|
 |
e354a5 |
commit 1a153e47fcc9401d8ea424ad86569a57ed0f8c52
|
|
|
e354a5 |
Author: Leonardo Sandoval <leonardo.sandoval.gonzalez@linux.intel.com>
|
|
|
e354a5 |
Date: Mon Oct 8 08:59:50 2018 -0500
|
|
|
e354a5 |
|
|
|
e354a5 |
x86-64: Optimize strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2
|
|
|
e354a5 |
|
|
|
e354a5 |
Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2.
|
|
|
e354a5 |
It uses vector comparison as much as possible. In general, the larger the
|
|
|
e354a5 |
source string, the greater performance gain observed, reaching speedups of
|
|
|
e354a5 |
1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
|
|
|
e354a5 |
strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper is
|
|
|
e354a5 |
preferred and AVX unaligned load is fast.
|
|
|
e354a5 |
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
|
|
|
e354a5 |
strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
|
|
|
e354a5 |
stpcpy-avx2 and stpncpy-avx2.
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/ifunc-impl-list.c:
|
|
|
e354a5 |
(__libc_ifunc_impl_list): Add tests for __strcat_avx2,
|
|
|
e354a5 |
__strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
|
|
|
e354a5 |
and __stpncpy_avx2.
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
|
|
|
e354a5 |
ifunc-strcpy.h}: rename header for a more generic name.
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/ifunc-strcpy.h:
|
|
|
e354a5 |
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
|
|
|
e354a5 |
AVX unaligned load is fast and vzeroupper is preferred.
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
|
|
|
e354a5 |
* sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise
|
|
|
e354a5 |
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
|
|
|
e354a5 |
index bb5e97073520ee51..395e432c092ca17c 100644
|
|
|
e354a5 |
--- a/sysdeps/x86_64/multiarch/Makefile
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/Makefile
|
|
|
e354a5 |
@@ -24,11 +24,14 @@ sysdep_routines += strncat-c stpncpy-c strncpy-c \
|
|
|
e354a5 |
strchr-sse2 strchrnul-sse2 strchr-avx2 strchrnul-avx2 \
|
|
|
e354a5 |
strrchr-sse2 strrchr-avx2 \
|
|
|
e354a5 |
strlen-sse2 strnlen-sse2 strlen-avx2 strnlen-avx2 \
|
|
|
e354a5 |
+ strcat-avx2 strncat-avx2 \
|
|
|
e354a5 |
strcat-ssse3 strncat-ssse3\
|
|
|
e354a5 |
+ strcpy-avx2 strncpy-avx2 \
|
|
|
e354a5 |
strcpy-sse2 stpcpy-sse2 \
|
|
|
e354a5 |
strcpy-ssse3 strncpy-ssse3 stpcpy-ssse3 stpncpy-ssse3 \
|
|
|
e354a5 |
strcpy-sse2-unaligned strncpy-sse2-unaligned \
|
|
|
e354a5 |
stpcpy-sse2-unaligned stpncpy-sse2-unaligned \
|
|
|
e354a5 |
+ stpcpy-avx2 stpncpy-avx2 \
|
|
|
e354a5 |
strcat-sse2 \
|
|
|
e354a5 |
strcat-sse2-unaligned strncat-sse2-unaligned \
|
|
|
e354a5 |
strchr-sse2-no-bsf memcmp-ssse3 strstr-sse2-unaligned \
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
e354a5 |
index 9aaaef7251b8edfe..8b55bb6954000cc2 100644
|
|
|
e354a5 |
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
e354a5 |
@@ -199,6 +199,8 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
e354a5 |
IFUNC_IMPL (i, name, stpncpy,
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, stpncpy, HAS_CPU_FEATURE (SSSE3),
|
|
|
e354a5 |
__stpncpy_ssse3)
|
|
|
e354a5 |
+ IFUNC_IMPL_ADD (array, i, stpncpy, HAS_ARCH_FEATURE (AVX2_Usable),
|
|
|
e354a5 |
+ __stpncpy_avx2)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, stpncpy, 1,
|
|
|
e354a5 |
__stpncpy_sse2_unaligned)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, stpncpy, 1, __stpncpy_sse2))
|
|
|
e354a5 |
@@ -207,6 +209,8 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
e354a5 |
IFUNC_IMPL (i, name, stpcpy,
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, stpcpy, HAS_CPU_FEATURE (SSSE3),
|
|
|
e354a5 |
__stpcpy_ssse3)
|
|
|
e354a5 |
+ IFUNC_IMPL_ADD (array, i, stpcpy, HAS_ARCH_FEATURE (AVX2_Usable),
|
|
|
e354a5 |
+ __stpcpy_avx2)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, stpcpy, 1, __stpcpy_sse2_unaligned)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, stpcpy, 1, __stpcpy_sse2))
|
|
|
e354a5 |
|
|
|
e354a5 |
@@ -239,6 +243,8 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
e354a5 |
|
|
|
e354a5 |
/* Support sysdeps/x86_64/multiarch/strcat.c. */
|
|
|
e354a5 |
IFUNC_IMPL (i, name, strcat,
|
|
|
e354a5 |
+ IFUNC_IMPL_ADD (array, i, strcat, HAS_ARCH_FEATURE (AVX2_Usable),
|
|
|
e354a5 |
+ __strcat_avx2)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, strcat, HAS_CPU_FEATURE (SSSE3),
|
|
|
e354a5 |
__strcat_ssse3)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, strcat, 1, __strcat_sse2_unaligned)
|
|
|
e354a5 |
@@ -280,6 +286,8 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
e354a5 |
|
|
|
e354a5 |
/* Support sysdeps/x86_64/multiarch/strcpy.c. */
|
|
|
e354a5 |
IFUNC_IMPL (i, name, strcpy,
|
|
|
e354a5 |
+ IFUNC_IMPL_ADD (array, i, strcpy, HAS_ARCH_FEATURE (AVX2_Usable),
|
|
|
e354a5 |
+ __strcpy_avx2)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, strcpy, HAS_CPU_FEATURE (SSSE3),
|
|
|
e354a5 |
__strcpy_ssse3)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, strcpy, 1, __strcpy_sse2_unaligned)
|
|
|
e354a5 |
@@ -321,6 +329,8 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
e354a5 |
|
|
|
e354a5 |
/* Support sysdeps/x86_64/multiarch/strncat.c. */
|
|
|
e354a5 |
IFUNC_IMPL (i, name, strncat,
|
|
|
e354a5 |
+ IFUNC_IMPL_ADD (array, i, strncat, HAS_ARCH_FEATURE (AVX2_Usable),
|
|
|
e354a5 |
+ __strncat_avx2)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, strncat, HAS_CPU_FEATURE (SSSE3),
|
|
|
e354a5 |
__strncat_ssse3)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, strncat, 1,
|
|
|
e354a5 |
@@ -329,6 +339,8 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
e354a5 |
|
|
|
e354a5 |
/* Support sysdeps/x86_64/multiarch/strncpy.c. */
|
|
|
e354a5 |
IFUNC_IMPL (i, name, strncpy,
|
|
|
e354a5 |
+ IFUNC_IMPL_ADD (array, i, strncpy, HAS_ARCH_FEATURE (AVX2_Usable),
|
|
|
e354a5 |
+ __strncpy_avx2)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, strncpy, HAS_CPU_FEATURE (SSSE3),
|
|
|
e354a5 |
__strncpy_ssse3)
|
|
|
e354a5 |
IFUNC_IMPL_ADD (array, i, strncpy, 1,
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/ifunc-unaligned-ssse3.h b/sysdeps/x86_64/multiarch/ifunc-strcpy.h
|
|
|
e354a5 |
similarity index 83%
|
|
|
e354a5 |
rename from sysdeps/x86_64/multiarch/ifunc-unaligned-ssse3.h
|
|
|
e354a5 |
rename to sysdeps/x86_64/multiarch/ifunc-strcpy.h
|
|
|
e354a5 |
index 81805f9832345923..4f2286fefccda069 100644
|
|
|
e354a5 |
--- a/sysdeps/x86_64/multiarch/ifunc-unaligned-ssse3.h
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/ifunc-strcpy.h
|
|
|
e354a5 |
@@ -24,12 +24,18 @@ extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2) attribute_hidden;
|
|
|
e354a5 |
extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2_unaligned)
|
|
|
e354a5 |
attribute_hidden;
|
|
|
e354a5 |
extern __typeof (REDIRECT_NAME) OPTIMIZE (ssse3) attribute_hidden;
|
|
|
e354a5 |
+extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2) attribute_hidden;
|
|
|
e354a5 |
|
|
|
e354a5 |
static inline void *
|
|
|
e354a5 |
IFUNC_SELECTOR (void)
|
|
|
e354a5 |
{
|
|
|
e354a5 |
const struct cpu_features* cpu_features = __get_cpu_features ();
|
|
|
e354a5 |
|
|
|
e354a5 |
+ if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER)
|
|
|
e354a5 |
+ && CPU_FEATURES_ARCH_P (cpu_features, AVX2_Usable)
|
|
|
e354a5 |
+ && CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load))
|
|
|
e354a5 |
+ return OPTIMIZE (avx2);
|
|
|
e354a5 |
+
|
|
|
e354a5 |
if (CPU_FEATURES_ARCH_P (cpu_features, Fast_Unaligned_Load))
|
|
|
e354a5 |
return OPTIMIZE (sse2_unaligned);
|
|
|
e354a5 |
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/stpcpy-avx2.S b/sysdeps/x86_64/multiarch/stpcpy-avx2.S
|
|
|
e354a5 |
new file mode 100644
|
|
|
e354a5 |
index 0000000000000000..f0bd3029fe3047ed
|
|
|
e354a5 |
--- /dev/null
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/stpcpy-avx2.S
|
|
|
e354a5 |
@@ -0,0 +1,3 @@
|
|
|
e354a5 |
+#define USE_AS_STPCPY
|
|
|
e354a5 |
+#define STRCPY __stpcpy_avx2
|
|
|
e354a5 |
+#include "strcpy-avx2.S"
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/stpcpy.c b/sysdeps/x86_64/multiarch/stpcpy.c
|
|
|
e354a5 |
index 1e340fca991a021c..8ffd13b48c83ca8e 100644
|
|
|
e354a5 |
--- a/sysdeps/x86_64/multiarch/stpcpy.c
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/stpcpy.c
|
|
|
e354a5 |
@@ -28,7 +28,7 @@
|
|
|
e354a5 |
# undef __stpcpy
|
|
|
e354a5 |
|
|
|
e354a5 |
# define SYMBOL_NAME stpcpy
|
|
|
e354a5 |
-# include "ifunc-unaligned-ssse3.h"
|
|
|
e354a5 |
+# include "ifunc-strcpy.h"
|
|
|
e354a5 |
|
|
|
e354a5 |
libc_ifunc_redirected (__redirect_stpcpy, __stpcpy, IFUNC_SELECTOR ());
|
|
|
e354a5 |
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/stpncpy-avx2.S b/sysdeps/x86_64/multiarch/stpncpy-avx2.S
|
|
|
e354a5 |
new file mode 100644
|
|
|
e354a5 |
index 0000000000000000..032b0407d08c6a9d
|
|
|
e354a5 |
--- /dev/null
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/stpncpy-avx2.S
|
|
|
e354a5 |
@@ -0,0 +1,4 @@
|
|
|
e354a5 |
+#define USE_AS_STPCPY
|
|
|
e354a5 |
+#define USE_AS_STRNCPY
|
|
|
e354a5 |
+#define STRCPY __stpncpy_avx2
|
|
|
e354a5 |
+#include "strcpy-avx2.S"
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/stpncpy.c b/sysdeps/x86_64/multiarch/stpncpy.c
|
|
|
e354a5 |
index 28842ece2b0998e3..f3e203f78cca2e61 100644
|
|
|
e354a5 |
--- a/sysdeps/x86_64/multiarch/stpncpy.c
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/stpncpy.c
|
|
|
e354a5 |
@@ -26,7 +26,7 @@
|
|
|
e354a5 |
# undef __stpncpy
|
|
|
e354a5 |
|
|
|
e354a5 |
# define SYMBOL_NAME stpncpy
|
|
|
e354a5 |
-# include "ifunc-unaligned-ssse3.h"
|
|
|
e354a5 |
+# include "ifunc-strcpy.h"
|
|
|
e354a5 |
|
|
|
e354a5 |
libc_ifunc_redirected (__redirect_stpncpy, __stpncpy, IFUNC_SELECTOR ());
|
|
|
e354a5 |
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/strcat-avx2.S b/sysdeps/x86_64/multiarch/strcat-avx2.S
|
|
|
e354a5 |
new file mode 100644
|
|
|
e354a5 |
index 0000000000000000..b062356427677ca6
|
|
|
e354a5 |
--- /dev/null
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/strcat-avx2.S
|
|
|
e354a5 |
@@ -0,0 +1,275 @@
|
|
|
e354a5 |
+/* strcat with AVX2
|
|
|
e354a5 |
+ Copyright (C) 2011-2018 Free Software Foundation, Inc.
|
|
|
e354a5 |
+ Contributed by Intel Corporation.
|
|
|
e354a5 |
+ This file is part of the GNU C Library.
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
e354a5 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
e354a5 |
+ License as published by the Free Software Foundation; either
|
|
|
e354a5 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
e354a5 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
e354a5 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
e354a5 |
+ Lesser General Public License for more details.
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
e354a5 |
+ License along with the GNU C Library; if not, see
|
|
|
e354a5 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+#if IS_IN (libc)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# include <sysdep.h>
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifndef STRCAT
|
|
|
e354a5 |
+# define STRCAT __strcat_avx2
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# define USE_AS_STRCAT
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* Number of bytes in a vector register */
|
|
|
e354a5 |
+# define VEC_SIZE 32
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .section .text.avx,"ax",@progbits
|
|
|
e354a5 |
+ENTRY (STRCAT)
|
|
|
e354a5 |
+ mov %rdi, %r9
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCAT
|
|
|
e354a5 |
+ mov %rdx, %r8
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ xor %eax, %eax
|
|
|
e354a5 |
+ mov %edi, %ecx
|
|
|
e354a5 |
+ and $((VEC_SIZE * 4) - 1), %ecx
|
|
|
e354a5 |
+ vpxor %xmm6, %xmm6, %xmm6
|
|
|
e354a5 |
+ cmp $(VEC_SIZE * 3), %ecx
|
|
|
e354a5 |
+ ja L(fourth_vector_boundary)
|
|
|
e354a5 |
+ vpcmpeqb (%rdi), %ymm6, %ymm0
|
|
|
e354a5 |
+ vpmovmskb %ymm0, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_first_vector)
|
|
|
e354a5 |
+ mov %rdi, %rax
|
|
|
e354a5 |
+ and $-VEC_SIZE, %rax
|
|
|
e354a5 |
+ jmp L(align_vec_size_start)
|
|
|
e354a5 |
+L(fourth_vector_boundary):
|
|
|
e354a5 |
+ mov %rdi, %rax
|
|
|
e354a5 |
+ and $-VEC_SIZE, %rax
|
|
|
e354a5 |
+ vpcmpeqb (%rax), %ymm6, %ymm0
|
|
|
e354a5 |
+ mov $-1, %r10d
|
|
|
e354a5 |
+ sub %rax, %rcx
|
|
|
e354a5 |
+ shl %cl, %r10d
|
|
|
e354a5 |
+ vpmovmskb %ymm0, %edx
|
|
|
e354a5 |
+ and %r10d, %edx
|
|
|
e354a5 |
+ jnz L(exit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+L(align_vec_size_start):
|
|
|
e354a5 |
+ vpcmpeqb VEC_SIZE(%rax), %ymm6, %ymm0
|
|
|
e354a5 |
+ vpmovmskb %ymm0, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_second_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
|
|
|
e354a5 |
+ vpmovmskb %ymm1, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_third_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
|
|
|
e354a5 |
+ vpmovmskb %ymm2, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_fourth_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
|
|
|
e354a5 |
+ vpmovmskb %ymm3, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_fifth_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 5)(%rax), %ymm6, %ymm0
|
|
|
e354a5 |
+ add $(VEC_SIZE * 4), %rax
|
|
|
e354a5 |
+ vpmovmskb %ymm0, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_second_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
|
|
|
e354a5 |
+ vpmovmskb %ymm1, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_third_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
|
|
|
e354a5 |
+ vpmovmskb %ymm2, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_fourth_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
|
|
|
e354a5 |
+ vpmovmskb %ymm3, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_fifth_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 5)(%rax), %ymm6, %ymm0
|
|
|
e354a5 |
+ add $(VEC_SIZE * 4), %rax
|
|
|
e354a5 |
+ vpmovmskb %ymm0, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_second_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
|
|
|
e354a5 |
+ vpmovmskb %ymm1, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_third_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
|
|
|
e354a5 |
+ vpmovmskb %ymm2, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_fourth_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
|
|
|
e354a5 |
+ vpmovmskb %ymm3, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_fifth_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 5)(%rax), %ymm6, %ymm0
|
|
|
e354a5 |
+ add $(VEC_SIZE * 4), %rax
|
|
|
e354a5 |
+ vpmovmskb %ymm0, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_second_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
|
|
|
e354a5 |
+ vpmovmskb %ymm1, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_third_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
|
|
|
e354a5 |
+ vpmovmskb %ymm2, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_fourth_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
|
|
|
e354a5 |
+ vpmovmskb %ymm3, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_fifth_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ test $((VEC_SIZE * 4) - 1), %rax
|
|
|
e354a5 |
+ jz L(align_four_vec_loop)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 5)(%rax), %ymm6, %ymm0
|
|
|
e354a5 |
+ add $(VEC_SIZE * 5), %rax
|
|
|
e354a5 |
+ vpmovmskb %ymm0, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ test $((VEC_SIZE * 4) - 1), %rax
|
|
|
e354a5 |
+ jz L(align_four_vec_loop)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb VEC_SIZE(%rax), %ymm6, %ymm1
|
|
|
e354a5 |
+ add $VEC_SIZE, %rax
|
|
|
e354a5 |
+ vpmovmskb %ymm1, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ test $((VEC_SIZE * 4) - 1), %rax
|
|
|
e354a5 |
+ jz L(align_four_vec_loop)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb VEC_SIZE(%rax), %ymm6, %ymm2
|
|
|
e354a5 |
+ add $VEC_SIZE, %rax
|
|
|
e354a5 |
+ vpmovmskb %ymm2, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ test $((VEC_SIZE * 4) - 1), %rax
|
|
|
e354a5 |
+ jz L(align_four_vec_loop)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb VEC_SIZE(%rax), %ymm6, %ymm3
|
|
|
e354a5 |
+ add $VEC_SIZE, %rax
|
|
|
e354a5 |
+ vpmovmskb %ymm3, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ add $VEC_SIZE, %rax
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(align_four_vec_loop):
|
|
|
e354a5 |
+ vmovaps (%rax), %ymm4
|
|
|
e354a5 |
+ vpminub VEC_SIZE(%rax), %ymm4, %ymm4
|
|
|
e354a5 |
+ vmovaps (VEC_SIZE * 2)(%rax), %ymm5
|
|
|
e354a5 |
+ vpminub (VEC_SIZE * 3)(%rax), %ymm5, %ymm5
|
|
|
e354a5 |
+ add $(VEC_SIZE * 4), %rax
|
|
|
e354a5 |
+ vpminub %ymm4, %ymm5, %ymm5
|
|
|
e354a5 |
+ vpcmpeqb %ymm5, %ymm6, %ymm5
|
|
|
e354a5 |
+ vpmovmskb %ymm5, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jz L(align_four_vec_loop)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb -(VEC_SIZE * 4)(%rax), %ymm6, %ymm0
|
|
|
e354a5 |
+ sub $(VEC_SIZE * 5), %rax
|
|
|
e354a5 |
+ vpmovmskb %ymm0, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_second_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
|
|
|
e354a5 |
+ vpmovmskb %ymm1, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_third_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
|
|
|
e354a5 |
+ vpmovmskb %ymm2, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(exit_null_on_fourth_vector)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
|
|
|
e354a5 |
+ vpmovmskb %ymm3, %edx
|
|
|
e354a5 |
+ sub %rdi, %rax
|
|
|
e354a5 |
+ bsf %rdx, %rdx
|
|
|
e354a5 |
+ add %rdx, %rax
|
|
|
e354a5 |
+ add $(VEC_SIZE * 4), %rax
|
|
|
e354a5 |
+ jmp L(StartStrcpyPart)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(exit):
|
|
|
e354a5 |
+ sub %rdi, %rax
|
|
|
e354a5 |
+L(exit_null_on_first_vector):
|
|
|
e354a5 |
+ bsf %rdx, %rdx
|
|
|
e354a5 |
+ add %rdx, %rax
|
|
|
e354a5 |
+ jmp L(StartStrcpyPart)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(exit_null_on_second_vector):
|
|
|
e354a5 |
+ sub %rdi, %rax
|
|
|
e354a5 |
+ bsf %rdx, %rdx
|
|
|
e354a5 |
+ add %rdx, %rax
|
|
|
e354a5 |
+ add $VEC_SIZE, %rax
|
|
|
e354a5 |
+ jmp L(StartStrcpyPart)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(exit_null_on_third_vector):
|
|
|
e354a5 |
+ sub %rdi, %rax
|
|
|
e354a5 |
+ bsf %rdx, %rdx
|
|
|
e354a5 |
+ add %rdx, %rax
|
|
|
e354a5 |
+ add $(VEC_SIZE * 2), %rax
|
|
|
e354a5 |
+ jmp L(StartStrcpyPart)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(exit_null_on_fourth_vector):
|
|
|
e354a5 |
+ sub %rdi, %rax
|
|
|
e354a5 |
+ bsf %rdx, %rdx
|
|
|
e354a5 |
+ add %rdx, %rax
|
|
|
e354a5 |
+ add $(VEC_SIZE * 3), %rax
|
|
|
e354a5 |
+ jmp L(StartStrcpyPart)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(exit_null_on_fifth_vector):
|
|
|
e354a5 |
+ sub %rdi, %rax
|
|
|
e354a5 |
+ bsf %rdx, %rdx
|
|
|
e354a5 |
+ add %rdx, %rax
|
|
|
e354a5 |
+ add $(VEC_SIZE * 4), %rax
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StartStrcpyPart):
|
|
|
e354a5 |
+ lea (%r9, %rax), %rdi
|
|
|
e354a5 |
+ mov %rsi, %rcx
|
|
|
e354a5 |
+ mov %r9, %rax /* save result */
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCAT
|
|
|
e354a5 |
+ test %r8, %r8
|
|
|
e354a5 |
+ jz L(ExitZero)
|
|
|
e354a5 |
+# define USE_AS_STRNCPY
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# include "strcpy-avx2.S"
|
|
|
e354a5 |
+#endif
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/strcat.c b/sysdeps/x86_64/multiarch/strcat.c
|
|
|
e354a5 |
index 1f7f6263f35ba402..694b9b2405827bd4 100644
|
|
|
e354a5 |
--- a/sysdeps/x86_64/multiarch/strcat.c
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/strcat.c
|
|
|
e354a5 |
@@ -24,7 +24,7 @@
|
|
|
e354a5 |
# undef strcat
|
|
|
e354a5 |
|
|
|
e354a5 |
# define SYMBOL_NAME strcat
|
|
|
e354a5 |
-# include "ifunc-unaligned-ssse3.h"
|
|
|
e354a5 |
+# include "ifunc-strcpy.h"
|
|
|
e354a5 |
|
|
|
e354a5 |
libc_ifunc_redirected (__redirect_strcat, strcat, IFUNC_SELECTOR ());
|
|
|
e354a5 |
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/strcpy-avx2.S b/sysdeps/x86_64/multiarch/strcpy-avx2.S
|
|
|
e354a5 |
new file mode 100644
|
|
|
e354a5 |
index 0000000000000000..81677f9060773a49
|
|
|
e354a5 |
--- /dev/null
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/strcpy-avx2.S
|
|
|
e354a5 |
@@ -0,0 +1,1022 @@
|
|
|
e354a5 |
+/* strcpy with AVX2
|
|
|
e354a5 |
+ Copyright (C) 2011-2018 Free Software Foundation, Inc.
|
|
|
e354a5 |
+ Contributed by Intel Corporation.
|
|
|
e354a5 |
+ This file is part of the GNU C Library.
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
e354a5 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
e354a5 |
+ License as published by the Free Software Foundation; either
|
|
|
e354a5 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
e354a5 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
e354a5 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
e354a5 |
+ Lesser General Public License for more details.
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
e354a5 |
+ License along with the GNU C Library; if not, see
|
|
|
e354a5 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+#if IS_IN (libc)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifndef USE_AS_STRCAT
|
|
|
e354a5 |
+# include <sysdep.h>
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifndef STRCPY
|
|
|
e354a5 |
+# define STRCPY __strcpy_avx2
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* Number of bytes in a vector register */
|
|
|
e354a5 |
+# ifndef VEC_SIZE
|
|
|
e354a5 |
+# define VEC_SIZE 32
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifndef VZEROUPPER
|
|
|
e354a5 |
+# define VZEROUPPER vzeroupper
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* zero register */
|
|
|
e354a5 |
+#define xmmZ xmm0
|
|
|
e354a5 |
+#define ymmZ ymm0
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* mask register */
|
|
|
e354a5 |
+#define ymmM ymm1
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifndef USE_AS_STRCAT
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .section .text.avx,"ax",@progbits
|
|
|
e354a5 |
+ENTRY (STRCPY)
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ mov %rdx, %r8
|
|
|
e354a5 |
+ test %r8, %r8
|
|
|
e354a5 |
+ jz L(ExitZero)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ mov %rsi, %rcx
|
|
|
e354a5 |
+# ifndef USE_AS_STPCPY
|
|
|
e354a5 |
+ mov %rdi, %rax /* save result */
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpxor %xmmZ, %xmmZ, %xmmZ
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ and $((VEC_SIZE * 4) - 1), %ecx
|
|
|
e354a5 |
+ cmp $(VEC_SIZE * 2), %ecx
|
|
|
e354a5 |
+ jbe L(SourceStringAlignmentLessTwoVecSize)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ and $-VEC_SIZE, %rsi
|
|
|
e354a5 |
+ and $(VEC_SIZE - 1), %ecx
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb (%rsi), %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ shr %cl, %rdx
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+# if defined USE_AS_STPCPY || defined USE_AS_STRCAT
|
|
|
e354a5 |
+ mov $VEC_SIZE, %r10
|
|
|
e354a5 |
+ sub %rcx, %r10
|
|
|
e354a5 |
+ cmp %r10, %r8
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ mov $(VEC_SIZE + 1), %r10
|
|
|
e354a5 |
+ sub %rcx, %r10
|
|
|
e354a5 |
+ cmp %r10, %r8
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ jbe L(CopyVecSizeTailCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(CopyVecSizeTail)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb VEC_SIZE(%rsi), %ymmZ, %ymm2
|
|
|
e354a5 |
+ vpmovmskb %ymm2, %edx
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ add $VEC_SIZE, %r10
|
|
|
e354a5 |
+ cmp %r10, %r8
|
|
|
e354a5 |
+ jbe L(CopyTwoVecSizeCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(CopyTwoVecSize)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vmovdqu (%rsi, %rcx), %ymm2 /* copy VEC_SIZE bytes */
|
|
|
e354a5 |
+ vmovdqu %ymm2, (%rdi)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* If source address alignment != destination address alignment */
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(UnalignVecSizeBoth):
|
|
|
e354a5 |
+ sub %rcx, %rdi
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ add %rcx, %r8
|
|
|
e354a5 |
+ sbb %rcx, %rcx
|
|
|
e354a5 |
+ or %rcx, %r8
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ mov $VEC_SIZE, %rcx
|
|
|
e354a5 |
+ vmovdqa (%rsi, %rcx), %ymm2
|
|
|
e354a5 |
+ vmovdqu %ymm2, (%rdi, %rcx)
|
|
|
e354a5 |
+ vmovdqa VEC_SIZE(%rsi, %rcx), %ymm2
|
|
|
e354a5 |
+ vpcmpeqb %ymm2, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ add $VEC_SIZE, %rcx
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ sub $(VEC_SIZE * 3), %r8
|
|
|
e354a5 |
+ jbe L(CopyVecSizeCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnalignedVec2)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jnz L(CopyVecSize)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vmovdqu %ymm2, (%rdi, %rcx)
|
|
|
e354a5 |
+ vmovdqa VEC_SIZE(%rsi, %rcx), %ymm3
|
|
|
e354a5 |
+ vpcmpeqb %ymm3, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ add $VEC_SIZE, %rcx
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jbe L(CopyVecSizeCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnalignedVec3)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jnz L(CopyVecSize)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vmovdqu %ymm3, (%rdi, %rcx)
|
|
|
e354a5 |
+ vmovdqa VEC_SIZE(%rsi, %rcx), %ymm4
|
|
|
e354a5 |
+ vpcmpeqb %ymm4, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ add $VEC_SIZE, %rcx
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jbe L(CopyVecSizeCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnalignedVec4)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jnz L(CopyVecSize)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vmovdqu %ymm4, (%rdi, %rcx)
|
|
|
e354a5 |
+ vmovdqa VEC_SIZE(%rsi, %rcx), %ymm2
|
|
|
e354a5 |
+ vpcmpeqb %ymm2, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ add $VEC_SIZE, %rcx
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jbe L(CopyVecSizeCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnalignedVec2)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jnz L(CopyVecSize)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vmovdqu %ymm2, (%rdi, %rcx)
|
|
|
e354a5 |
+ vmovdqa VEC_SIZE(%rsi, %rcx), %ymm2
|
|
|
e354a5 |
+ vpcmpeqb %ymm2, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ add $VEC_SIZE, %rcx
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jbe L(CopyVecSizeCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnalignedVec2)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jnz L(CopyVecSize)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vmovdqa VEC_SIZE(%rsi, %rcx), %ymm3
|
|
|
e354a5 |
+ vmovdqu %ymm2, (%rdi, %rcx)
|
|
|
e354a5 |
+ vpcmpeqb %ymm3, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ add $VEC_SIZE, %rcx
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jbe L(CopyVecSizeCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnalignedVec3)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jnz L(CopyVecSize)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vmovdqu %ymm3, (%rdi, %rcx)
|
|
|
e354a5 |
+ mov %rsi, %rdx
|
|
|
e354a5 |
+ lea VEC_SIZE(%rsi, %rcx), %rsi
|
|
|
e354a5 |
+ and $-(VEC_SIZE * 4), %rsi
|
|
|
e354a5 |
+ sub %rsi, %rdx
|
|
|
e354a5 |
+ sub %rdx, %rdi
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ lea (VEC_SIZE * 8)(%r8, %rdx), %r8
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+L(UnalignedFourVecSizeLoop):
|
|
|
e354a5 |
+ vmovdqa (%rsi), %ymm4
|
|
|
e354a5 |
+ vmovdqa VEC_SIZE(%rsi), %ymm5
|
|
|
e354a5 |
+ vmovdqa (VEC_SIZE * 2)(%rsi), %ymm6
|
|
|
e354a5 |
+ vmovdqa (VEC_SIZE * 3)(%rsi), %ymm7
|
|
|
e354a5 |
+ vpminub %ymm5, %ymm4, %ymm2
|
|
|
e354a5 |
+ vpminub %ymm7, %ymm6, %ymm3
|
|
|
e354a5 |
+ vpminub %ymm2, %ymm3, %ymm3
|
|
|
e354a5 |
+ vpcmpeqb %ymmM, %ymm3, %ymm3
|
|
|
e354a5 |
+ vpmovmskb %ymm3, %edx
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ sub $(VEC_SIZE * 4), %r8
|
|
|
e354a5 |
+ jbe L(UnalignedLeaveCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(UnalignedFourVecSizeLeave)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+L(UnalignedFourVecSizeLoop_start):
|
|
|
e354a5 |
+ add $(VEC_SIZE * 4), %rdi
|
|
|
e354a5 |
+ add $(VEC_SIZE * 4), %rsi
|
|
|
e354a5 |
+ vmovdqu %ymm4, -(VEC_SIZE * 4)(%rdi)
|
|
|
e354a5 |
+ vmovdqa (%rsi), %ymm4
|
|
|
e354a5 |
+ vmovdqu %ymm5, -(VEC_SIZE * 3)(%rdi)
|
|
|
e354a5 |
+ vmovdqa VEC_SIZE(%rsi), %ymm5
|
|
|
e354a5 |
+ vpminub %ymm5, %ymm4, %ymm2
|
|
|
e354a5 |
+ vmovdqu %ymm6, -(VEC_SIZE * 2)(%rdi)
|
|
|
e354a5 |
+ vmovdqa (VEC_SIZE * 2)(%rsi), %ymm6
|
|
|
e354a5 |
+ vmovdqu %ymm7, -VEC_SIZE(%rdi)
|
|
|
e354a5 |
+ vmovdqa (VEC_SIZE * 3)(%rsi), %ymm7
|
|
|
e354a5 |
+ vpminub %ymm7, %ymm6, %ymm3
|
|
|
e354a5 |
+ vpminub %ymm2, %ymm3, %ymm3
|
|
|
e354a5 |
+ vpcmpeqb %ymmM, %ymm3, %ymm3
|
|
|
e354a5 |
+ vpmovmskb %ymm3, %edx
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+ sub $(VEC_SIZE * 4), %r8
|
|
|
e354a5 |
+ jbe L(UnalignedLeaveCase2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jz L(UnalignedFourVecSizeLoop_start)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+L(UnalignedFourVecSizeLeave):
|
|
|
e354a5 |
+ vpcmpeqb %ymm4, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnaligned_0)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb %ymm5, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %ecx
|
|
|
e354a5 |
+ test %ecx, %ecx
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnaligned_16)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb %ymm6, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnaligned_32)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb %ymm7, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %ecx
|
|
|
e354a5 |
+ bsf %ecx, %edx
|
|
|
e354a5 |
+ vmovdqu %ymm4, (%rdi)
|
|
|
e354a5 |
+ vmovdqu %ymm5, VEC_SIZE(%rdi)
|
|
|
e354a5 |
+ vmovdqu %ymm6, (VEC_SIZE * 2)(%rdi)
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (VEC_SIZE * 3)(%rdi, %rdx), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ vmovdqu %ymm7, (VEC_SIZE * 3)(%rdi)
|
|
|
e354a5 |
+ add $(VEC_SIZE - 1), %r8
|
|
|
e354a5 |
+ sub %rdx, %r8
|
|
|
e354a5 |
+ lea ((VEC_SIZE * 3) + 1)(%rdi, %rdx), %rdi
|
|
|
e354a5 |
+ jmp L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ add $(VEC_SIZE * 3), %rsi
|
|
|
e354a5 |
+ add $(VEC_SIZE * 3), %rdi
|
|
|
e354a5 |
+ jmp L(CopyVecSizeExit)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* If source address alignment == destination address alignment */
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+L(SourceStringAlignmentLessTwoVecSize):
|
|
|
e354a5 |
+ vmovdqu (%rsi), %ymm3
|
|
|
e354a5 |
+ vmovdqu VEC_SIZE(%rsi), %ymm2
|
|
|
e354a5 |
+ vpcmpeqb %ymm3, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+# if defined USE_AS_STPCPY || defined USE_AS_STRCAT
|
|
|
e354a5 |
+ cmp $VEC_SIZE, %r8
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ cmp $(VEC_SIZE + 1), %r8
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ jbe L(CopyVecSizeTail1Case2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(CopyVecSizeTail1)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vmovdqu %ymm3, (%rdi)
|
|
|
e354a5 |
+ vpcmpeqb %ymm2, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+# if defined USE_AS_STPCPY || defined USE_AS_STRCAT
|
|
|
e354a5 |
+ cmp $(VEC_SIZE * 2), %r8
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ cmp $((VEC_SIZE * 2) + 1), %r8
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ jbe L(CopyTwoVecSize1Case2OrCase3)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+ jnz L(CopyTwoVecSize1)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ and $-VEC_SIZE, %rsi
|
|
|
e354a5 |
+ and $(VEC_SIZE - 1), %ecx
|
|
|
e354a5 |
+ jmp L(UnalignVecSizeBoth)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/*------End of main part with loops---------------------*/
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* Case1 */
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# if (!defined USE_AS_STRNCPY) || (defined USE_AS_STRCAT)
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSize):
|
|
|
e354a5 |
+ add %rcx, %rdi
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+L(CopyVecSizeTail):
|
|
|
e354a5 |
+ add %rcx, %rsi
|
|
|
e354a5 |
+L(CopyVecSizeTail1):
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+L(CopyVecSizeExit):
|
|
|
e354a5 |
+ cmp $32, %edx
|
|
|
e354a5 |
+ jae L(Exit32_63)
|
|
|
e354a5 |
+ cmp $16, %edx
|
|
|
e354a5 |
+ jae L(Exit16_31)
|
|
|
e354a5 |
+ cmp $8, %edx
|
|
|
e354a5 |
+ jae L(Exit8_15)
|
|
|
e354a5 |
+ cmp $4, %edx
|
|
|
e354a5 |
+ jae L(Exit4_7)
|
|
|
e354a5 |
+ cmp $3, %edx
|
|
|
e354a5 |
+ je L(Exit3)
|
|
|
e354a5 |
+ cmp $1, %edx
|
|
|
e354a5 |
+ ja L(Exit2)
|
|
|
e354a5 |
+ je L(Exit1)
|
|
|
e354a5 |
+ movb $0, (%rdi)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ sub $1, %r8
|
|
|
e354a5 |
+ lea 1(%rdi), %rdi
|
|
|
e354a5 |
+ jnz L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyTwoVecSize1):
|
|
|
e354a5 |
+ add $VEC_SIZE, %rsi
|
|
|
e354a5 |
+ add $VEC_SIZE, %rdi
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ jmp L(CopyVecSizeTail1)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyTwoVecSize):
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+ add %rcx, %rsi
|
|
|
e354a5 |
+ add $VEC_SIZE, %edx
|
|
|
e354a5 |
+ sub %ecx, %edx
|
|
|
e354a5 |
+ jmp L(CopyVecSizeExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeUnaligned_0):
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %rdx), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ vmovdqu %ymm4, (%rdi)
|
|
|
e354a5 |
+ add $((VEC_SIZE * 4) - 1), %r8
|
|
|
e354a5 |
+ sub %rdx, %r8
|
|
|
e354a5 |
+ lea 1(%rdi, %rdx), %rdi
|
|
|
e354a5 |
+ jmp L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jmp L(CopyVecSizeExit)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeUnaligned_16):
|
|
|
e354a5 |
+ bsf %ecx, %edx
|
|
|
e354a5 |
+ vmovdqu %ymm4, (%rdi)
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea VEC_SIZE(%rdi, %rdx), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ vmovdqu %ymm5, VEC_SIZE(%rdi)
|
|
|
e354a5 |
+ add $((VEC_SIZE * 3) - 1), %r8
|
|
|
e354a5 |
+ sub %rdx, %r8
|
|
|
e354a5 |
+ lea (VEC_SIZE + 1)(%rdi, %rdx), %rdi
|
|
|
e354a5 |
+ jmp L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ add $VEC_SIZE, %rsi
|
|
|
e354a5 |
+ add $VEC_SIZE, %rdi
|
|
|
e354a5 |
+ jmp L(CopyVecSizeExit)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeUnaligned_32):
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+ vmovdqu %ymm4, (%rdi)
|
|
|
e354a5 |
+ vmovdqu %ymm5, VEC_SIZE(%rdi)
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (VEC_SIZE * 2)(%rdi, %rdx), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ vmovdqu %ymm6, (VEC_SIZE * 2)(%rdi)
|
|
|
e354a5 |
+ add $((VEC_SIZE * 2) - 1), %r8
|
|
|
e354a5 |
+ sub %rdx, %r8
|
|
|
e354a5 |
+ lea ((VEC_SIZE * 2) + 1)(%rdi, %rdx), %rdi
|
|
|
e354a5 |
+ jmp L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ add $(VEC_SIZE * 2), %rsi
|
|
|
e354a5 |
+ add $(VEC_SIZE * 2), %rdi
|
|
|
e354a5 |
+ jmp L(CopyVecSizeExit)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+# ifndef USE_AS_STRCAT
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeUnalignedVec6):
|
|
|
e354a5 |
+ vmovdqu %ymm6, (%rdi, %rcx)
|
|
|
e354a5 |
+ jmp L(CopyVecSizeVecExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeUnalignedVec5):
|
|
|
e354a5 |
+ vmovdqu %ymm5, (%rdi, %rcx)
|
|
|
e354a5 |
+ jmp L(CopyVecSizeVecExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeUnalignedVec4):
|
|
|
e354a5 |
+ vmovdqu %ymm4, (%rdi, %rcx)
|
|
|
e354a5 |
+ jmp L(CopyVecSizeVecExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeUnalignedVec3):
|
|
|
e354a5 |
+ vmovdqu %ymm3, (%rdi, %rcx)
|
|
|
e354a5 |
+ jmp L(CopyVecSizeVecExit)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* Case2 */
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeCase2):
|
|
|
e354a5 |
+ add $VEC_SIZE, %r8
|
|
|
e354a5 |
+ add %rcx, %rdi
|
|
|
e354a5 |
+ add %rcx, %rsi
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+ cmp %r8d, %edx
|
|
|
e354a5 |
+ jb L(CopyVecSizeExit)
|
|
|
e354a5 |
+ jmp L(StrncpyExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyTwoVecSizeCase2):
|
|
|
e354a5 |
+ add %rcx, %rsi
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+ add $VEC_SIZE, %edx
|
|
|
e354a5 |
+ sub %ecx, %edx
|
|
|
e354a5 |
+ cmp %r8d, %edx
|
|
|
e354a5 |
+ jb L(CopyVecSizeExit)
|
|
|
e354a5 |
+ jmp L(StrncpyExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+L(CopyVecSizeTailCase2):
|
|
|
e354a5 |
+ add %rcx, %rsi
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+ cmp %r8d, %edx
|
|
|
e354a5 |
+ jb L(CopyVecSizeExit)
|
|
|
e354a5 |
+ jmp L(StrncpyExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+L(CopyVecSizeTail1Case2):
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+ cmp %r8d, %edx
|
|
|
e354a5 |
+ jb L(CopyVecSizeExit)
|
|
|
e354a5 |
+ jmp L(StrncpyExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* Case2 or Case3, Case3 */
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeCase2OrCase3):
|
|
|
e354a5 |
+ test %rdx, %rdx
|
|
|
e354a5 |
+ jnz L(CopyVecSizeCase2)
|
|
|
e354a5 |
+L(CopyVecSizeCase3):
|
|
|
e354a5 |
+ add $VEC_SIZE, %r8
|
|
|
e354a5 |
+ add %rcx, %rdi
|
|
|
e354a5 |
+ add %rcx, %rsi
|
|
|
e354a5 |
+ jmp L(StrncpyExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyTwoVecSizeCase2OrCase3):
|
|
|
e354a5 |
+ test %rdx, %rdx
|
|
|
e354a5 |
+ jnz L(CopyTwoVecSizeCase2)
|
|
|
e354a5 |
+ add %rcx, %rsi
|
|
|
e354a5 |
+ jmp L(StrncpyExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeTailCase2OrCase3):
|
|
|
e354a5 |
+ test %rdx, %rdx
|
|
|
e354a5 |
+ jnz L(CopyVecSizeTailCase2)
|
|
|
e354a5 |
+ add %rcx, %rsi
|
|
|
e354a5 |
+ jmp L(StrncpyExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyTwoVecSize1Case2OrCase3):
|
|
|
e354a5 |
+ add $VEC_SIZE, %rdi
|
|
|
e354a5 |
+ add $VEC_SIZE, %rsi
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+L(CopyVecSizeTail1Case2OrCase3):
|
|
|
e354a5 |
+ test %rdx, %rdx
|
|
|
e354a5 |
+ jnz L(CopyVecSizeTail1Case2)
|
|
|
e354a5 |
+ jmp L(StrncpyExit)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/*------------End labels regarding with copying 1-VEC_SIZE bytes--and 1-(VEC_SIZE*2) bytes----*/
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Exit1):
|
|
|
e354a5 |
+ movzwl (%rsi), %edx
|
|
|
e354a5 |
+ mov %dx, (%rdi)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea 1(%rdi), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ sub $2, %r8
|
|
|
e354a5 |
+ lea 2(%rdi), %rdi
|
|
|
e354a5 |
+ jnz L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Exit2):
|
|
|
e354a5 |
+ movzwl (%rsi), %ecx
|
|
|
e354a5 |
+ mov %cx, (%rdi)
|
|
|
e354a5 |
+ movb $0, 2(%rdi)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea 2(%rdi), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ sub $3, %r8
|
|
|
e354a5 |
+ lea 3(%rdi), %rdi
|
|
|
e354a5 |
+ jnz L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Exit3):
|
|
|
e354a5 |
+ mov (%rsi), %edx
|
|
|
e354a5 |
+ mov %edx, (%rdi)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea 3(%rdi), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ sub $4, %r8
|
|
|
e354a5 |
+ lea 4(%rdi), %rdi
|
|
|
e354a5 |
+ jnz L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Exit4_7):
|
|
|
e354a5 |
+ mov (%rsi), %ecx
|
|
|
e354a5 |
+ mov %ecx, (%rdi)
|
|
|
e354a5 |
+ mov -3(%rsi, %rdx), %ecx
|
|
|
e354a5 |
+ mov %ecx, -3(%rdi, %rdx)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %rdx), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ sub %rdx, %r8
|
|
|
e354a5 |
+ sub $1, %r8
|
|
|
e354a5 |
+ lea 1(%rdi, %rdx), %rdi
|
|
|
e354a5 |
+ jnz L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Exit8_15):
|
|
|
e354a5 |
+ mov (%rsi), %rcx
|
|
|
e354a5 |
+ mov -7(%rsi, %rdx), %r9
|
|
|
e354a5 |
+ mov %rcx, (%rdi)
|
|
|
e354a5 |
+ mov %r9, -7(%rdi, %rdx)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %rdx), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ sub %rdx, %r8
|
|
|
e354a5 |
+ sub $1, %r8
|
|
|
e354a5 |
+ lea 1(%rdi, %rdx), %rdi
|
|
|
e354a5 |
+ jnz L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Exit16_31):
|
|
|
e354a5 |
+ vmovdqu (%rsi), %xmm2
|
|
|
e354a5 |
+ vmovdqu -15(%rsi, %rdx), %xmm3
|
|
|
e354a5 |
+ vmovdqu %xmm2, (%rdi)
|
|
|
e354a5 |
+ vmovdqu %xmm3, -15(%rdi, %rdx)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %rdx), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ sub %rdx, %r8
|
|
|
e354a5 |
+ sub $1, %r8
|
|
|
e354a5 |
+ lea 1(%rdi, %rdx), %rdi
|
|
|
e354a5 |
+ jnz L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Exit32_63):
|
|
|
e354a5 |
+ vmovdqu (%rsi), %ymm2
|
|
|
e354a5 |
+ vmovdqu -31(%rsi, %rdx), %ymm3
|
|
|
e354a5 |
+ vmovdqu %ymm2, (%rdi)
|
|
|
e354a5 |
+ vmovdqu %ymm3, -31(%rdi, %rdx)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %rdx), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
|
|
|
e354a5 |
+ sub %rdx, %r8
|
|
|
e354a5 |
+ sub $1, %r8
|
|
|
e354a5 |
+ lea 1(%rdi, %rdx), %rdi
|
|
|
e354a5 |
+ jnz L(StrncpyFillTailWithZero)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifdef USE_AS_STRNCPY
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyExit1):
|
|
|
e354a5 |
+ movzbl (%rsi), %edx
|
|
|
e354a5 |
+ mov %dl, (%rdi)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea 1(%rdi), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, 1(%rdi)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyExit2):
|
|
|
e354a5 |
+ movzwl (%rsi), %edx
|
|
|
e354a5 |
+ mov %dx, (%rdi)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea 2(%rdi), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, 2(%rdi)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyExit3_4):
|
|
|
e354a5 |
+ movzwl (%rsi), %ecx
|
|
|
e354a5 |
+ movzwl -2(%rsi, %r8), %edx
|
|
|
e354a5 |
+ mov %cx, (%rdi)
|
|
|
e354a5 |
+ mov %dx, -2(%rdi, %r8)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %r8), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, (%rdi, %r8)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyExit5_8):
|
|
|
e354a5 |
+ mov (%rsi), %ecx
|
|
|
e354a5 |
+ mov -4(%rsi, %r8), %edx
|
|
|
e354a5 |
+ mov %ecx, (%rdi)
|
|
|
e354a5 |
+ mov %edx, -4(%rdi, %r8)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %r8), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, (%rdi, %r8)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyExit9_16):
|
|
|
e354a5 |
+ mov (%rsi), %rcx
|
|
|
e354a5 |
+ mov -8(%rsi, %r8), %rdx
|
|
|
e354a5 |
+ mov %rcx, (%rdi)
|
|
|
e354a5 |
+ mov %rdx, -8(%rdi, %r8)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %r8), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, (%rdi, %r8)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyExit17_32):
|
|
|
e354a5 |
+ vmovdqu (%rsi), %xmm2
|
|
|
e354a5 |
+ vmovdqu -16(%rsi, %r8), %xmm3
|
|
|
e354a5 |
+ vmovdqu %xmm2, (%rdi)
|
|
|
e354a5 |
+ vmovdqu %xmm3, -16(%rdi, %r8)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %r8), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, (%rdi, %r8)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyExit33_64):
|
|
|
e354a5 |
+ /* 0/32, 31/16 */
|
|
|
e354a5 |
+ vmovdqu (%rsi), %ymm2
|
|
|
e354a5 |
+ vmovdqu -VEC_SIZE(%rsi, %r8), %ymm3
|
|
|
e354a5 |
+ vmovdqu %ymm2, (%rdi)
|
|
|
e354a5 |
+ vmovdqu %ymm3, -VEC_SIZE(%rdi, %r8)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %r8), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, (%rdi, %r8)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyExit65):
|
|
|
e354a5 |
+ /* 0/32, 32/32, 64/1 */
|
|
|
e354a5 |
+ vmovdqu (%rsi), %ymm2
|
|
|
e354a5 |
+ vmovdqu 32(%rsi), %ymm3
|
|
|
e354a5 |
+ mov 64(%rsi), %cl
|
|
|
e354a5 |
+ vmovdqu %ymm2, (%rdi)
|
|
|
e354a5 |
+ vmovdqu %ymm3, 32(%rdi)
|
|
|
e354a5 |
+ mov %cl, 64(%rdi)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea 65(%rdi), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, 65(%rdi)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifndef USE_AS_STRCAT
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Fill1):
|
|
|
e354a5 |
+ mov %dl, (%rdi)
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Fill2):
|
|
|
e354a5 |
+ mov %dx, (%rdi)
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Fill3_4):
|
|
|
e354a5 |
+ mov %dx, (%rdi)
|
|
|
e354a5 |
+ mov %dx, -2(%rdi, %r8)
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Fill5_8):
|
|
|
e354a5 |
+ mov %edx, (%rdi)
|
|
|
e354a5 |
+ mov %edx, -4(%rdi, %r8)
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Fill9_16):
|
|
|
e354a5 |
+ mov %rdx, (%rdi)
|
|
|
e354a5 |
+ mov %rdx, -8(%rdi, %r8)
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(Fill17_32):
|
|
|
e354a5 |
+ vmovdqu %xmmZ, (%rdi)
|
|
|
e354a5 |
+ vmovdqu %xmmZ, -16(%rdi, %r8)
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeUnalignedVec2):
|
|
|
e354a5 |
+ vmovdqu %ymm2, (%rdi, %rcx)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(CopyVecSizeVecExit):
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+ add $(VEC_SIZE - 1), %r8
|
|
|
e354a5 |
+ add %rcx, %rdi
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (%rdi, %rdx), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ sub %rdx, %r8
|
|
|
e354a5 |
+ lea 1(%rdi, %rdx), %rdi
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyFillTailWithZero):
|
|
|
e354a5 |
+ xor %edx, %edx
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jbe L(StrncpyFillExit)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vmovdqu %ymmZ, (%rdi)
|
|
|
e354a5 |
+ add $VEC_SIZE, %rdi
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ mov %rdi, %rsi
|
|
|
e354a5 |
+ and $(VEC_SIZE - 1), %esi
|
|
|
e354a5 |
+ sub %rsi, %rdi
|
|
|
e354a5 |
+ add %rsi, %r8
|
|
|
e354a5 |
+ sub $(VEC_SIZE * 4), %r8
|
|
|
e354a5 |
+ jb L(StrncpyFillLessFourVecSize)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+L(StrncpyFillLoopVmovdqa):
|
|
|
e354a5 |
+ vmovdqa %ymmZ, (%rdi)
|
|
|
e354a5 |
+ vmovdqa %ymmZ, VEC_SIZE(%rdi)
|
|
|
e354a5 |
+ vmovdqa %ymmZ, (VEC_SIZE * 2)(%rdi)
|
|
|
e354a5 |
+ vmovdqa %ymmZ, (VEC_SIZE * 3)(%rdi)
|
|
|
e354a5 |
+ add $(VEC_SIZE * 4), %rdi
|
|
|
e354a5 |
+ sub $(VEC_SIZE * 4), %r8
|
|
|
e354a5 |
+ jae L(StrncpyFillLoopVmovdqa)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+L(StrncpyFillLessFourVecSize):
|
|
|
e354a5 |
+ add $(VEC_SIZE * 2), %r8
|
|
|
e354a5 |
+ jl L(StrncpyFillLessTwoVecSize)
|
|
|
e354a5 |
+ vmovdqa %ymmZ, (%rdi)
|
|
|
e354a5 |
+ vmovdqa %ymmZ, VEC_SIZE(%rdi)
|
|
|
e354a5 |
+ add $(VEC_SIZE * 2), %rdi
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jl L(StrncpyFillExit)
|
|
|
e354a5 |
+ vmovdqa %ymmZ, (%rdi)
|
|
|
e354a5 |
+ add $VEC_SIZE, %rdi
|
|
|
e354a5 |
+ jmp L(Fill)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyFillLessTwoVecSize):
|
|
|
e354a5 |
+ add $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jl L(StrncpyFillExit)
|
|
|
e354a5 |
+ vmovdqa %ymmZ, (%rdi)
|
|
|
e354a5 |
+ add $VEC_SIZE, %rdi
|
|
|
e354a5 |
+ jmp L(Fill)
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(StrncpyFillExit):
|
|
|
e354a5 |
+ add $VEC_SIZE, %r8
|
|
|
e354a5 |
+L(Fill):
|
|
|
e354a5 |
+ cmp $17, %r8d
|
|
|
e354a5 |
+ jae L(Fill17_32)
|
|
|
e354a5 |
+ cmp $9, %r8d
|
|
|
e354a5 |
+ jae L(Fill9_16)
|
|
|
e354a5 |
+ cmp $5, %r8d
|
|
|
e354a5 |
+ jae L(Fill5_8)
|
|
|
e354a5 |
+ cmp $3, %r8d
|
|
|
e354a5 |
+ jae L(Fill3_4)
|
|
|
e354a5 |
+ cmp $1, %r8d
|
|
|
e354a5 |
+ ja L(Fill2)
|
|
|
e354a5 |
+ je L(Fill1)
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+/* end of ifndef USE_AS_STRCAT */
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(UnalignedLeaveCase2OrCase3):
|
|
|
e354a5 |
+ test %rdx, %rdx
|
|
|
e354a5 |
+ jnz L(UnalignedFourVecSizeLeaveCase2)
|
|
|
e354a5 |
+L(UnalignedFourVecSizeLeaveCase3):
|
|
|
e354a5 |
+ lea (VEC_SIZE * 4)(%r8), %rcx
|
|
|
e354a5 |
+ and $-VEC_SIZE, %rcx
|
|
|
e354a5 |
+ add $(VEC_SIZE * 3), %r8
|
|
|
e354a5 |
+ jl L(CopyVecSizeCase3)
|
|
|
e354a5 |
+ vmovdqu %ymm4, (%rdi)
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jb L(CopyVecSizeCase3)
|
|
|
e354a5 |
+ vmovdqu %ymm5, VEC_SIZE(%rdi)
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jb L(CopyVecSizeCase3)
|
|
|
e354a5 |
+ vmovdqu %ymm6, (VEC_SIZE * 2)(%rdi)
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jb L(CopyVecSizeCase3)
|
|
|
e354a5 |
+ vmovdqu %ymm7, (VEC_SIZE * 3)(%rdi)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ lea (VEC_SIZE * 4)(%rdi), %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, (VEC_SIZE * 4)(%rdi)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(UnalignedFourVecSizeLeaveCase2):
|
|
|
e354a5 |
+ xor %ecx, %ecx
|
|
|
e354a5 |
+ vpcmpeqb %ymm4, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ add $(VEC_SIZE * 3), %r8
|
|
|
e354a5 |
+ jle L(CopyVecSizeCase2OrCase3)
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+# ifndef USE_AS_STRCAT
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnalignedVec4)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jnz L(CopyVecSize)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ vpcmpeqb %ymm5, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ vmovdqu %ymm4, (%rdi)
|
|
|
e354a5 |
+ add $VEC_SIZE, %rcx
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jbe L(CopyVecSizeCase2OrCase3)
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+# ifndef USE_AS_STRCAT
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnalignedVec5)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jnz L(CopyVecSize)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb %ymm6, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ vmovdqu %ymm5, VEC_SIZE(%rdi)
|
|
|
e354a5 |
+ add $VEC_SIZE, %rcx
|
|
|
e354a5 |
+ sub $VEC_SIZE, %r8
|
|
|
e354a5 |
+ jbe L(CopyVecSizeCase2OrCase3)
|
|
|
e354a5 |
+ test %edx, %edx
|
|
|
e354a5 |
+# ifndef USE_AS_STRCAT
|
|
|
e354a5 |
+ jnz L(CopyVecSizeUnalignedVec6)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+ jnz L(CopyVecSize)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ vpcmpeqb %ymm7, %ymmZ, %ymmM
|
|
|
e354a5 |
+ vpmovmskb %ymmM, %edx
|
|
|
e354a5 |
+ vmovdqu %ymm6, (VEC_SIZE * 2)(%rdi)
|
|
|
e354a5 |
+ lea VEC_SIZE(%rdi, %rcx), %rdi
|
|
|
e354a5 |
+ lea VEC_SIZE(%rsi, %rcx), %rsi
|
|
|
e354a5 |
+ bsf %edx, %edx
|
|
|
e354a5 |
+ cmp %r8d, %edx
|
|
|
e354a5 |
+ jb L(CopyVecSizeExit)
|
|
|
e354a5 |
+L(StrncpyExit):
|
|
|
e354a5 |
+ cmp $65, %r8d
|
|
|
e354a5 |
+ je L(StrncpyExit65)
|
|
|
e354a5 |
+ cmp $33, %r8d
|
|
|
e354a5 |
+ jae L(StrncpyExit33_64)
|
|
|
e354a5 |
+ cmp $17, %r8d
|
|
|
e354a5 |
+ jae L(StrncpyExit17_32)
|
|
|
e354a5 |
+ cmp $9, %r8d
|
|
|
e354a5 |
+ jae L(StrncpyExit9_16)
|
|
|
e354a5 |
+ cmp $5, %r8d
|
|
|
e354a5 |
+ jae L(StrncpyExit5_8)
|
|
|
e354a5 |
+ cmp $3, %r8d
|
|
|
e354a5 |
+ jae L(StrncpyExit3_4)
|
|
|
e354a5 |
+ cmp $1, %r8d
|
|
|
e354a5 |
+ ja L(StrncpyExit2)
|
|
|
e354a5 |
+ je L(StrncpyExit1)
|
|
|
e354a5 |
+# ifdef USE_AS_STPCPY
|
|
|
e354a5 |
+ mov %rdi, %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+# ifdef USE_AS_STRCAT
|
|
|
e354a5 |
+ movb $0, (%rdi)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+ .p2align 4
|
|
|
e354a5 |
+L(ExitZero):
|
|
|
e354a5 |
+# ifndef USE_AS_STRCAT
|
|
|
e354a5 |
+ mov %rdi, %rax
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+ VZEROUPPER
|
|
|
e354a5 |
+ ret
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+
|
|
|
e354a5 |
+# ifndef USE_AS_STRCAT
|
|
|
e354a5 |
+END (STRCPY)
|
|
|
e354a5 |
+# else
|
|
|
e354a5 |
+END (STRCAT)
|
|
|
e354a5 |
+# endif
|
|
|
e354a5 |
+#endif
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/strcpy.c b/sysdeps/x86_64/multiarch/strcpy.c
|
|
|
e354a5 |
index 12e0e3ffe20602c6..ecf90d4b044a1b01 100644
|
|
|
e354a5 |
--- a/sysdeps/x86_64/multiarch/strcpy.c
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/strcpy.c
|
|
|
e354a5 |
@@ -24,7 +24,7 @@
|
|
|
e354a5 |
# undef strcpy
|
|
|
e354a5 |
|
|
|
e354a5 |
# define SYMBOL_NAME strcpy
|
|
|
e354a5 |
-# include "ifunc-unaligned-ssse3.h"
|
|
|
e354a5 |
+# include "ifunc-strcpy.h"
|
|
|
e354a5 |
|
|
|
e354a5 |
libc_ifunc_redirected (__redirect_strcpy, strcpy, IFUNC_SELECTOR ());
|
|
|
e354a5 |
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/strncat-avx2.S b/sysdeps/x86_64/multiarch/strncat-avx2.S
|
|
|
e354a5 |
new file mode 100644
|
|
|
e354a5 |
index 0000000000000000..bfefa659bb6281fa
|
|
|
e354a5 |
--- /dev/null
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/strncat-avx2.S
|
|
|
e354a5 |
@@ -0,0 +1,3 @@
|
|
|
e354a5 |
+#define USE_AS_STRNCAT
|
|
|
e354a5 |
+#define STRCAT __strncat_avx2
|
|
|
e354a5 |
+#include "strcat-avx2.S"
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/strncat.c b/sysdeps/x86_64/multiarch/strncat.c
|
|
|
e354a5 |
index 841c165565add132..74f7d028ae23d700 100644
|
|
|
e354a5 |
--- a/sysdeps/x86_64/multiarch/strncat.c
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/strncat.c
|
|
|
e354a5 |
@@ -24,7 +24,7 @@
|
|
|
e354a5 |
# undef strncat
|
|
|
e354a5 |
|
|
|
e354a5 |
# define SYMBOL_NAME strncat
|
|
|
e354a5 |
-# include "ifunc-unaligned-ssse3.h"
|
|
|
e354a5 |
+# include "ifunc-strcpy.h"
|
|
|
e354a5 |
|
|
|
e354a5 |
libc_ifunc_redirected (__redirect_strncat, strncat, IFUNC_SELECTOR ());
|
|
|
e354a5 |
strong_alias (strncat, __strncat);
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/strncpy-avx2.S b/sysdeps/x86_64/multiarch/strncpy-avx2.S
|
|
|
e354a5 |
new file mode 100644
|
|
|
e354a5 |
index 0000000000000000..9ef8c87627dc4924
|
|
|
e354a5 |
--- /dev/null
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/strncpy-avx2.S
|
|
|
e354a5 |
@@ -0,0 +1,3 @@
|
|
|
e354a5 |
+#define USE_AS_STRNCPY
|
|
|
e354a5 |
+#define STRCPY __strncpy_avx2
|
|
|
e354a5 |
+#include "strcpy-avx2.S"
|
|
|
e354a5 |
diff --git a/sysdeps/x86_64/multiarch/strncpy.c b/sysdeps/x86_64/multiarch/strncpy.c
|
|
|
e354a5 |
index 3c3de8b18ebb177f..93dfb4cfde79467a 100644
|
|
|
e354a5 |
--- a/sysdeps/x86_64/multiarch/strncpy.c
|
|
|
e354a5 |
+++ b/sysdeps/x86_64/multiarch/strncpy.c
|
|
|
e354a5 |
@@ -24,7 +24,7 @@
|
|
|
e354a5 |
# undef strncpy
|
|
|
e354a5 |
|
|
|
e354a5 |
# define SYMBOL_NAME strncpy
|
|
|
e354a5 |
-# include "ifunc-unaligned-ssse3.h"
|
|
|
e354a5 |
+# include "ifunc-strcpy.h"
|
|
|
e354a5 |
|
|
|
e354a5 |
libc_ifunc_redirected (__redirect_strncpy, strncpy, IFUNC_SELECTOR ());
|
|
|
e354a5 |
|