[mirror_edk2.git] / MdePkg / Library / BaseMemoryLibSse2 / Ia32 / CopyMem.S

#------------------------------------------------------------------------------
#
# Copyright (c) 2006, Intel Corporation
# All rights reserved. This program and the accompanying materials
# are licensed and made available under the terms and conditions of the BSD License
# which accompanies this distribution.  The full text of the license may be found at
# http://opensource.org/licenses/bsd-license.php
#
# THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,
# WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.
#
# Module Name:
#
#   CopyMem.asm
#
# Abstract:
#
#   CopyMem function
#
# Notes:
#
#------------------------------------------------------------------------------

.globl _InternalMemCopyMem

#------------------------------------------------------------------------------
#  VOID *
#  EFIAPI
#  InternalMemCopyMem (
#    IN VOID   *Destination,
#    IN VOID   *Source,
#    IN UINTN  Count
#    );
#------------------------------------------------------------------------------
_InternalMemCopyMem:
    push    %esi
    push    %edi
    movl    16(%esp), %esi              # esi <- Source
    movl    12(%esp), %edi              # edi <- Destination
    movl    20(%esp), %edx              # edx <- Count
    leal    -1(%esi,%edx,), %eax        # eax <- End of Source
    cmpl    %edi, %esi
    jae     L0
    cmpl    %edi, %eax                  # Overlapped?
    jae     L_CopyBackward               # Copy backward if overlapped
L0:
    xorl    %ecx, %ecx
    subl    %edi, %ecx
    andl    $15, %ecx                   # ecx + edi aligns on 16-byte boundary
    jz      L1
    cmpl    %edx, %ecx
    cmova   %edx, %ecx
    subl    %ecx, %edx                  # edx <- remaining bytes to copy
    rep
    movsb
L1:
    movl    %edx, %ecx
    andl    $15, %edx
    shrl    $4, %ecx                    # ecx <- # of DQwords to copy
    jz      L_CopyBytes
    addl    $-16, %esp
    movdqu  %xmm0, (%esp)
L2:
    movdqu  (%esi), %xmm0
    movntdq %xmm0, (%edi)
    addl    $16, %esi
    addl    $16, %edi
    loop    L2
    mfence
    movdqu  (%esp),%xmm0
    addl    $16, %esp                   # stack cleanup
    jmp     L_CopyBytes
L_CopyBackward:
    movl    %eax, %esi                  # esi <- Last byte in Source
    leal    -1(%edi,%edx,), %edi        # edi <- Last byte in Destination
    std
L_CopyBytes:
    movl    %edx, %ecx
    rep
    movsb
    cld
    movl    12(%esp), %eax              # eax <- Destination as return value
    pop     %edi
    pop     %esi
    ret
Commit	Line	Data
f1baef62	1	#------------------------------------------------------------------------------
	2	#
	3	# Copyright (c) 2006, Intel Corporation
	4	# All rights reserved. This program and the accompanying materials
	5	# are licensed and made available under the terms and conditions of the BSD License
	6	# which accompanies this distribution. The full text of the license may be found at
	7	# http://opensource.org/licenses/bsd-license.php
	8	#
	9	# THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,
	10	# WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.
	11	#
	12	# Module Name:
	13	#
	14	# CopyMem.asm
	15	#
	16	# Abstract:
	17	#
	18	# CopyMem function
	19	#
	20	# Notes:
	21	#
	22	#------------------------------------------------------------------------------
	23
	24	.globl _InternalMemCopyMem
	25
	26	#------------------------------------------------------------------------------
	27	# VOID *
	28	# EFIAPI
	29	# InternalMemCopyMem (
	30	# IN VOID *Destination,
	31	# IN VOID *Source,
	32	# IN UINTN Count
	33	# );
	34	#------------------------------------------------------------------------------
	35	_InternalMemCopyMem:
	36	push %esi
	37	push %edi
	38	movl 16(%esp), %esi # esi <- Source
	39	movl 12(%esp), %edi # edi <- Destination
	40	movl 20(%esp), %edx # edx <- Count
	41	leal -1(%esi,%edx,), %eax # eax <- End of Source
	42	cmpl %edi, %esi
	43	jae L0
	44	cmpl %edi, %eax # Overlapped?
	45	jae L_CopyBackward # Copy backward if overlapped
	46	L0:
	47	xorl %ecx, %ecx
	48	subl %edi, %ecx
	49	andl $15, %ecx # ecx + edi aligns on 16-byte boundary
	50	jz L1
	51	cmpl %edx, %ecx
	52	cmova %edx, %ecx
	53	subl %ecx, %edx # edx <- remaining bytes to copy
	54	rep
	55	movsb
	56	L1:
	57	movl %edx, %ecx
	58	andl $15, %edx
	59	shrl $4, %ecx # ecx <- # of DQwords to copy
	60	jz L_CopyBytes
	61	addl $-16, %esp
	62	movdqu %xmm0, (%esp)
	63	L2:
	64	movdqu (%esi), %xmm0
65	movntdq %xmm0, (%edi)
66	addl $16, %esi
67	addl $16, %edi
68	loop L2
69	mfence
70	movdqu (%esp),%xmm0
71	addl $16, %esp # stack cleanup
72	jmp L_CopyBytes
73	L_CopyBackward:
74	movl %eax, %esi # esi <- Last byte in Source
75	leal -1(%edi,%edx,), %edi # edi <- Last byte in Destination
76	std
77	L_CopyBytes:
78	movl %edx, %ecx
79	rep
80	movsb
81	cld
82	movl 12(%esp), %eax # eax <- Destination as return value
83	pop %edi
84	pop %esi
85	ret