|
 |
08c3a6 |
commit 7079931c51547854323fe2ed6fdccf2a1b8b04d7
|
|
|
08c3a6 |
Author: Noah Goldstein <goldstein.w.n@gmail.com>
|
|
|
08c3a6 |
Date: Wed Jun 29 16:07:05 2022 -0700
|
|
|
08c3a6 |
|
|
|
08c3a6 |
x86: Move and slightly improve memset_erms
|
|
|
08c3a6 |
|
|
|
08c3a6 |
Implementation wise:
|
|
|
08c3a6 |
1. Remove the VZEROUPPER as memset_{impl}_unaligned_erms does not
|
|
|
08c3a6 |
use the L(stosb) label that was previously defined.
|
|
|
08c3a6 |
|
|
|
08c3a6 |
2. Don't give the hotpath (fallthrough) to zero size.
|
|
|
08c3a6 |
|
|
|
08c3a6 |
Code positioning wise:
|
|
|
08c3a6 |
|
|
|
08c3a6 |
Move memset_{chk}_erms to its own file. Leaving it in between the
|
|
|
08c3a6 |
memset_{impl}_unaligned both adds unnecessary complexity to the
|
|
|
08c3a6 |
file and wastes space in a relatively hot cache section.
|
|
|
08c3a6 |
|
|
|
08c3a6 |
(cherry picked from commit 4a3f29e7e475dd4e7cce2a24c187e6fb7b5b0a05)
|
|
|
08c3a6 |
|
|
|
08c3a6 |
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
|
|
|
08c3a6 |
index 0e39e63ef6be6a86..da9f16286a763556 100644
|
|
|
08c3a6 |
--- a/sysdeps/x86_64/multiarch/Makefile
|
|
|
08c3a6 |
+++ b/sysdeps/x86_64/multiarch/Makefile
|
|
|
08c3a6 |
@@ -29,6 +29,7 @@ sysdep_routines += \
|
|
|
08c3a6 |
memset-avx2-unaligned-erms-rtm \
|
|
|
08c3a6 |
memset-avx512-no-vzeroupper \
|
|
|
08c3a6 |
memset-avx512-unaligned-erms \
|
|
|
08c3a6 |
+ memset-erms \
|
|
|
08c3a6 |
memset-evex-unaligned-erms \
|
|
|
08c3a6 |
memset-sse2-unaligned-erms \
|
|
|
08c3a6 |
rawmemchr-avx2 \
|
|
|
08c3a6 |
diff --git a/sysdeps/x86_64/multiarch/memset-erms.S b/sysdeps/x86_64/multiarch/memset-erms.S
|
|
|
08c3a6 |
new file mode 100644
|
|
|
08c3a6 |
index 0000000000000000..e83cccc731f0a7ea
|
|
|
08c3a6 |
--- /dev/null
|
|
|
08c3a6 |
+++ b/sysdeps/x86_64/multiarch/memset-erms.S
|
|
|
08c3a6 |
@@ -0,0 +1,44 @@
|
|
|
08c3a6 |
+/* memset implement with rep stosb
|
|
|
08c3a6 |
+ Copyright (C) 2022 Free Software Foundation, Inc.
|
|
|
08c3a6 |
+ This file is part of the GNU C Library.
|
|
|
08c3a6 |
+
|
|
|
08c3a6 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
08c3a6 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
08c3a6 |
+ License as published by the Free Software Foundation; either
|
|
|
08c3a6 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
08c3a6 |
+
|
|
|
08c3a6 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
08c3a6 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
08c3a6 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
08c3a6 |
+ Lesser General Public License for more details.
|
|
|
08c3a6 |
+
|
|
|
08c3a6 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
08c3a6 |
+ License along with the GNU C Library; if not, see
|
|
|
08c3a6 |
+ <https://www.gnu.org/licenses/>. */
|
|
|
08c3a6 |
+
|
|
|
08c3a6 |
+
|
|
|
08c3a6 |
+#include <sysdep.h>
|
|
|
08c3a6 |
+
|
|
|
08c3a6 |
+#if defined USE_MULTIARCH && IS_IN (libc)
|
|
|
08c3a6 |
+ .text
|
|
|
08c3a6 |
+ENTRY (__memset_chk_erms)
|
|
|
08c3a6 |
+ cmp %RDX_LP, %RCX_LP
|
|
|
08c3a6 |
+ jb HIDDEN_JUMPTARGET (__chk_fail)
|
|
|
08c3a6 |
+END (__memset_chk_erms)
|
|
|
08c3a6 |
+
|
|
|
08c3a6 |
+/* Only used to measure performance of REP STOSB. */
|
|
|
08c3a6 |
+ENTRY (__memset_erms)
|
|
|
08c3a6 |
+ /* Skip zero length. */
|
|
|
08c3a6 |
+ test %RDX_LP, %RDX_LP
|
|
|
08c3a6 |
+ jz L(stosb_return_zero)
|
|
|
08c3a6 |
+ mov %RDX_LP, %RCX_LP
|
|
|
08c3a6 |
+ movzbl %sil, %eax
|
|
|
08c3a6 |
+ mov %RDI_LP, %RDX_LP
|
|
|
08c3a6 |
+ rep stosb
|
|
|
08c3a6 |
+ mov %RDX_LP, %RAX_LP
|
|
|
08c3a6 |
+ ret
|
|
|
08c3a6 |
+L(stosb_return_zero):
|
|
|
08c3a6 |
+ movq %rdi, %rax
|
|
|
08c3a6 |
+ ret
|
|
|
08c3a6 |
+END (__memset_erms)
|
|
|
08c3a6 |
+#endif
|
|
|
08c3a6 |
diff --git a/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S b/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
|
|
|
08c3a6 |
index abc12d9cda1b3843..905d0fa4643d5768 100644
|
|
|
08c3a6 |
--- a/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
|
|
|
08c3a6 |
+++ b/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
|
|
|
08c3a6 |
@@ -156,37 +156,6 @@ L(entry_from_wmemset):
|
|
|
08c3a6 |
#if defined USE_MULTIARCH && IS_IN (libc)
|
|
|
08c3a6 |
END (MEMSET_SYMBOL (__memset, unaligned))
|
|
|
08c3a6 |
|
|
|
08c3a6 |
-# if VEC_SIZE == 16
|
|
|
08c3a6 |
-ENTRY (__memset_chk_erms)
|
|
|
08c3a6 |
- cmp %RDX_LP, %RCX_LP
|
|
|
08c3a6 |
- jb HIDDEN_JUMPTARGET (__chk_fail)
|
|
|
08c3a6 |
-END (__memset_chk_erms)
|
|
|
08c3a6 |
-
|
|
|
08c3a6 |
-/* Only used to measure performance of REP STOSB. */
|
|
|
08c3a6 |
-ENTRY (__memset_erms)
|
|
|
08c3a6 |
- /* Skip zero length. */
|
|
|
08c3a6 |
- test %RDX_LP, %RDX_LP
|
|
|
08c3a6 |
- jnz L(stosb)
|
|
|
08c3a6 |
- movq %rdi, %rax
|
|
|
08c3a6 |
- ret
|
|
|
08c3a6 |
-# else
|
|
|
08c3a6 |
-/* Provide a hidden symbol to debugger. */
|
|
|
08c3a6 |
- .hidden MEMSET_SYMBOL (__memset, erms)
|
|
|
08c3a6 |
-ENTRY (MEMSET_SYMBOL (__memset, erms))
|
|
|
08c3a6 |
-# endif
|
|
|
08c3a6 |
-L(stosb):
|
|
|
08c3a6 |
- mov %RDX_LP, %RCX_LP
|
|
|
08c3a6 |
- movzbl %sil, %eax
|
|
|
08c3a6 |
- mov %RDI_LP, %RDX_LP
|
|
|
08c3a6 |
- rep stosb
|
|
|
08c3a6 |
- mov %RDX_LP, %RAX_LP
|
|
|
08c3a6 |
- VZEROUPPER_RETURN
|
|
|
08c3a6 |
-# if VEC_SIZE == 16
|
|
|
08c3a6 |
-END (__memset_erms)
|
|
|
08c3a6 |
-# else
|
|
|
08c3a6 |
-END (MEMSET_SYMBOL (__memset, erms))
|
|
|
08c3a6 |
-# endif
|
|
|
08c3a6 |
-
|
|
|
08c3a6 |
# if defined SHARED && IS_IN (libc)
|
|
|
08c3a6 |
ENTRY_CHK (MEMSET_CHK_SYMBOL (__memset_chk, unaligned_erms))
|
|
|
08c3a6 |
cmp %RDX_LP, %RCX_LP
|