Update libjpeg-turbo to r856.

simd/jdmrgss2.asm

 ;
 ; jdmrgss2.asm - merged upsampling/color conversion (SSE2)
 ;
-; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB
 ;
 ; Based on
 ; x86 SIMD extension for IJG JPEG library
 ; Copyright (C) 1999-2006, MIYASAKA Masaru.
 ; For conditions of distribution and use, see copyright notice in jsimdext.inc
 ;
 ; This file should be assembled with NASM (Netwide Assembler),
 ; can *not* be assembled with Microsoft's MASM or any compatible
 ; assembler (including Borland's Turbo Assembler).
 ; NASM is available from http://nasm.sourceforge.net/ or
@@ skipping 242 matching lines @@
 
         cmp     ecx, byte SIZEOF_XMMWORD
         jb      short .column_st32
 
         test    edi, SIZEOF_XMMWORD-1
         jnz     short .out1
         ; --(aligned)-------------------
         movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
         movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
         movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
-        add     edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD  ; outptr
         jmp     short .out0
 .out1:  ; --(unaligned)-----------------
-        pcmpeqb    xmmH,xmmH                    ; xmmH=(all 1's)
-        maskmovdqu xmmA,xmmH                    ; movntdqu XMMWORD [edi], xmmA
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
-        maskmovdqu xmmD,xmmH                    ; movntdqu XMMWORD [edi], xmmD
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
-        maskmovdqu xmmF,xmmH                    ; movntdqu XMMWORD [edi], xmmF
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
+        movdqu  XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+        movdqu  XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+        movdqu  XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
 .out0:
+        add     edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD  ; outptr
         sub     ecx, byte SIZEOF_XMMWORD
         jz      near .endcolumn
 
         add     esi, byte SIZEOF_XMMWORD        ; inptr0
         dec     al                      ; Yctr
         jnz     near .Yloop_2nd
 
         add     ebx, byte SIZEOF_XMMWORD        ; inptr1
         add     edx, byte SIZEOF_XMMWORD        ; inptr2
         jmp     near .columnloop
         alignx  16,7
 
 .column_st32:
-        pcmpeqb xmmH,xmmH                       ; xmmH=(all 1's)
         lea     ecx, [ecx+ecx*2]                ; imul ecx, RGB_PIXELSIZE
         cmp     ecx, byte 2*SIZEOF_XMMWORD
         jb      short .column_st16
-        maskmovdqu xmmA,xmmH                    ; movntdqu XMMWORD [edi], xmmA
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
-        maskmovdqu xmmD,xmmH                    ; movntdqu XMMWORD [edi], xmmD
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
+        movdqu  XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+        movdqu  XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+        add     edi, byte 2*SIZEOF_XMMWORD      ; outptr
         movdqa  xmmA,xmmF
         sub     ecx, byte 2*SIZEOF_XMMWORD
         jmp     short .column_st15
 .column_st16:
         cmp     ecx, byte SIZEOF_XMMWORD
         jb      short .column_st15
-        maskmovdqu xmmA,xmmH                    ; movntdqu XMMWORD [edi], xmmA
+        movdqu  XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
         add     edi, byte SIZEOF_XMMWORD        ; outptr
         movdqa  xmmA,xmmD
         sub     ecx, byte SIZEOF_XMMWORD
 .column_st15:
-%ifdef STRICT_MEMORY_ACCESS
         ; Store the lower 8 bytes of xmmA to the output when it has enough
         ; space.
         cmp     ecx, byte SIZEOF_MMWORD
         jb      short .column_st7
         movq    MMWORD [edi], xmmA
         add     edi, byte SIZEOF_MMWORD
         sub     ecx, byte SIZEOF_MMWORD
         psrldq  xmmA, SIZEOF_MMWORD
 .column_st7:
         ; Store the lower 4 bytes of xmmA to the output when it has enough
@@ skipping 13 matching lines @@
         mov     WORD [edi], ax
         add     edi, byte SIZEOF_WORD
         sub     ecx, byte SIZEOF_WORD
         shr     eax, 16
 .column_st1:
         ; Store the lower 1 byte of eax to the output when it has enough
         ; space.
         test    ecx, ecx
         jz      short .endcolumn
         mov     BYTE [edi], al
-%else
-        mov     eax,ecx
-        xor     ecx, byte 0x0F
-        shl     ecx, 2
-        movd    xmmB,ecx
-        psrlq   xmmH,4
-        pcmpeqb xmmE,xmmE
-        psrlq   xmmH,xmmB
-        psrlq   xmmE,xmmB
-        punpcklbw xmmE,xmmH
-        ; ----------------
-        mov     ecx,edi
-        and     ecx, byte SIZEOF_XMMWORD-1
-        jz      short .adj0
-        add     eax,ecx
-        cmp     eax, byte SIZEOF_XMMWORD
-        ja      short .adj0
-        and     edi, byte (-SIZEOF_XMMWORD)     ; align to 16-byte boundary
-        shl     ecx, 3                  ; pslldq xmmA,ecx & pslldq xmmE,ecx
-        movdqa  xmmG,xmmA
-        movdqa  xmmC,xmmE
-        pslldq  xmmA, SIZEOF_XMMWORD/2
-        pslldq  xmmE, SIZEOF_XMMWORD/2
-        movd    xmmD,ecx
-        sub     ecx, byte (SIZEOF_XMMWORD/2)*BYTE_BIT
-        jb      short .adj1
-        movd    xmmF,ecx
-        psllq   xmmA,xmmF
-        psllq   xmmE,xmmF
-        jmp     short .adj0
-.adj1:  neg     ecx
-        movd    xmmF,ecx
-        psrlq   xmmA,xmmF
-        psrlq   xmmE,xmmF
-        psllq   xmmG,xmmD
-        psllq   xmmC,xmmD
-        por     xmmA,xmmG
-        por     xmmE,xmmC
-.adj0:  ; ----------------
-        maskmovdqu xmmA,xmmE                    ; movntdqu XMMWORD [edi], xmmA
-%endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %else ; RGB_PIXELSIZE == 4 ; -----------
 
 %ifdef RGBX_FILLER_0XFF
         pcmpeqb   xmm6,xmm6             ; xmm6=XE=X(02468ACE********)
         pcmpeqb   xmm7,xmm7             ; xmm7=XO=X(13579BDF********)
 %else
         pxor      xmm6,xmm6             ; xmm6=XE=X(02468ACE********)
         pxor      xmm7,xmm7             ; xmm7=XO=X(13579BDF********)
 %endif
@@ skipping 24 matching lines @@
         cmp     ecx, byte SIZEOF_XMMWORD
         jb      short .column_st32
 
         test    edi, SIZEOF_XMMWORD-1
         jnz     short .out1
         ; --(aligned)-------------------
         movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
         movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
         movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
         movntdq XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
-        add     edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD  ; outptr
         jmp     short .out0
 .out1:  ; --(unaligned)-----------------
-        pcmpeqb    xmmE,xmmE                    ; xmmE=(all 1's)
-        maskmovdqu xmmA,xmmE                    ; movntdqu XMMWORD [edi], xmmA
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
-        maskmovdqu xmmD,xmmE                    ; movntdqu XMMWORD [edi], xmmD
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
-        maskmovdqu xmmC,xmmE                    ; movntdqu XMMWORD [edi], xmmC
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
-        maskmovdqu xmmH,xmmE                    ; movntdqu XMMWORD [edi], xmmH
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
+        movdqu  XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+        movdqu  XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+        movdqu  XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
+        movdqu  XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
 .out0:
+        add     edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD  ; outptr
         sub     ecx, byte SIZEOF_XMMWORD
         jz      near .endcolumn
 
         add     esi, byte SIZEOF_XMMWORD        ; inptr0
         dec     al                      ; Yctr
         jnz     near .Yloop_2nd
 
         add     ebx, byte SIZEOF_XMMWORD        ; inptr1
         add     edx, byte SIZEOF_XMMWORD        ; inptr2
         jmp     near .columnloop
         alignx  16,7
 
 .column_st32:
-        pcmpeqb xmmE,xmmE                       ; xmmE=(all 1's)
         cmp     ecx, byte SIZEOF_XMMWORD/2
         jb      short .column_st16
-        maskmovdqu xmmA,xmmE                    ; movntdqu XMMWORD [edi], xmmA
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
-        maskmovdqu xmmD,xmmE                    ; movntdqu XMMWORD [edi], xmmD
-        add     edi, byte SIZEOF_XMMWORD        ; outptr
+        movdqu  XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+        movdqu  XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+        add     edi, byte 2*SIZEOF_XMMWORD      ; outptr
         movdqa  xmmA,xmmC
         movdqa  xmmD,xmmH
         sub     ecx, byte SIZEOF_XMMWORD/2
 .column_st16:
         cmp     ecx, byte SIZEOF_XMMWORD/4
         jb      short .column_st15
-        maskmovdqu xmmA,xmmE                    ; movntdqu XMMWORD [edi], xmmA
+        movdqu  XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
         add     edi, byte SIZEOF_XMMWORD        ; outptr
         movdqa  xmmA,xmmD
         sub     ecx, byte SIZEOF_XMMWORD/4
 .column_st15:
-%ifdef STRICT_MEMORY_ACCESS
         ; Store two pixels (8 bytes) of xmmA to the output when it has enough
         ; space.
         cmp     ecx, byte SIZEOF_XMMWORD/8
         jb      short .column_st7
         movq    MMWORD [edi], xmmA
-        add     edi, byte SIZEOF_XMMWORD/2
+        add     edi, byte SIZEOF_XMMWORD/8*4
         sub     ecx, byte SIZEOF_XMMWORD/8
-        psrldq  xmmA, 64
+        psrldq  xmmA, SIZEOF_XMMWORD/8*4
 .column_st7:
         ; Store one pixel (4 bytes) of xmmA to the output when it has enough
         ; space.
         test    ecx, ecx
         jz      short .endcolumn
         movd    DWORD [edi], xmmA
-%else
-        cmp     ecx, byte SIZEOF_XMMWORD/16
-        jb      short .endcolumn
-        mov     eax,ecx
-        xor     ecx, byte 0x03
-        inc     ecx
-        shl     ecx, 4
-        movd    xmmF,ecx
-        psrlq   xmmE,xmmF
-        punpcklbw xmmE,xmmE
-        ; ----------------
-        mov     ecx,edi
-        and     ecx, byte SIZEOF_XMMWORD-1
-        jz      short .adj0
-        lea     eax, [ecx+eax*4]        ; RGB_PIXELSIZE
-        cmp     eax, byte SIZEOF_XMMWORD
-        ja      short .adj0
-        and     edi, byte (-SIZEOF_XMMWORD)     ; align to 16-byte boundary
-        shl     ecx, 3                  ; pslldq xmmA,ecx & pslldq xmmE,ecx
-        movdqa  xmmB,xmmA
-        movdqa  xmmG,xmmE
-        pslldq  xmmA, SIZEOF_XMMWORD/2
-        pslldq  xmmE, SIZEOF_XMMWORD/2
-        movd    xmmC,ecx
-        sub     ecx, byte (SIZEOF_XMMWORD/2)*BYTE_BIT
-        jb      short .adj1
-        movd    xmmH,ecx
-        psllq   xmmA,xmmH
-        psllq   xmmE,xmmH
-        jmp     short .adj0
-.adj1:  neg     ecx
-        movd    xmmH,ecx
-        psrlq   xmmA,xmmH
-        psrlq   xmmE,xmmH
-        psllq   xmmB,xmmC
-        psllq   xmmG,xmmC
-        por     xmmA,xmmB
-        por     xmmE,xmmG
-.adj0:  ; ----------------
-        maskmovdqu xmmA,xmmE                    ; movntdqu XMMWORD [edi], xmmA
-%endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %endif ; RGB_PIXELSIZE ; ---------------
 
 .endcolumn:
         sfence          ; flush the write buffer
 
 .return:
         pop     edi
         pop     esi
 ;       pop     edx             ; need not be preserved
@@ skipping 67 matching lines @@
         pop     esi
 ;       pop     edx             ; need not be preserved
 ;       pop     ecx             ; need not be preserved
         pop     ebx
         pop     ebp
         ret
 
 ; For some reason, the OS X linker does not honor the request to align the
 ; segment unless we do this.
         align   16