src/gallium/drivers/llvmpipe/lp_tile_soa.py

   1 #!/usr/bin/env python
   2
   3 '''
   4 /**************************************************************************
   5  *
   6  * Copyright 2009 VMware, Inc.
   7  * All Rights Reserved.
   8  *
   9  * Permission is hereby granted, free of charge, to any person obtaining a
  10  * copy of this software and associated documentation files (the
  11  * "Software"), to deal in the Software without restriction, including
  12  * without limitation the rights to use, copy, modify, merge, publish,
  13  * distribute, sub license, and/or sell copies of the Software, and to
  14  * permit persons to whom the Software is furnished to do so, subject to
  15  * the following conditions:
  16  *
  17  * The above copyright notice and this permission notice (including the
  18  * next paragraph) shall be included in all copies or substantial portions
  19  * of the Software.
  20  *
  21  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  22  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  23  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  24  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  25  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  26  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  27  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  28  *
  29  **************************************************************************/
  30
  31 /**
  32  * @file
  33  * Pixel format accessor functions.
  34  *
  35  * @author Jose Fonseca <jfonseca@vmware.com>
  36  */
  37 '''
  38
  39
  40 import sys
  41 import os.path
  42
  43 sys.path.insert(0, os.path.join(os.path.dirname(sys.argv[0]), '../../auxiliary/util'))
  44
  45 from u_format_access import *
  46
  47
  48 def generate_format_read(format, dst_channel, dst_native_type, dst_suffix):
  49     '''Generate the function to read pixels from a particular format'''
  50
  51     name = format.short_name()
  52
  53     src_native_type = native_type(format)
  54
  55     print 'static void'
  56     print 'lp_tile_%s_read_%s(%s *dst, const uint8_t *src, unsigned src_stride, unsigned x0, unsigned y0, unsigned w, unsigned h)' % (name, dst_suffix, dst_native_type)
  57     print '{'
  58     print '   unsigned x, y;'
  59     print '   const uint8_t *src_row = src + y0*src_stride;'
  60     print '   for (y = 0; y < h; ++y) {'
  61     print '      const %s *src_pixel = (const %s *)(src_row + x0*%u);' % (src_native_type, src_native_type, format.stride())
  62     print '      for (x = 0; x < w; ++x) {'
  63
  64     names = ['']*4
  65     if format.colorspace == 'rgb':
  66         for i in range(4):
  67             swizzle = format.swizzles[i]
  68             if swizzle < 4:
  69                 names[swizzle] += 'rgba'[i]
  70     elif format.colorspace == 'zs':
  71         swizzle = format.swizzles[0]
  72         if swizzle < 4:
  73             names[swizzle] = 'z'
  74         else:
  75             assert False
  76     else:
  77         assert False
  78
  79     if format.layout == PLAIN:
  80         if not format.is_array():
  81             print '         %s pixel = *src_pixel++;' % src_native_type
  82             shift = 0;
  83             for i in range(4):
  84                 src_channel = format.channels[i]
  85                 width = src_channel.size
  86                 if names[i]:
  87                     value = 'pixel'
  88                     mask = (1 << width) - 1
  89                     if shift:
  90                         value = '(%s >> %u)' % (value, shift)
  91                     if shift + width < format.block_size():
  92                         value = '(%s & 0x%x)' % (value, mask)
  93                     value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
  94                     print '         %s %s = %s;' % (dst_native_type, names[i], value)
  95                 shift += width
  96         else:
  97             for i in range(4):
  98                 src_channel = format.channels[i]
  99                 if names[i]:
 100                     value = '(*src_pixel++)'
 101                     value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
 102                     print '         %s %s = %s;' % (dst_native_type, names[i], value)
 103     else:
 104         assert False
 105
 106     for i in range(4):
 107         if format.colorspace == 'rgb':
 108             swizzle = format.swizzles[i]
 109             if swizzle < 4:
 110                 value = names[swizzle]
 111             elif swizzle == SWIZZLE_0:
 112                 value = '0'
 113             elif swizzle == SWIZZLE_1:
 114                 value = get_one(dst_channel)
 115             else:
 116                 assert False
 117         elif format.colorspace == 'zs':
 118             if i < 3:
 119                 value = 'z'
 120             else:
 121                 value = get_one(dst_channel)
 122         else:
 123             assert False
 124         print '         TILE_PIXEL(dst, x, y, %u) = %s; /* %s */' % (i, value, 'rgba'[i])
 125
 126     print '      }'
 127     print '      src_row += src_stride;'
 128     print '   }'
 129     print '}'
 130     print
 131
 132
 133 def pack_rgba(format, src_channel, r, g, b, a):
 134     """Return an expression for packing r, g, b, a into a pixel of the
 135     given format.  Ex: '(b << 24) | (g << 16) | (r << 8) | (a << 0)'
 136     """
 137     assert format.colorspace == 'rgb'
 138     inv_swizzle = format.inv_swizzles()
 139     shift = 0
 140     expr = None
 141     for i in range(4):
 142         # choose r, g, b, or a depending on the inverse swizzle term
 143         if inv_swizzle[i] == 0:
 144             value = r
 145         elif inv_swizzle[i] == 1:
 146             value = g
 147         elif inv_swizzle[i] == 2:
 148             value = b
 149         elif inv_swizzle[i] == 3:
 150             value = a
 151         else:
 152             value = None
 153
 154         if value:
 155             dst_channel = format.channels[i]
 156             dst_native_type = native_type(format)
 157             value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
 158             term = "((%s) << %d)" % (value, shift)
 159             if expr:
 160                 expr = expr + " | " + term
 161             else:
 162                 expr = term
 163
 164         width = format.channels[i].size
 165         shift = shift + width
 166     return expr
 167
 168
 169 def emit_unrolled_write_code(format, src_channel):
 170     '''Emit code for writing a block based on unrolled loops.
 171     This is considerably faster than the TILE_PIXEL-based code below.
 172     '''
 173     dst_native_type = 'uint%u_t' % format.block_size()
 174     print '   const unsigned dstpix_stride = dst_stride / %d;' % format.stride()
 175     print '   %s *dstpix = (%s *) dst;' % (dst_native_type, dst_native_type)
 176     print '   unsigned int qx, qy, i;'
 177     print
 178     print '   for (qy = 0; qy < h; qy += TILE_VECTOR_HEIGHT) {'
 179     print '      const unsigned py = y0 + qy;'
 180     print '      for (qx = 0; qx < w; qx += TILE_VECTOR_WIDTH) {'
 181     print '         const unsigned px = x0 + qx;'
 182     print '         const uint8_t *r = src + 0 * TILE_C_STRIDE;'
 183     print '         const uint8_t *g = src + 1 * TILE_C_STRIDE;'
 184     print '         const uint8_t *b = src + 2 * TILE_C_STRIDE;'
 185     print '         const uint8_t *a = src + 3 * TILE_C_STRIDE;'
 186     print '         (void) r; (void) g; (void) b; (void) a; /* silence warnings */'
 187     print '         for (i = 0; i < TILE_C_STRIDE; i += 2) {'
 188     print '            const uint32_t pixel0 = %s;' % pack_rgba(format, src_channel, "r[i+0]", "g[i+0]", "b[i+0]", "a[i+0]")
 189     print '            const uint32_t pixel1 = %s;' % pack_rgba(format, src_channel, "r[i+1]", "g[i+1]", "b[i+1]", "a[i+1]")
 190     print '            const unsigned offset = (py + tile_y_offset[i]) * dstpix_stride + (px + tile_x_offset[i]);'
 191     print '            dstpix[offset + 0] = pixel0;'
 192     print '            dstpix[offset + 1] = pixel1;'
 193     print '         }'
 194     print '         src += TILE_X_STRIDE;'
 195     print '      }'
 196     print '   }'
 197
 198
 199 def emit_tile_pixel_write_code(format, src_channel):
 200     '''Emit code for writing a block based on the TILE_PIXEL macro.'''
 201     dst_native_type = native_type(format)
 202
 203     inv_swizzle = format.inv_swizzles()
 204
 205     print '   unsigned x, y;'
 206     print '   uint8_t *dst_row = dst + y0*dst_stride;'
 207     print '   for (y = 0; y < h; ++y) {'
 208     print '      %s *dst_pixel = (%s *)(dst_row + x0*%u);' % (dst_native_type, dst_native_type, format.stride())
 209     print '      for (x = 0; x < w; ++x) {'
 210
 211     if format.layout == PLAIN:
 212         if not format.is_array():
 213             print '         %s pixel = 0;' % dst_native_type
 214             shift = 0;
 215             for i in range(4):
 216                 dst_channel = format.channels[i]
 217                 width = dst_channel.size
 218                 if inv_swizzle[i] is not None:
 219                     value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
 220                     value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
 221                     if shift:
 222                         value = '(%s << %u)' % (value, shift)
 223                     print '         pixel |= %s;' % value
 224                 shift += width
 225             print '         *dst_pixel++ = pixel;'
 226         else:
 227             for i in range(4):
 228                 dst_channel = format.channels[i]
 229                 if inv_swizzle[i] is not None:
 230                     value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
 231                     value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
 232                     print '         *dst_pixel++ = %s;' % value
 233     else:
 234         assert False
 235
 236     print '      }'
 237     print '      dst_row += dst_stride;'
 238     print '   }'
 239
 240
 241 def generate_format_write(format, src_channel, src_native_type, src_suffix):
 242     '''Generate the function to write pixels to a particular format'''
 243
 244     name = format.short_name()
 245
 246     print 'static void'
 247     print 'lp_tile_%s_write_%s(const %s *src, uint8_t *dst, unsigned dst_stride, unsigned x0, unsigned y0, unsigned w, unsigned h)' % (name, src_suffix, src_native_type)
 248     print '{'
 249     if format.layout == PLAIN \
 250         and format.colorspace == 'rgb' \
 251         and format.block_size() <= 32 \
 252         and format.is_pot() \
 253         and not format.is_mixed() \
 254         and format.channels[0].type == UNSIGNED:
 255         emit_unrolled_write_code(format, src_channel)
 256     else:
 257         emit_tile_pixel_write_code(format, src_channel)
 258     print '}'
 259     print
 260
 261
 262 def generate_read(formats, dst_channel, dst_native_type, dst_suffix):
 263     '''Generate the dispatch function to read pixels from any format'''
 264
 265     for format in formats:
 266         if is_format_supported(format):
 267             generate_format_read(format, dst_channel, dst_native_type, dst_suffix)
 268
 269     print 'void'
 270     print 'lp_tile_read_%s(enum pipe_format format, %s *dst, const void *src, unsigned src_stride, unsigned x, unsigned y, unsigned w, unsigned h)' % (dst_suffix, dst_native_type)
 271     print '{'
 272     print '   void (*func)(%s *dst, const uint8_t *src, unsigned src_stride, unsigned x0, unsigned y0, unsigned w, unsigned h);' % dst_native_type
 273     print '   switch(format) {'
 274     for format in formats:
 275         if is_format_supported(format):
 276             print '   case %s:' % format.name
 277             print '      func = &lp_tile_%s_read_%s;' % (format.short_name(), dst_suffix)
 278             print '      break;'
 279     print '   default:'
 280     print '      debug_printf("unsupported format\\n");'
 281     print '      return;'
 282     print '   }'
 283     print '   func(dst, (const uint8_t *)src, src_stride, x, y, w, h);'
 284     print '}'
 285     print
 286
 287
 288 def generate_write(formats, src_channel, src_native_type, src_suffix):
 289     '''Generate the dispatch function to write pixels to any format'''
 290
 291     for format in formats:
 292         if is_format_supported(format):
 293             generate_format_write(format, src_channel, src_native_type, src_suffix)
 294
 295     print 'void'
 296     print 'lp_tile_write_%s(enum pipe_format format, const %s *src, void *dst, unsigned dst_stride, unsigned x, unsigned y, unsigned w, unsigned h)' % (src_suffix, src_native_type)
 297
 298     print '{'
 299     print '   void (*func)(const %s *src, uint8_t *dst, unsigned dst_stride, unsigned x0, unsigned y0, unsigned w, unsigned h);' % src_native_type
 300     print '   switch(format) {'
 301     for format in formats:
 302         if is_format_supported(format):
 303             print '   case %s:' % format.name
 304             print '      func = &lp_tile_%s_write_%s;' % (format.short_name(), src_suffix)
 305             print '      break;'
 306     print '   default:'
 307     print '      debug_printf("unsupported format\\n");'
 308     print '      return;'
 309     print '   }'
 310     print '   func(src, (uint8_t *)dst, dst_stride, x, y, w, h);'
 311     print '}'
 312     print
 313
 314
 315 def main():
 316     formats = []
 317     for arg in sys.argv[1:]:
 318         formats.extend(parse(arg))
 319
 320     print '/* This file is autogenerated by lp_tile_soa.py from u_format.csv. Do not edit directly. */'
 321     print
 322     # This will print the copyright message on the top of this file
 323     print __doc__.strip()
 324     print
 325     print '#include "pipe/p_compiler.h"'
 326     print '#include "util/u_format.h"'
 327     print '#include "util/u_math.h"'
 328     print '#include "lp_tile_soa.h"'
 329     print
 330     print 'const unsigned char'
 331     print 'tile_offset[TILE_VECTOR_HEIGHT][TILE_VECTOR_WIDTH] = {'
 332     print '   {  0,  1,  4,  5},'
 333     print '   {  2,  3,  6,  7},'
 334     print '   {  8,  9, 12, 13},'
 335     print '   { 10, 11, 14, 15}'
 336     print '};'
 337     print
 338     print '/* Note: these lookup tables could be replaced with some'
 339     print ' * bit-twiddling code, but this is a little faster.'
 340     print ' */'
 341     print 'static unsigned tile_x_offset[TILE_VECTOR_WIDTH * TILE_VECTOR_HEIGHT] = {'
 342     print '   0, 1, 0, 1, 2, 3, 2, 3,'
 343     print '   0, 1, 0, 1, 2, 3, 2, 3'
 344     print '};'
 345     print
 346     print 'static unsigned tile_y_offset[TILE_VECTOR_WIDTH * TILE_VECTOR_HEIGHT] = {'
 347     print '   0, 0, 1, 1, 0, 0, 1, 1,'
 348     print '   2, 2, 3, 3, 2, 2, 3, 3'
 349     print '};'
 350     print
 351
 352     generate_clamp()
 353
 354     channel = Channel(UNSIGNED, True, 8)
 355     native_type = 'uint8_t'
 356     suffix = '4ub'
 357
 358     generate_read(formats, channel, native_type, suffix)
 359     generate_write(formats, channel, native_type, suffix)
 360
 361
 362 if __name__ == '__main__':
 363     main()