pnggccrd.c

Linux 下的图形编程环境。

#endif /* PNG_HAVE_ASSEMBLER_COMBINE_ROW */




//===========================================================================//
//                                                                           //
//                 P N G _ D O _ R E A D _ I N T E R L A C E                 //
//                                                                           //
//===========================================================================//

#if defined(PNG_READ_INTERLACING_SUPPORTED)
#if defined(PNG_HAVE_ASSEMBLER_READ_INTERLACE)

/* png_do_read_interlace() is called after any 16-bit to 8-bit conversion
 * has taken place.  [GRR: what other steps come before and/or after?]
 */

void /* PRIVATE */
png_do_read_interlace(png_structp png_ptr)
{
   png_row_infop row_info = &(png_ptr->row_info);
   png_bytep row = png_ptr->row_buf + 1;
   int pass = png_ptr->pass;
   png_uint_32 transformations = png_ptr->transformations;

   png_debug(1,"in png_do_read_interlace\n");

   if (_mmx_supported == 2) {
       png_mmx_support();
   }

   if (row != NULL && row_info != NULL)
   {
      png_uint_32 final_width;

      final_width = row_info->width * png_pass_inc[pass];

      switch (row_info->pixel_depth)
      {
         case 1:
         {
            png_bytep sp, dp;
            int sshift, dshift;
            int s_start, s_end, s_inc;
            png_byte v;
            png_uint_32 i;
            int j;

            sp = row + (png_size_t)((row_info->width - 1) >> 3);
            dp = row + (png_size_t)((final_width - 1) >> 3);
#if defined(PNG_READ_PACKSWAP_SUPPORTED)
            if (transformations & PNG_PACKSWAP)
            {
               sshift = (int)((row_info->width + 7) & 7);
               dshift = (int)((final_width + 7) & 7);
               s_start = 7;
               s_end = 0;
               s_inc = -1;
            }
            else
#endif
            {
               sshift = 7 - (int)((row_info->width + 7) & 7);
               dshift = 7 - (int)((final_width + 7) & 7);
               s_start = 0;
               s_end = 7;
               s_inc = 1;
            }

            for (i = row_info->width; i; i--)
            {
               v = (png_byte)((*sp >> sshift) & 0x1);
               for (j = 0; j < png_pass_inc[pass]; j++)
               {
                  *dp &= (png_byte)((0x7f7f >> (7 - dshift)) & 0xff);
                  *dp |= (png_byte)(v << dshift);
                  if (dshift == s_end)
                  {
                     dshift = s_start;
                     dp--;
                  }
                  else
                     dshift += s_inc;
               }
               if (sshift == s_end)
               {
                  sshift = s_start;
                  sp--;
               }
               else
                  sshift += s_inc;
            }
            break;
         }

         case 2:
         {
            png_bytep sp, dp;
            int sshift, dshift;
            int s_start, s_end, s_inc;
            png_uint_32 i;

            sp = row + (png_size_t)((row_info->width - 1) >> 2);
            dp = row + (png_size_t)((final_width - 1) >> 2);
#if defined(PNG_READ_PACKSWAP_SUPPORTED)
            if (transformations & PNG_PACKSWAP)
            {
               sshift = (png_size_t)(((row_info->width + 3) & 3) << 1);
               dshift = (png_size_t)(((final_width + 3) & 3) << 1);
               s_start = 6;
               s_end = 0;
               s_inc = -2;
            }
            else
#endif
            {
               sshift = (png_size_t)((3 - ((row_info->width + 3) & 3)) << 1);
               dshift = (png_size_t)((3 - ((final_width + 3) & 3)) << 1);
               s_start = 0;
               s_end = 6;
               s_inc = 2;
            }

            for (i = row_info->width; i; i--)
            {
               png_byte v;
               int j;

               v = (png_byte)((*sp >> sshift) & 0x3);
               for (j = 0; j < png_pass_inc[pass]; j++)
               {
                  *dp &= (png_byte)((0x3f3f >> (6 - dshift)) & 0xff);
                  *dp |= (png_byte)(v << dshift);
                  if (dshift == s_end)
                  {
                     dshift = s_start;
                     dp--;
                  }
                  else
                     dshift += s_inc;
               }
               if (sshift == s_end)
               {
                  sshift = s_start;
                  sp--;
               }
               else
                  sshift += s_inc;
            }
            break;
         }

         case 4:
         {
            png_bytep sp, dp;
            int sshift, dshift;
            int s_start, s_end, s_inc;
            png_uint_32 i;

            sp = row + (png_size_t)((row_info->width - 1) >> 1);
            dp = row + (png_size_t)((final_width - 1) >> 1);
#if defined(PNG_READ_PACKSWAP_SUPPORTED)
            if (transformations & PNG_PACKSWAP)
            {
               sshift = (png_size_t)(((row_info->width + 1) & 1) << 2);
               dshift = (png_size_t)(((final_width + 1) & 1) << 2);
               s_start = 4;
               s_end = 0;
               s_inc = -4;
            }
            else
#endif
            {
               sshift = (png_size_t)((1 - ((row_info->width + 1) & 1)) << 2);
               dshift = (png_size_t)((1 - ((final_width + 1) & 1)) << 2);
               s_start = 0;
               s_end = 4;
               s_inc = 4;
            }

            for (i = row_info->width; i; i--)
            {
               png_byte v;
               int j;

               v = (png_byte)((*sp >> sshift) & 0xf);
               for (j = 0; j < png_pass_inc[pass]; j++)
               {
                  *dp &= (png_byte)((0xf0f >> (4 - dshift)) & 0xff);
                  *dp |= (png_byte)(v << dshift);
                  if (dshift == s_end)
                  {
                     dshift = s_start;
                     dp--;
                  }
                  else
                     dshift += s_inc;
               }
               if (sshift == s_end)
               {
                  sshift = s_start;
                  sp--;
               }
               else
                  sshift += s_inc;
            }
            break;
         }

         //====================================================================

         default:  // 8-bit or larger (this is where the routine is modified)
         {
//          static unsigned long long _const4 = 0x0000000000FFFFFFLL;  no good
//          static unsigned long long const4 = 0x0000000000FFFFFFLL;   no good
//          unsigned long long _const4 = 0x0000000000FFFFFFLL;         no good
//          unsigned long long const4 = 0x0000000000FFFFFFLL;          no good
            png_bytep sptr, dp;
            png_uint_32 i;
            png_size_t pixel_bytes;
            int width = row_info->width;

            pixel_bytes = (row_info->pixel_depth >> 3);

            // point sptr at the last pixel in the pre-expanded row:
            sptr = row + (width - 1) * pixel_bytes;

            // point dp at the last pixel position in the expanded row:
            dp = row + (final_width - 1) * pixel_bytes;

            // New code by Nirav Chhatrapati - Intel Corporation

            if ( _mmx_supported )
            {
               //--------------------------------------------------------------
               if (pixel_bytes == 3)
               {
                  if (((pass == 0) || (pass == 1)) && width)
                  {
                     int dummy_value_c;   // fix 'forbidden register spilled'
                     int dummy_value_S;
                     int dummy_value_D;

                     __asm__ __volatile__ (
                        "subl $21, %%edi         \n\t"
                                     // (png_pass_inc[pass] - 1)*pixel_bytes

                     ".loop3_pass0:              \n\t"
                        "movd (%%esi), %%mm0     \n\t" // x x x x x 2 1 0
                        "pand _const4, %%mm0     \n\t" // z z z z z 2 1 0
                        "movq %%mm0, %%mm1       \n\t" // z z z z z 2 1 0
                        "psllq $16, %%mm0        \n\t" // z z z 2 1 0 z z
                        "movq %%mm0, %%mm2       \n\t" // z z z 2 1 0 z z
                        "psllq $24, %%mm0        \n\t" // 2 1 0 z z z z z
                        "psrlq $8, %%mm1         \n\t" // z z z z z z 2 1
                        "por %%mm2, %%mm0        \n\t" // 2 1 0 2 1 0 z z
                        "por %%mm1, %%mm0        \n\t" // 2 1 0 2 1 0 2 1
                        "movq %%mm0, %%mm3       \n\t" // 2 1 0 2 1 0 2 1
                        "psllq $16, %%mm0        \n\t" // 0 2 1 0 2 1 z z
                        "movq %%mm3, %%mm4       \n\t" // 2 1 0 2 1 0 2 1
                        "punpckhdq %%mm0, %%mm3  \n\t" // 0 2 1 0 2 1 0 2
                        "movq %%mm4, 16(%%edi)   \n\t"
                        "psrlq $32, %%mm0        \n\t" // z z z z 0 2 1 0
                        "movq %%mm3, 8(%%edi)    \n\t"
                        "punpckldq %%mm4, %%mm0  \n\t" // 1 0 2 1 0 2 1 0
                        "subl $3, %%esi          \n\t"
                        "movq %%mm0, (%%edi)     \n\t"
                        "subl $24, %%edi         \n\t"
                        "decl %%ecx              \n\t"
                        "jnz .loop3_pass0        \n\t"
                        "EMMS                    \n\t" // DONE

                        : "=c" (dummy_value_c),        // output regs (dummy)
                          "=S" (dummy_value_S),
                          "=D" (dummy_value_D)

                        : "1" (sptr),      // esi      // input regs
                          "2" (dp),        // edi
                          "0" (width)      // ecx
// doesn't work           "i" (0x0000000000FFFFFFLL)   // %1 (a.k.a. _const4)

#if 0  /* %mm0, ..., %mm4 not supported by gcc 2.7.2.3 or egcs 1.1 */
                        : "%mm0", "%mm1", "%mm2"       // clobber list
                        , "%mm3", "%mm4"
#endif
                     );
                  }
                  else if (((pass == 2) || (pass == 3)) && width)
                  {
                     int dummy_value_c;   // fix 'forbidden register spilled'
                     int dummy_value_S;
                     int dummy_value_D;

                     __asm__ __volatile__ (
                        "subl $9, %%edi          \n\t"
                                     // (png_pass_inc[pass] - 1)*pixel_bytes

                     ".loop3_pass2:              \n\t"
                        "movd (%%esi), %%mm0     \n\t" // x x x x x 2 1 0
                        "pand _const4, %%mm0     \n\t" // z z z z z 2 1 0
                        "movq %%mm0, %%mm1       \n\t" // z z z z z 2 1 0
                        "psllq $16, %%mm0        \n\t" // z z z 2 1 0 z z
                        "movq %%mm0, %%mm2       \n\t" // z z z 2 1 0 z z
                        "psllq $24, %%mm0        \n\t" // 2 1 0 z z z z z
                        "psrlq $8, %%mm1         \n\t" // z z z z z z 2 1
                        "por %%mm2, %%mm0        \n\t" // 2 1 0 2 1 0 z z
                        "por %%mm1, %%mm0        \n\t" // 2 1 0 2 1 0 2 1
                        "movq %%mm0, 4(%%edi)    \n\t"
                        "psrlq $16, %%mm0        \n\t" // z z 2 1 0 2 1 0
                        "subl $3, %%esi          \n\t"
                        "movd %%mm0, (%%edi)     \n\t"
                        "subl $12, %%edi         \n\t"
                        "decl %%ecx              \n\t"
                        "jnz .loop3_pass2        \n\t"
                        "EMMS                    \n\t" // DONE

                        : "=c" (dummy_value_c),        // output regs (dummy)
                          "=S" (dummy_value_S),
                          "=D" (dummy_val