// ----------------- original code using intrinsics -----------------

static __forceinline __m64 LoadPixel(const Pixel* p)
{
  return _mm_unpacklo_pi8(_mm_cvtsi32_si64(p->v), _mm_setzero_si64());
}

static __forceinline __m64 AddLowAndHigh(__m64 a, __m64 b)
{
  __m64 lh = _mm_unpacklo_pi32(a, b); // low halves
  __m64 hh = _mm_unpackhi_pi32(a, b); // high halves

  return _mm_add_pi32(lh, hh);
}

static __forceinline __m64 DoubleDot(__m64 a, __m64 b, __m64 x)
{
  return AddLowAndHigh(_mm_madd_pi16(a, x), _mm_madd_pi16(b, x));
}

static __forceinline __m64 AverageInts(__m64 a, __m64 b)
{
  return _mm_srai_pi32(_mm_add_pi32(a, b), 1);
}

static void MatchColorsBlockIntrin(const Pixel *block, const Pixel *color, sU32 &blockMask)
{
  static const __m64 maskc0 = { 0xfffffff8fffffffe };
  static const __m64 maskh  = { 0x0000000400000001 };
  static const __m64 maskc3 = { 0x0000000800000002 };

  __m64 col0 = LoadPixel(color+0);
  __m64 col1 = LoadPixel(color+1);
  __m64 dir = _mm_sub_pi16(col0, col1);
  __m64 stop01 = DoubleDot(col0, col1, dir);

  __m64 col2 = LoadPixel(color+2);
  __m64 col3 = LoadPixel(color+3);
  __m64 stop23 = DoubleDot(col2, col3, dir);

  __m64 c3c0p = AverageInts(stop01, stop23);
  __m64 halfp = AverageInts(_mm_unpackhi_pi32(stop23, stop23), stop23);

  __m64 c0c = _mm_unpackhi_pi32(c3c0p, c3c0p);
  __m64 c3c = _mm_unpacklo_pi32(c3c0p, c3c0p);
  __m64 halfc = _mm_unpacklo_pi32(halfp, halfp);
  __m64 mask = _mm_setzero_si64();

  for(sInt i=14;i>=0;i-=2)
  {
    __m64 pix0 = LoadPixel(block + i + 0);
    __m64 pix1 = LoadPixel(block + i + 1);
    __m64 dots = DoubleDot(pix0, pix1, dir);

    mask = _mm_slli_pi32(mask, 4);
    mask = _mm_add_pi32(mask, _mm_and_si64(_mm_cmpgt_pi32(c0c, dots), maskc0));
    mask = _mm_add_pi32(mask, _mm_and_si64(_mm_cmpgt_pi32(halfc, dots), maskh));
    mask = _mm_add_pi32(mask, _mm_and_si64(_mm_cmpgt_pi32(c3c, dots), maskc3));
  }

  mask = _mm_or_si64(mask, _mm_unpackhi_pi32(mask, mask));
  blockMask = _mm_cvtsi64_si32(mask);
  _mm_empty();
}

// ----------------- generated code -----------------

;	COMDAT ?MatchColorsBlockIntrin@@YAXPBTPixel@@0AAI@Z
_TEXT	SEGMENT
_halfc$ = -16						; size = 8
_c3c$ = -8						; size = 8
_blockMask$ = 8						; size = 4
?MatchColorsBlockIntrin@@YAXPBTPixel@@0AAI@Z PROC	; MatchColorsBlockIntrin, COMDAT
; _block$ = ecx
; _color$ = eax

; 212  : {

	push	ebp
	mov	ebp, esp
	and	esp, -8					; fffffff8H
	sub	esp, 16					; 00000010H

; 213  :   static const __m64 maskc0 = { 0xfffffff8fffffffe };
; 214  :   static const __m64 maskh  = { 0x0000000400000001 };
; 215  :   static const __m64 maskc3 = { 0x0000000800000002 };
; 216  : 
; 217  :   __m64 col0 = LoadPixel(color+0);

	mov	edx, DWORD PTR [eax]
	pxor	mm0, mm0
	movd	mm1, edx

; 218  :   __m64 col1 = LoadPixel(color+1);

	mov	edx, DWORD PTR [eax+4]
	movq	mm2, mm1
	movq	mm1, mm0
	punpcklbw mm2, mm1
	movd	mm1, edx

; 219  :   __m64 dir = _mm_sub_pi16(col0, col1);
; 220  :   __m64 stop01 = DoubleDot(col0, col1, dir);
; 221  : 
; 222  :   __m64 col2 = LoadPixel(color+2);
; 223  :   __m64 col3 = LoadPixel(color+3);

	mov	edx, DWORD PTR [eax+12]
	mov	eax, DWORD PTR [eax+8]
	movq	mm3, mm1
	movq	mm1, mm0
	punpcklbw mm3, mm1
	movq	mm1, mm2
	movq	mm4, mm3
	psubw	mm1, mm4
	movq	mm4, mm1
	pmaddwd	mm3, mm4
	pmaddwd	mm2, mm4
	movq	mm4, mm2
	movq	mm5, mm0
	movd	mm2, edx
	punpcklbw mm2, mm5

; 224  :   __m64 stop23 = DoubleDot(col2, col3, dir);

	movq	mm5, mm1
	pmaddwd	mm2, mm5
	movd	mm5, eax
	movq	mm6, mm0
	punpcklbw mm5, mm6
	movq	mm6, mm1
	pmaddwd	mm5, mm6
	movq	mm6, mm5
	movq	mm7, mm2
	punpckldq mm6, mm7
	punpckhdq mm5, mm2
	paddd	mm6, mm5
	movq	mm2, mm6
	movq	mm5, mm4
	movq	mm6, mm3
	punpckldq mm5, mm6
	punpckhdq mm4, mm3
	movq	mm3, mm4
	paddd	mm5, mm3

; 225  : 
; 226  :   __m64 c3c0p = AverageInts(stop01, stop23);

	movq	mm3, mm2
	paddd	mm5, mm3
	movq	mm3, mm5
	psrad	mm3, 1

; 227  :   __m64 halfp = AverageInts(_mm_unpackhi_pi32(stop23, stop23), stop23);

	movq	mm4, mm2
	movq	mm5, mm2
	punpckhdq mm4, mm5
	paddd	mm4, mm2
	movq	mm2, mm4
	psrad	mm2, 1

; 228  : 
; 229  :   __m64 c0c = _mm_unpackhi_pi32(c3c0p, c3c0p);

	movq	mm4, mm3
	movq	mm5, mm3
	punpckhdq mm4, mm5
	movq	mm5, mm4

; 230  :   __m64 c3c = _mm_unpacklo_pi32(c3c0p, c3c0p);

	movq	mm4, mm3
	punpckldq mm4, mm3
	movq	MMWORD PTR _c3c$[esp+16], mm4

; 231  :   __m64 halfc = _mm_unpacklo_pi32(halfp, halfp);

	movq	mm3, mm2
	punpckldq mm3, mm2
	movq	MMWORD PTR _halfc$[esp+16], mm3

; 232  :   __m64 mask = _mm_setzero_si64();

	movq	mm4, mm0

; 233  : 
; 234  :   for(sInt i=14;i>=0;i-=2)

	mov	eax, 14					; 0000000eH
	npad	1
$LL3@MatchColor:
	sub	eax, 2

; 235  :   {
; 236  :     __m64 pix0 = LoadPixel(block + i + 0);
; 237  :     __m64 pix1 = LoadPixel(block + i + 1);

	mov	edx, DWORD PTR [ecx+eax*4+12]
	movd	mm2, edx
	mov	edx, DWORD PTR [ecx+eax*4+8]
	movq	mm3, mm0
	punpcklbw mm2, mm3

; 238  :     __m64 dots = DoubleDot(pix0, pix1, dir);

	movq	mm3, mm1
	pmaddwd	mm2, mm3
	movq	mm6, mm0
	movd	mm3, edx
	punpcklbw mm3, mm6
	movq	mm6, mm1
	pmaddwd	mm3, mm6
	movq	mm6, mm3
	movq	mm7, mm2
	punpckldq mm6, mm7
	punpckhdq mm3, mm2
	paddd	mm6, mm3
	movq	mm2, mm6

; 239  : 
; 240  :     mask = _mm_slli_pi32(mask, 4);

	movq	mm3, mm4
	pslld	mm3, 4

; 241  :     mask = _mm_add_pi32(mask, _mm_and_si64(_mm_cmpgt_pi32(c0c, dots), maskc0));

	movq	mm4, mm5
	pcmpgtd	mm4, mm6
	movq	mm6, MMWORD PTR ?maskc0@?1??MatchColorsBlockIntrin@@YAXPBTPixel@@0AAI@Z@4T__m64@@B
	pand	mm4, mm6
	paddd	mm3, mm4

; 242  :     mask = _mm_add_pi32(mask, _mm_and_si64(_mm_cmpgt_pi32(halfc, dots), maskh));

	movq	mm4, MMWORD PTR _halfc$[esp+16]
	movq	mm6, mm2
	pcmpgtd	mm4, mm6
	movq	mm6, MMWORD PTR ?maskh@?1??MatchColorsBlockIntrin@@YAXPBTPixel@@0AAI@Z@4T__m64@@B
	pand	mm4, mm6
	paddd	mm3, mm4

; 243  :     mask = _mm_add_pi32(mask, _mm_and_si64(_mm_cmpgt_pi32(c3c, dots), maskc3));

	movq	mm4, MMWORD PTR _c3c$[esp+16]
	pcmpgtd	mm4, mm2
	movq	mm2, mm4
	movq	mm4, MMWORD PTR ?maskc3@?1??MatchColorsBlockIntrin@@YAXPBTPixel@@0AAI@Z@4T__m64@@B
	pand	mm2, mm4
	paddd	mm3, mm2
	movq	mm4, mm3
	jns	$LL3@MatchColor

; 244  :   }
; 245  : 
; 246  :   mask = _mm_or_si64(mask, _mm_unpackhi_pi32(mask, mask));
; 247  :   blockMask = _mm_cvtsi64_si32(mask);

	mov	ecx, DWORD PTR _blockMask$[ebp]
	movq	mm0, mm4
	movq	mm1, mm4
	movq	mm2, mm4
	punpckhdq mm1, mm2
	por	mm0, mm1
	movd	eax, mm0
	mov	DWORD PTR [ecx], eax

; 248  :   _mm_empty();

	emms

; 249  : }

	mov	esp, ebp
	pop	ebp
	ret	0
?MatchColorsBlockIntrin@@YAXPBTPixel@@0AAI@Z ENDP	; MatchColorsBlockIntrin
_TEXT	ENDS

// ----------------- implemented by hand -----------------

static void MatchColorsBlockMMX(const Pixel *block,const Pixel *color,sU32 &blockMask)
{
  static const __m64 maskc0 = { 0xfffffff8fffffffe };
  static const __m64 maskh  = { 0x0000000400000001 };
  static const __m64 maskc3 = { 0x0000000800000002 };

  __m64 c0c,c3c,halfc;

  __asm
  {
    mov       esi, [color];
    mov       edi, [block];

    pxor      mm7, mm7;
    movd      mm0, [esi+0];
    punpcklbw mm0, mm7;
    movd      mm1, [esi+4];
    punpcklbw mm1, mm7;
    
    movq      mm2, mm0;
    psubw     mm0, mm1; // mm0=dir

    pmaddwd   mm2, mm0; // dp0
    pmaddwd   mm1, mm0; // dp1
    movq      mm3, mm2;
    punpckldq mm2, mm1;
    punpckhdq mm3, mm1;
    paddd     mm2, mm3; // mm2=stop01

    movd      mm1, [esi+8];
    punpcklbw mm1, mm7;
    movd      mm3, [esi+12];
    punpcklbw mm3, mm7;
    
    pmaddwd   mm1, mm0; // dp0
    pmaddwd   mm3, mm0; // dp1
    movq      mm4, mm1;
    punpckldq mm1, mm3;
    punpckhdq mm4, mm3;
    paddd     mm1, mm4; // mm1=stop23
    movq      mm4, mm1;

    paddd     mm2, mm1;
    punpckhdq mm4, mm4;
    psrad     mm2, 1;   // mm2=c3c0p
    paddd     mm4, mm1;
    movq      mm1, mm2;
    psrad     mm4, 1;   // mm4=halfp

    punpckhdq mm1, mm1; // mm1=c0c
    punpckldq mm2, mm2; // mm2=c3c
    punpckldq mm4, mm4; // mm4=halfc
    movq      [c0c], mm1;
    movq      [c3c], mm2;
    movq      [halfc], mm4;

    pxor      mm1, mm1; // mm1=mask
    mov       ecx, 14;

matchlp:
    pslld     mm1, 4;

    movd      mm2, [edi+ecx*4+0];
    punpcklbw mm2, mm7;
    movd      mm3, [edi+ecx*4+4];
    punpcklbw mm3, mm7;

    pmaddwd   mm2, mm0; // dp0
    pmaddwd   mm3, mm0; // dp1
    movq      mm4, mm2;
    punpckldq mm2, mm3;
    punpckhdq mm4, mm3;
    paddd     mm2, mm4; // mm2=dots

    movq      mm3, [c0c];
    movq      mm4, [halfc];
    movq      mm5, [c3c];
    pcmpgtd   mm3, mm2;
    pcmpgtd   mm4, mm2;
    pcmpgtd   mm5, mm2;
    pand      mm3, [maskc0];
    pand      mm4, [maskh];
    pand      mm5, [maskc3];
    paddd     mm1, mm3;
    paddd     mm1, mm4;
    paddd     mm1, mm5;

    sub       ecx, 2;
    jns       matchlp;

    mov       eax, [blockMask];

    movq      mm0, mm1;
    punpckhdq mm0, mm0;
    por       mm0, mm1;
    movd      [eax], mm0;
    emms;
  }
}

// ----------------- line-by-line translation of asm code to intrinsics -----------------

static void MatchColorsBlockIntrin2(const Pixel *block,const Pixel *color,sU32 &blockMask)
{
  static const __m64 maskc0 = { 0xfffffff8fffffffe };
  static const __m64 maskh  = { 0x0000000400000001 };
  static const __m64 maskc3 = { 0x0000000800000002 };
  register __m64 mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7;
  __m64 c0c,c3c,halfc;

  mm7 = _mm_setzero_si64();
  mm0 = _mm_cvtsi32_si64(color[0].v);
  mm0 = _mm_unpacklo_pi8(mm0, mm7);
  mm1 = _mm_cvtsi32_si64(color[1].v);
  mm1 = _mm_unpacklo_pi8(mm1, mm7);
  
  mm2 = mm0;
  mm0 = _mm_sub_pi16(mm0, mm1); // mm0=dir

  mm2 = _mm_madd_pi16(mm2, mm0); // dp0
  mm1 = _mm_madd_pi16(mm1, mm0); // dp1
  mm3 = mm2;
  mm2 = _mm_unpacklo_pi32(mm2, mm1);
  mm3 = _mm_unpackhi_pi32(mm3, mm1);
  mm2 = _mm_add_pi32(mm2, mm3); // mm2=stop01

  mm1 = _mm_cvtsi32_si64(color[2].v);
  mm1 = _mm_unpacklo_pi8(mm1, mm7);
  mm3 = _mm_cvtsi32_si64(color[3].v);
  mm3 = _mm_unpacklo_pi8(mm3, mm7);

  mm1 = _mm_madd_pi16(mm1, mm0); // dp0
  mm3 = _mm_madd_pi16(mm3, mm0); // dp1
  mm4 = mm1;
  mm1 = _mm_unpacklo_pi32(mm1, mm3);
  mm4 = _mm_unpackhi_pi32(mm4, mm3);
  mm1 = _mm_add_pi32(mm1, mm4); // mm2=stop23
  mm4 = mm1;

  mm2 = _mm_add_pi32(mm2, mm1);
  mm4 = _mm_unpackhi_pi32(mm4, mm4);
  mm2 = _mm_srai_pi32(mm2, 1); // mm2=c3c0p
  mm4 = _mm_add_pi32(mm4, mm1);
  mm1 = mm2;
  mm4 = _mm_srai_pi32(mm4, 1); // mm4=halfp

  mm1 = _mm_unpackhi_pi32(mm1, mm1); // mm1=c0c
  mm2 = _mm_unpacklo_pi32(mm2, mm2); // mm2=c3c
  mm4 = _mm_unpacklo_pi32(mm4, mm4); // mm4=halfc
  c0c = mm1;
  c3c = mm2;
  halfc = mm4;

  mm1 = _mm_setzero_si64();
  for(sInt i=14;i>=0;i-=2)
  {
    mm1 = _mm_slli_pi32(mm1, 4);

    mm2 = _mm_cvtsi32_si64(block[i+0].v);
    mm2 = _mm_unpacklo_pi8(mm2, mm7);
    mm3 = _mm_cvtsi32_si64(block[i+1].v);
    mm3 = _mm_unpacklo_pi8(mm3, mm7);

    mm2 = _mm_madd_pi16(mm2, mm0); // dp0
    mm3 = _mm_madd_pi16(mm3, mm0); // dp1
    mm4 = mm2;
    mm2 = _mm_unpacklo_pi32(mm2, mm3);
    mm4 = _mm_unpackhi_pi32(mm4, mm3);
    mm2 = _mm_add_pi32(mm2, mm4); // mm2=dots

    mm3 = c0c;
    mm4 = halfc;
    mm5 = c3c;
    mm3 = _mm_cmpgt_pi32(mm3, mm2);
    mm4 = _mm_cmpgt_pi32(mm4, mm2);
    mm5 = _mm_cmpgt_pi32(mm5, mm2);
    mm3 = _mm_and_si64(mm3, maskc0);
    mm4 = _mm_and_si64(mm4, maskh);
    mm5 = _mm_and_si64(mm5, maskc3);
    mm1 = _mm_add_pi32(mm1, mm3);
    mm1 = _mm_add_pi32(mm1, mm4);
    mm1 = _mm_add_pi32(mm1, mm5);
  }

  mm0 = mm1;
  mm0 = _mm_unpackhi_pi32(mm0, mm0);
  mm0 = _mm_or_si64(mm0, mm1);
  blockMask = _mm_cvtsi64_si32(mm0);
  _mm_empty();
}

// ----------------- generated code -----------------

;	COMDAT ?MatchColorsBlockIntrin2@@YAXPBTPixel@@0AAI@Z
_TEXT	SEGMENT
_halfc$ = -16						; size = 8
_c3c$ = -8						; size = 8
_blockMask$ = 8						; size = 4
?MatchColorsBlockIntrin2@@YAXPBTPixel@@0AAI@Z PROC	; MatchColorsBlockIntrin2, COMDAT
; _block$ = ecx
; _color$ = eax

; 252  : {

	push	ebp
	mov	ebp, esp
	and	esp, -8					; fffffff8H
	sub	esp, 16					; 00000010H

; 253  :   static const __m64 maskc0 = { 0xfffffff8fffffffe };
; 254  :   static const __m64 maskh  = { 0x0000000400000001 };
; 255  :   static const __m64 maskc3 = { 0x0000000800000002 };
; 256  :   register __m64 mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7;
; 257  :   __m64 c0c,c3c,halfc;
; 258  : 
; 259  :   mm7 = _mm_setzero_si64();
; 260  :   mm0 = _mm_cvtsi32_si64(color[0].v);

	mov	edx, DWORD PTR [eax]
	pxor	mm0, mm0
	movd	mm1, edx

; 261  :   mm0 = _mm_unpacklo_pi8(mm0, mm7);
; 262  :   mm1 = _mm_cvtsi32_si64(color[1].v);

	mov	edx, DWORD PTR [eax+4]
	movq	mm2, mm0
	punpcklbw mm1, mm2
	movd	mm2, edx

; 263  :   mm1 = _mm_unpacklo_pi8(mm1, mm7);
; 264  :   
; 265  :   mm2 = mm0;
; 266  :   mm0 = _mm_sub_pi16(mm0, mm1); // mm0=dir
; 267  : 
; 268  :   mm2 = _mm_madd_pi16(mm2, mm0); // dp0
; 269  :   mm1 = _mm_madd_pi16(mm1, mm0); // dp1
; 270  :   mm3 = mm2;
; 271  :   mm2 = _mm_unpacklo_pi32(mm2, mm1);
; 272  :   mm3 = _mm_unpackhi_pi32(mm3, mm1);
; 273  :   mm2 = _mm_add_pi32(mm2, mm3); // mm2=stop01
; 274  : 
; 275  :   mm1 = _mm_cvtsi32_si64(color[2].v);

	mov	edx, DWORD PTR [eax+8]

; 276  :   mm1 = _mm_unpacklo_pi8(mm1, mm7);
; 277  :   mm3 = _mm_cvtsi32_si64(color[3].v);

	mov	eax, DWORD PTR [eax+12]
	movq	mm3, mm0
	punpcklbw mm2, mm3
	movq	mm3, mm1
	movq	mm4, mm2
	psubw	mm1, mm4
	movq	mm4, mm1
	pmaddwd	mm3, mm4
	pmaddwd	mm2, mm4
	movq	mm4, mm3
	movq	mm5, mm2
	punpckldq mm4, mm5
	punpckhdq mm3, mm2
	paddd	mm4, mm3
	movq	mm3, mm0
	movd	mm2, edx
	punpcklbw mm2, mm3

; 278  :   mm3 = _mm_unpacklo_pi8(mm3, mm7);
; 279  : 
; 280  :   mm1 = _mm_madd_pi16(mm1, mm0); // dp0

	movq	mm3, mm1
	pmaddwd	mm2, mm3
	movd	mm3, eax
	movq	mm5, mm0
	punpcklbw mm3, mm5

; 281  :   mm3 = _mm_madd_pi16(mm3, mm0); // dp1

	movq	mm5, mm1
	pmaddwd	mm3, mm5

; 282  :   mm4 = mm1;
; 283  :   mm1 = _mm_unpacklo_pi32(mm1, mm3);

	movq	mm5, mm2
	movq	mm6, mm3
	punpckldq mm5, mm6

; 284  :   mm4 = _mm_unpackhi_pi32(mm4, mm3);

	punpckhdq mm2, mm3

; 285  :   mm1 = _mm_add_pi32(mm1, mm4); // mm2=stop23

	paddd	mm5, mm2
	movq	mm2, mm5

; 286  :   mm4 = mm1;
; 287  : 
; 288  :   mm2 = _mm_add_pi32(mm2, mm1);

	movq	mm3, mm4
	movq	mm4, mm2
	paddd	mm3, mm4

; 289  :   mm4 = _mm_unpackhi_pi32(mm4, mm4);
; 290  :   mm2 = _mm_srai_pi32(mm2, 1); // mm2=c3c0p

	psrad	mm3, 1
	punpckhdq mm4, mm5

; 291  :   mm4 = _mm_add_pi32(mm4, mm1);

	paddd	mm4, mm2
	movq	mm2, mm4

; 292  :   mm1 = mm2;
; 293  :   mm4 = _mm_srai_pi32(mm4, 1); // mm4=halfp

	psrad	mm2, 1

; 294  : 
; 295  :   mm1 = _mm_unpackhi_pi32(mm1, mm1); // mm1=c0c

	movq	mm4, mm3
	movq	mm5, mm3
	punpckhdq mm4, mm5

; 296  :   mm2 = _mm_unpacklo_pi32(mm2, mm2); // mm2=c3c
; 297  :   mm4 = _mm_unpacklo_pi32(mm4, mm4); // mm4=halfc
; 298  :   c0c = mm1;

	movq	mm5, mm4
	movq	mm4, mm3
	punpckldq mm4, mm3

; 299  :   c3c = mm2;

	movq	MMWORD PTR _c3c$[esp+16], mm4
	movq	mm3, mm2
	punpckldq mm3, mm2

; 300  :   halfc = mm4;

	movq	MMWORD PTR _halfc$[esp+16], mm3

; 301  : 
; 302  :   mm1 = _mm_setzero_si64();

	movq	mm4, mm0

; 303  :   for(sInt i=14;i>=0;i-=2)

	mov	eax, 14					; 0000000eH
	npad	3
$LL3@MatchColor:
	sub	eax, 2

; 304  :   {
; 305  :     mm1 = _mm_slli_pi32(mm1, 4);
; 306  : 
; 307  :     mm2 = _mm_cvtsi32_si64(block[i+0].v);

	mov	edx, DWORD PTR [ecx+eax*4+8]
	movd	mm2, edx

; 308  :     mm2 = _mm_unpacklo_pi8(mm2, mm7);
; 309  :     mm3 = _mm_cvtsi32_si64(block[i+1].v);

	mov	edx, DWORD PTR [ecx+eax*4+12]
	movq	mm3, mm0
	punpcklbw mm2, mm3

; 310  :     mm3 = _mm_unpacklo_pi8(mm3, mm7);
; 311  : 
; 312  :     mm2 = _mm_madd_pi16(mm2, mm0); // dp0

	movq	mm3, mm1
	pmaddwd	mm2, mm3
	movq	mm6, mm0
	movd	mm3, edx
	punpcklbw mm3, mm6

; 313  :     mm3 = _mm_madd_pi16(mm3, mm0); // dp1

	movq	mm6, mm1
	pmaddwd	mm3, mm6

; 314  :     mm4 = mm2;
; 315  :     mm2 = _mm_unpacklo_pi32(mm2, mm3);

	movq	mm6, mm2
	movq	mm7, mm3
	punpckldq mm6, mm7

; 316  :     mm4 = _mm_unpackhi_pi32(mm4, mm3);

	punpckhdq mm2, mm3

; 317  :     mm2 = _mm_add_pi32(mm2, mm4); // mm2=dots

	paddd	mm6, mm2
	movq	mm2, mm6
	movq	mm3, mm4
	pslld	mm3, 4

; 318  : 
; 319  :     mm3 = c0c;
; 320  :     mm4 = halfc;
; 321  :     mm5 = c3c;
; 322  :     mm3 = _mm_cmpgt_pi32(mm3, mm2);

	movq	mm4, mm5
	pcmpgtd	mm4, mm6

; 323  :     mm4 = _mm_cmpgt_pi32(mm4, mm2);
; 324  :     mm5 = _mm_cmpgt_pi32(mm5, mm2);
; 325  :     mm3 = _mm_and_si64(mm3, maskc0);

	movq	mm6, MMWORD PTR ?maskc0@?1??MatchColorsBlockIntrin2@@YAXPBTPixel@@0AAI@Z@4T__m64@@B
	pand	mm4, mm6

; 326  :     mm4 = _mm_and_si64(mm4, maskh);
; 327  :     mm5 = _mm_and_si64(mm5, maskc3);
; 328  :     mm1 = _mm_add_pi32(mm1, mm3);

	paddd	mm3, mm4
	movq	mm4, MMWORD PTR _halfc$[esp+16]
	movq	mm6, mm2
	pcmpgtd	mm4, mm6
	movq	mm6, MMWORD PTR ?maskh@?1??MatchColorsBlockIntrin2@@YAXPBTPixel@@0AAI@Z@4T__m64@@B
	pand	mm4, mm6

; 329  :     mm1 = _mm_add_pi32(mm1, mm4);

	paddd	mm3, mm4
	movq	mm4, MMWORD PTR _c3c$[esp+16]
	pcmpgtd	mm4, mm2
	movq	mm2, mm4
	movq	mm4, MMWORD PTR ?maskc3@?1??MatchColorsBlockIntrin2@@YAXPBTPixel@@0AAI@Z@4T__m64@@B
	pand	mm2, mm4

; 330  :     mm1 = _mm_add_pi32(mm1, mm5);

	paddd	mm3, mm2
	movq	mm4, mm3
	jns	$LL3@MatchColor

; 331  :   }
; 332  : 
; 333  :   mm0 = mm1;
; 334  :   mm0 = _mm_unpackhi_pi32(mm0, mm0);
; 335  :   mm0 = _mm_or_si64(mm0, mm1);
; 336  :   blockMask = _mm_cvtsi64_si32(mm0);

	mov	ecx, DWORD PTR _blockMask$[ebp]
	movq	mm0, mm4
	movq	mm1, mm4
	punpckhdq mm0, mm1
	por	mm0, mm1
	movd	eax, mm0
	mov	DWORD PTR [ecx], eax

; 337  :   _mm_empty();

	emms

; 338  : }

	mov	esp, ebp
	pop	ebp
	ret	0
?MatchColorsBlockIntrin2@@YAXPBTPixel@@0AAI@Z ENDP	; MatchColorsBlockIntrin2
_TEXT	ENDS