* 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, 309 insertions, 3 deletions

diff --git a/sysdeps/x86_64/multiarch/strcat-ssse3.S b/sysdeps/x86_64/multiarch/strcat-ssse3.S
index fea9d11b40..901e66f2c8 100644
--- a/sysdeps/x86_64/multiarch/strcat-ssse3.S
+++ b/sysdeps/x86_64/multiarch/strcat-ssse3.S
@@ -33,11 +33,317 @@ ENTRY (STRCAT)
 	mov	%rdx, %r8
 # endif
 
-# define RETURN  jmp L(StartStrcpyPart)
-# include "strlen-sse2-no-bsf.S"
+	xor	%eax, %eax
+	cmpb	$0, (%rdi)
+	jz	L(exit_tail0)
+	cmpb	$0, 1(%rdi)
+	jz	L(exit_tail1)
+	cmpb	$0, 2(%rdi)
+	jz	L(exit_tail2)
+	cmpb	$0, 3(%rdi)
+	jz	L(exit_tail3)
+
+	cmpb	$0, 4(%rdi)
+	jz	L(exit_tail4)
+	cmpb	$0, 5(%rdi)
+	jz	L(exit_tail5)
+	cmpb	$0, 6(%rdi)
+	jz	L(exit_tail6)
+	cmpb	$0, 7(%rdi)
+	jz	L(exit_tail7)
+
+	cmpb	$0, 8(%rdi)
+	jz	L(exit_tail8)
+	cmpb	$0, 9(%rdi)
+	jz	L(exit_tail9)
+	cmpb	$0, 10(%rdi)
+	jz	L(exit_tail10)
+	cmpb	$0, 11(%rdi)
+	jz	L(exit_tail11)
+
+	cmpb	$0, 12(%rdi)
+	jz	L(exit_tail12)
+	cmpb	$0, 13(%rdi)
+	jz	L(exit_tail13)
+	cmpb	$0, 14(%rdi)
+	jz	L(exit_tail14)
+	cmpb	$0, 15(%rdi)
+	jz	L(exit_tail15)
+	pxor	%xmm0, %xmm0
+	lea	16(%rdi), %rcx
+	lea	16(%rdi), %rax
+	and	$-16, %rax
+
+	pcmpeqb	(%rax), %xmm0
+	pmovmskb %xmm0, %edx
+	pxor	%xmm1, %xmm1
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm1
+	pmovmskb %xmm1, %edx
+	pxor	%xmm2, %xmm2
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm2
+	pmovmskb %xmm2, %edx
+	pxor	%xmm3, %xmm3
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm3
+	pmovmskb %xmm3, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm0
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm1
+	pmovmskb %xmm1, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm2
+	pmovmskb %xmm2, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm3
+	pmovmskb %xmm3, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm0
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm1
+	pmovmskb %xmm1, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm2
+	pmovmskb %xmm2, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm3
+	pmovmskb %xmm3, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm0
+	pmovmskb %xmm0, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm1
+	pmovmskb %xmm1, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm2
+	pmovmskb %xmm2, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	pcmpeqb	(%rax), %xmm3
+	pmovmskb %xmm3, %edx
+	test	%edx, %edx
+	lea	16(%rax), %rax
+	jnz	L(exit)
+
+	and	$-0x40, %rax
 
-# undef RETURN
+	.p2align 4
+L(aligned_64):
+	pcmpeqb	(%rax), %xmm0
+	pcmpeqb	16(%rax), %xmm1
+	pcmpeqb	32(%rax), %xmm2
+	pcmpeqb	48(%rax), %xmm3
+	pmovmskb %xmm0, %edx
+	pmovmskb %xmm1, %r11d
+	pmovmskb %xmm2, %r10d
+	pmovmskb %xmm3, %r9d
+	or	%edx, %r9d
+	or	%r11d, %r9d
+	or	%r10d, %r9d
+	lea	64(%rax), %rax
+	jz	L(aligned_64)
+
+	test	%edx, %edx
+	jnz	L(aligned_64_exit_16)
+	test	%r11d, %r11d
+	jnz	L(aligned_64_exit_32)
+	test	%r10d, %r10d
+	jnz	L(aligned_64_exit_48)
+
+L(aligned_64_exit_64):
+	pmovmskb %xmm3, %edx
+	jmp	L(exit)
+
+L(aligned_64_exit_48):
+	lea	-16(%rax), %rax
+	mov	%r10d, %edx
+	jmp	L(exit)
+
+L(aligned_64_exit_32):
+	lea	-32(%rax), %rax
+	mov	%r11d, %edx
+	jmp	L(exit)
+
+L(aligned_64_exit_16):
+	lea	-48(%rax), %rax
+
+L(exit):
+	sub	%rcx, %rax
+	test	%dl, %dl
+	jz	L(exit_high)
+	test	$0x01, %dl
+	jnz	L(exit_tail0)
+
+	test	$0x02, %dl
+	jnz	L(exit_tail1)
+
+	test	$0x04, %dl
+	jnz	L(exit_tail2)
+
+	test	$0x08, %dl
+	jnz	L(exit_tail3)
+
+	test	$0x10, %dl
+	jnz	L(exit_tail4)
+
+	test	$0x20, %dl
+	jnz	L(exit_tail5)
+
+	test	$0x40, %dl
+	jnz	L(exit_tail6)
+	add	$7, %eax
+L(exit_tail0):
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_high):
+	add	$8, %eax
+	test	$0x01, %dh
+	jnz	L(exit_tail0)
+
+	test	$0x02, %dh
+	jnz	L(exit_tail1)
+
+	test	$0x04, %dh
+	jnz	L(exit_tail2)
+
+	test	$0x08, %dh
+	jnz	L(exit_tail3)
+
+	test	$0x10, %dh
+	jnz	L(exit_tail4)
+
+	test	$0x20, %dh
+	jnz	L(exit_tail5)
+
+	test	$0x40, %dh
+	jnz	L(exit_tail6)
+	add	$7, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail1):
+	add	$1, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail2):
+	add	$2, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail3):
+	add	$3, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail4):
+	add	$4, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail5):
+	add	$5, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail6):
+	add	$6, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail7):
+	add	$7, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail8):
+	add	$8, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail9):
+	add	$9, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail10):
+	add	$10, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail11):
+	add	$11, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail12):
+	add	$12, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail13):
+	add	$13, %eax
+	jmp	L(StartStrcpyPart)
 
+	.p2align 4
+L(exit_tail14):
+	add	$14, %eax
+	jmp	L(StartStrcpyPart)
+
+	.p2align 4
+L(exit_tail15):
+	add	$15, %eax
+
+	.p2align 4
 L(StartStrcpyPart):
 	mov	%rsi, %rcx
 	lea	(%rdi, %rax), %rdx