Update libvpx snapshot to v0.9.7-p1 (Cayuga).

source/libvpx/vpx_ports/x86_abi_support.asm

 ;
 ;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 ;
 ;  Use of this source code is governed by a BSD-style license
 ;  that can be found in the LICENSE file in the root of the source
 ;  tree. An additional intellectual property rights grant can be found
 ;  in the file PATENTS.  All contributing project authors may
 ;  be found in the AUTHORS file in the root of the source tree.
 ;
 
@@ skipping 129 matching lines @@
 ;
 %idefine PTR
 %idefine XMMWORD
 %idefine MMWORD
 
 ; PIC macros
 ;
 %if ABI_IS_32BIT
   %if CONFIG_PIC=1
   %ifidn __OUTPUT_FORMAT__,elf32
+    %define GET_GOT_SAVE_ARG 1
     %define WRT_PLT wrt ..plt
     %macro GET_GOT 1
       extern _GLOBAL_OFFSET_TABLE_
       push %1
       call %%get_got
       %%sub_offset:
       jmp %%exitGG
       %%get_got:
       mov %1, [esp]
       add %1, _GLOBAL_OFFSET_TABLE_ + $$ - %%sub_offset wrt ..gotpc
       ret
       %%exitGG:
       %undef GLOBAL
       %define GLOBAL(x) x + %1 wrt ..gotoff
       %undef RESTORE_GOT
       %define RESTORE_GOT pop %1
     %endmacro
   %elifidn __OUTPUT_FORMAT__,macho32
+    %define GET_GOT_SAVE_ARG 1
     %macro GET_GOT 1
       push %1
       call %%get_got
       %%get_got:
       pop  %1
       %undef GLOBAL
       %define GLOBAL(x) x + %1 - %%get_got
       %undef RESTORE_GOT
       %define RESTORE_GOT pop %1
     %endmacro
@@ skipping 70 matching lines @@
         mov rax,[rbp+off]
         push rax
         %assign off off+8
       %endrep
     %endif
   %endm
 %endif
   %define UNSHADOW_ARGS mov rsp, rbp
 %endif
 
-; must keep XMM6:XMM15 (libvpx uses XMM6 and XMM7) on Win64 ABI
-; rsp register has to be aligned
+; Win64 ABI requires that XMM6:XMM15 are callee saved
+; SAVE_XMM n, [u]
+; store registers 6-n on the stack
+; if u is specified, use unaligned movs.
+; Win64 ABI requires 16 byte stack alignment, but then pushes an 8 byte return
+; value. Typically we follow this up with 'push rbp' - re-aligning the stack -
+; but in some cases this is not done and unaligned movs must be used.
 %ifidn __OUTPUT_FORMAT__,x64
-%macro SAVE_XMM 0
-  sub rsp, 32
-  movdqa XMMWORD PTR [rsp], xmm6
-  movdqa XMMWORD PTR [rsp+16], xmm7
+%macro SAVE_XMM 1-2 a
+  %if %1 < 6
+    %error Only xmm registers 6-15 must be preserved
+  %else
+    %assign last_xmm %1
+    %define movxmm movdq %+ %2
+    %assign xmm_stack_space ((last_xmm - 5) * 16)
+    sub rsp, xmm_stack_space
+    %assign i 6
+    %rep (last_xmm - 5)
+      movxmm [rsp + ((i - 6) * 16)], xmm %+ i
+      %assign i i+1
+    %endrep
+  %endif
 %endmacro
 %macro RESTORE_XMM 0
-  movdqa xmm6, XMMWORD PTR [rsp]
-  movdqa xmm7, XMMWORD PTR [rsp+16]
-  add rsp, 32
+  %ifndef last_xmm
+    %error RESTORE_XMM must be paired with SAVE_XMM n
+  %else
+    %assign i last_xmm
+    %rep (last_xmm - 5)
+      movxmm xmm %+ i, [rsp +((i - 6) * 16)]
+      %assign i i-1
+    %endrep
+    add rsp, xmm_stack_space
+    ; there are a couple functions which return from multiple places.
+    ; otherwise, we could uncomment these:
+    ; %undef last_xmm
+    ; %undef xmm_stack_space
+    ; %undef movxmm
+  %endif
 %endmacro
 %else
-%macro SAVE_XMM 0
+%macro SAVE_XMM 1-2
 %endmacro
 %macro RESTORE_XMM 0
 %endmacro
 %endif
 
 ; Name of the rodata section
 ;
 ; .rodata seems to be an elf-ism, as it doesn't work on OSX.
 ;
 %ifidn __OUTPUT_FORMAT__,macho64
 %define SECTION_RODATA section .text
 %elifidn __OUTPUT_FORMAT__,macho32
 %macro SECTION_RODATA 0
 section .text
 %endmacro
 %else
 %define SECTION_RODATA section .rodata
 %endif
 
 
 ; Tell GNU ld that we don't require an executable stack.
 %ifidn __OUTPUT_FORMAT__,elf32
 section .note.GNU-stack noalloc noexec nowrite progbits
 section .text
 %elifidn __OUTPUT_FORMAT__,elf64
 section .note.GNU-stack noalloc noexec nowrite progbits
 section .text
 %endif