dsputil_mmx.c

tcpmp播放器的flv插件

	 "add %%"REG_a", %2		\n\t"
	 "movq (%1), %%mm0		\n\t"
	 "movq (%1, %3), %%mm1		\n\t"
	 "movq %%mm0, (%2)		\n\t"
	 "movq %%mm1, (%2, %3)		\n\t"
	 "add %%"REG_a", %1		\n\t"
	 "add %%"REG_a", %2		\n\t"
	 "subl $4, %0			\n\t"
	 "jnz 1b			\n\t"
	 : "+g"(h), "+r" (pixels),  "+r" (block)
	 : "r"((long)line_size)
	 : "%"REG_a, "memory"
	);
}

static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
    __asm __volatile(
	 "lea (%3, %3), %%"REG_a"	\n\t"
	 ".balign 8			\n\t"
	 "1:				\n\t"
	 "movq (%1), %%mm0		\n\t"
	 "movq 8(%1), %%mm4		\n\t"
	 "movq (%1, %3), %%mm1		\n\t"
	 "movq 8(%1, %3), %%mm5		\n\t"
     	 "movq %%mm0, (%2)		\n\t"
     	 "movq %%mm4, 8(%2)		\n\t"
	 "movq %%mm1, (%2, %3)		\n\t"
	 "movq %%mm5, 8(%2, %3)		\n\t"
	 "add %%"REG_a", %1		\n\t"
	 "add %%"REG_a", %2       	\n\t"
	 "movq (%1), %%mm0		\n\t"
	 "movq 8(%1), %%mm4		\n\t"
	 "movq (%1, %3), %%mm1		\n\t"
	 "movq 8(%1, %3), %%mm5		\n\t"
	 "movq %%mm0, (%2)		\n\t"
	 "movq %%mm4, 8(%2)		\n\t"
	 "movq %%mm1, (%2, %3)		\n\t"
	 "movq %%mm5, 8(%2, %3)		\n\t"
	 "add %%"REG_a", %1		\n\t"
	 "add %%"REG_a", %2       	\n\t"
	 "subl $4, %0			\n\t"
	 "jnz 1b			\n\t"
	 : "+g"(h), "+r" (pixels),  "+r" (block)
	 : "r"((long)line_size)
	 : "%"REG_a, "memory"
	);
}

static void clear_blocks_mmx(DCTELEM *blocks)
{
    __asm __volatile(
                "pxor %%mm7, %%mm7		\n\t"
                "mov $-128*6, %%"REG_a"	\n\t"
                "1:				\n\t"
                "movq %%mm7, (%0, %%"REG_a")	\n\t"
                "movq %%mm7, 8(%0, %%"REG_a")	\n\t"
                "movq %%mm7, 16(%0, %%"REG_a")	\n\t"
                "movq %%mm7, 24(%0, %%"REG_a")	\n\t"
                "add $32, %%"REG_a"		\n\t"
                " js 1b				\n\t"
                : : "r" (((uint8_t *)blocks)+128*6)
                : "%"REG_a
        );
}

#ifdef CONFIG_ENCODERS
static int pix_sum16_mmx(uint8_t * pix, int line_size){
    const int h=16;
    int sum;
    long index= -line_size*h;

    __asm __volatile(
                "pxor %%mm7, %%mm7		\n\t"
                "pxor %%mm6, %%mm6		\n\t"
                "1:				\n\t"
                "movq (%2, %1), %%mm0		\n\t"
                "movq (%2, %1), %%mm1		\n\t"
                "movq 8(%2, %1), %%mm2		\n\t"
                "movq 8(%2, %1), %%mm3		\n\t"
                "punpcklbw %%mm7, %%mm0		\n\t"
                "punpckhbw %%mm7, %%mm1		\n\t"
                "punpcklbw %%mm7, %%mm2		\n\t"
                "punpckhbw %%mm7, %%mm3		\n\t"
                "paddw %%mm0, %%mm1		\n\t"
                "paddw %%mm2, %%mm3		\n\t"
                "paddw %%mm1, %%mm3		\n\t"
                "paddw %%mm3, %%mm6		\n\t"
                "add %3, %1			\n\t"
                " js 1b				\n\t"
                "movq %%mm6, %%mm5		\n\t"
                "psrlq $32, %%mm6		\n\t"
                "paddw %%mm5, %%mm6		\n\t"
                "movq %%mm6, %%mm5		\n\t"
                "psrlq $16, %%mm6		\n\t"
                "paddw %%mm5, %%mm6		\n\t"
                "movd %%mm6, %0			\n\t"
                "andl $0xFFFF, %0		\n\t"
                : "=&r" (sum), "+r" (index)
                : "r" (pix - index), "r" ((long)line_size)
        );

        return sum;
}
#endif //CONFIG_ENCODERS

static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
    long i=0;
    asm volatile(
        "1:				\n\t"
        "movq  (%1, %0), %%mm0		\n\t"
        "movq  (%2, %0), %%mm1		\n\t"
        "paddb %%mm0, %%mm1		\n\t"
        "movq %%mm1, (%2, %0)		\n\t"
        "movq 8(%1, %0), %%mm0		\n\t"
        "movq 8(%2, %0), %%mm1		\n\t"
        "paddb %%mm0, %%mm1		\n\t"
        "movq %%mm1, 8(%2, %0)		\n\t"
        "add $16, %0			\n\t"
        "cmp %3, %0			\n\t"
        " jb 1b				\n\t"
        : "+r" (i)
        : "r"(src), "r"(dst), "r"((long)w-15)
    );
    for(; i<w; i++)
        dst[i+0] += src[i+0];
}

#define H263_LOOP_FILTER \
        "pxor %%mm7, %%mm7		\n\t"\
        "movq  %0, %%mm0		\n\t"\
        "movq  %0, %%mm1		\n\t"\
        "movq  %3, %%mm2		\n\t"\
        "movq  %3, %%mm3		\n\t"\
        "punpcklbw %%mm7, %%mm0		\n\t"\
        "punpckhbw %%mm7, %%mm1		\n\t"\
        "punpcklbw %%mm7, %%mm2		\n\t"\
        "punpckhbw %%mm7, %%mm3		\n\t"\
        "psubw %%mm2, %%mm0		\n\t"\
        "psubw %%mm3, %%mm1		\n\t"\
        "movq  %1, %%mm2		\n\t"\
        "movq  %1, %%mm3		\n\t"\
        "movq  %2, %%mm4		\n\t"\
        "movq  %2, %%mm5		\n\t"\
        "punpcklbw %%mm7, %%mm2		\n\t"\
        "punpckhbw %%mm7, %%mm3		\n\t"\
        "punpcklbw %%mm7, %%mm4		\n\t"\
        "punpckhbw %%mm7, %%mm5		\n\t"\
        "psubw %%mm2, %%mm4		\n\t"\
        "psubw %%mm3, %%mm5		\n\t"\
        "psllw $2, %%mm4		\n\t"\
        "psllw $2, %%mm5		\n\t"\
        "paddw %%mm0, %%mm4		\n\t"\
        "paddw %%mm1, %%mm5		\n\t"\
        "pxor %%mm6, %%mm6		\n\t"\
        "pcmpgtw %%mm4, %%mm6		\n\t"\
        "pcmpgtw %%mm5, %%mm7		\n\t"\
        "pxor %%mm6, %%mm4		\n\t"\
        "pxor %%mm7, %%mm5		\n\t"\
        "psubw %%mm6, %%mm4		\n\t"\
        "psubw %%mm7, %%mm5		\n\t"\
        "psrlw $3, %%mm4		\n\t"\
        "psrlw $3, %%mm5		\n\t"\
        "packuswb %%mm5, %%mm4		\n\t"\
        "packsswb %%mm7, %%mm6		\n\t"\
        "pxor %%mm7, %%mm7		\n\t"\
        "movd %4, %%mm2			\n\t"\
        "punpcklbw %%mm2, %%mm2		\n\t"\
        "punpcklbw %%mm2, %%mm2		\n\t"\
        "punpcklbw %%mm2, %%mm2		\n\t"\
        "psubusb %%mm4, %%mm2		\n\t"\
        "movq %%mm2, %%mm3		\n\t"\
        "psubusb %%mm4, %%mm3		\n\t"\
        "psubb %%mm3, %%mm2		\n\t"\
        "movq %1, %%mm3			\n\t"\
        "movq %2, %%mm4			\n\t"\
        "pxor %%mm6, %%mm3		\n\t"\
        "pxor %%mm6, %%mm4		\n\t"\
        "paddusb %%mm2, %%mm3		\n\t"\
        "psubusb %%mm2, %%mm4		\n\t"\
        "pxor %%mm6, %%mm3		\n\t"\
        "pxor %%mm6, %%mm4		\n\t"\
        "paddusb %%mm2, %%mm2		\n\t"\
        "packsswb %%mm1, %%mm0		\n\t"\
        "pcmpgtb %%mm0, %%mm7		\n\t"\
        "pxor %%mm7, %%mm0		\n\t"\
        "psubb %%mm7, %%mm0		\n\t"\
        "movq %%mm0, %%mm1		\n\t"\
        "psubusb %%mm2, %%mm0		\n\t"\
        "psubb %%mm0, %%mm1		\n\t"\
        "pand %5, %%mm1			\n\t"\
        "psrlw $2, %%mm1		\n\t"\
        "pxor %%mm7, %%mm1		\n\t"\
        "psubb %%mm7, %%mm1		\n\t"\
        "movq %0, %%mm5			\n\t"\
        "movq %3, %%mm6			\n\t"\
        "psubb %%mm1, %%mm5		\n\t"\
        "paddb %%mm1, %%mm6		\n\t"

static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
    const int strength= ff_h263_loop_filter_strength[qscale];

    asm volatile(
    
        H263_LOOP_FILTER
        
        "movq %%mm3, %1			\n\t"
        "movq %%mm4, %2			\n\t"
        "movq %%mm5, %0			\n\t"
        "movq %%mm6, %3			\n\t"
        : "+m" (*(uint64_t*)(src - 2*stride)),
          "+m" (*(uint64_t*)(src - 1*stride)),
          "+m" (*(uint64_t*)(src + 0*stride)),
          "+m" (*(uint64_t*)(src + 1*stride))
        : "g" (2*strength), "m"(ff_pb_FC)
    );
}

static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){
    asm volatile( //FIXME could save 1 instruction if done as 8x4 ...
        "movd  %4, %%mm0		\n\t"
        "movd  %5, %%mm1		\n\t"
        "movd  %6, %%mm2		\n\t"
        "movd  %7, %%mm3		\n\t"
        "punpcklbw %%mm1, %%mm0		\n\t"
        "punpcklbw %%mm3, %%mm2		\n\t"
        "movq %%mm0, %%mm1		\n\t"
        "punpcklwd %%mm2, %%mm0		\n\t"
        "punpckhwd %%mm2, %%mm1		\n\t"
        "movd  %%mm0, %0		\n\t"
        "punpckhdq %%mm0, %%mm0		\n\t"
        "movd  %%mm0, %1		\n\t"
        "movd  %%mm1, %2		\n\t"
        "punpckhdq %%mm1, %%mm1		\n\t"
        "movd  %%mm1, %3		\n\t"
        
        : "=m" (*(uint32_t*)(dst + 0*dst_stride)),
          "=m" (*(uint32_t*)(dst + 1*dst_stride)),
          "=m" (*(uint32_t*)(dst + 2*dst_stride)),
          "=m" (*(uint32_t*)(dst + 3*dst_stride))
        :  "m" (*(uint32_t*)(src + 0*src_stride)),
           "m" (*(uint32_t*)(src + 1*src_stride)),
           "m" (*(uint32_t*)(src + 2*src_stride)),
           "m" (*(uint32_t*)(src + 3*src_stride))
    );
}

static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
    const int strength= ff_h263_loop_filter_strength[qscale];
    uint64_t temp[4] __attribute__ ((aligned(8)));
    uint8_t *btemp= (uint8_t*)temp;
    
    src -= 2;

    transpose4x4(btemp  , src           , 8, stride);
    transpose4x4(btemp+4, src + 4*stride, 8, stride);
    asm volatile(
        H263_LOOP_FILTER // 5 3 4 6
        
        : "+m" (temp[0]),
          "+m" (temp[1]),
          "+m" (temp[2]),
          "+m" (temp[3])
        : "g" (2*strength), "m"(ff_pb_FC)
    );

    asm volatile(
        "movq %%mm5, %%mm1		\n\t"
        "movq %%mm4, %%mm0		\n\t"
        "punpcklbw %%mm3, %%mm5		\n\t"
        "punpcklbw %%mm6, %%mm4		\n\t"
        "punpckhbw %%mm3, %%mm1		\n\t"
        "punpckhbw %%mm6, %%mm0		\n\t"
        "movq %%mm5, %%mm3		\n\t"
        "movq %%mm1, %%mm6		\n\t"
        "punpcklwd %%mm4, %%mm5		\n\t"
        "punpcklwd %%mm0, %%mm1		\n\t"
        "punpckhwd %%mm4, %%mm3		\n\t"
        "punpckhwd %%mm0, %%mm6		\n\t"
        "movd %%mm5, (%0)		\n\t"
        "punpckhdq %%mm5, %%mm5		\n\t"
        "movd %%mm5, (%0,%2)		\n\t"
        "movd %%mm3, (%0,%2,2)		\n\t"
        "punpckhdq %%mm3, %%mm3		\n\t"
        "movd %%mm3, (%0,%3)		\n\t"
        "movd %%mm1, (%1)		\n\t"
        "punpckhdq %%mm1, %%mm1		\n\t"
        "movd %%mm1, (%1,%2)		\n\t"
        "movd %%mm6, (%1,%2,2)		\n\t"
        "punpckhdq %%mm6, %%mm6		\n\t"
        "movd %%mm6, (%1,%3)		\n\t"
        :: "r" (src),
           "r" (src + 4*stride),
           "r" ((long)   stride ),
           "r" ((long)(3*stride))
    );
}

#ifdef CONFIG_ENCODERS
static int pix_norm1_mmx(uint8_t *pix, int line_size) {
    int tmp;
  asm volatile (
      "movl $16,%%ecx\n"
      "pxor %%mm0,%%mm0\n"
      "pxor %%mm7,%%mm7\n"
      "1:\n"
      "movq (%0),%%mm2\n"	/* mm2 = pix[0-7] */
      "movq 8(%0),%%mm3\n"	/* mm3 = pix[8-15] */

      "movq %%mm2,%%mm1\n"	/* mm1 = mm2 = pix[0-7] */

      "punpckhbw %%mm0,%%mm1\n"	/* mm1 = [pix4-7] */
      "punpcklbw %%mm0,%%mm2\n"	/* mm2 = [pix0-3] */

      "movq %%mm3,%%mm4\n"	/* mm4 = mm3 = pix[8-15] */
      "punpckhbw %%mm0,%%mm3\n"	/* mm3 = [pix12-15] */
      "punpcklbw %%mm0,%%mm4\n"	/* mm4 = [pix8-11] */

      "pmaddwd %%mm1,%%mm1\n"	/* mm1 = (pix0^2+pix1^2,pix2^2+pix3^2) */
      "pmaddwd %%mm2,%%mm2\n"	/* mm2 = (pix4^2+pix5^2,pix6^2+pix7^2) */

      "pmaddwd %%mm3,%%mm3\n"
      "pmaddwd %%mm4,%%mm4\n"

      "paddd %%mm1,%%mm2\n"	/* mm2 = (pix0^2+pix1^2+pix4^2+pix5^2,
					  pix2^2+pix3^2+pix6^2+pix7^2) */
      "paddd %%mm3,%%mm4\n"
      "paddd %%mm2,%%mm7\n"

      "add %2, %0\n"
      "paddd %%mm4,%%mm7\n"
      "dec %%ecx\n"
      "jnz 1b\n"

      "movq %%mm7,%%mm1\n"
      "psrlq $32, %%mm7\n"	/* shift hi dword to lo */
      "paddd %%mm7,%%mm1\n"
      "movd %%mm1,%1\n"
      : "+r" (pix), "=r"(tmp) : "r" ((long)line_size) : "%ecx" );
    return tmp;
}

static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    int tmp;
  asm volatile (
      "movl %4,%%ecx\n"
      "pxor %%mm0,%%mm0\n"	/* mm0 = 0 */
      "pxor %%mm7,%%mm7\n"	/* mm7 holds the sum */
      "1:\n"
      "movq (%0),%%mm1\n"	/* mm1 = pix1[0-7] */
      "movq (%1),%%mm2\n"	/* mm2 = pix2[0-7] */

      "movq %%mm1,%%mm5\n"
      "psubusb %%mm2,%%mm1\n"
      "psubusb %%mm5,%%mm2\n"

      "por %%mm1,%%mm2\n"

      "movq %%mm2,%%mm1\n"

      "punpckhbw %%mm0,%%mm2\n"
      "punpcklbw %%mm0,%%mm1\n"	/* mm1 now spread over (mm1,mm2) */

      "pmaddwd %%mm2,%%mm2\n"
      "pmaddwd %%mm1,%%mm1\n"

      "add %3,%0\n"
      "add %3,%1\n"

      "paddd %%mm2,%%mm1\n"
      "paddd %%mm1,%%mm7\n"

      "decl %%ecx\n"
      "jnz 1b\n"

      "movq %%mm7,%%mm1\n"
      "psrlq $32, %%mm7\n"	/* shift hi dword to lo */
      "paddd %%mm7,%%mm1\n"
      "movd %%mm1,%2\n"
      : "+r" (pix1), "+r" (pix2), "=r"(tmp) 
      : "r" ((long)line_size) , "m" (h)
      : "%ecx");
    return tmp;
}

static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    int tmp;
  asm volatile (
      "movl %4,%%ecx\n"
      "pxor %%mm0,%%mm0\n"	/* mm0 = 0 */
      "pxor %%mm7,%%mm7\n"	/* mm7 holds the sum */
      "1:\n"
      "movq (%0),%%mm1\n"	/* mm1 = pix1[0-7] */
      "movq (%1),%%mm2\n"	/* mm2 = pix2[0-7] */
      "movq 8(%0),%%mm3\n"	/* mm3 = pix1[8-15] */
      "movq 8(%1),%%mm4\n"	/* mm4 = pix2[8-15] */

      /* todo: mm1-mm2, mm3-mm4 */
      /* algo: substract mm1 from mm2 with saturation and vice versa */
      /*       OR the results to get absolute difference */