From: Noah Goldstein <goldstein.w.n@gmail.com>
To: libc-alpha@sourceware.org
Subject: [PATCH v1] x86: Move strcat SSE2 implementation to multiarch/strcat-sse2.S
Date: Tue, 12 Jul 2022 12:29:06 -0700 [thread overview]
Message-ID: <20220712192910.351121-6-goldstein.w.n@gmail.com> (raw)
In-Reply-To: <20220712192910.351121-1-goldstein.w.n@gmail.com>
This commit doesn't affect libc.so.6, its just housekeeping to prepare
for adding explicit ISA level support.
Tested build on x86_64 and x86_32 with/without multiarch.
---
sysdeps/x86_64/multiarch/strcat-sse2.S | 242 ++++++++++++++++++++++++-
sysdeps/x86_64/strcat.S | 239 +-----------------------
2 files changed, 238 insertions(+), 243 deletions(-)
diff --git a/sysdeps/x86_64/multiarch/strcat-sse2.S b/sysdeps/x86_64/multiarch/strcat-sse2.S
index 449e102438..244c4a6d74 100644
--- a/sysdeps/x86_64/multiarch/strcat-sse2.S
+++ b/sysdeps/x86_64/multiarch/strcat-sse2.S
@@ -17,12 +17,242 @@
<https://www.gnu.org/licenses/>. */
#if IS_IN (libc)
+# ifndef STRCAT
+# define STRCAT __strcat_sse2
+# endif
+#endif
-# include <sysdep.h>
-# define strcat __strcat_sse2
+#include <sysdep.h>
+
+ .text
+ENTRY (STRCAT)
+ movq %rdi, %rcx /* Dest. register. */
+ andl $7, %ecx /* mask alignment bits */
+ movq %rdi, %rax /* Duplicate destination pointer. */
+ movq $0xfefefefefefefeff,%r8
+
+ /* First step: Find end of destination. */
+ jz 4f /* aligned => start loop */
+
+ neg %ecx /* We need to align to 8 bytes. */
+ addl $8,%ecx
+ /* Search the first bytes directly. */
+0: cmpb $0x0,(%rax) /* is byte NUL? */
+ je 2f /* yes => start copy */
+ incq %rax /* increment pointer */
+ decl %ecx
+ jnz 0b
+
+
+
+ /* Now the source is aligned. Scan for NUL byte. */
+ .p2align 4
+4:
+ /* First unroll. */
+ movq (%rax), %rcx /* get double word (= 8 bytes) in question */
+ addq $8,%rax /* adjust pointer for next word */
+ movq %r8, %rdx /* magic value */
+ addq %rcx, %rdx /* add the magic value to the word. We get
+ carry bits reported for each byte which
+ is *not* 0 */
+ jnc 3f /* highest byte is NUL => return pointer */
+ xorq %rcx, %rdx /* (word+magic)^word */
+ orq %r8, %rdx /* set all non-carry bits */
+ incq %rdx /* add 1: if one carry bit was *not* set
+ the addition will not result in 0. */
+ jnz 3f /* found NUL => return pointer */
+
+ /* Second unroll. */
+ movq (%rax), %rcx /* get double word (= 8 bytes) in question */
+ addq $8,%rax /* adjust pointer for next word */
+ movq %r8, %rdx /* magic value */
+ addq %rcx, %rdx /* add the magic value to the word. We get
+ carry bits reported for each byte which
+ is *not* 0 */
+ jnc 3f /* highest byte is NUL => return pointer */
+ xorq %rcx, %rdx /* (word+magic)^word */
+ orq %r8, %rdx /* set all non-carry bits */
+ incq %rdx /* add 1: if one carry bit was *not* set
+ the addition will not result in 0. */
+ jnz 3f /* found NUL => return pointer */
+
+ /* Third unroll. */
+ movq (%rax), %rcx /* get double word (= 8 bytes) in question */
+ addq $8,%rax /* adjust pointer for next word */
+ movq %r8, %rdx /* magic value */
+ addq %rcx, %rdx /* add the magic value to the word. We get
+ carry bits reported for each byte which
+ is *not* 0 */
+ jnc 3f /* highest byte is NUL => return pointer */
+ xorq %rcx, %rdx /* (word+magic)^word */
+ orq %r8, %rdx /* set all non-carry bits */
+ incq %rdx /* add 1: if one carry bit was *not* set
+ the addition will not result in 0. */
+ jnz 3f /* found NUL => return pointer */
+
+ /* Fourth unroll. */
+ movq (%rax), %rcx /* get double word (= 8 bytes) in question */
+ addq $8,%rax /* adjust pointer for next word */
+ movq %r8, %rdx /* magic value */
+ addq %rcx, %rdx /* add the magic value to the word. We get
+ carry bits reported for each byte which
+ is *not* 0 */
+ jnc 3f /* highest byte is NUL => return pointer */
+ xorq %rcx, %rdx /* (word+magic)^word */
+ orq %r8, %rdx /* set all non-carry bits */
+ incq %rdx /* add 1: if one carry bit was *not* set
+ the addition will not result in 0. */
+ jz 4b /* no NUL found => continue loop */
+
+ .p2align 4 /* Align, it's a jump target. */
+3: subq $8,%rax /* correct pointer increment. */
+
+ testb %cl, %cl /* is first byte NUL? */
+ jz 2f /* yes => return */
+ incq %rax /* increment pointer */
+
+ testb %ch, %ch /* is second byte NUL? */
+ jz 2f /* yes => return */
+ incq %rax /* increment pointer */
+
+ testl $0x00ff0000, %ecx /* is third byte NUL? */
+ jz 2f /* yes => return pointer */
+ incq %rax /* increment pointer */
+
+ testl $0xff000000, %ecx /* is fourth byte NUL? */
+ jz 2f /* yes => return pointer */
+ incq %rax /* increment pointer */
+
+ shrq $32, %rcx /* look at other half. */
+
+ testb %cl, %cl /* is first byte NUL? */
+ jz 2f /* yes => return */
+ incq %rax /* increment pointer */
+
+ testb %ch, %ch /* is second byte NUL? */
+ jz 2f /* yes => return */
+ incq %rax /* increment pointer */
+
+ testl $0xff0000, %ecx /* is third byte NUL? */
+ jz 2f /* yes => return pointer */
+ incq %rax /* increment pointer */
+
+2:
+ /* Second step: Copy source to destination. */
+
+ movq %rsi, %rcx /* duplicate */
+ andl $7,%ecx /* mask alignment bits */
+ movq %rax, %rdx /* move around */
+ jz 22f /* aligned => start loop */
+
+ neg %ecx /* align to 8 bytes. */
+ addl $8, %ecx
+ /* Align the source pointer. */
+21:
+ movb (%rsi), %al /* Fetch a byte */
+ testb %al, %al /* Is it NUL? */
+ movb %al, (%rdx) /* Store it */
+ jz 24f /* If it was NUL, done! */
+ incq %rsi
+ incq %rdx
+ decl %ecx
+ jnz 21b
+
+ /* Now the sources is aligned. Unfortunatly we cannot force
+ to have both source and destination aligned, so ignore the
+ alignment of the destination. */
+ .p2align 4
+22:
+ /* 1st unroll. */
+ movq (%rsi), %rax /* Read double word (8 bytes). */
+ addq $8, %rsi /* Adjust pointer for next word. */
+ movq %rax, %r9 /* Save a copy for NUL finding. */
+ addq %r8, %r9 /* add the magic value to the word. We get
+ carry bits reported for each byte which
+ is *not* 0 */
+ jnc 23f /* highest byte is NUL => return pointer */
+ xorq %rax, %r9 /* (word+magic)^word */
+ orq %r8, %r9 /* set all non-carry bits */
+ incq %r9 /* add 1: if one carry bit was *not* set
+ the addition will not result in 0. */
+
+ jnz 23f /* found NUL => return pointer */
+
+ movq %rax, (%rdx) /* Write value to destination. */
+ addq $8, %rdx /* Adjust pointer. */
+
+ /* 2nd unroll. */
+ movq (%rsi), %rax /* Read double word (8 bytes). */
+ addq $8, %rsi /* Adjust pointer for next word. */
+ movq %rax, %r9 /* Save a copy for NUL finding. */
+ addq %r8, %r9 /* add the magic value to the word. We get
+ carry bits reported for each byte which
+ is *not* 0 */
+ jnc 23f /* highest byte is NUL => return pointer */
+ xorq %rax, %r9 /* (word+magic)^word */
+ orq %r8, %r9 /* set all non-carry bits */
+ incq %r9 /* add 1: if one carry bit was *not* set
+ the addition will not result in 0. */
+
+ jnz 23f /* found NUL => return pointer */
+
+ movq %rax, (%rdx) /* Write value to destination. */
+ addq $8, %rdx /* Adjust pointer. */
+
+ /* 3rd unroll. */
+ movq (%rsi), %rax /* Read double word (8 bytes). */
+ addq $8, %rsi /* Adjust pointer for next word. */
+ movq %rax, %r9 /* Save a copy for NUL finding. */
+ addq %r8, %r9 /* add the magic value to the word. We get
+ carry bits reported for each byte which
+ is *not* 0 */
+ jnc 23f /* highest byte is NUL => return pointer */
+ xorq %rax, %r9 /* (word+magic)^word */
+ orq %r8, %r9 /* set all non-carry bits */
+ incq %r9 /* add 1: if one carry bit was *not* set
+ the addition will not result in 0. */
+
+ jnz 23f /* found NUL => return pointer */
+
+ movq %rax, (%rdx) /* Write value to destination. */
+ addq $8, %rdx /* Adjust pointer. */
+
+ /* 4th unroll. */
+ movq (%rsi), %rax /* Read double word (8 bytes). */
+ addq $8, %rsi /* Adjust pointer for next word. */
+ movq %rax, %r9 /* Save a copy for NUL finding. */
+ addq %r8, %r9 /* add the magic value to the word. We get
+ carry bits reported for each byte which
+ is *not* 0 */
+ jnc 23f /* highest byte is NUL => return pointer */
+ xorq %rax, %r9 /* (word+magic)^word */
+ orq %r8, %r9 /* set all non-carry bits */
+ incq %r9 /* add 1: if one carry bit was *not* set
+ the addition will not result in 0. */
+
+ jnz 23f /* found NUL => return pointer */
+
+ movq %rax, (%rdx) /* Write value to destination. */
+ addq $8, %rdx /* Adjust pointer. */
+ jmp 22b /* Next iteration. */
+
+ /* Do the last few bytes. %rax contains the value to write.
+ The loop is unrolled twice. */
+ .p2align 4
+23:
+ movb %al, (%rdx) /* 1st byte. */
+ testb %al, %al /* Is it NUL. */
+ jz 24f /* yes, finish. */
+ incq %rdx /* Increment destination. */
+ movb %ah, (%rdx) /* 2nd byte. */
+ testb %ah, %ah /* Is it NUL?. */
+ jz 24f /* yes, finish. */
+ incq %rdx /* Increment destination. */
+ shrq $16, %rax /* Shift... */
+ jmp 23b /* and look at next two bytes in %rax. */
-# undef libc_hidden_builtin_def
-# define libc_hidden_builtin_def(strcat)
-#endif
-#include <sysdeps/x86_64/strcat.S>
+24:
+ movq %rdi, %rax /* Source is return value. */
+ retq
+END (STRCAT)
diff --git a/sysdeps/x86_64/strcat.S b/sysdeps/x86_64/strcat.S
index 565a9c785a..fc3e8a9bcf 100644
--- a/sysdeps/x86_64/strcat.S
+++ b/sysdeps/x86_64/strcat.S
@@ -17,241 +17,6 @@
License along with the GNU C Library; if not, see
<https://www.gnu.org/licenses/>. */
-#include <sysdep.h>
-#include "asm-syntax.h"
-
-/* Will be removed when new strcpy implementation gets merged. */
-
- .text
-ENTRY (strcat)
- movq %rdi, %rcx /* Dest. register. */
- andl $7, %ecx /* mask alignment bits */
- movq %rdi, %rax /* Duplicate destination pointer. */
- movq $0xfefefefefefefeff,%r8
-
- /* First step: Find end of destination. */
- jz 4f /* aligned => start loop */
-
- neg %ecx /* We need to align to 8 bytes. */
- addl $8,%ecx
- /* Search the first bytes directly. */
-0: cmpb $0x0,(%rax) /* is byte NUL? */
- je 2f /* yes => start copy */
- incq %rax /* increment pointer */
- decl %ecx
- jnz 0b
-
-
-
- /* Now the source is aligned. Scan for NUL byte. */
- .p2align 4
-4:
- /* First unroll. */
- movq (%rax), %rcx /* get double word (= 8 bytes) in question */
- addq $8,%rax /* adjust pointer for next word */
- movq %r8, %rdx /* magic value */
- addq %rcx, %rdx /* add the magic value to the word. We get
- carry bits reported for each byte which
- is *not* 0 */
- jnc 3f /* highest byte is NUL => return pointer */
- xorq %rcx, %rdx /* (word+magic)^word */
- orq %r8, %rdx /* set all non-carry bits */
- incq %rdx /* add 1: if one carry bit was *not* set
- the addition will not result in 0. */
- jnz 3f /* found NUL => return pointer */
-
- /* Second unroll. */
- movq (%rax), %rcx /* get double word (= 8 bytes) in question */
- addq $8,%rax /* adjust pointer for next word */
- movq %r8, %rdx /* magic value */
- addq %rcx, %rdx /* add the magic value to the word. We get
- carry bits reported for each byte which
- is *not* 0 */
- jnc 3f /* highest byte is NUL => return pointer */
- xorq %rcx, %rdx /* (word+magic)^word */
- orq %r8, %rdx /* set all non-carry bits */
- incq %rdx /* add 1: if one carry bit was *not* set
- the addition will not result in 0. */
- jnz 3f /* found NUL => return pointer */
-
- /* Third unroll. */
- movq (%rax), %rcx /* get double word (= 8 bytes) in question */
- addq $8,%rax /* adjust pointer for next word */
- movq %r8, %rdx /* magic value */
- addq %rcx, %rdx /* add the magic value to the word. We get
- carry bits reported for each byte which
- is *not* 0 */
- jnc 3f /* highest byte is NUL => return pointer */
- xorq %rcx, %rdx /* (word+magic)^word */
- orq %r8, %rdx /* set all non-carry bits */
- incq %rdx /* add 1: if one carry bit was *not* set
- the addition will not result in 0. */
- jnz 3f /* found NUL => return pointer */
-
- /* Fourth unroll. */
- movq (%rax), %rcx /* get double word (= 8 bytes) in question */
- addq $8,%rax /* adjust pointer for next word */
- movq %r8, %rdx /* magic value */
- addq %rcx, %rdx /* add the magic value to the word. We get
- carry bits reported for each byte which
- is *not* 0 */
- jnc 3f /* highest byte is NUL => return pointer */
- xorq %rcx, %rdx /* (word+magic)^word */
- orq %r8, %rdx /* set all non-carry bits */
- incq %rdx /* add 1: if one carry bit was *not* set
- the addition will not result in 0. */
- jz 4b /* no NUL found => continue loop */
-
- .p2align 4 /* Align, it's a jump target. */
-3: subq $8,%rax /* correct pointer increment. */
-
- testb %cl, %cl /* is first byte NUL? */
- jz 2f /* yes => return */
- incq %rax /* increment pointer */
-
- testb %ch, %ch /* is second byte NUL? */
- jz 2f /* yes => return */
- incq %rax /* increment pointer */
-
- testl $0x00ff0000, %ecx /* is third byte NUL? */
- jz 2f /* yes => return pointer */
- incq %rax /* increment pointer */
-
- testl $0xff000000, %ecx /* is fourth byte NUL? */
- jz 2f /* yes => return pointer */
- incq %rax /* increment pointer */
-
- shrq $32, %rcx /* look at other half. */
-
- testb %cl, %cl /* is first byte NUL? */
- jz 2f /* yes => return */
- incq %rax /* increment pointer */
-
- testb %ch, %ch /* is second byte NUL? */
- jz 2f /* yes => return */
- incq %rax /* increment pointer */
-
- testl $0xff0000, %ecx /* is third byte NUL? */
- jz 2f /* yes => return pointer */
- incq %rax /* increment pointer */
-
-2:
- /* Second step: Copy source to destination. */
-
- movq %rsi, %rcx /* duplicate */
- andl $7,%ecx /* mask alignment bits */
- movq %rax, %rdx /* move around */
- jz 22f /* aligned => start loop */
-
- neg %ecx /* align to 8 bytes. */
- addl $8, %ecx
- /* Align the source pointer. */
-21:
- movb (%rsi), %al /* Fetch a byte */
- testb %al, %al /* Is it NUL? */
- movb %al, (%rdx) /* Store it */
- jz 24f /* If it was NUL, done! */
- incq %rsi
- incq %rdx
- decl %ecx
- jnz 21b
-
- /* Now the sources is aligned. Unfortunatly we cannot force
- to have both source and destination aligned, so ignore the
- alignment of the destination. */
- .p2align 4
-22:
- /* 1st unroll. */
- movq (%rsi), %rax /* Read double word (8 bytes). */
- addq $8, %rsi /* Adjust pointer for next word. */
- movq %rax, %r9 /* Save a copy for NUL finding. */
- addq %r8, %r9 /* add the magic value to the word. We get
- carry bits reported for each byte which
- is *not* 0 */
- jnc 23f /* highest byte is NUL => return pointer */
- xorq %rax, %r9 /* (word+magic)^word */
- orq %r8, %r9 /* set all non-carry bits */
- incq %r9 /* add 1: if one carry bit was *not* set
- the addition will not result in 0. */
-
- jnz 23f /* found NUL => return pointer */
-
- movq %rax, (%rdx) /* Write value to destination. */
- addq $8, %rdx /* Adjust pointer. */
-
- /* 2nd unroll. */
- movq (%rsi), %rax /* Read double word (8 bytes). */
- addq $8, %rsi /* Adjust pointer for next word. */
- movq %rax, %r9 /* Save a copy for NUL finding. */
- addq %r8, %r9 /* add the magic value to the word. We get
- carry bits reported for each byte which
- is *not* 0 */
- jnc 23f /* highest byte is NUL => return pointer */
- xorq %rax, %r9 /* (word+magic)^word */
- orq %r8, %r9 /* set all non-carry bits */
- incq %r9 /* add 1: if one carry bit was *not* set
- the addition will not result in 0. */
-
- jnz 23f /* found NUL => return pointer */
-
- movq %rax, (%rdx) /* Write value to destination. */
- addq $8, %rdx /* Adjust pointer. */
-
- /* 3rd unroll. */
- movq (%rsi), %rax /* Read double word (8 bytes). */
- addq $8, %rsi /* Adjust pointer for next word. */
- movq %rax, %r9 /* Save a copy for NUL finding. */
- addq %r8, %r9 /* add the magic value to the word. We get
- carry bits reported for each byte which
- is *not* 0 */
- jnc 23f /* highest byte is NUL => return pointer */
- xorq %rax, %r9 /* (word+magic)^word */
- orq %r8, %r9 /* set all non-carry bits */
- incq %r9 /* add 1: if one carry bit was *not* set
- the addition will not result in 0. */
-
- jnz 23f /* found NUL => return pointer */
-
- movq %rax, (%rdx) /* Write value to destination. */
- addq $8, %rdx /* Adjust pointer. */
-
- /* 4th unroll. */
- movq (%rsi), %rax /* Read double word (8 bytes). */
- addq $8, %rsi /* Adjust pointer for next word. */
- movq %rax, %r9 /* Save a copy for NUL finding. */
- addq %r8, %r9 /* add the magic value to the word. We get
- carry bits reported for each byte which
- is *not* 0 */
- jnc 23f /* highest byte is NUL => return pointer */
- xorq %rax, %r9 /* (word+magic)^word */
- orq %r8, %r9 /* set all non-carry bits */
- incq %r9 /* add 1: if one carry bit was *not* set
- the addition will not result in 0. */
-
- jnz 23f /* found NUL => return pointer */
-
- movq %rax, (%rdx) /* Write value to destination. */
- addq $8, %rdx /* Adjust pointer. */
- jmp 22b /* Next iteration. */
-
- /* Do the last few bytes. %rax contains the value to write.
- The loop is unrolled twice. */
- .p2align 4
-23:
- movb %al, (%rdx) /* 1st byte. */
- testb %al, %al /* Is it NUL. */
- jz 24f /* yes, finish. */
- incq %rdx /* Increment destination. */
- movb %ah, (%rdx) /* 2nd byte. */
- testb %ah, %ah /* Is it NUL?. */
- jz 24f /* yes, finish. */
- incq %rdx /* Increment destination. */
- shrq $16, %rax /* Shift... */
- jmp 23b /* and look at next two bytes in %rax. */
-
-
-24:
- movq %rdi, %rax /* Source is return value. */
- retq
-END (strcat)
+#define STRCAT strcat
+#include "multiarch/strcat-sse2.S"
libc_hidden_builtin_def (strcat)
--
2.34.1
next prev parent reply other threads:[~2022-07-12 19:29 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-07-12 19:29 [PATCH v1] x86: Move strlen SSE2 implementation to multiarch/strlen-sse2.S Noah Goldstein
2022-07-12 19:29 ` [PATCH v1] x86: Move strcpy SSE2 implementation to multiarch/strcpy-sse2.S Noah Goldstein
2022-07-12 23:23 ` H.J. Lu
2022-07-12 19:29 ` [PATCH v1] x86: Move memrchr SSE2 implementation to multiarch/memrchr-sse2.S Noah Goldstein
2022-07-12 22:58 ` H.J. Lu
2022-07-12 19:29 ` [PATCH v1] x86: Move strrchr SSE2 implementation to multiarch/strrchr-sse2.S Noah Goldstein
2022-07-12 22:28 ` H.J. Lu
2022-07-12 19:29 ` [PATCH v1] x86: Move strchr SSE2 implementation to multiarch/strchr-sse2.S Noah Goldstein
2022-07-12 21:27 ` H.J. Lu
2022-07-12 19:29 ` Noah Goldstein [this message]
2022-07-12 21:16 ` [PATCH v1] x86: Move strcat SSE2 implementation to multiarch/strcat-sse2.S H.J. Lu
2022-07-12 19:29 ` [PATCH v1] x86: Move wcschr SSE2 implementation to multiarch/wcschr-sse2.S Noah Goldstein
2022-07-12 20:55 ` H.J. Lu
2022-07-12 19:29 ` [PATCH v1] x86: Move wcslen SSE2 implementation to multiarch/wcslen-sse2.S Noah Goldstein
2022-07-12 20:26 ` H.J. Lu
2022-07-12 19:29 ` [PATCH v1] x86: Remove unneeded rtld-wmemcmp Noah Goldstein
2022-07-12 19:44 ` H.J. Lu
2022-07-12 19:29 ` [PATCH v1] x86: Add missing rtm tests for strcmp family Noah Goldstein
2022-07-12 19:59 ` H.J. Lu
2022-07-12 23:29 ` [PATCH v1] x86: Move strlen SSE2 implementation to multiarch/strlen-sse2.S H.J. Lu
2022-07-13 4:06 ` Noah Goldstein
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20220712192910.351121-6-goldstein.w.n@gmail.com \
--to=goldstein.w.n@gmail.com \
--cc=libc-alpha@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).