diff mbox series

[v2,4/4] math: Optimize float ilogb/llogb

Message ID 20250428170430.2030400-5-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_invalidf_i/__math_invalidf_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 <__ilogbf>:
   0:   1e260000        fmov    w0, s0
   4:   d3577801        ubfx    x1, x0, #23, #8
   8:   340000e1        cbz     w1, 24 <__ilogbf+0x24>
   c:   5101fc20        sub     w0, w1, #0x7f
  10:   7103fc3f        cmp     w1, #0xff
  14:   54000040        b.eq    1c <__ilogbf+0x1c>  // b.none
  18:   d65f03c0        ret
  1c:   12b00000        mov     w0, #0x7fffffff                 // #2147483647
  20:   14000000        b       0 <__math_invalidf_i>
  24:   53175800        lsl     w0, w0, #9
  28:   340000a0        cbz     w0, 3c <__ilogbf+0x3c>
  2c:   5ac01000        clz     w0, w0
  30:   12800fc1        mov     w1, #0xffffff81                 // #-127
  34:   4b000020        sub     w0, w1, w0
  38:   d65f03c0        ret
  3c:   320107e0        mov     w0, #0x80000001                 // #-2147483647
  40:   14000000        b       0 <__math_invalidf_i>

Checked on aarch64-linux-gnu and x86_64-linux-gnu.
---
 sysdeps/ieee754/flt-32/e_ilogbf.c    | 42 +-----------------
 sysdeps/ieee754/flt-32/math_config.h |  2 +
 sysdeps/ieee754/flt-32/math_errf.c   | 33 ++++++++++++++
 sysdeps/ieee754/flt-32/w_ilogbf.c    | 66 ++++++++++++++++++++++++++++
 sysdeps/ieee754/flt-32/w_llogbf.c    |  2 +
 5 files changed, 104 insertions(+), 41 deletions(-)
 create mode 100644 sysdeps/ieee754/flt-32/w_ilogbf.c
 create mode 100644 sysdeps/ieee754/flt-32/w_llogbf.c
diff mbox series

Patch

diff --git a/sysdeps/ieee754/flt-32/e_ilogbf.c b/sysdeps/ieee754/flt-32/e_ilogbf.c
index 024b114638..a27fb94732 100644
--- a/sysdeps/ieee754/flt-32/e_ilogbf.c
+++ b/sysdeps/ieee754/flt-32/e_ilogbf.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_ilogbf (float x)
-{
-  uint32_t ux = asuint (x);
-  int ex = (ux & ~SIGN_MASK) >> MANTISSA_WIDTH;
-  if (ex == 0) /* zero or subnormal */
-    {
-      /* Clear sign and exponent.  */
-      ux <<= 1 + EXPONENT_WIDTH;
-      if (ux == 0)
-	return FP_ILOGB0;
-      /* sbunormal */
-      return -127 - stdc_leading_zeros (ux);
-    }
-  if (ex == EXPONENT_MASK >> MANTISSA_WIDTH) /* NaN or Inf */
-    return ux << (1 + EXPONENT_WIDTH) ? FP_ILOGBNAN : INT_MAX;
-  return ex - 127;
-}
+/* ilogbf is implemented at w_ilogbf.c  */
diff --git a/sysdeps/ieee754/flt-32/math_config.h b/sysdeps/ieee754/flt-32/math_config.h
index 8d9c8ee3ad..fd9481a016 100644
--- a/sysdeps/ieee754/flt-32/math_config.h
+++ b/sysdeps/ieee754/flt-32/math_config.h
@@ -208,6 +208,8 @@  attribute_hidden float __math_uflowf (uint32_t);
 attribute_hidden float __math_may_uflowf (uint32_t);
 attribute_hidden float __math_divzerof (uint32_t);
 attribute_hidden float __math_invalidf (float);
+attribute_hidden int __math_invalidf_i (int);
+attribute_hidden long int __math_invalidf_li (long int);
 attribute_hidden float __math_edomf (float x);
 
 /* Shared between expf, exp2f, exp10f, and powf.  */
diff --git a/sysdeps/ieee754/flt-32/math_errf.c b/sysdeps/ieee754/flt-32/math_errf.c
index edcc4c094a..244e38a4ce 100644
--- a/sysdeps/ieee754/flt-32/math_errf.c
+++ b/sysdeps/ieee754/flt-32/math_errf.c
@@ -16,6 +16,7 @@ 
    License along with the GNU C Library; if not, see
    <https://www.gnu.org/licenses/>.  */
 
+#include <math-barriers.h>
 #include "math_config.h"
 
 #if WANT_ERRNO
@@ -27,8 +28,24 @@  with_errnof (float y, int e)
   errno = e;
   return y;
 }
+
+NOINLINE static int
+with_errnof_i (int y, int e)
+{
+  errno = e;
+  return y;
+}
+
+NOINLINE static long int
+with_errnof_li (long int y, int e)
+{
+  errno = e;
+  return y;
+}
 #else
 # define with_errnof(x, e) (x)
+# define with_errnof_i(x, x) (x)
+# define with_errnof_li(x, x) (x)
 #endif
 
 attribute_hidden float
@@ -80,3 +97,19 @@  __math_invalidf (float x)
   float y = (x - x) / (x - x);
   return isnan (x) ? y : with_errnof (y, EDOM);
 }
+
+attribute_hidden int
+__math_invalidf_i (int x)
+{
+  float y = 0.0f / 0.0f;
+  math_force_eval (y);
+  return with_errnof_i (x, EDOM);
+}
+
+attribute_hidden long int
+__math_invalidf_li (long int x)
+{
+  float y = 0.0f / 0.0f;
+  math_force_eval (y);
+  return with_errnof_li (x, EDOM);
+}
diff --git a/sysdeps/ieee754/flt-32/w_ilogbf.c b/sysdeps/ieee754/flt-32/w_ilogbf.c
new file mode 100644
index 0000000000..7f7314797d
--- /dev/null
+++ b/sysdeps/ieee754/flt-32/w_ilogbf.c
@@ -0,0 +1,66 @@ 
+/* 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-float.h>
+#include <math-type-macros-float.h>
+#include "math_config.h"
+
+#ifdef DEF_AS_LLOGBF
+# 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_invalidf_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_invalidf_i
+#endif
+
+RET_TYPE
+M_DECL_FUNC (IMPL_NAME) (float x)
+{
+  uint32_t ux = asuint (x);
+  int ex = (ux & ~SIGN_MASK) >> MANTISSA_WIDTH;
+  if (__glibc_unlikely (ex == 0))
+    {
+      /* Zero or subnormal.
+         Clear sign and exponent.  */
+      ux <<= 1 + EXPONENT_WIDTH;
+      if (ux == 0)
+	return RET_INVALID (RET_LOGB0);
+      /* sbunormal */
+      return (RET_TYPE)-127 - stdc_leading_zeros (ux);
+    }
+  if (__glibc_unlikely (ex == EXPONENT_MASK >> MANTISSA_WIDTH))
+    /* NaN or Inf */
+    return RET_INVALID (ux << (1 + EXPONENT_WIDTH) ? RET_LOGBNAN : RET_LOGMAX);
+  return ex - 127;
+}
+libm_alias_float (IMPL_NAME, FUNC_NAME);
diff --git a/sysdeps/ieee754/flt-32/w_llogbf.c b/sysdeps/ieee754/flt-32/w_llogbf.c
new file mode 100644
index 0000000000..8676434f6e
--- /dev/null
+++ b/sysdeps/ieee754/flt-32/w_llogbf.c
@@ -0,0 +1,2 @@ 
+#define DEF_AS_LLOGBF
+#include "w_ilogbf.c"