Use generic sinf/cosf in lgammaf_r
Commit Message
The internal functions __kernel_sinf and __kernel_cosf are used only by
lgammaf_r. Removing the internal functions and using the generic sinf
and cosf is better overall. Benchmarking on Cortex-A72 shows the generic
sinf and cosf are 1.4x and 2.3x faster in the range |x| < PI/4, and 0.66x
and 1.1x for |x| < PI/2, so it should make lgammaf_r faster on average.
GLIBC regression tests pass on AArch64.
ChangeLog:
2018-08-14 Wilco Dijkstra <wdijkstr@arm.com>
* sysdeps/ieee754/flt-32/e_lgammaf_r.c (sin_pif): Use __sinf/__cosf.
* sysdeps/ieee754/flt-32/k_cosf.c (__kernel_cosf): Remove all code.
* sysdeps/ieee754/flt-32/k_sinf.c (__kernel_sinf): Likewise.
--
Comments
On Tue, 14 Aug 2018, Wilco Dijkstra wrote:
> The internal functions __kernel_sinf and __kernel_cosf are used only by
> lgammaf_r. Removing the internal functions and using the generic sinf
> and cosf is better overall. Benchmarking on Cortex-A72 shows the generic
> sinf and cosf are 1.4x and 2.3x faster in the range |x| < PI/4, and 0.66x
> and 1.1x for |x| < PI/2, so it should make lgammaf_r faster on average.
I'd expect the sysdeps/powerpc/fpu versions of k_cosf.c and k_sinf.c to be
removed as well, since this patch makes them useless.
Joseph Myers wrote:
> I'd expect the sysdeps/powerpc/fpu versions of k_cosf.c and k_sinf.c to be
> removed as well, since this patch makes them useless.
It seems sysdeps/powerpc/fpu/s_sinf.c still needs that __kernel_sinf.
It would be nice if powerpc moved to the generic sinf/cosf instead (this may
need the PREFER_FLOAT_COMPARISON feature added again).
It also seems there is only one use left of __ieee754_rem_pio2f.
tanf could use reduce_large instead, allowing removal of 400 lines of
horribly complicated code...
Wilco
On Wed, 15 Aug 2018, Wilco Dijkstra wrote:
> Joseph Myers wrote:
>
> > I'd expect the sysdeps/powerpc/fpu versions of k_cosf.c and k_sinf.c to be
> > removed as well, since this patch makes them useless.
>
> It seems sysdeps/powerpc/fpu/s_sinf.c still needs that __kernel_sinf.
> It would be nice if powerpc moved to the generic sinf/cosf instead (this may
> need the PREFER_FLOAT_COMPARISON feature added again).
Thanks. In that case the patch is OK as-is, but I strongly encourage
powerpc maintainers to look at eliminating these powerpc/fpu function
implementations after doing anything necessary to make the generic ones at
least as fast on powerpc.
Joseph Myers <joseph@codesourcery.com> writes:
> On Wed, 15 Aug 2018, Wilco Dijkstra wrote:
>
>> Joseph Myers wrote:
>>
>> > I'd expect the sysdeps/powerpc/fpu versions of k_cosf.c and k_sinf.c to be
>> > removed as well, since this patch makes them useless.
>>
>> It seems sysdeps/powerpc/fpu/s_sinf.c still needs that __kernel_sinf.
>> It would be nice if powerpc moved to the generic sinf/cosf instead (this may
>> need the PREFER_FLOAT_COMPARISON feature added again).
>
> Thanks. In that case the patch is OK as-is, but I strongly encourage
> powerpc maintainers to look at eliminating these powerpc/fpu function
> implementations after doing anything necessary to make the generic ones at
> least as fast on powerpc.
Ack!
Thanks for the patch and the reminder!
@@ -97,7 +97,7 @@ sin_pif(float x)
GET_FLOAT_WORD(ix,x);
ix &= 0x7fffffff;
- if(ix<0x3e800000) return __kernel_sinf(pi*x,zero,0);
+ if(ix<0x3e800000) return __sinf (pi*x);
y = -x; /* x is assume negative */
/*
@@ -121,14 +121,14 @@ sin_pif(float x)
}
}
switch (n) {
- case 0: y = __kernel_sinf(pi*y,zero,0); break;
+ case 0: y = __sinf (pi*y); break;
case 1:
- case 2: y = __kernel_cosf(pi*((float)0.5-y),zero); break;
+ case 2: y = __cosf (pi*((float)0.5-y)); break;
case 3:
- case 4: y = __kernel_sinf(pi*(one-y),zero,0); break;
+ case 4: y = __sinf (pi*(one-y)); break;
case 5:
- case 6: y = -__kernel_cosf(pi*(y-(float)1.5),zero); break;
- default: y = __kernel_sinf(pi*(y-(float)2.0),zero,0); break;
+ case 6: y = -__cosf (pi*(y-(float)1.5)); break;
+ default: y = __sinf (pi*(y-(float)2.0)); break;
}
return -y;
}
@@ -1,55 +1 @@
-/* k_cosf.c -- float version of k_cos.c
- * Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
- */
-
-/*
- * ====================================================
- * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
- *
- * Developed at SunPro, a Sun Microsystems, Inc. business.
- * Permission to use, copy, modify, and distribute this
- * software is freely granted, provided that this notice
- * is preserved.
- * ====================================================
- */
-
-#if defined(LIBM_SCCS) && !defined(lint)
-static char rcsid[] = "$NetBSD: k_cosf.c,v 1.4 1995/05/10 20:46:23 jtc Exp $";
-#endif
-
-#include <math.h>
-#include <math_private.h>
-
-static const float
-one = 1.0000000000e+00, /* 0x3f800000 */
-C1 = 4.1666667908e-02, /* 0x3d2aaaab */
-C2 = -1.3888889225e-03, /* 0xbab60b61 */
-C3 = 2.4801587642e-05, /* 0x37d00d01 */
-C4 = -2.7557314297e-07, /* 0xb493f27c */
-C5 = 2.0875723372e-09, /* 0x310f74f6 */
-C6 = -1.1359647598e-11; /* 0xad47d74e */
-
-float __kernel_cosf(float x, float y)
-{
- float a,hz,z,r,qx;
- int32_t ix;
- GET_FLOAT_WORD(ix,x);
- ix &= 0x7fffffff; /* ix = |x|'s high word*/
- if(ix<0x32000000) { /* if x < 2**27 */
- if(((int)x)==0) return one; /* generate inexact */
- }
- z = x*x;
- r = z*(C1+z*(C2+z*(C3+z*(C4+z*(C5+z*C6)))));
- if(ix < 0x3e99999a) /* if |x| < 0.3 */
- return one - ((float)0.5*z - (z*r - x*y));
- else {
- if(ix > 0x3f480000) { /* x > 0.78125 */
- qx = (float)0.28125;
- } else {
- SET_FLOAT_WORD(qx,ix-0x01000000); /* x/4 */
- }
- hz = (float)0.5*z-qx;
- a = one-qx;
- return a - (hz - (z*r-x*y));
- }
-}
+/* Not needed. */
@@ -1,51 +1 @@
-/* k_sinf.c -- float version of k_sin.c
- * Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
- */
-
-/*
- * ====================================================
- * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
- *
- * Developed at SunPro, a Sun Microsystems, Inc. business.
- * Permission to use, copy, modify, and distribute this
- * software is freely granted, provided that this notice
- * is preserved.
- * ====================================================
- */
-
-#if defined(LIBM_SCCS) && !defined(lint)
-static char rcsid[] = "$NetBSD: k_sinf.c,v 1.4 1995/05/10 20:46:33 jtc Exp $";
-#endif
-
-#include <float.h>
-#include <math.h>
-#include <math_private.h>
-#include <math-underflow.h>
-
-static const float
-half = 5.0000000000e-01,/* 0x3f000000 */
-S1 = -1.6666667163e-01, /* 0xbe2aaaab */
-S2 = 8.3333337680e-03, /* 0x3c088889 */
-S3 = -1.9841270114e-04, /* 0xb9500d01 */
-S4 = 2.7557314297e-06, /* 0x3638ef1b */
-S5 = -2.5050759689e-08, /* 0xb2d72f34 */
-S6 = 1.5896910177e-10; /* 0x2f2ec9d3 */
-
-float __kernel_sinf(float x, float y, int iy)
-{
- float z,r,v;
- int32_t ix;
- GET_FLOAT_WORD(ix,x);
- ix &= 0x7fffffff; /* high word of x */
- if(ix<0x32000000) /* |x| < 2**-27 */
- {
- math_check_force_underflow (x);
- if ((int) x == 0)
- return x; /* generate inexact */
- }
- z = x*x;
- v = z*x;
- r = S2+z*(S3+z*(S4+z*(S5+z*S6)));
- if(iy==0) return x+v*(S1+z*r);
- else return x-((z*(half*y-v*r)-y)-v*S1);
-}
+/* Not needed. */