[PATCH 6/8] math: Sync log2 with Arm Optimized Routines

[Adhemerval Zanella]

Sync with commit 712aa21 for log2.c and 189dfef for log2_data.c  No
functional change.

Checked on x86_64-linux-gnu, i686-linux-gnu, and aarch64-linux-gnu.
---
 sysdeps/ieee754/dbl-64/e_log2.c      | 14 ++++++++++----
 sysdeps/ieee754/dbl-64/math_config.h |  3 +--
 2 files changed, 11 insertions(+), 6 deletions(-)

diff --git a/sysdeps/ieee754/dbl-64/e_log2.c b/sysdeps/ieee754/dbl-64/e_log2.c
index 872cebddd8..3f9a846ff2 100644
--- a/sysdeps/ieee754/dbl-64/e_log2.c
+++ b/sysdeps/ieee754/dbl-64/e_log2.c
@@ -50,8 +50,10 @@ __log2 (double x)
   ix = asuint64 (x);
   top = top16 (x);
 
-#define LO asuint64 (1.0 - 0x1.5b51p-5)
-#define HI asuint64 (1.0 + 0x1.6ab2p-5)
+#if LOG2_POLY1_ORDER == 11
+# define LO asuint64 (1.0 - 0x1.5b51p-5)
+# define HI asuint64 (1.0 + 0x1.6ab2p-5)
+#endif
   if (__glibc_unlikely (ix - LO < HI - LO))
     {
       /* Handle close to 1.0 inputs separately.  */
@@ -71,6 +73,7 @@ __log2 (double x)
 #endif
       r2 = r * r; /* rounding error: 0x1p-62.  */
       r4 = r2 * r2;
+#if LOG2_POLY1_ORDER == 11
       /* Worst-case error is less than 0.54 ULP (0.55 ULP without fma).  */
       p = r2 * (B[0] + r * B[1]);
       y = hi + p;
@@ -78,6 +81,7 @@ __log2 (double x)
       lo += r4 * (B[2] + r * B[3] + r2 * (B[4] + r * B[5])
 		  + r4 * (B[6] + r * B[7] + r2 * (B[8] + r * B[9])));
       y += lo;
+#endif
       return y;
     }
   if (__glibc_unlikely (top - 0x0010 >= 0x7ff0 - 0x0010))
@@ -110,9 +114,9 @@ __log2 (double x)
   /* r ~= z/c - 1, |r| < 1/(2*N).  */
 #ifdef __FP_FAST_FMA
   /* rounding error: 0x1p-55/N.  */
-  r = __builtin_fma (z, invc, -1.0);
+  r = fma (z, invc, -1.0);
   t1 = r * InvLn2hi;
-  t2 = r * InvLn2lo + __builtin_fma (r, InvLn2hi, -t1);
+  t2 = r * InvLn2lo + fma (r, InvLn2hi, -t1);
 #else
   double rhi, rlo;
   /* rounding error: 0x1p-55/N + 0x1p-65.  */
@@ -132,10 +136,12 @@ __log2 (double x)
   /* Evaluation is optimized assuming superscalar pipelined execution.  */
   r2 = r * r; /* rounding error: 0x1p-54/N^2.  */
   r4 = r2 * r2;
+#if LOG2_POLY_ORDER == 7
   /* Worst-case error if |y| > 0x1p-4: 0.547 ULP (0.550 ULP without fma).
      ~ 0.5 + 2/N/ln2 + abs-poly-error*0x1p56 ULP (+ 0.003 ULP without fma).  */
   p = A[0] + r * A[1] + r2 * (A[2] + r * A[3]) + r4 * (A[4] + r * A[5]);
   y = lo + r2 * p + hi;
+#endif
   return y;
 }
 #ifndef __log2
diff --git a/sysdeps/ieee754/dbl-64/math_config.h b/sysdeps/ieee754/dbl-64/math_config.h
index ea92cf313a..a3506d77eb 100644
--- a/sysdeps/ieee754/dbl-64/math_config.h
+++ b/sysdeps/ieee754/dbl-64/math_config.h
@@ -287,12 +287,11 @@ extern const struct log2_data
   double invln2lo;
   double poly[LOG2_POLY_ORDER - 1];
   double poly1[LOG2_POLY1_ORDER - 1];
-  /* See e_log2_data.c for details.  */
   struct {double invc, logc;} tab[1 << LOG2_TABLE_BITS];
 #ifndef __FP_FAST_FMA
   struct {double chi, clo;} tab2[1 << LOG2_TABLE_BITS];
 #endif
-} __log2_data attribute_hidden;
+} __log2_data __attribute__ ((__aligned__ (16))) attribute_hidden;
 
 #define POW_LOG_TABLE_BITS 7
 #define POW_LOG_POLY_ORDER 8
-- 
2.43.0