[14/28] math: Use wordsize-64 version for isinf
diff mbox series

Message ID 20190329133529.22523-15-adhemerval.zanella@linaro.org
State New
Headers show
Series
  • powerpc floating-point optimization refactor
Related show

Commit Message

Adhemerval Zanella March 29, 2019, 1:35 p.m. UTC
- math.h will use compiler builtin for gcc 4.4+ when built without
    -fsignaling-nans and the builtin is expanded inline for all
    supported architectures.  As an example, there is no intra isnan
    call on libm for the architecture I checked (x86, arm, aarch64,
    and powerpc).

  - The resulting binary difference on 32 bits architecture is minimum
    for the non hotspot symbol.

  - It helps wordsize-64 architectures that use ldbl-opt.

  - It add some code simplification with reduction of duplicated
    implementations.

Checked on powerpc-linux-gnu (built without --with-cpu, with
--with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
and --disable-multi-arch).

        * sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c: Move to ...
        * sysdeps/ieee754/dbl-64/s_isinf.c: ... here and format code.
---
 sysdeps/ieee754/dbl-64/s_isinf.c             | 15 ++++-----
 sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c | 34 --------------------
 2 files changed, 6 insertions(+), 43 deletions(-)
 delete mode 100644 sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c

Comments

Gabriel F. T. Gomes June 11, 2019, 1:13 p.m. UTC | #1
On Fri, Mar 29 2019, Adhemerval Zanella wrote:
> 
>   - The resulting binary difference on 32 bits architecture is minimum
>     for the non hotspot symbol.
> 
>   - It helps wordsize-64 architectures that use ldbl-opt.

Same comments I made for patch 12/28 [1], apply here.

[1] https://sourceware.org/ml/libc-alpha/2019-06/msg00085.html
Szabolcs Nagy June 13, 2019, 8:57 a.m. UTC | #2
On 29/03/2019 13:35, Adhemerval Zanella wrote:
>   - math.h will use compiler builtin for gcc 4.4+ when built without
>     -fsignaling-nans and the builtin is expanded inline for all
>     supported architectures.  As an example, there is no intra isnan
>     call on libm for the architecture I checked (x86, arm, aarch64,
>     and powerpc).
> 
>   - The resulting binary difference on 32 bits architecture is minimum
>     for the non hotspot symbol.
> 
>   - It helps wordsize-64 architectures that use ldbl-opt.
> 
>   - It add some code simplification with reduction of duplicated
>     implementations.
> 
> Checked on powerpc-linux-gnu (built without --with-cpu, with
> --with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
> powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
> and --disable-multi-arch).
> 
>         * sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c: Move to ...
>         * sysdeps/ieee754/dbl-64/s_isinf.c: ... here and format code.


on aarch64 buildbot i see

make[2]: *** No rule to make target '../sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c', needed by '/B/math/s_isinf.o'.
make[2]: *** No rule to make target '../sysdeps/ieee754/dbl-64/wordsize-64/s_isnan.c', needed by '/B/math/s_isnan.o'.
make[2]: *** No rule to make target '../sysdeps/ieee754/dbl-64/wordsize-64/s_finite.c', needed by '/B/math/s_finite.o'.
make[2]: Target 'subdir_lib' not remade because of errors.
make[1]: *** [Makefile:259: math/subdir_lib] Error 2
Szabolcs Nagy June 13, 2019, 9:11 a.m. UTC | #3
On 13/06/2019 09:57, Szabolcs Nagy wrote:
> on aarch64 buildbot i see
> 
> make[2]: *** No rule to make target '../sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c', needed by '/B/math/s_isinf.o'.
> make[2]: *** No rule to make target '../sysdeps/ieee754/dbl-64/wordsize-64/s_isnan.c', needed by '/B/math/s_isnan.o'.
> make[2]: *** No rule to make target '../sysdeps/ieee754/dbl-64/wordsize-64/s_finite.c', needed by '/B/math/s_finite.o'.
> make[2]: Target 'subdir_lib' not remade because of errors.
> make[1]: *** [Makefile:259: math/subdir_lib] Error 2

sorry this is just a buildbot thing.

(it first runs the build in an unclean dir, which of
course had the deps for removed files, when it rerun
in a clean dir there was no failure.)

Patch
diff mbox series

diff --git a/sysdeps/ieee754/dbl-64/s_isinf.c b/sysdeps/ieee754/dbl-64/s_isinf.c
index 93eb65c147..e5300fd2a5 100644
--- a/sysdeps/ieee754/dbl-64/s_isinf.c
+++ b/sysdeps/ieee754/dbl-64/s_isinf.c
@@ -4,10 +4,6 @@ 
  * Public domain.
  */
 
-#if defined(LIBM_SCCS) && !defined(lint)
-static char rcsid[] = "$NetBSD: s_isinf.c,v 1.3 1995/05/11 23:20:14 jtc Exp $";
-#endif
-
 /*
  * isinf(x) returns 1 is x is inf, -1 if x is -inf, else 0;
  * no branching!
@@ -21,11 +17,12 @@  static char rcsid[] = "$NetBSD: s_isinf.c,v 1.3 1995/05/11 23:20:14 jtc Exp $";
 int
 __isinf (double x)
 {
-  int32_t hx, lx;
-  EXTRACT_WORDS (hx, lx, x);
-  lx |= (hx & 0x7fffffff) ^ 0x7ff00000;
-  lx |= -lx;
-  return ~(lx >> 31) & (hx >> 30);
+  int64_t ix;
+  EXTRACT_WORDS64 (ix,x);
+  int64_t t = ix & UINT64_C (0x7fffffffffffffff);
+  t ^= UINT64_C (0x7ff0000000000000);
+  t |= -t;
+  return ~(t >> 63) & (ix >> 62);
 }
 hidden_def (__isinf)
 weak_alias (__isinf, isinf)
diff --git a/sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c b/sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c
deleted file mode 100644
index 2b427a8b4c..0000000000
--- a/sysdeps/ieee754/dbl-64/wordsize-64/s_isinf.c
+++ /dev/null
@@ -1,34 +0,0 @@ 
-/*
- * Written by J.T. Conklin <jtc@netbsd.org>.
- * Changed to return -1 for -Inf by Ulrich Drepper <drepper@cygnus.com>.
- * Public domain.
- */
-
-/*
- * isinf(x) returns 1 is x is inf, -1 if x is -inf, else 0;
- * no branching!
- */
-
-#include <math.h>
-#include <math_private.h>
-#include <ldbl-classify-compat.h>
-#include <shlib-compat.h>
-
-int
-__isinf (double x)
-{
-	int64_t ix;
-	EXTRACT_WORDS64(ix,x);
-	int64_t t = ix & UINT64_C(0x7fffffffffffffff);
-	t ^= UINT64_C(0x7ff0000000000000);
-	t |= -t;
-	return ~(t >> 63) & (ix >> 62);
-}
-hidden_def (__isinf)
-weak_alias (__isinf, isinf)
-#ifdef NO_LONG_DOUBLE
-# if LDBL_CLASSIFY_COMPAT && SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_23)
-compat_symbol (libc, __isinf, __isinfl, GLIBC_2_0);
-# endif
-weak_alias (__isinf, isinfl)
-#endif