dsputil_mmx.c

Trolltech公司发布的图形界面操作系统。可在qt-embedded-2.3.10平台上编译为嵌入式图形界面操作系统。

	 : "+g"(h), "+r" (pixels),  "+r" (block)
	 : "r"(line_size)
	 : "%eax", "memory"
	);
}

static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
    __asm __volatile(
	 "lea (%3, %3), %%eax		\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"
	 "addl %%eax, %1		\n\t"
         "addl %%eax, %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"
	 "addl %%eax, %1		\n\t"
	 "addl %%eax, %2       		\n\t"
	 "subl $4, %0			\n\t"
	 "jnz 1b			\n\t"
	 : "+g"(h), "+r" (pixels),  "+r" (block)
	 : "r"(line_size)
	 : "%eax", "memory"
	);
}

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

#ifdef CONFIG_ENCODERS
static int pix_sum16_mmx(uint8_t * pix, int line_size){
    const int h=16;
    int sum;
    int 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"
                "addl %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" (line_size)
        );

        return sum;
}
#endif //CONFIG_ENCODERS

static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
    int 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"
        "addl $16, %0			\n\t"
        "cmpl %3, %0			\n\t"
        " jb 1b				\n\t"
        : "+r" (i)
        : "r"(src), "r"(dst), "r"(w-15)
    );
    for(; i<w; i++)
        dst[i+0] += src[i+0];
}

#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"

      "addl %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" (line_size) : "%ecx" );
    return tmp;
}

static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size) {
    int tmp;
  asm volatile (
      "movl $16,%%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 */
      "movq %%mm1,%%mm5\n"
      "movq %%mm3,%%mm6\n"
      "psubusb %%mm2,%%mm1\n"
      "psubusb %%mm4,%%mm3\n"
      "psubusb %%mm5,%%mm2\n"
      "psubusb %%mm6,%%mm4\n"

      "por %%mm1,%%mm2\n"
      "por %%mm3,%%mm4\n"

      /* now convert to 16-bit vectors so we can square them */
      "movq %%mm2,%%mm1\n"
      "movq %%mm4,%%mm3\n"

      "punpckhbw %%mm0,%%mm2\n"
      "punpckhbw %%mm0,%%mm4\n"
      "punpcklbw %%mm0,%%mm1\n"	/* mm1 now spread over (mm1,mm2) */
      "punpcklbw %%mm0,%%mm3\n"	/* mm4 now spread over (mm3,mm4) */

      "pmaddwd %%mm2,%%mm2\n"
      "pmaddwd %%mm4,%%mm4\n"
      "pmaddwd %%mm1,%%mm1\n"
      "pmaddwd %%mm3,%%mm3\n"

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

      "paddd %%mm2,%%mm1\n"
      "paddd %%mm4,%%mm3\n"
      "paddd %%mm1,%%mm7\n"
      "paddd %%mm3,%%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" (line_size) : "%ecx");
    return tmp;
}

static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
    int i=0;
    asm volatile(
        "1:				\n\t"
        "movq  (%2, %0), %%mm0		\n\t"
        "movq  (%1, %0), %%mm1		\n\t"
        "psubb %%mm0, %%mm1		\n\t"
        "movq %%mm1, (%3, %0)		\n\t"
        "movq 8(%2, %0), %%mm0		\n\t"
        "movq 8(%1, %0), %%mm1		\n\t"
        "psubb %%mm0, %%mm1		\n\t"
        "movq %%mm1, 8(%3, %0)		\n\t"
        "addl $16, %0			\n\t"
        "cmpl %4, %0			\n\t"
        " jb 1b				\n\t"
        : "+r" (i)
        : "r"(src1), "r"(src2), "r"(dst), "r"(w-15)
    );
    for(; i<w; i++)
        dst[i+0] = src1[i+0]-src2[i+0];
}

static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){
    int i=0;
    uint8_t l, lt;

    asm volatile(
        "1:				\n\t"
        "movq  -1(%1, %0), %%mm0	\n\t" // LT
        "movq  (%1, %0), %%mm1		\n\t" // T
        "movq  -1(%2, %0), %%mm2	\n\t" // L
        "movq  (%2, %0), %%mm3		\n\t" // X
        "movq %%mm2, %%mm4		\n\t" // L
        "psubb %%mm0, %%mm2		\n\t"
        "paddb %%mm1, %%mm2		\n\t" // L + T - LT
        "movq %%mm4, %%mm5		\n\t" // L
        "pmaxub %%mm1, %%mm4		\n\t" // max(T, L)
        "pminub %%mm5, %%mm1		\n\t" // min(T, L)
        "pminub %%mm2, %%mm4		\n\t" 
        "pmaxub %%mm1, %%mm4		\n\t"
        "psubb %%mm4, %%mm3		\n\t" // dst - pred
        "movq %%mm3, (%3, %0)		\n\t"
        "addl $8, %0			\n\t"
        "cmpl %4, %0			\n\t"
        " jb 1b				\n\t"
        : "+r" (i)
        : "r"(src1), "r"(src2), "r"(dst), "r"(w)
    );

    l= *left;
    lt= *left_top;
    
    dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);
    
    *left_top= src1[w-1];
    *left    = src2[w-1];
}

#define LBUTTERFLY2(a1,b1,a2,b2)\
    "paddw " #b1 ", " #a1 "		\n\t"\
    "paddw " #b2 ", " #a2 "		\n\t"\
    "paddw " #b1 ", " #b1 "		\n\t"\
    "paddw " #b2 ", " #b2 "		\n\t"\
    "psubw " #a1 ", " #b1 "		\n\t"\
    "psubw " #a2 ", " #b2 "		\n\t"

#define HADAMARD48\
        LBUTTERFLY2(%%mm0, %%mm1, %%mm2, %%mm3)\
        LBUTTERFLY2(%%mm4, %%mm5, %%mm6, %%mm7)\
        LBUTTERFLY2(%%mm0, %%mm2, %%mm1, %%mm3)\
        LBUTTERFLY2(%%mm4, %%mm6, %%mm5, %%mm7)\
        LBUTTERFLY2(%%mm0, %%mm4, %%mm1, %%mm5)\
        LBUTTERFLY2(%%mm2, %%mm6, %%mm3, %%mm7)\

#define MMABS(a,z)\
    "pxor " #z ", " #z "		\n\t"\
    "pcmpgtw " #a ", " #z "		\n\t"\
    "pxor " #z ", " #a "		\n\t"\
    "psubw " #z ", " #a "		\n\t"

#define MMABS_SUM(a,z, sum)\
    "pxor " #z ", " #z "		\n\t"\
    "pcmpgtw " #a ", " #z "		\n\t"\
    "pxor " #z ", " #a "		\n\t"\
    "psubw " #z ", " #a "		\n\t"\
    "paddusw " #a ", " #sum "		\n\t"

#define MMABS_MMX2(a,z)\
    "pxor " #z ", " #z "		\n\t"\
    "psubw " #a ", " #z "		\n\t"\
    "pmaxsw " #z ", " #a "		\n\t"

#define MMABS_SUM_MMX2(a,z, sum)\
    "pxor " #z ", " #z "		\n\t"\
    "psubw " #a ", " #z "		\n\t"\
    "pmaxsw " #z ", " #a "		\n\t"\
    "paddusw " #a ", " #sum "		\n\t"
        
#define SBUTTERFLY(a,b,t,n)\
    "movq " #a ", " #t "		\n\t" /* abcd */\
    "punpckl" #n " " #b ", " #a "	\n\t" /* aebf */\
    "punpckh" #n " " #b ", " #t "	\n\t" /* cgdh */\

#define TRANSPOSE4(a,b,c,d,t)\
    SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
    SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
    SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
    SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

#define LOAD4(o, a, b, c, d)\
        "movq "#o"(%1), " #a "		\n\t"\
        "movq "#o"+16(%1), " #b "	\n\t"\
        "movq "#o"+32(%1), " #c "	\n\t"\
        "movq "#o"+48(%1), " #d "	\n\t"

#define STORE4(o, a, b, c, d)\
        "movq "#a", "#o"(%1)		\n\t"\
        "movq "#b", "#o"+16(%1)		\n\t"\
        "movq "#c", "#o"+32(%1)		\n\t"\
        "movq "#d", "#o"+48(%1)		\n\t"\

static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride){
    uint64_t temp[16] __align8;
    int sum=0;

    diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);