dsputilenc_mmx.c

ffmpeg移植到symbian的全部源代码

/*
 * MMX optimized DSP utils
 * Copyright (c) 2000, 2001 Fabrice Bellard.
 * Copyright (c) 2002-2004 Michael Niedermayer <[e-mail]>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 *
 * MMX optimization by Nick Kurshev <[e-mail]>
 */

#include "libavutil/x86_cpu.h"
#include "libavcodec/dsputil.h"
#include "libavcodec/mpegvideo.h"
#include "dsputil_mmx.h"


static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
{
    asm volatile(
        "mov $-128, %%"REG_a"           \n\t"
        "pxor %%mm7, %%mm7              \n\t"
        ASMALIGN(4)
        "1:                             \n\t"
        "movq (%0), %%mm0               \n\t"
        "movq (%0, %2), %%mm2           \n\t"
        "movq %%mm0, %%mm1              \n\t"
        "movq %%mm2, %%mm3              \n\t"
        "punpcklbw %%mm7, %%mm0         \n\t"
        "punpckhbw %%mm7, %%mm1         \n\t"
        "punpcklbw %%mm7, %%mm2         \n\t"
        "punpckhbw %%mm7, %%mm3         \n\t"
        "movq %%mm0, (%1, %%"REG_a")    \n\t"
        "movq %%mm1, 8(%1, %%"REG_a")   \n\t"
        "movq %%mm2, 16(%1, %%"REG_a")  \n\t"
        "movq %%mm3, 24(%1, %%"REG_a")  \n\t"
        "add %3, %0                     \n\t"
        "add $32, %%"REG_a"             \n\t"
        "js 1b                          \n\t"
        : "+r" (pixels)
        : "r" (block+64), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*2)
        : "%"REG_a
    );
}

static inline void diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride)
{
    asm volatile(
        "pxor %%mm7, %%mm7              \n\t"
        "mov $-128, %%"REG_a"           \n\t"
        ASMALIGN(4)
        "1:                             \n\t"
        "movq (%0), %%mm0               \n\t"
        "movq (%1), %%mm2               \n\t"
        "movq %%mm0, %%mm1              \n\t"
        "movq %%mm2, %%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 %%mm0, (%2, %%"REG_a")    \n\t"
        "movq %%mm1, 8(%2, %%"REG_a")   \n\t"
        "add %3, %0                     \n\t"
        "add %3, %1                     \n\t"
        "add $16, %%"REG_a"             \n\t"
        "jnz 1b                         \n\t"
        : "+r" (s1), "+r" (s2)
        : "r" (block+64), "r" ((x86_reg)stride)
        : "%"REG_a
    );
}

static int pix_sum16_mmx(uint8_t * pix, int line_size){
    const int h=16;
    int sum;
    x86_reg 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" ((x86_reg)line_size)
        );

        return sum;
}

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" ((x86_reg)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"
      "shr $1,%%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][0-7] */
      "movq (%1),%%mm2\n"       /* mm2 = pix2[0][0-7] */
      "movq (%0,%3),%%mm3\n"    /* mm3 = pix1[1][0-7] */
      "movq (%1,%3),%%mm4\n"    /* mm4 = pix2[1][0-7] */

      /* todo: mm1-mm2, mm3-mm4 */
      /* algo: subtract 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"

      "lea (%0,%3,2), %0\n"     /* pix1 += 2*line_size */
      "lea (%1,%3,2), %1\n"     /* pix2 += 2*line_size */

      "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" ((x86_reg)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: subtract 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"

      "add %3,%0\n"
      "add %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" ((x86_reg)line_size) , "m" (h)
      : "%ecx");
    return tmp;
}

static int sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    int tmp;
  asm volatile (
      "shr $1,%2\n"
      "pxor %%xmm0,%%xmm0\n"    /* mm0 = 0 */
      "pxor %%xmm7,%%xmm7\n"    /* mm7 holds the sum */
      "1:\n"
      "movdqu (%0),%%xmm1\n"    /* mm1 = pix1[0][0-15] */
      "movdqu (%1),%%xmm2\n"    /* mm2 = pix2[0][0-15] */
      "movdqu (%0,%4),%%xmm3\n" /* mm3 = pix1[1][0-15] */
      "movdqu (%1,%4),%%xmm4\n" /* mm4 = pix2[1][0-15] */

      /* todo: mm1-mm2, mm3-mm4 */
      /* algo: subtract mm1 from mm2 with saturation and vice versa */
      /*       OR the results to get absolute difference */
      "movdqa %%xmm1,%%xmm5\n"
      "movdqa %%xmm3,%%xmm6\n"
      "psubusb %%xmm2,%%xmm1\n"
      "psubusb %%xmm4,%%xmm3\n"
      "psubusb %%xmm5,%%xmm2\n"
      "psubusb %%xmm6,%%xmm4\n"

      "por %%xmm1,%%xmm2\n"
      "por %%xmm3,%%xmm4\n"

      /* now convert to 16-bit vectors so we can square them */
      "movdqa %%xmm2,%%xmm1\n"
      "movdqa %%xmm4,%%xmm3\n"

      "punpckhbw %%xmm0,%%xmm2\n"
      "punpckhbw %%xmm0,%%xmm4\n"
      "punpcklbw %%xmm0,%%xmm1\n"  /* mm1 now spread over (mm1,mm2) */
      "punpcklbw %%xmm0,%%xmm3\n"  /* mm4 now spread over (mm3,mm4) */

      "pmaddwd %%xmm2,%%xmm2\n"
      "pmaddwd %%xmm4,%%xmm4\n"
      "pmaddwd %%xmm1,%%xmm1\n"
      "pmaddwd %%xmm3,%%xmm3\n"

      "lea (%0,%4,2), %0\n"        /* pix1 += 2*line_size */
      "lea (%1,%4,2), %1\n"        /* pix2 += 2*line_size */

      "paddd %%xmm2,%%xmm1\n"
      "paddd %%xmm4,%%xmm3\n"
      "paddd %%xmm1,%%xmm7\n"
      "paddd %%xmm3,%%xmm7\n"

      "decl %2\n"
      "jnz 1b\n"

      "movdqa %%xmm7,%%xmm1\n"
      "psrldq $8, %%xmm7\n"        /* shift hi qword to lo */
      "paddd %%xmm1,%%xmm7\n"
      "movdqa %%xmm7,%%xmm1\n"
      "psrldq $4, %%xmm7\n"        /* shift hi dword to lo */
      "paddd %%xmm1,%%xmm7\n"
      "movd %%xmm7,%3\n"
      : "+r" (pix1), "+r" (pix2), "+r"(h), "=r"(tmp)
      : "r" ((x86_reg)line_size));
    return tmp;
}

static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
    int tmp;
  asm volatile (
      "movl %3,%%ecx\n"
      "pxor %%mm7,%%mm7\n"
      "pxor %%mm6,%%mm6\n"

      "movq (%0),%%mm0\n"
      "movq %%mm0, %%mm1\n"
      "psllq $8, %%mm0\n"
      "psrlq $8, %%mm1\n"
      "psrlq $8, %%mm0\n"
      "movq %%mm0, %%mm2\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm0\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm2\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm0\n"
      "psubw %%mm3, %%mm2\n"

      "add %2,%0\n"

      "movq (%0),%%mm4\n"
      "movq %%mm4, %%mm1\n"
      "psllq $8, %%mm4\n"
      "psrlq $8, %%mm1\n"
      "psrlq $8, %%mm4\n"
      "movq %%mm4, %%mm5\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm4\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm5\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm4\n"
      "psubw %%mm3, %%mm5\n"
      "psubw %%mm4, %%mm0\n"
      "psubw %%mm5, %%mm2\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm0, %%mm3\n\t"
      "pcmpgtw %%mm2, %%mm1\n\t"
      "pxor %%mm3, %%mm0\n"
      "pxor %%mm1, %%mm2\n"
      "psubw %%mm3, %%mm0\n"
      "psubw %%mm1, %%mm2\n"
      "paddw %%mm0, %%mm2\n"
      "paddw %%mm2, %%mm6\n"

      "add %2,%0\n"
      "1:\n"

      "movq (%0),%%mm0\n"
      "movq %%mm0, %%mm1\n"
      "psllq $8, %%mm0\n"
      "psrlq $8, %%mm1\n"
      "psrlq $8, %%mm0\n"
      "movq %%mm0, %%mm2\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm0\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm2\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm0\n"
      "psubw %%mm3, %%mm2\n"
      "psubw %%mm0, %%mm4\n"
      "psubw %%mm2, %%mm5\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm4, %%mm3\n\t"
      "pcmpgtw %%mm5, %%mm1\n\t"
      "pxor %%mm3, %%mm4\n"
      "pxor %%mm1, %%mm5\n"
      "psubw %%mm3, %%mm4\n"
      "psubw %%mm1, %%mm5\n"
      "paddw %%mm4, %%mm5\n"
      "paddw %%mm5, %%mm6\n"

      "add %2,%0\n"

      "movq (%0),%%mm4\n"
      "movq %%mm4, %%mm1\n"
      "psllq $8, %%mm4\n"
      "psrlq $8, %%mm1\n"
      "psrlq $8, %%mm4\n"
      "movq %%mm4, %%mm5\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm4\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm5\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm4\n"
      "psubw %%mm3, %%mm5\n"
      "psubw %%mm4, %%mm0\n"
      "psubw %%mm5, %%mm2\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm0, %%mm3\n\t"
      "pcmpgtw %%mm2, %%mm1\n\t"
      "pxor %%mm3, %%mm0\n"
      "pxor %%mm1, %%mm2\n"
      "psubw %%mm3, %%mm0\n"
      "psubw %%mm1, %%mm2\n"
      "paddw %%mm0, %%mm2\n"
      "paddw %%mm2, %%mm6\n"

      "add %2,%0\n"
      "subl $2, %%ecx\n"
      " jnz 1b\n"

      "movq %%mm6, %%mm0\n"
      "punpcklwd %%mm7,%%mm0\n"
      "punpckhwd %%mm7,%%mm6\n"
      "paddd %%mm0, %%mm6\n"

      "movq %%mm6,%%mm0\n"
      "psrlq $32, %%mm6\n"
      "paddd %%mm6,%%mm0\n"