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

//\r
// Include common header file for this module.\r
//\r
#include "CommonHeader.h"\r
\r
#------------------------------------------------------------------------------\r
#\r
# Copyright (c) 2006, Intel Corporation\r
# All rights reserved. This program and the accompanying materials\r
# are licensed and made available under the terms and conditions of the BSD License\r
# which accompanies this distribution.  The full text of the license may be found at\r
# http://opensource.org/licenses/bsd-license.php\r
#\r
# THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,\r
# WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.\r
#\r
# Module Name:\r
#\r
#   CopyMem.asm\r
#\r
# Abstract:\r
#\r
#   CopyMem function\r
#\r
# Notes:\r
#\r
#------------------------------------------------------------------------------\r
\r
.globl _InternalMemCopyMem\r
\r
#------------------------------------------------------------------------------\r
#  VOID *\r
#  EFIAPI\r
#  InternalMemCopyMem (\r
#    IN VOID   *Destination,\r
#    IN VOID   *Source,\r
#    IN UINTN  Count\r
#    );\r
#------------------------------------------------------------------------------\r
_InternalMemCopyMem:\r
    push    %esi\r
    push    %edi\r
    movl    16(%esp), %esi              # esi <- Source\r
    movl    12(%esp), %edi              # edi <- Destination\r
    movl    20(%esp), %edx              # edx <- Count\r
    leal    -1(%esi,%edx,), %eax        # eax <- End of Source\r
    cmpl    %edi, %esi\r
    jae     L0\r
    cmpl    %edi, %eax                  # Overlapped?\r
    jae     L_CopyBackward               # Copy backward if overlapped\r
L0:\r
    xorl    %ecx, %ecx\r
    subl    %edi, %ecx\r
    andl    $15, %ecx                   # ecx + edi aligns on 16-byte boundary\r
    jz      L1\r
    cmpl    %edx, %ecx\r
    cmova   %edx, %ecx\r
    subl    %ecx, %edx                  # edx <- remaining bytes to copy\r
    rep\r
    movsb\r
L1:\r
    movl    %edx, %ecx\r
    andl    $15, %edx\r
    shrl    $4, %ecx                    # ecx <- # of DQwords to copy\r
    jz      L_CopyBytes\r
    addl    $-16, %esp\r
    movdqu  %xmm0, (%esp)\r
L2:\r
    movdqu  (%esi), %xmm0\r
    movntdq %xmm0, (%edi)\r
    addl    $16, %esi\r
    addl    $16, %edi\r
    loop    L2\r
    mfence\r
    movdqu  (%esp),%xmm0\r
    addl    $16, %esp                   # stack cleanup\r
    jmp     L_CopyBytes\r
L_CopyBackward:\r
    movl    %eax, %esi                  # esi <- Last byte in Source\r
    leal    -1(%edi,%edx,), %edi        # edi <- Last byte in Destination\r
    std\r
L_CopyBytes:\r
    movl    %edx, %ecx\r
    rep\r
    movsb\r
    cld\r
    movl    12(%esp), %eax              # eax <- Destination as return value\r
    pop     %edi\r
    pop     %esi\r
    ret\r
Commit	Line	Data
e1f414b6	1	//\r
	2	// Include common header file for this module.\r
	3	//\r
	4	#include "CommonHeader.h"\r
	5	\r
	6	#------------------------------------------------------------------------------\r
	7	#\r
	8	# Copyright (c) 2006, Intel Corporation\r
	9	# All rights reserved. This program and the accompanying materials\r
	10	# are licensed and made available under the terms and conditions of the BSD License\r
	11	# which accompanies this distribution. The full text of the license may be found at\r
	12	# http://opensource.org/licenses/bsd-license.php\r
	13	#\r
	14	# THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,\r
	15	# WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.\r
	16	#\r
	17	# Module Name:\r
	18	#\r
	19	# CopyMem.asm\r
	20	#\r
	21	# Abstract:\r
	22	#\r
	23	# CopyMem function\r
	24	#\r
	25	# Notes:\r
	26	#\r
	27	#------------------------------------------------------------------------------\r
	28	\r
	29	.globl _InternalMemCopyMem\r
	30	\r
	31	#------------------------------------------------------------------------------\r
	32	# VOID *\r
	33	# EFIAPI\r
	34	# InternalMemCopyMem (\r
	35	# IN VOID *Destination,\r
	36	# IN VOID *Source,\r
	37	# IN UINTN Count\r
	38	# );\r
	39	#------------------------------------------------------------------------------\r
	40	_InternalMemCopyMem:\r
	41	push %esi\r
	42	push %edi\r
	43	movl 16(%esp), %esi # esi <- Source\r
	44	movl 12(%esp), %edi # edi <- Destination\r
	45	movl 20(%esp), %edx # edx <- Count\r
	46	leal -1(%esi,%edx,), %eax # eax <- End of Source\r
	47	cmpl %edi, %esi\r
	48	jae L0\r
	49	cmpl %edi, %eax # Overlapped?\r
	50	jae L_CopyBackward # Copy backward if overlapped\r
	51	L0:\r
	52	xorl %ecx, %ecx\r
	53	subl %edi, %ecx\r
	54	andl $15, %ecx # ecx + edi aligns on 16-byte boundary\r
	55	jz L1\r
	56	cmpl %edx, %ecx\r
	57	cmova %edx, %ecx\r
	58	subl %ecx, %edx # edx <- remaining bytes to copy\r
	59	rep\r
	60	movsb\r
	61	L1:\r
	62	movl %edx, %ecx\r
	63	andl $15, %edx\r
	64	shrl $4, %ecx # ecx <- # of DQwords to copy\r
65	jz L_CopyBytes\r
66	addl $-16, %esp\r
67	movdqu %xmm0, (%esp)\r
68	L2:\r
69	movdqu (%esi), %xmm0\r
70	movntdq %xmm0, (%edi)\r
71	addl $16, %esi\r
72	addl $16, %edi\r
73	loop L2\r
74	mfence\r
75	movdqu (%esp),%xmm0\r
76	addl $16, %esp # stack cleanup\r
77	jmp L_CopyBytes\r
78	L_CopyBackward:\r
79	movl %eax, %esi # esi <- Last byte in Source\r
80	leal -1(%edi,%edx,), %edi # edi <- Last byte in Destination\r
81	std\r
82	L_CopyBytes:\r
83	movl %edx, %ecx\r
84	rep\r
85	movsb\r
86	cld\r
87	movl 12(%esp), %eax # eax <- Destination as return value\r
88	pop %edi\r
89	pop %esi\r
90	ret\r