📄 postprocess_template.c
字号:
"movq " #addr ", %%mm0 \n\t"\ "pminub %%mm0, %%mm7 \n\t"\ "pmaxub %%mm0, %%mm6 \n\t"#else#define REAL_FIND_MIN_MAX(addr)\ "movq " #addr ", %%mm0 \n\t"\ "movq %%mm7, %%mm1 \n\t"\ "psubusb %%mm0, %%mm6 \n\t"\ "paddb %%mm0, %%mm6 \n\t"\ "psubusb %%mm0, %%mm1 \n\t"\ "psubb %%mm1, %%mm7 \n\t"#endif#define FIND_MIN_MAX(addr) REAL_FIND_MIN_MAX(addr)FIND_MIN_MAX((%%REGa))FIND_MIN_MAX((%%REGa, %1))FIND_MIN_MAX((%%REGa, %1, 2))FIND_MIN_MAX((%0, %1, 4))FIND_MIN_MAX((%%REGd))FIND_MIN_MAX((%%REGd, %1))FIND_MIN_MAX((%%REGd, %1, 2))FIND_MIN_MAX((%0, %1, 8)) "movq %%mm7, %%mm4 \n\t" "psrlq $8, %%mm7 \n\t"#ifdef HAVE_MMX2 "pminub %%mm4, %%mm7 \n\t" // min of pixels "pshufw $0xF9, %%mm7, %%mm4 \n\t" "pminub %%mm4, %%mm7 \n\t" // min of pixels "pshufw $0xFE, %%mm7, %%mm4 \n\t" "pminub %%mm4, %%mm7 \n\t"#else "movq %%mm7, %%mm1 \n\t" "psubusb %%mm4, %%mm1 \n\t" "psubb %%mm1, %%mm7 \n\t" "movq %%mm7, %%mm4 \n\t" "psrlq $16, %%mm7 \n\t" "movq %%mm7, %%mm1 \n\t" "psubusb %%mm4, %%mm1 \n\t" "psubb %%mm1, %%mm7 \n\t" "movq %%mm7, %%mm4 \n\t" "psrlq $32, %%mm7 \n\t" "movq %%mm7, %%mm1 \n\t" "psubusb %%mm4, %%mm1 \n\t" "psubb %%mm1, %%mm7 \n\t"#endif "movq %%mm6, %%mm4 \n\t" "psrlq $8, %%mm6 \n\t"#ifdef HAVE_MMX2 "pmaxub %%mm4, %%mm6 \n\t" // max of pixels "pshufw $0xF9, %%mm6, %%mm4 \n\t" "pmaxub %%mm4, %%mm6 \n\t" "pshufw $0xFE, %%mm6, %%mm4 \n\t" "pmaxub %%mm4, %%mm6 \n\t"#else "psubusb %%mm4, %%mm6 \n\t" "paddb %%mm4, %%mm6 \n\t" "movq %%mm6, %%mm4 \n\t" "psrlq $16, %%mm6 \n\t" "psubusb %%mm4, %%mm6 \n\t" "paddb %%mm4, %%mm6 \n\t" "movq %%mm6, %%mm4 \n\t" "psrlq $32, %%mm6 \n\t" "psubusb %%mm4, %%mm6 \n\t" "paddb %%mm4, %%mm6 \n\t"#endif "movq %%mm6, %%mm0 \n\t" // max "psubb %%mm7, %%mm6 \n\t" // max - min "movd %%mm6, %%ecx \n\t" "cmpb "MANGLE(deringThreshold)", %%cl \n\t" " jb 1f \n\t" "lea -24(%%"REG_SP"), %%"REG_c" \n\t" "and "ALIGN_MASK", %%"REG_c" \n\t" PAVGB(%%mm0, %%mm7) // a=(max + min)/2 "punpcklbw %%mm7, %%mm7 \n\t" "punpcklbw %%mm7, %%mm7 \n\t" "punpcklbw %%mm7, %%mm7 \n\t" "movq %%mm7, (%%"REG_c") \n\t" "movq (%0), %%mm0 \n\t" // L10 "movq %%mm0, %%mm1 \n\t" // L10 "movq %%mm0, %%mm2 \n\t" // L10 "psllq $8, %%mm1 \n\t" "psrlq $8, %%mm2 \n\t" "movd -4(%0), %%mm3 \n\t" "movd 8(%0), %%mm4 \n\t" "psrlq $24, %%mm3 \n\t" "psllq $56, %%mm4 \n\t" "por %%mm3, %%mm1 \n\t" // L00 "por %%mm4, %%mm2 \n\t" // L20 "movq %%mm1, %%mm3 \n\t" // L00 PAVGB(%%mm2, %%mm1) // (L20 + L00)/2 PAVGB(%%mm0, %%mm1) // (L20 + L00 + 2L10)/4 "psubusb %%mm7, %%mm0 \n\t" "psubusb %%mm7, %%mm2 \n\t" "psubusb %%mm7, %%mm3 \n\t" "pcmpeqb "MANGLE(b00)", %%mm0 \n\t" // L10 > a ? 0 : -1 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L20 > a ? 0 : -1 "pcmpeqb "MANGLE(b00)", %%mm3 \n\t" // L00 > a ? 0 : -1 "paddb %%mm2, %%mm0 \n\t" "paddb %%mm3, %%mm0 \n\t" "movq (%%"REG_a"), %%mm2 \n\t" // L11 "movq %%mm2, %%mm3 \n\t" // L11 "movq %%mm2, %%mm4 \n\t" // L11 "psllq $8, %%mm3 \n\t" "psrlq $8, %%mm4 \n\t" "movd -4(%%"REG_a"), %%mm5 \n\t" "movd 8(%%"REG_a"), %%mm6 \n\t" "psrlq $24, %%mm5 \n\t" "psllq $56, %%mm6 \n\t" "por %%mm5, %%mm3 \n\t" // L01 "por %%mm6, %%mm4 \n\t" // L21 "movq %%mm3, %%mm5 \n\t" // L01 PAVGB(%%mm4, %%mm3) // (L21 + L01)/2 PAVGB(%%mm2, %%mm3) // (L21 + L01 + 2L11)/4 "psubusb %%mm7, %%mm2 \n\t" "psubusb %%mm7, %%mm4 \n\t" "psubusb %%mm7, %%mm5 \n\t" "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L11 > a ? 0 : -1 "pcmpeqb "MANGLE(b00)", %%mm4 \n\t" // L21 > a ? 0 : -1 "pcmpeqb "MANGLE(b00)", %%mm5 \n\t" // L01 > a ? 0 : -1 "paddb %%mm4, %%mm2 \n\t" "paddb %%mm5, %%mm2 \n\t"// 0, 2, 3, 1#define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \ "movq " #src ", " #sx " \n\t" /* src[0] */\ "movq " #sx ", " #lx " \n\t" /* src[0] */\ "movq " #sx ", " #t0 " \n\t" /* src[0] */\ "psllq $8, " #lx " \n\t"\ "psrlq $8, " #t0 " \n\t"\ "movd -4" #src ", " #t1 " \n\t"\ "psrlq $24, " #t1 " \n\t"\ "por " #t1 ", " #lx " \n\t" /* src[-1] */\ "movd 8" #src ", " #t1 " \n\t"\ "psllq $56, " #t1 " \n\t"\ "por " #t1 ", " #t0 " \n\t" /* src[+1] */\ "movq " #lx ", " #t1 " \n\t" /* src[-1] */\ PAVGB(t0, lx) /* (src[-1] + src[+1])/2 */\ PAVGB(sx, lx) /* (src[-1] + 2src[0] + src[+1])/4 */\ PAVGB(lx, pplx) \ "movq " #lx ", 8(%%"REG_c") \n\t"\ "movq (%%"REG_c"), " #lx " \n\t"\ "psubusb " #lx ", " #t1 " \n\t"\ "psubusb " #lx ", " #t0 " \n\t"\ "psubusb " #lx ", " #sx " \n\t"\ "movq "MANGLE(b00)", " #lx " \n\t"\ "pcmpeqb " #lx ", " #t1 " \n\t" /* src[-1] > a ? 0 : -1*/\ "pcmpeqb " #lx ", " #t0 " \n\t" /* src[+1] > a ? 0 : -1*/\ "pcmpeqb " #lx ", " #sx " \n\t" /* src[0] > a ? 0 : -1*/\ "paddb " #t1 ", " #t0 " \n\t"\ "paddb " #t0 ", " #sx " \n\t"\\ PAVGB(plx, pplx) /* filtered */\ "movq " #dst ", " #t0 " \n\t" /* dst */\ "movq " #t0 ", " #t1 " \n\t" /* dst */\ "psubusb %3, " #t0 " \n\t"\ "paddusb %3, " #t1 " \n\t"\ PMAXUB(t0, pplx)\ PMINUB(t1, pplx, t0)\ "paddb " #sx ", " #ppsx " \n\t"\ "paddb " #psx ", " #ppsx " \n\t"\ "#paddb "MANGLE(b02)", " #ppsx " \n\t"\ "pand "MANGLE(b08)", " #ppsx " \n\t"\ "pcmpeqb " #lx ", " #ppsx " \n\t"\ "pand " #ppsx ", " #pplx " \n\t"\ "pandn " #dst ", " #ppsx " \n\t"\ "por " #pplx ", " #ppsx " \n\t"\ "movq " #ppsx ", " #dst " \n\t"\ "movq 8(%%"REG_c"), " #lx " \n\t"#define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \ REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)/*0000000111111111111101111101111110011110111111010111100111110001110111*///DERING_CORE(dst ,src ,ppsx ,psx ,sx ,pplx ,plx ,lx ,t0 ,t1)DERING_CORE((%%REGa) ,(%%REGa, %1) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)DERING_CORE((%%REGa, %1) ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)DERING_CORE((%%REGa, %1, 2),(%0, %1, 4) ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)DERING_CORE((%0, %1, 4) ,(%%REGd) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)DERING_CORE((%%REGd) ,(%%REGd, %1) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)DERING_CORE((%%REGd, %1) ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)DERING_CORE((%%REGd, %1, 2),(%0, %1, 8) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)DERING_CORE((%0, %1, 8) ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7) "1: \n\t" : : "r" (src), "r" ((long)stride), "m" (c->pQPb), "m"(c->pQPb2) : "%"REG_a, "%"REG_d, "%"REG_c );#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW) int y; int min=255; int max=0; int avg; uint8_t *p; int s[10]; const int QP2= c->QP/2 + 1; for(y=1; y<9; y++){ int x; p= src + stride*y; for(x=1; x<9; x++){ p++; if(*p > max) max= *p; if(*p < min) min= *p; } } avg= (min + max + 1)>>1; if(max - min <deringThreshold) return; for(y=0; y<10; y++){ int t = 0; if(src[stride*y + 0] > avg) t+= 1; if(src[stride*y + 1] > avg) t+= 2; if(src[stride*y + 2] > avg) t+= 4; if(src[stride*y + 3] > avg) t+= 8; if(src[stride*y + 4] > avg) t+= 16; if(src[stride*y + 5] > avg) t+= 32; if(src[stride*y + 6] > avg) t+= 64; if(src[stride*y + 7] > avg) t+= 128; if(src[stride*y + 8] > avg) t+= 256; if(src[stride*y + 9] > avg) t+= 512; t |= (~t)<<16; t &= (t<<1) & (t>>1); s[y] = t; } for(y=1; y<9; y++){ int t = s[y-1] & s[y] & s[y+1]; t|= t>>16; s[y-1]= t; } for(y=1; y<9; y++){ int x; int t = s[y-1]; p= src + stride*y; for(x=1; x<9; x++){ p++; if(t & (1<<x)){ int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1)) +2*(*(p -1)) + 4*(*p ) + 2*(*(p +1)) +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1)); f= (f + 8)>>4;#ifdef DEBUG_DERING_THRESHOLD asm volatile("emms\n\t":); { static long long numPixels=0; if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;// if((max-min)<20 || (max-min)*QP<200)// if((max-min)*QP < 500)// if(max-min<QP/2) if(max-min < 20){ static int numSkipped=0; static int errorSum=0; static int worstQP=0; static int worstRange=0; static int worstDiff=0; int diff= (f - *p); int absDiff= FFABS(diff); int error= diff*diff; if(x==1 || x==8 || y==1 || y==8) continue; numSkipped++; if(absDiff > worstDiff){ worstDiff= absDiff;
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -