[mirror_edk2.git] / MdePkg / Library / BaseMemoryLibSse2 / Ia32 / SetMem.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
#   SetMem.asm\r
#\r
# Abstract:\r
#\r
#   SetMem function\r
#\r
# Notes:\r
#\r
#------------------------------------------------------------------------------\r
\r
    .686:\r
    #.MODEL flat,C\r
    .xmm:\r
    .code:\r
\r
#------------------------------------------------------------------------------\r
#  VOID *\r
#  _mem_SetMem (\r
#    IN VOID   *Buffer,\r
#    IN UINTN  Count,\r
#    IN UINT8  Value\r
#    );\r
#------------------------------------------------------------------------------\r
.globl _InternalMemSetMem\r
_InternalMemSetMem:\r
    push    %edi\r
    movl    12(%esp), %edx              # edx <- Count\r
    movl    8(%esp), %edi               # edi <- Buffer\r
    movb    16(%esp), %al               # al <- Value\r
    xorl    %ecx, %ecx\r
    subl    %edi, %ecx\r
    andl    $15, %ecx                   # ecx + edi aligns on 16-byte boundary\r
    jz      L0\r
    cmpl    %edx, %ecx\r
    cmova   %edx, %ecx\r
    subl    %ecx, %edx\r
    rep\r
    stosb\r
L0:\r
    movl    %edx, %ecx\r
    andl    $15, %edx\r
    shrl    $4, %ecx                    # ecx <- # of DQwords to set\r
    jz      L_SetBytes\r
    movb    %al, %ah                    # ax <- Value | (Value << 8)\r
    addl    $-16, %esp\r
    movdqu  %xmm0, (%esp)\r
    movd    %eax, %xmm0\r
    pshuflw $0, %xmm0, %xmm0\r
    movlhps %xmm0, %xmm0\r
L1:\r
    movntdq %xmm0, (%edi)\r
    addl    $16, %edi\r
    loop   L1\r
    mfence\r
    movdqu  (%esp), %xmm0\r
    addl    $16, %esp                   # stack cleanup\r
L_SetBytes:\r
    movl    %edx, %ecx\r
    rep\r
    stosb\r
    movl    8(%esp), %eax               # eax <- Buffer as return value\r
    pop     %edi\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	# SetMem.asm\r
	20	#\r
	21	# Abstract:\r
	22	#\r
	23	# SetMem function\r
	24	#\r
	25	# Notes:\r
	26	#\r
	27	#------------------------------------------------------------------------------\r
	28	\r
	29	.686:\r
	30	#.MODEL flat,C\r
	31	.xmm:\r
	32	.code:\r
	33	\r
	34	#------------------------------------------------------------------------------\r
	35	# VOID *\r
	36	# _mem_SetMem (\r
	37	# IN VOID *Buffer,\r
	38	# IN UINTN Count,\r
	39	# IN UINT8 Value\r
	40	# );\r
	41	#------------------------------------------------------------------------------\r
	42	.globl _InternalMemSetMem\r
	43	_InternalMemSetMem:\r
	44	push %edi\r
	45	movl 12(%esp), %edx # edx <- Count\r
	46	movl 8(%esp), %edi # edi <- Buffer\r
	47	movb 16(%esp), %al # al <- Value\r
	48	xorl %ecx, %ecx\r
	49	subl %edi, %ecx\r
	50	andl $15, %ecx # ecx + edi aligns on 16-byte boundary\r
	51	jz L0\r
	52	cmpl %edx, %ecx\r
	53	cmova %edx, %ecx\r
	54	subl %ecx, %edx\r
	55	rep\r
	56	stosb\r
	57	L0:\r
	58	movl %edx, %ecx\r
	59	andl $15, %edx\r
	60	shrl $4, %ecx # ecx <- # of DQwords to set\r
	61	jz L_SetBytes\r
	62	movb %al, %ah # ax <- Value \| (Value << 8)\r
	63	addl $-16, %esp\r
	64	movdqu %xmm0, (%esp)\r
65	movd %eax, %xmm0\r
66	pshuflw $0, %xmm0, %xmm0\r
67	movlhps %xmm0, %xmm0\r
68	L1:\r
69	movntdq %xmm0, (%edi)\r
70	addl $16, %edi\r
71	loop L1\r
72	mfence\r
73	movdqu (%esp), %xmm0\r
74	addl $16, %esp # stack cleanup\r
75	L_SetBytes:\r
76	movl %edx, %ecx\r
77	rep\r
78	stosb\r
79	movl 8(%esp), %eax # eax <- Buffer as return value\r
80	pop %edi\r
81	ret\r