math: Optimize f{max,min}imum_mag_num{f,l,f128}

Simplify order signal comparison and reorganize check to reduce
branches and allow targets to use conditional select/move.

With gcc-15 on aarch64 for fmaximum_mag_num:

* master:

0000000000000000 <__fmaximum_mag_num>:
   0:   d503245f        bti     c
   4:   1e60c004        fabs    d4, d0
   8:   1e60c023        fabs    d3, d1
   c:   1e632080        fcmp    d4, d3
  10:   5400008d        b.le    20 <__fmaximum_mag_num+0x20>
  14:   1e60401f        fmov    d31, d0
  18:   1e6043e0        fmov    d0, d31
  1c:   d65f03c0        ret
  20:   54000085        b.pl    30 <__fmaximum_mag_num+0x30>  // b.nfrst
  24:   1e60403f        fmov    d31, d1
  28:   1e6043e0        fmov    d0, d31
  2c:   d65f03c0        ret
  30:   54000161        b.ne    5c <__fmaximum_mag_num+0x5c>  // b.any
  34:   4f000402        movi    v2.4s, #0x0
  38:   1e6e101d        fmov    d29, #1.000000000000000000e+00
  3c:   6ee0f842        fneg    v2.2d, v2.2d
  40:   4ea21c5e        mov     v30.16b, v2.16b
  44:   2e7d1c22        bsl     v2.8b, v1.8b, v29.8b
  48:   2e7d1c1e        bsl     v30.8b, v0.8b, v29.8b
  4c:   1e6223d0        fcmpe   d30, d2
  50:   1e61ac1f        fcsel   d31, d0, d1, ge // ge = tcont
  54:   1e6043e0        fmov    d0, d31
  58:   d65f03c0        ret
  5c:   1e612020        fcmp    d1, d1
  60:   1e60403f        fmov    d31, d1
  64:   54ffff87        b.vc    54 <__fmaximum_mag_num+0x54>
  68:   1e602000        fcmp    d0, d0
  6c:   1e60401f        fmov    d31, d0
  70:   54ffff27        b.vc    54 <__fmaximum_mag_num+0x54>
  74:   1e61281f        fadd    d31, d0, d1
  78:   17fffff7        b       54 <__fmaximum_mag_num+0x54>

* patch:

0000000000000000 <__fminimum_mag_num>:
   0:   d503245f        bti     c
   4:   1e612000        fcmp    d0, d1
   8:   1e60401f        fmov    d31, d0
   c:   540000e6        b.vs    28 <__fminimum_mag_num+0x28>
  10:   1e60c01e        fabs    d30, d0
  14:   1e60c03d        fabs    d29, d1
  18:   1e7d23c0        fcmp    d30, d29
  1c:   54000160        b.eq    48 <__fminimum_mag_num+0x48>  // b.none
  20:   1e605c20        fcsel   d0, d1, d0, pl  // pl = nfrst
  24:   d65f03c0        ret
  28:   1e612020        fcmp    d1, d1
  2c:   1e604020        fmov    d0, d1
  30:   54ffffa7        b.vc    24 <__fminimum_mag_num+0x24>
  34:   1e7f23e0        fcmp    d31, d31
  38:   1e6043e0        fmov    d0, d31
  3c:   54ffff47        b.vc    24 <__fminimum_mag_num+0x24>
  40:   1e612be0        fadd    d0, d31, d1
  44:   d65f03c0        ret
  48:   9e660000        fmov    x0, d0
  4c:   f100001f        cmp     x0, #0x0
  50:   1e61bc00        fcsel   d0, d0, d1, lt  // lt = tstop
  54:   d65f03c0        ret

Checked on x86_64-linux-gnu, aarch64-linux-gnu, i686-linux-gnu,
arm-linux-gnueabihf, powerpc64le-linux-gnu,
riscv64-linux-gnu-rv64imafdc-lp64d, and loongarch64-linux-gnuf64.

Reviewed-by: Wilco Dijkstra  <Wilco.Dijkstra@arm.com>

diff --git a/math/s_fmaximum_mag_num_template.c b/math/s_fmaximum_mag_num_template.c
index ee63f87f37638c8d91e86243c366f32ecf3cbed4..c9b6fbbdb5e9958afa35783614d5bb68cb8f4e40 100644 (file)
--- a/math/s_fmaximum_mag_num_template.c
+++ b/math/s_fmaximum_mag_num_template.c
@@ -21,14 +21,14 @@
 FLOAT
 M_DECL_FUNC (__fmaximum_mag_num) (FLOAT x, FLOAT y)
 {
-  FLOAT ax = M_FABS (x);
-  FLOAT ay = M_FABS (y);
-  if (isgreater (ax, ay))
-    return x;
-  else if (isless (ax, ay))
-    return y;
-  else if (ax == ay)
-    return (M_COPYSIGN (1, x) >= M_COPYSIGN (1, y) ? x : y);
+  if (__glibc_likely (!isunordered (x, y)))
+    {
+      FLOAT ax = M_FABS (x);
+      FLOAT ay = M_FABS (y);
+      if (__glibc_unlikely (ax == ay))
+        return signbit (x) ? y : x;
+      return isgreater (ax, ay) ? x : y;
+    }
   else
     return isnan (y) ? (isnan (x) ? x + y : x) : y;
 }

diff --git a/math/s_fminimum_mag_num_template.c b/math/s_fminimum_mag_num_template.c
index 1e9c117751e304f9ba47c20a4f0e11e6f360d1de..b097d54ffb211ecf8d7ef77f6da2ffad46dbad24 100644 (file)
--- a/math/s_fminimum_mag_num_template.c
+++ b/math/s_fminimum_mag_num_template.c
@@ -21,14 +21,14 @@
 FLOAT
 M_DECL_FUNC (__fminimum_mag_num) (FLOAT x, FLOAT y)
 {
-  FLOAT ax = M_FABS (x);
-  FLOAT ay = M_FABS (y);
-  if (isless (ax, ay))
-    return x;
-  else if (isgreater (ax, ay))
-    return y;
-  else if (ax == ay)
-    return (M_COPYSIGN (1, x) <= M_COPYSIGN (1, y) ? x : y);
+  if (__glibc_likely (!isunordered (x, y)))
+    {
+      FLOAT ax = M_FABS (x);
+      FLOAT ay = M_FABS (y);
+      if (__glibc_unlikely (ax == ay))
+        return signbit (x) ? x : y;
+      return isless (ax, ay) ? x : y;
+    }
   else
     return isnan (y) ? (isnan (x) ? x + y : x) : y;
 }