[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:
#
#------------------------------------------------------------------------------

    .686: 
    #.MODEL flat,C
    .xmm: 
    .code: 

#------------------------------------------------------------------------------
#  VOID *
#  _mem_CopyMem (
#    IN VOID   *Destination,
#    IN VOID   *Source,
#    IN UINTN  Count
#    )
#------------------------------------------------------------------------------
.global _InternalMemCopyMem
_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(%edi,%edx,), %eax        # eax <- End of Destination
    cmpl    %edi, %esi
    jae     L0
    cmpl    %esi, %eax                  # Overlapped?
    jae     @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      @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     @CopyBytes
@CopyBackward: 
    movl    %eax, %edi                  # edi <- Last byte in Destination
    leal    -1(%esi,%edx,), %esi        # esi <- Last byte in Source
    std
@CopyBytes: 
    movl    %edx, %ecx
    rep
    movsb
    cld
    movl    12(%esp), %eax              # eax <- Destination as return value
    pop     %edi
    pop     %esi
    ret
Commit	Line	Data
6e3b8c47	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	.686:
	25	#.MODEL flat,C
	26	.xmm:
	27	.code:
	28
	29	#------------------------------------------------------------------------------
	30	# VOID *
	31	# _mem_CopyMem (
	32	# IN VOID *Destination,
	33	# IN VOID *Source,
	34	# IN UINTN Count
	35	# )
	36	#------------------------------------------------------------------------------
	37	.global _InternalMemCopyMem
	38	_InternalMemCopyMem:
	39	push %esi
	40	push %edi
	41	movl 16(%esp), %esi # esi <- Source
	42	movl 12(%esp), %edi # edi <- Destination
	43	movl 20(%esp), %edx # edx <- Count
	44	leal -1(%edi,%edx,), %eax # eax <- End of Destination
	45	cmpl %edi, %esi
	46	jae L0
	47	cmpl %esi, %eax # Overlapped?
	48	jae @CopyBackward # Copy backward if overlapped
	49	L0:
	50	xorl %ecx, %ecx
	51	subl %edi, %ecx
	52	andl $15, %ecx # ecx + edi aligns on 16-byte boundary
	53	jz L1
	54	cmpl %edx, %ecx
	55	cmova %edx, %ecx
	56	subl %ecx, %edx # edx <- remaining bytes to copy
	57	rep
	58	movsb
	59	L1:
	60	movl %edx, %ecx
	61	andl $15, %edx
	62	shrl $4, %ecx # ecx <- # of DQwords to copy
	63	jz @CopyBytes
	64	addl $-16, %esp
65	movdqu %xmm0, (%esp)
66	L2:
67	movdqu (%esi), %xmm0
68	movntdq %xmm0, (%edi)
69	addl $16, %esi
70	addl $16, %edi
71	loop L2
72	mfence
73	movdqu (%esp),%xmm0
74	addl $16, %esp # stack cleanup
75	jmp @CopyBytes
76	@CopyBackward:
77	movl %eax, %edi # edi <- Last byte in Destination
78	leal -1(%esi,%edx,), %esi # esi <- Last byte in Source
79	std
80	@CopyBytes:
81	movl %edx, %ecx
82	rep
83	movsb
84	cld
85	movl 12(%esp), %eax # eax <- Destination as return value
86	pop %edi
87	pop %esi
88	ret