Port to GNU assembly.

git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@189 6f19259b-4bc3-4df7-8a09-765794883524

11 files changed, 671 insertions, 0 deletions

diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/CompareMem.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/CompareMem.s
new file mode 100644
index 0000000000..c01b0c1358
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/CompareMem.s
@@ -0,0 +1,48 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   CompareMem.Asm
+#
+# Abstract:
+#
+#   CompareMem function
+#
+# Notes:
+#
+#   The following BaseMemoryLib instances share the same version of this file:
+#
+#       BaseMemoryLibRepStr
+#       BaseMemoryLibMmx
+#       BaseMemoryLibSse2
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .code: 
+
+.global _InternalMemCompareMem
+_InternalMemCompareMem:
+    push    %esi
+    push    %edi
+    movl    12(%esp), %esi
+    movl    16(%esp), %edi
+    movl    20(%esp), %ecx
+    repe    cmpsb
+    movzbl  -1(%esi), %eax
+    movzbl  -1(%edi), %edx
+    subl    %edx, %eax
+    subl    %edx, %eax
+    pop     %edi
+    pop     %esi
+    ret
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/CopyMem.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/CopyMem.s
new file mode 100644
index 0000000000..deb9860339
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/CopyMem.s
@@ -0,0 +1,88 @@
+#------------------------------------------------------------------------------
+#
+# 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
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem16.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem16.s
new file mode 100644
index 0000000000..c433a28b72
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem16.s
@@ -0,0 +1,44 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   ScanMem16.Asm
+#
+# Abstract:
+#
+#   ScanMem16 function
+#
+# Notes:
+#
+#   The following BaseMemoryLib instances share the same version of this file:
+#
+#       BaseMemoryLibRepStr
+#       BaseMemoryLibMmx
+#       BaseMemoryLibSse2
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .code: 
+
+.global _InternalMemScanMem16
+_InternalMemScanMem16:
+    push    %edi
+    movl    12(%esp), %ecx
+    movl    8(%esp), %edi
+    movl    16(%esp), %eax
+    repne   scasw
+    leal    -2(%edi), %eax
+    cmovnz  %ecx, %eax
+    pop     %edi
+    ret
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem32.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem32.s
new file mode 100644
index 0000000000..1de283c788
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem32.s
@@ -0,0 +1,44 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   ScanMem32.Asm
+#
+# Abstract:
+#
+#   ScanMem32 function
+#
+# Notes:
+#
+#   The following BaseMemoryLib instances share the same version of this file:
+#
+#       BaseMemoryLibRepStr
+#       BaseMemoryLibMmx
+#       BaseMemoryLibSse2
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .code: 
+
+.global _InternalMemScanMem32
+_InternalMemScanMem32:
+    push    %edi
+    movl    12(%esp), %ecx
+    movl    8(%esp), %edi
+    movl    16(%esp), %eax
+    repnz   scasl
+    leal    -4(%edi), %eax
+    cmovnz  %ecx, %eax
+    pop     %edi
+    ret
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem64.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem64.s
new file mode 100644
index 0000000000..00f6d74436
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem64.s
@@ -0,0 +1,53 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   ScanMem64.Asm
+#
+# Abstract:
+#
+#   ScanMem64 function
+#
+# Notes:
+#
+#   The following BaseMemoryLib instances share the same version of this file:
+#
+#       BaseMemoryLibRepStr
+#       BaseMemoryLibMmx
+#       BaseMemoryLibSse2
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .code: 
+
+.global _InternalMemScanMem64
+_InternalMemScanMem64:
+    push    %edi
+    movl    12(%esp), %ecx
+    movl    16(%esp), %eax
+    movl    20(%esp), %edx
+    movl    8(%esp), %edi
+L0: 
+    cmpl    (%edi), %eax
+    leal    8(%edi), %edi
+    loopne  L0
+    jne     L1
+    cmpl    -4(%edi), %edx
+    jecxz   L1
+    jne     L0
+L1: 
+    leal    -8(%edi), %eax
+    cmovne  %ecx, %eax
+    pop     %edi
+    ret
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem8.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem8.s
new file mode 100644
index 0000000000..57640378b4
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/ScanMem8.s
@@ -0,0 +1,44 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   ScanMem8.Asm
+#
+# Abstract:
+#
+#   ScanMem8 function
+#
+# Notes:
+#
+#   The following BaseMemoryLib instances share the same version of this file:
+#
+#       BaseMemoryLibRepStr
+#       BaseMemoryLibMmx
+#       BaseMemoryLibSse2
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .code: 
+
+.global _InternalMemScanMem8
+_InternalMemScanMem8:
+    push    %edi
+    movl    12(%esp), %ecx
+    movl    8(%esp), %edi
+    movb    16(%esp), %al
+    repne   scasb
+    leal    -1(%edi), %eax
+    cmovnz  %ecx, %eax
+    pop     %edi
+    ret
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem.s
new file mode 100644
index 0000000000..e778cd8b86
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem.s
@@ -0,0 +1,76 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   SetMem.asm
+#
+# Abstract:
+#
+#   SetMem function
+#
+# Notes:
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .xmm: 
+    .code: 
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  _mem_SetMem (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count,
+#    IN UINT8  Value
+#    )
+#------------------------------------------------------------------------------
+.global _InternalMemSetMem
+_InternalMemSetMem:
+    push    %edi
+    movl    12(%esp), %edx              # edx <- Count
+    movl    8(%esp), %edi               # edi <- Buffer
+    movb    16(%esp), %al               # al <- Value
+    xorl    %ecx, %ecx
+    subl    %edi, %ecx
+    andl    $15, %ecx                   # ecx + edi aligns on 16-byte boundary
+    jz      L0
+    cmpl    %edx, %ecx
+    cmova   %edx, %ecx
+    subl    %ecx, %edx
+    rep
+    stosb
+L0: 
+    movl    %edx, %ecx
+    andl    $15, %edx
+    shrl    $4, %ecx                    # ecx <- # of DQwords to set
+    jz      @SetBytes
+    movb    %al, %ah                    # ax <- Value | (Value << 8)
+    addl    $-16, %esp
+    movdqu  %xmm0, (%esp)
+    movd    %eax, %xmm0
+    pshuflw $0, %xmm0, %xmm0
+    movlhps %xmm0, %xmm0
+L1: 
+    movntdq %xmm0, (%edi)
+    addl    $16, %edi
+    loop   L1
+    mfence
+    movdqu  (%esp), %xmm0
+    addl    $16, %esp                   # stack cleanup
+@SetBytes: 
+    movl    %edx, %ecx
+    rep
+    stosb
+    movl    8(%esp), %eax               # eax <- Buffer as return value
+    pop     %edi
+    ret
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem16.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem16.s
new file mode 100644
index 0000000000..27110b32d3
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem16.s
@@ -0,0 +1,72 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   SetMem16.asm
+#
+# Abstract:
+#
+#   SetMem16 function
+#
+# Notes:
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .xmm: 
+    .code: 
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  _mem_SetMem16 (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count,
+#    IN UINT16 Value
+#    )
+#------------------------------------------------------------------------------
+.global _InternalMemSetMem16
+_InternalMemSetMem16:
+    push    %edi
+    movl    12(%esp), %edx
+    movl    8(%esp), %edi
+    xorl    %ecx, %ecx
+    subl    %edi, %ecx
+    andl    $15, %ecx                   # ecx + edi aligns on 16-byte boundary
+    movl    16(%esp), %eax
+    jz      L0
+    shrl    %ecx
+    cmpl    %edx, %ecx
+    cmova   %edx, %ecx
+    subl    %ecx, %edx
+    rep
+    stosw
+L0: 
+    movl    %edx, %ecx
+    andl    $7, %edx
+    shrl    $3, %ecx
+    jz      @SetWords
+    movd    %eax, %xmm0
+    pshuflw $0, %xmm0, %xmm0
+    movlhps %xmm0, %xmm0
+L1:
+    movntdq %xmm0, (%edi)
+    addl    $16, %edi
+    loop    L1
+    mfence
+@SetWords: 
+    movl    %edx, %ecx
+    rep
+    stosw
+    movl    8(%esp), %eax
+    pop     %edi
+    ret
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem32.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem32.s
new file mode 100644
index 0000000000..a1a6317115
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem32.s
@@ -0,0 +1,71 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   SetMem32.asm
+#
+# Abstract:
+#
+#   SetMem32 function
+#
+# Notes:
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .xmm: 
+    .code: 
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  _mem_SetMem32 (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count,
+#    IN UINT32 Value
+#    )
+#------------------------------------------------------------------------------
+.global _InternalMemSetMem32
+_InternalMemSetMem32:
+    push    %edi
+    movl    12(%esp), %edx
+    movl    8(%esp), %edi
+    xorl    %ecx, %ecx
+    subl    %edi, %ecx
+    andl    $15, %ecx                   # ecx + edi aligns on 16-byte boundary
+    movl    16(%esp), %eax
+    jz      L0
+    shrl    $2, %ecx
+    cmpl    %edx, %ecx
+    cmova   %edx, %ecx
+    subl    %ecx, %edx
+    rep
+    stosl
+L0: 
+    movl    %edx, %ecx
+    andl    $3, %edx
+    shrl    $2, %ecx
+    jz      @SetDwords
+    movd    %eax, %xmm0
+    pshufd  $0, %xmm0, %xmm0
+L1: 
+    movntdq %xmm0, (%edi)
+    addl    $16, %edi
+    loop    L1
+    mfence
+@SetDwords: 
+    movl    %edx, %ecx
+    rep
+    stosl
+    movl    8(%esp), %eax
+    pop     %edi
+    ret
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem64.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem64.s
new file mode 100644
index 0000000000..2535a61aca
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/SetMem64.s
@@ -0,0 +1,63 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   SetMem64.asm
+#
+# Abstract:
+#
+#   SetMem64 function
+#
+# Notes:
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .xmm: 
+    .code: 
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  _mem_SetMem64 (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count,
+#    IN UINT64 Value
+#    )
+#------------------------------------------------------------------------------
+.global _InternalMemSetMem64
+_InternalMemSetMem64:
+    push    %edi
+    movl    12(%esp), %ecx
+    movl    8(%esp), %edi
+    testl   $8, %edi
+    movddup 16(%esp), %xmm0
+    jz      L0
+    movq    %xmm0, (%edi)
+    addl    $8, %edi
+    decl    %ecx
+L0: 
+    movl    %ecx, %edx
+    shrl    %ecx
+    jz      @SetQwords
+L1: 
+    movntdq %xmm0, (%edi)
+    addl    $16, %edi
+    loop    L1
+    mfence
+@SetQwords: 
+    testb   $1, %dl
+    jz      L2
+    movq    %xmm0, (%edi)
+L2: 
+    pop     %edi
+    ret
diff --git a/MdePkg/Library/BaseMemoryLibSse2/ia32/ZeroMem.s b/MdePkg/Library/BaseMemoryLibSse2/ia32/ZeroMem.s
new file mode 100644
index 0000000000..f927f9128f
--- /dev/null
+++ b/MdePkg/Library/BaseMemoryLibSse2/ia32/ZeroMem.s
@@ -0,0 +1,68 @@
+#------------------------------------------------------------------------------
+#
+# 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:
+#
+#   ZeroMem.asm
+#
+# Abstract:
+#
+#   ZeroMem function
+#
+# Notes:
+#
+#------------------------------------------------------------------------------
+
+    .686: 
+    #.MODEL flat,C
+    .xmm: 
+    .code: 
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  _mem_ZeroMem (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count
+#    )
+#------------------------------------------------------------------------------
+.global _InternalMemZeroMem
+_InternalMemZeroMem:
+    push    %edi
+    movl    8(%esp), %edi
+    movl    12(%esp), %edx
+    xorl    %ecx, %ecx
+    subl    %edi, %ecx
+    xorl    %eax, %eax
+    andl    $15, %ecx
+    jz      L0
+    cmpl    %edx, %ecx
+    cmova   %edx, %ecx
+    subl    %ecx, %edx
+    rep
+    stosb
+L0: 
+    movl    %edx, %ecx
+    andl    $15, %edx
+    shrl    $4, %ecx
+    jz      @ZeroBytes
+    pxor    %xmm0, %xmm0
+L1: 
+    movntdq %xmm0, (%edi)
+    addl    $16, %edi
+    loop    L1
+    mfence
+@ZeroBytes: 
+    movl    %edx, %ecx
+    rep
+    stosb
+    movl    8(%esp), %eax
+    pop     %edi
+    ret