/*
Copyright (c) 2024, Intel Corporation
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

    * Redistributions of source code must retain the above copyright notice,
    * this list of conditions and the following disclaimer.

    * Redistributions in binary form must reproduce the above copyright notice,
    * this list of conditions and the following disclaimer in the documentation
    * and/or other materials provided with the distribution.

    * Neither the name of Intel Corporation nor the names of its contributors
    * may be used to endorse or promote products derived from this software
    * without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/


#ifndef MEMMOVE
# define MEMMOVE		memmove_avx2
#endif

#ifndef L
# define L(label)	.L##label
#endif

#ifndef cfi_startproc
# define cfi_startproc	.cfi_startproc
#endif

#ifndef cfi_endproc
# define cfi_endproc	.cfi_endproc
#endif

#ifndef cfi_rel_offset
# define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
#endif

#ifndef cfi_restore
# define cfi_restore(reg)	.cfi_restore reg
#endif

#ifndef cfi_adjust_cfa_offset
# define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
#endif

#ifndef ENTRY
# define ENTRY(name)		\
	.type name,  @function;		\
	.globl name;		\
	.p2align 4;		\
name:		\
	cfi_startproc
#endif

#ifndef ALIAS_SYMBOL
# define ALIAS_SYMBOL(alias, original) \
	.globl alias; \
	.equ alias, original
#endif

#ifndef END
# define END(name)		\
	cfi_endproc;		\
	.size name, .-name
#endif

#define CFI_PUSH(REG)		\
	cfi_adjust_cfa_offset (8);		\
	cfi_rel_offset (REG, 0)

#define CFI_POP(REG)		\
	cfi_adjust_cfa_offset (-8);		\
	cfi_restore (REG)

#define PUSH(REG)	push REG;
#define POP(REG)	pop REG;

#define ENTRANCE	\
	PUSH(%rbx);		\
	CFI_PUSH (%rbx);
#define RETURN_END	\
	POP(%rbx);		\
	CFI_POP (%rbx);	\
	ret
#define RETURN		RETURN_END;

	.section .text.avx2,"ax",@progbits
ENTRY (__memcpy_chk_avx2)
	cmp	%rcx, %rdx
	ja	__memcpy_chk_fail
/* Fall through to memcpy/memmove. */
END (__memcpy_chk_avx2)

ENTRY (MEMMOVE)
	ENTRANCE
	mov	%rdi, %rax

/* Check whether we should copy backward or forward.  */
	cmp	%rsi, %rdi
	je	L(mm_return)
	jg	L(mm_len_0_or_more_backward)

/* Now do checks for lengths. We do [0..16], [0..32], [0..64], [0..128]
	separately.  */
	cmp	$16, %rdx
	jbe	L(mm_len_0_16_bytes_forward)

	cmp	$32, %rdx
	ja	L(mm_len_32_or_more_forward)

/* Copy [0..32] and return.  */
	movdqu	(%rsi), %xmm0
	movdqu	-16(%rsi, %rdx), %xmm1
	movdqu	%xmm0, (%rdi)
	movdqu	%xmm1, -16(%rdi, %rdx)
	jmp	L(mm_return)

L(mm_len_32_or_more_forward):
	cmp	$64, %rdx
	ja	L(mm_len_64_or_more_forward)

/* Copy [0..64] and return.  */
	movdqu	(%rsi), %xmm0
	movdqu	16(%rsi), %xmm1
	movdqu	-16(%rsi, %rdx), %xmm2
	movdqu	-32(%rsi, %rdx), %xmm3
	movdqu	%xmm0, (%rdi)
	movdqu	%xmm1, 16(%rdi)
	movdqu	%xmm2, -16(%rdi, %rdx)
	movdqu	%xmm3, -32(%rdi, %rdx)
	jmp	L(mm_return)

L(mm_len_64_or_more_forward):
	cmp	$128, %rdx
	ja	L(mm_len_128_or_more_forward)

/* Copy [0..128] and return.  */
	movdqu	(%rsi), %xmm0
	movdqu	16(%rsi), %xmm1
	movdqu	32(%rsi), %xmm2
	movdqu	48(%rsi), %xmm3
	movdqu	-64(%rsi, %rdx), %xmm4
	movdqu	-48(%rsi, %rdx), %xmm5
	movdqu	-32(%rsi, %rdx), %xmm6
	movdqu	-16(%rsi, %rdx), %xmm7
	movdqu	%xmm0, (%rdi)
	movdqu	%xmm1, 16(%rdi)
	movdqu	%xmm2, 32(%rdi)
	movdqu	%xmm3, 48(%rdi)
	movdqu	%xmm4, -64(%rdi, %rdx)
	movdqu	%xmm5, -48(%rdi, %rdx)
	movdqu	%xmm6, -32(%rdi, %rdx)
	movdqu	%xmm7, -16(%rdi, %rdx)
	jmp	L(mm_return)

L(mm_len_128_or_more_forward):
	cmp	$256, %rdx
	ja	L(mm_len_256_or_more_forward)

/* Copy [0..256] and return.  */
	movdqu	(%rsi), %xmm0
	movdqu	16(%rsi), %xmm1
	movdqu	32(%rsi), %xmm2
	movdqu	48(%rsi), %xmm3
	movdqu	64(%rsi), %xmm4
	movdqu	80(%rsi), %xmm5
	movdqu	96(%rsi), %xmm6
	movdqu	112(%rsi), %xmm7
	movdqu	-128(%rsi, %rdx), %xmm8
	movdqu	-112(%rsi, %rdx), %xmm9
	movdqu	-96(%rsi, %rdx), %xmm10
	movdqu	-80(%rsi, %rdx), %xmm11
	movdqu	-64(%rsi, %rdx), %xmm12
	movdqu	-48(%rsi, %rdx), %xmm13
	movdqu	-32(%rsi, %rdx), %xmm14
	movdqu	-16(%rsi, %rdx), %xmm15
	movdqu	%xmm0, (%rdi)
	movdqu	%xmm1, 16(%rdi)
	movdqu	%xmm2, 32(%rdi)
	movdqu	%xmm3, 48(%rdi)
	movdqu	%xmm4, 64(%rdi)
	movdqu	%xmm5, 80(%rdi)
	movdqu	%xmm6, 96(%rdi)
	movdqu	%xmm7, 112(%rdi)
	movdqu	%xmm8, -128(%rdi, %rdx)
	movdqu	%xmm9, -112(%rdi, %rdx)
	movdqu	%xmm10, -96(%rdi, %rdx)
	movdqu	%xmm11, -80(%rdi, %rdx)
	movdqu	%xmm12, -64(%rdi, %rdx)
	movdqu	%xmm13, -48(%rdi, %rdx)
	movdqu	%xmm14, -32(%rdi, %rdx)
	movdqu	%xmm15, -16(%rdi, %rdx)
	jmp	L(mm_return)

L(mm_len_256_or_more_forward):
/* Aligning the address of destination.  */
/*  save first unaligned 128 bytes */
	vmovdqu	(%rsi), %ymm0
	vmovdqu	32(%rsi), %ymm1
	vmovdqu	64(%rsi), %ymm2
	vmovdqu	96(%rsi), %ymm3

	lea	128(%rdi), %r8
	and	$-128, %r8  /* r8 now aligned to next 128 byte boundary */
	sub	%rdi, %rsi  /* rsi = src - dst = diff */

	vmovdqu	(%r8, %rsi), %ymm4
	vmovdqu	32(%r8, %rsi), %ymm5
	vmovdqu	64(%r8, %rsi), %ymm6
	vmovdqu	96(%r8, %rsi), %ymm7

	vmovdqu	%ymm0, (%rdi)
	vmovdqu	%ymm1, 32(%rdi)
	vmovdqu	%ymm2, 64(%rdi)
	vmovdqu	%ymm3, 96(%rdi)
	vmovdqa	%ymm4, (%r8)
	vmovaps	%ymm5, 32(%r8)
	vmovaps	%ymm6, 64(%r8)
	vmovaps	%ymm7, 96(%r8)
	add	$128, %r8

	lea	(%rdi, %rdx), %rbx
	and	$-128, %rbx
	cmp	%r8, %rbx
	jbe	L(mm_copy_remaining_forward)

	cmp	__x86_shared_cache_size_half(%rip), %rdx
	jae	L(mm_overlapping_check_forward)

		.p2align 4
L(mm_main_loop_forward):
	prefetcht0	128(%r8, %rsi)
	vmovdqu	(%r8, %rsi), %ymm0
	vmovdqu	32(%r8, %rsi), %ymm1
	vmovdqa	%ymm0, (%r8)
	vmovaps	%ymm1, 32(%r8)
	lea	64(%r8), %r8
	cmp	%r8, %rbx
	ja	L(mm_main_loop_forward)

L(mm_copy_remaining_forward):
	add	%rdi, %rdx
	sub	%r8, %rdx
/* We copied all up till %rdi position in the dst.
	In %rdx now is how many bytes are left to copy.
	Now we need to advance %r8. */
	lea	(%r8, %rsi), %r9

L(mm_remaining_0_128_bytes_forward):
	cmp	$64, %rdx
	ja	L(mm_remaining_65_128_bytes_forward)
	cmp	$32, %rdx
	ja	L(mm_remaining_33_64_bytes_forward)
	vzeroupper
	cmp	$16, %rdx
	ja	L(mm_remaining_17_32_bytes_forward)
	test	%rdx, %rdx
	.p2align 4,,2
	je	L(mm_return)

	cmpb	$8, %dl
	ja	L(mm_remaining_9_16_bytes_forward)
	cmpb	$4, %dl
	.p2align 4,,5
	ja	L(mm_remaining_5_8_bytes_forward)
	cmpb	$2, %dl
	.p2align 4,,1
	ja	L(mm_remaining_3_4_bytes_forward)
	movzbl	-1(%r9,%rdx), %esi
	movzbl	(%r9), %ebx
	movb	%sil, -1(%r8,%rdx)
	movb	%bl, (%r8)
	jmp	L(mm_return)

L(mm_remaining_65_128_bytes_forward):
	vmovdqu (%r9), %ymm0
	vmovdqu 32(%r9), %ymm1
	vmovdqu -64(%r9, %rdx), %ymm2
	vmovdqu -32(%r9, %rdx), %ymm3
	vmovdqu %ymm0, (%r8)
	vmovdqu %ymm1, 32(%r8)
	vmovdqu %ymm2, -64(%r8, %rdx)
	vmovdqu %ymm3, -32(%r8, %rdx)
	jmp L(mm_return_vzeroupper)

L(mm_remaining_33_64_bytes_forward):
	vmovdqu (%r9), %ymm0
	vmovdqu -32(%r9, %rdx), %ymm1
	vmovdqu %ymm0, (%r8)
	vmovdqu %ymm1, -32(%r8, %rdx)
	jmp	L(mm_return_vzeroupper)

L(mm_remaining_17_32_bytes_forward):
	movdqu	(%r9), %xmm0
	movdqu	-16(%r9, %rdx), %xmm1
	movdqu	%xmm0, (%r8)
	movdqu	%xmm1, -16(%r8, %rdx)
	jmp	L(mm_return)

L(mm_remaining_5_8_bytes_forward):
	movl	(%r9), %esi
	movl	-4(%r9,%rdx), %ebx
	movl	%esi, (%r8)
	movl	%ebx, -4(%r8,%rdx)
	jmp	L(mm_return)

L(mm_remaining_9_16_bytes_forward):
	mov	(%r9), %rsi
	mov	-8(%r9, %rdx), %rbx
	mov	%rsi, (%r8)
	mov	%rbx, -8(%r8, %rdx)
	jmp	L(mm_return)

L(mm_remaining_3_4_bytes_forward):
	movzwl	-2(%r9,%rdx), %esi
	movzwl	(%r9), %ebx
	movw	%si, -2(%r8,%rdx)
	movw	%bx, (%r8)
	jmp	L(mm_return)

L(mm_len_0_16_bytes_forward):
	testb	$24, %dl
	jne	L(mm_len_9_16_bytes_forward)
	testb	$4, %dl
	.p2align 4,,5
	jne	L(mm_len_5_8_bytes_forward)
	test	%rdx, %rdx
	.p2align 4,,2
	je	L(mm_return)
	testb	$2, %dl
	.p2align 4,,1
	jne	L(mm_len_2_4_bytes_forward)
	movzbl	-1(%rsi,%rdx), %ebx
	movzbl	(%rsi), %esi
	movb	%bl, -1(%rdi,%rdx)
	movb	%sil, (%rdi)
	jmp	L(mm_return)

L(mm_len_2_4_bytes_forward):
	movzwl	-2(%rsi,%rdx), %ebx
	movzwl	(%rsi), %esi
	movw	%bx, -2(%rdi,%rdx)
	movw	%si, (%rdi)
	jmp	L(mm_return)

L(mm_len_5_8_bytes_forward):
	movl	(%rsi), %ebx
	movl	-4(%rsi,%rdx), %esi
	movl	%ebx, (%rdi)
	movl	%esi, -4(%rdi,%rdx)
	jmp	L(mm_return)

L(mm_len_9_16_bytes_forward):
	mov	(%rsi), %rbx
	mov	-8(%rsi, %rdx), %rsi
	mov	%rbx, (%rdi)
	mov	%rsi, -8(%rdi, %rdx)
	jmp	L(mm_return)

L(mm_recalc_len):
/* Compute in %rdx how many bytes are left to copy after
	the main loop stops.  */
	vzeroupper
	mov 	%rbx, %rdx
	sub 	%rdi, %rdx
/* The code for copying backwards.  */
L(mm_len_0_or_more_backward):

/* Now do checks for lengths. We do [0..16], [16..32], [32..64], [64..128]
	separately.  */
	cmp	$16, %rdx
	jbe	L(mm_len_0_16_bytes_backward)

	cmp	$32, %rdx
	ja	L(mm_len_32_or_more_backward)

/* Copy [0..32] and return.  */
	movdqu	(%rsi), %xmm0
	movdqu	-16(%rsi, %rdx), %xmm1
	movdqu	%xmm0, (%rdi)
	movdqu	%xmm1, -16(%rdi, %rdx)
	jmp	L(mm_return)

L(mm_len_32_or_more_backward):
	cmp	$64, %rdx
	ja	L(mm_len_64_or_more_backward)

/* Copy [0..64] and return.  */
	movdqu	(%rsi), %xmm0
	movdqu	16(%rsi), %xmm1
	movdqu	-16(%rsi, %rdx), %xmm2
	movdqu	-32(%rsi, %rdx), %xmm3
	movdqu	%xmm0, (%rdi)
	movdqu	%xmm1, 16(%rdi)
	movdqu	%xmm2, -16(%rdi, %rdx)
	movdqu	%xmm3, -32(%rdi, %rdx)
	jmp	L(mm_return)

L(mm_len_64_or_more_backward):
	cmp	$128, %rdx
	ja	L(mm_len_128_or_more_backward)

/* Copy [0..128] and return.  */
	movdqu	(%rsi), %xmm0
	movdqu	16(%rsi), %xmm1
	movdqu	32(%rsi), %xmm2
	movdqu	48(%rsi), %xmm3
	movdqu	-64(%rsi, %rdx), %xmm4
	movdqu	-48(%rsi, %rdx), %xmm5
	movdqu	-32(%rsi, %rdx), %xmm6
	movdqu	-16(%rsi, %rdx), %xmm7
	movdqu	%xmm0, (%rdi)
	movdqu	%xmm1, 16(%rdi)
	movdqu	%xmm2, 32(%rdi)
	movdqu	%xmm3, 48(%rdi)
	movdqu	%xmm4, -64(%rdi, %rdx)
	movdqu	%xmm5, -48(%rdi, %rdx)
	movdqu	%xmm6, -32(%rdi, %rdx)
	movdqu	%xmm7, -16(%rdi, %rdx)
	jmp	L(mm_return)

L(mm_len_128_or_more_backward):
	cmp	$256, %rdx
	ja	L(mm_len_256_or_more_backward)

/* Copy [0..256] and return.  */
	movdqu	(%rsi), %xmm0
	movdqu	16(%rsi), %xmm1
	movdqu	32(%rsi), %xmm2
	movdqu	48(%rsi), %xmm3
	movdqu	64(%rsi), %xmm4
	movdqu	80(%rsi), %xmm5
	movdqu	96(%rsi), %xmm6
	movdqu	112(%rsi), %xmm7
	movdqu	-128(%rsi, %rdx), %xmm8
	movdqu	-112(%rsi, %rdx), %xmm9
	movdqu	-96(%rsi, %rdx), %xmm10
	movdqu	-80(%rsi, %rdx), %xmm11
	movdqu	-64(%rsi, %rdx), %xmm12
	movdqu	-48(%rsi, %rdx), %xmm13
	movdqu	-32(%rsi, %rdx), %xmm14
	movdqu	-16(%rsi, %rdx), %xmm15
	movdqu	%xmm0, (%rdi)
	movdqu	%xmm1, 16(%rdi)
	movdqu	%xmm2, 32(%rdi)
	movdqu	%xmm3, 48(%rdi)
	movdqu	%xmm4, 64(%rdi)
	movdqu	%xmm5, 80(%rdi)
	movdqu	%xmm6, 96(%rdi)
	movdqu	%xmm7, 112(%rdi)
	movdqu	%xmm8, -128(%rdi, %rdx)
	movdqu	%xmm9, -112(%rdi, %rdx)
	movdqu	%xmm10, -96(%rdi, %rdx)
	movdqu	%xmm11, -80(%rdi, %rdx)
	movdqu	%xmm12, -64(%rdi, %rdx)
	movdqu	%xmm13, -48(%rdi, %rdx)
	movdqu	%xmm14, -32(%rdi, %rdx)
	movdqu	%xmm15, -16(%rdi, %rdx)
	jmp	L(mm_return)

L(mm_len_256_or_more_backward):
/* Aligning the address of destination. We need to save
	128 bytes from the source in order not to overwrite them.  */
	vmovdqu	-32(%rsi, %rdx), %ymm0
	vmovdqu	-64(%rsi, %rdx), %ymm1
	vmovdqu	-96(%rsi, %rdx), %ymm2
	vmovdqu	-128(%rsi, %rdx), %ymm3

	lea	(%rdi, %rdx), %r9
	and	$-128, %r9 /* r9 = aligned dst */

	mov	%rsi, %r8
	sub	%rdi, %r8 /* r8 = src - dst, diff */

	vmovdqu	-32(%r9, %r8), %ymm4
	vmovdqu	-64(%r9, %r8), %ymm5
	vmovdqu	-96(%r9, %r8), %ymm6
	vmovdqu	-128(%r9, %r8), %ymm7

	vmovdqu	%ymm0, -32(%rdi, %rdx)
	vmovdqu	%ymm1, -64(%rdi, %rdx)
	vmovdqu	%ymm2, -96(%rdi, %rdx)
	vmovdqu	%ymm3, -128(%rdi, %rdx)
	vmovdqa	%ymm4, -32(%r9)
	vmovdqa	%ymm5, -64(%r9)
	vmovdqa	%ymm6, -96(%r9)
	vmovdqa	%ymm7, -128(%r9)
	lea	-128(%r9), %r9

	lea	128(%rdi), %rbx
	and	$-128, %rbx

	cmp	%r9, %rbx
	jae	L(mm_recalc_len)

	cmp	__x86_shared_cache_size_half(%rip), %rdx
	jae	L(mm_overlapping_check_backward)

	.p2align 4
L(mm_main_loop_backward):
	prefetcht0 -128(%r9, %r8)

	vmovdqu	-64(%r9, %r8), %ymm0
	vmovdqu	-32(%r9, %r8), %ymm1
	vmovdqa	%ymm0, -64(%r9)
	vmovaps	%ymm1, -32(%r9)
	lea	-64(%r9), %r9
	cmp	%r9, %rbx
	jb	L(mm_main_loop_backward)
	jmp	L(mm_recalc_len)

/* Copy [0..16] and return.  */
L(mm_len_0_16_bytes_backward):
	testb	$24, %dl
	jnz	L(mm_len_9_16_bytes_backward)
	testb	$4, %dl
	.p2align 4,,5
	jnz	L(mm_len_5_8_bytes_backward)
	test	%rdx, %rdx
	.p2align 4,,2
	je	L(mm_return)
	testb	$2, %dl
	.p2align 4,,1
	jne	L(mm_len_3_4_bytes_backward)
	movzbl	-1(%rsi,%rdx), %ebx
	movzbl	(%rsi), %ecx
	movb	%bl, -1(%rdi,%rdx)
	movb	%cl, (%rdi)
	jmp	L(mm_return)

L(mm_len_3_4_bytes_backward):
	movzwl	-2(%rsi,%rdx), %ebx
	movzwl	(%rsi), %ecx
	movw	%bx, -2(%rdi,%rdx)
	movw	%cx, (%rdi)
	jmp	L(mm_return)

L(mm_len_9_16_bytes_backward):
	movl	-4(%rsi,%rdx), %ebx
	movl	-8(%rsi,%rdx), %ecx
	movl	%ebx, -4(%rdi,%rdx)
	movl	%ecx, -8(%rdi,%rdx)
	sub	$8, %rdx
	jmp	L(mm_len_0_16_bytes_backward)

L(mm_len_5_8_bytes_backward):
	movl	(%rsi), %ebx
	movl	-4(%rsi,%rdx), %ecx
	movl	%ebx, (%rdi)
	movl	%ecx, -4(%rdi,%rdx)

L(mm_return):
	RETURN

L(mm_return_vzeroupper):
	vzeroupper
	RETURN

/* Big length copy forward part.  */

	.p2align 4

L(mm_overlapping_check_forward):
	mov	%rsi, %r9
	add	%rdx, %r9
	cmp	__x86_shared_cache_size(%rip), %r9
	jbe	L(mm_main_loop_forward)

L(mm_large_page_loop_forward):
	vmovdqu	  (%r8, %rsi), %ymm0
	vmovdqu	  32(%r8, %rsi), %ymm1
	vmovdqu	  64(%r8, %rsi), %ymm2
	vmovdqu	  96(%r8, %rsi), %ymm3
	vmovntdq  %ymm0, (%r8)
	vmovntdq  %ymm1, 32(%r8)
	vmovntdq  %ymm2, 64(%r8)
	vmovntdq  %ymm3, 96(%r8)
	lea 	  128(%r8), %r8
	cmp	  %r8, %rbx
	ja	  L(mm_large_page_loop_forward)
	sfence
	jmp	  L(mm_copy_remaining_forward)

/* Big length copy backward part.  */
	.p2align 4

L(mm_overlapping_check_backward):
	mov	%rdi, %r11
	sub	%rsi, %r11 /* r11 = dst - src, diff */
	add	%rdx, %r11
	cmp	__x86_shared_cache_size(%rip), %r11
	jbe	L(mm_main_loop_backward)

L(mm_large_page_loop_backward):
	vmovdqu	  -64(%r9, %r8), %ymm0
	vmovdqu	  -32(%r9, %r8), %ymm1
	vmovntdq  %ymm0, -64(%r9)
	vmovntdq  %ymm1, -32(%r9)
	lea 	  -64(%r9), %r9
	cmp	  %r9, %rbx
	jb	  L(mm_large_page_loop_backward)
	sfence
	jmp	  L(mm_recalc_len)

END (MEMMOVE)