Move __HAVE_FAST_FMA to math_config.h

Define it consistently with other HAVE_* macros that only affect code
using math_config.h.  This is also closer to the Arm Optimized Routines
code.

newlib/libc/include/machine/ieeefp.h

@@ -65,17 +65,6 @@
 	double and single precision arithmetics has similar latency and it
 	has no legacy SVID matherr support, only POSIX errno and fenv
 	exception based error handling.
-
-   __HAVE_FAST_FMA_DEFAULT
-
-	Default value for __HAVE_FAST_FMA if that's not set by the user.
-	It should be set here based on predefined feature macros.
-
-   __HAVE_FAST_FMA
-
-	It should be set to 1 if the compiler can inline an fma call as a
-	single instruction.  Some math code has a separate faster code
-	path assuming the target has single instruction fma.
 */
 
 #if (defined(__arm__) || defined(__thumb__)) && !defined(__MAVERICK__)
@@ -91,9 +80,6 @@
 # endif
 # if __ARM_FP & 0x8
 #  define __OBSOLETE_MATH_DEFAULT 0
-#  if __ARM_FEATURE_FMA
-#   define __HAVE_FAST_FMA_DEFAULT 1
-#  endif
 # endif
 #else
 # define __IEEE_BIG_ENDIAN
@@ -110,7 +96,6 @@
 #define __IEEE_BIG_ENDIAN
 #endif
 #define __OBSOLETE_MATH_DEFAULT 0
-#define __HAVE_FAST_FMA_DEFAULT 1
 #endif
 
 #ifdef __epiphany__
@@ -479,14 +464,6 @@
 #define __OBSOLETE_MATH __OBSOLETE_MATH_DEFAULT
 #endif
 
-#ifndef __HAVE_FAST_FMA_DEFAULT
-/* Assume slow fma by default.  */
-#define __HAVE_FAST_FMA_DEFAULT 0
-#endif
-#ifndef __HAVE_FAST_FMA
-#define __HAVE_FAST_FMA __HAVE_FAST_FMA_DEFAULT
-#endif
-
 #ifndef __IEEE_BIG_ENDIAN
 #ifndef __IEEE_LITTLE_ENDIAN
 #error Endianess not declared!!

newlib/libm/common/log.c

@@ -146,7 +146,7 @@ log (double x)
 
   /* log(x) = log1p(z/c-1) + log(c) + k*Ln2.  */
   /* r ~= z/c - 1, |r| < 1/(2*N).  */
-#if __HAVE_FAST_FMA
+#if HAVE_FAST_FMA
   /* rounding error: 0x1p-55/N.  */
   r = fma (z, invc, -1.0);
 #else

newlib/libm/common/log2.c

@@ -72,7 +72,7 @@ double
       if (WANT_ROUNDING && unlikely (ix == asuint64 (1.0)))
 	return 0;
       r = x - 1.0;
-#if __HAVE_FAST_FMA
+#if HAVE_FAST_FMA
       hi = r * InvLn2hi;
       lo = r * InvLn2lo + fma (r, InvLn2hi, -hi);
 #else
@@ -123,7 +123,7 @@ double
 
   /* log2(x) = log2(z/c) + log2(c) + k.  */
   /* r ~= z/c - 1, |r| < 1/(2*N).  */
-#if __HAVE_FAST_FMA
+#if HAVE_FAST_FMA
   /* rounding error: 0x1p-55/N.  */
   r = fma (z, invc, -1.0);
   t1 = r * InvLn2hi;

newlib/libm/common/log2_data.c

@@ -134,7 +134,7 @@ const struct log2_data __log2_data = {
 {0x1.767dcf99eff8cp-1, 0x1.ce0a43dbf4000p-2},
 #endif
 },
-#if !__HAVE_FAST_FMA
+#if !HAVE_FAST_FMA
 .tab2 = {
 # if N == 64
 {0x1.6200012b90a8ep-1, 0x1.904ab0644b605p-55},
@@ -203,6 +203,6 @@ const struct log2_data __log2_data = {
 {0x1.5dfffebfc3481p+0, -0x1.180902e30e93ep-54},
 # endif
 },
-#endif /* !__HAVE_FAST_FMA */
+#endif /* !HAVE_FAST_FMA */
 };
 #endif /* __OBSOLETE_MATH */

newlib/libm/common/log_data.c

@@ -307,7 +307,7 @@ const struct log_data __log_data = {
 {0x1.756cadbd6130cp-1, 0x1.432eee32fe000p-2},
 #endif
 },
-#if !__HAVE_FAST_FMA
+#if !HAVE_FAST_FMA
 .tab2 = {
 # if N == 64
 {0x1.61ffff94c4fecp-1, -0x1.9fe4fc998f325p-56},
@@ -505,6 +505,6 @@ const struct log_data __log_data = {
 {0x1.5efffe7b87a89p+0, -0x1.47eb780ed6904p-54},
 #endif
 },
-#endif /* !__HAVE_FAST_FMA */
+#endif /* !HAVE_FAST_FMA */
 };
 #endif /* __OBSOLETE_MATH */

newlib/libm/common/math_config.h

@@ -61,6 +61,15 @@
 # endif
 #endif
 
+/* Compiler can inline fma as a single instruction.  */
+#ifndef HAVE_FAST_FMA
+# if __aarch64__ || __ARM_FEATURE_FMA
+#   define HAVE_FAST_FMA 1
+# else
+#   define HAVE_FAST_FMA 0
+# endif
+#endif
+
 #if HAVE_FAST_ROUND
 # define TOINT_INTRINSICS 1
 
@@ -366,7 +375,7 @@ extern const struct log_data
   double poly[LOG_POLY_ORDER - 1]; /* First coefficient is 1.  */
   double poly1[LOG_POLY1_ORDER - 1];
   struct {double invc, logc;} tab[1 << LOG_TABLE_BITS];
-#if !__HAVE_FAST_FMA
+#if !HAVE_FAST_FMA
   struct {double chi, clo;} tab2[1 << LOG_TABLE_BITS];
 #endif
 } __log_data HIDDEN;
@@ -381,7 +390,7 @@ extern const struct log2_data
   double poly[LOG2_POLY_ORDER - 1];
   double poly1[LOG2_POLY1_ORDER - 1];
   struct {double invc, logc;} tab[1 << LOG2_TABLE_BITS];
-#if !__HAVE_FAST_FMA
+#if !HAVE_FAST_FMA
   struct {double chi, clo;} tab2[1 << LOG2_TABLE_BITS];
 #endif
 } __log2_data HIDDEN;

newlib/libm/common/pow.c

@@ -80,7 +80,7 @@ log_inline (uint64_t ix, double_t *tail)
   logctail = T[i].logctail;
 
   /* r = z/c - 1, arranged to be exact.  */
-#if __HAVE_FAST_FMA
+#if HAVE_FAST_FMA
   r = fma (z, invc, -1.0);
 #else
   double_t zhi = asdouble (iz & (-1ULL << 32));
@@ -102,7 +102,7 @@ log_inline (uint64_t ix, double_t *tail)
   ar2 = r * ar;
   ar3 = r * ar2;
   /* k*Ln2 + log(c) + r + A[0]*r*r.  */
-#if __HAVE_FAST_FMA
+#if HAVE_FAST_FMA
   hi = t2 + ar2;
   lo3 = fma (ar, r, -ar2);
   lo4 = t2 - hi + ar2;
@@ -376,7 +376,7 @@ pow (double x, double y)
   double_t lo;
   double_t hi = log_inline (ix, &lo);
   double_t ehi, elo;
-#if __HAVE_FAST_FMA
+#if HAVE_FAST_FMA
   ehi = y * hi;
   elo = y * lo + fma (y, hi, -ehi);
 #else