aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/s390/wcpcpy-vx.S
diff options
context:
space:
mode:
authorStefan Liebler <stli@linux.ibm.com>2018-12-18 13:57:19 +0100
committerStefan Liebler <stli@linux.ibm.com>2018-12-18 13:57:19 +0100
commit0582e4284529b4ea0fcd1a8973ccab7d95ec0e87 (patch)
tree08dd0fbd46d8468091946bcfeb4e91ea1abdd50d /sysdeps/s390/wcpcpy-vx.S
parent804f2e5c73b1363836ce5db29a0abb3d36e1286a (diff)
downloadglibc-0582e4284529b4ea0fcd1a8973ccab7d95ec0e87.tar
glibc-0582e4284529b4ea0fcd1a8973ccab7d95ec0e87.tar.gz
glibc-0582e4284529b4ea0fcd1a8973ccab7d95ec0e87.tar.bz2
glibc-0582e4284529b4ea0fcd1a8973ccab7d95ec0e87.zip
S390: Refactor wcpcpy ifunc handling.
The ifunc handling for wcpcpy is adjusted in order to omit ifunc if the minimum architecture level already supports newer CPUs by default. Unfortunately the c ifunc variant can't be omitted at all as it is used by the z13 ifunc variant as fallback if the pointers are not 4-byte aligned. ChangeLog: * sysdeps/s390/multiarch/Makefile (sysdep_routines): Remove wcpcpy variants. * sysdeps/s390/Makefile (sysdep_routines): Add wcpcpy variants. * sysdeps/s390/multiarch/ifunc-impl-list.c (__libc_ifunc_impl_list): Refactor ifunc handling for wcpcpy. * sysdeps/s390/multiarch/wcpcpy-c.c: Move to ... * sysdeps/s390/wcpcpy-c.c: ... here and adjust ifunc handling. * sysdeps/s390/multiarch/wcpcpy-vx.S: Move to ... * sysdeps/s390/wcpcpy-vx.S: ... here and adjust ifunc handling. * sysdeps/s390/multiarch/wcpcpy.c: Move to ... * sysdeps/s390/wcpcpy.c: ... here and adjust ifunc handling. * sysdeps/s390/ifunc-wcpcpy.h: New file.
Diffstat (limited to 'sysdeps/s390/wcpcpy-vx.S')
-rw-r--r--sysdeps/s390/wcpcpy-vx.S120
1 files changed, 120 insertions, 0 deletions
diff --git a/sysdeps/s390/wcpcpy-vx.S b/sysdeps/s390/wcpcpy-vx.S
new file mode 100644
index 0000000000..5154ad4461
--- /dev/null
+++ b/sysdeps/s390/wcpcpy-vx.S
@@ -0,0 +1,120 @@
+/* Vector optimized 32/64 bit S/390 version of wcpcpy.
+ Copyright (C) 2015-2018 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <ifunc-wcpcpy.h>
+#if HAVE_WCPCPY_Z13
+
+# include "sysdep.h"
+# include "asm-syntax.h"
+
+ .text
+
+/* wchar_t * wcpcpy (const wchar_t *dest, const wchar_t *src)
+ Copy string src to dest returning a pointer to its end.
+
+ Register usage:
+ -r0=border-len for switching to vector-instructions
+ -r1=tmp
+ -r2=dest and return value
+ -r3=src
+ -r4=tmp
+ -r5=current_len
+ -v16=part of src
+ -v17=index of zero
+ -v18=part of src
+*/
+ENTRY(WCPCPY_Z13)
+ .machine "z13"
+ .machinemode "zarch_nohighgprs"
+
+ vlbb %v16,0(%r3),6 /* Load s until next 4k-byte boundary. */
+ lcbb %r1,0(%r3),6 /* Get bytes to 4k-byte boundary or 16. */
+
+ tmll %r3,3 /* Test if s is 4-byte aligned? */
+ jne .Lfallback /* And use common-code variant if not. */
+
+ vfenezf %v17,%v16,%v16 /* Find element not equal with zero search. */
+ vlgvb %r5,%v17,7 /* Load zero index or 16 if not found. */
+ clrjl %r5,%r1,.Lfound_align /* If found zero within loaded bytes,
+ copy bytes before and return. */
+
+ /* Align s to 16 byte. */
+ risbgn %r4,%r3,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */
+ lghi %r5,15 /* current_len = 15. */
+ slr %r5,%r4 /* Compute highest index to 16byte boundary. */
+
+ vstl %v16,%r5,0(%r2) /* Copy loaded characters - no zero. */
+ ahi %r5,1 /* Start loop at next character. */
+
+ /* Find zero in 16byte aligned loop. */
+.Lloop:
+ vl %v16,0(%r5,%r3) /* Load s. */
+ vfenezfs %v17,%v16,%v16 /* Find element not equal with zero search. */
+ je .Lfound_v16_0 /* Jump away if zero was found. */
+ vl %v18,16(%r5,%r3) /* Load next part of s. */
+ vst %v16,0(%r5,%r2) /* Store previous part without zero to dst. */
+ vfenezfs %v17,%v18,%v18
+ je .Lfound_v18_16
+ vl %v16,32(%r5,%r3)
+ vst %v18,16(%r5,%r2)
+ vfenezfs %v17,%v16,%v16
+ je .Lfound_v16_32
+ vl %v18,48(%r5,%r3)
+ vst %v16,32(%r5,%r2)
+ vfenezfs %v17,%v18,%v18
+ je .Lfound_v18_48
+ vst %v18,48(%r5,%r2)
+
+ aghi %r5,64
+ j .Lloop /* No zero found -> loop. */
+
+.Lfound_v16_32:
+ aghi %r5,32
+.Lfound_v16_0:
+ la %r3,0(%r5,%r2)
+ vlgvb %r1,%v17,7 /* Load byte index of zero. */
+ aghi %r1,3 /* Also copy remaining bytes of zero. */
+ vstl %v16,%r1,0(%r3) /* Copy characters including zero. */
+ lay %r2,-3(%r1,%r3) /* Return pointer to zero. */
+ br %r14
+
+.Lfound_v18_48:
+ aghi %r5,32
+.Lfound_v18_16:
+ la %r3,16(%r5,%r2)
+ vlgvb %r1,%v17,7 /* Load byte index of zero. */
+ aghi %r1,3 /* Also copy remaining bytes of zero. */
+ vstl %v18,%r1,0(%r3) /* Copy characters including zero. */
+ lay %r2,-3(%r1,%r3) /* Return pointer to zero. */
+ br %r14
+
+.Lfound_align:
+ aghi %r5,3 /* Also copy remaining bytes of zero. */
+ vstl %v16,%r5,0(%r2) /* Copy characters including zero. */
+ lay %r2,-3(%r5,%r2) /* Return pointer to zero. */
+ br %r14
+
+.Lfallback:
+ jg WCPCPY_C
+END(WCPCPY_Z13)
+
+# if ! HAVE_WCPCPY_IFUNC
+strong_alias (WCPCPY_Z13, __wcpcpy)
+weak_alias (__wcpcpy, wcpcpy)
+# endif
+#endif