Optimized wcschr and wcscpy for x86-64 and x86-32

1 files changed, 621 insertions, 0 deletions

diff --git a/sysdeps/i386/i686/multiarch/wcscpy-ssse3.S b/sysdeps/i386/i686/multiarch/wcscpy-ssse3.S
new file mode 100644
index 0000000000..84d92a8bde
--- /dev/null
+++ b/sysdeps/i386/i686/multiarch/wcscpy-ssse3.S
@@ -0,0 +1,621 @@
+/* wcscpy with SSSE3
+   Copyright (C) 2011 Free Software Foundation, Inc.
+   Contributed by Intel Corporation.
+   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.  */
+
+#ifndef NOT_IN_libc
+# include <sysdep.h>
+
+# define CFI_PUSH(REG)	\
+	cfi_adjust_cfa_offset (4);	\
+	cfi_rel_offset (REG, 0)
+
+# define CFI_POP(REG)	\
+	cfi_adjust_cfa_offset (-4);	\
+	cfi_restore (REG)
+
+# define PUSH(REG)	pushl REG; CFI_PUSH (REG)
+# define POP(REG)	popl REG; CFI_POP (REG)
+
+# define PARMS	4
+# define RETURN	POP (%edi); ret; CFI_PUSH (%edi)
+# define STR1	PARMS
+# define STR2	STR1+4
+# define LEN	STR2+4
+
+	atom_text_section
+ENTRY (__wcscpy_ssse3)
+	mov	STR1(%esp), %edx
+	mov	STR2(%esp), %ecx
+
+	cmp	$0, (%ecx)
+	jz	L(ExitTail4)
+	cmp	$0, 4(%ecx)
+	jz	L(ExitTail8)
+	cmp	$0, 8(%ecx)
+	jz	L(ExitTail12)
+	cmp	$0, 12(%ecx)
+	jz	L(ExitTail16)
+
+	PUSH	(%edi)
+	mov	%edx, %edi
+
+	PUSH	(%esi)
+	lea	16(%ecx), %esi
+
+	and	$-16, %esi
+
+	pxor	%xmm0, %xmm0
+	pcmpeqd	(%esi), %xmm0
+	movdqu	(%ecx), %xmm1
+	movdqu	%xmm1, (%edx)
+
+	pmovmskb %xmm0, %eax
+	sub	%ecx, %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	mov	%edx, %eax
+	lea	16(%edx), %edx
+	and	$-16, %edx
+	sub	%edx, %eax
+
+	sub	%eax, %ecx
+	mov	%ecx, %eax
+	and	$0xf, %eax
+	mov	$0, %esi
+
+	jz	L(Align16Both)
+	cmp	$4, %eax
+	je	L(Shl4)
+	cmp	$8, %eax
+	je	L(Shl8)
+	jmp	L(Shl12)
+
+L(Align16Both):
+	movaps	(%ecx), %xmm1
+	movaps	16(%ecx), %xmm2
+	movaps	%xmm1, (%edx)
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm3
+	movaps	%xmm2, (%edx, %esi)
+	pcmpeqd	%xmm3, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm4
+	movaps	%xmm3, (%edx, %esi)
+	pcmpeqd	%xmm4, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm1
+	movaps	%xmm4, (%edx, %esi)
+	pcmpeqd	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm2
+	movaps	%xmm1, (%edx, %esi)
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm3
+	movaps	%xmm2, (%edx, %esi)
+	pcmpeqd	%xmm3, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	%xmm3, (%edx, %esi)
+	mov	%ecx, %eax
+	lea	16(%ecx, %esi), %ecx
+	and	$-0x40, %ecx
+	sub	%ecx, %eax
+	sub	%eax, %edx
+
+	mov	$-0x40, %esi
+
+L(Aligned64Loop):
+	movaps	(%ecx), %xmm2
+	movaps	32(%ecx), %xmm3
+	movaps	%xmm2, %xmm4
+	movaps	16(%ecx), %xmm5
+	movaps	%xmm3, %xmm6
+	movaps	48(%ecx), %xmm7
+	pminub	%xmm5, %xmm2
+	pminub	%xmm7, %xmm3
+	pminub	%xmm2, %xmm3
+	lea	64(%edx), %edx
+	pcmpeqd	%xmm0, %xmm3
+	lea	64(%ecx), %ecx
+	pmovmskb %xmm3, %eax
+
+	test	%eax, %eax
+	jnz	L(Aligned64Leave)
+	movaps	%xmm4, -64(%edx)
+	movaps	%xmm5, -48(%edx)
+	movaps	%xmm6, -32(%edx)
+	movaps	%xmm7, -16(%edx)
+	jmp	L(Aligned64Loop)
+
+L(Aligned64Leave):
+	pcmpeqd	%xmm4, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	pcmpeqd	%xmm5, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm4, -64(%edx)
+	test	%eax, %eax
+	lea	16(%esi), %esi
+	jnz	L(CopyFrom1To16Bytes)
+
+	pcmpeqd	%xmm6, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm5, -48(%edx)
+	test	%eax, %eax
+	lea	16(%esi), %esi
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	%xmm6, -32(%edx)
+	pcmpeqd	%xmm7, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	lea	16(%esi), %esi
+	jnz	L(CopyFrom1To16Bytes)
+
+	mov	$-0x40, %esi
+	movaps	%xmm7, -16(%edx)
+	jmp	L(Aligned64Loop)
+
+	.p2align 4
+L(Shl4):
+	movaps	-4(%ecx), %xmm1
+	movaps	12(%ecx), %xmm2
+L(Shl4Start):
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl4LoopExit)
+
+	palignr	$4, %xmm1, %xmm2
+	movaps	%xmm3, %xmm1
+	movaps	%xmm2, (%edx)
+	movaps	28(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl4LoopExit)
+
+	palignr	$4, %xmm1, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	28(%ecx), %xmm2
+	movaps	%xmm3, %xmm1
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl4LoopExit)
+
+	palignr	$4, %xmm1, %xmm2
+	movaps	%xmm3, %xmm1
+	movaps	%xmm2, (%edx)
+	movaps	28(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl4LoopExit)
+
+	palignr	$4, %xmm1, %xmm2
+	movaps	%xmm3, %xmm1
+	movaps	%xmm2, (%edx)
+	lea	28(%ecx), %ecx
+	lea	16(%edx), %edx
+
+	mov	%ecx, %eax
+	and	$-0x40, %ecx
+	sub	%ecx, %eax
+	lea	-12(%ecx), %ecx
+	sub	%eax, %edx
+
+	movaps	-4(%ecx), %xmm1
+
+L(Shl4LoopStart):
+	movaps	12(%ecx), %xmm2
+	movaps	28(%ecx), %xmm3
+	movaps	%xmm3, %xmm6
+	movaps	44(%ecx), %xmm4
+	movaps	%xmm4, %xmm7
+	movaps	60(%ecx), %xmm5
+	pminub	%xmm2, %xmm6
+	pminub	%xmm5, %xmm7
+	pminub	%xmm6, %xmm7
+	pcmpeqd	%xmm0, %xmm7
+	pmovmskb %xmm7, %eax
+	movaps	%xmm5, %xmm7
+	palignr	$4, %xmm4, %xmm5
+	test	%eax, %eax
+	palignr	$4, %xmm3, %xmm4
+	jnz	L(Shl4Start)
+
+	palignr	$4, %xmm2, %xmm3
+	lea	64(%ecx), %ecx
+	palignr	$4, %xmm1, %xmm2
+	movaps	%xmm7, %xmm1
+	movaps	%xmm5, 48(%edx)
+	movaps	%xmm4, 32(%edx)
+	movaps	%xmm3, 16(%edx)
+	movaps	%xmm2, (%edx)
+	lea	64(%edx), %edx
+	jmp	L(Shl4LoopStart)
+
+L(Shl4LoopExit):
+	movaps	(%edx), %xmm6
+	psrldq	$12, %xmm6
+	palignr	$4, %xmm1, %xmm6
+	movaps	%xmm6, (%edx)
+	add	$12, %edx
+	add	$12, %ecx
+
+	POP	(%esi)
+	test	%al, %al
+	jz	L(ExitHigh)
+	test	$0x01, %al
+	jnz	L(Exit4)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(Shl8):
+	movaps	-8(%ecx), %xmm1
+	movaps	8(%ecx), %xmm2
+L(Shl8Start):
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl8LoopExit)
+
+	palignr	$8, %xmm1, %xmm2
+	movaps	%xmm3, %xmm1
+	movaps	%xmm2, (%edx)
+	movaps	24(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl8LoopExit)
+
+	palignr	$8, %xmm1, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	24(%ecx), %xmm2
+	movaps	%xmm3, %xmm1
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl8LoopExit)
+
+	palignr	$8, %xmm1, %xmm2
+	movaps	%xmm3, %xmm1
+	movaps	%xmm2, (%edx)
+	movaps	24(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl8LoopExit)
+
+	palignr	$8, %xmm1, %xmm2
+	movaps	%xmm3, %xmm1
+	movaps	%xmm2, (%edx)
+	lea	24(%ecx), %ecx
+	lea	16(%edx), %edx
+
+	mov	%ecx, %eax
+	and	$-0x40, %ecx
+	sub	%ecx, %eax
+	lea	-8(%ecx), %ecx
+	sub	%eax, %edx
+
+	movaps	-8(%ecx), %xmm1
+
+L(Shl8LoopStart):
+	movaps	8(%ecx), %xmm2
+	movaps	24(%ecx), %xmm3
+	movaps	%xmm3, %xmm6
+	movaps	40(%ecx), %xmm4
+	movaps	%xmm4, %xmm7
+	movaps	56(%ecx), %xmm5
+	pminub	%xmm2, %xmm6
+	pminub	%xmm5, %xmm7
+	pminub	%xmm6, %xmm7
+	pcmpeqd	%xmm0, %xmm7
+	pmovmskb %xmm7, %eax
+	movaps	%xmm5, %xmm7
+	palignr	$8, %xmm4, %xmm5
+	test	%eax, %eax
+	palignr	$8, %xmm3, %xmm4
+	jnz	L(Shl8Start)
+
+	palignr	$8, %xmm2, %xmm3
+	lea	64(%ecx), %ecx
+	palignr	$8, %xmm1, %xmm2
+	movaps	%xmm7, %xmm1
+	movaps	%xmm5, 48(%edx)
+	movaps	%xmm4, 32(%edx)
+	movaps	%xmm3, 16(%edx)
+	movaps	%xmm2, (%edx)
+	lea	64(%edx), %edx
+	jmp	L(Shl8LoopStart)
+
+L(Shl8LoopExit):
+	movaps	(%edx), %xmm6
+	psrldq	$8, %xmm6
+	palignr	$8, %xmm1, %xmm6
+	movaps	%xmm6, (%edx)
+	add	$8, %edx
+	add	$8, %ecx
+
+	POP	(%esi)
+	test	%al, %al
+	jz	L(ExitHigh)
+	test	$0x01, %al
+	jnz	L(Exit4)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(Shl12):
+	movaps	-12(%ecx), %xmm1
+	movaps	4(%ecx), %xmm2
+L(Shl12Start):
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl12LoopExit)
+
+	palignr	$12, %xmm1, %xmm2
+	movaps	%xmm3, %xmm1
+	movaps	%xmm2, (%edx)
+	movaps	20(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl12LoopExit)
+
+	palignr	$12, %xmm1, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	20(%ecx), %xmm2
+	movaps	%xmm3, %xmm1
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl12LoopExit)
+
+	palignr	$12, %xmm1, %xmm2
+	movaps	%xmm3, %xmm1
+	movaps	%xmm2, (%edx)
+	movaps	20(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl12LoopExit)
+
+	palignr	$12, %xmm1, %xmm2
+	movaps	%xmm3, %xmm1
+	movaps	%xmm2, (%edx)
+	lea	20(%ecx), %ecx
+	lea	16(%edx), %edx
+
+	mov	%ecx, %eax
+	and	$-0x40, %ecx
+	sub	%ecx, %eax
+	lea	-4(%ecx), %ecx
+	sub	%eax, %edx
+
+	movaps	-12(%ecx), %xmm1
+
+L(Shl12LoopStart):
+	movaps	4(%ecx), %xmm2
+	movaps	20(%ecx), %xmm3
+	movaps	%xmm3, %xmm6
+	movaps	36(%ecx), %xmm4
+	movaps	%xmm4, %xmm7
+	movaps	52(%ecx), %xmm5
+	pminub	%xmm2, %xmm6
+	pminub	%xmm5, %xmm7
+	pminub	%xmm6, %xmm7
+	pcmpeqd	%xmm0, %xmm7
+	pmovmskb %xmm7, %eax
+	movaps	%xmm5, %xmm7
+	palignr	$12, %xmm4, %xmm5
+	test	%eax, %eax
+	palignr	$12, %xmm3, %xmm4
+	jnz	L(Shl12Start)
+
+	palignr	$12, %xmm2, %xmm3
+	lea	64(%ecx), %ecx
+	palignr	$12, %xmm1, %xmm2
+	movaps	%xmm7, %xmm1
+	movaps	%xmm5, 48(%edx)
+	movaps	%xmm4, 32(%edx)
+	movaps	%xmm3, 16(%edx)
+	movaps	%xmm2, (%edx)
+	lea	64(%edx), %edx
+	jmp	L(Shl12LoopStart)
+
+L(Shl12LoopExit):
+	movaps	(%edx), %xmm6
+	psrldq	$4, %xmm6
+	mov	$4, %esi
+	palignr	$12, %xmm1, %xmm6
+	movaps	%xmm6, (%edx)
+
+	.p2align 4
+L(CopyFrom1To16Bytes):
+	add	%esi, %edx
+	add	%esi, %ecx
+
+	POP	(%esi)
+	test	%al, %al
+	jz	L(ExitHigh)
+	test	$0x01, %al
+	jnz	L(Exit4)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	.p2align 4
+L(ExitHigh):
+	test	$0x01, %ah
+	jnz	L(Exit12)
+	movdqu	(%ecx), %xmm0
+	movdqu	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	.p2align 4
+L(Exit4):
+	movl	(%ecx), %eax
+	movl	%eax, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	.p2align 4
+L(Exit12):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	8(%ecx), %eax
+	movl	%eax, 8(%edx)
+	movl	%edi, %eax
+	RETURN
+
+CFI_POP	(%edi)
+
+	.p2align 4
+L(ExitTail4):
+	movl	(%ecx), %eax
+	movl	%eax, (%edx)
+	movl	%edx, %eax
+	ret
+
+	.p2align 4
+L(ExitTail8):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edx, %eax
+	ret
+
+	.p2align 4
+L(ExitTail12):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	8(%ecx), %eax
+	movl	%eax, 8(%edx)
+	movl	%edx, %eax
+	ret
+
+	.p2align 4
+L(ExitTail16):
+	movdqu	(%ecx), %xmm0
+	movdqu	%xmm0, (%edx)
+	movl	%edx, %eax
+	ret
+
+END (__wcscpy_ssse3)
+#endif