Browse code
temp denoiser changes: (a-b)^2 instead of |a-b| and MMX2/3DNOW version
Originally committed as revision 2896 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc
Michael Niedermayer authored on 2001/11/14 11:46:58Showing 2 changed files
| ... | ... |
@@ -35,7 +35,7 @@ LinIpolDeinterlace e E E* |
| 35 | 35 |
CubicIpolDeinterlace a e e* |
| 36 | 36 |
LinBlendDeinterlace e E E* |
| 37 | 37 |
MedianDeinterlace# Ec Ec |
| 38 |
-TempDeNoiser# a |
|
| 38 |
+TempDeNoiser# E e e |
|
| 39 | 39 |
|
| 40 | 40 |
* i dont have a 3dnow CPU -> its untested, but noone said it doesnt work so it seems to work |
| 41 | 41 |
# more or less selfinvented filters so the exactness isnt too meaningfull |
| ... | ... |
@@ -61,6 +61,7 @@ split this huge file |
| 61 | 61 |
border remover |
| 62 | 62 |
optimize c versions |
| 63 | 63 |
try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks |
| 64 |
+smart blur |
|
| 64 | 65 |
... |
| 65 | 66 |
|
| 66 | 67 |
Notes: |
| ... | ... |
@@ -2592,10 +2593,294 @@ static inline void transpose2(uint8_t *dst, int dstStride, uint8_t *src) |
| 2592 | 2592 |
); |
| 2593 | 2593 |
} |
| 2594 | 2594 |
#endif |
| 2595 |
+//static int test=0; |
|
| 2595 | 2596 |
|
| 2596 | 2597 |
static void inline tempNoiseReducer(uint8_t *src, int stride, |
| 2597 | 2598 |
uint8_t *tempBlured, int *maxNoise) |
| 2598 | 2599 |
{
|
| 2600 |
+#define FAST_L2_DIFF |
|
| 2601 |
+//#define L1_DIFF //u should change the thresholds too if u try that one |
|
| 2602 |
+#if defined (HAVE_MMX2) || defined (HAVE_3DNOW) |
|
| 2603 |
+ asm volatile( |
|
| 2604 |
+ "leal (%2, %2, 2), %%eax \n\t" // 3*stride |
|
| 2605 |
+ "leal (%2, %2, 4), %%ebx \n\t" // 5*stride |
|
| 2606 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2607 |
+// 0 1 2 3 4 5 6 7 8 9 |
|
| 2608 |
+// %x %x+%2 %x+2%2 %x+eax %x+4%2 %x+ebx %x+2eax %x+ecx %x+8%2 |
|
| 2609 |
+//FIXME reorder? |
|
| 2610 |
+#ifdef L1_DIFF //needs mmx2 |
|
| 2611 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2612 |
+ "psadbw (%1), %%mm0 \n\t" // |L0-R0| |
|
| 2613 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2614 |
+ "psadbw (%1, %2), %%mm1 \n\t" // |L1-R1| |
|
| 2615 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2616 |
+ "psadbw (%1, %2, 2), %%mm2 \n\t" // |L2-R2| |
|
| 2617 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2618 |
+ "psadbw (%1, %%eax), %%mm3 \n\t" // |L3-R3| |
|
| 2619 |
+ |
|
| 2620 |
+ "movq (%0, %2, 4), %%mm4 \n\t" // L4 |
|
| 2621 |
+ "paddw %%mm1, %%mm0 \n\t" |
|
| 2622 |
+ "psadbw (%1, %2, 4), %%mm4 \n\t" // |L4-R4| |
|
| 2623 |
+ "movq (%0, %%ebx), %%mm5 \n\t" // L5 |
|
| 2624 |
+ "paddw %%mm2, %%mm0 \n\t" |
|
| 2625 |
+ "psadbw (%1, %%ebx), %%mm5 \n\t" // |L5-R5| |
|
| 2626 |
+ "movq (%0, %%eax, 2), %%mm6 \n\t" // L6 |
|
| 2627 |
+ "paddw %%mm3, %%mm0 \n\t" |
|
| 2628 |
+ "psadbw (%1, %%eax, 2), %%mm6 \n\t" // |L6-R6| |
|
| 2629 |
+ "movq (%0, %%ecx), %%mm7 \n\t" // L7 |
|
| 2630 |
+ "paddw %%mm4, %%mm0 \n\t" |
|
| 2631 |
+ "psadbw (%1, %%ecx), %%mm7 \n\t" // |L7-R7| |
|
| 2632 |
+ "paddw %%mm5, %%mm6 \n\t" |
|
| 2633 |
+ "paddw %%mm7, %%mm6 \n\t" |
|
| 2634 |
+ "paddw %%mm6, %%mm0 \n\t" |
|
| 2635 |
+#elif defined (FAST_L2_DIFF) |
|
| 2636 |
+ "pcmpeqb %%mm7, %%mm7 \n\t" |
|
| 2637 |
+ "movq b80, %%mm6 \n\t" |
|
| 2638 |
+ "pxor %%mm0, %%mm0 \n\t" |
|
| 2639 |
+#define L2_DIFF_CORE(a, b)\ |
|
| 2640 |
+ "movq " #a ", %%mm5 \n\t"\ |
|
| 2641 |
+ "movq " #b ", %%mm2 \n\t"\ |
|
| 2642 |
+ "pxor %%mm7, %%mm2 \n\t"\ |
|
| 2643 |
+ PAVGB(%%mm2, %%mm5)\ |
|
| 2644 |
+ "paddb %%mm6, %%mm5 \n\t"\ |
|
| 2645 |
+ "movq %%mm5, %%mm2 \n\t"\ |
|
| 2646 |
+ "psllw $8, %%mm5 \n\t"\ |
|
| 2647 |
+ "pmaddwd %%mm5, %%mm5 \n\t"\ |
|
| 2648 |
+ "pmaddwd %%mm2, %%mm2 \n\t"\ |
|
| 2649 |
+ "paddd %%mm2, %%mm5 \n\t"\ |
|
| 2650 |
+ "psrld $14, %%mm5 \n\t"\ |
|
| 2651 |
+ "paddd %%mm5, %%mm0 \n\t" |
|
| 2652 |
+ |
|
| 2653 |
+L2_DIFF_CORE((%0), (%1)) |
|
| 2654 |
+L2_DIFF_CORE((%0, %2), (%1, %2)) |
|
| 2655 |
+L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2)) |
|
| 2656 |
+L2_DIFF_CORE((%0, %%eax), (%1, %%eax)) |
|
| 2657 |
+L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4)) |
|
| 2658 |
+L2_DIFF_CORE((%0, %%ebx), (%1, %%ebx)) |
|
| 2659 |
+L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2)) |
|
| 2660 |
+L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx)) |
|
| 2661 |
+ |
|
| 2662 |
+#else |
|
| 2663 |
+ "pxor %%mm7, %%mm7 \n\t" |
|
| 2664 |
+ "pxor %%mm0, %%mm0 \n\t" |
|
| 2665 |
+#define L2_DIFF_CORE(a, b)\ |
|
| 2666 |
+ "movq " #a ", %%mm5 \n\t"\ |
|
| 2667 |
+ "movq " #b ", %%mm2 \n\t"\ |
|
| 2668 |
+ "movq %%mm5, %%mm1 \n\t"\ |
|
| 2669 |
+ "movq %%mm2, %%mm3 \n\t"\ |
|
| 2670 |
+ "punpcklbw %%mm7, %%mm5 \n\t"\ |
|
| 2671 |
+ "punpckhbw %%mm7, %%mm1 \n\t"\ |
|
| 2672 |
+ "punpcklbw %%mm7, %%mm2 \n\t"\ |
|
| 2673 |
+ "punpckhbw %%mm7, %%mm3 \n\t"\ |
|
| 2674 |
+ "psubw %%mm2, %%mm5 \n\t"\ |
|
| 2675 |
+ "psubw %%mm3, %%mm1 \n\t"\ |
|
| 2676 |
+ "pmaddwd %%mm5, %%mm5 \n\t"\ |
|
| 2677 |
+ "pmaddwd %%mm1, %%mm1 \n\t"\ |
|
| 2678 |
+ "paddd %%mm1, %%mm5 \n\t"\ |
|
| 2679 |
+ "paddd %%mm5, %%mm0 \n\t" |
|
| 2680 |
+ |
|
| 2681 |
+L2_DIFF_CORE((%0), (%1)) |
|
| 2682 |
+L2_DIFF_CORE((%0, %2), (%1, %2)) |
|
| 2683 |
+L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2)) |
|
| 2684 |
+L2_DIFF_CORE((%0, %%eax), (%1, %%eax)) |
|
| 2685 |
+L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4)) |
|
| 2686 |
+L2_DIFF_CORE((%0, %%ebx), (%1, %%ebx)) |
|
| 2687 |
+L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2)) |
|
| 2688 |
+L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx)) |
|
| 2689 |
+ |
|
| 2690 |
+#endif |
|
| 2691 |
+ |
|
| 2692 |
+ "movq %%mm0, %%mm4 \n\t" |
|
| 2693 |
+ "psrlq $32, %%mm0 \n\t" |
|
| 2694 |
+ "paddd %%mm0, %%mm4 \n\t" |
|
| 2695 |
+ "movd %%mm4, %%ecx \n\t" |
|
| 2696 |
+// "movl %3, %%ecx \n\t" |
|
| 2697 |
+// "movl %%ecx, test \n\t" |
|
| 2698 |
+// "jmp 4f \n\t" |
|
| 2699 |
+ "cmpl %4, %%ecx \n\t" |
|
| 2700 |
+ " jb 2f \n\t" |
|
| 2701 |
+ "cmpl %5, %%ecx \n\t" |
|
| 2702 |
+ " jb 1f \n\t" |
|
| 2703 |
+ |
|
| 2704 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2705 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2706 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2707 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2708 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2709 |
+ "movq (%0, %2, 4), %%mm4 \n\t" // L4 |
|
| 2710 |
+ "movq (%0, %%ebx), %%mm5 \n\t" // L5 |
|
| 2711 |
+ "movq (%0, %%eax, 2), %%mm6 \n\t" // L6 |
|
| 2712 |
+ "movq (%0, %%ecx), %%mm7 \n\t" // L7 |
|
| 2713 |
+ "movq %%mm0, (%1) \n\t" // L0 |
|
| 2714 |
+ "movq %%mm1, (%1, %2) \n\t" // L1 |
|
| 2715 |
+ "movq %%mm2, (%1, %2, 2) \n\t" // L2 |
|
| 2716 |
+ "movq %%mm3, (%1, %%eax) \n\t" // L3 |
|
| 2717 |
+ "movq %%mm4, (%1, %2, 4) \n\t" // L4 |
|
| 2718 |
+ "movq %%mm5, (%1, %%ebx) \n\t" // L5 |
|
| 2719 |
+ "movq %%mm6, (%1, %%eax, 2) \n\t" // L6 |
|
| 2720 |
+ "movq %%mm7, (%1, %%ecx) \n\t" // L7 |
|
| 2721 |
+ "jmp 4f \n\t" |
|
| 2722 |
+ |
|
| 2723 |
+ "1: \n\t" |
|
| 2724 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2725 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2726 |
+ "pavgb (%1), %%mm0 \n\t" // L0 |
|
| 2727 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2728 |
+ "pavgb (%1, %2), %%mm1 \n\t" // L1 |
|
| 2729 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2730 |
+ "pavgb (%1, %2, 2), %%mm2 \n\t" // L2 |
|
| 2731 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2732 |
+ "pavgb (%1, %%eax), %%mm3 \n\t" // L3 |
|
| 2733 |
+ "movq (%0, %2, 4), %%mm4 \n\t" // L4 |
|
| 2734 |
+ "pavgb (%1, %2, 4), %%mm4 \n\t" // L4 |
|
| 2735 |
+ "movq (%0, %%ebx), %%mm5 \n\t" // L5 |
|
| 2736 |
+ "pavgb (%1, %%ebx), %%mm5 \n\t" // L5 |
|
| 2737 |
+ "movq (%0, %%eax, 2), %%mm6 \n\t" // L6 |
|
| 2738 |
+ "pavgb (%1, %%eax, 2), %%mm6 \n\t" // L6 |
|
| 2739 |
+ "movq (%0, %%ecx), %%mm7 \n\t" // L7 |
|
| 2740 |
+ "pavgb (%1, %%ecx), %%mm7 \n\t" // L7 |
|
| 2741 |
+ "movq %%mm0, (%1) \n\t" // R0 |
|
| 2742 |
+ "movq %%mm1, (%1, %2) \n\t" // R1 |
|
| 2743 |
+ "movq %%mm2, (%1, %2, 2) \n\t" // R2 |
|
| 2744 |
+ "movq %%mm3, (%1, %%eax) \n\t" // R3 |
|
| 2745 |
+ "movq %%mm4, (%1, %2, 4) \n\t" // R4 |
|
| 2746 |
+ "movq %%mm5, (%1, %%ebx) \n\t" // R5 |
|
| 2747 |
+ "movq %%mm6, (%1, %%eax, 2) \n\t" // R6 |
|
| 2748 |
+ "movq %%mm7, (%1, %%ecx) \n\t" // R7 |
|
| 2749 |
+ "movq %%mm0, (%0) \n\t" // L0 |
|
| 2750 |
+ "movq %%mm1, (%0, %2) \n\t" // L1 |
|
| 2751 |
+ "movq %%mm2, (%0, %2, 2) \n\t" // L2 |
|
| 2752 |
+ "movq %%mm3, (%0, %%eax) \n\t" // L3 |
|
| 2753 |
+ "movq %%mm4, (%0, %2, 4) \n\t" // L4 |
|
| 2754 |
+ "movq %%mm5, (%0, %%ebx) \n\t" // L5 |
|
| 2755 |
+ "movq %%mm6, (%0, %%eax, 2) \n\t" // L6 |
|
| 2756 |
+ "movq %%mm7, (%0, %%ecx) \n\t" // L7 |
|
| 2757 |
+ "jmp 4f \n\t" |
|
| 2758 |
+ |
|
| 2759 |
+ "2: \n\t" |
|
| 2760 |
+ "cmpl %3, %%ecx \n\t" |
|
| 2761 |
+ " jb 3f \n\t" |
|
| 2762 |
+ |
|
| 2763 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2764 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2765 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2766 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2767 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2768 |
+ "movq (%1), %%mm4 \n\t" // R0 |
|
| 2769 |
+ "movq (%1, %2), %%mm5 \n\t" // R1 |
|
| 2770 |
+ "movq (%1, %2, 2), %%mm6 \n\t" // R2 |
|
| 2771 |
+ "movq (%1, %%eax), %%mm7 \n\t" // R3 |
|
| 2772 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2773 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2774 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2775 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2776 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2777 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2778 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2779 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2780 |
+ "movq %%mm0, (%1) \n\t" // R0 |
|
| 2781 |
+ "movq %%mm1, (%1, %2) \n\t" // R1 |
|
| 2782 |
+ "movq %%mm2, (%1, %2, 2) \n\t" // R2 |
|
| 2783 |
+ "movq %%mm3, (%1, %%eax) \n\t" // R3 |
|
| 2784 |
+ "movq %%mm0, (%0) \n\t" // L0 |
|
| 2785 |
+ "movq %%mm1, (%0, %2) \n\t" // L1 |
|
| 2786 |
+ "movq %%mm2, (%0, %2, 2) \n\t" // L2 |
|
| 2787 |
+ "movq %%mm3, (%0, %%eax) \n\t" // L3 |
|
| 2788 |
+ |
|
| 2789 |
+ "movq (%0, %2, 4), %%mm0 \n\t" // L4 |
|
| 2790 |
+ "movq (%0, %%ebx), %%mm1 \n\t" // L5 |
|
| 2791 |
+ "movq (%0, %%eax, 2), %%mm2 \n\t" // L6 |
|
| 2792 |
+ "movq (%0, %%ecx), %%mm3 \n\t" // L7 |
|
| 2793 |
+ "movq (%1, %2, 4), %%mm4 \n\t" // R4 |
|
| 2794 |
+ "movq (%1, %%ebx), %%mm5 \n\t" // R5 |
|
| 2795 |
+ "movq (%1, %%eax, 2), %%mm6 \n\t" // R6 |
|
| 2796 |
+ "movq (%1, %%ecx), %%mm7 \n\t" // R7 |
|
| 2797 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2798 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2799 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2800 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2801 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2802 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2803 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2804 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2805 |
+ "movq %%mm0, (%1, %2, 4) \n\t" // R4 |
|
| 2806 |
+ "movq %%mm1, (%1, %%ebx) \n\t" // R5 |
|
| 2807 |
+ "movq %%mm2, (%1, %%eax, 2) \n\t" // R6 |
|
| 2808 |
+ "movq %%mm3, (%1, %%ecx) \n\t" // R7 |
|
| 2809 |
+ "movq %%mm0, (%0, %2, 4) \n\t" // L4 |
|
| 2810 |
+ "movq %%mm1, (%0, %%ebx) \n\t" // L5 |
|
| 2811 |
+ "movq %%mm2, (%0, %%eax, 2) \n\t" // L6 |
|
| 2812 |
+ "movq %%mm3, (%0, %%ecx) \n\t" // L7 |
|
| 2813 |
+ "jmp 4f \n\t" |
|
| 2814 |
+ |
|
| 2815 |
+ "3: \n\t" |
|
| 2816 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2817 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2818 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2819 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2820 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2821 |
+ "movq (%1), %%mm4 \n\t" // R0 |
|
| 2822 |
+ "movq (%1, %2), %%mm5 \n\t" // R1 |
|
| 2823 |
+ "movq (%1, %2, 2), %%mm6 \n\t" // R2 |
|
| 2824 |
+ "movq (%1, %%eax), %%mm7 \n\t" // R3 |
|
| 2825 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2826 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2827 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2828 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2829 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2830 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2831 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2832 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2833 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2834 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2835 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2836 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2837 |
+ "movq %%mm0, (%1) \n\t" // R0 |
|
| 2838 |
+ "movq %%mm1, (%1, %2) \n\t" // R1 |
|
| 2839 |
+ "movq %%mm2, (%1, %2, 2) \n\t" // R2 |
|
| 2840 |
+ "movq %%mm3, (%1, %%eax) \n\t" // R3 |
|
| 2841 |
+ "movq %%mm0, (%0) \n\t" // L0 |
|
| 2842 |
+ "movq %%mm1, (%0, %2) \n\t" // L1 |
|
| 2843 |
+ "movq %%mm2, (%0, %2, 2) \n\t" // L2 |
|
| 2844 |
+ "movq %%mm3, (%0, %%eax) \n\t" // L3 |
|
| 2845 |
+ |
|
| 2846 |
+ "movq (%0, %2, 4), %%mm0 \n\t" // L4 |
|
| 2847 |
+ "movq (%0, %%ebx), %%mm1 \n\t" // L5 |
|
| 2848 |
+ "movq (%0, %%eax, 2), %%mm2 \n\t" // L6 |
|
| 2849 |
+ "movq (%0, %%ecx), %%mm3 \n\t" // L7 |
|
| 2850 |
+ "movq (%1, %2, 4), %%mm4 \n\t" // R4 |
|
| 2851 |
+ "movq (%1, %%ebx), %%mm5 \n\t" // R5 |
|
| 2852 |
+ "movq (%1, %%eax, 2), %%mm6 \n\t" // R6 |
|
| 2853 |
+ "movq (%1, %%ecx), %%mm7 \n\t" // R7 |
|
| 2854 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2855 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2856 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2857 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2858 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2859 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2860 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2861 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2862 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2863 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2864 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2865 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2866 |
+ "movq %%mm0, (%1, %2, 4) \n\t" // R4 |
|
| 2867 |
+ "movq %%mm1, (%1, %%ebx) \n\t" // R5 |
|
| 2868 |
+ "movq %%mm2, (%1, %%eax, 2) \n\t" // R6 |
|
| 2869 |
+ "movq %%mm3, (%1, %%ecx) \n\t" // R7 |
|
| 2870 |
+ "movq %%mm0, (%0, %2, 4) \n\t" // L4 |
|
| 2871 |
+ "movq %%mm1, (%0, %%ebx) \n\t" // L5 |
|
| 2872 |
+ "movq %%mm2, (%0, %%eax, 2) \n\t" // L6 |
|
| 2873 |
+ "movq %%mm3, (%0, %%ecx) \n\t" // L7 |
|
| 2874 |
+ |
|
| 2875 |
+ "4: \n\t" |
|
| 2876 |
+ |
|
| 2877 |
+ :: "r" (src), "r" (tempBlured), "r"(stride), |
|
| 2878 |
+ "m"(maxNoise[0]), "m"(maxNoise[1]), "m"(maxNoise[2]) |
|
| 2879 |
+ : "%eax", "%ebx", "%ecx", "memory" |
|
| 2880 |
+ ); |
|
| 2881 |
+//printf("%d\n", test);
|
|
| 2882 |
+#else |
|
| 2599 | 2883 |
int y; |
| 2600 | 2884 |
int d=0; |
| 2601 | 2885 |
int sysd=0; |
| ... | ... |
@@ -2608,7 +2893,10 @@ static void inline tempNoiseReducer(uint8_t *src, int stride, |
| 2608 | 2608 |
int ref= tempBlured[ x + y*stride ]; |
| 2609 | 2609 |
int cur= src[ x + y*stride ]; |
| 2610 | 2610 |
int d1=ref - cur; |
| 2611 |
- d+= ABS(d1); //d1*d1; |
|
| 2611 |
+// if(x==0 || x==7) d1+= d1>>1; |
|
| 2612 |
+// if(y==0 || y==7) d1+= d1>>1; |
|
| 2613 |
+// d+= ABS(d1); |
|
| 2614 |
+ d+= d1*d1; |
|
| 2612 | 2615 |
sysd+= d1; |
| 2613 | 2616 |
} |
| 2614 | 2617 |
} |
| ... | ... |
@@ -2682,6 +2970,7 @@ Switch between |
| 2682 | 2682 |
} |
| 2683 | 2683 |
} |
| 2684 | 2684 |
} |
| 2685 |
+#endif |
|
| 2685 | 2686 |
} |
| 2686 | 2687 |
|
| 2687 | 2688 |
#ifdef HAVE_ODIVX_POSTPROCESS |
| ... | ... |
@@ -2914,9 +3203,9 @@ void postprocess(unsigned char * src[], int src_stride, |
| 2914 | 2914 |
ppMode.lumMode= mode; |
| 2915 | 2915 |
mode= ((mode&0xFF)>>4) | (mode&0xFFFFFF00); |
| 2916 | 2916 |
ppMode.chromMode= mode; |
| 2917 |
- ppMode.maxTmpNoise[0]= 150; |
|
| 2918 |
- ppMode.maxTmpNoise[1]= 200; |
|
| 2919 |
- ppMode.maxTmpNoise[2]= 400; |
|
| 2917 |
+ ppMode.maxTmpNoise[0]= 700; |
|
| 2918 |
+ ppMode.maxTmpNoise[1]= 1500; |
|
| 2919 |
+ ppMode.maxTmpNoise[2]= 3000; |
|
| 2920 | 2920 |
|
| 2921 | 2921 |
#ifdef HAVE_ODIVX_POSTPROCESS |
| 2922 | 2922 |
// Note: I could make this shit outside of this file, but it would mean one |
| ... | ... |
@@ -35,7 +35,7 @@ LinIpolDeinterlace e E E* |
| 35 | 35 |
CubicIpolDeinterlace a e e* |
| 36 | 36 |
LinBlendDeinterlace e E E* |
| 37 | 37 |
MedianDeinterlace# Ec Ec |
| 38 |
-TempDeNoiser# a |
|
| 38 |
+TempDeNoiser# E e e |
|
| 39 | 39 |
|
| 40 | 40 |
* i dont have a 3dnow CPU -> its untested, but noone said it doesnt work so it seems to work |
| 41 | 41 |
# more or less selfinvented filters so the exactness isnt too meaningfull |
| ... | ... |
@@ -61,6 +61,7 @@ split this huge file |
| 61 | 61 |
border remover |
| 62 | 62 |
optimize c versions |
| 63 | 63 |
try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks |
| 64 |
+smart blur |
|
| 64 | 65 |
... |
| 65 | 66 |
|
| 66 | 67 |
Notes: |
| ... | ... |
@@ -2592,10 +2593,294 @@ static inline void transpose2(uint8_t *dst, int dstStride, uint8_t *src) |
| 2592 | 2592 |
); |
| 2593 | 2593 |
} |
| 2594 | 2594 |
#endif |
| 2595 |
+//static int test=0; |
|
| 2595 | 2596 |
|
| 2596 | 2597 |
static void inline tempNoiseReducer(uint8_t *src, int stride, |
| 2597 | 2598 |
uint8_t *tempBlured, int *maxNoise) |
| 2598 | 2599 |
{
|
| 2600 |
+#define FAST_L2_DIFF |
|
| 2601 |
+//#define L1_DIFF //u should change the thresholds too if u try that one |
|
| 2602 |
+#if defined (HAVE_MMX2) || defined (HAVE_3DNOW) |
|
| 2603 |
+ asm volatile( |
|
| 2604 |
+ "leal (%2, %2, 2), %%eax \n\t" // 3*stride |
|
| 2605 |
+ "leal (%2, %2, 4), %%ebx \n\t" // 5*stride |
|
| 2606 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2607 |
+// 0 1 2 3 4 5 6 7 8 9 |
|
| 2608 |
+// %x %x+%2 %x+2%2 %x+eax %x+4%2 %x+ebx %x+2eax %x+ecx %x+8%2 |
|
| 2609 |
+//FIXME reorder? |
|
| 2610 |
+#ifdef L1_DIFF //needs mmx2 |
|
| 2611 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2612 |
+ "psadbw (%1), %%mm0 \n\t" // |L0-R0| |
|
| 2613 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2614 |
+ "psadbw (%1, %2), %%mm1 \n\t" // |L1-R1| |
|
| 2615 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2616 |
+ "psadbw (%1, %2, 2), %%mm2 \n\t" // |L2-R2| |
|
| 2617 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2618 |
+ "psadbw (%1, %%eax), %%mm3 \n\t" // |L3-R3| |
|
| 2619 |
+ |
|
| 2620 |
+ "movq (%0, %2, 4), %%mm4 \n\t" // L4 |
|
| 2621 |
+ "paddw %%mm1, %%mm0 \n\t" |
|
| 2622 |
+ "psadbw (%1, %2, 4), %%mm4 \n\t" // |L4-R4| |
|
| 2623 |
+ "movq (%0, %%ebx), %%mm5 \n\t" // L5 |
|
| 2624 |
+ "paddw %%mm2, %%mm0 \n\t" |
|
| 2625 |
+ "psadbw (%1, %%ebx), %%mm5 \n\t" // |L5-R5| |
|
| 2626 |
+ "movq (%0, %%eax, 2), %%mm6 \n\t" // L6 |
|
| 2627 |
+ "paddw %%mm3, %%mm0 \n\t" |
|
| 2628 |
+ "psadbw (%1, %%eax, 2), %%mm6 \n\t" // |L6-R6| |
|
| 2629 |
+ "movq (%0, %%ecx), %%mm7 \n\t" // L7 |
|
| 2630 |
+ "paddw %%mm4, %%mm0 \n\t" |
|
| 2631 |
+ "psadbw (%1, %%ecx), %%mm7 \n\t" // |L7-R7| |
|
| 2632 |
+ "paddw %%mm5, %%mm6 \n\t" |
|
| 2633 |
+ "paddw %%mm7, %%mm6 \n\t" |
|
| 2634 |
+ "paddw %%mm6, %%mm0 \n\t" |
|
| 2635 |
+#elif defined (FAST_L2_DIFF) |
|
| 2636 |
+ "pcmpeqb %%mm7, %%mm7 \n\t" |
|
| 2637 |
+ "movq b80, %%mm6 \n\t" |
|
| 2638 |
+ "pxor %%mm0, %%mm0 \n\t" |
|
| 2639 |
+#define L2_DIFF_CORE(a, b)\ |
|
| 2640 |
+ "movq " #a ", %%mm5 \n\t"\ |
|
| 2641 |
+ "movq " #b ", %%mm2 \n\t"\ |
|
| 2642 |
+ "pxor %%mm7, %%mm2 \n\t"\ |
|
| 2643 |
+ PAVGB(%%mm2, %%mm5)\ |
|
| 2644 |
+ "paddb %%mm6, %%mm5 \n\t"\ |
|
| 2645 |
+ "movq %%mm5, %%mm2 \n\t"\ |
|
| 2646 |
+ "psllw $8, %%mm5 \n\t"\ |
|
| 2647 |
+ "pmaddwd %%mm5, %%mm5 \n\t"\ |
|
| 2648 |
+ "pmaddwd %%mm2, %%mm2 \n\t"\ |
|
| 2649 |
+ "paddd %%mm2, %%mm5 \n\t"\ |
|
| 2650 |
+ "psrld $14, %%mm5 \n\t"\ |
|
| 2651 |
+ "paddd %%mm5, %%mm0 \n\t" |
|
| 2652 |
+ |
|
| 2653 |
+L2_DIFF_CORE((%0), (%1)) |
|
| 2654 |
+L2_DIFF_CORE((%0, %2), (%1, %2)) |
|
| 2655 |
+L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2)) |
|
| 2656 |
+L2_DIFF_CORE((%0, %%eax), (%1, %%eax)) |
|
| 2657 |
+L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4)) |
|
| 2658 |
+L2_DIFF_CORE((%0, %%ebx), (%1, %%ebx)) |
|
| 2659 |
+L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2)) |
|
| 2660 |
+L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx)) |
|
| 2661 |
+ |
|
| 2662 |
+#else |
|
| 2663 |
+ "pxor %%mm7, %%mm7 \n\t" |
|
| 2664 |
+ "pxor %%mm0, %%mm0 \n\t" |
|
| 2665 |
+#define L2_DIFF_CORE(a, b)\ |
|
| 2666 |
+ "movq " #a ", %%mm5 \n\t"\ |
|
| 2667 |
+ "movq " #b ", %%mm2 \n\t"\ |
|
| 2668 |
+ "movq %%mm5, %%mm1 \n\t"\ |
|
| 2669 |
+ "movq %%mm2, %%mm3 \n\t"\ |
|
| 2670 |
+ "punpcklbw %%mm7, %%mm5 \n\t"\ |
|
| 2671 |
+ "punpckhbw %%mm7, %%mm1 \n\t"\ |
|
| 2672 |
+ "punpcklbw %%mm7, %%mm2 \n\t"\ |
|
| 2673 |
+ "punpckhbw %%mm7, %%mm3 \n\t"\ |
|
| 2674 |
+ "psubw %%mm2, %%mm5 \n\t"\ |
|
| 2675 |
+ "psubw %%mm3, %%mm1 \n\t"\ |
|
| 2676 |
+ "pmaddwd %%mm5, %%mm5 \n\t"\ |
|
| 2677 |
+ "pmaddwd %%mm1, %%mm1 \n\t"\ |
|
| 2678 |
+ "paddd %%mm1, %%mm5 \n\t"\ |
|
| 2679 |
+ "paddd %%mm5, %%mm0 \n\t" |
|
| 2680 |
+ |
|
| 2681 |
+L2_DIFF_CORE((%0), (%1)) |
|
| 2682 |
+L2_DIFF_CORE((%0, %2), (%1, %2)) |
|
| 2683 |
+L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2)) |
|
| 2684 |
+L2_DIFF_CORE((%0, %%eax), (%1, %%eax)) |
|
| 2685 |
+L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4)) |
|
| 2686 |
+L2_DIFF_CORE((%0, %%ebx), (%1, %%ebx)) |
|
| 2687 |
+L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2)) |
|
| 2688 |
+L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx)) |
|
| 2689 |
+ |
|
| 2690 |
+#endif |
|
| 2691 |
+ |
|
| 2692 |
+ "movq %%mm0, %%mm4 \n\t" |
|
| 2693 |
+ "psrlq $32, %%mm0 \n\t" |
|
| 2694 |
+ "paddd %%mm0, %%mm4 \n\t" |
|
| 2695 |
+ "movd %%mm4, %%ecx \n\t" |
|
| 2696 |
+// "movl %3, %%ecx \n\t" |
|
| 2697 |
+// "movl %%ecx, test \n\t" |
|
| 2698 |
+// "jmp 4f \n\t" |
|
| 2699 |
+ "cmpl %4, %%ecx \n\t" |
|
| 2700 |
+ " jb 2f \n\t" |
|
| 2701 |
+ "cmpl %5, %%ecx \n\t" |
|
| 2702 |
+ " jb 1f \n\t" |
|
| 2703 |
+ |
|
| 2704 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2705 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2706 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2707 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2708 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2709 |
+ "movq (%0, %2, 4), %%mm4 \n\t" // L4 |
|
| 2710 |
+ "movq (%0, %%ebx), %%mm5 \n\t" // L5 |
|
| 2711 |
+ "movq (%0, %%eax, 2), %%mm6 \n\t" // L6 |
|
| 2712 |
+ "movq (%0, %%ecx), %%mm7 \n\t" // L7 |
|
| 2713 |
+ "movq %%mm0, (%1) \n\t" // L0 |
|
| 2714 |
+ "movq %%mm1, (%1, %2) \n\t" // L1 |
|
| 2715 |
+ "movq %%mm2, (%1, %2, 2) \n\t" // L2 |
|
| 2716 |
+ "movq %%mm3, (%1, %%eax) \n\t" // L3 |
|
| 2717 |
+ "movq %%mm4, (%1, %2, 4) \n\t" // L4 |
|
| 2718 |
+ "movq %%mm5, (%1, %%ebx) \n\t" // L5 |
|
| 2719 |
+ "movq %%mm6, (%1, %%eax, 2) \n\t" // L6 |
|
| 2720 |
+ "movq %%mm7, (%1, %%ecx) \n\t" // L7 |
|
| 2721 |
+ "jmp 4f \n\t" |
|
| 2722 |
+ |
|
| 2723 |
+ "1: \n\t" |
|
| 2724 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2725 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2726 |
+ "pavgb (%1), %%mm0 \n\t" // L0 |
|
| 2727 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2728 |
+ "pavgb (%1, %2), %%mm1 \n\t" // L1 |
|
| 2729 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2730 |
+ "pavgb (%1, %2, 2), %%mm2 \n\t" // L2 |
|
| 2731 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2732 |
+ "pavgb (%1, %%eax), %%mm3 \n\t" // L3 |
|
| 2733 |
+ "movq (%0, %2, 4), %%mm4 \n\t" // L4 |
|
| 2734 |
+ "pavgb (%1, %2, 4), %%mm4 \n\t" // L4 |
|
| 2735 |
+ "movq (%0, %%ebx), %%mm5 \n\t" // L5 |
|
| 2736 |
+ "pavgb (%1, %%ebx), %%mm5 \n\t" // L5 |
|
| 2737 |
+ "movq (%0, %%eax, 2), %%mm6 \n\t" // L6 |
|
| 2738 |
+ "pavgb (%1, %%eax, 2), %%mm6 \n\t" // L6 |
|
| 2739 |
+ "movq (%0, %%ecx), %%mm7 \n\t" // L7 |
|
| 2740 |
+ "pavgb (%1, %%ecx), %%mm7 \n\t" // L7 |
|
| 2741 |
+ "movq %%mm0, (%1) \n\t" // R0 |
|
| 2742 |
+ "movq %%mm1, (%1, %2) \n\t" // R1 |
|
| 2743 |
+ "movq %%mm2, (%1, %2, 2) \n\t" // R2 |
|
| 2744 |
+ "movq %%mm3, (%1, %%eax) \n\t" // R3 |
|
| 2745 |
+ "movq %%mm4, (%1, %2, 4) \n\t" // R4 |
|
| 2746 |
+ "movq %%mm5, (%1, %%ebx) \n\t" // R5 |
|
| 2747 |
+ "movq %%mm6, (%1, %%eax, 2) \n\t" // R6 |
|
| 2748 |
+ "movq %%mm7, (%1, %%ecx) \n\t" // R7 |
|
| 2749 |
+ "movq %%mm0, (%0) \n\t" // L0 |
|
| 2750 |
+ "movq %%mm1, (%0, %2) \n\t" // L1 |
|
| 2751 |
+ "movq %%mm2, (%0, %2, 2) \n\t" // L2 |
|
| 2752 |
+ "movq %%mm3, (%0, %%eax) \n\t" // L3 |
|
| 2753 |
+ "movq %%mm4, (%0, %2, 4) \n\t" // L4 |
|
| 2754 |
+ "movq %%mm5, (%0, %%ebx) \n\t" // L5 |
|
| 2755 |
+ "movq %%mm6, (%0, %%eax, 2) \n\t" // L6 |
|
| 2756 |
+ "movq %%mm7, (%0, %%ecx) \n\t" // L7 |
|
| 2757 |
+ "jmp 4f \n\t" |
|
| 2758 |
+ |
|
| 2759 |
+ "2: \n\t" |
|
| 2760 |
+ "cmpl %3, %%ecx \n\t" |
|
| 2761 |
+ " jb 3f \n\t" |
|
| 2762 |
+ |
|
| 2763 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2764 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2765 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2766 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2767 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2768 |
+ "movq (%1), %%mm4 \n\t" // R0 |
|
| 2769 |
+ "movq (%1, %2), %%mm5 \n\t" // R1 |
|
| 2770 |
+ "movq (%1, %2, 2), %%mm6 \n\t" // R2 |
|
| 2771 |
+ "movq (%1, %%eax), %%mm7 \n\t" // R3 |
|
| 2772 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2773 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2774 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2775 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2776 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2777 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2778 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2779 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2780 |
+ "movq %%mm0, (%1) \n\t" // R0 |
|
| 2781 |
+ "movq %%mm1, (%1, %2) \n\t" // R1 |
|
| 2782 |
+ "movq %%mm2, (%1, %2, 2) \n\t" // R2 |
|
| 2783 |
+ "movq %%mm3, (%1, %%eax) \n\t" // R3 |
|
| 2784 |
+ "movq %%mm0, (%0) \n\t" // L0 |
|
| 2785 |
+ "movq %%mm1, (%0, %2) \n\t" // L1 |
|
| 2786 |
+ "movq %%mm2, (%0, %2, 2) \n\t" // L2 |
|
| 2787 |
+ "movq %%mm3, (%0, %%eax) \n\t" // L3 |
|
| 2788 |
+ |
|
| 2789 |
+ "movq (%0, %2, 4), %%mm0 \n\t" // L4 |
|
| 2790 |
+ "movq (%0, %%ebx), %%mm1 \n\t" // L5 |
|
| 2791 |
+ "movq (%0, %%eax, 2), %%mm2 \n\t" // L6 |
|
| 2792 |
+ "movq (%0, %%ecx), %%mm3 \n\t" // L7 |
|
| 2793 |
+ "movq (%1, %2, 4), %%mm4 \n\t" // R4 |
|
| 2794 |
+ "movq (%1, %%ebx), %%mm5 \n\t" // R5 |
|
| 2795 |
+ "movq (%1, %%eax, 2), %%mm6 \n\t" // R6 |
|
| 2796 |
+ "movq (%1, %%ecx), %%mm7 \n\t" // R7 |
|
| 2797 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2798 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2799 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2800 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2801 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2802 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2803 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2804 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2805 |
+ "movq %%mm0, (%1, %2, 4) \n\t" // R4 |
|
| 2806 |
+ "movq %%mm1, (%1, %%ebx) \n\t" // R5 |
|
| 2807 |
+ "movq %%mm2, (%1, %%eax, 2) \n\t" // R6 |
|
| 2808 |
+ "movq %%mm3, (%1, %%ecx) \n\t" // R7 |
|
| 2809 |
+ "movq %%mm0, (%0, %2, 4) \n\t" // L4 |
|
| 2810 |
+ "movq %%mm1, (%0, %%ebx) \n\t" // L5 |
|
| 2811 |
+ "movq %%mm2, (%0, %%eax, 2) \n\t" // L6 |
|
| 2812 |
+ "movq %%mm3, (%0, %%ecx) \n\t" // L7 |
|
| 2813 |
+ "jmp 4f \n\t" |
|
| 2814 |
+ |
|
| 2815 |
+ "3: \n\t" |
|
| 2816 |
+ "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride |
|
| 2817 |
+ "movq (%0), %%mm0 \n\t" // L0 |
|
| 2818 |
+ "movq (%0, %2), %%mm1 \n\t" // L1 |
|
| 2819 |
+ "movq (%0, %2, 2), %%mm2 \n\t" // L2 |
|
| 2820 |
+ "movq (%0, %%eax), %%mm3 \n\t" // L3 |
|
| 2821 |
+ "movq (%1), %%mm4 \n\t" // R0 |
|
| 2822 |
+ "movq (%1, %2), %%mm5 \n\t" // R1 |
|
| 2823 |
+ "movq (%1, %2, 2), %%mm6 \n\t" // R2 |
|
| 2824 |
+ "movq (%1, %%eax), %%mm7 \n\t" // R3 |
|
| 2825 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2826 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2827 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2828 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2829 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2830 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2831 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2832 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2833 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2834 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2835 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2836 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2837 |
+ "movq %%mm0, (%1) \n\t" // R0 |
|
| 2838 |
+ "movq %%mm1, (%1, %2) \n\t" // R1 |
|
| 2839 |
+ "movq %%mm2, (%1, %2, 2) \n\t" // R2 |
|
| 2840 |
+ "movq %%mm3, (%1, %%eax) \n\t" // R3 |
|
| 2841 |
+ "movq %%mm0, (%0) \n\t" // L0 |
|
| 2842 |
+ "movq %%mm1, (%0, %2) \n\t" // L1 |
|
| 2843 |
+ "movq %%mm2, (%0, %2, 2) \n\t" // L2 |
|
| 2844 |
+ "movq %%mm3, (%0, %%eax) \n\t" // L3 |
|
| 2845 |
+ |
|
| 2846 |
+ "movq (%0, %2, 4), %%mm0 \n\t" // L4 |
|
| 2847 |
+ "movq (%0, %%ebx), %%mm1 \n\t" // L5 |
|
| 2848 |
+ "movq (%0, %%eax, 2), %%mm2 \n\t" // L6 |
|
| 2849 |
+ "movq (%0, %%ecx), %%mm3 \n\t" // L7 |
|
| 2850 |
+ "movq (%1, %2, 4), %%mm4 \n\t" // R4 |
|
| 2851 |
+ "movq (%1, %%ebx), %%mm5 \n\t" // R5 |
|
| 2852 |
+ "movq (%1, %%eax, 2), %%mm6 \n\t" // R6 |
|
| 2853 |
+ "movq (%1, %%ecx), %%mm7 \n\t" // R7 |
|
| 2854 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2855 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2856 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2857 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2858 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2859 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2860 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2861 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2862 |
+ PAVGB(%%mm4, %%mm0) |
|
| 2863 |
+ PAVGB(%%mm5, %%mm1) |
|
| 2864 |
+ PAVGB(%%mm6, %%mm2) |
|
| 2865 |
+ PAVGB(%%mm7, %%mm3) |
|
| 2866 |
+ "movq %%mm0, (%1, %2, 4) \n\t" // R4 |
|
| 2867 |
+ "movq %%mm1, (%1, %%ebx) \n\t" // R5 |
|
| 2868 |
+ "movq %%mm2, (%1, %%eax, 2) \n\t" // R6 |
|
| 2869 |
+ "movq %%mm3, (%1, %%ecx) \n\t" // R7 |
|
| 2870 |
+ "movq %%mm0, (%0, %2, 4) \n\t" // L4 |
|
| 2871 |
+ "movq %%mm1, (%0, %%ebx) \n\t" // L5 |
|
| 2872 |
+ "movq %%mm2, (%0, %%eax, 2) \n\t" // L6 |
|
| 2873 |
+ "movq %%mm3, (%0, %%ecx) \n\t" // L7 |
|
| 2874 |
+ |
|
| 2875 |
+ "4: \n\t" |
|
| 2876 |
+ |
|
| 2877 |
+ :: "r" (src), "r" (tempBlured), "r"(stride), |
|
| 2878 |
+ "m"(maxNoise[0]), "m"(maxNoise[1]), "m"(maxNoise[2]) |
|
| 2879 |
+ : "%eax", "%ebx", "%ecx", "memory" |
|
| 2880 |
+ ); |
|
| 2881 |
+//printf("%d\n", test);
|
|
| 2882 |
+#else |
|
| 2599 | 2883 |
int y; |
| 2600 | 2884 |
int d=0; |
| 2601 | 2885 |
int sysd=0; |
| ... | ... |
@@ -2608,7 +2893,10 @@ static void inline tempNoiseReducer(uint8_t *src, int stride, |
| 2608 | 2608 |
int ref= tempBlured[ x + y*stride ]; |
| 2609 | 2609 |
int cur= src[ x + y*stride ]; |
| 2610 | 2610 |
int d1=ref - cur; |
| 2611 |
- d+= ABS(d1); //d1*d1; |
|
| 2611 |
+// if(x==0 || x==7) d1+= d1>>1; |
|
| 2612 |
+// if(y==0 || y==7) d1+= d1>>1; |
|
| 2613 |
+// d+= ABS(d1); |
|
| 2614 |
+ d+= d1*d1; |
|
| 2612 | 2615 |
sysd+= d1; |
| 2613 | 2616 |
} |
| 2614 | 2617 |
} |
| ... | ... |
@@ -2682,6 +2970,7 @@ Switch between |
| 2682 | 2682 |
} |
| 2683 | 2683 |
} |
| 2684 | 2684 |
} |
| 2685 |
+#endif |
|
| 2685 | 2686 |
} |
| 2686 | 2687 |
|
| 2687 | 2688 |
#ifdef HAVE_ODIVX_POSTPROCESS |
| ... | ... |
@@ -2914,9 +3203,9 @@ void postprocess(unsigned char * src[], int src_stride, |
| 2914 | 2914 |
ppMode.lumMode= mode; |
| 2915 | 2915 |
mode= ((mode&0xFF)>>4) | (mode&0xFFFFFF00); |
| 2916 | 2916 |
ppMode.chromMode= mode; |
| 2917 |
- ppMode.maxTmpNoise[0]= 150; |
|
| 2918 |
- ppMode.maxTmpNoise[1]= 200; |
|
| 2919 |
- ppMode.maxTmpNoise[2]= 400; |
|
| 2917 |
+ ppMode.maxTmpNoise[0]= 700; |
|
| 2918 |
+ ppMode.maxTmpNoise[1]= 1500; |
|
| 2919 |
+ ppMode.maxTmpNoise[2]= 3000; |
|
| 2920 | 2920 |
|
| 2921 | 2921 |
#ifdef HAVE_ODIVX_POSTPROCESS |
| 2922 | 2922 |
// Note: I could make this shit outside of this file, but it would mean one |