reconstruct_mmx.h

一个很好用的MPEG1/4的开源编码器

/*
    libfame - Fast Assembly MPEG Encoder Library
    Copyright (C) 2000-2001 Vivien Chappelier

    This library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Library General Public
    License as published by the Free Software Foundation; either
    version 2 of the License, or (at your option) any later version.

    This library is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
    Library General Public License for more details.

    You should have received a copy of the GNU Library General Public
    License along with this library; if not, write to the Free
    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
/*************************** reconstruct reference frame *********************/

static void inline reconstruct(unsigned char *plane,
			       dct_t *block,
			       int pitch)
{
  int dummy;

  asm volatile ("movq 0x00(%1), %%mm0\n"
		"movq 0x08(%1), %%mm1\n"
		"movq 0x10(%1), %%mm2\n"
		"movq 0x18(%1), %%mm3\n"
		"movq 0x20(%1), %%mm4\n"
		"movq 0x28(%1), %%mm5\n"
		"movq 0x30(%1), %%mm6\n"
		"movq 0x38(%1), %%mm7\n"
		"psraw $0x04, %%mm0\n"
		"psraw $0x04, %%mm1\n"
		"psraw $0x04, %%mm2\n"
		"psraw $0x04, %%mm3\n"
		"psraw $0x04, %%mm4\n"
		"psraw $0x04, %%mm5\n"
		"psraw $0x04, %%mm6\n"
		"psraw $0x04, %%mm7\n"
		"packuswb %%mm1, %%mm0\n"
		"packuswb %%mm3, %%mm2\n"
		"packuswb %%mm5, %%mm4\n"
		"packuswb %%mm7, %%mm6\n"
		"movq %%mm0, (%0)\n"
		"addl %2, %0\n"
		"movq %%mm2, (%0)\n"
		"addl %2, %0\n"
		"movq %%mm4, (%0)\n"
		"addl %2, %0\n"
		"movq %%mm6, (%0)\n"
		"addl %2, %0\n"
		"movq 0x40(%1), %%mm0\n"
		"movq 0x48(%1), %%mm1\n"
		"movq 0x50(%1), %%mm2\n"
		"movq 0x58(%1), %%mm3\n"
		"movq 0x60(%1), %%mm4\n"
		"movq 0x68(%1), %%mm5\n"
		"movq 0x70(%1), %%mm6\n"
		"movq 0x78(%1), %%mm7\n"
		"psraw $0x04, %%mm0\n"
		"psraw $0x04, %%mm1\n"
		"psraw $0x04, %%mm2\n"
		"psraw $0x04, %%mm3\n"
		"psraw $0x04, %%mm4\n"
		"psraw $0x04, %%mm5\n"
		"psraw $0x04, %%mm6\n"
		"psraw $0x04, %%mm7\n"
		"packuswb %%mm1, %%mm0\n"
		"packuswb %%mm3, %%mm2\n"
		"packuswb %%mm5, %%mm4\n"
		"packuswb %%mm7, %%mm6\n"
		"movq %%mm0, (%0)\n"
		"addl %2, %0\n"
		"movq %%mm2, (%0)\n"
		"addl %2, %0\n"
		"movq %%mm4, (%0)\n"
		"addl %2, %0\n"
		"movq %%mm6, (%0)\n"
		: "=r"(dummy), "=r"(block), "=r"(pitch)
		: "0"(plane), "1"(block), "2"(pitch)
		: "memory");
}

static void inline sum(unsigned char *plane,
		       unsigned char *ref,
		       unsigned int *sum,
		       dct_t *block,
		       int pitch)
{
  int dummy;
  
#define SUM_STEP(x)							\
    "movq (%3), %%mm2\n"            /* mm2 = [ref]b */			\
    "movq 0x" #x "0(%1), %%mm0\n"   /* mm0 = [0-3]w */			\
    "movq 0x" #x "8(%1), %%mm1\n"   /* mm1 = [4-7]w */			\
    "movq %%mm2, %%mm5\n"           /* mm5 = [ref]b */			\
    "punpcklbw %%mm7, %%mm2\n"      /* mm2 = [ref0-3]w */		\
    "punpckhbw %%mm7, %%mm5\n"      /* mm5 = [ref4-7]w */		\
    "movq %%mm0, %%mm3\n"           /* copy [0-3] for rounding */	\
    "movq %%mm1, %%mm4\n"           /* copy [4-7] for rounding */	\
    "psraw $0x0f, %%mm3\n"          /* mm3 = (sign(mm0) - 1) / 2 */	\
    "psraw $0x0f, %%mm4\n"          /* mm4 = (sign(mm0) - 1) / 2 */	\
    "paddsw %%mm3, %%mm0\n"         /* adjust sign before rounding */	\
    "paddsw %%mm4, %%mm1\n"         /* adjust sign before rounding */	\
    "movq %%mm0, %%mm3\n"           /* copy [0-3] for rounding */	\
    "movq %%mm1, %%mm4\n"           /* copy [4-7] for rounding */	\
    "psllw $0x0c, %%mm3\n"          /* keep 4 bits right aligned */	\
    "psllw $0x0c, %%mm4\n"          /* keep 4 bits right aligned */	\
    "psrlw $0x0c, %%mm3\n"          /* keep only 'fixed point' part */	\
    "psrlw $0x0c, %%mm4\n"          /* keep only 'fixed point' part */	\
    "paddsw %%mm3, %%mm0\n"         /* add fixed point to number */	\
    "paddsw %%mm4, %%mm1\n"         /* add fixed point to number */	\
    "psraw $0x04, %%mm0\n"          /* mm0 /= 16 (rounded) */		\
    "psraw $0x04, %%mm1\n"          /* mm1 /= 16 (rounded) */		\
    "por %%mm0, %%mm6\n"            /* accumulate sum */		\
    "por %%mm1, %%mm6\n"            /* accumulate sum */		\
    "paddw %%mm0, %%mm2\n"          /* add to ref */			\
    "paddw %%mm1, %%mm5\n"          /* add to ref */			\
    "packuswb %%mm5, %%mm2\n"       /* pack to byte and saturate */	\
    "addl %2, %3\n"                 /* increment frame line */		\
    "movq %%mm2, (%0)\n"            /* store in frame */		\
    "addl %2, %0\n"                 /* increment frame line */

  asm volatile ("pxor %%mm7, %%mm7\n"      /* mm7 = zero */
		"pxor %%mm6, %%mm6\n"      /* mm6 = sum accumulator */
		SUM_STEP(0)
		SUM_STEP(1)
		SUM_STEP(2)
		SUM_STEP(3)
		SUM_STEP(4)
		SUM_STEP(5)
		SUM_STEP(6)
		SUM_STEP(7)
		: "=r"(dummy), "=r"(block), "=r"(pitch), "=r"(ref)
		: "0"(plane), "1"(block), "2"(pitch), "3"(ref)
		: "memory");

  asm volatile ("movq %%mm6, %%mm5\n"
		"psrlq $32, %%mm6\n"
		"por %%mm6, %%mm5\n"
		"movd %%mm5, %0\n"
		: "=r"(*sum)
		: "0"(*sum)
		);
}

static void inline move(unsigned char *plane,
			unsigned char *ref,
			int pitch)
{
  int dummy1, dummy2;

#define MOVE_STEP() 					\
  "movq (%2), %%mm0\n"      /* mm0 = [ref] */		\
  "addl %1, %2\n"           /* increment frame line */	\
  "movq %%mm0, (%0)\n"      /* store in frame */	\
  "addl %1, %0\n"           /* increment frame line */

  asm volatile (MOVE_STEP()
		MOVE_STEP()
		MOVE_STEP()
		MOVE_STEP()
		MOVE_STEP()
		MOVE_STEP()
		MOVE_STEP()
		MOVE_STEP()
		: "=r"(dummy1), "=r"(pitch), "=r"(dummy2)
		: "0"(plane), "1"(pitch), "2"(ref)
		: "memory");
}