git.maemo.org Git - mafwsubrenderer/blob - gst-plugins-base-subtitles0.10/debian/patches/0015-ARM-NEON-optimized-version-of-uyvy422_to_yuv420p-fun.patch

   1 From 04c686998b70b616d6c3a13cc694cf0811ba8650 Mon Sep 17 00:00:00 2001
   2 From: Siarhei Siamashka <siarhei.siamashka@nokia.com>
   3 Date: Tue, 14 Jul 2009 21:50:18 +0300
   4 Subject: [PATCH] ARM NEON optimized version of uyvy422_to_yuv420p function
   5
   6 ---
   7  gst/ffmpegcolorspace/imgconvert.c |  153 +++++++++++++++++++++++++++++++++++++
   8  1 files changed, 153 insertions(+), 0 deletions(-)
   9
  10 diff --git a/gst/ffmpegcolorspace/imgconvert.c b/gst/ffmpegcolorspace/imgconvert.c
  11 index 2436a20..67cb19d 100644
  12 --- a/gst/ffmpegcolorspace/imgconvert.c
  13 +++ b/gst/ffmpegcolorspace/imgconvert.c
  14 @@ -956,6 +956,158 @@ uyvy422_to_rgb24_neon (AVPicture * dst, const AVPicture * src,
  15    }
  16  }
  17
  18 +/*
  19 + * ARM NEON optimized implementation of UYVY -> YV12 convertor
  20 + */
  21 +static void
  22 +uyvy422_to_yuv420p_neon (AVPicture * dst, const AVPicture * src,
  23 +    int width, int height)
  24 +{
  25 +  asm volatile
  26 +      (".macro convert_macroblock_uyvy422_to_yuv420p size, store_c, src, lum, cb, cr\n"
  27 +      /* load up to 16 source pixels in UYVY format */
  28 +      ".if \\size == 16\n" "  vld1.8      {d0, d1, d2, d3}, [\\src]!\n" "  pld [\\src, #256]\n" ".elseif \\size == 8\n" "  vld1.8      {d0, d1}, [\\src]!\n" ".elseif \\size == 4\n" "  vld1.8      {d0}, [\\src]!\n" ".elseif \\size == 2\n" "  vld1.8      {d0[0]}, [\\src]!\n" "  vld1.8      {d0[1]}, [\\src]!\n" "  vld1.8      {d0[2]}, [\\src]!\n" "  vld1.8      {d0[3]}, [\\src]!\n" ".elseif \\size == 1\n" "  vld1.8      {d0[0]}, [\\src]!\n" "  vld1.8      {d0[1]}, [\\src]!\n" "  vld1.8      {d0[2]}, [\\src]!\n" ".else\n" "  .error \"unsupported macroblock size\"\n" ".endif\n" "  vuzp.8      d0, d1\n"    /* d1 - separated Y (first 8 bytes) */
  29 +      "  vuzp.8      d2, d3\n"  /* d3 - separated Y (next 8 bytes) */
  30 +      "  vuzp.8      d0, d2\n"  /* d0 - separated U, d2 - separated V */
  31 +      "  vswp        d1, d2\n"  /* exchange d1 and d2 */
  32 +      ".if \\size == 16\n"
  33 +      "  vst1.8      {d2, d3}, [\\lum]!\n"
  34 +      ".if \\store_c\n"
  35 +      "  vst1.8      {d0}, [\\cb]!\n"
  36 +      "  vst1.8      {d1}, [\\cr]!\n"
  37 +      ".endif\n"
  38 +      ".elseif \\size == 8\n"
  39 +      "  vst1.8      {d2}, [\\lum]!\n"
  40 +      ".if \\store_c\n"
  41 +      "  vst1.8      {d0[0]}, [\\cb]!\n"
  42 +      "  vst1.8      {d0[1]}, [\\cb]!\n"
  43 +      "  vst1.8      {d0[2]}, [\\cb]!\n"
  44 +      "  vst1.8      {d0[3]}, [\\cb]!\n"
  45 +      "  vst1.8      {d1[0]}, [\\cr]!\n"
  46 +      "  vst1.8      {d1[1]}, [\\cr]!\n"
  47 +      "  vst1.8      {d1[2]}, [\\cr]!\n"
  48 +      "  vst1.8      {d1[3]}, [\\cr]!\n"
  49 +      ".endif\n"
  50 +      ".elseif \\size == 4\n"
  51 +      "  vst1.8      {d2[0]}, [\\lum]!\n"
  52 +      "  vst1.8      {d2[1]}, [\\lum]!\n"
  53 +      "  vst1.8      {d2[2]}, [\\lum]!\n"
  54 +      "  vst1.8      {d2[3]}, [\\lum]!\n"
  55 +      ".if \\store_c\n"
  56 +      "  vst1.8      {d0[0]}, [\\cb]!\n"
  57 +      "  vst1.8      {d0[1]}, [\\cb]!\n"
  58 +      "  vst1.8      {d1[0]}, [\\cr]!\n"
  59 +      "  vst1.8      {d1[1]}, [\\cr]!\n"
  60 +      ".endif\n"
  61 +      ".elseif \\size == 2\n"
  62 +      "  vst1.8      {d2[0]}, [\\lum]!\n"
  63 +      "  vst1.8      {d2[1]}, [\\lum]!\n"
  64 +      ".if \\store_c\n"
  65 +      "  vst1.8      {d0[0]}, [\\cb]!\n"
  66 +      "  vst1.8      {d1[0]}, [\\cr]!\n"
  67 +      ".endif\n"
  68 +      ".elseif \\size == 1\n"
  69 +      "  vst1.8      {d2[0]}, [\\lum]!\n"
  70 +      ".if \\store_c\n"
  71 +      "  vst1.8      {d0[0]}, [\\cb]!\n"
  72 +      "  vst1.8      {d1[0]}, [\\cr]!\n"
  73 +      ".endif\n"
  74 +      ".else\n"
  75 +      "  .error \"unsupported macroblock size\"\n" ".endif\n" ".endm\n");
  76 +
  77 +  const uint8_t *p, *p1;
  78 +  uint8_t *lum, *cr, *cb, *lum1, *cr1, *cb1;
  79 +  int w;
  80 +
  81 +  p1 = src->data[0];
  82 +
  83 +  lum1 = dst->data[0];
  84 +  cb1 = dst->data[1];
  85 +  cr1 = dst->data[2];
  86 +
  87 +  for (; height >= 1; height -= 2) {
  88 +    p = p1;
  89 +    lum = lum1;
  90 +    cb = cb1;
  91 +    cr = cr1;
  92 +    w = width;
  93 +#if 1
  94 +    asm volatile ("   subs        %[w], %[w], #16\n"
  95 +        "   blt         2f\n"
  96 +        "1:\n"
  97 +        "   convert_macroblock_uyvy422_to_yuv420p 16, 1, %[p], %[lum], %[cb], %[cr]\n"
  98 +        "   subs        %[w], %[w], #16\n"
  99 +        "   bge         1b\n"
 100 +        "2:\n"
 101 +        "   .irp        size, 8, 4, 2, 1\n"
 102 +        "   tst         %[w], #\\size\n"
 103 +        "   beq         3f\n"
 104 +        "   convert_macroblock_uyvy422_to_yuv420p \\size, 1, %[p], %[lum], %[cb], %[cr]\n"
 105 +        "3:\n"
 106 +        "  .endr\n":[w] "+&r" (w),[p] "+&r" (p),[lum] "+&r" (lum),
 107 +        [cb] "+&r" (cb),[cr] "+&r" (cr)
 108 +        ::"cc", "memory", "d0", "d1", "d2", "d3");
 109 +#else
 110 +    for (; w >= 2; w -= 2) {
 111 +      lum[0] = p[1];
 112 +      cb[0] = p[0];
 113 +      lum[1] = p[3];
 114 +      cr[0] = p[2];
 115 +      p += 4;
 116 +      lum += 2;
 117 +      cb++;
 118 +      cr++;
 119 +    }
 120 +    if (w) {
 121 +      lum[0] = p[1];
 122 +      cb[0] = p[0];
 123 +      cr[0] = p[2];
 124 +      cb++;
 125 +      cr++;
 126 +    }
 127 +#endif
 128 +    p1 += src->linesize[0];
 129 +    lum1 += dst->linesize[0];
 130 +    if (height > 1) {
 131 +      p = p1;
 132 +      lum = lum1;
 133 +      w = width;
 134 +#if 1
 135 +      asm volatile ("   subs        %[w], %[w], #16\n"
 136 +          "   blt         2f\n"
 137 +          "1:\n"
 138 +          "   convert_macroblock_uyvy422_to_yuv420p 16, 0, %[p], %[lum], %[cb], %[cr]\n"
 139 +          "   subs        %[w], %[w], #16\n"
 140 +          "   bge         1b\n"
 141 +          "2:\n"
 142 +          "   .irp        size, 8, 4, 2, 1\n"
 143 +          "   tst         %[w], #\\size\n"
 144 +          "   beq         3f\n"
 145 +          "   convert_macroblock_uyvy422_to_yuv420p \\size, 0, %[p], %[lum], %[cb], %[cr]\n"
 146 +          "3:\n"
 147 +          "  .endr\n":[w] "+&r" (w),[p] "+&r" (p),[lum] "+&r" (lum),
 148 +          [cb] "+&r" (cb),[cr] "+&r" (cr)
 149 +          ::"cc", "memory", "d0", "d1", "d2", "d3");
 150 +#else
 151 +      for (w = width; w >= 2; w -= 2) {
 152 +        lum[0] = p[1];
 153 +        lum[1] = p[3];
 154 +        p += 4;
 155 +        lum += 2;
 156 +      }
 157 +      if (w) {
 158 +        lum[0] = p[1];
 159 +      }
 160 +#endif
 161 +      p1 += src->linesize[0];
 162 +      lum1 += dst->linesize[0];
 163 +    }
 164 +    cb1 += dst->linesize[1];
 165 +    cr1 += dst->linesize[2];
 166 +  }
 167 +  asm volatile (".purgem convert_macroblock_uyvy422_to_yuv420p\n");
 168 +}
 169 +
 170  #endif
 171
 172  /* XXX: totally non optimized */
 173 @@ -3311,6 +3463,7 @@ static ConvertEntry convert_table[] = {
 174
 175  #ifdef __ARM_NEON__
 176    {PIX_FMT_UYVY422, PIX_FMT_RGB24, uyvy422_to_rgb24_neon},
 177 +  {PIX_FMT_UYVY422, PIX_FMT_YUV420P, uyvy422_to_yuv420p_neon},
 178  #endif
 179    {PIX_FMT_UYVY422, PIX_FMT_YUV420P, uyvy422_to_yuv420p},
 180    {PIX_FMT_UYVY422, PIX_FMT_YUV422P, uyvy422_to_yuv422p},