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

11 files changed, 179 insertions, 49 deletions

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/CompareMem.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/CompareMem.asm
index b8e289bb27..554bc07c64 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/CompareMem.asm
+++ b/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

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/CopyMem.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/CopyMem.asm
index 801cf92750..b57192edbe 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/CopyMem.asm
+++ b/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      48h, 0fh, 0e7h, 07h         ; movntq  [rdi], mm0

+    DB      0fh, 6fh, 06h               ; movd mm0, [rsi]

+    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

-    lea     rsi, [rsi + r8 - 1]         ; rsi <- End of Source

+    mov     rsi, r9                     ; rsi <- End of Source

+    lea     rdi, [rdi + r8 - 1]         ; rdi <- End of Destination

     std                                 ; set direction flag

 @CopyBytes:

     mov     rcx, r8

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem16.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem16.asm
index 6af88fae5b..a2538c2cca 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem16.asm
+++ b/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

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem32.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem32.asm
index f9c9feb00e..e0523d8547 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem32.asm
+++ b/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

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem64.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem64.asm
index 2f286c9e2f..0e7e9c782f 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem64.asm
+++ b/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

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem8.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem8.asm
index 4027ece768..879273f960 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/ScanMem8.asm
+++ b/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

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem.asm
index 9688c3d65b..923beef6c1 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem.asm
+++ b/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      48h, 0fh, 0e7h, 07h        ;  movntq  [rdi], mm0

+    DB      0fh, 70h, 0C0h, 00h         ; pshufw mm0, mm0, 0h

+@@:

+    DB      0fh, 0e7h, 07h              ; movntq [rdi], mm0

     add     rdi, 8

     loop    @B

     mfence

 @SetBytes:

-    and     rdx, 7

-    mov     rcx, rdx

+    mov     ecx, edx

     rep     stosb

     mov     rax, r8

     ret

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem16.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem16.asm
index a95020876c..01b72c55d5 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem16.asm
+++ b/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      48h, 0fh, 0e7h, 07h        ;  movntq  [rdi], mm0

+    DB      0fh, 70h, 0C0h, 00h         ; pshufw mm0, mm0, 0h

+@@:

+    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

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem32.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem32.asm
index d835543767..89cf175608 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem32.asm
+++ b/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

-    mov     rax, rcx                                            ; rax <- Buffer

-    mov     rdi, rax

-    mov     rcx, rdx

-    shr     rcx, 1                                      

+;------------------------------------------------------------------------------

+;  VOID *

+;  InternalMemSetMem32 (

+;    IN VOID   *Buffer,

+;    IN UINTN  Count,

+;    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

-    add     rdi, 8

+    DB      0fh, 0e7h, 02h              ; movntq [rdx], mm0

+    lea     rdx, [rdx + 8]              ; use "lea" to avoid flag changes

     loop    @B

     mfence

 @SetDwords:

-    test    dl, 1

-    jz      @F

-    DB      0fh, 7eh, 07h              ;  movd    [rdi], mm0

+    jnc     @F

+    DB      0fh, 7eh, 02h               ; movd [rdx], mm0

 @@:

     ret

 InternalMemSetMem32 ENDP

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem64.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem64.asm
index 5780647445..454d6234bf 100644
--- a/MdePkg/Library/BaseMemoryLibMmx/X64/SetMem64.asm
+++ b/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

-    mov     rax, rcx                                    ; rax <- Buffer

-    xchg    rcx, rdx                                    ; rcx <- Count

-    mov     rdi, rax

+;------------------------------------------------------------------------------

+;  VOID *

+;  InternalMemSetMem64 (

+;    IN VOID   *Buffer,

+;    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

-    add     rdi, 8

+    DB      0fh, 0e7h, 02h              ; movntq  [rdx], mm0

+    add     rdx, 8

     loop    @B

     mfence

     ret

diff --git a/MdePkg/Library/BaseMemoryLibMmx/X64/ZeroMem.asm b/MdePkg/Library/BaseMemoryLibMmx/X64/ZeroMem.asm
new file mode 100644
index 0000000000..e9035a3c14
--- /dev/null
+++ b/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