Mess up the "kernel" float trig function .c files with ifdefs so that

they can be #included in other .c files to give inline functions, and
use them to inline the functions in most callers (not in e_lgammaf_r.c).
__kernel_tanf() is too large and complicated for gcc to inline very well.

An athlons, this gives a speed increase under favourable pipeline
conditions of about 10% overall (larger for AXP, smaller for A64).
E.g., on AXP, sinf() on uniformly distributed args in [-2Pi, 2Pi]
now takes 30-56 cycles; it used to take 45-61 cycles; hardware fsin
takes 65-129.

lib/msun/src/k_cosf.c

@@ -14,9 +14,11 @@
  * ====================================================
  */
 
+#ifndef INLINE_KERNEL_COSF
 #ifndef lint
 static char rcsid[] = "$FreeBSD$";
 #endif
+#endif
 
 #include "math.h"
 #include "math_private.h"
@@ -28,6 +30,9 @@ C1  =  0xaaaaa5.0p-28,		/*  0.041666645557 */
 C2  = -0xb60615.0p-33,		/* -0.0013887310633 */
 C3  =  0xccf47d.0p-39;		/*  0.000024432542887 */
 
+#ifdef INLINE_KERNEL_COSF
+extern inline
+#endif
 float
 __kernel_cosf(float x, float y)
 {

lib/msun/src/k_sinf.c

@@ -14,9 +14,11 @@
  * ====================================================
  */
 
+#ifndef INLINE_KERNEL_SINF
 #ifndef lint
 static char rcsid[] = "$FreeBSD$";
 #endif
+#endif
 
 #include "math.h"
 #include "math_private.h"
@@ -29,6 +31,9 @@ S2  =  0x8888bb.0p-30,		/*  0.0083333803341 */
 S3  = -0xd02de1.0p-36,		/* -0.00019853517006 */
 S4  =  0xbe6dbe.0p-42;		/*  0.0000028376084629 */
 
+#ifdef INLINE_KERNEL_SINF
+extern inline
+#endif
 float
 __kernel_sinf(float x, float y, int iy)
 {

lib/msun/src/k_tanf.c

@@ -12,9 +12,11 @@
  * ====================================================
  */
 
+#ifndef INLINE_KERNEL_TANF
 #ifndef lint
 static char rcsid[] = "$FreeBSD$";
 #endif
+#endif
 
 #include "math.h"
 #include "math_private.h"
@@ -32,6 +34,9 @@ T[] =  {
   0xeaf97e.0p-31,		/* 0.0071708550677 */
 };
 
+#ifdef INLINE_KERNEL_TANF
+extern inline
+#endif
 float
 __kernel_tanf(float x, float y, int iy)
 {

lib/msun/src/s_cosf.c

@@ -18,7 +18,11 @@ static char rcsid[] = "$FreeBSD$";
 #endif
 
 #include "math.h"
+#define	INLINE_KERNEL_COSF
+#define	INLINE_KERNEL_SINF
 #include "math_private.h"
+#include "k_cosf.c"
+#include "k_sinf.c"
 
 /* Small multiples of pi/2 rounded to double precision. */
 static const double

lib/msun/src/s_sinf.c

@@ -18,7 +18,11 @@ static char rcsid[] = "$FreeBSD$";
 #endif
 
 #include "math.h"
+#define	INLINE_KERNEL_COSF
+#define	INLINE_KERNEL_SINF
 #include "math_private.h"
+#include "k_cosf.c"
+#include "k_sinf.c"
 
 /* Small multiples of pi/2 rounded to double precision. */
 static const double

lib/msun/src/s_tanf.c

@@ -18,7 +18,9 @@ static char rcsid[] = "$FreeBSD$";
 #endif
 
 #include "math.h"
+#define	INLINE_KERNEL_TANF
 #include "math_private.h"
+#include "k_tanf.c"
 
 /* Small multiples of pi/2 rounded to double precision. */
 static const double