diff --git a/common/jpeg/jdmaster.c b/common/jpeg/jdmaster.c
index 2802c5b..38db792 100644
--- a/common/jpeg/jdmaster.c
+++ b/common/jpeg/jdmaster.c
@@ -49,8 +49,14 @@
     return FALSE;
   /* jdmerge.c only supports YCC=>RGB color conversion */
   if (cinfo->jpeg_color_space != JCS_YCbCr || cinfo->num_components != 3 ||
-      cinfo->out_color_space != JCS_RGB ||
-      cinfo->out_color_components != RGB_PIXELSIZE)
+      (cinfo->out_color_space != JCS_RGB &&
+      cinfo->out_color_space != JCS_EXT_RGB &&
+      cinfo->out_color_space != JCS_EXT_RGBX &&
+      cinfo->out_color_space != JCS_EXT_BGR &&
+      cinfo->out_color_space != JCS_EXT_BGRX &&
+      cinfo->out_color_space != JCS_EXT_XBGR &&
+      cinfo->out_color_space != JCS_EXT_XRGB) ||
+      cinfo->out_color_components != rgb_pixelsize[cinfo->out_color_space])
     return FALSE;
   /* and it only handles 2h1v or 2h2v sampling ratios */
   if (cinfo->comp_info[0].h_samp_factor != 2 ||
@@ -175,10 +181,14 @@
     cinfo->out_color_components = 1;
     break;
   case JCS_RGB:
-#if RGB_PIXELSIZE != 3
-    cinfo->out_color_components = RGB_PIXELSIZE;
+  case JCS_EXT_RGB:
+  case JCS_EXT_RGBX:
+  case JCS_EXT_BGR:
+  case JCS_EXT_BGRX:
+  case JCS_EXT_XBGR:
+  case JCS_EXT_XRGB:
+    cinfo->out_color_components = rgb_pixelsize[cinfo->out_color_space];
     break;
-#endif /* else share code with YCbCr */
   case JCS_YCbCr:
     cinfo->out_color_components = 3;
     break;
diff --git a/common/jpeg/jsimd.c b/common/jpeg/jsimd.c
index 2c49a6b..861309a 100644
--- a/common/jpeg/jsimd.c
+++ b/common/jpeg/jsimd.c
@@ -461,12 +461,45 @@
                             JSAMPARRAY output_buf)
 {
 #ifdef WITH_SIMD
+  void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+  void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+  switch(cinfo->out_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_h2v2_extrgb_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extrgb_merged_upsample_mmx;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_h2v2_extrgbx_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extrgbx_merged_upsample_mmx;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_h2v2_extbgr_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extbgr_merged_upsample_mmx;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_h2v2_extbgrx_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extbgrx_merged_upsample_mmx;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_h2v2_extxbgr_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extxbgr_merged_upsample_mmx;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_h2v2_extxrgb_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_extxrgb_merged_upsample_mmx;
+      break;
+    default:
+      sse2fct=jsimd_h2v2_merged_upsample_sse2;
+      mmxfct=jsimd_h2v2_merged_upsample_mmx;
+      break;
+  }
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
-    jsimd_h2v2_merged_upsample_sse2(cinfo->output_width, input_buf,
+    sse2fct(cinfo->output_width, input_buf,
         in_row_group_ctr, output_buf);
   else if (simd_support & JSIMD_MMX)
-    jsimd_h2v2_merged_upsample_mmx(cinfo->output_width, input_buf,
+    mmxfct(cinfo->output_width, input_buf,
         in_row_group_ctr, output_buf);
 #endif
 }
@@ -478,12 +511,45 @@
                             JSAMPARRAY output_buf)
 {
 #ifdef WITH_SIMD
+  void (*sse2fct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+  void (*mmxfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
+  switch(cinfo->out_color_space)
+  {
+    case JCS_EXT_RGB:
+      sse2fct=jsimd_h2v1_extrgb_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extrgb_merged_upsample_mmx;
+      break;
+    case JCS_EXT_RGBX:
+      sse2fct=jsimd_h2v1_extrgbx_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extrgbx_merged_upsample_mmx;
+      break;
+    case JCS_EXT_BGR:
+      sse2fct=jsimd_h2v1_extbgr_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extbgr_merged_upsample_mmx;
+      break;
+    case JCS_EXT_BGRX:
+      sse2fct=jsimd_h2v1_extbgrx_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extbgrx_merged_upsample_mmx;
+      break;
+    case JCS_EXT_XBGR:
+      sse2fct=jsimd_h2v1_extxbgr_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extxbgr_merged_upsample_mmx;
+      break;
+    case JCS_EXT_XRGB:
+      sse2fct=jsimd_h2v1_extxrgb_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_extxrgb_merged_upsample_mmx;
+      break;
+    default:
+      sse2fct=jsimd_h2v1_merged_upsample_sse2;
+      mmxfct=jsimd_h2v1_merged_upsample_mmx;
+      break;
+  }
   if ((simd_support & JSIMD_SSE2) &&
       IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
-    jsimd_h2v1_merged_upsample_sse2(cinfo->output_width, input_buf,
+    sse2fct(cinfo->output_width, input_buf,
         in_row_group_ctr, output_buf);
   else if (simd_support & JSIMD_MMX)
-    jsimd_h2v1_merged_upsample_mmx(cinfo->output_width, input_buf,
+    mmxfct(cinfo->output_width, input_buf,
         in_row_group_ctr, output_buf);
 #endif
 }
diff --git a/common/jpeg/simd/jdmermmx.asm b/common/jpeg/simd/jdmermmx.asm
index 8ebe243..fd587fb 100644
--- a/common/jpeg/simd/jdmermmx.asm
+++ b/common/jpeg/simd/jdmermmx.asm
@@ -2,6 +2,7 @@
 ; jdmermmx.asm - merged upsampling/color conversion (MMX)
 ;
 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright 2009 D. R. Commander
 ;
 ; Based on
 ; x86 SIMD extension for IJG JPEG library
@@ -17,7 +18,6 @@
 ; [TAB8]
 
 %include "jsimdext.inc"
-%include "jcolsamp.inc"
 
 ; --------------------------------------------------------------------------
 
@@ -48,445 +48,76 @@
 	alignz	16
 
 ; --------------------------------------------------------------------------
-	SECTION	SEG_TEXT
-	BITS	32
-;
-; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
-;
-; GLOBAL(void)
-; jsimd_h2v1_merged_upsample_mmx (JDIMENSION output_width,
-;                                 JSAMPIMAGE input_buf,
-;                                 JDIMENSION in_row_group_ctr,
-;                                 JSAMPARRAY output_buf);
-;
+%include "jdmrgmmx.asm"
 
-%define output_width(b)	(b)+8			; JDIMENSION output_width
-%define input_buf(b)		(b)+12		; JSAMPIMAGE input_buf
-%define in_row_group_ctr(b)	(b)+16		; JDIMENSION in_row_group_ctr
-%define output_buf(b)		(b)+20		; JSAMPARRAY output_buf
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 3
+%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extrgb_merged_upsample_mmx
+%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extrgb_merged_upsample_mmx
+%include "jdmrgmmx.asm"
 
-%define original_ebp	ebp+0
-%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_MMWORD	; mmword wk[WK_NUM]
-%define WK_NUM		3
-%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 4
+%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extrgbx_merged_upsample_mmx
+%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extrgbx_merged_upsample_mmx
+%include "jdmrgmmx.asm"
 
-	align	16
-	global	EXTN(jsimd_h2v1_merged_upsample_mmx)
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 3
+%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extbgr_merged_upsample_mmx
+%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extbgr_merged_upsample_mmx
+%include "jdmrgmmx.asm"
 
-EXTN(jsimd_h2v1_merged_upsample_mmx):
-	push	ebp
-	mov	eax,esp				; eax = original ebp
-	sub	esp, byte 4
-	and	esp, byte (-SIZEOF_MMWORD)	; align to 64 bits
-	mov	[esp],eax
-	mov	ebp,esp				; ebp = aligned ebp
-	lea	esp, [wk(0)]
-	pushpic	eax		; make a room for GOT address
-	push	ebx
-;	push	ecx		; need not be preserved
-;	push	edx		; need not be preserved
-	push	esi
-	push	edi
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 4
+%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extbgrx_merged_upsample_mmx
+%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extbgrx_merged_upsample_mmx
+%include "jdmrgmmx.asm"
 
-	get_GOT	ebx			; get GOT address
-	movpic	POINTER [gotptr], ebx	; save GOT address
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 3
+%define RGB_GREEN 2
+%define RGB_BLUE 1
+%define RGB_PIXELSIZE 4
+%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extxbgr_merged_upsample_mmx
+%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extxbgr_merged_upsample_mmx
+%include "jdmrgmmx.asm"
 
-	mov	ecx, JDIMENSION [output_width(eax)]	; col
-	test	ecx,ecx
-	jz	near .return
-
-	push	ecx
-
-	mov	edi, JSAMPIMAGE [input_buf(eax)]
-	mov	ecx, JDIMENSION [in_row_group_ctr(eax)]
-	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
-	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
-	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
-	mov	edi, JSAMPARRAY [output_buf(eax)]
-	mov	esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW]		; inptr0
-	mov	ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW]		; inptr1
-	mov	edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW]		; inptr2
-	mov	edi, JSAMPROW [edi]				; outptr
-
-	pop	ecx			; col
-
-	alignx	16,7
-.columnloop:
-	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
-
-	movq      mm6, MMWORD [ebx]	; mm6=Cb(01234567)
-	movq      mm7, MMWORD [edx]	; mm7=Cr(01234567)
-
-	pxor      mm1,mm1		; mm1=(all 0's)
-	pcmpeqw   mm3,mm3
-	psllw     mm3,7			; mm3={0xFF80 0xFF80 0xFF80 0xFF80}
-
-	movq      mm4,mm6
-	punpckhbw mm6,mm1		; mm6=Cb(4567)=CbH
-	punpcklbw mm4,mm1		; mm4=Cb(0123)=CbL
-	movq      mm0,mm7
-	punpckhbw mm7,mm1		; mm7=Cr(4567)=CrH
-	punpcklbw mm0,mm1		; mm0=Cr(0123)=CrL
-
-	paddw     mm6,mm3
-	paddw     mm4,mm3
-	paddw     mm7,mm3
-	paddw     mm0,mm3
-
-	; (Original)
-	; R = Y                + 1.40200 * Cr
-	; G = Y - 0.34414 * Cb - 0.71414 * Cr
-	; B = Y + 1.77200 * Cb
-	;
-	; (This implementation)
-	; R = Y                + 0.40200 * Cr + Cr
-	; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
-	; B = Y - 0.22800 * Cb + Cb + Cb
-
-	movq	mm5,mm6			; mm5=CbH
-	movq	mm2,mm4			; mm2=CbL
-	paddw	mm6,mm6			; mm6=2*CbH
-	paddw	mm4,mm4			; mm4=2*CbL
-	movq	mm1,mm7			; mm1=CrH
-	movq	mm3,mm0			; mm3=CrL
-	paddw	mm7,mm7			; mm7=2*CrH
-	paddw	mm0,mm0			; mm0=2*CrL
-
-	pmulhw	mm6,[GOTOFF(eax,PW_MF0228)]	; mm6=(2*CbH * -FIX(0.22800))
-	pmulhw	mm4,[GOTOFF(eax,PW_MF0228)]	; mm4=(2*CbL * -FIX(0.22800))
-	pmulhw	mm7,[GOTOFF(eax,PW_F0402)]	; mm7=(2*CrH * FIX(0.40200))
-	pmulhw	mm0,[GOTOFF(eax,PW_F0402)]	; mm0=(2*CrL * FIX(0.40200))
-
-	paddw	mm6,[GOTOFF(eax,PW_ONE)]
-	paddw	mm4,[GOTOFF(eax,PW_ONE)]
-	psraw	mm6,1			; mm6=(CbH * -FIX(0.22800))
-	psraw	mm4,1			; mm4=(CbL * -FIX(0.22800))
-	paddw	mm7,[GOTOFF(eax,PW_ONE)]
-	paddw	mm0,[GOTOFF(eax,PW_ONE)]
-	psraw	mm7,1			; mm7=(CrH * FIX(0.40200))
-	psraw	mm0,1			; mm0=(CrL * FIX(0.40200))
-
-	paddw	mm6,mm5
-	paddw	mm4,mm2
-	paddw	mm6,mm5			; mm6=(CbH * FIX(1.77200))=(B-Y)H
-	paddw	mm4,mm2			; mm4=(CbL * FIX(1.77200))=(B-Y)L
-	paddw	mm7,mm1			; mm7=(CrH * FIX(1.40200))=(R-Y)H
-	paddw	mm0,mm3			; mm0=(CrL * FIX(1.40200))=(R-Y)L
-
-	movq	MMWORD [wk(0)], mm6	; wk(0)=(B-Y)H
-	movq	MMWORD [wk(1)], mm7	; wk(1)=(R-Y)H
-
-	movq      mm6,mm5
-	movq      mm7,mm2
-	punpcklwd mm5,mm1
-	punpckhwd mm6,mm1
-	pmaddwd   mm5,[GOTOFF(eax,PW_MF0344_F0285)]
-	pmaddwd   mm6,[GOTOFF(eax,PW_MF0344_F0285)]
-	punpcklwd mm2,mm3
-	punpckhwd mm7,mm3
-	pmaddwd   mm2,[GOTOFF(eax,PW_MF0344_F0285)]
-	pmaddwd   mm7,[GOTOFF(eax,PW_MF0344_F0285)]
-
-	paddd     mm5,[GOTOFF(eax,PD_ONEHALF)]
-	paddd     mm6,[GOTOFF(eax,PD_ONEHALF)]
-	psrad     mm5,SCALEBITS
-	psrad     mm6,SCALEBITS
-	paddd     mm2,[GOTOFF(eax,PD_ONEHALF)]
-	paddd     mm7,[GOTOFF(eax,PD_ONEHALF)]
-	psrad     mm2,SCALEBITS
-	psrad     mm7,SCALEBITS
-
-	packssdw  mm5,mm6	; mm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
-	packssdw  mm2,mm7	; mm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
-	psubw     mm5,mm1	; mm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
-	psubw     mm2,mm3	; mm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
-
-	movq	MMWORD [wk(2)], mm5	; wk(2)=(G-Y)H
-
-	mov	al,2			; Yctr
-	jmp	short .Yloop_1st
-	alignx	16,7
-
-.Yloop_2nd:
-	movq	mm0, MMWORD [wk(1)]	; mm0=(R-Y)H
-	movq	mm2, MMWORD [wk(2)]	; mm2=(G-Y)H
-	movq	mm4, MMWORD [wk(0)]	; mm4=(B-Y)H
-	alignx	16,7
-
-.Yloop_1st:
-	movq	mm7, MMWORD [esi]	; mm7=Y(01234567)
-
-	pcmpeqw	mm6,mm6
-	psrlw	mm6,BYTE_BIT		; mm6={0xFF 0x00 0xFF 0x00 ..}
-	pand	mm6,mm7			; mm6=Y(0246)=YE
-	psrlw	mm7,BYTE_BIT		; mm7=Y(1357)=YO
-
-	movq	mm1,mm0			; mm1=mm0=(R-Y)(L/H)
-	movq	mm3,mm2			; mm3=mm2=(G-Y)(L/H)
-	movq	mm5,mm4			; mm5=mm4=(B-Y)(L/H)
-
-	paddw     mm0,mm6		; mm0=((R-Y)+YE)=RE=(R0 R2 R4 R6)
-	paddw     mm1,mm7		; mm1=((R-Y)+YO)=RO=(R1 R3 R5 R7)
-	packuswb  mm0,mm0		; mm0=(R0 R2 R4 R6 ** ** ** **)
-	packuswb  mm1,mm1		; mm1=(R1 R3 R5 R7 ** ** ** **)
-
-	paddw     mm2,mm6		; mm2=((G-Y)+YE)=GE=(G0 G2 G4 G6)
-	paddw     mm3,mm7		; mm3=((G-Y)+YO)=GO=(G1 G3 G5 G7)
-	packuswb  mm2,mm2		; mm2=(G0 G2 G4 G6 ** ** ** **)
-	packuswb  mm3,mm3		; mm3=(G1 G3 G5 G7 ** ** ** **)
-
-	paddw     mm4,mm6		; mm4=((B-Y)+YE)=BE=(B0 B2 B4 B6)
-	paddw     mm5,mm7		; mm5=((B-Y)+YO)=BO=(B1 B3 B5 B7)
-	packuswb  mm4,mm4		; mm4=(B0 B2 B4 B6 ** ** ** **)
-	packuswb  mm5,mm5		; mm5=(B1 B3 B5 B7 ** ** ** **)
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-	; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
-	; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
-	; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
-	; mmG=(** ** ** ** ** ** ** **), mmH=(** ** ** ** ** ** ** **)
-
-	punpcklbw mmA,mmC		; mmA=(00 10 02 12 04 14 06 16)
-	punpcklbw mmE,mmB		; mmE=(20 01 22 03 24 05 26 07)
-	punpcklbw mmD,mmF		; mmD=(11 21 13 23 15 25 17 27)
-
-	movq      mmG,mmA
-	movq      mmH,mmA
-	punpcklwd mmA,mmE		; mmA=(00 10 20 01 02 12 22 03)
-	punpckhwd mmG,mmE		; mmG=(04 14 24 05 06 16 26 07)
-
-	psrlq     mmH,2*BYTE_BIT	; mmH=(02 12 04 14 06 16 -- --)
-	psrlq     mmE,2*BYTE_BIT	; mmE=(22 03 24 05 26 07 -- --)
-
-	movq      mmC,mmD
-	movq      mmB,mmD
-	punpcklwd mmD,mmH		; mmD=(11 21 02 12 13 23 04 14)
-	punpckhwd mmC,mmH		; mmC=(15 25 06 16 17 27 -- --)
-
-	psrlq     mmB,2*BYTE_BIT	; mmB=(13 23 15 25 17 27 -- --)
-
-	movq      mmF,mmE
-	punpcklwd mmE,mmB		; mmE=(22 03 13 23 24 05 15 25)
-	punpckhwd mmF,mmB		; mmF=(26 07 17 27 -- -- -- --)
-
-	punpckldq mmA,mmD		; mmA=(00 10 20 01 11 21 02 12)
-	punpckldq mmE,mmG		; mmE=(22 03 13 23 04 14 24 05)
-	punpckldq mmC,mmF		; mmC=(15 25 06 16 26 07 17 27)
-
-	cmp	ecx, byte SIZEOF_MMWORD
-	jb	short .column_st16
-
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmE
-	movq	MMWORD [edi+2*SIZEOF_MMWORD], mmC
-
-	sub	ecx, byte SIZEOF_MMWORD
-	jz	short .endcolumn
-
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; outptr
-	add	esi, byte SIZEOF_MMWORD			; inptr0
-	dec	al			; Yctr
-	jnz	near .Yloop_2nd
-
-	add	ebx, byte SIZEOF_MMWORD			; inptr1
-	add	edx, byte SIZEOF_MMWORD			; inptr2
-	jmp	near .columnloop
-	alignx	16,7
-
-.column_st16:
-	lea	ecx, [ecx+ecx*2]	; imul ecx, RGB_PIXELSIZE
-	cmp	ecx, byte 2*SIZEOF_MMWORD
-	jb	short .column_st8
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmE
-	movq	mmA,mmC
-	sub	ecx, byte 2*SIZEOF_MMWORD
-	add	edi, byte 2*SIZEOF_MMWORD
-	jmp	short .column_st4
-.column_st8:
-	cmp	ecx, byte SIZEOF_MMWORD
-	jb	short .column_st4
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	mmA,mmE
-	sub	ecx, byte SIZEOF_MMWORD
-	add	edi, byte SIZEOF_MMWORD
-.column_st4:
-	movd	eax,mmA
-	cmp	ecx, byte SIZEOF_DWORD
-	jb	short .column_st2
-	mov	DWORD [edi+0*SIZEOF_DWORD], eax
-	psrlq	mmA,DWORD_BIT
-	movd	eax,mmA
-	sub	ecx, byte SIZEOF_DWORD
-	add	edi, byte SIZEOF_DWORD
-.column_st2:
-	cmp	ecx, byte SIZEOF_WORD
-	jb	short .column_st1
-	mov	WORD [edi+0*SIZEOF_WORD], ax
-	shr	eax,WORD_BIT
-	sub	ecx, byte SIZEOF_WORD
-	add	edi, byte SIZEOF_WORD
-.column_st1:
-	cmp	ecx, byte SIZEOF_BYTE
-	jb	short .endcolumn
-	mov	BYTE [edi+0*SIZEOF_BYTE], al
-
-%else ; RGB_PIXELSIZE == 4 ; -----------
-
-%ifdef RGBX_FILLER_0XFF
-	pcmpeqb   mm6,mm6		; mm6=(X0 X2 X4 X6 ** ** ** **)
-	pcmpeqb   mm7,mm7		; mm7=(X1 X3 X5 X7 ** ** ** **)
-%else
-	pxor      mm6,mm6		; mm6=(X0 X2 X4 X6 ** ** ** **)
-	pxor      mm7,mm7		; mm7=(X1 X3 X5 X7 ** ** ** **)
-%endif
-	; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
-	; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
-	; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
-	; mmG=(30 32 34 36 ** ** ** **), mmH=(31 33 35 37 ** ** ** **)
-
-	punpcklbw mmA,mmC		; mmA=(00 10 02 12 04 14 06 16)
-	punpcklbw mmE,mmG		; mmE=(20 30 22 32 24 34 26 36)
-	punpcklbw mmB,mmD		; mmB=(01 11 03 13 05 15 07 17)
-	punpcklbw mmF,mmH		; mmF=(21 31 23 33 25 35 27 37)
-
-	movq      mmC,mmA
-	punpcklwd mmA,mmE		; mmA=(00 10 20 30 02 12 22 32)
-	punpckhwd mmC,mmE		; mmC=(04 14 24 34 06 16 26 36)
-	movq      mmG,mmB
-	punpcklwd mmB,mmF		; mmB=(01 11 21 31 03 13 23 33)
-	punpckhwd mmG,mmF		; mmG=(05 15 25 35 07 17 27 37)
-
-	movq      mmD,mmA
-	punpckldq mmA,mmB		; mmA=(00 10 20 30 01 11 21 31)
-	punpckhdq mmD,mmB		; mmD=(02 12 22 32 03 13 23 33)
-	movq      mmH,mmC
-	punpckldq mmC,mmG		; mmC=(04 14 24 34 05 15 25 35)
-	punpckhdq mmH,mmG		; mmH=(06 16 26 36 07 17 27 37)
-
-	cmp	ecx, byte SIZEOF_MMWORD
-	jb	short .column_st16
-
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmD
-	movq	MMWORD [edi+2*SIZEOF_MMWORD], mmC
-	movq	MMWORD [edi+3*SIZEOF_MMWORD], mmH
-
-	sub	ecx, byte SIZEOF_MMWORD
-	jz	short .endcolumn
-
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; outptr
-	add	esi, byte SIZEOF_MMWORD			; inptr0
-	dec	al			; Yctr
-	jnz	near .Yloop_2nd
-
-	add	ebx, byte SIZEOF_MMWORD			; inptr1
-	add	edx, byte SIZEOF_MMWORD			; inptr2
-	jmp	near .columnloop
-	alignx	16,7
-
-.column_st16:
-	cmp	ecx, byte SIZEOF_MMWORD/2
-	jb	short .column_st8
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmD
-	movq	mmA,mmC
-	movq	mmD,mmH
-	sub	ecx, byte SIZEOF_MMWORD/2
-	add	edi, byte 2*SIZEOF_MMWORD
-.column_st8:
-	cmp	ecx, byte SIZEOF_MMWORD/4
-	jb	short .column_st4
-	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
-	movq	mmA,mmD
-	sub	ecx, byte SIZEOF_MMWORD/4
-	add	edi, byte 1*SIZEOF_MMWORD
-.column_st4:
-	cmp	ecx, byte SIZEOF_MMWORD/8
-	jb	short .endcolumn
-	movd	DWORD [edi+0*SIZEOF_DWORD], mmA
-
-%endif ; RGB_PIXELSIZE ; ---------------
-
-.endcolumn:
-	emms		; empty MMX state
-
-.return:
-	pop	edi
-	pop	esi
-;	pop	edx		; need not be preserved
-;	pop	ecx		; need not be preserved
-	pop	ebx
-	mov	esp,ebp		; esp <- aligned ebp
-	pop	esp		; esp <- original ebp
-	pop	ebp
-	ret
-
-; --------------------------------------------------------------------------
-;
-; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
-;
-; GLOBAL(void)
-; jsimd_h2v2_merged_upsample_mmx (JDIMENSION output_width,
-;                                 JSAMPIMAGE input_buf,
-;                                 JDIMENSION in_row_group_ctr,
-;                                 JSAMPARRAY output_buf);
-;
-
-%define output_width(b)	(b)+8			; JDIMENSION output_width
-%define input_buf(b)		(b)+12		; JSAMPIMAGE input_buf
-%define in_row_group_ctr(b)	(b)+16		; JDIMENSION in_row_group_ctr
-%define output_buf(b)		(b)+20		; JSAMPARRAY output_buf
-
-	align	16
-	global	EXTN(jsimd_h2v2_merged_upsample_mmx)
-
-EXTN(jsimd_h2v2_merged_upsample_mmx):
-	push	ebp
-	mov	ebp,esp
-	push	ebx
-;	push	ecx		; need not be preserved
-;	push	edx		; need not be preserved
-	push	esi
-	push	edi
-
-	mov	eax, JDIMENSION [output_width(ebp)]
-
-	mov	edi, JSAMPIMAGE [input_buf(ebp)]
-	mov	ecx, JDIMENSION [in_row_group_ctr(ebp)]
-	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
-	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
-	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
-	mov	edi, JSAMPARRAY [output_buf(ebp)]
-	lea	esi, [esi+ecx*SIZEOF_JSAMPROW]
-
-	push	edx			; inptr2
-	push	ebx			; inptr1
-	push	esi			; inptr00
-	mov	ebx,esp
-
-	push	edi			; output_buf (outptr0)
-	push	ecx			; in_row_group_ctr
-	push	ebx			; input_buf
-	push	eax			; output_width
-
-	call	near EXTN(jsimd_h2v1_merged_upsample_mmx)
-
-	add	esi, byte SIZEOF_JSAMPROW	; inptr01
-	add	edi, byte SIZEOF_JSAMPROW	; outptr1
-	mov	POINTER [ebx+0*SIZEOF_POINTER], esi
-	mov	POINTER [ebx-1*SIZEOF_POINTER], edi
-
-	call	near EXTN(jsimd_h2v1_merged_upsample_mmx)
-
-	add	esp, byte 7*SIZEOF_DWORD
-
-	pop	edi
-	pop	esi
-;	pop	edx		; need not be preserved
-;	pop	ecx		; need not be preserved
-	pop	ebx
-	pop	ebp
-	ret
-
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 1
+%define RGB_GREEN 2
+%define RGB_BLUE 3
+%define RGB_PIXELSIZE 4
+%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extxrgb_merged_upsample_mmx
+%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extxrgb_merged_upsample_mmx
+%include "jdmrgmmx.asm"
diff --git a/common/jpeg/simd/jdmerss2.asm b/common/jpeg/simd/jdmerss2.asm
index 8a7c68f..2294e0d 100644
--- a/common/jpeg/simd/jdmerss2.asm
+++ b/common/jpeg/simd/jdmerss2.asm
@@ -2,6 +2,7 @@
 ; jdmerss2.asm - merged upsampling/color conversion (SSE2)
 ;
 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+; Copyright 2009 D. R. Commander
 ;
 ; Based on
 ; x86 SIMD extension for IJG JPEG library
@@ -17,7 +18,6 @@
 ; [TAB8]
 
 %include "jsimdext.inc"
-%include "jcolsamp.inc"
 
 ; --------------------------------------------------------------------------
 
@@ -48,543 +48,76 @@
 	alignz	16
 
 ; --------------------------------------------------------------------------
-	SECTION	SEG_TEXT
-	BITS	32
-;
-; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
-;
-; GLOBAL(void)
-; jsimd_h2v1_merged_upsample_sse2 (JDIMENSION output_width,
-;                                  JSAMPIMAGE input_buf,
-;                                  JDIMENSION in_row_group_ctr,
-;                                  JSAMPARRAY output_buf);
-;
+%include "jdmrgss2.asm"
 
-%define output_width(b)	(b)+8			; JDIMENSION output_width
-%define input_buf(b)		(b)+12		; JSAMPIMAGE input_buf
-%define in_row_group_ctr(b)	(b)+16		; JDIMENSION in_row_group_ctr
-%define output_buf(b)		(b)+20		; JSAMPARRAY output_buf
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 3
+%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extrgb_merged_upsample_sse2
+%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extrgb_merged_upsample_sse2
+%include "jdmrgss2.asm"
 
-%define original_ebp	ebp+0
-%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_XMMWORD	; xmmword wk[WK_NUM]
-%define WK_NUM		3
-%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 0
+%define RGB_GREEN 1
+%define RGB_BLUE 2
+%define RGB_PIXELSIZE 4
+%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extrgbx_merged_upsample_sse2
+%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extrgbx_merged_upsample_sse2
+%include "jdmrgss2.asm"
 
-	align	16
-	global	EXTN(jsimd_h2v1_merged_upsample_sse2)
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 3
+%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extbgr_merged_upsample_sse2
+%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extbgr_merged_upsample_sse2
+%include "jdmrgss2.asm"
 
-EXTN(jsimd_h2v1_merged_upsample_sse2):
-	push	ebp
-	mov	eax,esp				; eax = original ebp
-	sub	esp, byte 4
-	and	esp, byte (-SIZEOF_XMMWORD)	; align to 128 bits
-	mov	[esp],eax
-	mov	ebp,esp				; ebp = aligned ebp
-	lea	esp, [wk(0)]
-	pushpic	eax		; make a room for GOT address
-	push	ebx
-;	push	ecx		; need not be preserved
-;	push	edx		; need not be preserved
-	push	esi
-	push	edi
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 2
+%define RGB_GREEN 1
+%define RGB_BLUE 0
+%define RGB_PIXELSIZE 4
+%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extbgrx_merged_upsample_sse2
+%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extbgrx_merged_upsample_sse2
+%include "jdmrgss2.asm"
 
-	get_GOT	ebx			; get GOT address
-	movpic	POINTER [gotptr], ebx	; save GOT address
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 3
+%define RGB_GREEN 2
+%define RGB_BLUE 1
+%define RGB_PIXELSIZE 4
+%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extxbgr_merged_upsample_sse2
+%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extxbgr_merged_upsample_sse2
+%include "jdmrgss2.asm"
 
-	mov	ecx, JDIMENSION [output_width(eax)]	; col
-	test	ecx,ecx
-	jz	near .return
-
-	push	ecx
-
-	mov	edi, JSAMPIMAGE [input_buf(eax)]
-	mov	ecx, JDIMENSION [in_row_group_ctr(eax)]
-	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
-	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
-	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
-	mov	edi, JSAMPARRAY [output_buf(eax)]
-	mov	esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW]		; inptr0
-	mov	ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW]		; inptr1
-	mov	edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW]		; inptr2
-	mov	edi, JSAMPROW [edi]				; outptr
-
-	pop	ecx			; col
-
-	alignx	16,7
-.columnloop:
-	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
-
-	movdqa    xmm6, XMMWORD [ebx]	; xmm6=Cb(0123456789ABCDEF)
-	movdqa    xmm7, XMMWORD [edx]	; xmm7=Cr(0123456789ABCDEF)
-
-	pxor      xmm1,xmm1		; xmm1=(all 0's)
-	pcmpeqw   xmm3,xmm3
-	psllw     xmm3,7		; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}
-
-	movdqa    xmm4,xmm6
-	punpckhbw xmm6,xmm1		; xmm6=Cb(89ABCDEF)=CbH
-	punpcklbw xmm4,xmm1		; xmm4=Cb(01234567)=CbL
-	movdqa    xmm0,xmm7
-	punpckhbw xmm7,xmm1		; xmm7=Cr(89ABCDEF)=CrH
-	punpcklbw xmm0,xmm1		; xmm0=Cr(01234567)=CrL
-
-	paddw     xmm6,xmm3
-	paddw     xmm4,xmm3
-	paddw     xmm7,xmm3
-	paddw     xmm0,xmm3
-
-	; (Original)
-	; R = Y                + 1.40200 * Cr
-	; G = Y - 0.34414 * Cb - 0.71414 * Cr
-	; B = Y + 1.77200 * Cb
-	;
-	; (This implementation)
-	; R = Y                + 0.40200 * Cr + Cr
-	; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
-	; B = Y - 0.22800 * Cb + Cb + Cb
-
-	movdqa	xmm5,xmm6		; xmm5=CbH
-	movdqa	xmm2,xmm4		; xmm2=CbL
-	paddw	xmm6,xmm6		; xmm6=2*CbH
-	paddw	xmm4,xmm4		; xmm4=2*CbL
-	movdqa	xmm1,xmm7		; xmm1=CrH
-	movdqa	xmm3,xmm0		; xmm3=CrL
-	paddw	xmm7,xmm7		; xmm7=2*CrH
-	paddw	xmm0,xmm0		; xmm0=2*CrL
-
-	pmulhw	xmm6,[GOTOFF(eax,PW_MF0228)]	; xmm6=(2*CbH * -FIX(0.22800))
-	pmulhw	xmm4,[GOTOFF(eax,PW_MF0228)]	; xmm4=(2*CbL * -FIX(0.22800))
-	pmulhw	xmm7,[GOTOFF(eax,PW_F0402)]	; xmm7=(2*CrH * FIX(0.40200))
-	pmulhw	xmm0,[GOTOFF(eax,PW_F0402)]	; xmm0=(2*CrL * FIX(0.40200))
-
-	paddw	xmm6,[GOTOFF(eax,PW_ONE)]
-	paddw	xmm4,[GOTOFF(eax,PW_ONE)]
-	psraw	xmm6,1			; xmm6=(CbH * -FIX(0.22800))
-	psraw	xmm4,1			; xmm4=(CbL * -FIX(0.22800))
-	paddw	xmm7,[GOTOFF(eax,PW_ONE)]
-	paddw	xmm0,[GOTOFF(eax,PW_ONE)]
-	psraw	xmm7,1			; xmm7=(CrH * FIX(0.40200))
-	psraw	xmm0,1			; xmm0=(CrL * FIX(0.40200))
-
-	paddw	xmm6,xmm5
-	paddw	xmm4,xmm2
-	paddw	xmm6,xmm5		; xmm6=(CbH * FIX(1.77200))=(B-Y)H
-	paddw	xmm4,xmm2		; xmm4=(CbL * FIX(1.77200))=(B-Y)L
-	paddw	xmm7,xmm1		; xmm7=(CrH * FIX(1.40200))=(R-Y)H
-	paddw	xmm0,xmm3		; xmm0=(CrL * FIX(1.40200))=(R-Y)L
-
-	movdqa	XMMWORD [wk(0)], xmm6	; wk(0)=(B-Y)H
-	movdqa	XMMWORD [wk(1)], xmm7	; wk(1)=(R-Y)H
-
-	movdqa    xmm6,xmm5
-	movdqa    xmm7,xmm2
-	punpcklwd xmm5,xmm1
-	punpckhwd xmm6,xmm1
-	pmaddwd   xmm5,[GOTOFF(eax,PW_MF0344_F0285)]
-	pmaddwd   xmm6,[GOTOFF(eax,PW_MF0344_F0285)]
-	punpcklwd xmm2,xmm3
-	punpckhwd xmm7,xmm3
-	pmaddwd   xmm2,[GOTOFF(eax,PW_MF0344_F0285)]
-	pmaddwd   xmm7,[GOTOFF(eax,PW_MF0344_F0285)]
-
-	paddd     xmm5,[GOTOFF(eax,PD_ONEHALF)]
-	paddd     xmm6,[GOTOFF(eax,PD_ONEHALF)]
-	psrad     xmm5,SCALEBITS
-	psrad     xmm6,SCALEBITS
-	paddd     xmm2,[GOTOFF(eax,PD_ONEHALF)]
-	paddd     xmm7,[GOTOFF(eax,PD_ONEHALF)]
-	psrad     xmm2,SCALEBITS
-	psrad     xmm7,SCALEBITS
-
-	packssdw  xmm5,xmm6	; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
-	packssdw  xmm2,xmm7	; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
-	psubw     xmm5,xmm1	; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
-	psubw     xmm2,xmm3	; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
-
-	movdqa	XMMWORD [wk(2)], xmm5	; wk(2)=(G-Y)H
-
-	mov	al,2			; Yctr
-	jmp	short .Yloop_1st
-	alignx	16,7
-
-.Yloop_2nd:
-	movdqa	xmm0, XMMWORD [wk(1)]	; xmm0=(R-Y)H
-	movdqa	xmm2, XMMWORD [wk(2)]	; xmm2=(G-Y)H
-	movdqa	xmm4, XMMWORD [wk(0)]	; xmm4=(B-Y)H
-	alignx	16,7
-
-.Yloop_1st:
-	movdqa	xmm7, XMMWORD [esi]	; xmm7=Y(0123456789ABCDEF)
-
-	pcmpeqw	xmm6,xmm6
-	psrlw	xmm6,BYTE_BIT		; xmm6={0xFF 0x00 0xFF 0x00 ..}
-	pand	xmm6,xmm7		; xmm6=Y(02468ACE)=YE
-	psrlw	xmm7,BYTE_BIT		; xmm7=Y(13579BDF)=YO
-
-	movdqa	xmm1,xmm0		; xmm1=xmm0=(R-Y)(L/H)
-	movdqa	xmm3,xmm2		; xmm3=xmm2=(G-Y)(L/H)
-	movdqa	xmm5,xmm4		; xmm5=xmm4=(B-Y)(L/H)
-
-	paddw     xmm0,xmm6		; xmm0=((R-Y)+YE)=RE=R(02468ACE)
-	paddw     xmm1,xmm7		; xmm1=((R-Y)+YO)=RO=R(13579BDF)
-	packuswb  xmm0,xmm0		; xmm0=R(02468ACE********)
-	packuswb  xmm1,xmm1		; xmm1=R(13579BDF********)
-
-	paddw     xmm2,xmm6		; xmm2=((G-Y)+YE)=GE=G(02468ACE)
-	paddw     xmm3,xmm7		; xmm3=((G-Y)+YO)=GO=G(13579BDF)
-	packuswb  xmm2,xmm2		; xmm2=G(02468ACE********)
-	packuswb  xmm3,xmm3		; xmm3=G(13579BDF********)
-
-	paddw     xmm4,xmm6		; xmm4=((B-Y)+YE)=BE=B(02468ACE)
-	paddw     xmm5,xmm7		; xmm5=((B-Y)+YO)=BO=B(13579BDF)
-	packuswb  xmm4,xmm4		; xmm4=B(02468ACE********)
-	packuswb  xmm5,xmm5		; xmm5=B(13579BDF********)
-
-%if RGB_PIXELSIZE == 3 ; ---------------
-
-	; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
-	; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
-	; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
-	; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
-
-	punpcklbw xmmA,xmmC	; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
-	punpcklbw xmmE,xmmB	; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
-	punpcklbw xmmD,xmmF	; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
-
-	movdqa    xmmG,xmmA
-	movdqa    xmmH,xmmA
-	punpcklwd xmmA,xmmE	; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
-	punpckhwd xmmG,xmmE	; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
-
-	psrldq    xmmH,2	; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
-	psrldq    xmmE,2	; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
-
-	movdqa    xmmC,xmmD
-	movdqa    xmmB,xmmD
-	punpcklwd xmmD,xmmH	; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
-	punpckhwd xmmC,xmmH	; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
-
-	psrldq    xmmB,2	; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
-
-	movdqa    xmmF,xmmE
-	punpcklwd xmmE,xmmB	; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
-	punpckhwd xmmF,xmmB	; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
-
-	pshufd    xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
-	movdqa    xmmB,xmmE
-	punpckldq xmmA,xmmD	; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
-	punpckldq xmmE,xmmH	; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
-	punpckhdq xmmD,xmmB	; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
-
-	pshufd    xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
-	movdqa    xmmB,xmmF
-	punpckldq xmmG,xmmC	; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
-	punpckldq xmmF,xmmH	; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
-	punpckhdq xmmC,xmmB	; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
-
-	punpcklqdq xmmA,xmmE	; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
-	punpcklqdq xmmD,xmmG	; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
-	punpcklqdq xmmF,xmmC	; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
-
-	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
-.out0:
-	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
-	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
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	movdqa	xmmA,xmmD
-	sub	ecx, byte SIZEOF_XMMWORD
-.column_st15:
-	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
-
-%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
-	; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
-	; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
-	; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
-	; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
-
-	punpcklbw xmmA,xmmC	; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
-	punpcklbw xmmE,xmmG	; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
-	punpcklbw xmmB,xmmD	; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
-	punpcklbw xmmF,xmmH	; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
-
-	movdqa    xmmC,xmmA
-	punpcklwd xmmA,xmmE	; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
-	punpckhwd xmmC,xmmE	; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
-	movdqa    xmmG,xmmB
-	punpcklwd xmmB,xmmF	; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
-	punpckhwd xmmG,xmmF	; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
-
-	movdqa    xmmD,xmmA
-	punpckldq xmmA,xmmB	; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
-	punpckhdq xmmD,xmmB	; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
-	movdqa    xmmH,xmmC
-	punpckldq xmmC,xmmG	; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
-	punpckhdq xmmH,xmmG	; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
-
-	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
-.out0:
-	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
-	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
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	movdqa	xmmA,xmmD
-	sub	ecx, byte SIZEOF_XMMWORD/4
-.column_st15:
-	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 ; RGB_PIXELSIZE ; ---------------
-
-.endcolumn:
-	sfence		; flush the write buffer
-
-.return:
-	pop	edi
-	pop	esi
-;	pop	edx		; need not be preserved
-;	pop	ecx		; need not be preserved
-	pop	ebx
-	mov	esp,ebp		; esp <- aligned ebp
-	pop	esp		; esp <- original ebp
-	pop	ebp
-	ret
-
-; --------------------------------------------------------------------------
-;
-; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
-;
-; GLOBAL(void)
-; jsimd_h2v2_merged_upsample_sse2 (JDIMENSION output_width,
-;                                  JSAMPIMAGE input_buf,
-;                                  JDIMENSION in_row_group_ctr,
-;                                  JSAMPARRAY output_buf);
-;
-
-%define output_width(b)	(b)+8			; JDIMENSION output_width
-%define input_buf(b)		(b)+12		; JSAMPIMAGE input_buf
-%define in_row_group_ctr(b)	(b)+16		; JDIMENSION in_row_group_ctr
-%define output_buf(b)		(b)+20		; JSAMPARRAY output_buf
-
-	align	16
-	global	EXTN(jsimd_h2v2_merged_upsample_sse2)
-
-EXTN(jsimd_h2v2_merged_upsample_sse2):
-	push	ebp
-	mov	ebp,esp
-	push	ebx
-;	push	ecx		; need not be preserved
-;	push	edx		; need not be preserved
-	push	esi
-	push	edi
-
-	mov	eax, POINTER [output_width(ebp)]
-
-	mov	edi, JSAMPIMAGE [input_buf(ebp)]
-	mov	ecx, JDIMENSION [in_row_group_ctr(ebp)]
-	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
-	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
-	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
-	mov	edi, JSAMPARRAY [output_buf(ebp)]
-	lea	esi, [esi+ecx*SIZEOF_JSAMPROW]
-
-	push	edx			; inptr2
-	push	ebx			; inptr1
-	push	esi			; inptr00
-	mov	ebx,esp
-
-	push	edi			; output_buf (outptr0)
-	push	ecx			; in_row_group_ctr
-	push	ebx			; input_buf
-	push	eax			; output_width
-
-	call	near EXTN(jsimd_h2v1_merged_upsample_sse2)
-
-	add	esi, byte SIZEOF_JSAMPROW	; inptr01
-	add	edi, byte SIZEOF_JSAMPROW	; outptr1
-	mov	POINTER [ebx+0*SIZEOF_POINTER], esi
-	mov	POINTER [ebx-1*SIZEOF_POINTER], edi
-
-	call	near EXTN(jsimd_h2v1_merged_upsample_sse2)
-
-	add	esp, byte 7*SIZEOF_DWORD
-
-	pop	edi
-	pop	esi
-;	pop	edx		; need not be preserved
-;	pop	ecx		; need not be preserved
-	pop	ebx
-	pop	ebp
-	ret
-
+%undef RGB_RED
+%undef RGB_GREEN
+%undef RGB_BLUE
+%undef RGB_PIXELSIZE
+%define RGB_RED 1
+%define RGB_GREEN 2
+%define RGB_BLUE 3
+%define RGB_PIXELSIZE 4
+%define jsimd_h2v1_merged_upsample_sse2 jsimd_h2v1_extxrgb_merged_upsample_sse2
+%define jsimd_h2v2_merged_upsample_sse2 jsimd_h2v2_extxrgb_merged_upsample_sse2
+%include "jdmrgss2.asm"
diff --git a/common/jpeg/simd/jdmrgmmx.asm b/common/jpeg/simd/jdmrgmmx.asm
new file mode 100644
index 0000000..08b0985
--- /dev/null
+++ b/common/jpeg/simd/jdmrgmmx.asm
@@ -0,0 +1,463 @@
+;
+; jdmrgmmx.asm - merged upsampling/color conversion (MMX)
+;
+; Copyright 2009 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
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+
+; --------------------------------------------------------------------------
+	SECTION	SEG_TEXT
+	BITS	32
+;
+; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v1_merged_upsample_mmx (JDIMENSION output_width,
+;                                 JSAMPIMAGE input_buf,
+;                                 JDIMENSION in_row_group_ctr,
+;                                 JSAMPARRAY output_buf);
+;
+
+%define output_width(b)	(b)+8			; JDIMENSION output_width
+%define input_buf(b)		(b)+12		; JSAMPIMAGE input_buf
+%define in_row_group_ctr(b)	(b)+16		; JDIMENSION in_row_group_ctr
+%define output_buf(b)		(b)+20		; JSAMPARRAY output_buf
+
+%define original_ebp	ebp+0
+%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_MMWORD	; mmword wk[WK_NUM]
+%define WK_NUM		3
+%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+
+	align	16
+	global	EXTN(jsimd_h2v1_merged_upsample_mmx)
+
+EXTN(jsimd_h2v1_merged_upsample_mmx):
+	push	ebp
+	mov	eax,esp				; eax = original ebp
+	sub	esp, byte 4
+	and	esp, byte (-SIZEOF_MMWORD)	; align to 64 bits
+	mov	[esp],eax
+	mov	ebp,esp				; ebp = aligned ebp
+	lea	esp, [wk(0)]
+	pushpic	eax		; make a room for GOT address
+	push	ebx
+;	push	ecx		; need not be preserved
+;	push	edx		; need not be preserved
+	push	esi
+	push	edi
+
+	get_GOT	ebx			; get GOT address
+	movpic	POINTER [gotptr], ebx	; save GOT address
+
+	mov	ecx, JDIMENSION [output_width(eax)]	; col
+	test	ecx,ecx
+	jz	near .return
+
+	push	ecx
+
+	mov	edi, JSAMPIMAGE [input_buf(eax)]
+	mov	ecx, JDIMENSION [in_row_group_ctr(eax)]
+	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+	mov	edi, JSAMPARRAY [output_buf(eax)]
+	mov	esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW]		; inptr0
+	mov	ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW]		; inptr1
+	mov	edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW]		; inptr2
+	mov	edi, JSAMPROW [edi]				; outptr
+
+	pop	ecx			; col
+
+	alignx	16,7
+.columnloop:
+	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
+
+	movq      mm6, MMWORD [ebx]	; mm6=Cb(01234567)
+	movq      mm7, MMWORD [edx]	; mm7=Cr(01234567)
+
+	pxor      mm1,mm1		; mm1=(all 0's)
+	pcmpeqw   mm3,mm3
+	psllw     mm3,7			; mm3={0xFF80 0xFF80 0xFF80 0xFF80}
+
+	movq      mm4,mm6
+	punpckhbw mm6,mm1		; mm6=Cb(4567)=CbH
+	punpcklbw mm4,mm1		; mm4=Cb(0123)=CbL
+	movq      mm0,mm7
+	punpckhbw mm7,mm1		; mm7=Cr(4567)=CrH
+	punpcklbw mm0,mm1		; mm0=Cr(0123)=CrL
+
+	paddw     mm6,mm3
+	paddw     mm4,mm3
+	paddw     mm7,mm3
+	paddw     mm0,mm3
+
+	; (Original)
+	; R = Y                + 1.40200 * Cr
+	; G = Y - 0.34414 * Cb - 0.71414 * Cr
+	; B = Y + 1.77200 * Cb
+	;
+	; (This implementation)
+	; R = Y                + 0.40200 * Cr + Cr
+	; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+	; B = Y - 0.22800 * Cb + Cb + Cb
+
+	movq	mm5,mm6			; mm5=CbH
+	movq	mm2,mm4			; mm2=CbL
+	paddw	mm6,mm6			; mm6=2*CbH
+	paddw	mm4,mm4			; mm4=2*CbL
+	movq	mm1,mm7			; mm1=CrH
+	movq	mm3,mm0			; mm3=CrL
+	paddw	mm7,mm7			; mm7=2*CrH
+	paddw	mm0,mm0			; mm0=2*CrL
+
+	pmulhw	mm6,[GOTOFF(eax,PW_MF0228)]	; mm6=(2*CbH * -FIX(0.22800))
+	pmulhw	mm4,[GOTOFF(eax,PW_MF0228)]	; mm4=(2*CbL * -FIX(0.22800))
+	pmulhw	mm7,[GOTOFF(eax,PW_F0402)]	; mm7=(2*CrH * FIX(0.40200))
+	pmulhw	mm0,[GOTOFF(eax,PW_F0402)]	; mm0=(2*CrL * FIX(0.40200))
+
+	paddw	mm6,[GOTOFF(eax,PW_ONE)]
+	paddw	mm4,[GOTOFF(eax,PW_ONE)]
+	psraw	mm6,1			; mm6=(CbH * -FIX(0.22800))
+	psraw	mm4,1			; mm4=(CbL * -FIX(0.22800))
+	paddw	mm7,[GOTOFF(eax,PW_ONE)]
+	paddw	mm0,[GOTOFF(eax,PW_ONE)]
+	psraw	mm7,1			; mm7=(CrH * FIX(0.40200))
+	psraw	mm0,1			; mm0=(CrL * FIX(0.40200))
+
+	paddw	mm6,mm5
+	paddw	mm4,mm2
+	paddw	mm6,mm5			; mm6=(CbH * FIX(1.77200))=(B-Y)H
+	paddw	mm4,mm2			; mm4=(CbL * FIX(1.77200))=(B-Y)L
+	paddw	mm7,mm1			; mm7=(CrH * FIX(1.40200))=(R-Y)H
+	paddw	mm0,mm3			; mm0=(CrL * FIX(1.40200))=(R-Y)L
+
+	movq	MMWORD [wk(0)], mm6	; wk(0)=(B-Y)H
+	movq	MMWORD [wk(1)], mm7	; wk(1)=(R-Y)H
+
+	movq      mm6,mm5
+	movq      mm7,mm2
+	punpcklwd mm5,mm1
+	punpckhwd mm6,mm1
+	pmaddwd   mm5,[GOTOFF(eax,PW_MF0344_F0285)]
+	pmaddwd   mm6,[GOTOFF(eax,PW_MF0344_F0285)]
+	punpcklwd mm2,mm3
+	punpckhwd mm7,mm3
+	pmaddwd   mm2,[GOTOFF(eax,PW_MF0344_F0285)]
+	pmaddwd   mm7,[GOTOFF(eax,PW_MF0344_F0285)]
+
+	paddd     mm5,[GOTOFF(eax,PD_ONEHALF)]
+	paddd     mm6,[GOTOFF(eax,PD_ONEHALF)]
+	psrad     mm5,SCALEBITS
+	psrad     mm6,SCALEBITS
+	paddd     mm2,[GOTOFF(eax,PD_ONEHALF)]
+	paddd     mm7,[GOTOFF(eax,PD_ONEHALF)]
+	psrad     mm2,SCALEBITS
+	psrad     mm7,SCALEBITS
+
+	packssdw  mm5,mm6	; mm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
+	packssdw  mm2,mm7	; mm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
+	psubw     mm5,mm1	; mm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
+	psubw     mm2,mm3	; mm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
+
+	movq	MMWORD [wk(2)], mm5	; wk(2)=(G-Y)H
+
+	mov	al,2			; Yctr
+	jmp	short .Yloop_1st
+	alignx	16,7
+
+.Yloop_2nd:
+	movq	mm0, MMWORD [wk(1)]	; mm0=(R-Y)H
+	movq	mm2, MMWORD [wk(2)]	; mm2=(G-Y)H
+	movq	mm4, MMWORD [wk(0)]	; mm4=(B-Y)H
+	alignx	16,7
+
+.Yloop_1st:
+	movq	mm7, MMWORD [esi]	; mm7=Y(01234567)
+
+	pcmpeqw	mm6,mm6
+	psrlw	mm6,BYTE_BIT		; mm6={0xFF 0x00 0xFF 0x00 ..}
+	pand	mm6,mm7			; mm6=Y(0246)=YE
+	psrlw	mm7,BYTE_BIT		; mm7=Y(1357)=YO
+
+	movq	mm1,mm0			; mm1=mm0=(R-Y)(L/H)
+	movq	mm3,mm2			; mm3=mm2=(G-Y)(L/H)
+	movq	mm5,mm4			; mm5=mm4=(B-Y)(L/H)
+
+	paddw     mm0,mm6		; mm0=((R-Y)+YE)=RE=(R0 R2 R4 R6)
+	paddw     mm1,mm7		; mm1=((R-Y)+YO)=RO=(R1 R3 R5 R7)
+	packuswb  mm0,mm0		; mm0=(R0 R2 R4 R6 ** ** ** **)
+	packuswb  mm1,mm1		; mm1=(R1 R3 R5 R7 ** ** ** **)
+
+	paddw     mm2,mm6		; mm2=((G-Y)+YE)=GE=(G0 G2 G4 G6)
+	paddw     mm3,mm7		; mm3=((G-Y)+YO)=GO=(G1 G3 G5 G7)
+	packuswb  mm2,mm2		; mm2=(G0 G2 G4 G6 ** ** ** **)
+	packuswb  mm3,mm3		; mm3=(G1 G3 G5 G7 ** ** ** **)
+
+	paddw     mm4,mm6		; mm4=((B-Y)+YE)=BE=(B0 B2 B4 B6)
+	paddw     mm5,mm7		; mm5=((B-Y)+YO)=BO=(B1 B3 B5 B7)
+	packuswb  mm4,mm4		; mm4=(B0 B2 B4 B6 ** ** ** **)
+	packuswb  mm5,mm5		; mm5=(B1 B3 B5 B7 ** ** ** **)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+	; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
+	; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
+	; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
+	; mmG=(** ** ** ** ** ** ** **), mmH=(** ** ** ** ** ** ** **)
+
+	punpcklbw mmA,mmC		; mmA=(00 10 02 12 04 14 06 16)
+	punpcklbw mmE,mmB		; mmE=(20 01 22 03 24 05 26 07)
+	punpcklbw mmD,mmF		; mmD=(11 21 13 23 15 25 17 27)
+
+	movq      mmG,mmA
+	movq      mmH,mmA
+	punpcklwd mmA,mmE		; mmA=(00 10 20 01 02 12 22 03)
+	punpckhwd mmG,mmE		; mmG=(04 14 24 05 06 16 26 07)
+
+	psrlq     mmH,2*BYTE_BIT	; mmH=(02 12 04 14 06 16 -- --)
+	psrlq     mmE,2*BYTE_BIT	; mmE=(22 03 24 05 26 07 -- --)
+
+	movq      mmC,mmD
+	movq      mmB,mmD
+	punpcklwd mmD,mmH		; mmD=(11 21 02 12 13 23 04 14)
+	punpckhwd mmC,mmH		; mmC=(15 25 06 16 17 27 -- --)
+
+	psrlq     mmB,2*BYTE_BIT	; mmB=(13 23 15 25 17 27 -- --)
+
+	movq      mmF,mmE
+	punpcklwd mmE,mmB		; mmE=(22 03 13 23 24 05 15 25)
+	punpckhwd mmF,mmB		; mmF=(26 07 17 27 -- -- -- --)
+
+	punpckldq mmA,mmD		; mmA=(00 10 20 01 11 21 02 12)
+	punpckldq mmE,mmG		; mmE=(22 03 13 23 04 14 24 05)
+	punpckldq mmC,mmF		; mmC=(15 25 06 16 26 07 17 27)
+
+	cmp	ecx, byte SIZEOF_MMWORD
+	jb	short .column_st16
+
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmE
+	movq	MMWORD [edi+2*SIZEOF_MMWORD], mmC
+
+	sub	ecx, byte SIZEOF_MMWORD
+	jz	short .endcolumn
+
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; outptr
+	add	esi, byte SIZEOF_MMWORD			; inptr0
+	dec	al			; Yctr
+	jnz	near .Yloop_2nd
+
+	add	ebx, byte SIZEOF_MMWORD			; inptr1
+	add	edx, byte SIZEOF_MMWORD			; inptr2
+	jmp	near .columnloop
+	alignx	16,7
+
+.column_st16:
+	lea	ecx, [ecx+ecx*2]	; imul ecx, RGB_PIXELSIZE
+	cmp	ecx, byte 2*SIZEOF_MMWORD
+	jb	short .column_st8
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmE
+	movq	mmA,mmC
+	sub	ecx, byte 2*SIZEOF_MMWORD
+	add	edi, byte 2*SIZEOF_MMWORD
+	jmp	short .column_st4
+.column_st8:
+	cmp	ecx, byte SIZEOF_MMWORD
+	jb	short .column_st4
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	mmA,mmE
+	sub	ecx, byte SIZEOF_MMWORD
+	add	edi, byte SIZEOF_MMWORD
+.column_st4:
+	movd	eax,mmA
+	cmp	ecx, byte SIZEOF_DWORD
+	jb	short .column_st2
+	mov	DWORD [edi+0*SIZEOF_DWORD], eax
+	psrlq	mmA,DWORD_BIT
+	movd	eax,mmA
+	sub	ecx, byte SIZEOF_DWORD
+	add	edi, byte SIZEOF_DWORD
+.column_st2:
+	cmp	ecx, byte SIZEOF_WORD
+	jb	short .column_st1
+	mov	WORD [edi+0*SIZEOF_WORD], ax
+	shr	eax,WORD_BIT
+	sub	ecx, byte SIZEOF_WORD
+	add	edi, byte SIZEOF_WORD
+.column_st1:
+	cmp	ecx, byte SIZEOF_BYTE
+	jb	short .endcolumn
+	mov	BYTE [edi+0*SIZEOF_BYTE], al
+
+%else ; RGB_PIXELSIZE == 4 ; -----------
+
+%ifdef RGBX_FILLER_0XFF
+	pcmpeqb   mm6,mm6		; mm6=(X0 X2 X4 X6 ** ** ** **)
+	pcmpeqb   mm7,mm7		; mm7=(X1 X3 X5 X7 ** ** ** **)
+%else
+	pxor      mm6,mm6		; mm6=(X0 X2 X4 X6 ** ** ** **)
+	pxor      mm7,mm7		; mm7=(X1 X3 X5 X7 ** ** ** **)
+%endif
+	; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
+	; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
+	; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
+	; mmG=(30 32 34 36 ** ** ** **), mmH=(31 33 35 37 ** ** ** **)
+
+	punpcklbw mmA,mmC		; mmA=(00 10 02 12 04 14 06 16)
+	punpcklbw mmE,mmG		; mmE=(20 30 22 32 24 34 26 36)
+	punpcklbw mmB,mmD		; mmB=(01 11 03 13 05 15 07 17)
+	punpcklbw mmF,mmH		; mmF=(21 31 23 33 25 35 27 37)
+
+	movq      mmC,mmA
+	punpcklwd mmA,mmE		; mmA=(00 10 20 30 02 12 22 32)
+	punpckhwd mmC,mmE		; mmC=(04 14 24 34 06 16 26 36)
+	movq      mmG,mmB
+	punpcklwd mmB,mmF		; mmB=(01 11 21 31 03 13 23 33)
+	punpckhwd mmG,mmF		; mmG=(05 15 25 35 07 17 27 37)
+
+	movq      mmD,mmA
+	punpckldq mmA,mmB		; mmA=(00 10 20 30 01 11 21 31)
+	punpckhdq mmD,mmB		; mmD=(02 12 22 32 03 13 23 33)
+	movq      mmH,mmC
+	punpckldq mmC,mmG		; mmC=(04 14 24 34 05 15 25 35)
+	punpckhdq mmH,mmG		; mmH=(06 16 26 36 07 17 27 37)
+
+	cmp	ecx, byte SIZEOF_MMWORD
+	jb	short .column_st16
+
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmD
+	movq	MMWORD [edi+2*SIZEOF_MMWORD], mmC
+	movq	MMWORD [edi+3*SIZEOF_MMWORD], mmH
+
+	sub	ecx, byte SIZEOF_MMWORD
+	jz	short .endcolumn
+
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD	; outptr
+	add	esi, byte SIZEOF_MMWORD			; inptr0
+	dec	al			; Yctr
+	jnz	near .Yloop_2nd
+
+	add	ebx, byte SIZEOF_MMWORD			; inptr1
+	add	edx, byte SIZEOF_MMWORD			; inptr2
+	jmp	near .columnloop
+	alignx	16,7
+
+.column_st16:
+	cmp	ecx, byte SIZEOF_MMWORD/2
+	jb	short .column_st8
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	MMWORD [edi+1*SIZEOF_MMWORD], mmD
+	movq	mmA,mmC
+	movq	mmD,mmH
+	sub	ecx, byte SIZEOF_MMWORD/2
+	add	edi, byte 2*SIZEOF_MMWORD
+.column_st8:
+	cmp	ecx, byte SIZEOF_MMWORD/4
+	jb	short .column_st4
+	movq	MMWORD [edi+0*SIZEOF_MMWORD], mmA
+	movq	mmA,mmD
+	sub	ecx, byte SIZEOF_MMWORD/4
+	add	edi, byte 1*SIZEOF_MMWORD
+.column_st4:
+	cmp	ecx, byte SIZEOF_MMWORD/8
+	jb	short .endcolumn
+	movd	DWORD [edi+0*SIZEOF_DWORD], mmA
+
+%endif ; RGB_PIXELSIZE ; ---------------
+
+.endcolumn:
+	emms		; empty MMX state
+
+.return:
+	pop	edi
+	pop	esi
+;	pop	edx		; need not be preserved
+;	pop	ecx		; need not be preserved
+	pop	ebx
+	mov	esp,ebp		; esp <- aligned ebp
+	pop	esp		; esp <- original ebp
+	pop	ebp
+	ret
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v2_merged_upsample_mmx (JDIMENSION output_width,
+;                                 JSAMPIMAGE input_buf,
+;                                 JDIMENSION in_row_group_ctr,
+;                                 JSAMPARRAY output_buf);
+;
+
+%define output_width(b)	(b)+8			; JDIMENSION output_width
+%define input_buf(b)		(b)+12		; JSAMPIMAGE input_buf
+%define in_row_group_ctr(b)	(b)+16		; JDIMENSION in_row_group_ctr
+%define output_buf(b)		(b)+20		; JSAMPARRAY output_buf
+
+	align	16
+	global	EXTN(jsimd_h2v2_merged_upsample_mmx)
+
+EXTN(jsimd_h2v2_merged_upsample_mmx):
+	push	ebp
+	mov	ebp,esp
+	push	ebx
+;	push	ecx		; need not be preserved
+;	push	edx		; need not be preserved
+	push	esi
+	push	edi
+
+	mov	eax, JDIMENSION [output_width(ebp)]
+
+	mov	edi, JSAMPIMAGE [input_buf(ebp)]
+	mov	ecx, JDIMENSION [in_row_group_ctr(ebp)]
+	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+	mov	edi, JSAMPARRAY [output_buf(ebp)]
+	lea	esi, [esi+ecx*SIZEOF_JSAMPROW]
+
+	push	edx			; inptr2
+	push	ebx			; inptr1
+	push	esi			; inptr00
+	mov	ebx,esp
+
+	push	edi			; output_buf (outptr0)
+	push	ecx			; in_row_group_ctr
+	push	ebx			; input_buf
+	push	eax			; output_width
+
+	call	near EXTN(jsimd_h2v1_merged_upsample_mmx)
+
+	add	esi, byte SIZEOF_JSAMPROW	; inptr01
+	add	edi, byte SIZEOF_JSAMPROW	; outptr1
+	mov	POINTER [ebx+0*SIZEOF_POINTER], esi
+	mov	POINTER [ebx-1*SIZEOF_POINTER], edi
+
+	call	near EXTN(jsimd_h2v1_merged_upsample_mmx)
+
+	add	esp, byte 7*SIZEOF_DWORD
+
+	pop	edi
+	pop	esi
+;	pop	edx		; need not be preserved
+;	pop	ecx		; need not be preserved
+	pop	ebx
+	pop	ebp
+	ret
+
diff --git a/common/jpeg/simd/jdmrgss2.asm b/common/jpeg/simd/jdmrgss2.asm
new file mode 100644
index 0000000..5018a23
--- /dev/null
+++ b/common/jpeg/simd/jdmrgss2.asm
@@ -0,0 +1,561 @@
+;
+; jdmrgss2.asm - merged upsampling/color conversion (SSE2)
+;
+; Copyright 2009 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
+; http://sourceforge.net/project/showfiles.php?group_id=6208
+;
+; [TAB8]
+
+%include "jcolsamp.inc"
+				
+; --------------------------------------------------------------------------
+	SECTION	SEG_TEXT
+	BITS	32
+;
+; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v1_merged_upsample_sse2 (JDIMENSION output_width,
+;                                  JSAMPIMAGE input_buf,
+;                                  JDIMENSION in_row_group_ctr,
+;                                  JSAMPARRAY output_buf);
+;
+
+%define output_width(b)	(b)+8			; JDIMENSION output_width
+%define input_buf(b)		(b)+12		; JSAMPIMAGE input_buf
+%define in_row_group_ctr(b)	(b)+16		; JDIMENSION in_row_group_ctr
+%define output_buf(b)		(b)+20		; JSAMPARRAY output_buf
+
+%define original_ebp	ebp+0
+%define wk(i)		ebp-(WK_NUM-(i))*SIZEOF_XMMWORD	; xmmword wk[WK_NUM]
+%define WK_NUM		3
+%define gotptr		wk(0)-SIZEOF_POINTER	; void * gotptr
+
+	align	16
+	global	EXTN(jsimd_h2v1_merged_upsample_sse2)
+
+EXTN(jsimd_h2v1_merged_upsample_sse2):
+	push	ebp
+	mov	eax,esp				; eax = original ebp
+	sub	esp, byte 4
+	and	esp, byte (-SIZEOF_XMMWORD)	; align to 128 bits
+	mov	[esp],eax
+	mov	ebp,esp				; ebp = aligned ebp
+	lea	esp, [wk(0)]
+	pushpic	eax		; make a room for GOT address
+	push	ebx
+;	push	ecx		; need not be preserved
+;	push	edx		; need not be preserved
+	push	esi
+	push	edi
+
+	get_GOT	ebx			; get GOT address
+	movpic	POINTER [gotptr], ebx	; save GOT address
+
+	mov	ecx, JDIMENSION [output_width(eax)]	; col
+	test	ecx,ecx
+	jz	near .return
+
+	push	ecx
+
+	mov	edi, JSAMPIMAGE [input_buf(eax)]
+	mov	ecx, JDIMENSION [in_row_group_ctr(eax)]
+	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+	mov	edi, JSAMPARRAY [output_buf(eax)]
+	mov	esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW]		; inptr0
+	mov	ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW]		; inptr1
+	mov	edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW]		; inptr2
+	mov	edi, JSAMPROW [edi]				; outptr
+
+	pop	ecx			; col
+
+	alignx	16,7
+.columnloop:
+	movpic	eax, POINTER [gotptr]	; load GOT address (eax)
+
+	movdqa    xmm6, XMMWORD [ebx]	; xmm6=Cb(0123456789ABCDEF)
+	movdqa    xmm7, XMMWORD [edx]	; xmm7=Cr(0123456789ABCDEF)
+
+	pxor      xmm1,xmm1		; xmm1=(all 0's)
+	pcmpeqw   xmm3,xmm3
+	psllw     xmm3,7		; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..}
+
+	movdqa    xmm4,xmm6
+	punpckhbw xmm6,xmm1		; xmm6=Cb(89ABCDEF)=CbH
+	punpcklbw xmm4,xmm1		; xmm4=Cb(01234567)=CbL
+	movdqa    xmm0,xmm7
+	punpckhbw xmm7,xmm1		; xmm7=Cr(89ABCDEF)=CrH
+	punpcklbw xmm0,xmm1		; xmm0=Cr(01234567)=CrL
+
+	paddw     xmm6,xmm3
+	paddw     xmm4,xmm3
+	paddw     xmm7,xmm3
+	paddw     xmm0,xmm3
+
+	; (Original)
+	; R = Y                + 1.40200 * Cr
+	; G = Y - 0.34414 * Cb - 0.71414 * Cr
+	; B = Y + 1.77200 * Cb
+	;
+	; (This implementation)
+	; R = Y                + 0.40200 * Cr + Cr
+	; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+	; B = Y - 0.22800 * Cb + Cb + Cb
+
+	movdqa	xmm5,xmm6		; xmm5=CbH
+	movdqa	xmm2,xmm4		; xmm2=CbL
+	paddw	xmm6,xmm6		; xmm6=2*CbH
+	paddw	xmm4,xmm4		; xmm4=2*CbL
+	movdqa	xmm1,xmm7		; xmm1=CrH
+	movdqa	xmm3,xmm0		; xmm3=CrL
+	paddw	xmm7,xmm7		; xmm7=2*CrH
+	paddw	xmm0,xmm0		; xmm0=2*CrL
+
+	pmulhw	xmm6,[GOTOFF(eax,PW_MF0228)]	; xmm6=(2*CbH * -FIX(0.22800))
+	pmulhw	xmm4,[GOTOFF(eax,PW_MF0228)]	; xmm4=(2*CbL * -FIX(0.22800))
+	pmulhw	xmm7,[GOTOFF(eax,PW_F0402)]	; xmm7=(2*CrH * FIX(0.40200))
+	pmulhw	xmm0,[GOTOFF(eax,PW_F0402)]	; xmm0=(2*CrL * FIX(0.40200))
+
+	paddw	xmm6,[GOTOFF(eax,PW_ONE)]
+	paddw	xmm4,[GOTOFF(eax,PW_ONE)]
+	psraw	xmm6,1			; xmm6=(CbH * -FIX(0.22800))
+	psraw	xmm4,1			; xmm4=(CbL * -FIX(0.22800))
+	paddw	xmm7,[GOTOFF(eax,PW_ONE)]
+	paddw	xmm0,[GOTOFF(eax,PW_ONE)]
+	psraw	xmm7,1			; xmm7=(CrH * FIX(0.40200))
+	psraw	xmm0,1			; xmm0=(CrL * FIX(0.40200))
+
+	paddw	xmm6,xmm5
+	paddw	xmm4,xmm2
+	paddw	xmm6,xmm5		; xmm6=(CbH * FIX(1.77200))=(B-Y)H
+	paddw	xmm4,xmm2		; xmm4=(CbL * FIX(1.77200))=(B-Y)L
+	paddw	xmm7,xmm1		; xmm7=(CrH * FIX(1.40200))=(R-Y)H
+	paddw	xmm0,xmm3		; xmm0=(CrL * FIX(1.40200))=(R-Y)L
+
+	movdqa	XMMWORD [wk(0)], xmm6	; wk(0)=(B-Y)H
+	movdqa	XMMWORD [wk(1)], xmm7	; wk(1)=(R-Y)H
+
+	movdqa    xmm6,xmm5
+	movdqa    xmm7,xmm2
+	punpcklwd xmm5,xmm1
+	punpckhwd xmm6,xmm1
+	pmaddwd   xmm5,[GOTOFF(eax,PW_MF0344_F0285)]
+	pmaddwd   xmm6,[GOTOFF(eax,PW_MF0344_F0285)]
+	punpcklwd xmm2,xmm3
+	punpckhwd xmm7,xmm3
+	pmaddwd   xmm2,[GOTOFF(eax,PW_MF0344_F0285)]
+	pmaddwd   xmm7,[GOTOFF(eax,PW_MF0344_F0285)]
+
+	paddd     xmm5,[GOTOFF(eax,PD_ONEHALF)]
+	paddd     xmm6,[GOTOFF(eax,PD_ONEHALF)]
+	psrad     xmm5,SCALEBITS
+	psrad     xmm6,SCALEBITS
+	paddd     xmm2,[GOTOFF(eax,PD_ONEHALF)]
+	paddd     xmm7,[GOTOFF(eax,PD_ONEHALF)]
+	psrad     xmm2,SCALEBITS
+	psrad     xmm7,SCALEBITS
+
+	packssdw  xmm5,xmm6	; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285)
+	packssdw  xmm2,xmm7	; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285)
+	psubw     xmm5,xmm1	; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H
+	psubw     xmm2,xmm3	; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L
+
+	movdqa	XMMWORD [wk(2)], xmm5	; wk(2)=(G-Y)H
+
+	mov	al,2			; Yctr
+	jmp	short .Yloop_1st
+	alignx	16,7
+
+.Yloop_2nd:
+	movdqa	xmm0, XMMWORD [wk(1)]	; xmm0=(R-Y)H
+	movdqa	xmm2, XMMWORD [wk(2)]	; xmm2=(G-Y)H
+	movdqa	xmm4, XMMWORD [wk(0)]	; xmm4=(B-Y)H
+	alignx	16,7
+
+.Yloop_1st:
+	movdqa	xmm7, XMMWORD [esi]	; xmm7=Y(0123456789ABCDEF)
+
+	pcmpeqw	xmm6,xmm6
+	psrlw	xmm6,BYTE_BIT		; xmm6={0xFF 0x00 0xFF 0x00 ..}
+	pand	xmm6,xmm7		; xmm6=Y(02468ACE)=YE
+	psrlw	xmm7,BYTE_BIT		; xmm7=Y(13579BDF)=YO
+
+	movdqa	xmm1,xmm0		; xmm1=xmm0=(R-Y)(L/H)
+	movdqa	xmm3,xmm2		; xmm3=xmm2=(G-Y)(L/H)
+	movdqa	xmm5,xmm4		; xmm5=xmm4=(B-Y)(L/H)
+
+	paddw     xmm0,xmm6		; xmm0=((R-Y)+YE)=RE=R(02468ACE)
+	paddw     xmm1,xmm7		; xmm1=((R-Y)+YO)=RO=R(13579BDF)
+	packuswb  xmm0,xmm0		; xmm0=R(02468ACE********)
+	packuswb  xmm1,xmm1		; xmm1=R(13579BDF********)
+
+	paddw     xmm2,xmm6		; xmm2=((G-Y)+YE)=GE=G(02468ACE)
+	paddw     xmm3,xmm7		; xmm3=((G-Y)+YO)=GO=G(13579BDF)
+	packuswb  xmm2,xmm2		; xmm2=G(02468ACE********)
+	packuswb  xmm3,xmm3		; xmm3=G(13579BDF********)
+
+	paddw     xmm4,xmm6		; xmm4=((B-Y)+YE)=BE=B(02468ACE)
+	paddw     xmm5,xmm7		; xmm5=((B-Y)+YO)=BO=B(13579BDF)
+	packuswb  xmm4,xmm4		; xmm4=B(02468ACE********)
+	packuswb  xmm5,xmm5		; xmm5=B(13579BDF********)
+
+%if RGB_PIXELSIZE == 3 ; ---------------
+
+	; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+	; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+	; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+	; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **)
+
+	punpcklbw xmmA,xmmC	; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+	punpcklbw xmmE,xmmB	; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F)
+	punpcklbw xmmD,xmmF	; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F)
+
+	movdqa    xmmG,xmmA
+	movdqa    xmmH,xmmA
+	punpcklwd xmmA,xmmE	; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07)
+	punpckhwd xmmG,xmmE	; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F)
+
+	psrldq    xmmH,2	; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --)
+	psrldq    xmmE,2	; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --)
+
+	movdqa    xmmC,xmmD
+	movdqa    xmmB,xmmD
+	punpcklwd xmmD,xmmH	; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18)
+	punpckhwd xmmC,xmmH	; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --)
+
+	psrldq    xmmB,2	; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --)
+
+	movdqa    xmmF,xmmE
+	punpcklwd xmmE,xmmB	; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29)
+	punpckhwd xmmF,xmmB	; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --)
+
+	pshufd    xmmH,xmmA,0x4E; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03)
+	movdqa    xmmB,xmmE
+	punpckldq xmmA,xmmD	; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14)
+	punpckldq xmmE,xmmH	; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07)
+	punpckhdq xmmD,xmmB	; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29)
+
+	pshufd    xmmH,xmmG,0x4E; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B)
+	movdqa    xmmB,xmmF
+	punpckldq xmmG,xmmC	; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C)
+	punpckldq xmmF,xmmH	; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F)
+	punpckhdq xmmC,xmmB	; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --)
+
+	punpcklqdq xmmA,xmmE	; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05)
+	punpcklqdq xmmD,xmmG	; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A)
+	punpcklqdq xmmF,xmmC	; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F)
+
+	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
+.out0:
+	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
+	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
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqa	xmmA,xmmD
+	sub	ecx, byte SIZEOF_XMMWORD
+.column_st15:
+	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
+
+%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
+	; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **)
+	; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **)
+	; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **)
+	; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **)
+
+	punpcklbw xmmA,xmmC	; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E)
+	punpcklbw xmmE,xmmG	; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E)
+	punpcklbw xmmB,xmmD	; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F)
+	punpcklbw xmmF,xmmH	; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F)
+
+	movdqa    xmmC,xmmA
+	punpcklwd xmmA,xmmE	; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36)
+	punpckhwd xmmC,xmmE	; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E)
+	movdqa    xmmG,xmmB
+	punpcklwd xmmB,xmmF	; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37)
+	punpckhwd xmmG,xmmF	; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F)
+
+	movdqa    xmmD,xmmA
+	punpckldq xmmA,xmmB	; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33)
+	punpckhdq xmmD,xmmB	; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37)
+	movdqa    xmmH,xmmC
+	punpckldq xmmC,xmmG	; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B)
+	punpckhdq xmmH,xmmG	; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F)
+
+	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
+.out0:
+	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
+	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
+	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqa	xmmA,xmmD
+	sub	ecx, byte SIZEOF_XMMWORD/4
+.column_st15:
+	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 ; RGB_PIXELSIZE ; ---------------
+
+.endcolumn:
+	sfence		; flush the write buffer
+
+.return:
+	pop	edi
+	pop	esi
+;	pop	edx		; need not be preserved
+;	pop	ecx		; need not be preserved
+	pop	ebx
+	mov	esp,ebp		; esp <- aligned ebp
+	pop	esp		; esp <- original ebp
+	pop	ebp
+	ret
+
+; --------------------------------------------------------------------------
+;
+; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
+;
+; GLOBAL(void)
+; jsimd_h2v2_merged_upsample_sse2 (JDIMENSION output_width,
+;                                  JSAMPIMAGE input_buf,
+;                                  JDIMENSION in_row_group_ctr,
+;                                  JSAMPARRAY output_buf);
+;
+
+%define output_width(b)	(b)+8			; JDIMENSION output_width
+%define input_buf(b)		(b)+12		; JSAMPIMAGE input_buf
+%define in_row_group_ctr(b)	(b)+16		; JDIMENSION in_row_group_ctr
+%define output_buf(b)		(b)+20		; JSAMPARRAY output_buf
+
+	align	16
+	global	EXTN(jsimd_h2v2_merged_upsample_sse2)
+
+EXTN(jsimd_h2v2_merged_upsample_sse2):
+	push	ebp
+	mov	ebp,esp
+	push	ebx
+;	push	ecx		; need not be preserved
+;	push	edx		; need not be preserved
+	push	esi
+	push	edi
+
+	mov	eax, POINTER [output_width(ebp)]
+
+	mov	edi, JSAMPIMAGE [input_buf(ebp)]
+	mov	ecx, JDIMENSION [in_row_group_ctr(ebp)]
+	mov	esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
+	mov	ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
+	mov	edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
+	mov	edi, JSAMPARRAY [output_buf(ebp)]
+	lea	esi, [esi+ecx*SIZEOF_JSAMPROW]
+
+	push	edx			; inptr2
+	push	ebx			; inptr1
+	push	esi			; inptr00
+	mov	ebx,esp
+
+	push	edi			; output_buf (outptr0)
+	push	ecx			; in_row_group_ctr
+	push	ebx			; input_buf
+	push	eax			; output_width
+
+	call	near EXTN(jsimd_h2v1_merged_upsample_sse2)
+
+	add	esi, byte SIZEOF_JSAMPROW	; inptr01
+	add	edi, byte SIZEOF_JSAMPROW	; outptr1
+	mov	POINTER [ebx+0*SIZEOF_POINTER], esi
+	mov	POINTER [ebx-1*SIZEOF_POINTER], edi
+
+	call	near EXTN(jsimd_h2v1_merged_upsample_sse2)
+
+	add	esp, byte 7*SIZEOF_DWORD
+
+	pop	edi
+	pop	esi
+;	pop	edx		; need not be preserved
+;	pop	ecx		; need not be preserved
+	pop	ebx
+	pop	ebp
+	ret
+
diff --git a/common/jpeg/simd/jsimd.h b/common/jpeg/simd/jsimd.h
index a5b432f..c21cf29 100644
--- a/common/jpeg/simd/jsimd.h
+++ b/common/jpeg/simd/jsimd.h
@@ -59,7 +59,19 @@
 #define jsimd_h2v2_fancy_upsample_mmx         jSFUpH2V2M
 #define jsimd_h2v1_fancy_upsample_mmx         jSFUpH2V1M
 #define jsimd_h2v2_merged_upsample_mmx        jSMUpH2V2M
+#define jsimd_h2v2_extrgb_merged_upsample_mmx jSMUpH2V2EXTRGBM
+#define jsimd_h2v2_extrgbx_merged_upsample_mmx jSMUpH2V2EXTRGBXM
+#define jsimd_h2v2_extbgr_merged_upsample_mmx jSMUpH2V2EXTBGRM
+#define jsimd_h2v2_extbgrx_merged_upsample_mmx jSMUpH2V2EXTBGRXM
+#define jsimd_h2v2_extxbgr_merged_upsample_mmx jSMUpH2V2EXTXBGRM
+#define jsimd_h2v2_extxrgb_merged_upsample_mmx jSMUpH2V2EXTXRGBM
 #define jsimd_h2v1_merged_upsample_mmx        jSMUpH2V1M
+#define jsimd_h2v1_extrgb_merged_upsample_mmx jSMUpH2V1EXTRGBM
+#define jsimd_h2v1_extrgbx_merged_upsample_mmx jSMUpH2V1EXTRGBXM
+#define jsimd_h2v1_extbgr_merged_upsample_mmx jSMUpH2V1EXTBGRM
+#define jsimd_h2v1_extbgrx_merged_upsample_mmx jSMUpH2V1EXTBGRXM
+#define jsimd_h2v1_extxbgr_merged_upsample_mmx jSMUpH2V1EXTXBGRM
+#define jsimd_h2v1_extxrgb_merged_upsample_mmx jSMUpH2V1EXTXRGBM
 #define jsimd_h2v2_upsample_sse2              jSUpH2V2S2
 #define jsimd_h2v1_upsample_sse2              jSUpH2V1S2
 #define jconst_fancy_upsample_sse2            jSCFUpS2
@@ -67,7 +79,19 @@
 #define jsimd_h2v1_fancy_upsample_sse2        jSFUpH2V1S2
 #define jconst_merged_upsample_sse2           jSCMUpS2
 #define jsimd_h2v2_merged_upsample_sse2       jSMUpH2V2S2
+#define jsimd_h2v2_extrgb_merged_upsample_sse2 jSMUpH2V2EXTRGBS2
+#define jsimd_h2v2_extrgbx_merged_upsample_sse2 jSMUpH2V2EXTRGBXS2
+#define jsimd_h2v2_extbgr_merged_upsample_sse2 jSMUpH2V2EXTBGRS2
+#define jsimd_h2v2_extbgrx_merged_upsample_sse2 jSMUpH2V2EXTBGRXS2
+#define jsimd_h2v2_extxbgr_merged_upsample_sse2 jSMUpH2V2EXTXBGRS2
+#define jsimd_h2v2_extxrgb_merged_upsample_sse2 jSMUpH2V2EXTXRGBS2
 #define jsimd_h2v1_merged_upsample_sse2       jSMUpH2V1S2
+#define jsimd_h2v1_extrgb_merged_upsample_sse2 jSMUpH2V1EXTRGBS2
+#define jsimd_h2v1_extrgbx_merged_upsample_sse2 jSMUpH2V1EXTRGBXS2
+#define jsimd_h2v1_extbgr_merged_upsample_sse2 jSMUpH2V1EXTBGRS2
+#define jsimd_h2v1_extbgrx_merged_upsample_sse2 jSMUpH2V1EXTBGRXS2
+#define jsimd_h2v1_extxbgr_merged_upsample_sse2 jSMUpH2V1EXTXBGRS2
+#define jsimd_h2v1_extxrgb_merged_upsample_sse2 jSMUpH2V1EXTXRGBS2
 #define jsimd_convsamp_mmx                    jSConvM
 #define jsimd_convsamp_sse2                   jSConvS2
 #define jsimd_convsamp_float_3dnow            jSConvF3D
@@ -264,9 +288,45 @@
 EXTERN(void) jsimd_h2v2_merged_upsample_mmx
         JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
              JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
 EXTERN(void) jsimd_h2v1_merged_upsample_mmx
         JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
              JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_mmx
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
 
 EXTERN(void) jsimd_h2v2_upsample_sse2
         JPP((int max_v_samp_factor, JDIMENSION output_width,
@@ -287,9 +347,45 @@
 EXTERN(void) jsimd_h2v2_merged_upsample_sse2
         JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
              JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
 EXTERN(void) jsimd_h2v1_merged_upsample_sse2
         JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
              JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
+EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_sse2
+        JPP((JDIMENSION output_width, JSAMPIMAGE input_buf,
+             JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf));
 
 /* SIMD Sample Conversion */
 EXTERN(void) jsimd_convsamp_mmx JPP((JSAMPARRAY sample_data,