1. Updated function headers for all assembly function

2. Optimized register usage in SetMemXX functions in all lib instances
3. Fixed a logical error in CopyMem for all lib instances


git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@1139 6f19259b-4bc3-4df7-8a09-765794883524

MdePkg/Library/BaseMemoryLib/ZeroMemWrapper.c

@@ -47,5 +47,5 @@ ZeroMem (
 {
   ASSERT (!(Buffer == NULL && Length > 0));
   ASSERT (Length <= (MAX_ADDRESS - (UINTN)Buffer + 1));
-  return InternalMemSetMem (Buffer, Length, 0);
+  return InternalMemZeroMem (Buffer, Length);
 }

MdePkg/Library/BaseMemoryLibMmx/BaseMemoryLibMmx.msa

@@ -89,6 +89,7 @@ WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.
     <Filename SupArchList="X64">X64/ScanMem16.asm</Filename>
     <Filename SupArchList="X64">X64/ScanMem32.asm</Filename>
     <Filename SupArchList="X64">X64/ScanMem64.asm</Filename>
+    <Filename SupArchList="X64">X64/ZeroMem.asm</Filename>
   </SourceFiles>
   <PackageDependencies>
     <Package PackageGuid="5e0e9358-46b6-4ae2-8218-4ab8b9bbdcec"/>

MdePkg/Library/BaseMemoryLibMmx/Ia32/CompareMem.S

@@ -27,11 +27,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemCompareMem
+
+#------------------------------------------------------------------------------
+# INTN
+# EFIAPI
+# InternalMemCompareMem (
+#   IN      CONST VOID                *DestinationBuffer,
+#   IN      CONST VOID                *SourceBuffer,
+#   IN      UINTN                     Length
+#   );
+#------------------------------------------------------------------------------
 _InternalMemCompareMem:
     push    %esi
     push    %edi
@@ -45,5 +51,3 @@ _InternalMemCompareMem:
     pop     %edi
     pop     %esi
     ret
-
-

MdePkg/Library/BaseMemoryLibMmx/Ia32/CompareMem.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; INTN
+; EFIAPI
+; InternalMemCompareMem (
+;   IN      CONST VOID                *DestinationBuffer,
+;   IN      CONST VOID                *SourceBuffer,
+;   IN      UINTN                     Length
+;   );
+;------------------------------------------------------------------------------
 InternalMemCompareMem   PROC    USES    esi edi
     mov     esi, [esp + 12]
     mov     edi, [esp + 16]

MdePkg/Library/BaseMemoryLibMmx/Ia32/CopyMem.S

@@ -21,30 +21,27 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemCopyMem
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_CopyMem (
+#  EFIAPI
+#  InternalMemCopyMem (
 #    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
+    leal    -1(%esi,%edx,), %eax        # eax <- End of Source
     cmpl    %edi, %esi
     jae     L0
-    cmpl    %esi, %eax                  # Overlapped?
+    cmpl    %edi, %eax                  # Overlapped?
     jae     @CopyBackward               # Copy backward if overlapped
 L0:
     xorl    %ecx, %ecx
@@ -66,18 +63,17 @@ L1:
     movq    %mm0, (%esp)                # save mm0
 L2:
     movq    (%esi), %mm0
-    movntq  %mm0, (%edi)
+    movq    %mm0, (%edi)
     addl    $8, %esi
     addl    $8, %edi
     loop    L2
-    mfence
     movq    (%esp), %mm0                # restore mm0
     popl    %ecx                        # stack cleanup
     popl    %ecx                        # stack cleanup
     jmp     @CopyBytes
 @CopyBackward:
-    movl    %eax, %edi                  # edi <- Last byte in Destination
+    movl    %eax, %esi                  # esi <- Last byte in Source
-    leal    -1(%esi,%edx,), %esi        # esi <- Last byte in Source
+    leal    -1(%edi,%edx,), %edi        # edi <- Last byte in Destination
     std
 @CopyBytes:
     movl    %edx, %ecx

MdePkg/Library/BaseMemoryLibMmx/Ia32/CopyMem.asm

@@ -23,35 +23,27 @@
 
     .686
     .model  flat,C
-    .xmm
+    .mmx
     .code
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_CopyMem (
+;  EFIAPI
+;  InternalMemCopyMem (
 ;    IN VOID   *Destination,
 ;    IN VOID   *Source,
 ;    IN UINTN  Count
-;    )
+;    );
 ;------------------------------------------------------------------------------
 InternalMemCopyMem  PROC    USES    esi edi
     mov     esi, [esp + 16]             ; esi <- Source
     mov     edi, [esp + 12]             ; edi <- Destination
     mov     edx, [esp + 20]             ; edx <- Count
-    lea     eax, [edi + edx - 1]        ; eax <- End of Destination
+    lea     eax, [esi + edx - 1]        ; eax <- End of Source
     cmp     esi, edi
     jae     @F
-    cmp     eax, esi                    ; Overlapped?
+    cmp     eax, edi                    ; Overlapped?
     jae     @CopyBackward               ; Copy backward if overlapped
-@@:
-    xor     ecx, ecx
-    sub     ecx, esi
-    and     ecx, 7                      ; ecx + esi aligns on 8-byte boundary
-    jz      @F
-    cmp     ecx, edx
-    cmova   ecx, edx
-    sub     edx, ecx                    ; edx <- remaining bytes to copy
-    rep     movsb
 @@:
     mov     ecx, edx
     and     edx, 7
@@ -62,18 +54,17 @@ InternalMemCopyMem  PROC    USES    esi edi
     movq    [esp], mm0                  ; save mm0
 @@:
     movq    mm0, [esi]
-    movntq  [edi], mm0
+    movq    [edi], mm0
     add     esi, 8
     add     edi, 8
     loop    @B
-    mfence
     movq    mm0, [esp]                  ; restore mm0
     pop     ecx                         ; stack cleanup
     pop     ecx                         ; stack cleanup
     jmp     @CopyBytes
 @CopyBackward:
-    mov     edi, eax                    ; edi <- Last byte in Destination
+    mov     esi, eax                    ; esi <- Last byte in Source
-    lea     esi, [esi + edx - 1]        ; esi <- Last byte in Source
+    lea     edi, [edi + edx - 1]        ; edi <- Last byte in Destination
     std
 @CopyBytes:
     mov     ecx, edx

MdePkg/Library/BaseMemoryLibMmx/Ia32/ScanMem16.S

@@ -27,11 +27,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemScanMem16
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem16 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT16                    Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem16:
     push    %edi
     movl    12(%esp), %ecx
@@ -42,5 +48,3 @@ _InternalMemScanMem16:
     cmovnz  %ecx, %eax
     pop     %edi
     ret
-
-

MdePkg/Library/BaseMemoryLibMmx/Ia32/ScanMem16.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem16 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT16                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem16    PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibMmx/Ia32/ScanMem32.S

@@ -27,11 +27,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemScanMem32
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem32 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT32                    Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem32:
     push    %edi
     movl    12(%esp), %ecx

MdePkg/Library/BaseMemoryLibMmx/Ia32/ScanMem32.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem32 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT32                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem32    PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibMmx/Ia32/ScanMem64.S

@@ -27,11 +27,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemScanMem64
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem64 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT64                    Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem64:
     push    %edi
     movl    12(%esp), %ecx

MdePkg/Library/BaseMemoryLibMmx/Ia32/ScanMem64.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem64 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT64                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem64    PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     eax, [esp + 16]

MdePkg/Library/BaseMemoryLibMmx/Ia32/ScanMem8.S

@@ -27,11 +27,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemScanMem8
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem8 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT8                     Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem8:
     push    %edi
     movl    12(%esp), %ecx

MdePkg/Library/BaseMemoryLibMmx/Ia32/ScanMem8.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem8 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT8                     Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem8 PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibMmx/Ia32/SetMem.S

@@ -21,44 +21,43 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemSetMem
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_SetMem (
+#  InternalMemSetMem (
 #    IN VOID   *Buffer,
 #    IN UINTN  Count,
 #    IN UINT8  Value
 #    )
 #------------------------------------------------------------------------------
-.global _InternalMemSetMem
 _InternalMemSetMem:
     push    %edi
+    movb    16(%esp), %al
+    movb    %al, %ah
+    shrdl   $16, %eax, %edx
+    shldl   $16, %edx, %eax
     movl    12(%esp), %ecx              # ecx <- Count
     movl    8(%esp), %edi               # edi <- Buffer
     movl    %ecx, %edx
-    shrl    $3, %ecx                    # # of Qwords to set
-    movb    16(%esp), %al               # al <- Value
-    jz      @SetBytes
-    movb    %al, %ah                    # ax <- Value | (Value << 8)
-    pushl   %ecx
-    pushl   %ecx
-    movq    %mm0, (%esp)                # save mm0
-    movd    %eax, %mm0
-    pshufw  $0x0,%mm0,%mm0
-L0: 
-    movntq  %mm0, (%edi)
-    addl    $8, %edi
-    loop    L0
-    mfence
-    movq    (%esp), %mm0                # restore mm0
-    popl    %ecx                        # stack cleanup
-    popl    %ecx
-@SetBytes: 
     andl    $7, %edx
+    shrl    $3, %ecx                    # # of Qwords to set
+    jz      L1
+    addl    $0x-10, %esp
+    movq    %mm0, (%esp)                # save mm0
+    movq    %mm1, 8(%esp)               # save mm1
+    movd    %eax, %mm0
+    movd    %eax, %mm1
+    psllq   $32, %mm0
+    por     %mm1, %mm0                  # fill mm0 with 8 Value's
+L0:
+    movq    %mm0, (%edi)
+    addl    $8, %edi
+    loopl   L0
+    movq    (%esp), %mm0                # restore mm0
+    movq    8(%esp), %mm1               # restore mm1
+    addl    $0x10, %esp                 # stack cleanup
+L1:
     movl    %edx, %ecx
     rep
     stosb

MdePkg/Library/BaseMemoryLibMmx/Ia32/SetMem.asm

@@ -23,40 +23,44 @@
 
     .686
     .model  flat,C
-    .xmm
+    .mmx
     .code
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_SetMem (
+;  EFIAPI
+;  InternalMemSetMem (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count,
 ;    IN UINT8  Value
 ;    )
 ;------------------------------------------------------------------------------
 InternalMemSetMem   PROC    USES    edi
+    mov     al, [esp + 16]
+    mov     ah, al
+    shrd    edx, eax, 16
+    shld    eax, edx, 16
     mov     ecx, [esp + 12]             ; ecx <- Count
     mov     edi, [esp + 8]              ; edi <- Buffer
     mov     edx, ecx
+    and     edx, 7
     shr     ecx, 3                      ; # of Qwords to set
-    mov     al, [esp + 16]              ; al <- Value
     jz      @SetBytes
-    mov     ah, al                      ; ax <- Value | (Value << 8)
+    add     esp, -10h
-    push    ecx
-    push    ecx
     movq    [esp], mm0                  ; save mm0
+    movq    [esp + 8], mm1              ; save mm1
     movd    mm0, eax
-    pshufw  mm0, mm0, 0                 ; fill mm0 with 8 Value's
+    movd    mm1, eax
+    psllq   mm0, 32
+    por     mm0, mm1                    ; fill mm0 with 8 Value's
 @@:
-    movntq  [edi], mm0
+    movq    [edi], mm0
     add     edi, 8
     loop    @B
-    mfence
     movq    mm0, [esp]                  ; restore mm0
-    pop     ecx                         ; stack cleanup
+    movq    mm1, [esp + 8]              ; restore mm1
-    pop     ecx
+    add     esp, 10h                    ; stack cleanup
 @SetBytes:
-    and     edx, 7
     mov     ecx, edx
     rep     stosb
     mov     eax, [esp + 8]              ; eax <- Buffer as return value

MdePkg/Library/BaseMemoryLibMmx/Ia32/SetMem16.S

@@ -21,37 +21,36 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemSetMem16
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_SetMem16 (
+#  InternalMemSetMem16 (
 #    IN VOID   *Buffer,
 #    IN UINTN  Count,
 #    IN UINT16 Value
 #    )
 #------------------------------------------------------------------------------
-.global _InternalMemSetMem16
 _InternalMemSetMem16:
     push    %edi
+    movl    16(%esp), %eax
+    shrdl   $16, %eax, %edx
+    shldl   $16, %edx, %eax
     movl    12(%esp), %edx
     movl    8(%esp), %edi
     movl    %edx, %ecx
     andl    $3, %edx
     shrl    $2, %ecx
-    movl    16(%esp), %eax
+    jz      L1
-    jz      @SetWords
     movd    %eax, %mm0
-    pshufw  $0, %mm0, %mm0
+    movd    %eax, %mm1
+    psllq   $32, %mm0
+    por     %mm1, %mm0
 L0:
-    movntq  %mm0, (%edi)
+    movq    %mm0, (%edi)
     addl    $8, %edi
-    loop    L0
+    loopl   L0
-    mfence
+L1:
-@SetWords: 
     movl    %edx, %ecx
     rep
     stosw

MdePkg/Library/BaseMemoryLibMmx/Ia32/SetMem16.asm

@@ -23,32 +23,36 @@
 
     .686
     .model  flat,C
-    .xmm
+    .mmx
     .code
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_SetMem16 (
+;  EFIAPI
+;  InternalMemSetMem16 (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count,
 ;    IN UINT16 Value
 ;    )
 ;------------------------------------------------------------------------------
 InternalMemSetMem16 PROC    USES    edi
+    mov     eax, [esp + 16]
+    shrd    edx, eax, 16
+    shld    eax, edx, 16
     mov     edx, [esp + 12]
     mov     edi, [esp + 8]
     mov     ecx, edx
     and     edx, 3
     shr     ecx, 2
-    mov     eax, [esp + 16]
     jz      @SetWords
     movd    mm0, eax
-    pshufw  mm0, mm0, 0
+    movd    mm1, eax
+    psllq   mm0, 32
+    por     mm0, mm1
 @@:
-    movntq  [edi], mm0
+    movq    [edi], mm0
     add     edi, 8
     loop    @B
-    mfence
 @SetWords:
     mov     ecx, edx
     rep     stosw

MdePkg/Library/BaseMemoryLibMmx/Ia32/SetMem32.S

@@ -21,39 +21,32 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemSetMem32
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_SetMem32 (
+#  InternalMemSetMem32 (
 #    IN VOID   *Buffer,
 #    IN UINTN  Count,
 #    IN UINT32 Value
 #    )
 #------------------------------------------------------------------------------
-.global _InternalMemSetMem32
 _InternalMemSetMem32:
-    push    %edi
+    movl    4(%esp), %eax
-    movl    12(%esp), %edx
+    movl    8(%esp), %ecx
-    movl    8(%esp), %edi
+    movd    12(%esp), %mm0
-    movl    %edx, %ecx
     shrl    %ecx
-    movd    16(%esp), %mm0
+    movl    %eax, %edx
-    movl    %edi, %eax
-    jz      @SetDwords
-    pshufw  $0x44, %mm0, %mm0
-L0: 
-    movntq  %mm0, (%edi)
-    addl    $8, %edi
-    loopl   L0
-    mfence
-@SetDwords: 
-    testb   $1, %dl
     jz      L1
-    movd    %mm0, (%edi)
+    movq    %mm0, %mm1
+    psllq   $32, %mm1
+    por     %mm1, %mm0
+L0:
+    movq    %mm0, (%edx)
+    lea     8(%edx), %edx
+    loopl   L0
 L1:
-    pop     %edi
+    jnc     L2
+    movd    %mm0, (%edx)
+L2:
     ret

MdePkg/Library/BaseMemoryLibMmx/Ia32/SetMem32.asm

@@ -23,35 +23,35 @@
 
     .686
     .model  flat,C
-    .xmm
+    .mmx
     .code
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_SetMem32 (
+;  EFIAPI
+;  InternalMemSetMem32 (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count,
 ;    IN UINT32 Value
-;    )
+;    );
 ;------------------------------------------------------------------------------
-InternalMemSetMem32 PROC    USES    edi
+InternalMemSetMem32 PROC
-    mov     edx, [esp + 12]
+    mov     eax, [esp + 4]              ; eax <- Buffer as return value
-    mov     edi, [esp + 8]
+    mov     ecx, [esp + 8]              ; ecx <- Count
-    mov     ecx, edx
+    movd    mm0, [esp + 12]             ; mm0 <- Value
-    shr     ecx, 1
+    shr     ecx, 1                      ; ecx <- number of qwords to set
-    movd    mm0, [esp + 16]
+    mov     edx, eax                    ; edx <- Buffer
-    mov     eax, edi
     jz      @SetDwords
-    pshufw  mm0, mm0, 44h
+    movq    mm1, mm0
+    psllq   mm1, 32
+    por     mm0, mm1
 @@:
-    movntq  [edi], mm0
+    movq    [edx], mm0
-    add     edi, 8
+    lea     edx, [edx + 8]              ; use "lea" to avoid change in flags
     loop    @B
-    mfence
 @SetDwords:
-    test    dl, 1
+    jnc     @F
-    jz      @F
+    movd    [edx], mm0
-    movd    [edi], mm0
 @@:
     ret
 InternalMemSetMem32 ENDP

MdePkg/Library/BaseMemoryLibMmx/Ia32/SetMem64.S

@@ -21,30 +21,23 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemSetMem64
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_SetMem64 (
+#  InternalMemSetMem64 (
 #    IN VOID   *Buffer,
 #    IN UINTN  Count,
 #    IN UINT64 Value
 #    )
 #------------------------------------------------------------------------------
-.global _InternalMemSetMem64
 _InternalMemSetMem64:
-    push    %edi
+    movl    4(%esp), %eax
-    movq    16(%esp), %mm0
+    movl    8(%esp), %ecx
-    movl    12(%esp), %ecx
+    movq    12(%esp), %mm0
-    movl    8(%esp), %edi
+    movl    %eax, %edx
-    movl    %edi, %eax
 L0:
-    movntq  %mm0, (%edi)
+    movq    %mm0, (%edx)
-    addl    $8, %edi
+    lea     8(%edx), %edx
     loopl   L0
-    mfence
-    pop     %edi
     ret

MdePkg/Library/BaseMemoryLibMmx/Ia32/SetMem64.asm

@@ -23,12 +23,13 @@
 
     .686
     .model  flat,C
-    .xmm
+    .mmx
     .code
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_SetMem64 (
+;  EFIAPI
+;  InternalMemSetMem64 (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count,
 ;    IN UINT64 Value
@@ -40,10 +41,9 @@ InternalMemSetMem64 PROC
     movq    mm0, [esp + 12]
     mov     edx, eax
 @@:
-    movntq  [edx], mm0
+    movq    [edx], mm0
     add     edx, 8
     loop    @B
-    mfence
     ret
 InternalMemSetMem64 ENDP
 

MdePkg/Library/BaseMemoryLibMmx/Ia32/ZeroMem.S

@@ -21,19 +21,16 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemZeroMem
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_ZeroMem (
+#  EFIAPI
+#  InternalMemZeroMem (
 #    IN VOID   *Buffer,
 #    IN UINTN  Count
-#    )
+#    );
 #------------------------------------------------------------------------------
-.global _InternalMemZeroMem
 _InternalMemZeroMem:
     push    %edi
     movl    8(%esp), %edi
@@ -43,10 +40,9 @@ _InternalMemZeroMem:
     jz      @ZeroBytes
     pxor    %mm0, %mm0
 L0:
-    movntq  %mm0, (%edi)
+    movq    %mm0, (%edi)
     addl    $8, %edi
     loop    L0
-    mfence
 @ZeroBytes:
     andl    $7, %edx
     xorl    %eax, %eax

MdePkg/Library/BaseMemoryLibMmx/Ia32/ZeroMem.asm

@@ -23,15 +23,15 @@
 
     .686
     .model  flat,C
-    .xmm
+    .mmx
     .code
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_ZeroMem (
+;  InternalMemZeroMem (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count
-;    )
+;    );
 ;------------------------------------------------------------------------------
 InternalMemZeroMem  PROC    USES    edi
     mov     edi, [esp + 8]
@@ -41,10 +41,9 @@ InternalMemZeroMem  PROC    USES    edi
     jz      @ZeroBytes
     pxor    mm0, mm0
 @@:
-    movntq  [edi], mm0
+    movq    [edi], mm0
     add     edi, 8
     loop    @B
-    mfence
 @ZeroBytes:
     and     edx, 7
     xor     eax, eax

MdePkg/Library/BaseMemoryLibMmx/X64/CompareMem.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; INTN
+; EFIAPI
+; InternalMemCompareMem (
+;   IN      CONST VOID                *DestinationBuffer,
+;   IN      CONST VOID                *SourceBuffer,
+;   IN      UINTN                     Length
+;   );
+;------------------------------------------------------------------------------
 InternalMemCompareMem   PROC    USES    rsi rdi
     mov     rsi, rcx
     mov     rdi, rdx

MdePkg/Library/BaseMemoryLibMmx/X64/CopyMem.asm

@@ -23,42 +23,42 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; VOID *
+; EFIAPI
+; InternalMemCopyMem (
+;   OUT     VOID                      *DestinationBuffer,
+;   IN      CONST VOID                *SourceBuffer,
+;   IN      UINTN                     Length
+;   );
+;------------------------------------------------------------------------------
 InternalMemCopyMem  PROC    USES    rsi rdi
     mov     rsi, rdx                    ; rsi <- Source
     mov     rdi, rcx                    ; rdi <- Destination
-    lea     r9, [rdi + r8 - 1]          ; r9 <- End of Destination
+    lea     r9, [rsi + r8 - 1]          ; r9 <- End of Source
     cmp     rsi, rdi
     mov     rax, rdi                    ; rax <- Destination as return value
     jae     @F
-    cmp     r9, rsi
+    cmp     r9, rdi
     jae     @CopyBackward               ; Copy backward if overlapped
-@@:
-    xor     rcx, rcx
-    sub     rcx, rsi
-    and     rcx, 7                      ; rcx + rsi aligns on 8-byte boundary
-    jz      @F
-    cmp     rcx, r8
-    cmova   rcx, r8
-    sub     r8, rcx                     ; r8 <- remaining bytes to copy
-    rep     movsb
 @@:
     mov     rcx, r8
     and     r8, 7
     shr     rcx, 3                      ; rcx <- # of Qwords to copy
     jz      @CopyBytes
-    DB      49h, 0fh, 7eh, 0c2h         ; movq    r10,  mm0    ; save mm0
+    DB      49h, 0fh, 7eh, 0c2h         ; movd r10, mm0 (Save mm0 in r10)
 @@:
-    DB      48h, 0fh, 6fh, 06h          ; movq    mm0, [rsi]
+    DB      0fh, 6fh, 06h               ; movd mm0, [rsi]
-    DB      48h, 0fh, 0e7h, 07h         ; movntq  [rdi], mm0
+    DB      0fh, 0e7h, 07h              ; movntq [rdi], mm0
     add     rsi, 8
     add     rdi, 8
     loop    @B
     mfence
-    DB      49h, 0fh, 6eh, 0c2h          ; movq    mm0, r10    ; restore mm0
+    DB      49h, 0fh, 6eh, 0c2h         ; movd mm0, r10 (Restore mm0)
     jmp     @CopyBytes
 @CopyBackward:
-    mov     rdi, r9                     ; rdi <- End of Destination
+    mov     rsi, r9                     ; rsi <- End of Source
-    lea     rsi, [rsi + r8 - 1]         ; rsi <- End of Source
+    lea     rdi, [rdi + r8 - 1]         ; rdi <- End of Destination
     std                                 ; set direction flag
 @CopyBytes:
     mov     rcx, r8

MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem16.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem16 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT16                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem16    PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem32.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem32 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT32                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem32    PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem64.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem64 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT64                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem64    PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem8.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem8 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT8                     Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem8 PROC    USES    rdi
     mov     rdi, rcx
     mov     rcx, rdx

MdePkg/Library/BaseMemoryLibMmx/X64/SetMem.asm

@@ -23,24 +23,33 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; VOID *
+; EFIAPI
+; InternalMemSetMem (
+;   OUT     VOID                      *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT8                     Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemSetMem   PROC    USES    rdi
     mov     rax, r8
     mov     ah, al
-    DB      48h, 0fh, 6eh, 0c0h        ;  movq    mm0, rax
+    DB      48h, 0fh, 6eh, 0c0h         ; movd mm0, rax
     mov     r8, rcx
-    mov     rdi, r8
+    mov     rdi, r8                     ; rdi <- Buffer
     mov     rcx, rdx
+    and     edx, 7
     shr     rcx, 3
     jz      @SetBytes
-    DB      0fh, 70h, 0C0h, 00h        ;  pshufw  mm0, mm0, 0h
+    DB      0fh, 70h, 0C0h, 00h         ; pshufw mm0, mm0, 0h
 @@:
-    DB      48h, 0fh, 0e7h, 07h        ;  movntq  [rdi], mm0
+    DB      0fh, 0e7h, 07h              ; movntq [rdi], mm0
     add     rdi, 8
     loop    @B
     mfence
 @SetBytes:
-    and     rdx, 7
+    mov     ecx, edx
-    mov     rcx, rdx
     rep     stosb
     mov     rax, r8
     ret

MdePkg/Library/BaseMemoryLibMmx/X64/SetMem16.asm

@@ -23,23 +23,32 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; VOID *
+; EFIAPI
+; InternalMemSetMem16 (
+;   OUT     VOID                      *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT16                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemSetMem16 PROC    USES    rdi
     mov     rax, r8
-    DB      48h, 0fh, 6eh, 0c0h        ;  movq    mm0, rax
+    DB      48h, 0fh, 6eh, 0c0h         ; movd mm0, rax
     mov     r8, rcx
     mov     rdi, r8
     mov     rcx, rdx
-    and     rdx, 3
+    and     edx, 3
     shr     rcx, 2
     jz      @SetWords
-    DB      0fh, 70h, 0C0h, 00h        ;  pshufw  mm0, mm0, 0h
+    DB      0fh, 70h, 0C0h, 00h         ; pshufw mm0, mm0, 0h
 @@:
-    DB      48h, 0fh, 0e7h, 07h        ;  movntq  [rdi], mm0
+    DB      0fh, 0e7h, 07h              ; movntq [rdi], mm0
     add     rdi, 8
     loop    @B
     mfence
 @SetWords:
-    mov     rcx, rdx
+    mov     ecx, edx
     rep     stosw
     mov     rax, r8
     ret

MdePkg/Library/BaseMemoryLibMmx/X64/SetMem32.asm

@@ -23,23 +23,29 @@
 
     .code
 
-InternalMemSetMem32 PROC    USES    rdi
+;------------------------------------------------------------------------------
-    DB      49h, 0fh, 6eh, 0c0h        ;  movq    mm0, r8       ; mm0 <- Value
+;  VOID *
-    mov     rax, rcx                                            ; rax <- Buffer
+;  InternalMemSetMem32 (
-    mov     rdi, rax
+;    IN VOID   *Buffer,
-    mov     rcx, rdx
+;    IN UINTN  Count,
-    shr     rcx, 1                                      
+;    IN UINT32 Value
+;    )
+;------------------------------------------------------------------------------
+InternalMemSetMem32 PROC
+    DB      49h, 0fh, 6eh, 0c0h         ; movd mm0, r8 (Value)
+    mov     rax, rcx                    ; rax <- Buffer
+    xchg    rcx, rdx                    ; rcx <- Count  rdx <- Buffer
+    shr     rcx, 1                      ; rcx <- # of qwords to set
     jz      @SetDwords
-    DB      0fh, 70h, 0C0h, 44h        ;  pshufw  mm0, mm0, 44h
+    DB      0fh, 70h, 0C0h, 44h         ; pshufw mm0, mm0, 44h
 @@:
-    DB      48h, 0fh, 0e7h, 07h        ;  movntq  [rdi], mm0
+    DB      0fh, 0e7h, 02h              ; movntq [rdx], mm0
-    add     rdi, 8
+    lea     rdx, [rdx + 8]              ; use "lea" to avoid flag changes
     loop    @B
     mfence
 @SetDwords:
-    test    dl, 1
+    jnc     @F
-    jz      @F
+    DB      0fh, 7eh, 02h               ; movd [rdx], mm0
-    DB      0fh, 7eh, 07h              ;  movd    [rdi], mm0
 @@:
     ret
 InternalMemSetMem32 ENDP

MdePkg/Library/BaseMemoryLibMmx/X64/SetMem64.asm

@@ -23,14 +23,21 @@
 
     .code
 
-InternalMemSetMem64 PROC    USES    rdi
+;------------------------------------------------------------------------------
-    DB      49h, 0fh, 6eh, 0c0h;  movq    mm0, r8       ; mm0 <- Value
+;  VOID *
-    mov     rax, rcx                                    ; rax <- Buffer
+;  InternalMemSetMem64 (
-    xchg    rcx, rdx                                    ; rcx <- Count
+;    IN VOID   *Buffer,
-    mov     rdi, rax
+;    IN UINTN  Count,
+;    IN UINT64 Value
+;    )
+;------------------------------------------------------------------------------
+InternalMemSetMem64 PROC
+    DB      49h, 0fh, 6eh, 0c0h         ; movd mm0, r8 (Value)
+    mov     rax, rcx                    ; rax <- Buffer
+    xchg    rcx, rdx                    ; rcx <- Count
 @@:
-    DB      48h, 0fh, 0e7h, 07h;  movntq  [rdi], mm0
+    DB      0fh, 0e7h, 02h              ; movntq  [rdx], mm0
-    add     rdi, 8
+    add     rdx, 8
     loop    @B
     mfence
     ret

MdePkg/Library/BaseMemoryLibMmx/X64/ZeroMem.asm

@@ -0,0 +1,54 @@
+;------------------------------------------------------------------------------
+;
+; 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:
+;
+;------------------------------------------------------------------------------
+
+    .code
+
+;------------------------------------------------------------------------------
+;  VOID *
+;  InternalMemZeroMem (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count
+;    );
+;------------------------------------------------------------------------------
+InternalMemZeroMem  PROC    USES    rdi
+    mov     rdi, rcx
+    mov     rcx, rdx
+    mov     r8, rdi
+    and     edx, 7
+    shr     rcx, 3
+    jz      @ZeroBytes
+    DB      0fh, 0efh, 0c0h             ; pxor mm0, mm0
+@@:
+    DB      0fh, 0e7h, 7                ; movntq [rdi], mm0
+    add     rdi, 8
+    loop    @B
+    DB      0fh, 0aeh, 0f0h             ; mfence
+@ZeroBytes:
+    xor     eax, eax
+    mov     ecx, edx
+    rep     stosb
+    mov     rax, r8
+    ret
+InternalMemZeroMem  ENDP
+
+    END

MdePkg/Library/BaseMemoryLibMmx/ZeroMemWrapper.c

@@ -47,5 +47,5 @@ ZeroMem (
 {
   ASSERT (!(Buffer == NULL && Length > 0));
   ASSERT (Length <= (MAX_ADDRESS - (UINTN)Buffer + 1));
-  return InternalMemSetMem (Buffer, Length, 0);
+  return InternalMemZeroMem (Buffer, Length);
 }

MdePkg/Library/BaseMemoryLibRepStr/Ia32/CompareMem.S

@@ -27,20 +27,27 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    .code: 
-
 .global _InternalMemCompareMem
+
+#------------------------------------------------------------------------------
+# INTN
+# EFIAPI
+# InternalMemCompareMem (
+#   IN      CONST VOID                *DestinationBuffer,
+#   IN      CONST VOID                *SourceBuffer,
+#   IN      UINTN                     Length
+#   );
+#------------------------------------------------------------------------------
 _InternalMemCompareMem:
     push    %esi
     push    %edi
-    movl    12(%esp),%esi
+    movl    12(%esp), %esi
-    movl    16(%esp),%edi
+    movl    16(%esp), %edi
-    movl    20(%esp),%ecx
+    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

MdePkg/Library/BaseMemoryLibRepStr/Ia32/CompareMem.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; INTN
+; EFIAPI
+; InternalMemCompareMem (
+;   IN      CONST VOID                *DestinationBuffer,
+;   IN      CONST VOID                *SourceBuffer,
+;   IN      UINTN                     Length
+;   );
+;------------------------------------------------------------------------------
 InternalMemCompareMem   PROC    USES    esi edi
     mov     esi, [esp + 12]
     mov     edi, [esp + 16]

MdePkg/Library/BaseMemoryLibRepStr/Ia32/CopyMem.S

@@ -21,38 +21,45 @@
 #
 #------------------------------------------------------------------------------
 
-    .386: 
-    .code: 
-
 .global _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    16(%esp), %esi              # esi <- Source
-    movl    12(%esp),%edi               # edi <- Destination
+    movl    12(%esp), %edi              # edi <- Destination
-    movl    20(%esp),%edx               # edx <- Count
+    movl    20(%esp), %edx              # edx <- Count
-    leal    -1(%edi,%edx),%eax          # eax <- End of Destination
+    leal    -1(%esi, %edx), %eax        # eax <- End of Source
-    cmpl    %edi,%esi
+    cmpl    %edi, %esi
     jae     L0
-    cmpl    %esi,%eax
+    cmpl    %edi, %eax
     jae     @CopyBackward               # Copy backward if overlapped
 L0:
-    movl    %edx,%ecx
+    movl    %edx, %ecx
-    andl    $3,%edx
+    andl    $3, %edx
-    shrl    $2,%ecx
+    shrl    $2, %ecx
     rep
     movsl                               # Copy as many Dwords as possible
     jmp     @CopyBytes
 @CopyBackward:
-    movl    %eax,%edi                   # edi <- End of Destination
+    movl    %eax, %esi                  # esi <- End of Source
-    leal    -1(%esi,%edx),%esi          # esi <- End of Source
+    leal    -1(%edi, %edx), %edi        # edi <- End of Destination
     std
 @CopyBytes:
-    movl    %edx,%ecx
+    movl    %edx, %ecx
     rep
     movsb                               # Copy bytes backward
     cld
-    movl    12(%esp),%eax               # eax <- Destination as return value
+    movl    12(%esp), %eax              # eax <- Destination as return value
     pop     %edi
     pop     %esi
     ret

MdePkg/Library/BaseMemoryLibRepStr/Ia32/CopyMem.asm

@@ -25,14 +25,22 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  InternalMemCopyMem (
+;    IN VOID   *Destination,
+;    IN VOID   *Source,
+;    IN UINTN  Count
+;    )
+;------------------------------------------------------------------------------
 InternalMemCopyMem  PROC    USES    esi edi
     mov     esi, [esp + 16]             ; esi <- Source
     mov     edi, [esp + 12]             ; edi <- Destination
     mov     edx, [esp + 20]             ; edx <- Count
-    lea     eax, [edi + edx - 1]        ; eax <- End of Destination
+    lea     eax, [esi + edx - 1]        ; eax <- End of Source
     cmp     esi, edi
     jae     @F
-    cmp     eax, esi
+    cmp     eax, edi
     jae     @CopyBackward               ; Copy backward if overlapped
 @@:
     mov     ecx, edx
@@ -41,8 +49,8 @@ InternalMemCopyMem  PROC    USES    esi edi
     rep     movsd                       ; Copy as many Dwords as possible
     jmp     @CopyBytes
 @CopyBackward:
-    mov     edi, eax                    ; edi <- End of Destination
+    mov     esi, eax                    ; esi <- End of Source
-    lea     esi, [esi + edx - 1]        ; esi <- End of Source
+    lea     edi, [edi + edx - 1]        ; edi <- End of Destination
     std
 @CopyBytes:
     mov     ecx, edx

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ScanMem16.S

@@ -27,17 +27,24 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    .code: 
-
 .global _InternalMemScanMem16
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem16 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT16                    Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem16:
     push    %edi
-    movl    12(%esp),%ecx
+    movl    12(%esp), %ecx
-    movl    8(%esp),%edi
+    movl    8(%esp), %edi
-    movl    16(%esp),%eax
+    movl    16(%esp), %eax
     repne   scasw
-    leal    -2(%edi),%eax
+    leal    -2(%edi), %eax
     cmovnz  %ecx, %eax
     pop     %edi
     ret

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ScanMem16.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem16 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT16                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem16    PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ScanMem32.S

@@ -27,17 +27,24 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    .code: 
-
 .global _InternalMemScanMem32
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem32 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT32                    Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem32:
     push    %edi
-    movl    12(%esp),%ecx
+    movl    12(%esp), %ecx
-    movl    8(%esp),%edi
+    movl    8(%esp), %edi
-    movl    16(%esp),%eax
+    movl    16(%esp), %eax
     repne   scasl
-    leal    -4(%edi),%eax
+    leal    -4(%edi), %eax
     cmovnz  %ecx, %eax
     pop     %edi
     ret

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ScanMem32.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem32 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT32                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem32    PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ScanMem64.S

@@ -27,27 +27,33 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    .code: 
-
 .global _InternalMemScanMem64
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem64 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT64                    Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem64:
     push    %edi
-    movl    12(%esp),%ecx
+    movl    12(%esp), %ecx
-    movl    16(%esp),%eax
+    movl    16(%esp), %eax
-    movl    20(%esp),%edx
+    movl    20(%esp), %edx
-    movl    8(%esp),%edi
+    movl    8(%esp), %edi
 L0:
-    cmpl    (%edi),%eax
+    cmpl    (%edi), %eax
-    leal    8(%edi),%edi
+    leal    8(%edi), %edi
     loopne  L0
     jne     L1
-    cmpl    -4(%edi),%edx
+    cmpl    -4(%edi), %edx
     jecxz   L1
     jne     L0
 L1:
-    leal    -8(%edi),%eax
+    leal    -8(%edi), %eax
     cmovne  %ecx, %eax
     pop     %edi
     ret
-

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ScanMem64.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem64 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT64                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem64    PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     eax, [esp + 16]

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ScanMem8.S

@@ -27,17 +27,24 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    .code: 
-
 .global _InternalMemScanMem8
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem8 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT8                     Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem8:
     push    %edi
-    movl    12(%esp),%ecx
+    movl    12(%esp), %ecx
-    movl    8(%esp),%edi
+    movl    8(%esp), %edi
-    movb    16(%esp),%al
+    movb    16(%esp), %al
     repne   scasb
-    leal    -1(%edi),%eax
+    leal    -1(%edi), %eax
     cmovnz  %ecx, %eax
     pop     %edi
     ret

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ScanMem8.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem8 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT8                     Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem8 PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibRepStr/Ia32/SetMem.S

@@ -25,6 +25,15 @@
     .code:
 
 .global _InternalMemSetMem
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  InternalMemSetMem (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count,
+#    IN UINT8  Value
+#    )
+#------------------------------------------------------------------------------
 _InternalMemSetMem:
     push    %edi
     movl    16(%esp),%eax

MdePkg/Library/BaseMemoryLibRepStr/Ia32/SetMem.asm

@@ -25,6 +25,14 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  InternalMemSetMem (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count,
+;    IN UINT8  Value
+;    )
+;------------------------------------------------------------------------------
 InternalMemSetMem   PROC    USES    edi
     mov     eax, [esp + 16]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibRepStr/Ia32/SetMem16.S

@@ -21,17 +21,23 @@
 #
 #------------------------------------------------------------------------------
 
-    .386: 
-    .code: 
-
 .global _InternalMemSetMem16
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  InternalMemSetMem16 (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count,
+#    IN UINT16 Value
+#    )
+#------------------------------------------------------------------------------
 _InternalMemSetMem16:
     push    %edi
-    movl    16(%esp),%eax
+    movl    16(%esp), %eax
-    movl    8(%esp),%edi
+    movl    8(%esp), %edi
-    movl    12(%esp),%ecx
+    movl    12(%esp), %ecx
     rep
     stosw
-    movl    8(%esp),%eax
+    movl    8(%esp), %eax
     pop     %edi
     ret

MdePkg/Library/BaseMemoryLibRepStr/Ia32/SetMem16.asm

@@ -25,6 +25,14 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  InternalMemSetMem16 (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count,
+;    IN UINT16 Value
+;    )
+;------------------------------------------------------------------------------
 InternalMemSetMem16 PROC    USES    edi
     mov     eax, [esp + 16]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibRepStr/Ia32/SetMem32.S

@@ -21,10 +21,16 @@
 #
 #------------------------------------------------------------------------------
 
-    .386: 
-    .code: 
-
 .global _InternalMemSetMem32
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  InternalMemSetMem32 (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count,
+#    IN UINT32 Value
+#    )
+#------------------------------------------------------------------------------
 _InternalMemSetMem32:
     push    %edi
     movl    16(%esp),%eax

MdePkg/Library/BaseMemoryLibRepStr/Ia32/SetMem32.asm

@@ -25,6 +25,14 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  InternalMemSetMem32 (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count,
+;    IN UINT32 Value
+;    )
+;------------------------------------------------------------------------------
 InternalMemSetMem32 PROC    USES    edi
     mov     eax, [esp + 16]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibRepStr/Ia32/SetMem64.S

@@ -21,11 +21,16 @@
 #
 #------------------------------------------------------------------------------
 
-    .386: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemSetMem64
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  InternalMemSetMem64 (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count,
+#    IN UINT64 Value
+#    )
+#------------------------------------------------------------------------------
 _InternalMemSetMem64:
     push    %edi
     movl    12(%esp), %ecx
@@ -33,8 +38,8 @@ _InternalMemSetMem64:
     movl    20(%esp), %edx
     movl    8(%esp), %edi
 L0:
-    mov     %eax,-8(%edi,%ecx,8)
+    mov     %eax, -8(%edi, %ecx, 8)
-    mov     %edx,-4(%edi,%ecx,8)
+    mov     %edx, -4(%edi, %ecx, 8)
     loop    L0
     movl    %edi, %eax
     pop     %edi

MdePkg/Library/BaseMemoryLibRepStr/Ia32/SetMem64.asm

@@ -25,6 +25,14 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  InternalMemSetMem64 (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count,
+;    IN UINT64 Value
+;    )
+;------------------------------------------------------------------------------
 InternalMemSetMem64 PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     eax, [esp + 16]

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ZeroMem.S

@@ -21,10 +21,15 @@
 #
 #------------------------------------------------------------------------------
 
-    .386: 
-    .code: 
-
 .global _InternalMemZeroMem
+
+#------------------------------------------------------------------------------
+#  VOID *
+#  InternalMemZeroMem (
+#    IN VOID   *Buffer,
+#    IN UINTN  Count
+#    );
+#------------------------------------------------------------------------------
 _InternalMemZeroMem:
     push    %edi
     xorl    %eax,%eax

MdePkg/Library/BaseMemoryLibRepStr/Ia32/ZeroMem.asm

@@ -25,6 +25,13 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  InternalMemZeroMem (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count
+;    );
+;------------------------------------------------------------------------------
 InternalMemZeroMem  PROC    USES    edi
     xor     eax, eax
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibRepStr/ZeroMemWrapper.c

@@ -47,5 +47,5 @@ ZeroMem (
 {
   ASSERT (!(Buffer == NULL && Length > 0));
   ASSERT (Length <= (MAX_ADDRESS - (UINTN)Buffer + 1));
-  return InternalMemSetMem (Buffer, Length, 0);
+  return InternalMemZeroMem (Buffer, Length);
 }

MdePkg/Library/BaseMemoryLibRepStr/x64/CompareMem.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; INTN
+; EFIAPI
+; InternalMemCompareMem (
+;   IN      CONST VOID                *DestinationBuffer,
+;   IN      CONST VOID                *SourceBuffer,
+;   IN      UINTN                     Length
+;   );
+;------------------------------------------------------------------------------
 InternalMemCompareMem   PROC    USES    rsi rdi
     mov     rsi, rcx
     mov     rdi, rdx

MdePkg/Library/BaseMemoryLibRepStr/x64/CopyMem.asm

@@ -23,10 +23,19 @@
 
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  EFIAPI
+;  InternalMemCopyMem (
+;    IN VOID   *Destination,
+;    IN VOID   *Source,
+;    IN UINTN  Count
+;    )
+;------------------------------------------------------------------------------
 InternalMemCopyMem  PROC    USES    rsi rdi
     mov     rsi, rdx                    ; rsi <- Source
     mov     rdi, rcx                    ; rdi <- Destination
-    lea     r9, [rdi + r8 - 1]          ; r9 <- End of Destination
+    lea     r9, [rsi + r8 - 1]          ; r9 <- End of Source
     cmp     rsi, rdi
     mov     rax, rdi                    ; rax <- Destination as return value
     jae     @F
@@ -39,8 +48,8 @@ InternalMemCopyMem  PROC    USES    rsi rdi
     rep     movsq                       ; Copy as many Qwords as possible
     jmp     @CopyBytes
 @CopyBackward:
-    mov     rdi, r9                     ; rdi <- End of Destination
+    mov     rsi, r9                     ; rsi <- End of Source
-    lea     rsi, [rsi + r8 - 1]         ; esi <- End of Source
+    lea     rdi, [rdi + r8 - 1]         ; esi <- End of Destination
     std                                 ; set direction flag
 @CopyBytes:
     mov     rcx, r8

MdePkg/Library/BaseMemoryLibRepStr/x64/ScanMem16.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem16 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT16                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem16    PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibRepStr/x64/ScanMem32.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem32 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT32                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem32    PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibRepStr/x64/ScanMem64.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem64 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT64                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem64    PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibRepStr/x64/ScanMem8.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem8 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT8                     Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem8 PROC    USES    rdi
     mov     rdi, rcx
     mov     rcx, rdx

MdePkg/Library/BaseMemoryLibRepStr/x64/SetMem.asm

@@ -23,6 +23,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  EFIAPI
+;  InternalMemSetMem (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count,
+;    IN UINT8  Value
+;    )
+;------------------------------------------------------------------------------
 InternalMemSetMem   PROC    USES    rdi
     mov     rax, r8
     mov     rdi, rcx

MdePkg/Library/BaseMemoryLibRepStr/x64/SetMem16.asm

@@ -23,6 +23,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  EFIAPI
+;  InternalMemSetMem16 (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count,
+;    IN UINT16 Value
+;    )
+;------------------------------------------------------------------------------
 InternalMemSetMem16 PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibRepStr/x64/SetMem32.asm

@@ -23,6 +23,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  EFIAPI
+;  InternalMemSetMem32 (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count,
+;    IN UINT32 Value
+;    );
+;------------------------------------------------------------------------------
 InternalMemSetMem32 PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibRepStr/x64/SetMem64.asm

@@ -23,6 +23,14 @@
 
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  InternalMemSetMem64 (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count,
+;    IN UINT64 Value
+;    )
+;------------------------------------------------------------------------------
 InternalMemSetMem64 PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibRepStr/x64/ZeroMem.asm

@@ -23,6 +23,13 @@
 
     .code
 
+;------------------------------------------------------------------------------
+;  VOID *
+;  InternalMemZeroMem (
+;    IN VOID   *Buffer,
+;    IN UINTN  Count
+;    );
+;------------------------------------------------------------------------------
 InternalMemZeroMem  PROC    USES    rdi
     push    rcx
     xor     rax, rax

MdePkg/Library/BaseMemoryLibSse2/Ia32/CompareMem.S

@@ -27,11 +27,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemCompareMem
+
+#------------------------------------------------------------------------------
+# INTN
+# EFIAPI
+# InternalMemCompareMem (
+#   IN      CONST VOID                *DestinationBuffer,
+#   IN      CONST VOID                *SourceBuffer,
+#   IN      UINTN                     Length
+#   );
+#------------------------------------------------------------------------------
 _InternalMemCompareMem:
     push    %esi
     push    %edi

MdePkg/Library/BaseMemoryLibSse2/Ia32/CompareMem.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; INTN
+; EFIAPI
+; InternalMemCompareMem (
+;   IN      CONST VOID                *DestinationBuffer,
+;   IN      CONST VOID                *SourceBuffer,
+;   IN      UINTN                     Length
+;   );
+;------------------------------------------------------------------------------
 InternalMemCompareMem   PROC    USES    esi edi
     mov     esi, [esp + 12]
     mov     edi, [esp + 16]

MdePkg/Library/BaseMemoryLibSse2/Ia32/CopyMem.S

@@ -21,30 +21,27 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemCopyMem
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_CopyMem (
+#  EFIAPI
+#  InternalMemCopyMem (
 #    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
+    leal    -1(%esi,%edx,), %eax        # eax <- End of Source
     cmpl    %edi, %esi
     jae     L0
-    cmpl    %esi, %eax                  # Overlapped?
+    cmpl    %edi, %eax                  # Overlapped?
     jae     @CopyBackward               # Copy backward if overlapped
 L0:
     xorl    %ecx, %ecx
@@ -74,8 +71,8 @@ L2:
     addl    $16, %esp                   # stack cleanup
     jmp     @CopyBytes
 @CopyBackward:
-    movl    %eax, %edi                  # edi <- Last byte in Destination
+    movl    %eax, %esi                  # esi <- Last byte in Source
-    leal    -1(%esi,%edx,), %esi        # esi <- Last byte in Source
+    leal    -1(%edi,%edx,), %edi        # edi <- Last byte in Destination
     std
 @CopyBytes:
     movl    %edx, %ecx

MdePkg/Library/BaseMemoryLibSse2/Ia32/CopyMem.asm

@@ -28,20 +28,20 @@
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_CopyMem (
+;  InternalMemCopyMem (
 ;    IN VOID   *Destination,
 ;    IN VOID   *Source,
 ;    IN UINTN  Count
-;    )
+;    );
 ;------------------------------------------------------------------------------
 InternalMemCopyMem  PROC    USES    esi edi
     mov     esi, [esp + 16]             ; esi <- Source
     mov     edi, [esp + 12]             ; edi <- Destination
     mov     edx, [esp + 20]             ; edx <- Count
-    lea     eax, [edi + edx - 1]        ; eax <- End of Destination
+    lea     eax, [esi + edx - 1]        ; eax <- End of Source
     cmp     esi, edi
     jae     @F
-    cmp     eax, esi                    ; Overlapped?
+    cmp     eax, edi                    ; Overlapped?
     jae     @CopyBackward               ; Copy backward if overlapped
 @@:
     xor     ecx, ecx
@@ -70,8 +70,8 @@ InternalMemCopyMem  PROC    USES    esi edi
     add     esp, 16                     ; stack cleanup
     jmp     @CopyBytes
 @CopyBackward:
-    mov     edi, eax                    ; edi <- Last byte in Destination
+    mov     esi, eax                    ; esi <- Last byte in Source
-    lea     esi, [esi + edx - 1]        ; esi <- Last byte in Source
+    lea     edi, [edi + edx - 1]        ; edi <- Last byte in Destination
     std
 @CopyBytes:
     mov     ecx, edx

MdePkg/Library/BaseMemoryLibSse2/Ia32/ScanMem16.S

@@ -27,11 +27,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemScanMem16
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem16 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT16                    Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem16:
     push    %edi
     movl    12(%esp), %ecx

MdePkg/Library/BaseMemoryLibSse2/Ia32/ScanMem16.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem16 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT16                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem16    PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibSse2/Ia32/ScanMem32.S

@@ -27,17 +27,23 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemScanMem32
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem32 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT32                    Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem32:
     push    %edi
     movl    12(%esp), %ecx
     movl    8(%esp), %edi
     movl    16(%esp), %eax
-    repnz   scasl
+    repne   scasl
     leal    -4(%edi), %eax
     cmovnz  %ecx, %eax
     pop     %edi

MdePkg/Library/BaseMemoryLibSse2/Ia32/ScanMem32.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem32 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT32                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem32    PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibSse2/Ia32/ScanMem64.S

@@ -27,11 +27,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemScanMem64
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem64 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT64                    Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem64:
     push    %edi
     movl    12(%esp), %ecx

MdePkg/Library/BaseMemoryLibSse2/Ia32/ScanMem64.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem64 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT64                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem64    PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     eax, [esp + 16]

MdePkg/Library/BaseMemoryLibSse2/Ia32/ScanMem8.S

@@ -27,11 +27,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
-    #.MODEL flat,C
-    .code: 
-
 .global _InternalMemScanMem8
+
+#------------------------------------------------------------------------------
+# CONST VOID *
+# EFIAPI
+# InternalMemScanMem8 (
+#   IN      CONST VOID                *Buffer,
+#   IN      UINTN                     Length,
+#   IN      UINT8                     Value
+#   );
+#------------------------------------------------------------------------------
 _InternalMemScanMem8:
     push    %edi
     movl    12(%esp), %ecx

MdePkg/Library/BaseMemoryLibSse2/Ia32/ScanMem8.asm

@@ -31,6 +31,15 @@
     .model  flat,C
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem8 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT8                     Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem8 PROC    USES    edi
     mov     ecx, [esp + 12]
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem.S

@@ -32,7 +32,7 @@
 #    IN VOID   *Buffer,
 #    IN UINTN  Count,
 #    IN UINT8  Value
-#    )
+#    );
 #------------------------------------------------------------------------------
 .global _InternalMemSetMem
 _InternalMemSetMem:

MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem.asm

@@ -28,11 +28,12 @@
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_SetMem (
+;  EFIAPI
+;  InternalMemSetMem (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count,
 ;    IN UINT8  Value
-;    )
+;    );
 ;------------------------------------------------------------------------------
 InternalMemSetMem   PROC    USES    edi
     mov     edx, [esp + 12]             ; edx <- Count

MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem16.S

@@ -21,20 +21,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemSetMem16
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_SetMem16 (
+#  EFIAPI
+#  InternalMemSetMem16 (
 #    IN VOID   *Buffer,
 #    IN UINTN  Count,
 #    IN UINT16 Value
 #    )
 #------------------------------------------------------------------------------
-.global _InternalMemSetMem16
 _InternalMemSetMem16:
     push    %edi
     movl    12(%esp), %edx

MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem16.asm

@@ -28,11 +28,12 @@
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_SetMem16 (
+;  EFIAPI
+;  InternalMemSetMem16 (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count,
 ;    IN UINT16 Value
-;    )
+;    );
 ;------------------------------------------------------------------------------
 InternalMemSetMem16 PROC    USES    edi
     mov     edx, [esp + 12]

MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem32.S

@@ -21,20 +21,17 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemSetMem32
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_SetMem32 (
+#  EFIAPI
+#  InternalMemSetMem32 (
 #    IN VOID   *Buffer,
 #    IN UINTN  Count,
 #    IN UINT32 Value
-#    )
+#    );
 #------------------------------------------------------------------------------
-.global _InternalMemSetMem32
 _InternalMemSetMem32:
     push    %edi
     movl    12(%esp), %edx

MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem32.asm

@@ -28,11 +28,12 @@
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_SetMem32 (
+;  EFIAPI
+;  InternalMemSetMem32 (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count,
 ;    IN UINT32 Value
-;    )
+;    );
 ;------------------------------------------------------------------------------
 InternalMemSetMem32 PROC    USES    edi
     mov     edx, [esp + 12]

MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem64.S

@@ -21,15 +21,17 @@
 #
 #------------------------------------------------------------------------------
 
+.globl _InternalMemSetMem64
+
 #------------------------------------------------------------------------------
 #  VOID *
+#  EFIAPI
 #  InternalMemSetMem64 (
 #    IN VOID   *Buffer,
 #    IN UINTN  Count,
 #    IN UINT64 Value
-#    )
+#    );
 #------------------------------------------------------------------------------
-.globl _InternalMemSetMem64
 _InternalMemSetMem64:
     movl    4(%esp), %eax
     movl    8(%esp), %ecx
@@ -54,6 +56,3 @@ L2:
     movq    %xmm0, (%edx)
 L3:
     ret
-
-
-

MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem64.asm

@@ -28,6 +28,7 @@
 
 ;------------------------------------------------------------------------------
 ;  VOID *
+;  EFIAPI
 ;  InternalMemSetMem64 (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count,
@@ -35,8 +36,8 @@
 ;    )
 ;------------------------------------------------------------------------------
 InternalMemSetMem64 PROC
-    mov     eax, [esp + 4]
+    mov     eax, [esp + 4]              ; eax <- Buffer
-    mov     ecx, [esp + 8]
+    mov     ecx, [esp + 8]              ; ecx <- Count
     test    al, 8
     mov     edx, eax
     movq    xmm0, [esp + 12]

MdePkg/Library/BaseMemoryLibSse2/Ia32/ZeroMem.S

@@ -21,19 +21,16 @@
 #
 #------------------------------------------------------------------------------
 
-    .686: 
+.global _InternalMemZeroMem
-    #.MODEL flat,C
-    .xmm: 
-    .code: 
 
 #------------------------------------------------------------------------------
 #  VOID *
-#  _mem_ZeroMem (
+#  EFIAPI
+#  InternalMemZeroMem (
 #    IN VOID   *Buffer,
 #    IN UINTN  Count
-#    )
+#    );
 #------------------------------------------------------------------------------
-.global _InternalMemZeroMem
 _InternalMemZeroMem:
     push    %edi
     movl    8(%esp), %edi

MdePkg/Library/BaseMemoryLibSse2/Ia32/ZeroMem.asm

@@ -28,10 +28,11 @@
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_ZeroMem (
+;  EFIAPI
+;  InternalMemZeroMem (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count
-;    )
+;    );
 ;------------------------------------------------------------------------------
 InternalMemZeroMem  PROC    USES    edi
     mov     edi, [esp + 8]

MdePkg/Library/BaseMemoryLibSse2/ZeroMemWrapper.c

@@ -47,5 +47,5 @@ ZeroMem (
 {
   ASSERT (!(Buffer == NULL && Length > 0));
   ASSERT (Length <= (MAX_ADDRESS - (UINTN)Buffer + 1));
-  return InternalMemSetMem (Buffer, Length, 0);
+  return InternalMemZeroMem (Buffer, Length);
 }

MdePkg/Library/BaseMemoryLibSse2/x64/CompareMem.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; INTN
+; EFIAPI
+; InternalMemCompareMem (
+;   IN      CONST VOID                *DestinationBuffer,
+;   IN      CONST VOID                *SourceBuffer,
+;   IN      UINTN                     Length
+;   );
+;------------------------------------------------------------------------------
 InternalMemCompareMem   PROC    USES    rsi rdi
     mov     rsi, rcx
     mov     rdi, rdx

MdePkg/Library/BaseMemoryLibSse2/x64/CopyMem.asm

@@ -25,20 +25,21 @@
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_CopyMem (
+;  EFIAPI
+;  InternalMemCopyMem (
 ;    IN VOID   *Destination,
 ;    IN VOID   *Source,
 ;    IN UINTN  Count
-;    )
+;    );
 ;------------------------------------------------------------------------------
 InternalMemCopyMem  PROC    USES    rsi rdi
     mov     rsi, rdx                    ; rsi <- Source
     mov     rdi, rcx                    ; rdi <- Destination
-    lea     r9, [rdi + r8 - 1]          ; r9 <- Last byte of Destination
+    lea     r9, [rsi + r8 - 1]          ; r9 <- Last byte of Source
     cmp     rsi, rdi
     mov     rax, rdi                    ; rax <- Destination as return value
     jae     @F                          ; Copy forward if Source > Destination
-    cmp     r9, rsi                     ; Overlapped?
+    cmp     r9, rdi                     ; Overlapped?
     jae     @CopyBackward               ; Copy backward if overlapped
 @@:
     xor     rcx, rcx
@@ -65,8 +66,8 @@ InternalMemCopyMem  PROC    USES    rsi rdi
     movdqa  xmm0, [rsp + 18h]           ; restore xmm0
     jmp     @CopyBytes                  ; copy remaining bytes
 @CopyBackward:
-    mov     rdi, r9                     ; rdi <- Last byte of Destination
+    mov     rsi, r9                     ; rsi <- Last byte of Source
-    lea     rsi, [rsi + r8 - 1]         ; rsi <- Last byte of Source
+    lea     rdi, [rdi + r8 - 1]         ; rdi <- Last byte of Destination
     std
 @CopyBytes:
     mov     rcx, r8

MdePkg/Library/BaseMemoryLibSse2/x64/ScanMem16.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem16 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT16                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem16    PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibSse2/x64/ScanMem32.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem32 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT32                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem32    PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibSse2/x64/ScanMem64.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem64 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT64                    Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem64    PROC    USES    rdi
     mov     rdi, rcx
     mov     rax, r8

MdePkg/Library/BaseMemoryLibSse2/x64/ScanMem8.asm

@@ -29,6 +29,15 @@
 
     .code
 
+;------------------------------------------------------------------------------
+; CONST VOID *
+; EFIAPI
+; InternalMemScanMem8 (
+;   IN      CONST VOID                *Buffer,
+;   IN      UINTN                     Length,
+;   IN      UINT8                     Value
+;   );
+;------------------------------------------------------------------------------
 InternalMemScanMem8 PROC    USES    rdi
     mov     rdi, rcx
     mov     rcx, rdx

MdePkg/Library/BaseMemoryLibSse2/x64/SetMem.asm

@@ -25,7 +25,7 @@
 
 ;------------------------------------------------------------------------------
 ;  VOID *
-;  _mem_SetMem (
+;  InternalMemSetMem (
 ;    IN VOID   *Buffer,
 ;    IN UINTN  Count,
 ;    IN UINT8  Value