jdmerge.c

jpeg编解码器

	  punpcklbw mm0, mm0		; Cr3 Cr3 Cr2 Cr2 Cr1 Cr1 Cr0 Cr0

	  movq mm7, const128

	  punpcklwd mm0, mm0		; Cr1 Cr1 Cr1 Cr1 Cr0 Cr0 Cr0 Cr0

	  movq mm4, mm0

	  punpcklbw mm0, mm6		; Cr0 Cr0 Cr0 Cr0

	  psubsw mm0, mm7			; Cr0 - 128:Cr0-128:Cr0-128:Cr0 -128
	  
	  movd mm1, [ecx]			; Cb7 Cb6...... Cb1 Cb0
	  	   
	  psllw mm0, 2				; left shift by 2 bits

	  punpcklbw mm1, mm1		; Cb3 Cb3 Cb2 Cb2 Cb1 Cb1 Cb0 Cb0
	  
	  paddsw mm0, const05		; add (one_half/fix(x)) << 2

	  punpcklwd mm1, mm1		; Cb1 Cb1 Cb1 Cb1 Cb0 Cb0 Cb0 Cb0

	  movq mm5, mm1

	  pmulhw mm0, const1		; multiply by (fix(x) >> 1) 

	  punpcklbw mm1, mm6		; Cb0 Cb0 Cb0 Cb0

	  punpckhbw mm4, mm6		; Cr1 Cr1 Cr1 Cr1

	  psubsw mm1, mm7			; Cb0 - 128:Cb0-128:Cb0-128:Cb0 -128

	  punpckhbw mm5, mm6		; Cb1 Cb1 Cb1 Cb1

	  psllw mm1, 2				; left shift by 2 bits
 
	  paddsw mm1, const15		; add (one_half/fix(x)) << 2

	  psubsw mm4, mm7			; Cr1 - 128:Cr1-128:Cr1-128:Cr1 -128
						
	  psubsw mm5, mm7			; Cb1 - 128:Cb1-128:Cb1-128:Cb1 -128

	  pmulhw mm1, const2		; multiply by (fix(x) >> 1) 

	  psllw mm4, 2				; left shift by 2 bits

	  psllw mm5, 2				; left shift by 2 bits

	  paddsw mm4, const45		; add (one_half/fix(x)) << 2

	  movd mm7, [esi]			;  Y13 Y12 Y9 Y8 Y5 Y4 Y1 Y0

	  pmulhw mm4, const5		; multiply by (fix(x) >> 1) 

	  movq mm6, mm7

	  punpcklbw mm7, mm7		; Y5 Y5 Y4 Y4 Y1 Y1 Y0 Y0

	  paddsw mm5, const55		; add (one_half/fix(x)) << 2

	  paddsw  mm0, mm1			; cred0 cbl0 cgr0 cred0

	  movq mm1, mm7

	  pmulhw mm5, const6		; multiply by (fix(x) >> 1) 

	  movq	mm2, mm0			; cred0 cbl0 cgr0 cred0

	  punpcklwd mm7, mm6		; Y5 Y4 Y1 Y1 Y1 Y0 Y0 Y0

	  pand mm2, davemask		; 0 cbl0 cgr0 0

	  psrlq mm1, 16				; 0 0 Y5 Y5 Y4 Y4 Y1 Y1

	  psrlq	mm2, 16				; 0 0 cbl0 cgr0

	  punpcklbw mm7, empty		; Y1 Y0 Y0 Y0

	  paddsw mm4, mm5			; cbl1 cgr1 cred1 cbl1

	  movq	mm3, mm4			; cbl1 cgr1 cred1 cbl1

	  pand	mm3, davemask		; 0 cgr1 cred1 0

	  paddsw mm7, mm0			; r1 b0 g0 r0

	  psllq	mm3, 16				; cgr1 cred1 0 0

	  movq mm6, mm1				; 0 0 Y5 Y5 Y4 Y4 Y1 Y1
	
	  por	mm2, mm3			; cgr1 cred1 cbl0 cgr0

	  punpcklbw mm6, empty		; Y4 Y4 Y1 Y1

	  movd mm3, [eax]			; Y15 Y14 Y11 Y10 Y7 Y6 Y3 Y2
	  
	  paddsw mm6, mm2			; g4 r4 b1 g1

	  packuswb mm7, mm6			; g4 r4 b1 g1 r1 b0 g0 r0

	  movq mm6, mm3				; Y15 Y14 Y11 Y10 Y7 Y6 Y3 Y2

	  punpcklbw mm3, mm3		; Y7 Y7 Y6 Y6 Y3 Y3 Y2 Y2

	  movq [edi], mm7			; move to memory g4 r4 b1 g1 r1 b0 g0 r0

	  movq mm5, mm3				; Y7 Y7 Y6 Y6 Y3 Y3 Y2 Y2

	  punpcklwd mm3, mm6		; X X X X Y3 Y2 Y2 Y2

	  punpcklbw mm3, empty		; Y3 Y2 Y2 Y2

	  psrlq mm5, 16				; 0 0 Y7 Y7 Y6 Y6 Y3 Y3

	  paddsw mm3, mm0			; r3 b2 g2 r2

	  movq mm6, mm5				; 0 0 Y7 Y7 Y6 Y6 Y3 Y3

	  movq mm0, mm1				; 0 0 Y5 Y5 Y4 Y4 Y1 Y1

	  punpckldq mm6, mm6		; X X X X Y6 Y6 Y3 Y3

	  punpcklbw mm6, empty		; Y6 Y6 Y3 Y3

	  psrlq mm1, 24				; 0 0 0 0 0 Y5 Y5 Y4
	  
	  paddsw mm6, mm2			; g6 r6 b3 g3

	  packuswb mm3, mm6			; g6 r6 b3 g3 r3 b2 g2 r2

	  movq mm2, mm5				; 0 0 Y7 Y7 Y6 Y6 Y3 Y3

	  psrlq mm0, 32				; 0 0 0 0 0 0 Y5 Y5

	  movq [edx], mm3			; move to memory g6 r6 b3 g3 r3 b2 g2 r2
	  
	  punpcklwd mm1, mm0		; X X X X Y5 Y5 Y5 Y4

	  psrlq mm5, 24				; 0 0 0 0 0 Y7 Y7 Y6 

	  movd mm0, [ebx]			; Cr9 Cr8.....Cr3 Cr2

	  psrlq mm2, 32	   			; 0 0 0 0 0 0 Y7 Y7	 
	  
	  psrlq	mm0, 16		

	  punpcklbw mm1, empty		; Y5 Y5 Y5 Y4

	  punpcklwd mm5, mm2		; X X X X Y7 Y7 Y7 Y6

	  paddsw mm1, mm4			; b5 g5 r5 b4
	 
	  punpcklbw mm5, empty		; Y7 Y7 Y7 Y6	    

	  pxor mm6, mm6				; clear mm6 registr
	  
	  punpcklbw mm0, mm0		; X X X X Cr3 Cr3 Cr2 Cr2
  
	  paddsw mm5, mm4			; b7 g7 r7 b6
	  
	  punpcklwd mm0, mm0		; Cr3 Cr3 Cr3 Cr3 Cr2 Cr2 Cr2 Cr2

	  movq mm4, mm0

	  movd mm3, [ecx]			; Cb9 Cb8...... Cb3 Cb2
	  
	  punpcklbw mm0, mm6		; Cr2 Cr2 Cr2 Cr2

	  psrlq	mm3, 16

	  psubsw mm0, const128		; Cr2 - 128:Cr2-128:Cr2-128:Cr2 -128

	  punpcklbw mm3, mm3		; X X X X Cb3 Cb3 Cb2 Cb2

	  psllw mm0, 2				; left shift by 2 bits

	  paddsw mm0, const05		; add (one_half/fix(x)) << 2

	  punpcklwd mm3, mm3		; Cb3 Cb3 Cb3 Cb3 Cb2 Cb2 Cb2 Cb2

	  movq mm7, mm3
	  
	  pmulhw mm0, const1		; multiply by (fix(x) >> 1) 	  	  

	  punpcklbw mm3, mm6		; Cb2 Cb2 Cb2 Cb2

	  psubsw mm3, const128		; Cb0 - 128:Cb0-128:Cb0-128:Cb0 -128

	  punpckhbw mm4, mm6		; Cr3 Cr3 Cr3 Cr3
	  
	  psllw mm3, 2				; left shift by 2 bits

	  paddsw mm3, const15		; add (one_half/fix(x)) << 2

	  punpckhbw mm7, mm6		; Cb3 Cb3 Cb3 Cb3

	  pmulhw mm3, const2		; multiply by (fix(x) >> 1) 
	  
	  psubsw mm7, const128		; Cb3 - 128:Cb3-128:Cb3-128:Cb3 -128

	  paddsw  mm0, mm3			; cred2 cbl2 cgr2 cred2
	    
	  psllw mm7, 2				; left shift by 2 bits

	  psubsw mm4, const128		; Cr3 - 128:Cr3-128:Cr3-128:Cr3 -128
	  
	  movd mm3, [esi+4]			;  Y21 Y20 Y17 Y16 Y13 Y12 Y9 Y8
	  
	  psllw mm4, 2				; left shift by 2 bits

	  paddsw mm7, const55		; add (one_half/fix(x)) << 2
	  	  
	  movq mm6, mm3				;  Y21 Y20 Y17 Y16 Y13 Y12 Y9 Y8

	  movq	mm2, mm0
	  	  
	  pand mm2, davemask

	  punpcklbw mm3, mm3		; Y13 Y13 Y12 Y12 Y9 Y9 Y8 Y8

	  psrlq	mm2, 16
	    	  
	  paddsw mm4, const45		; add (one_half/fix(x)) << 2

	  punpcklwd mm3, mm6		; X X X X Y9 Y8 Y8 Y8
	  
	  pmulhw mm4, const5		; multiply by (fix(x) >> 1) 

	  pmulhw mm7, const6		; multiply by (fix(x) >> 1) 

	  punpcklbw mm3, empty		; Y9 Y8 Y8 Y8
	  
	  paddsw mm4, mm7			; cbl3 cgr3 cred3 cbl3

	  paddsw mm3, mm0			; r9 b8 g8 r8

	  movq	mm7, mm4

	  packuswb mm1, mm3			; r9 b8 g8 r8 b5 g5 r5 b4

	  movd mm3, [eax+4]			; Y23 Y22 Y19 Y18 Y15 Y14 Y11 Y10
 	  
	  pand	mm7, davemask

	  psrlq mm6, 8				; 0 Y21 Y20 Y17 Y16 Y13 Y12 Y9

	  psllq	mm7, 16
						   
	  movq [edi+8], mm1			; move to memory r9 b8 g8 r8 b5 g5 r5 b4

	  por	mm2, mm7

	  movq mm7, mm3				; Y23 Y22 Y19 Y18 Y15 Y14 Y11 Y10

	  punpcklbw mm3, mm3		; X X X X Y11 Y11 Y10 Y10

	  pxor mm1, mm1

	  punpcklwd mm3, mm7		; X X X X Y11 Y10 Y10 Y10

	  punpcklbw mm3, mm1		; Y11 Y10 Y10 Y10

	  psrlq mm7, 8				; 0 Y23 Y22 Y19 Y18 Y15 Y14 Y11
	  
	  paddsw mm3, mm0			; r11 b10 g10 r10

	  movq mm0, mm7				; 0 Y23 Y22 Y19 Y18 Y15 Y14 Y11

	  packuswb mm5, mm3			; r11 b10 g10 r10 b7 g7 r7 b6

	  punpcklbw mm7, mm7		; X X X X Y14 Y14 Y11 Y11

	  movq [edx+8], mm5			; move to memory r11 b10 g10 r10 b7 g7 r7 b6

	  movq mm3, mm6				; 0 Y21 Y20 Y17 Y16 Y13 Y12 Y9

	  punpcklbw mm6, mm6		; X X X X Y12 Y12 Y9 Y9

	  punpcklbw mm7, mm1		; Y14 Y14 Y11 Y11

	  punpcklbw mm6, mm1		; Y12 Y12 Y9 Y9

	  paddsw mm7, mm2			; g14 r14 b11 g11

	  paddsw mm6, mm2			; g12 r12 b9 g9

	  psrlq mm3, 8				; 0 0 Y21 Y20 Y17 Y16 Y13 Y12

	  movq mm1, mm3				; 0 0 Y21 Y20 Y17 Y16 Y13 Y12

	  punpcklbw mm3, mm3		; X X X X Y13 Y13 Y12 Y12

	  add esi, 8

	  psrlq mm3, 16				; X X X X X X Y13 Y13 modified on 09/24

	  punpcklwd mm1, mm3		; X X X X Y13 Y13 Y13 Y12

	  add eax, 8

	  psrlq mm0, 8				; 0 0 Y23 Y22 Y19 Y18 Y15 Y14	

	  punpcklbw mm1, empty		; Y13 Y13 Y13 Y12

	  movq mm5, mm0				; 0 0 Y23 Y22 Y19 Y18 Y15 Y14	

	  punpcklbw mm0, mm0		; X X X X Y15 Y15 Y14 Y14

	  paddsw mm1, mm4			; b13 g13 r13 b12

	  psrlq mm0, 16				; X X X X X X Y15 Y15

	  add edi, 24
	  
	  punpcklwd mm5, mm0		; X X X X Y15 Y15 Y15 Y14

	  packuswb mm6, mm1			; b13 g13 r13 b12 g12 r12 b9 g9

	  add edx, 24
	  
	  punpcklbw mm5, empty		; Y15 Y15 Y15 Y14

	  add ebx, 4
	  	  
	  paddsw mm5, mm4			; b15 g15 r15 b14

	  movq [edi-8], mm6		; move to memory b13 g13 r13 b12 g12 r12 b9 g9

	  packuswb mm7, mm5			; b15 g15 r15 b14 g14 r14 b11 g11

	  add ecx, 4
  
	  movq [edx-8], mm7		; move to memory b15 g15 r15 b14 g14 r14 b11 g11

	  dec cols_asm
	  
	  jnz do_next16

	  EMMS
	  	  
	  }
#endif
#ifdef HAVE_MMX_ATT_MNEMONICS
  fprintf(stderr, "Using accelerated MMX code for merge !\n");

          __asm__ (
          "pushl %%ebx            \n\t"

          "movl %0, %%esi         \n\t"

          "movl %1, %%eax         \n\t"

          "movl %2, %%ebx          \n\t"

          "movl %3, %%ecx          \n\t"

          "movl %4, %%edi         \n\t"

          "movl %5, %%edx         \n\t"

	  "do_next16:                  \n\t"
	  
          "movd (%%ebx),%%mm0            \n\t"    // Cr7 Cr6.....Cr1 Cr0

          "pxor %%mm6,%%mm6              \n\t"

          "punpcklbw %%mm0,%%mm0         \n\t" // Cr3 Cr3 Cr2 Cr2 Cr1 Cr1 Cr0 Cr0

          "movq const128,%%mm7         \n\t"

          "punpcklwd %%mm0,%%mm0         \n\t" // Cr1 Cr1 Cr1 Cr1 Cr0 Cr0 Cr0 Cr0

          "movq %%mm0,%%mm4              \n\t"

          "punpcklbw %%mm6,%%mm0         \n\t" // Cr0 Cr0 Cr0 Cr0

          "psubsw %%mm7,%%mm0            \n\t"    // Cr0 - 128:Cr0-128:Cr0-128:Cr0 -128

          "movd (%%ecx),%%mm1            \n\t"    // Cb7 Cb6...... Cb1 Cb0

          "psllw $2,%%mm0               \n\t"    // left shift by 2 bits

          "punpcklbw %%mm1,%%mm1         \n\t" // Cb3 Cb3 Cb2 Cb2 Cb1 Cb1 Cb0 Cb0

          "paddsw const05,%%mm0        \n\t" // add (one_half/fix(x)) << 2

          "punpcklwd %%mm1,%%mm1         \n\t" // Cb1 Cb1 Cb1 Cb1 Cb0 Cb0 Cb0 Cb0

          "movq %%mm1,%%mm5              \n\t"

          "pmulhw const1,%%mm0         \n\t" // multiply by (fix(x) >> 1) 

          "punpcklbw %%mm6,%%mm1         \n\t" // Cb0 Cb0 Cb0 Cb0

          "punpckhbw %%mm6,%%mm4         \n\t" // Cr1 Cr1 Cr1 Cr1

          "psubsw %%mm7,%%mm1            \n\t"    // Cb0 - 128:Cb0-128:Cb0-128:Cb0 -128

          "punpckhbw %%mm6,%%mm5         \n\t" // Cb1 Cb1 Cb1 Cb1

          "psllw $2,%%mm1               \n\t"    // left shift by 2 bits

          "paddsw const15,%%mm1        \n\t" // add (one_half/fix(x)) << 2

          "psubsw %%mm7,%%mm4            \n\t"    // Cr1 - 128:Cr1-128:Cr1-128:Cr1 -128

          "psubsw %%mm7,%%mm5            \n\t"    // Cb1 - 128:Cb1-128:Cb1-128:Cb1 -128

          "pmulhw const2,%%mm1         \n\t" // multiply by (fix(x) >> 1) 

          "psllw $2,%%mm4               \n\t"    // left shift by 2 bits

          "psllw $2,%%mm5               \n\t"    // left shift by 2 bits

          "paddsw const45,%%mm4        \n\t" // add (one_half/fix(x)) << 2

          "movd (%%esi),%%mm7            \n\t"    //  Y13 Y12 Y9 Y8 Y5 Y4 Y1 Y0

          "pmulhw const5,%%mm4         \n\t" // multiply by (fix(x) >> 1) 

          "movq %%mm7,%%mm6              \n\t"

          "punpcklbw %%mm7,%%mm7         \n\t" // Y5 Y5 Y4 Y4 Y1 Y1 Y0 Y0

          "paddsw const55,%%mm5        \n\t" // add (one_half/fix(x)) << 2

          "paddsw  %%mm1,%%mm0           \n\t"    // cred0 cbl0 cgr0 cred0

          "movq %%mm7,%%mm1              \n\t"

          "pmulhw const6,%%mm5         \n\t" // multiply by (fix(x) >> 1) 

          "movq  %%mm0,%%mm2             \n\t"    // cred0 cbl0 cgr0 cred0

          "punpcklwd %%mm6,%%mm7         \n\t" // Y5 Y4 Y1 Y1 Y1 Y0 Y0 Y0

          "pand davemask,%%mm2         \n\t" // 0 cbl0 cgr0 0

          "psrlq $16,%%mm1              \n\t"    // 0 0 Y5 Y5 Y4 Y4 Y1 Y1

          "psrlq $16,%%mm2              \n\t"    // 0 0 cbl0 cgr0

          "punpcklbw empty,%%mm7       \n\t" // Y1 Y0 Y0 Y0

          "paddsw %%mm5,%%mm4            \n\t"    // cbl1 cgr1 cred1 cbl1

          "movq  %%mm4,%%mm3             \n\t"    // cbl1 cgr1 cred1 cbl1

          "pand  davemask,%%mm3        \n\t" // 0 cgr1 cred1 0

          "paddsw %%mm0,%%mm7            \n\t"    // r1 b0 g0 r0

          "psllq $16,%%mm3              \n\t"    // cgr1 cred1 0 0

          "movq %%mm1,%%mm6              \n\t"    // 0 0 Y5 Y5 Y4 Y4 Y1 Y1

          "por   %%mm3,%%mm2             \n\t"    // cgr1 cred1 cbl0 cgr0

          "punpcklbw empty,%%mm6       \n\t" // Y4 Y4 Y1 Y1

          "movd (%%eax),%%mm3            \n\t"    // Y15 Y14 Y11 Y10 Y7 Y6 Y3 Y2

          "paddsw %%mm2,%%mm6            \n\t"    // g4 r4 b1 g1

          "packuswb %%mm6,%%mm7          \n\t"    // g4 r4 b1 g1 r1 b0 g0 r0

          "movq %%mm3,%%mm6              \n\t"    // Y15 Y14 Y11 Y10 Y7 Y6 Y3 Y2

          "punpcklbw %%mm3,%%mm3         \n\t" // Y7 Y7 Y6 Y6 Y3 Y3 Y2 Y2

          "movq %%mm7,(%%edi)            \n\t"    // move to memory g4 r4 b1 g1 r1 b0 g0 r0