* sysdeps/x86_64/strlen.S: Replace with new SSE2 based implementation

	which is faster on all x86_64 architectures.
	Tested on AMD, Intel Nehalem, SNB, IVB.

1 files changed, 226 insertions, 3 deletions

diff --git a/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S b/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
index 72bb609949..6d9951e89f 100644
--- a/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
+++ b/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
@@ -34,10 +34,233 @@ ENTRY (STRCAT)
 	mov	%rdx, %r8
 # endif
 
-# define RETURN  jmp L(StartStrcpyPart)
-# include "strlen-sse2-pminub.S"
-# undef RETURN
+	xor	%rax, %rax
+	mov	%edi, %ecx
+	and	$0x3f, %ecx
+	pxor	%xmm0, %xmm0
+	cmp	$0x30, %ecx
+	ja	L(next)
+	movdqu	(%rdi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	jnz	L(exit_less16)
+	mov	%rdi, %rax
+	and	$-16, %rax
+	jmp	L(align16_start)
+L(next):
+	mov	%rdi, %rax
+	and	$-16, %rax
+	pcmpeqb	(%rax), %xmm0
+	mov	$-1, %r10d
+	sub	%rax, %rcx
+	shl	%cl, %r10d
+	pmovmskb %xmm0, %edx
+	and	%r10d, %edx
+	jnz	L(exit)
 
+L(align16_start):
+	pxor	%xmm0, %xmm0
+	pxor	%xmm1, %xmm1
+	pxor	%xmm2, %xmm2
+	pxor	%xmm3, %xmm3
+	pcmpeqb	16(%rax), %xmm0
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	jnz	L(exit16)
+
+	pcmpeqb	32(%rax), %xmm1
+	pmovmskb %xmm1, %edx
+	test	%edx, %edx
+	jnz	L(exit32)
+
+	pcmpeqb	48(%rax), %xmm2
+	pmovmskb %xmm2, %edx
+	test	%edx, %edx
+	jnz	L(exit48)
+
+	pcmpeqb	64(%rax), %xmm3
+	pmovmskb %xmm3, %edx
+	test	%edx, %edx
+	jnz	L(exit64)
+
+	pcmpeqb	80(%rax), %xmm0
+	add	$64, %rax
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	jnz	L(exit16)
+
+	pcmpeqb	32(%rax), %xmm1
+	pmovmskb %xmm1, %edx
+	test	%edx, %edx
+	jnz	L(exit32)
+
+	pcmpeqb	48(%rax), %xmm2
+	pmovmskb %xmm2, %edx
+	test	%edx, %edx
+	jnz	L(exit48)
+
+	pcmpeqb	64(%rax), %xmm3
+	pmovmskb %xmm3, %edx
+	test	%edx, %edx
+	jnz	L(exit64)
+
+	pcmpeqb	80(%rax), %xmm0
+	add	$64, %rax
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	jnz	L(exit16)
+
+	pcmpeqb	32(%rax), %xmm1
+	pmovmskb %xmm1, %edx
+	test	%edx, %edx
+	jnz	L(exit32)
+
+	pcmpeqb	48(%rax), %xmm2
+	pmovmskb %xmm2, %edx
+	test	%edx, %edx
+	jnz	L(exit48)
+
+	pcmpeqb	64(%rax), %xmm3
+	pmovmskb %xmm3, %edx
+	test	%edx, %edx
+	jnz	L(exit64)
+
+	pcmpeqb	80(%rax), %xmm0
+	add	$64, %rax
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	jnz	L(exit16)
+
+	pcmpeqb	32(%rax), %xmm1
+	pmovmskb %xmm1, %edx
+	test	%edx, %edx
+	jnz	L(exit32)
+
+	pcmpeqb	48(%rax), %xmm2
+	pmovmskb %xmm2, %edx
+	test	%edx, %edx
+	jnz	L(exit48)
+
+	pcmpeqb	64(%rax), %xmm3
+	pmovmskb %xmm3, %edx
+	test	%edx, %edx
+	jnz	L(exit64)
+
+	test	$0x3f, %rax
+	jz	L(align64_loop)
+
+	pcmpeqb	80(%rax), %xmm0
+	add	$80, %rax
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	jnz	L(exit)
+
+	test	$0x3f, %rax
+	jz	L(align64_loop)
+
+	pcmpeqb	16(%rax), %xmm1
+	add	$16, %rax
+	pmovmskb %xmm1, %edx
+	test	%edx, %edx
+	jnz	L(exit)
+
+	test	$0x3f, %rax
+	jz	L(align64_loop)
+
+	pcmpeqb	16(%rax), %xmm2
+	add	$16, %rax
+	pmovmskb %xmm2, %edx
+	test	%edx, %edx
+	jnz	L(exit)
+
+	test	$0x3f, %rax
+	jz	L(align64_loop)
+
+	pcmpeqb	16(%rax), %xmm3
+	add	$16, %rax
+	pmovmskb %xmm3, %edx
+	test	%edx, %edx
+	jnz	L(exit)
+
+	add	$16, %rax
+	.p2align 4
+	L(align64_loop):
+	movaps	(%rax),	%xmm4
+	pminub	16(%rax),	%xmm4
+	movaps	32(%rax),	%xmm5
+	pminub	48(%rax),	%xmm5
+	add	$64,	%rax
+	pminub	%xmm4,	%xmm5
+	pcmpeqb	%xmm0,	%xmm5
+	pmovmskb %xmm5,	%edx
+	test	%edx,	%edx
+	jz	L(align64_loop)
+
+	pcmpeqb	-64(%rax), %xmm0
+	sub	$80,	%rax
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	jnz	L(exit16)
+
+	pcmpeqb	32(%rax), %xmm1
+	pmovmskb %xmm1, %edx
+	test	%edx, %edx
+	jnz	L(exit32)
+
+	pcmpeqb	48(%rax), %xmm2
+	pmovmskb %xmm2, %edx
+	test	%edx, %edx
+	jnz	L(exit48)
+
+	pcmpeqb	64(%rax), %xmm3
+	pmovmskb %xmm3, %edx
+	sub	%rdi, %rax
+	bsf	%rdx, %rdx
+	add	%rdx, %rax
+	add	$64, %rax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit):
+	sub	%rdi, %rax
+L(exit_less16):
+	bsf	%rdx, %rdx
+	add	%rdx, %rax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit16):
+	sub	%rdi, %rax
+	bsf	%rdx, %rdx
+	add	%rdx, %rax
+	add	$16, %rax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit32):
+	sub	%rdi, %rax
+	bsf	%rdx, %rdx
+	add	%rdx, %rax
+	add	$32, %rax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit48):
+	sub	%rdi, %rax
+	bsf	%rdx, %rdx
+	add	%rdx, %rax
+	add	$48, %rax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit64):
+	sub	%rdi, %rax
+	bsf	%rdx, %rdx
+	add	%rdx, %rax
+	add	$64, %rax
+
+	.p2align 4
 L(StartStrcpyPart):
 	lea	(%r9, %rax), %rdi
 	mov	%rsi, %rcx