Add SSE4.2 optimized rawmemchr implementation for x86-64.

ChangeLog

@@ -1,5 +1,9 @@
 2009-06-05  Ulrich Drepper  <drepper@redhat.com>
 
+	* sysdeps/x86_64/rawmemchr.S: Minor optimization.
+
+	* sysdeps/x86_64/multiarch/rawmemchr.S: New file.
+
 	* sysdeps/x86_64/multiarch/strlen.S: New file.
 	* sysdeps/x86_64/multiarch/ifunc-defines.sym: New file.
 	* sysdeps/x86_64/multiarch/Makefile: Add rule to build ifunc-defines.h.

sysdeps/x86_64/multiarch/rawmemchr.S

@@ -0,0 +1,93 @@
+/* Copyright (C) 2009 Free Software Foundation, Inc.
+   Contributed by Ulrich Drepper <drepper@redhat.com>.
+   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, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include <ifunc-defines.h>
+
+
+/* Define multiple versions only for the definition in lib.  */
+#ifndef NOT_IN_libc
+	.text
+ENTRY(rawmemchr)
+	.type	rawmemchr, @gnu_indirect_function
+	cmpl	$0, __cpu_features+KIND_OFFSET(%rip)
+	jne	1f
+	call	__init_cpu_features
+1:	leaq	__rawmemchr_sse2(%rip), %rax
+	testl	$(1<<20), __cpu_features+CPUID_OFFSET+COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET(%rip)
+	jz	2f
+	leaq	__rawmemchr_sse42(%rip), %rax
+2:	ret
+END(rawmemchr)
+strong_alias (rawmemchr, __rawmemchr)
+
+
+	.align 	16
+	.type	__rawmemchr_sse42, @function
+__rawmemchr_sse42:
+	cfi_startproc
+	CALL_MCOUNT
+	movd	%esi, %xmm1
+	movq	%rdi, %rcx
+	punpcklbw %xmm1, %xmm1
+	andq	$~15, %rdi
+	punpcklbw %xmm1, %xmm1
+	orl	$0xffffffff, %esi
+	movdqa	(%rdi), %xmm0
+	pshufd	$0, %xmm1, %xmm1
+	subq	%rdi, %rcx
+	pcmpeqb	%xmm1, %xmm0
+	shl	%cl, %esi
+	pmovmskb %xmm0, %ecx
+	movl	$16, %eax
+	movl	$16, %edx
+	andl	%esi, %ecx
+	jnz	1f
+
+2:	pcmpestri $0x08, 16(%rdi), %xmm1
+	leaq	16(%rdi), %rdi
+	jnc	2b
+
+	leaq	(%rdi,%rcx), %rax
+	ret
+
+1:	bsfl	%ecx, %eax
+	addq	%rdi, %rax
+	ret
+	cfi_endproc
+	.size	__rawmemchr_sse42, .-__rawmemchr_sse42
+
+
+# undef ENTRY
+# define ENTRY(name) \
+	.type __rawmemchr_sse2, @function; \
+	__rawmemchr_sse2: cfi_startproc; \
+	CALL_MCOUNT
+# undef END
+# define END(name) \
+	cfi_endproc; .size __rawmemchr_sse2, .-__rawmemchr_sse2
+# undef libc_hidden_builtin_def
+/* It doesn't make sense to send libc-internal rawmemchr calls through a PLT.
+   The speedup we get from using SSE4.2 instruction is likely eaten away
+   by the indirect call in the PLT.  */
+# define libc_hidden_builtin_def(name) \
+	.globl __GI___rawmemchr; __GI___rawmemchr = __rawmemchr_sse2
+#endif
+
+#include "../rawmemchr.S"

sysdeps/x86_64/rawmemchr.S

@@ -34,19 +34,18 @@ ENTRY (rawmemchr)
 	pcmpeqb	%xmm1, %xmm0
 	shl	%cl, %esi
 	pmovmskb %xmm0, %ecx
-	leaq	16(%rdi), %rdi
 	andl	%esi, %ecx
 	jnz	1f
 
-2:	movdqa	(%rdi), %xmm0
+2:	movdqa	16(%rdi), %xmm0
 	leaq	16(%rdi), %rdi
 	pcmpeqb	%xmm1, %xmm0
 	pmovmskb %xmm0, %ecx
 	testl	%ecx, %ecx
 	jz	2b
 
-1:	bsfl	%ecx, %ecx
-	leaq	-16(%rcx,%rdi), %rax
+1:	bsfl	%ecx, %eax
+	addq	%rdi, %rax
 	ret
 END (rawmemchr)