|
 |
588e70 |
commit 56c81132ccc6f468fa4fc29c536db060e18e9d87
|
|
|
588e70 |
Author: Raphael Moreira Zinsly <rzinsly@linux.ibm.com>
|
|
|
588e70 |
Date: Tue Feb 23 14:14:37 2021 -0300
|
|
|
588e70 |
|
|
|
588e70 |
powerpc: Add optimized ilogb* for POWER9
|
|
|
588e70 |
|
|
|
588e70 |
The instructions xsxexpdp and xsxexpqp introduced on POWER9 extract
|
|
|
588e70 |
the exponent from a double-precision and quad-precision floating-point
|
|
|
588e70 |
respectively, thus they can be used to improve ilogb, ilogbf and ilogbf128.
|
|
|
588e70 |
|
|
|
588e70 |
diff --git a/sysdeps/powerpc/fpu/math_private.h b/sysdeps/powerpc/fpu/math_private.h
|
|
|
588e70 |
index e642d6c8237578ea..5bbc468829062a48 100644
|
|
|
588e70 |
--- a/sysdeps/powerpc/fpu/math_private.h
|
|
|
588e70 |
+++ b/sysdeps/powerpc/fpu/math_private.h
|
|
|
588e70 |
@@ -26,7 +26,28 @@
|
|
|
588e70 |
|
|
|
588e70 |
#include_next <math_private.h>
|
|
|
588e70 |
|
|
|
588e70 |
-#if defined _ARCH_PWR9 && __HAVE_DISTINCT_FLOAT128
|
|
|
588e70 |
+#ifdef _ARCH_PWR9
|
|
|
588e70 |
+
|
|
|
588e70 |
+#if __GNUC_PREREQ (8, 0)
|
|
|
588e70 |
+# define _GL_HAS_BUILTIN_ILOGB 1
|
|
|
588e70 |
+#elif defined __has_builtin
|
|
|
588e70 |
+# define _GL_HAS_BUILTIN_ILOGB __has_builtin (__builtin_vsx_scalar_extract_exp)
|
|
|
588e70 |
+#else
|
|
|
588e70 |
+# define _GL_HAS_BUILTIN_ILOGB 0
|
|
|
588e70 |
+#endif
|
|
|
588e70 |
+
|
|
|
588e70 |
+#define __builtin_test_dc_ilogbf __builtin_test_dc_ilogb
|
|
|
588e70 |
+#define __builtin_ilogbf __builtin_ilogb
|
|
|
588e70 |
+
|
|
|
588e70 |
+#define __builtin_test_dc_ilogb(x, y) \
|
|
|
588e70 |
+ __builtin_vsx_scalar_test_data_class_dp(x, y)
|
|
|
588e70 |
+#define __builtin_ilogb(x) __builtin_vsx_scalar_extract_exp(x) - 0x3ff
|
|
|
588e70 |
+
|
|
|
588e70 |
+#define __builtin_test_dc_ilogbf128(x, y) \
|
|
|
588e70 |
+ __builtin_vsx_scalar_test_data_class_qp(x, y)
|
|
|
588e70 |
+#define __builtin_ilogbf128(x) __builtin_vsx_scalar_extract_expq(x) - 0x3fff
|
|
|
588e70 |
+
|
|
|
588e70 |
+#if __HAVE_DISTINCT_FLOAT128
|
|
|
588e70 |
extern __always_inline _Float128
|
|
|
588e70 |
__ieee754_sqrtf128 (_Float128 __x)
|
|
|
588e70 |
{
|
|
|
588e70 |
@@ -35,6 +56,9 @@ __ieee754_sqrtf128 (_Float128 __x)
|
|
|
588e70 |
return __z;
|
|
|
588e70 |
}
|
|
|
588e70 |
#endif
|
|
|
588e70 |
+#else /* !_ARCH_PWR9 */
|
|
|
588e70 |
+#define _GL_HAS_BUILTIN_ILOGB 0
|
|
|
588e70 |
+#endif
|
|
|
588e70 |
|
|
|
588e70 |
#if defined _ARCH_PWR5X
|
|
|
588e70 |
|
|
|
588e70 |
diff --git a/sysdeps/powerpc/powerpc64/le/fpu/w_ilogb_template.c b/sysdeps/powerpc/powerpc64/le/fpu/w_ilogb_template.c
|
|
|
588e70 |
new file mode 100644
|
|
|
588e70 |
index 0000000000000000..b5c1c0aa9db86f3d
|
|
|
588e70 |
--- /dev/null
|
|
|
588e70 |
+++ b/sysdeps/powerpc/powerpc64/le/fpu/w_ilogb_template.c
|
|
|
588e70 |
@@ -0,0 +1,30 @@
|
|
|
588e70 |
+#include <math.h>
|
|
|
588e70 |
+#include <errno.h>
|
|
|
588e70 |
+#include <limits.h>
|
|
|
588e70 |
+#include <math_private.h>
|
|
|
588e70 |
+#include <fenv.h>
|
|
|
588e70 |
+
|
|
|
588e70 |
+#if _GL_HAS_BUILTIN_ILOGB
|
|
|
588e70 |
+int
|
|
|
588e70 |
+M_DECL_FUNC (__ilogb) (FLOAT x)
|
|
|
588e70 |
+{
|
|
|
588e70 |
+ int r;
|
|
|
588e70 |
+ /* Check for exceptional cases. */
|
|
|
588e70 |
+ if (! M_SUF(__builtin_test_dc_ilogb) (x, 0x7f))
|
|
|
588e70 |
+ r = M_SUF (__builtin_ilogb) (x);
|
|
|
588e70 |
+ else
|
|
|
588e70 |
+ /* Fallback to the generic ilogb if x is NaN, Inf or subnormal. */
|
|
|
588e70 |
+ r = M_SUF (__ieee754_ilogb) (x);
|
|
|
588e70 |
+ if (__builtin_expect (r == FP_ILOGB0, 0)
|
|
|
588e70 |
+ || __builtin_expect (r == FP_ILOGBNAN, 0)
|
|
|
588e70 |
+ || __builtin_expect (r == INT_MAX, 0))
|
|
|
588e70 |
+ {
|
|
|
588e70 |
+ __set_errno (EDOM);
|
|
|
588e70 |
+ __feraiseexcept (FE_INVALID);
|
|
|
588e70 |
+ }
|
|
|
588e70 |
+ return r;
|
|
|
588e70 |
+}
|
|
|
588e70 |
+declare_mgen_alias (__ilogb, ilogb)
|
|
|
588e70 |
+#else
|
|
|
588e70 |
+#include <math/w_ilogb_template.c>
|
|
|
588e70 |
+#endif
|
|
|
588e70 |
diff --git a/sysdeps/powerpc/powerpc64/le/fpu/w_ilogbl.c b/sysdeps/powerpc/powerpc64/le/fpu/w_ilogbl.c
|
|
|
588e70 |
new file mode 100644
|
|
|
588e70 |
index 0000000000000000..205f154f0089a269
|
|
|
588e70 |
--- /dev/null
|
|
|
588e70 |
+++ b/sysdeps/powerpc/powerpc64/le/fpu/w_ilogbl.c
|
|
|
588e70 |
@@ -0,0 +1,4 @@
|
|
|
588e70 |
+/* Skip the optimization for long double as ibm128 does not provide an
|
|
|
588e70 |
+ optimized builtin. */
|
|
|
588e70 |
+#include <math-type-macros-ldouble.h>
|
|
|
588e70 |
+#include <math/w_ilogb_template.c>
|