internal_bfin.s

ffmpeg的完整源代码和作者自己写的文档。不但有在Linux的工程哦

        /* G = Y+ cgu*(Cb-128)+cgv*(Cr-128) */
                a1 += r1.h*r6.l,          a0 += r1.l*r6.l  || r1=[i1++]; // cgv
        r2.h = (a1 += r1.h*r7.l), r2.l = (a0 += r1.l*r7.l);
        r2 = byteop3p(r3:2, r1:0)(LO)                      || r5=[i1++m0]; // gmask
        r2 = r2 << 2 (v);
        r2 = r2 & r5;
        r3 = r3 | r2;
        [p1++]=r3                                          || r1=[i1++]; // cy

        /* Y' = y*cy */

        a1 = r1.h*r4.h, a0 = r1.l*r4.l                     || r1=[i1++]; // crv

        /* R = Y+ crv*(Cr-128) */
        r2.h = (a1 += r1.h*r7.h), r2.l = (a0 += r1.l*r7.h);
                a1 -= r1.h*r7.h,          a0 -= r1.l*r7.h  || r5=[i1++]; // rmask
        r2 = byteop3p(r3:2, r1:0)(LO)                      || r1=[i1++]; // cbu
        r2 = r2 >> 3 (v);
        r3 = r2 & r5;

        /* B = Y+ cbu*(Cb-128) */
        r2.h = (a1 += r1.h*r6.h), r2.l = (a0 += r1.l*r6.h);
                a1 -= r1.h*r6.h,          a0 -= r1.l*r6.h  || r5=[i1++]; // bmask
        r2 = byteop3p(r3:2, r1:0)(LO)                      || r1=[i1++]; // cgu
        r2 = r2 << 7 (v);
        r2 = r2 & r5;
        r3 = r3 | r2;

        /* G = Y+ cgu*(Cb-128)+cgv*(Cr-128) */
                a1 += r1.h*r6.h,          a0 += r1.l*r6.h  || r1=[i1++]; // cgv
        r2.h = (a1 += r1.h*r7.h), r2.l = (a0 += r1.l*r7.h) || r5=[i1++]; // gmask
        r2 = byteop3p(r3:2, r1:0)(LO)                      || r0=[i0++];     // 4Y
        r2 = r2 << 2 (v)                                   || r1.l=w[i2++];  // 2u
        r2 = r2 & r5;
        r3 = r3 | r2;
        [p1++]=r3                                          || r1.h=w[i3++]; // 2v

.L1555:                                                       r2=[i1++]; // oy

        l1 = 0;

        (r7:4) = [sp++];
        unlink;
        rts;
DEFUN_END(yuv2rgb555_line)

DEFUN(yuv2rgb24_line,MEM,
   (uint8_t *Y, uint8_t *U, uint8_t *V, int *out, int dW, uint32_t *coeffs)):
        link 0;
        [--sp] = (r7:4);
        p1 = [fp+ARG_OUT];
        r3 = [fp+ARG_W];
        p2 = p1;
        p2 += 3;

        i0 = r0;
        i2 = r1;
        i3 = r2;

        r0 = [fp+ARG_COEFF]; // coeff buffer
        i1 = r0;
        b1 = i1;
        l1 = COEFF_LEN;
        m0 = COEFF_REL_CY_OFF;
        p0 = r3;

        r0   = [i0++];         // 2Y
        r1.l = w[i2++];        // 2u
        r1.h = w[i3++];        // 2v
        p0 = p0>>2;

        lsetup (.L0888, .L1888) lc0 = p0;

        /*
           uint32_t oy,oc,zero,cy,crv,rmask,cbu,bmask,cgu,cgv
           r0 -- used to load 4ys
           r1 -- used to load 2us,2vs
           r4 -- y3,y2
           r5 -- y1,y0
           r6 -- u1,u0
           r7 -- v1,v0
        */
                                                              r2=[i1++]; // oy
.L0888:
        (r4,r5) = byteop16m (r1:0, r3:2)                   || r3=[i1++]; // oc
        (r7,r6) = byteop16m (r1:0, r3:2) (r);
        r5 = r5 << 2 (v);               // y1,y0
        r4 = r4 << 2 (v);               // y3,y2
        r6 = r6 << 2 (v) || r0=[i1++];  // u1,u0, r0=zero
        r7 = r7 << 2 (v) || r1=[i1++];  // v1,v0  r1=cy

        /* Y' = y*cy */
        a1 = r1.h*r5.h, a0 = r1.l*r5.l                     || r1=[i1++]; // crv

        /* R = Y+ crv*(Cr-128) */
        r2.h = (a1 += r1.h*r7.l), r2.l = (a0 += r1.l*r7.l);
                a1 -= r1.h*r7.l,          a0 -= r1.l*r7.l  || r5=[i1++]; // rmask
        r2 = byteop3p(r3:2, r1:0)(LO)                      || r1=[i1++]; // cbu
        r2=r2>>16 || B[p1++]=r2;
                     B[p2++]=r2;

        /* B = Y+ cbu*(Cb-128) */
        r2.h = (a1 += r1.h*r6.l), r2.l = (a0 += r1.l*r6.l);
                a1 -= r1.h*r6.l,          a0 -= r1.l*r6.l  || r5=[i1++]; // bmask
        r3 = byteop3p(r3:2, r1:0)(LO)                      || r1=[i1++]; // cgu

        /* G = Y+ cgu*(Cb-128)+cgv*(Cr-128) */
                a1 += r1.h*r6.l,          a0 += r1.l*r6.l  || r1=[i1++]; // cgv
        r2.h = (a1 += r1.h*r7.l), r2.l = (a0 += r1.l*r7.l);
        r2 = byteop3p(r3:2, r1:0)(LO)                      || r5=[i1++m0]; // gmask, oy,cy,zero

        r2=r2>>16 || B[p1++]=r2;
                     B[p2++]=r2;

        r3=r3>>16 || B[p1++]=r3;
                     B[p2++]=r3                            || r1=[i1++]; // cy

        p1+=3;
        p2+=3;
        /* Y' = y*cy */
        a1 = r1.h*r4.h, a0 = r1.l*r4.l                     || r1=[i1++]; // crv

        /* R = Y+ crv*(Cr-128) */
        r2.h = (a1 += r1.h*r7.h), r2.l = (a0 += r1.l*r7.h);
                a1 -= r1.h*r7.h,          a0 -= r1.l*r7.h  || r5=[i1++]; // rmask
        r2 = byteop3p(r3:2, r1:0)(LO)                      || r1=[i1++]; // cbu
        r2=r2>>16 || B[p1++]=r2;
        B[p2++]=r2;

        /* B = Y+ cbu*(Cb-128) */
        r2.h = (a1 += r1.h*r6.h), r2.l = (a0 += r1.l*r6.h);
                a1 -= r1.h*r6.h,          a0 -= r1.l*r6.h  || r5=[i1++]; // bmask
        r3 = byteop3p(r3:2, r1:0)(LO)                      || r1=[i1++]; // cgu

        /* G = Y+ cgu*(Cb-128)+cgv*(Cr-128) */
                a1 += r1.h*r6.h,          a0 += r1.l*r6.h  || r1=[i1++]; // cgv
        r2.h = (a1 += r1.h*r7.h), r2.l = (a0 += r1.l*r7.h);
        r2 = byteop3p(r3:2, r1:0)(LO)                      || r5=[i1++]; // gmask
        r2=r2>>16 || B[p1++]=r2 || r0 = [i0++];    // 4y
                     B[p2++]=r2 || r1.l = w[i2++]; // 2u
        r3=r3>>16 || B[p1++]=r3 || r1.h = w[i3++]; // 2v
                     B[p2++]=r3 || r2=[i1++];      // oy

        p1+=3;
.L1888: p2+=3;

        l1 = 0;

        (r7:4) = [sp++];
        unlink;
        rts;
DEFUN_END(yuv2rgb24_line)



#define ARG_vdst        20
#define ARG_width       24
#define ARG_height      28
#define ARG_lumStride   32
#define ARG_chromStride 36
#define ARG_srcStride   40

DEFUN(uyvytoyv12, mL3,  (const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
                         long width, long height,
                         long lumStride, long chromStride, long srcStride)):
        link 0;
        [--sp] = (r7:4,p5:4);

        p0 = r1;       // Y top even

        i2 = r2; // *u
        r2 = [fp + ARG_vdst];
        i3 = r2; // *v

        r1 = [fp + ARG_srcStride];
        r2 = r0 + r1;
        r1 += -8;  // i0,i1 is pre read need to correct
        m0 = r1;

        i0 = r0;  // uyvy_T even
        i1 = r2;  // uyvy_B odd

        p2 = [fp + ARG_lumStride];
        p1 = p0 + p2;  // Y bot odd

        p5 = [fp + ARG_width];
        p4 = [fp + ARG_height];
        r0 = p5;
        p4 = p4 >> 1;
        p5 = p5 >> 2;

        r2 = [fp + ARG_chromStride];
        r0 = r0 >> 1;
        r2 = r2 - r0;
        m1 = r2;

        /*   I0,I1 - src input line pointers
         *   p0,p1 - luma output line pointers
         *   I2    - dstU
         *   I3    - dstV
         */

        lsetup (0f, 1f) lc1 = p4;   // H/2
0:        r0 = [i0++] || r2 = [i1++];
          r1 = [i0++] || r3 = [i1++];
          r4 = byteop1p(r1:0, r3:2);
          r5 = byteop1p(r1:0, r3:2) (r);
          lsetup (2f, 3f) lc0 = p5; // W/4
2:          r0 = r0 >> 8(v);
            r1 = r1 >> 8(v);
            r2 = r2 >> 8(v);
            r3 = r3 >> 8(v);
            r0 = bytepack(r0, r1);
            r2 = bytepack(r2, r3)         ||  [p0++] = r0;    // yyyy
            r6 = pack(r5.l, r4.l)         ||  [p1++] = r2;    // yyyy
            r7 = pack(r5.h, r4.h)         ||  r0 = [i0++] || r2 = [i1++];
            r6 = bytepack(r6, r7)         ||  r1 = [i0++] || r3 = [i1++];
            r4 = byteop1p(r1:0, r3:2)     ||  w[i2++] = r6.l; // uu
3:          r5 = byteop1p(r1:0, r3:2) (r) ||  w[i3++] = r6.h; // vv

          i0 += m0;
          i1 += m0;
          i2 += m1;
          i3 += m1;
          p0 = p0 + p2;
1:        p1 = p1 + p2;

        (r7:4,p5:4) = [sp++];
        unlink;
        rts;
DEFUN_END(uyvytoyv12)

DEFUN(yuyvtoyv12, mL3,  (const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
                         long width, long height,
                         long lumStride, long chromStride, long srcStride)):
        link 0;
        [--sp] = (r7:4,p5:4);

        p0 = r1;       // Y top even

        i2 = r2; // *u
        r2 = [fp + ARG_vdst];
        i3 = r2; // *v

        r1 = [fp + ARG_srcStride];
        r2 = r0 + r1;
        r1 += -8;  // i0,i1 is pre read need to correct
        m0 = r1;

        i0 = r0;  // uyvy_T even
        i1 = r2;  // uyvy_B odd

        p2 = [fp + ARG_lumStride];
        p1 = p0 + p2;  // Y bot odd

        p5 = [fp + ARG_width];
        p4 = [fp + ARG_height];
        r0 = p5;
        p4 = p4 >> 1;
        p5 = p5 >> 2;

        r2 = [fp + ARG_chromStride];
        r0 = r0 >> 1;
        r2 = r2 - r0;
        m1 = r2;

        /*   I0,I1 - src input line pointers
         *   p0,p1 - luma output line pointers
         *   I2    - dstU
         *   I3    - dstV
         */

        lsetup (0f, 1f) lc1 = p4;   // H/2
0:        r0 = [i0++] || r2 = [i1++];
          r1 = [i0++] || r3 = [i1++];
          r4 = bytepack(r0, r1);
          r5 = bytepack(r2, r3);
          lsetup (2f, 3f) lc0 = p5; // W/4
2:          r0 = r0 >> 8(v) || [p0++] = r4;  // yyyy-even
            r1 = r1 >> 8(v) || [p1++] = r5;  // yyyy-odd
            r2 = r2 >> 8(v);
            r3 = r3 >> 8(v);
            r4 = byteop1p(r1:0, r3:2);
            r5 = byteop1p(r1:0, r3:2) (r);
            r6 = pack(r5.l, r4.l);
            r7 = pack(r5.h, r4.h)         ||  r0 = [i0++] || r2 = [i1++];
            r6 = bytepack(r6, r7)         ||  r1 = [i0++] || r3 = [i1++];
            r4 = bytepack(r0, r1)         ||  w[i2++] = r6.l; // uu
3:          r5 = bytepack(r2, r3)         ||  w[i3++] = r6.h; // vv

          i0 += m0;
          i1 += m0;
          i2 += m1;
          i3 += m1;
          p0 = p0 + p2;
1:        p1 = p1 + p2;

        (r7:4,p5:4) = [sp++];
        unlink;
        rts;
DEFUN_END(yuyvtoyv12)