src/gallium/tests/unit/translate_test.c

   1 /**************************************************************************
   2  *
   3  * Copyright © 2010 Luca Barbieri
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  22  * DEALINGS IN THE SOFTWARE.
  23  *
  24  **************************************************************************/
  25
  26 #include <stdio.h>
  27 #include "translate/translate.h"
  28 #include "util/u_memory.h"
  29 #include "util/u_format.h"
  30 #include "util/u_cpu_detect.h"
  31 #include "rtasm/rtasm_cpu.h"
  32
  33 /* don't use this for serious use */
  34 static double rand_double()
  35 {
  36    const double rm = (double)RAND_MAX + 1;
  37    double div = 1;
  38    double v = 0;
  39    unsigned i;
  40    for(i = 0; i < 4; ++i)
  41    {
  42       div *= rm;
  43       v += (double)rand() / div;
  44    }
  45    return v;
  46 }
  47
  48 int main(int argc, char** argv)
  49 {
  50    struct translate *(*create_fn)(const struct translate_key *key) = 0;
  51
  52    struct translate_key key;
  53    unsigned output_format;
  54    unsigned input_format;
  55    unsigned buffer_size = 4096;
  56    unsigned char* buffer[5];
  57    unsigned char* byte_buffer;
  58    float* float_buffer;
  59    double* double_buffer;
  60    unsigned * elts;
  61    unsigned count = 4;
  62    unsigned i, j, k;
  63    unsigned passed = 0;
  64    unsigned total = 0;
  65    const float error = 0.03125;
  66
  67    create_fn = 0;
  68
  69    util_cpu_detect();
  70
  71    if(argc <= 1)
  72    {}
  73    else if (!strcmp(argv[1], "generic"))
  74       create_fn = translate_generic_create;
  75    else if (!strcmp(argv[1], "x86"))
  76       create_fn = translate_sse2_create;
  77    else if (!strcmp(argv[1], "nosse"))
  78    {
  79       util_cpu_caps.has_sse = 0;
  80       util_cpu_caps.has_sse2 = 0;
  81       util_cpu_caps.has_sse3 = 0;
  82       util_cpu_caps.has_sse4_1 = 0;
  83       create_fn = translate_sse2_create;
  84    }
  85    else if (!strcmp(argv[1], "sse"))
  86    {
  87       if(!util_cpu_caps.has_sse || !rtasm_cpu_has_sse())
  88       {
  89          printf("Error: CPU doesn't support SSE (test with qemu)\n");
  90          return 2;
  91       }
  92       util_cpu_caps.has_sse2 = 0;
  93       util_cpu_caps.has_sse3 = 0;
  94       util_cpu_caps.has_sse4_1 = 0;
  95       create_fn = translate_sse2_create;
  96    }
  97    else if (!strcmp(argv[1], "sse2"))
  98    {
  99       if(!util_cpu_caps.has_sse2 || !rtasm_cpu_has_sse())
 100       {
 101          printf("Error: CPU doesn't support SSE2 (test with qemu)\n");
 102          return 2;
 103       }
 104       util_cpu_caps.has_sse3 = 0;
 105       util_cpu_caps.has_sse4_1 = 0;
 106       create_fn = translate_sse2_create;
 107    }
 108    else if (!strcmp(argv[1], "sse3"))
 109    {
 110       if(!util_cpu_caps.has_sse3 || !rtasm_cpu_has_sse())
 111       {
 112          printf("Error: CPU doesn't support SSE3 (test with qemu)\n");
 113          return 2;
 114       }
 115       util_cpu_caps.has_sse4_1 = 0;
 116       create_fn = translate_sse2_create;
 117    }
 118    else if (!strcmp(argv[1], "sse4.1"))
 119    {
 120       if(!util_cpu_caps.has_sse4_1 || !rtasm_cpu_has_sse())
 121       {
 122          printf("Error: CPU doesn't support SSE4.1 (test with qemu)\n");
 123          return 2;
 124       }
 125       create_fn = translate_sse2_create;
 126    }
 127
 128    if (!create_fn)
 129    {
 130       printf("Usage: ./translate_test [generic|x86|nosse|sse|sse2|sse3|sse4.1]\n");
 131       return 2;
 132    }
 133
 134    for (i = 1; i < Elements(buffer); ++i)
 135       buffer[i] = align_malloc(buffer_size, 4096);
 136
 137    byte_buffer = align_malloc(buffer_size, 4096);
 138    float_buffer = align_malloc(buffer_size, 4096);
 139    double_buffer = align_malloc(buffer_size, 4096);
 140
 141    elts = align_malloc(count * sizeof *elts, 4096);
 142
 143    key.nr_elements = 1;
 144    key.element[0].input_buffer = 0;
 145    key.element[0].input_offset = 0;
 146    key.element[0].output_offset = 0;
 147    key.element[0].type = TRANSLATE_ELEMENT_NORMAL;
 148    key.element[0].instance_divisor = 0;
 149
 150    srand(4359025);
 151
 152    /* avoid negative values that work badly when converted to unsigned format*/
 153    for (i = 0; i < buffer_size; ++i)
 154       byte_buffer[i] = rand() & 0x7f7f7f7f;
 155
 156    for (i = 0; i < buffer_size / sizeof(float); ++i)
 157       float_buffer[i] = (float)rand_double();
 158
 159    for (i = 0; i < buffer_size / sizeof(double); ++i)
 160       double_buffer[i] = rand_double();
 161
 162    for (i = 0; i < count; ++i)
 163       elts[i] = i;
 164
 165    for (output_format = 1; output_format < PIPE_FORMAT_COUNT; ++output_format)
 166    {
 167       const struct util_format_description* output_format_desc = util_format_description(output_format);
 168       unsigned output_format_size;
 169       unsigned output_normalized = 0;
 170
 171       if (!output_format_desc
 172             || !output_format_desc->fetch_rgba_float
 173             || !output_format_desc->pack_rgba_float
 174             || output_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
 175             || output_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
 176             || !translate_is_output_format_supported(output_format))
 177          continue;
 178
 179       for(i = 0; i < output_format_desc->nr_channels; ++i)
 180       {
 181          if(output_format_desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT)
 182             output_normalized |= (1 << output_format_desc->channel[i].normalized);
 183       }
 184
 185       output_format_size = util_format_get_stride(output_format, 1);
 186
 187       for (input_format = 1; input_format < PIPE_FORMAT_COUNT; ++input_format)
 188       {
 189          const struct util_format_description* input_format_desc = util_format_description(input_format);
 190          unsigned input_format_size;
 191          struct translate* translate[2];
 192          unsigned fail = 0;
 193          unsigned used_generic = 0;
 194          unsigned input_normalized = 0;
 195          boolean input_is_float = FALSE;
 196
 197          if (!input_format_desc
 198                || !input_format_desc->fetch_rgba_float
 199                || !input_format_desc->pack_rgba_float
 200                || input_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
 201                || input_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
 202                || !translate_is_output_format_supported(input_format))
 203             continue;
 204
 205          input_format_size = util_format_get_stride(input_format, 1);
 206
 207          for(i = 0; i < input_format_desc->nr_channels; ++i)
 208          {
 209             if(input_format_desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT)
 210             {
 211                input_is_float = 1;
 212                input_normalized |= 1 << 1;
 213             }
 214             else
 215                input_normalized |= (1 << input_format_desc->channel[i].normalized);
 216          }
 217
 218          if(((input_normalized | output_normalized) == 3)
 219                || ((input_normalized & 1) && (output_normalized & 1)
 220                      && input_format_size * output_format_desc->nr_channels > output_format_size * input_format_desc->nr_channels))
 221             continue;
 222
 223          key.element[0].input_format = input_format;
 224          key.element[0].output_format = output_format;
 225          key.output_stride = output_format_size;
 226          translate[0] = create_fn(&key);
 227          if (!translate[0])
 228             continue;
 229
 230          key.element[0].input_format = output_format;
 231          key.element[0].output_format = input_format;
 232          key.output_stride = input_format_size;
 233          translate[1] = create_fn(&key);
 234          if(!translate[1])
 235          {
 236             used_generic = 1;
 237             translate[1] = translate_generic_create(&key);
 238             if(!translate[1])
 239                continue;
 240          }
 241
 242          for(i = 1; i < 5; ++i)
 243             memset(buffer[i], 0xcd - (0x22 * i), 4096);
 244
 245          if(input_is_float && input_format_desc->channel[0].size == 32)
 246             buffer[0] = (unsigned char*)float_buffer;
 247          else if(input_is_float && input_format_desc->channel[0].size == 64)
 248             buffer[0] = (unsigned char*)double_buffer;
 249          else if(input_is_float)
 250             abort();
 251          else
 252             buffer[0] = byte_buffer;
 253
 254          translate[0]->set_buffer(translate[0], 0, buffer[0], input_format_size, count - 1);
 255          translate[0]->run_elts(translate[0], elts, count, 0, buffer[1]);
 256          translate[1]->set_buffer(translate[1], 0, buffer[1], output_format_size, count - 1);
 257          translate[1]->run_elts(translate[1], elts, count, 0, buffer[2]);
 258          translate[0]->set_buffer(translate[0], 0, buffer[2], input_format_size, count - 1);
 259          translate[0]->run_elts(translate[0], elts, count, 0, buffer[3]);
 260          translate[1]->set_buffer(translate[1], 0, buffer[3], output_format_size, count - 1);
 261          translate[1]->run_elts(translate[1], elts, count, 0, buffer[4]);
 262
 263          for (i = 0; i < count; ++i)
 264          {
 265             float a[4];
 266             float b[4];
 267             input_format_desc->fetch_rgba_float(a, buffer[2] + i * input_format_size, 0, 0);
 268             input_format_desc->fetch_rgba_float(b, buffer[4] + i * input_format_size, 0, 0);
 269
 270             for (j = 0; j < count; ++j)
 271             {
 272                float d = a[j] - b[j];
 273                if (d > error || d < -error)
 274                {
 275                   fail = 1;
 276                   break;
 277                }
 278             }
 279          }
 280
 281          printf("%s%s: %s -> %s -> %s -> %s -> %s\n",
 282                fail ? "FAIL" : "PASS",
 283                used_generic ? "[GENERIC]" : "",
 284                input_format_desc->name, output_format_desc->name, input_format_desc->name, output_format_desc->name, input_format_desc->name);
 285
 286          if (1)
 287          {
 288             for (i = 0; i < Elements(buffer); ++i)
 289             {
 290                unsigned format_size = (i & 1) ? output_format_size : input_format_size;
 291                printf("%c ", (i == 2 || i == 4) ? '*' : ' ');
 292                for (j = 0; j < count; ++j)
 293                {
 294                   for (k = 0; k < format_size; ++k)
 295                   {
 296                      printf("%02x", buffer[i][j * format_size + k]);
 297                   }
 298                   printf(" ");
 299                }
 300                printf("\n");
 301             }
 302          }
 303
 304          if (!fail)
 305             ++passed;
 306          ++total;
 307
 308          if(translate[1])
 309             translate[1]->release(translate[1]);
 310          translate[0]->release(translate[0]);
 311       }
 312    }
 313
 314    printf("%u/%u tests passed for translate_%s\n", passed, total, argv[1]);
 315    return passed != total;
 316 }