unofficial mirror of libc-alpha@sourceware.org
 help / color / mirror / Atom feed
* [PATCH v2 00/10] LoongArch: Use builtins with GCC >= 13
@ 2022-11-23  3:49 Xiaolin Tang
  2022-11-23  3:49 ` [PATCH v2 08/10] LoongArch: Add support for scalbn[f] Xiaolin Tang
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Xiaolin Tang @ 2022-11-23  3:49 UTC (permalink / raw)
  To: adhemerval.zanella, libc-alpha, caiyinyu, xry111
  Cc: xuchenghua, chenglulu, Xiaolin Tang

Patch 0001-0007:
GCC 13, LoongArch support builtins for {,l,ll}rint, logb,
use these built-ins instead of generic function implementation.

Patch 0008:
GCC 13, LoongArch support ___builtin_scalbn{,f} with -fno-math-errno,
but only "libm" can use -fno-math-errno in GLIBC, and scalbn is in 
libc instead of libm because __printf_fp calls it. So inline assembler
of the scalbn functions is a better choice for LoongArch.

Patch 0009: Add LoongArch support for scalb[f].


From v1 to v2:
Add patch 0010: Add LoongArch support for ilogb[f].


Discuss with Xi Ruoyao<xry111@xry111.site>
Link:
https://sourceware.org/pipermail/libc-alpha/2022-November/143425.html

Xi Ruoyao (1):
  LoongArch: Use __builtin_rint{,f} with GCC >= 13

Xiaolin Tang (9):
  Use GCC builtins for lrint functions if desired.
  LoongArch: Use __builtin_lrint{,f} with GCC >= 13
  Use GCC builtins for llrint functions if desired.
  LoongArch: Use __builtin_llrint{,f} with GCC >= 13
  Use GCC builtins for logb functions if desired.
  LoongArch: Use __builtin_logb{,f} with GCC >= 13
  LoongArch: Add support for scalbn[f]
  LoongArch: Add support for scalb[f]
  LoongArch: Add support for ilogb[f]

 sysdeps/generic/math-use-builtins-llrint.h    |  4 ++
 sysdeps/generic/math-use-builtins-logb.h      |  4 ++
 sysdeps/generic/math-use-builtins-lrint.h     |  4 ++
 sysdeps/generic/math-use-builtins.h           |  3 +
 sysdeps/ieee754/dbl-64/s_llrint.c             | 18 ++++--
 sysdeps/ieee754/dbl-64/s_logb.c               |  5 ++
 sysdeps/ieee754/dbl-64/s_lrint.c              | 18 ++++--
 sysdeps/ieee754/float128/float128_private.h   |  6 ++
 sysdeps/ieee754/flt-32/s_llrintf.c            | 18 ++++--
 sysdeps/ieee754/flt-32/s_logbf.c              |  5 ++
 sysdeps/ieee754/flt-32/s_lrintf.c             | 18 ++++--
 sysdeps/ieee754/ldbl-128/s_llrintl.c          | 17 ++++--
 sysdeps/ieee754/ldbl-128/s_logbl.c            |  6 ++
 sysdeps/ieee754/ldbl-128/s_lrintl.c           | 17 ++++--
 sysdeps/loongarch/fpu/e_ilogb.c               | 39 ++++++++++++
 sysdeps/loongarch/fpu/e_ilogbf.c              | 39 ++++++++++++
 sysdeps/loongarch/fpu/e_scalb.c               | 60 +++++++++++++++++++
 sysdeps/loongarch/fpu/e_scalbf.c              | 60 +++++++++++++++++++
 .../loongarch/fpu/math-use-builtins-llrint.h  | 10 ++++
 .../loongarch/fpu/math-use-builtins-logb.h    | 10 ++++
 .../loongarch/fpu/math-use-builtins-lrint.h   | 10 ++++
 .../loongarch/fpu/math-use-builtins-rint.h    |  9 +++
 sysdeps/loongarch/fpu/s_scalbn.c              | 29 +++++++++
 sysdeps/loongarch/fpu/s_scalbnf.c             | 29 +++++++++
 24 files changed, 404 insertions(+), 34 deletions(-)
 create mode 100644 sysdeps/generic/math-use-builtins-llrint.h
 create mode 100644 sysdeps/generic/math-use-builtins-logb.h
 create mode 100644 sysdeps/generic/math-use-builtins-lrint.h
 create mode 100644 sysdeps/loongarch/fpu/e_ilogb.c
 create mode 100644 sysdeps/loongarch/fpu/e_ilogbf.c
 create mode 100644 sysdeps/loongarch/fpu/e_scalb.c
 create mode 100644 sysdeps/loongarch/fpu/e_scalbf.c
 create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-llrint.h
 create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-logb.h
 create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-lrint.h
 create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-rint.h
 create mode 100644 sysdeps/loongarch/fpu/s_scalbn.c
 create mode 100644 sysdeps/loongarch/fpu/s_scalbnf.c

-- 
2.31.1


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH v2 08/10] LoongArch: Add support for scalbn[f]
  2022-11-23  3:49 [PATCH v2 00/10] LoongArch: Use builtins with GCC >= 13 Xiaolin Tang
@ 2022-11-23  3:49 ` Xiaolin Tang
  2022-11-23  3:49 ` [PATCH v2 09/10] LoongArch: Add support for scalb[f] Xiaolin Tang
  2022-11-23  3:49 ` [PATCH v2 10/10] LoongArch: Add support for ilogb[f] Xiaolin Tang
  2 siblings, 0 replies; 4+ messages in thread
From: Xiaolin Tang @ 2022-11-23  3:49 UTC (permalink / raw)
  To: adhemerval.zanella, libc-alpha, caiyinyu, xry111
  Cc: xuchenghua, chenglulu, Xiaolin Tang

Add inline assembler for the scalbn functions. Passes GLIBC regression.

GCC 13, LoongArch support ___builtin_scalbn{,f} with -fno-math-errno,
but only "libm" can use -fno-math-errno in GLIBC, and scalbn is in libc
instead of libm because __printf_fp calls it.
---
 sysdeps/loongarch/fpu/s_scalbn.c  | 29 +++++++++++++++++++++++++++++
 sysdeps/loongarch/fpu/s_scalbnf.c | 29 +++++++++++++++++++++++++++++
 2 files changed, 58 insertions(+)
 create mode 100644 sysdeps/loongarch/fpu/s_scalbn.c
 create mode 100644 sysdeps/loongarch/fpu/s_scalbnf.c

diff --git a/sysdeps/loongarch/fpu/s_scalbn.c b/sysdeps/loongarch/fpu/s_scalbn.c
new file mode 100644
index 0000000000..c03e81a31e
--- /dev/null
+++ b/sysdeps/loongarch/fpu/s_scalbn.c
@@ -0,0 +1,29 @@
+/* scalbn().  LoongArch version.
+   Copyright (C) 2022 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/>.  */
+
+#define NO_MATH_REDIRECT
+#include <math.h>
+
+double
+__scalbn (double x, int fn)
+{
+  double tmp;
+  asm volatile ("movgr2fr.d \t%0, %1" : "=f" (tmp) : "r" (fn));
+  asm volatile ("fscaleb.d \t%0, %1, %2" : "=f" (x) : "f" (x), "f" (tmp));
+  return x;
+}
diff --git a/sysdeps/loongarch/fpu/s_scalbnf.c b/sysdeps/loongarch/fpu/s_scalbnf.c
new file mode 100644
index 0000000000..15e6428044
--- /dev/null
+++ b/sysdeps/loongarch/fpu/s_scalbnf.c
@@ -0,0 +1,29 @@
+/* scalbnf().  LoongArch version.
+   Copyright (C) 2022 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/>.  */
+
+#define NO_MATH_REDIRECT
+#include <math.h>
+
+float
+__scalbnf (float x, int fn)
+{
+  float tmp;
+  asm volatile ("movgr2fr.w \t%0, %1" : "=f" (tmp) : "r" (fn));
+  asm volatile ("fscaleb.s \t%0, %1, %2" : "=f" (x) : "f" (x), "f" (tmp));
+  return x;
+}
-- 
2.31.1


^ permalink raw reply related	[flat|nested] 4+ messages in thread

* [PATCH v2 09/10] LoongArch: Add support for scalb[f]
  2022-11-23  3:49 [PATCH v2 00/10] LoongArch: Use builtins with GCC >= 13 Xiaolin Tang
  2022-11-23  3:49 ` [PATCH v2 08/10] LoongArch: Add support for scalbn[f] Xiaolin Tang
@ 2022-11-23  3:49 ` Xiaolin Tang
  2022-11-23  3:49 ` [PATCH v2 10/10] LoongArch: Add support for ilogb[f] Xiaolin Tang
  2 siblings, 0 replies; 4+ messages in thread
From: Xiaolin Tang @ 2022-11-23  3:49 UTC (permalink / raw)
  To: adhemerval.zanella, libc-alpha, caiyinyu, xry111
  Cc: xuchenghua, chenglulu, Xiaolin Tang

Add inline assembler for the scalb functions. Passes GLIBC regression.
---
 sysdeps/loongarch/fpu/e_scalb.c  | 60 ++++++++++++++++++++++++++++++++
 sysdeps/loongarch/fpu/e_scalbf.c | 60 ++++++++++++++++++++++++++++++++
 2 files changed, 120 insertions(+)
 create mode 100644 sysdeps/loongarch/fpu/e_scalb.c
 create mode 100644 sysdeps/loongarch/fpu/e_scalbf.c

diff --git a/sysdeps/loongarch/fpu/e_scalb.c b/sysdeps/loongarch/fpu/e_scalb.c
new file mode 100644
index 0000000000..9fdf21fa3a
--- /dev/null
+++ b/sysdeps/loongarch/fpu/e_scalb.c
@@ -0,0 +1,60 @@
+/* scalb().  LoongArch version.
+   Copyright (C) 2022 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/>.  */
+
+#define NO_MATH_REDIRECT
+#include <math.h>
+#include <libm-alias-finite.h>
+#include <fpu_control.h>
+#include <float.h>
+
+double
+__ieee754_scalb (double x, double fn)
+{
+  int x_cond;
+  int fn_cond;
+  asm volatile ("fclass.d \t%0, %1" : "=f" (x_cond) : "f" (x));
+  asm volatile ("fclass.d \t%0, %1" : "=f" (fn_cond) : "f" (fn));
+
+  if (__glibc_unlikely(( x_cond | fn_cond) & _FCLASS_NAN))
+      return x * fn;
+  else if (__glibc_unlikely(fn_cond & _FCLASS_INF))
+    {
+      if (!(fn_cond & _FCLASS_MINF))
+	  return x * fn;
+      else
+	  return x / -fn;
+    }
+  else if (__glibc_likely(-DBL_MAX < fn && fn < DBL_MAX))
+    {
+      double rint_fn, tmp;
+
+      /* rint_fn = rint(fn) */
+      asm volatile ("frint.d \t%0, %1" : "=f" (rint_fn) : "f" (fn));
+
+      if (rint_fn != fn )
+	  return (x - x) / (x - x);
+
+      asm volatile ("ftintrz.l.d \t%0, %1" : "=f" (tmp) : "f" (rint_fn));
+      asm volatile ("fscaleb.d \t%0, %1, %2" : "=f" (x) : "f" (x), "f" (tmp));
+    }
+  else
+    asm volatile ("fscaleb.d \t%0, %1, %2" : "=f" (x) : "f" (x), "f" (fn));
+
+  return x;
+}
+libm_alias_finite (__ieee754_scalb, __scalb)
diff --git a/sysdeps/loongarch/fpu/e_scalbf.c b/sysdeps/loongarch/fpu/e_scalbf.c
new file mode 100644
index 0000000000..fae2e92e20
--- /dev/null
+++ b/sysdeps/loongarch/fpu/e_scalbf.c
@@ -0,0 +1,60 @@
+/* scalbf().  LoongArch version.
+   Copyright (C) 2022 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/>.  */
+
+#define NO_MATH_REDIRECT
+#include <math.h>
+#include <libm-alias-finite.h>
+#include <fpu_control.h>
+#include <float.h>
+
+float
+__ieee754_scalbf (float x, float fn)
+{
+  int x_cond;
+  int fn_cond;
+  asm volatile ("fclass.s \t%0, %1" : "=f" (x_cond) : "f" (x));
+  asm volatile ("fclass.s \t%0, %1" : "=f" (fn_cond) : "f" (fn));
+
+  if (__glibc_unlikely(( x_cond | fn_cond) & _FCLASS_NAN))
+      return x * fn;
+  else if (__glibc_unlikely(fn_cond & _FCLASS_INF))
+    {
+      if (!(fn_cond & _FCLASS_MINF))
+	  return x * fn;
+      else
+	  return x / -fn;
+    }
+  else if (__glibc_likely(-FLT_MAX < fn && fn < FLT_MAX))
+    {
+      float rintf_fn, tmp;
+
+      /* rintf_fn = rintf(fn) */
+      asm volatile ("frint.s \t%0, %1" : "=f" (rintf_fn) : "f" (fn));
+
+      if (rintf_fn != fn )
+	  return (x - x) / (x - x);
+
+      asm volatile ("ftintrz.w.s \t%0, %1" : "=f" (tmp) : "f" (rintf_fn));
+      asm volatile ("fscaleb.s \t%0, %1, %2" : "=f" (x) : "f" (x), "f" (tmp));
+    }
+  else
+    asm volatile ("fscaleb.s \t%0, %1, %2" : "=f" (x) : "f" (x), "f" (fn));
+
+  return x;
+}
+libm_alias_finite (__ieee754_scalb, __scalb)
-- 
2.31.1


^ permalink raw reply related	[flat|nested] 4+ messages in thread

* [PATCH v2 10/10] LoongArch: Add support for ilogb[f]
  2022-11-23  3:49 [PATCH v2 00/10] LoongArch: Use builtins with GCC >= 13 Xiaolin Tang
  2022-11-23  3:49 ` [PATCH v2 08/10] LoongArch: Add support for scalbn[f] Xiaolin Tang
  2022-11-23  3:49 ` [PATCH v2 09/10] LoongArch: Add support for scalb[f] Xiaolin Tang
@ 2022-11-23  3:49 ` Xiaolin Tang
  2 siblings, 0 replies; 4+ messages in thread
From: Xiaolin Tang @ 2022-11-23  3:49 UTC (permalink / raw)
  To: adhemerval.zanella, libc-alpha, caiyinyu, xry111
  Cc: xuchenghua, chenglulu, Xiaolin Tang

Add inline assembler for the ilogb functions. Passes GLIBC regression.
---
 sysdeps/loongarch/fpu/e_ilogb.c  | 39 ++++++++++++++++++++++++++++++++
 sysdeps/loongarch/fpu/e_ilogbf.c | 39 ++++++++++++++++++++++++++++++++
 2 files changed, 78 insertions(+)
 create mode 100644 sysdeps/loongarch/fpu/e_ilogb.c
 create mode 100644 sysdeps/loongarch/fpu/e_ilogbf.c

diff --git a/sysdeps/loongarch/fpu/e_ilogb.c b/sysdeps/loongarch/fpu/e_ilogb.c
new file mode 100644
index 0000000000..f9ada6921a
--- /dev/null
+++ b/sysdeps/loongarch/fpu/e_ilogb.c
@@ -0,0 +1,39 @@
+/* __ieee754_ilogb().  LoongArch version.
+   Copyright (C) 2022 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/>.  */
+
+#define NO_MATH_REDIRECT
+#include <math.h>
+#include <fpu_control.h>
+
+int
+__ieee754_ilogb (double x)
+{
+  int x_cond;
+  asm volatile ("fclass.d \t%0, %1" : "=f" (x_cond) : "f" (x));
+
+  if (__glibc_unlikely (x_cond & _FCLASS_ZERO))
+      return FP_ILOGB0;
+  else if (__glibc_unlikely (x_cond & ( _FCLASS_NAN | _FCLASS_INF)))
+      return FP_ILOGBNAN;
+  else
+    {
+      asm volatile ("fabs.d \t%0, %1" : "=f" (x) : "f" (x));
+      asm volatile ("flogb.d \t%0, %1" : "=f" (x) : "f" (x));
+      return x;
+    }
+}
diff --git a/sysdeps/loongarch/fpu/e_ilogbf.c b/sysdeps/loongarch/fpu/e_ilogbf.c
new file mode 100644
index 0000000000..e1da48ecfc
--- /dev/null
+++ b/sysdeps/loongarch/fpu/e_ilogbf.c
@@ -0,0 +1,39 @@
+/* __ieee754_ilogbf().  LoongArch version.
+   Copyright (C) 2022 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/>.  */
+
+#define NO_MATH_REDIRECT
+#include <math.h>
+#include <fpu_control.h>
+
+int
+__ieee754_ilogbf (float x)
+{
+  int x_cond;
+  asm volatile ("fclass.s \t%0, %1" : "=f" (x_cond) : "f" (x));
+
+  if (__glibc_unlikely (x_cond & _FCLASS_ZERO))
+      return FP_ILOGB0;
+  else if (__glibc_unlikely (x_cond & ( _FCLASS_NAN | _FCLASS_INF)))
+      return FP_ILOGBNAN;
+  else
+    {
+      asm volatile ("fabs.s \t%0, %1" : "=f" (x) : "f" (x));
+      asm volatile ("flogb.s \t%0, %1" : "=f" (x) : "f" (x));
+      return x;
+    }
+}
-- 
2.31.1


^ permalink raw reply related	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2022-11-23  3:50 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-11-23  3:49 [PATCH v2 00/10] LoongArch: Use builtins with GCC >= 13 Xiaolin Tang
2022-11-23  3:49 ` [PATCH v2 08/10] LoongArch: Add support for scalbn[f] Xiaolin Tang
2022-11-23  3:49 ` [PATCH v2 09/10] LoongArch: Add support for scalb[f] Xiaolin Tang
2022-11-23  3:49 ` [PATCH v2 10/10] LoongArch: Add support for ilogb[f] Xiaolin Tang

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).