|
 |
00db10 |
Backport of:
|
|
|
00db10 |
commit 487972aea52004f604c2878c8c9d3e77670f2c32
|
|
|
00db10 |
Author: Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
Date: Thu Feb 27 09:43:51 2014 -0600
|
|
|
00db10 |
|
|
|
00db10 |
PowerPC: Optimized isnan/isnanf for POWER8
|
|
|
00db10 |
|
|
|
00db10 |
This patch add a optimized isnan/isnanf implementation for POWER8
|
|
|
00db10 |
using the new Move From VSR Doubleword instruction to gains some
|
|
|
00db10 |
cycles from FP to GRP register move.
|
|
|
00db10 |
|
|
|
00db10 |
ChangeLog:
|
|
|
00db10 |
2014-02-27 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
|
|
|
00db10 |
* sysdeps/powerpc/powerpc32/power4/multiarch/init-arch.h (INIT_ARCH):
|
|
|
00db10 |
Add hwcap2 initialization.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile: Add isnan power8
|
|
|
00db10 |
implementation.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power8.S: New file:
|
|
|
00db10 |
POWER8 isnan ifunc implementation.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c (__isnan): Add
|
|
|
00db10 |
POWER8 implementation.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnanf.c (__isnanf):
|
|
|
00db10 |
Likewise.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/power8/fpu/s_isnan.S: New file:
|
|
|
00db10 |
POWER8 isnan implementation.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/power8/fpu/s_isnanf.S: New file.
|
|
|
00db10 |
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc32/power4/multiarch/init-arch.h b/sysdeps/powerpc/powerpc32/power4/multiarch/init-arch.h
|
|
|
00db10 |
index 51a34f2..72d720d 100644
|
|
|
00db10 |
--- a/sysdeps/powerpc/powerpc32/power4/multiarch/init-arch.h
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc32/power4/multiarch/init-arch.h
|
|
|
00db10 |
@@ -36,6 +36,7 @@
|
|
|
00db10 |
and fills the previous ones. */
|
|
|
00db10 |
#define INIT_ARCH() \
|
|
|
00db10 |
unsigned long int hwcap = __GLRO(dl_hwcap); \
|
|
|
00db10 |
+ unsigned long int __attribute__((unused)) hwcap2 = __GLRO(dl_hwcap2); \
|
|
|
00db10 |
if (hwcap & PPC_FEATURE_ARCH_2_06) \
|
|
|
00db10 |
hwcap |= PPC_FEATURE_ARCH_2_05 | \
|
|
|
00db10 |
PPC_FEATURE_POWER5_PLUS | \
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile b/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
|
|
|
00db10 |
index 1e04f21..4cd1c5d 100644
|
|
|
00db10 |
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
|
|
|
00db10 |
@@ -4,7 +4,7 @@ sysdep_routines += s_isnan-power7 s_isnan-power6x s_isnan-power6 \
|
|
|
00db10 |
s_copysign-ppc64 s_finite-power7 s_finite-ppc64 \
|
|
|
00db10 |
s_finitef-ppc64 s_isinff-ppc64 s_isinf-power7 \
|
|
|
00db10 |
s_isinf-ppc64 s_modf-power5+ s_modf-ppc64 \
|
|
|
00db10 |
- s_modff-power5+ s_modff-ppc64
|
|
|
00db10 |
+ s_modff-power5+ s_modff-ppc64 s_isnan-power8
|
|
|
00db10 |
|
|
|
00db10 |
libm-sysdep_routines += s_isnan-power7 s_isnan-power6x s_isnan-power6 \
|
|
|
00db10 |
s_isnan-power5 s_isnan-ppc64 s_llround-power6x \
|
|
|
00db10 |
@@ -21,7 +21,8 @@ libm-sysdep_routines += s_isnan-power7 s_isnan-power6x s_isnan-power6 \
|
|
|
00db10 |
s_logbl-power7 s_logb-ppc64 s_logbf-ppc64 \
|
|
|
00db10 |
s_logbl-ppc64 s_modf-power5+ s_modf-ppc64 \
|
|
|
00db10 |
s_modff-power5+ s_modff-ppc64 e_hypot-ppc64 \
|
|
|
00db10 |
- e_hypot-power7 e_hypotf-ppc64 e_hypotf-power7
|
|
|
00db10 |
+ e_hypot-power7 e_hypotf-ppc64 e_hypotf-power7 \
|
|
|
00db10 |
+ s_isnan-power8
|
|
|
00db10 |
|
|
|
00db10 |
CFLAGS-s_logbf-power7.c = -mcpu=power7
|
|
|
00db10 |
CFLAGS-s_logbl-power7.c = -mcpu=power7
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power8.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power8.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..c176d5a
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan-power8.S
|
|
|
00db10 |
@@ -0,0 +1,33 @@
|
|
|
00db10 |
+/* isnan(). PowerPC64/POWER7 version.
|
|
|
00db10 |
+ Copyright (C) 2014 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+#include <math_ldbl_opt.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef hidden_def
|
|
|
00db10 |
+#define hidden_def(name)
|
|
|
00db10 |
+#undef weak_alias
|
|
|
00db10 |
+#define weak_alias(name, alias)
|
|
|
00db10 |
+#undef strong_alias
|
|
|
00db10 |
+#define strong_alias(name, alias)
|
|
|
00db10 |
+#undef compat_symbol
|
|
|
00db10 |
+#define compat_symbol(lib, name, symbol, ver)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#define __isnan __isnan_power8
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdeps/powerpc/powerpc64/power8/fpu/s_isnan.S>
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c
|
|
|
00db10 |
index 0de833e..65a5ca0 100644
|
|
|
00db10 |
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c
|
|
|
00db10 |
@@ -26,16 +26,19 @@ extern __typeof (__isnan) __isnan_power5 attribute_hidden;
|
|
|
00db10 |
extern __typeof (__isnan) __isnan_power6 attribute_hidden;
|
|
|
00db10 |
extern __typeof (__isnan) __isnan_power6x attribute_hidden;
|
|
|
00db10 |
extern __typeof (__isnan) __isnan_power7 attribute_hidden;
|
|
|
00db10 |
+extern __typeof (__isnan) __isnan_power8 attribute_hidden;
|
|
|
00db10 |
|
|
|
00db10 |
libc_ifunc (__isnan,
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_ARCH_2_06)
|
|
|
00db10 |
- ? __isnan_power7 :
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_POWER6_EXT)
|
|
|
00db10 |
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07)
|
|
|
00db10 |
+ ? __isnan_power8 :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_ARCH_2_06)
|
|
|
00db10 |
+ ? __isnan_power7 :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_POWER6_EXT)
|
|
|
00db10 |
? __isnan_power6x :
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_ARCH_2_05)
|
|
|
00db10 |
- ? __isnan_power6 :
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_POWER5)
|
|
|
00db10 |
- ? __isnan_power5
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_ARCH_2_05)
|
|
|
00db10 |
+ ? __isnan_power6 :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_POWER5)
|
|
|
00db10 |
+ ? __isnan_power5
|
|
|
00db10 |
: __isnan_ppc64);
|
|
|
00db10 |
|
|
|
00db10 |
weak_alias (__isnan, isnan)
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnanf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnanf.c
|
|
|
00db10 |
index b237455..eb68a50 100644
|
|
|
00db10 |
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnanf.c
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnanf.c
|
|
|
00db10 |
@@ -25,16 +25,19 @@ extern __typeof (__isnanf) __isnan_power5 attribute_hidden;
|
|
|
00db10 |
extern __typeof (__isnanf) __isnan_power6 attribute_hidden;
|
|
|
00db10 |
extern __typeof (__isnanf) __isnan_power6x attribute_hidden;
|
|
|
00db10 |
extern __typeof (__isnanf) __isnan_power7 attribute_hidden;
|
|
|
00db10 |
+extern __typeof (__isnanf) __isnan_power8 attribute_hidden;
|
|
|
00db10 |
|
|
|
00db10 |
libc_ifunc (__isnanf,
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_ARCH_2_06)
|
|
|
00db10 |
- ? __isnan_power7 :
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_POWER6_EXT)
|
|
|
00db10 |
- ? __isnan_power6x :
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_ARCH_2_05)
|
|
|
00db10 |
- ? __isnan_power6 :
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_POWER5)
|
|
|
00db10 |
- ? __isnan_power5
|
|
|
00db10 |
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07)
|
|
|
00db10 |
+ ? __isnan_power8 :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_ARCH_2_06)
|
|
|
00db10 |
+ ? __isnan_power7 :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_POWER6_EXT)
|
|
|
00db10 |
+ ? __isnan_power6x :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_ARCH_2_05)
|
|
|
00db10 |
+ ? __isnan_power6 :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_POWER5)
|
|
|
00db10 |
+ ? __isnan_power5
|
|
|
00db10 |
: __isnan_ppc64);
|
|
|
00db10 |
|
|
|
00db10 |
weak_alias (__isnanf, isnanf)
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/power8/fpu/s_isnan.S b/sysdeps/powerpc/powerpc64/power8/fpu/s_isnan.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..c1ca9a5
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/power8/fpu/s_isnan.S
|
|
|
00db10 |
@@ -0,0 +1,53 @@
|
|
|
00db10 |
+/* isnan(). PowerPC64/POWER8 version.
|
|
|
00db10 |
+ Copyright (C) 2014 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+#include <math_ldbl_opt.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#define MFVSRD_R3_V1 .byte 0x7c,0x23,0x00,0x66 /* mfvsrd r3,vs1 */
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* int [r3] __isnan([f1] x) */
|
|
|
00db10 |
+
|
|
|
00db10 |
+EALIGN (__isnan, 4, 0)
|
|
|
00db10 |
+ CALL_MCOUNT 0
|
|
|
00db10 |
+ MFVSRD_R3_V1
|
|
|
00db10 |
+ lis r9,0x7ff0
|
|
|
00db10 |
+ clrldi r3,r3,1 /* r3 = r3 & 0x8000000000000000 */
|
|
|
00db10 |
+ rldicr r9,r9,32,31 /* r9 = (r9 << 32) & 0xffffffff */
|
|
|
00db10 |
+ subf r3,r3,r9
|
|
|
00db10 |
+ rldicl r3,r3,1,63
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
+END (__isnan)
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* It turns out that the 'double' version will also always work for
|
|
|
00db10 |
+ single-precision. */
|
|
|
00db10 |
+strong_alias (__isnan, __isnanf)
|
|
|
00db10 |
+hidden_def (__isnanf)
|
|
|
00db10 |
+weak_alias (__isnanf, isnanf)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#ifdef NO_LONG_DOUBLE
|
|
|
00db10 |
+strong_alias (__isnan, __isnanl)
|
|
|
00db10 |
+weak_alias (__isnan, isnanl)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+
|
|
|
00db10 |
+#ifndef IS_IN_libm
|
|
|
00db10 |
+# if LONG_DOUBLE_COMPAT(libc, GLIBC_2_0)
|
|
|
00db10 |
+compat_symbol (libc, __isnan, __isnanl, GLIBC_2_0);
|
|
|
00db10 |
+compat_symbol (libc, isnan, isnanl, GLIBC_2_0);
|
|
|
00db10 |
+# endif
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/power8/fpu/s_isnanf.S b/sysdeps/powerpc/powerpc64/power8/fpu/s_isnanf.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..b48c85e
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/power8/fpu/s_isnanf.S
|
|
|
00db10 |
@@ -0,0 +1 @@
|
|
|
00db10 |
+/* This function uses the same code as s_isnan.S. */
|