diff mbox series

[v2,2/4] math: Optimize double ilogb/llogb

Message ID 20250428170430.2030400-3-adhemerval.zanella@linaro.org
State New
Headers show
Series Remove UB and optimize ilogbf/ilogb | expand

Commit Message

Adhemerval Zanella Netto April 28, 2025, 5:03 p.m. UTC
It removes the wrapper by moving the error/EDOM handling to an
out-of-line implementation (__math_invalid_i/__math_invalid_li).

Also, __glibc_unlikely is used on errors case since it helps
code generation on recent gcc.

The code now builds to with gcc-14 on aarch64:

0000000000000000 <__ilogb>:
   0:   9e660000        fmov    x0, d0
   4:   d374f801        ubfx    x1, x0, #52, #11
   8:   340000e1        cbz     w1, 24 <__ilogb+0x24>
   c:   510ffc20        sub     w0, w1, #0x3ff
  10:   711ffc3f        cmp     w1, #0x7ff
  14:   54000040        b.eq    1c <__ilogb+0x1c>  // b.none
  18:   d65f03c0        ret
  1c:   12b00000        mov     w0, #0x7fffffff                 // #2147483647
  20:   14000000        b       0 <__math_invalid_i>
  24:   d374cc00        lsl     x0, x0, #12
  28:   b40000a0        cbz     x0, 3c <__ilogb+0x3c>
  2c:   dac01000        clz     x0, x0
  30:   12807fc1        mov     w1, #0xfffffc01                 // #-1023
  34:   4b000020        sub     w0, w1, w0
  38:   d65f03c0        ret
  3c:   320107e0        mov     w0, #0x80000001                 // #-2147483647
  40:   14000000        b       0 <__math_invalid_i>

Checked on aarch64-linux-gnu and x86_64-linux-gnu.
---
 sysdeps/i386/fpu/math_err.c          |  1 -
 sysdeps/ieee754/dbl-64/e_ilogb.c     | 42 +------------------
 sysdeps/ieee754/dbl-64/math_config.h |  2 +
 sysdeps/ieee754/dbl-64/math_err.c    | 32 ++++++++++++++
 sysdeps/ieee754/dbl-64/w_ilogb.c     | 63 ++++++++++++++++++++++++++++
 sysdeps/ieee754/dbl-64/w_llogb.c     |  2 +
 6 files changed, 100 insertions(+), 42 deletions(-)
 delete mode 100644 sysdeps/i386/fpu/math_err.c
 create mode 100644 sysdeps/ieee754/dbl-64/w_ilogb.c
 create mode 100644 sysdeps/ieee754/dbl-64/w_llogb.c
diff mbox series

Patch

diff --git a/sysdeps/i386/fpu/math_err.c b/sysdeps/i386/fpu/math_err.c
deleted file mode 100644
index 1cc8931700..0000000000
--- a/sysdeps/i386/fpu/math_err.c
+++ /dev/null
@@ -1 +0,0 @@ 
-/* Not needed.  */
diff --git a/sysdeps/ieee754/dbl-64/e_ilogb.c b/sysdeps/ieee754/dbl-64/e_ilogb.c
index 89e7498266..1ea2f23e2f 100644
--- a/sysdeps/ieee754/dbl-64/e_ilogb.c
+++ b/sysdeps/ieee754/dbl-64/e_ilogb.c
@@ -1,41 +1 @@ 
-/* Get integer exponent of a floating-point value.
-   Copyright (C) 1999-2025 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 <limits.h>
-#include <math.h>
-#include <stdbit.h>
-#include "math_config.h"
-
-int
-__ieee754_ilogb (double x)
-{
-  uint64_t ux = asuint64 (x);
-  int ex = (ux & ~SIGN_MASK) >> MANTISSA_WIDTH;
-  if (ex == 0) /* zero or subnormal */
-    {
-      /* Clear sign and exponent */
-      ux <<= 12;
-      if (ux == 0)
-	return FP_ILOGB0;
-      /* subnormal  */
-      return -1023 - stdc_leading_zeros (ux);
-    }
-  if (ex == EXPONENT_MASK >> MANTISSA_WIDTH) /* NaN or Inf */
-    return ux << 12 ? FP_ILOGBNAN : INT_MAX;
-  return ex - 1023;
-}
+/* ilogb is implemented at w_ilogb.c  */
diff --git a/sysdeps/ieee754/dbl-64/math_config.h b/sysdeps/ieee754/dbl-64/math_config.h
index 3382e385f9..5766f6a1db 100644
--- a/sysdeps/ieee754/dbl-64/math_config.h
+++ b/sysdeps/ieee754/dbl-64/math_config.h
@@ -164,6 +164,8 @@  attribute_hidden double __math_divzero (uint32_t);
 
 /* Invalid input unless it is a quiet NaN.  */
 attribute_hidden double __math_invalid (double);
+attribute_hidden int __math_invalid_i (int);
+attribute_hidden long int __math_invalid_li (long int);
 
 /* Error handling using output checking, only for errno setting.  */
 
diff --git a/sysdeps/ieee754/dbl-64/math_err.c b/sysdeps/ieee754/dbl-64/math_err.c
index 4a07fd5714..b8c645a3df 100644
--- a/sysdeps/ieee754/dbl-64/math_err.c
+++ b/sysdeps/ieee754/dbl-64/math_err.c
@@ -29,8 +29,24 @@  with_errno (double y, int e)
   errno = e;
   return y;
 }
+
+NOINLINE static int
+with_errno_i (int y, int e)
+{
+  errno = e;
+  return y;
+}
+
+NOINLINE static long int
+with_errno_li (long int y, int e)
+{
+  errno = e;
+  return y;
+}
 #else
 #define with_errno(x, e) (x)
+#define with_errno_i(x, e) (x)
+#define with_errno_li(x, e) (x)
 #endif
 
 attribute_hidden double
@@ -83,6 +99,22 @@  __math_invalid (double x)
   return isnan (x) ? y : with_errno (y, EDOM);
 }
 
+attribute_hidden int
+__math_invalid_i (int r)
+{
+  double y = 0.0 / 0.0;
+  math_force_eval (y);
+  return with_errno_i (r, EDOM);
+}
+
+attribute_hidden long int
+__math_invalid_li (long int r)
+{
+  double y = 0.0 / 0.0;
+  math_force_eval (y);
+  return with_errno_li (r, EDOM);
+}
+
 /* Check result and set errno if necessary.  */
 
 attribute_hidden double
diff --git a/sysdeps/ieee754/dbl-64/w_ilogb.c b/sysdeps/ieee754/dbl-64/w_ilogb.c
new file mode 100644
index 0000000000..8f9641be45
--- /dev/null
+++ b/sysdeps/ieee754/dbl-64/w_ilogb.c
@@ -0,0 +1,63 @@ 
+/* Get integer exponent of a floating-point value.
+   Copyright (C) 1999-2025 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 <fenv.h>
+#include <errno.h>
+#include <limits.h>
+#include <math.h>
+#include <stdbit.h>
+#include <libm-alias-double.h>
+#include "math_config.h"
+
+#ifdef DEF_AS_LLOGB
+# define IMPL_NAME   __llogb
+# define FUNC_NAME   llogb
+# define RET_TYPE    long int
+# define RET_LOGB0   FP_LLOGB0
+# define RET_LOGBNAN FP_LLOGBNAN
+# define RET_LOGMAX  LONG_MAX
+# define RET_INVALID __math_invalid_li
+#else
+# define IMPL_NAME   __ilogb
+# define FUNC_NAME   ilogb
+# define RET_TYPE    int
+# define RET_LOGB0   FP_ILOGB0
+# define RET_LOGBNAN FP_ILOGBNAN
+# define RET_LOGMAX  INT_MAX
+# define RET_INVALID __math_invalid_i
+#endif
+
+RET_TYPE
+IMPL_NAME (double x)
+{
+  uint64_t ux = asuint64 (x);
+  int ex = (ux & ~SIGN_MASK) >> MANTISSA_WIDTH;
+  if (__glibc_unlikely (ex == 0)) /* zero or subnormal */
+    {
+      /* Clear sign and exponent */
+      ux <<= 12;
+      if (ux == 0)
+	return RET_INVALID (RET_LOGB0);
+      /* subnormal  */
+      return (RET_TYPE)-1023 - stdc_leading_zeros (ux);
+    }
+  if (__glibc_unlikely (ex == EXPONENT_MASK >> MANTISSA_WIDTH)) /* NaN or Inf */
+    return RET_INVALID (ux << 12 ? RET_LOGBNAN : RET_LOGMAX);
+  return ex - 1023;
+}
+libm_alias_double (IMPL_NAME, FUNC_NAME)
diff --git a/sysdeps/ieee754/dbl-64/w_llogb.c b/sysdeps/ieee754/dbl-64/w_llogb.c
new file mode 100644
index 0000000000..c984cd151a
--- /dev/null
+++ b/sysdeps/ieee754/dbl-64/w_llogb.c
@@ -0,0 +1,2 @@ 
+#define DEF_AS_LLOGB
+#include "w_ilogb.c"