit
> | | roundeven | roundevenf |
> |------------|--------------|--------------|
> | duration | 3.77659e+09 | 3.77504e+09 |
> | iterations | 3.97043e+08 | 4.36752e+08 |
> | max | 83.714 | 58.861 |
> | min | 7.144 | 6.27 |
> | mean | 9.51179 | 8.64345 |
>
> * benchmark result after this commit
> | | roundeven | roundevenf |
> |------------|--------------|--------------|
> | duration | 3.76913e+09 | 3.76923e+09 |
> | iterations | 5.55921e+08 | 5.64822e+08 |
> | max | 211.698 | 439.09 |
> | min | 6.498 | 6.422 |
> | mean | 6.77998 | 6.6733 |
>
> Signed-off-by: Shen-Ta Hsieh <ibmibmibm.tw@gmail.com>
> ---
> sysdeps/x86_64/fpu/multiarch/Makefile | 5 +--
> sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c | 2 ++
> .../x86_64/fpu/multiarch/s_roundeven-sse4_1.S | 24 ++++++++++++++
> sysdeps/x86_64/fpu/multiarch/s_roundeven.c | 31 +++++++++++++++++++
> sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c | 3 ++
> .../fpu/multiarch/s_roundevenf-sse4_1.S | 24 ++++++++++++++
> sysdeps/x86_64/fpu/multiarch/s_roundevenf.c | 31 +++++++++++++++++++
> 7 files changed, 118 insertions(+), 2 deletions(-)
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundeven.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundevenf.c
>
> diff --git a/sysdeps/x86_64/fpu/multiarch/Makefile b/sysdeps/x86_64/fpu/multiarch/Makefile
> index 57892c56bb..d425ffd6d3 100644
> --- a/sysdeps/x86_64/fpu/multiarch/Makefile
> +++ b/sysdeps/x86_64/fpu/multiarch/Makefile
> @@ -1,11 +1,12 @@
> ifeq ($(subdir),math)
> libm-sysdep_routines += s_floor-c s_ceil-c s_floorf-c s_ceilf-c \
> s_rint-c s_rintf-c s_nearbyint-c s_nearbyintf-c \
> - s_trunc-c s_truncf-c
> + s_roundeven-c s_roundevenf-c s_trunc-c s_truncf-c
>
> libm-sysdep_routines += s_ceil-sse4_1 s_ceilf-sse4_1 s_floor-sse4_1 \
> s_floorf-sse4_1 s_nearbyint-sse4_1 \
> - s_nearbyintf-sse4_1 s_rint-sse4_1 s_rintf-sse4_1 \
> + s_nearbyintf-sse4_1 s_roundeven-sse4_1 \
> + s_roundevenf-sse4_1 s_rint-sse4_1 s_rintf-sse4_1 \
> s_trunc-sse4_1 s_truncf-sse4_1
>
> libm-sysdep_routines += e_exp-fma e_log-fma e_pow-fma s_atan-fma \
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c b/sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c
> new file mode 100644
> index 0000000000..c7be43cb22
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c
> @@ -0,0 +1,2 @@
> +#define __roundeven __roundeven_c
> +#include <sysdeps/ieee754/dbl-64/s_roundeven.c>
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S b/sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S
> new file mode 100644
> index 0000000000..6ae8f6b1d3
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S
> @@ -0,0 +1,24 @@
> +/* Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +
> + .section .text.sse4.1,"ax",@progbits
> +ENTRY(__roundeven_sse41)
> + roundsd $8, %xmm0, %xmm0
> + ret
> +END(__roundeven_sse41)
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundeven.c b/sysdeps/x86_64/fpu/multiarch/s_roundeven.c
> new file mode 100644
> index 0000000000..d92eda652a
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundeven.c
> @@ -0,0 +1,31 @@
> +/* Multiple versions of __roundeven.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <libm-alias-double.h>
> +
> +#define roundeven __redirect_roundeven
> +#define __roundeven __redirect___roundeven
> +#include <math.h>
> +#undef roundeven
> +#undef __roundeven
> +
> +#define SYMBOL_NAME roundeven
> +#include "ifunc-sse4_1.h"
> +
> +libc_ifunc_redirected (__redirect_roundeven, __roundeven, IFUNC_SELECTOR ());
> +libm_alias_double (__roundeven, roundeven)
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c
> new file mode 100644
> index 0000000000..72a6e7d1fb
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c
> @@ -0,0 +1,3 @@
> +#undef __roundevenf
> +#define __roundevenf __roundevenf_c
> +#include <sysdeps/ieee754/flt-32/s_roundevenf.c>
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S
> new file mode 100644
> index 0000000000..a76e10807e
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S
> @@ -0,0 +1,24 @@
> +/* Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +
> + .section .text.sse4.1,"ax",@progbits
> +ENTRY(__roundevenf_sse41)
> + roundss $8, %xmm0, %xmm0
> + ret
> +END(__roundevenf_sse41)
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundevenf.c b/sysdeps/x86_64/fpu/multiarch/s_roundevenf.c
> new file mode 100644
> index 0000000000..2ee196e68f
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundevenf.c
> @@ -0,0 +1,31 @@
> +/* Multiple versions of __roundevenf.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <libm-alias-float.h>
> +
> +#define roundevenf __redirect_roundevenf
> +#define __roundevenf __redirect___roundevenf
> +#include <math.h>
> +#undef roundevenf
> +#undef __roundevenf
> +
> +#define SYMBOL_NAME roundevenf
> +#include "ifunc-sse4_1.h"
> +
> +libc_ifunc_redirected (__redirect_roundevenf, __roundevenf, IFUNC_SELECTOR ());
> +libm_alias_float (__roundeven, roundeven)
> --
> 2.31.1
>
LGTM.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
Thanks.
--
H.J.
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v6 2/3] x86_64: roundeven with sse4.1 support
2021-05-22 0:22 ` [PATCH v6 2/3] x86_64: roundeven with sse4.1 support Shen-Ta Hsieh
2021-05-22 4:52 ` Paul Zimmermann
2021-05-23 17:10 ` H.J. Lu
@ 2021-05-23 17:12 ` Noah Goldstein
2 siblings, 0 replies; 12+ messages in thread
From: Noah Goldstein @ 2021-05-23 17:12 UTC (permalink / raw)
To: Shen-Ta Hsieh; +Cc: GNU C Library
On Fri, May 21, 2021 at 6:08 PM Shen-Ta Hsieh via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
> This patch adds support for the sse4.1 hardware floating point
> roundeven.
>
> Here is a benchmark result on my AMD Ryzen 9 3900X system:
>
> * benchmark result before this commit
> | | roundeven | roundevenf |
> |------------|--------------|--------------|
> | duration | 3.77659e+09 | 3.77504e+09 |
> | iterations | 3.97043e+08 | 4.36752e+08 |
> | max | 83.714 | 58.861 |
> | min | 7.144 | 6.27 |
> | mean | 9.51179 | 8.64345 |
>
> * benchmark result after this commit
> | | roundeven | roundevenf |
> |------------|--------------|--------------|
> | duration | 3.76913e+09 | 3.76923e+09 |
> | iterations | 5.55921e+08 | 5.64822e+08 |
> | max | 211.698 | 439.09 |
> | min | 6.498 | 6.422 |
> | mean | 6.77998 | 6.6733 |
>
> Signed-off-by: Shen-Ta Hsieh <ibmibmibm.tw@gmail.com>
You should probably update the commit message with the
new numbers and a mention that the relevant tests are
passing.
> ---
> sysdeps/x86_64/fpu/multiarch/Makefile | 5 +--
> sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c | 2 ++
> .../x86_64/fpu/multiarch/s_roundeven-sse4_1.S | 24 ++++++++++++++
> sysdeps/x86_64/fpu/multiarch/s_roundeven.c | 31 +++++++++++++++++++
> sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c | 3 ++
> .../fpu/multiarch/s_roundevenf-sse4_1.S | 24 ++++++++++++++
> sysdeps/x86_64/fpu/multiarch/s_roundevenf.c | 31 +++++++++++++++++++
> 7 files changed, 118 insertions(+), 2 deletions(-)
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundeven.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S
> create mode 100644 sysdeps/x86_64/fpu/multiarch/s_roundevenf.c
>
> diff --git a/sysdeps/x86_64/fpu/multiarch/Makefile b/sysdeps/x86_64/fpu/multiarch/Makefile
> index 57892c56bb..d425ffd6d3 100644
> --- a/sysdeps/x86_64/fpu/multiarch/Makefile
> +++ b/sysdeps/x86_64/fpu/multiarch/Makefile
> @@ -1,11 +1,12 @@
> ifeq ($(subdir),math)
> libm-sysdep_routines += s_floor-c s_ceil-c s_floorf-c s_ceilf-c \
> s_rint-c s_rintf-c s_nearbyint-c s_nearbyintf-c \
> - s_trunc-c s_truncf-c
> + s_roundeven-c s_roundevenf-c s_trunc-c s_truncf-c
>
> libm-sysdep_routines += s_ceil-sse4_1 s_ceilf-sse4_1 s_floor-sse4_1 \
> s_floorf-sse4_1 s_nearbyint-sse4_1 \
> - s_nearbyintf-sse4_1 s_rint-sse4_1 s_rintf-sse4_1 \
> + s_nearbyintf-sse4_1 s_roundeven-sse4_1 \
> + s_roundevenf-sse4_1 s_rint-sse4_1 s_rintf-sse4_1 \
> s_trunc-sse4_1 s_truncf-sse4_1
>
> libm-sysdep_routines += e_exp-fma e_log-fma e_pow-fma s_atan-fma \
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c b/sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c
> new file mode 100644
> index 0000000000..c7be43cb22
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundeven-c.c
> @@ -0,0 +1,2 @@
> +#define __roundeven __roundeven_c
> +#include <sysdeps/ieee754/dbl-64/s_roundeven.c>
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S b/sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S
> new file mode 100644
> index 0000000000..6ae8f6b1d3
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundeven-sse4_1.S
> @@ -0,0 +1,24 @@
> +/* Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +
> + .section .text.sse4.1,"ax",@progbits
> +ENTRY(__roundeven_sse41)
> + roundsd $8, %xmm0, %xmm0
> + ret
> +END(__roundeven_sse41)
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundeven.c b/sysdeps/x86_64/fpu/multiarch/s_roundeven.c
> new file mode 100644
> index 0000000000..d92eda652a
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundeven.c
> @@ -0,0 +1,31 @@
> +/* Multiple versions of __roundeven.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <libm-alias-double.h>
> +
> +#define roundeven __redirect_roundeven
> +#define __roundeven __redirect___roundeven
> +#include <math.h>
> +#undef roundeven
> +#undef __roundeven
> +
> +#define SYMBOL_NAME roundeven
> +#include "ifunc-sse4_1.h"
> +
> +libc_ifunc_redirected (__redirect_roundeven, __roundeven, IFUNC_SELECTOR ());
> +libm_alias_double (__roundeven, roundeven)
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c
> new file mode 100644
> index 0000000000..72a6e7d1fb
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-c.c
> @@ -0,0 +1,3 @@
> +#undef __roundevenf
> +#define __roundevenf __roundevenf_c
> +#include <sysdeps/ieee754/flt-32/s_roundevenf.c>
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S
> new file mode 100644
> index 0000000000..a76e10807e
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundevenf-sse4_1.S
> @@ -0,0 +1,24 @@
> +/* Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <sysdep.h>
> +
> + .section .text.sse4.1,"ax",@progbits
> +ENTRY(__roundevenf_sse41)
> + roundss $8, %xmm0, %xmm0
> + ret
> +END(__roundevenf_sse41)
> diff --git a/sysdeps/x86_64/fpu/multiarch/s_roundevenf.c b/sysdeps/x86_64/fpu/multiarch/s_roundevenf.c
> new file mode 100644
> index 0000000000..2ee196e68f
> --- /dev/null
> +++ b/sysdeps/x86_64/fpu/multiarch/s_roundevenf.c
> @@ -0,0 +1,31 @@
> +/* Multiple versions of __roundevenf.
> + Copyright (C) 2021 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <libm-alias-float.h>
> +
> +#define roundevenf __redirect_roundevenf
> +#define __roundevenf __redirect___roundevenf
> +#include <math.h>
> +#undef roundevenf
> +#undef __roundevenf
> +
> +#define SYMBOL_NAME roundevenf
> +#include "ifunc-sse4_1.h"
> +
> +libc_ifunc_redirected (__redirect_roundevenf, __roundevenf, IFUNC_SELECTOR ());
> +libm_alias_float (__roundeven, roundeven)
> --
> 2.31.1
>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v6 1/3] math: redirect roundeven function
2021-05-22 0:22 [PATCH v6 1/3] math: redirect roundeven function Shen-Ta Hsieh
2021-05-22 0:22 ` [PATCH v6 2/3] x86_64: roundeven with sse4.1 support Shen-Ta Hsieh
2021-05-22 0:22 ` [PATCH v6 3/3] Use GCC builtins for roundeven functions if desired Shen-Ta Hsieh
@ 2021-05-23 18:54 ` H.J. Lu
2021-05-24 1:39 ` 謝昇達(Shen-Ta Hsieh)
2 siblings, 1 reply; 12+ messages in thread
From: H.J. Lu @ 2021-05-23 18:54 UTC (permalink / raw)
To: Shen-Ta Hsieh; +Cc: GNU C Library
On Fri, May 21, 2021 at 5:23 PM Shen-Ta Hsieh <ibmibmibm.tw@gmail.com> wrote:
>
> This patch redirect roundeven function for futhermore changes.
>
> Signed-off-by: Shen-Ta Hsieh <ibmibmibm.tw@gmail.com>
> ---
> include/math.h | 2 +-
> sysdeps/ieee754/dbl-64/s_roundeven.c | 4 +++-
> sysdeps/ieee754/float128/s_roundevenf128.c | 1 +
> sysdeps/ieee754/flt-32/s_roundevenf.c | 3 +++
> sysdeps/ieee754/ldbl-128/s_roundevenl.c | 1 +
> sysdeps/ieee754/ldbl-96/s_roundevenl.c | 1 +
> 6 files changed, 10 insertions(+), 2 deletions(-)
>
> diff --git a/include/math.h b/include/math.h
> index e1c2a4eb64..b4772d3d3b 100644
> --- a/include/math.h
> +++ b/include/math.h
> @@ -38,7 +38,6 @@ libm_hidden_proto (__issignaling)
> libm_hidden_proto (__issignalingf)
> libm_hidden_proto (__exp)
> libm_hidden_proto (__expf)
> -libm_hidden_proto (__roundeven)
>
> # if !defined __NO_LONG_DOUBLE_MATH \
> && __LDOUBLE_REDIRECTS_TO_FLOAT128_ABI == 0
> @@ -159,6 +158,7 @@ fabsf128 (_Float128 x)
> MATH_REDIRECT (sqrt, "__ieee754_", MATH_REDIRECT_UNARY_ARGS)
> MATH_REDIRECT (ceil, "__", MATH_REDIRECT_UNARY_ARGS)
> MATH_REDIRECT (floor, "__", MATH_REDIRECT_UNARY_ARGS)
> +MATH_REDIRECT (roundeven, "__", MATH_REDIRECT_UNARY_ARGS)
> MATH_REDIRECT (rint, "__", MATH_REDIRECT_UNARY_ARGS)
> MATH_REDIRECT (trunc, "__", MATH_REDIRECT_UNARY_ARGS)
> MATH_REDIRECT (round, "__", MATH_REDIRECT_UNARY_ARGS)
> diff --git a/sysdeps/ieee754/dbl-64/s_roundeven.c b/sysdeps/ieee754/dbl-64/s_roundeven.c
> index 943b2c634c..5deff3bb8a 100644
> --- a/sysdeps/ieee754/dbl-64/s_roundeven.c
> +++ b/sysdeps/ieee754/dbl-64/s_roundeven.c
> @@ -16,6 +16,7 @@
> License along with the GNU C Library; if not, see
> <https://www.gnu.org/licenses/>. */
>
> +#define NO_MATH_REDIRECT
> #include <math.h>
> #include <math_private.h>
> #include <libm-alias-double.h>
> @@ -66,5 +67,6 @@ __roundeven (double x)
> INSERT_WORDS64 (x, ix);
> return x;
> }
> -hidden_def (__roundeven)
> +#ifndef __roundeven
> libm_alias_double (__roundeven, roundeven)
> +#endif
> diff --git a/sysdeps/ieee754/float128/s_roundevenf128.c b/sysdeps/ieee754/float128/s_roundevenf128.c
> index 5a9b3f395f..e0faf727f6 100644
> --- a/sysdeps/ieee754/float128/s_roundevenf128.c
> +++ b/sysdeps/ieee754/float128/s_roundevenf128.c
> @@ -1,2 +1,3 @@
> +#define NO_MATH_REDIRECT
> #include <float128_private.h>
> #include "../ldbl-128/s_roundevenl.c"
> diff --git a/sysdeps/ieee754/flt-32/s_roundevenf.c b/sysdeps/ieee754/flt-32/s_roundevenf.c
> index ae54365bbe..2cc0d7b001 100644
> --- a/sysdeps/ieee754/flt-32/s_roundevenf.c
> +++ b/sysdeps/ieee754/flt-32/s_roundevenf.c
> @@ -17,6 +17,7 @@
> License along with the GNU C Library; if not, see
> <https://www.gnu.org/licenses/>. */
>
> +#define NO_MATH_REDIRECT
> #include <math.h>
> #include <math_private.h>
> #include <libm-alias-float.h>
> @@ -67,4 +68,6 @@ __roundevenf (float x)
> SET_FLOAT_WORD (x, ix);
> return x;
> }
> +#ifndef __roundevenf
> libm_alias_float (__roundeven, roundeven)
> +#endif
> diff --git a/sysdeps/ieee754/ldbl-128/s_roundevenl.c b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> index 02765be0e0..7c9c11fad2 100644
> --- a/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> +++ b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> @@ -17,6 +17,7 @@
> License along with the GNU C Library; if not, see
> <https://www.gnu.org/licenses/>. */
>
> +#define NO_MATH_REDIRECT
Is this really needed?
> #include <math.h>
> #include <math_private.h>
> #include <libm-alias-ldouble.h>
> diff --git a/sysdeps/ieee754/ldbl-96/s_roundevenl.c b/sysdeps/ieee754/ldbl-96/s_roundevenl.c
> index 8340116ed8..f826829c31 100644
> --- a/sysdeps/ieee754/ldbl-96/s_roundevenl.c
> +++ b/sysdeps/ieee754/ldbl-96/s_roundevenl.c
> @@ -17,6 +17,7 @@
> License along with the GNU C Library; if not, see
> <https://www.gnu.org/licenses/>. */
>
> +#define NO_MATH_REDIRECT
Is this really needed?
> #include <math.h>
> #include <math_private.h>
> #include <libm-alias-ldouble.h>
> --
> 2.31.1
>
--
H.J.
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v6 3/3] Use GCC builtins for roundeven functions if desired.
2021-05-22 0:22 ` [PATCH v6 3/3] Use GCC builtins for roundeven functions if desired Shen-Ta Hsieh
@ 2021-05-23 19:03 ` H.J. Lu
2021-05-24 1:22 ` 謝昇達(Shen-Ta Hsieh)
0 siblings, 1 reply; 12+ messages in thread
From: H.J. Lu @ 2021-05-23 19:03 UTC (permalink / raw)
To: Shen-Ta Hsieh; +Cc: GNU C Library
On Fri, May 21, 2021 at 5:23 PM Shen-Ta Hsieh <ibmibmibm.tw@gmail.com> wrote:
>
> This patch is using the corresponding GCC builtin for roundevenf,
> roundeven and roundevenl if the USE_FUNCTION_BUILTIN macros are defined
> to one in math-use-builtins.h.
>
> These builtin functions is supported since GCC 10.
>
> The code of the generic implementation is not changed.
>
> Signed-off-by: Shen-Ta Hsieh <ibmibmibm.tw@gmail.com>
> ---
> sysdeps/generic/math-use-builtins-roundeven.h | 4 ++++
> sysdeps/generic/math-use-builtins.h | 1 +
> sysdeps/ieee754/dbl-64/s_roundeven.c | 5 +++++
> sysdeps/ieee754/flt-32/s_roundevenf.c | 5 +++++
> sysdeps/ieee754/ldbl-128/s_roundevenl.c | 5 +++++
> 5 files changed, 20 insertions(+)
> create mode 100644 sysdeps/generic/math-use-builtins-roundeven.h
>
> diff --git a/sysdeps/generic/math-use-builtins-roundeven.h b/sysdeps/generic/math-use-builtins-roundeven.h
> new file mode 100644
> index 0000000000..bd2e84c93f
> --- /dev/null
> +++ b/sysdeps/generic/math-use-builtins-roundeven.h
> @@ -0,0 +1,4 @@
> +#define USE_ROUNDEVEN_BUILTIN 0
> +#define USE_ROUNDEVENF_BUILTIN 0
> +#define USE_ROUNDEVENL_BUILTIN 0
> +#define USE_ROUNDEVENF128_BUILTIN 0
> diff --git a/sysdeps/generic/math-use-builtins.h b/sysdeps/generic/math-use-builtins.h
> index cc3e9f2201..19d2d1cf3c 100644
> --- a/sysdeps/generic/math-use-builtins.h
> +++ b/sysdeps/generic/math-use-builtins.h
> @@ -30,6 +30,7 @@
> #include <math-use-builtins-ceil.h>
> #include <math-use-builtins-trunc.h>
> #include <math-use-builtins-round.h>
> +#include <math-use-builtins-roundeven.h>
> #include <math-use-builtins-copysign.h>
> #include <math-use-builtins-sqrt.h>
> #include <math-use-builtins-fma.h>
> diff --git a/sysdeps/ieee754/dbl-64/s_roundeven.c b/sysdeps/ieee754/dbl-64/s_roundeven.c
> index 5deff3bb8a..88c051e9fe 100644
> --- a/sysdeps/ieee754/dbl-64/s_roundeven.c
> +++ b/sysdeps/ieee754/dbl-64/s_roundeven.c
> @@ -21,6 +21,7 @@
> #include <math_private.h>
> #include <libm-alias-double.h>
> #include <stdint.h>
> +#include <math-use-builtins.h>
>
> #define BIAS 0x3ff
> #define MANT_DIG 53
> @@ -29,6 +30,9 @@
> double
> __roundeven (double x)
> {
> +#if USE_ROUNDEVEN_BUILTIN
> + return __builtin_roundeven (x);
> +#else
> uint64_t ix, ux;
> EXTRACT_WORDS64 (ix, x);
> ux = ix & 0x7fffffffffffffffULL;
> @@ -66,6 +70,7 @@ __roundeven (double x)
> ix &= 0x8000000000000000ULL;
> INSERT_WORDS64 (x, ix);
> return x;
> +#endif /* ! USE_ROUNDEVEN_BUILTIN */
> }
> #ifndef __roundeven
> libm_alias_double (__roundeven, roundeven)
> diff --git a/sysdeps/ieee754/flt-32/s_roundevenf.c b/sysdeps/ieee754/flt-32/s_roundevenf.c
> index 2cc0d7b001..4f568579d8 100644
> --- a/sysdeps/ieee754/flt-32/s_roundevenf.c
> +++ b/sysdeps/ieee754/flt-32/s_roundevenf.c
> @@ -21,6 +21,7 @@
> #include <math.h>
> #include <math_private.h>
> #include <libm-alias-float.h>
> +#include <math-use-builtins.h>
> #include <stdint.h>
>
> #define BIAS 0x7f
> @@ -30,6 +31,9 @@
> float
> __roundevenf (float x)
> {
> +#if USE_ROUNDEVENF_BUILTIN
> + return __builtin_roundevenf (x);
> +#else
> uint32_t ix, ux;
> GET_FLOAT_WORD (ix, x);
> ux = ix & 0x7fffffff;
> @@ -67,6 +71,7 @@ __roundevenf (float x)
> ix &= 0x80000000;
> SET_FLOAT_WORD (x, ix);
> return x;
> +#endif /* ! USE_ROUNDEVENF_BUILTIN */
> }
> #ifndef __roundevenf
> libm_alias_float (__roundeven, roundeven)
> diff --git a/sysdeps/ieee754/ldbl-128/s_roundevenl.c b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> index 7c9c11fad2..afbf13d4f2 100644
> --- a/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> +++ b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> @@ -21,6 +21,7 @@
> #include <math.h>
> #include <math_private.h>
> #include <libm-alias-ldouble.h>
> +#include <math-use-builtins.h>
> #include <stdint.h>
>
> #define BIAS 0x3fff
> @@ -30,6 +31,9 @@
> _Float128
> __roundevenl (_Float128 x)
> {
> +#if USE_ROUNDEVENL_BUILTIN
> + return __builtin_roundevenl (x);
> +#else
> uint64_t hx, lx, uhx;
> GET_LDOUBLE_WORDS64 (hx, lx, x);
> uhx = hx & 0x7fffffffffffffffULL;
> @@ -101,5 +105,6 @@ __roundevenl (_Float128 x)
> }
> SET_LDOUBLE_WORDS64 (x, hx, lx);
> return x;
> +#endif /* ! USE_ROUNDEVENL_BUILTIN */
> }
> libm_alias_ldouble (__roundeven, roundeven)
> --
> 2.31.1
>
Do you need to map USE_ROUNDEVENL_BUILTIN to
USE_ROUNDEVENF128_BUILTIN in float128_private.h?
--
H.J.
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v6 3/3] Use GCC builtins for roundeven functions if desired.
2021-05-23 19:03 ` H.J. Lu
@ 2021-05-24 1:22 ` 謝昇達(Shen-Ta Hsieh)
0 siblings, 0 replies; 12+ messages in thread
From: 謝昇達(Shen-Ta Hsieh) @ 2021-05-24 1:22 UTC (permalink / raw)
To: H.J. Lu; +Cc: GNU C Library
Oops, I missed that. I'll add this in this patch, thanks!
H.J. Lu <hjl.tools@gmail.com> 於 2021年5月24日 週一 上午3:04寫道:
>
> On Fri, May 21, 2021 at 5:23 PM Shen-Ta Hsieh <ibmibmibm.tw@gmail.com> wrote:
> >
> > This patch is using the corresponding GCC builtin for roundevenf,
> > roundeven and roundevenl if the USE_FUNCTION_BUILTIN macros are defined
> > to one in math-use-builtins.h.
> >
> > These builtin functions is supported since GCC 10.
> >
> > The code of the generic implementation is not changed.
> >
> > Signed-off-by: Shen-Ta Hsieh <ibmibmibm.tw@gmail.com>
> > ---
> > sysdeps/generic/math-use-builtins-roundeven.h | 4 ++++
> > sysdeps/generic/math-use-builtins.h | 1 +
> > sysdeps/ieee754/dbl-64/s_roundeven.c | 5 +++++
> > sysdeps/ieee754/flt-32/s_roundevenf.c | 5 +++++
> > sysdeps/ieee754/ldbl-128/s_roundevenl.c | 5 +++++
> > 5 files changed, 20 insertions(+)
> > create mode 100644 sysdeps/generic/math-use-builtins-roundeven.h
> >
> > diff --git a/sysdeps/generic/math-use-builtins-roundeven.h b/sysdeps/generic/math-use-builtins-roundeven.h
> > new file mode 100644
> > index 0000000000..bd2e84c93f
> > --- /dev/null
> > +++ b/sysdeps/generic/math-use-builtins-roundeven.h
> > @@ -0,0 +1,4 @@
> > +#define USE_ROUNDEVEN_BUILTIN 0
> > +#define USE_ROUNDEVENF_BUILTIN 0
> > +#define USE_ROUNDEVENL_BUILTIN 0
> > +#define USE_ROUNDEVENF128_BUILTIN 0
> > diff --git a/sysdeps/generic/math-use-builtins.h b/sysdeps/generic/math-use-builtins.h
> > index cc3e9f2201..19d2d1cf3c 100644
> > --- a/sysdeps/generic/math-use-builtins.h
> > +++ b/sysdeps/generic/math-use-builtins.h
> > @@ -30,6 +30,7 @@
> > #include <math-use-builtins-ceil.h>
> > #include <math-use-builtins-trunc.h>
> > #include <math-use-builtins-round.h>
> > +#include <math-use-builtins-roundeven.h>
> > #include <math-use-builtins-copysign.h>
> > #include <math-use-builtins-sqrt.h>
> > #include <math-use-builtins-fma.h>
> > diff --git a/sysdeps/ieee754/dbl-64/s_roundeven.c b/sysdeps/ieee754/dbl-64/s_roundeven.c
> > index 5deff3bb8a..88c051e9fe 100644
> > --- a/sysdeps/ieee754/dbl-64/s_roundeven.c
> > +++ b/sysdeps/ieee754/dbl-64/s_roundeven.c
> > @@ -21,6 +21,7 @@
> > #include <math_private.h>
> > #include <libm-alias-double.h>
> > #include <stdint.h>
> > +#include <math-use-builtins.h>
> >
> > #define BIAS 0x3ff
> > #define MANT_DIG 53
> > @@ -29,6 +30,9 @@
> > double
> > __roundeven (double x)
> > {
> > +#if USE_ROUNDEVEN_BUILTIN
> > + return __builtin_roundeven (x);
> > +#else
> > uint64_t ix, ux;
> > EXTRACT_WORDS64 (ix, x);
> > ux = ix & 0x7fffffffffffffffULL;
> > @@ -66,6 +70,7 @@ __roundeven (double x)
> > ix &= 0x8000000000000000ULL;
> > INSERT_WORDS64 (x, ix);
> > return x;
> > +#endif /* ! USE_ROUNDEVEN_BUILTIN */
> > }
> > #ifndef __roundeven
> > libm_alias_double (__roundeven, roundeven)
> > diff --git a/sysdeps/ieee754/flt-32/s_roundevenf.c b/sysdeps/ieee754/flt-32/s_roundevenf.c
> > index 2cc0d7b001..4f568579d8 100644
> > --- a/sysdeps/ieee754/flt-32/s_roundevenf.c
> > +++ b/sysdeps/ieee754/flt-32/s_roundevenf.c
> > @@ -21,6 +21,7 @@
> > #include <math.h>
> > #include <math_private.h>
> > #include <libm-alias-float.h>
> > +#include <math-use-builtins.h>
> > #include <stdint.h>
> >
> > #define BIAS 0x7f
> > @@ -30,6 +31,9 @@
> > float
> > __roundevenf (float x)
> > {
> > +#if USE_ROUNDEVENF_BUILTIN
> > + return __builtin_roundevenf (x);
> > +#else
> > uint32_t ix, ux;
> > GET_FLOAT_WORD (ix, x);
> > ux = ix & 0x7fffffff;
> > @@ -67,6 +71,7 @@ __roundevenf (float x)
> > ix &= 0x80000000;
> > SET_FLOAT_WORD (x, ix);
> > return x;
> > +#endif /* ! USE_ROUNDEVENF_BUILTIN */
> > }
> > #ifndef __roundevenf
> > libm_alias_float (__roundeven, roundeven)
> > diff --git a/sysdeps/ieee754/ldbl-128/s_roundevenl.c b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> > index 7c9c11fad2..afbf13d4f2 100644
> > --- a/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> > +++ b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> > @@ -21,6 +21,7 @@
> > #include <math.h>
> > #include <math_private.h>
> > #include <libm-alias-ldouble.h>
> > +#include <math-use-builtins.h>
> > #include <stdint.h>
> >
> > #define BIAS 0x3fff
> > @@ -30,6 +31,9 @@
> > _Float128
> > __roundevenl (_Float128 x)
> > {
> > +#if USE_ROUNDEVENL_BUILTIN
> > + return __builtin_roundevenl (x);
> > +#else
> > uint64_t hx, lx, uhx;
> > GET_LDOUBLE_WORDS64 (hx, lx, x);
> > uhx = hx & 0x7fffffffffffffffULL;
> > @@ -101,5 +105,6 @@ __roundevenl (_Float128 x)
> > }
> > SET_LDOUBLE_WORDS64 (x, hx, lx);
> > return x;
> > +#endif /* ! USE_ROUNDEVENL_BUILTIN */
> > }
> > libm_alias_ldouble (__roundeven, roundeven)
> > --
> > 2.31.1
> >
>
> Do you need to map USE_ROUNDEVENL_BUILTIN to
> USE_ROUNDEVENF128_BUILTIN in float128_private.h?
>
> --
> H.J.
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v6 1/3] math: redirect roundeven function
2021-05-23 18:54 ` [PATCH v6 1/3] math: redirect roundeven function H.J. Lu
@ 2021-05-24 1:39 ` 謝昇達(Shen-Ta Hsieh)
0 siblings, 0 replies; 12+ messages in thread
From: 謝昇達(Shen-Ta Hsieh) @ 2021-05-24 1:39 UTC (permalink / raw)
To: H.J. Lu; +Cc: GNU C Library
I don't know if it is not needed, but I saw the same line at
"sysdeps/ieee754/ldbl-96/s_roundl.c", which is added by this commit:
https://sourceware.org/git/?p=glibc.git;a=commit;h=9755bc4686d8cd6a0e9539040b903e9e9291c319
H.J. Lu <hjl.tools@gmail.com> 於 2021年5月24日 週一 上午2:55寫道:
>
> On Fri, May 21, 2021 at 5:23 PM Shen-Ta Hsieh <ibmibmibm.tw@gmail.com> wrote:
> >
> > This patch redirect roundeven function for futhermore changes.
> >
> > Signed-off-by: Shen-Ta Hsieh <ibmibmibm.tw@gmail.com>
> > ---
> > include/math.h | 2 +-
> > sysdeps/ieee754/dbl-64/s_roundeven.c | 4 +++-
> > sysdeps/ieee754/float128/s_roundevenf128.c | 1 +
> > sysdeps/ieee754/flt-32/s_roundevenf.c | 3 +++
> > sysdeps/ieee754/ldbl-128/s_roundevenl.c | 1 +
> > sysdeps/ieee754/ldbl-96/s_roundevenl.c | 1 +
> > 6 files changed, 10 insertions(+), 2 deletions(-)
> >
> > diff --git a/include/math.h b/include/math.h
> > index e1c2a4eb64..b4772d3d3b 100644
> > --- a/include/math.h
> > +++ b/include/math.h
> > @@ -38,7 +38,6 @@ libm_hidden_proto (__issignaling)
> > libm_hidden_proto (__issignalingf)
> > libm_hidden_proto (__exp)
> > libm_hidden_proto (__expf)
> > -libm_hidden_proto (__roundeven)
> >
> > # if !defined __NO_LONG_DOUBLE_MATH \
> > && __LDOUBLE_REDIRECTS_TO_FLOAT128_ABI == 0
> > @@ -159,6 +158,7 @@ fabsf128 (_Float128 x)
> > MATH_REDIRECT (sqrt, "__ieee754_", MATH_REDIRECT_UNARY_ARGS)
> > MATH_REDIRECT (ceil, "__", MATH_REDIRECT_UNARY_ARGS)
> > MATH_REDIRECT (floor, "__", MATH_REDIRECT_UNARY_ARGS)
> > +MATH_REDIRECT (roundeven, "__", MATH_REDIRECT_UNARY_ARGS)
> > MATH_REDIRECT (rint, "__", MATH_REDIRECT_UNARY_ARGS)
> > MATH_REDIRECT (trunc, "__", MATH_REDIRECT_UNARY_ARGS)
> > MATH_REDIRECT (round, "__", MATH_REDIRECT_UNARY_ARGS)
> > diff --git a/sysdeps/ieee754/dbl-64/s_roundeven.c b/sysdeps/ieee754/dbl-64/s_roundeven.c
> > index 943b2c634c..5deff3bb8a 100644
> > --- a/sysdeps/ieee754/dbl-64/s_roundeven.c
> > +++ b/sysdeps/ieee754/dbl-64/s_roundeven.c
> > @@ -16,6 +16,7 @@
> > License along with the GNU C Library; if not, see
> > <https://www.gnu.org/licenses/>. */
> >
> > +#define NO_MATH_REDIRECT
> > #include <math.h>
> > #include <math_private.h>
> > #include <libm-alias-double.h>
> > @@ -66,5 +67,6 @@ __roundeven (double x)
> > INSERT_WORDS64 (x, ix);
> > return x;
> > }
> > -hidden_def (__roundeven)
> > +#ifndef __roundeven
> > libm_alias_double (__roundeven, roundeven)
> > +#endif
> > diff --git a/sysdeps/ieee754/float128/s_roundevenf128.c b/sysdeps/ieee754/float128/s_roundevenf128.c
> > index 5a9b3f395f..e0faf727f6 100644
> > --- a/sysdeps/ieee754/float128/s_roundevenf128.c
> > +++ b/sysdeps/ieee754/float128/s_roundevenf128.c
> > @@ -1,2 +1,3 @@
> > +#define NO_MATH_REDIRECT
> > #include <float128_private.h>
> > #include "../ldbl-128/s_roundevenl.c"
> > diff --git a/sysdeps/ieee754/flt-32/s_roundevenf.c b/sysdeps/ieee754/flt-32/s_roundevenf.c
> > index ae54365bbe..2cc0d7b001 100644
> > --- a/sysdeps/ieee754/flt-32/s_roundevenf.c
> > +++ b/sysdeps/ieee754/flt-32/s_roundevenf.c
> > @@ -17,6 +17,7 @@
> > License along with the GNU C Library; if not, see
> > <https://www.gnu.org/licenses/>. */
> >
> > +#define NO_MATH_REDIRECT
> > #include <math.h>
> > #include <math_private.h>
> > #include <libm-alias-float.h>
> > @@ -67,4 +68,6 @@ __roundevenf (float x)
> > SET_FLOAT_WORD (x, ix);
> > return x;
> > }
> > +#ifndef __roundevenf
> > libm_alias_float (__roundeven, roundeven)
> > +#endif
> > diff --git a/sysdeps/ieee754/ldbl-128/s_roundevenl.c b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> > index 02765be0e0..7c9c11fad2 100644
> > --- a/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> > +++ b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
> > @@ -17,6 +17,7 @@
> > License along with the GNU C Library; if not, see
> > <https://www.gnu.org/licenses/>. */
> >
> > +#define NO_MATH_REDIRECT
>
> Is this really needed?
>
> > #include <math.h>
> > #include <math_private.h>
> > #include <libm-alias-ldouble.h>
> > diff --git a/sysdeps/ieee754/ldbl-96/s_roundevenl.c b/sysdeps/ieee754/ldbl-96/s_roundevenl.c
> > index 8340116ed8..f826829c31 100644
> > --- a/sysdeps/ieee754/ldbl-96/s_roundevenl.c
> > +++ b/sysdeps/ieee754/ldbl-96/s_roundevenl.c
> > @@ -17,6 +17,7 @@
> > License along with the GNU C Library; if not, see
> > <https://www.gnu.org/licenses/>. */
> >
> > +#define NO_MATH_REDIRECT
>
> Is this really needed?
>
> > #include <math.h>
> > #include <math_private.h>
> > #include <libm-alias-ldouble.h>
> > --
> > 2.31.1
> >
>
>
> --
> H.J.
^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2021-05-24 1:39 UTC | newest]
Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-05-22 0:22 [PATCH v6 1/3] math: redirect roundeven function Shen-Ta Hsieh
2021-05-22 0:22 ` [PATCH v6 2/3] x86_64: roundeven with sse4.1 support Shen-Ta Hsieh
2021-05-22 4:52 ` Paul Zimmermann
2021-05-23 0:41 ` 謝昇達(Shen-Ta Hsieh)
2021-05-23 16:14 ` Paul Zimmermann
2021-05-23 17:10 ` H.J. Lu
2021-05-23 17:12 ` Noah Goldstein
2021-05-22 0:22 ` [PATCH v6 3/3] Use GCC builtins for roundeven functions if desired Shen-Ta Hsieh
2021-05-23 19:03 ` H.J. Lu
2021-05-24 1:22 ` 謝昇達(Shen-Ta Hsieh)
2021-05-23 18:54 ` [PATCH v6 1/3] math: redirect roundeven function H.J. Lu
2021-05-24 1:39 ` 謝昇達(Shen-Ta Hsieh)
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).