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/format/u_format.h"
  30 #include "util/u_half.h"
  31 #include "util/u_cpu_detect.h"
  32 #include "rtasm/rtasm_cpu.h"
  33
  34 /* don't use this for serious use */
  35 static double rand_double()
  36 {
  37    const double rm = (double)RAND_MAX + 1;
  38    double div = 1;
  39    double v = 0;
  40    unsigned i;
  41    for(i = 0; i < 4; ++i)
  42    {
  43       div *= rm;
  44       v += (double)rand() / div;
  45    }
  46    return v;
  47 }
  48
  49 int main(int argc, char** argv)
  50 {
  51    struct translate *(*create_fn)(const struct translate_key *key) = 0;
  52
  53    struct translate_key key;
  54    unsigned output_format;
  55    unsigned input_format;
  56    unsigned buffer_size = 4096;
  57    unsigned char* buffer[5];
  58    unsigned char* byte_buffer;
  59    float* float_buffer;
  60    double* double_buffer;
  61    uint16_t *half_buffer;
  62    unsigned * elts;
  63    unsigned count = 4;
  64    unsigned i, j, k;
  65    unsigned passed = 0;
  66    unsigned total = 0;
  67    const float error = 0.03125;
  68
  69    create_fn = 0;
  70
  71    util_cpu_detect();
  72
  73    if (argc <= 1 ||
  74        !strcmp(argv[1], "default") )
  75       create_fn = translate_create;
  76    else if (!strcmp(argv[1], "generic"))
  77       create_fn = translate_generic_create;
  78    else if (!strcmp(argv[1], "x86"))
  79       create_fn = translate_sse2_create;
  80    else if (!strcmp(argv[1], "nosse"))
  81    {
  82       util_cpu_caps.has_sse = 0;
  83       util_cpu_caps.has_sse2 = 0;
  84       util_cpu_caps.has_sse3 = 0;
  85       util_cpu_caps.has_sse4_1 = 0;
  86       create_fn = translate_sse2_create;
  87    }
  88    else if (!strcmp(argv[1], "sse"))
  89    {
  90       if(!util_cpu_caps.has_sse || !rtasm_cpu_has_sse())
  91       {
  92          printf("Error: CPU doesn't support SSE (test with qemu)\n");
  93          return 2;
  94       }
  95       util_cpu_caps.has_sse2 = 0;
  96       util_cpu_caps.has_sse3 = 0;
  97       util_cpu_caps.has_sse4_1 = 0;
  98       create_fn = translate_sse2_create;
  99    }
 100    else if (!strcmp(argv[1], "sse2"))
 101    {
 102       if(!util_cpu_caps.has_sse2 || !rtasm_cpu_has_sse())
 103       {
 104          printf("Error: CPU doesn't support SSE2 (test with qemu)\n");
 105          return 2;
 106       }
 107       util_cpu_caps.has_sse3 = 0;
 108       util_cpu_caps.has_sse4_1 = 0;
 109       create_fn = translate_sse2_create;
 110    }
 111    else if (!strcmp(argv[1], "sse3"))
 112    {
 113       if(!util_cpu_caps.has_sse3 || !rtasm_cpu_has_sse())
 114       {
 115          printf("Error: CPU doesn't support SSE3 (test with qemu)\n");
 116          return 2;
 117       }
 118       util_cpu_caps.has_sse4_1 = 0;
 119       create_fn = translate_sse2_create;
 120    }
 121    else if (!strcmp(argv[1], "sse4.1"))
 122    {
 123       if(!util_cpu_caps.has_sse4_1 || !rtasm_cpu_has_sse())
 124       {
 125          printf("Error: CPU doesn't support SSE4.1 (test with qemu)\n");
 126          return 2;
 127       }
 128       create_fn = translate_sse2_create;
 129    }
 130
 131    if (!create_fn)
 132    {
 133       printf("Usage: ./translate_test [default|generic|x86|nosse|sse|sse2|sse3|sse4.1]\n");
 134       return 2;
 135    }
 136
 137    for (i = 1; i < ARRAY_SIZE(buffer); ++i)
 138       buffer[i] = align_malloc(buffer_size, 4096);
 139
 140    byte_buffer = align_malloc(buffer_size, 4096);
 141    float_buffer = align_malloc(buffer_size, 4096);
 142    double_buffer = align_malloc(buffer_size, 4096);
 143    half_buffer = align_malloc(buffer_size, 4096);
 144
 145    elts = align_malloc(count * sizeof *elts, 4096);
 146
 147    key.nr_elements = 1;
 148    key.element[0].input_buffer = 0;
 149    key.element[0].input_offset = 0;
 150    key.element[0].output_offset = 0;
 151    key.element[0].type = TRANSLATE_ELEMENT_NORMAL;
 152    key.element[0].instance_divisor = 0;
 153
 154    srand(4359025);
 155
 156    /* avoid negative values that work badly when converted to unsigned format*/
 157    for (i = 0; i < buffer_size; ++i)
 158       byte_buffer[i] = rand() & 0x7f7f7f7f;
 159
 160    for (i = 0; i < buffer_size / sizeof(float); ++i)
 161       float_buffer[i] = (float)rand_double();
 162
 163    for (i = 0; i < buffer_size / sizeof(double); ++i)
 164       double_buffer[i] = rand_double();
 165
 166    for (i = 0; i < buffer_size / sizeof(double); ++i)
 167       half_buffer[i] = util_float_to_half((float) rand_double());
 168
 169    for (i = 0; i < count; ++i)
 170       elts[i] = i;
 171
 172    for (output_format = 1; output_format < PIPE_FORMAT_COUNT; ++output_format)
 173    {
 174       const struct util_format_description* output_format_desc = util_format_description(output_format);
 175       const struct util_format_pack_description* output_format_pack = util_format_pack_description(output_format);
 176       const struct util_format_unpack_description* output_format_unpack = util_format_unpack_description(output_format);
 177       unsigned output_format_size;
 178       unsigned output_normalized = 0;
 179
 180       if (!output_format_desc
 181             || !output_format_unpack->fetch_rgba
 182             || !output_format_pack->pack_rgba_float
 183             || output_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
 184             || output_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
 185             || !translate_is_output_format_supported(output_format))
 186          continue;
 187
 188       for(i = 0; i < output_format_desc->nr_channels; ++i)
 189       {
 190          if(output_format_desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT)
 191             output_normalized |= (1 << output_format_desc->channel[i].normalized);
 192       }
 193
 194       output_format_size = util_format_get_stride(output_format, 1);
 195
 196       for (input_format = 1; input_format < PIPE_FORMAT_COUNT; ++input_format)
 197       {
 198          const struct util_format_description* input_format_desc = util_format_description(input_format);
 199       const struct util_format_pack_description* input_format_pack = util_format_pack_description(input_format);
 200       const struct util_format_unpack_description* input_format_unpack = util_format_unpack_description(input_format);
 201          unsigned input_format_size;
 202          struct translate* translate[2];
 203          unsigned fail = 0;
 204          unsigned used_generic = 0;
 205          unsigned input_normalized = 0;
 206          boolean input_is_float = FALSE;
 207
 208          if (!input_format_desc
 209                || !input_format_unpack->fetch_rgba
 210                || !input_format_pack->pack_rgba_float
 211                || input_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
 212                || input_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
 213                || !translate_is_output_format_supported(input_format))
 214             continue;
 215
 216          input_format_size = util_format_get_stride(input_format, 1);
 217
 218          for(i = 0; i < input_format_desc->nr_channels; ++i)
 219          {
 220             if(input_format_desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT)
 221             {
 222                input_is_float = 1;
 223                input_normalized |= 1 << 1;
 224             }
 225             else
 226                input_normalized |= (1 << input_format_desc->channel[i].normalized);
 227          }
 228
 229          if(((input_normalized | output_normalized) == 3)
 230                || ((input_normalized & 1) && (output_normalized & 1)
 231                      && input_format_size * output_format_desc->nr_channels > output_format_size * input_format_desc->nr_channels))
 232             continue;
 233
 234          key.element[0].input_format = input_format;
 235          key.element[0].output_format = output_format;
 236          key.output_stride = output_format_size;
 237          translate[0] = create_fn(&key);
 238          if (!translate[0])
 239             continue;
 240
 241          key.element[0].input_format = output_format;
 242          key.element[0].output_format = input_format;
 243          key.output_stride = input_format_size;
 244          translate[1] = create_fn(&key);
 245          if(!translate[1])
 246          {
 247             used_generic = 1;
 248             translate[1] = translate_generic_create(&key);
 249             if(!translate[1])
 250                continue;
 251          }
 252
 253          for(i = 1; i < 5; ++i)
 254             memset(buffer[i], 0xcd - (0x22 * i), 4096);
 255
 256          if(input_is_float && input_format_desc->channel[0].size == 32)
 257             buffer[0] = (unsigned char*)float_buffer;
 258          else if(input_is_float && input_format_desc->channel[0].size == 64)
 259             buffer[0] = (unsigned char*)double_buffer;
 260          else if(input_is_float && input_format_desc->channel[0].size == 16)
 261             buffer[0] = (unsigned char*)half_buffer;
 262          else if(input_is_float)
 263             abort();
 264          else
 265             buffer[0] = byte_buffer;
 266
 267          translate[0]->set_buffer(translate[0], 0, buffer[0], input_format_size, count - 1);
 268          translate[0]->run_elts(translate[0], elts, count, 0, 0, buffer[1]);
 269          translate[1]->set_buffer(translate[1], 0, buffer[1], output_format_size, count - 1);
 270          translate[1]->run_elts(translate[1], elts, count, 0, 0, buffer[2]);
 271          translate[0]->set_buffer(translate[0], 0, buffer[2], input_format_size, count - 1);
 272          translate[0]->run_elts(translate[0], elts, count, 0, 0, buffer[3]);
 273          translate[1]->set_buffer(translate[1], 0, buffer[3], output_format_size, count - 1);
 274          translate[1]->run_elts(translate[1], elts, count, 0, 0, buffer[4]);
 275
 276          for (i = 0; i < count; ++i)
 277          {
 278             float a[4];
 279             float b[4];
 280             input_format_unpack->fetch_rgba(a, buffer[2] + i * input_format_size, 0, 0);
 281             input_format_unpack->fetch_rgba(b, buffer[4] + i * input_format_size, 0, 0);
 282
 283             for (j = 0; j < count; ++j)
 284             {
 285                float d = a[j] - b[j];
 286                if (d > error || d < -error)
 287                {
 288                   fail = 1;
 289                   break;
 290                }
 291             }
 292          }
 293
 294          printf("%s%s: %s -> %s -> %s -> %s -> %s\n",
 295                fail ? "FAIL" : "PASS",
 296                used_generic ? "[GENERIC]" : "",
 297                input_format_desc->name, output_format_desc->name, input_format_desc->name, output_format_desc->name, input_format_desc->name);
 298
 299          if (1)
 300          {
 301             for (i = 0; i < ARRAY_SIZE(buffer); ++i)
 302             {
 303                unsigned format_size = (i & 1) ? output_format_size : input_format_size;
 304                printf("%c ", (i == 2 || i == 4) ? '*' : ' ');
 305                for (j = 0; j < count; ++j)
 306                {
 307                   for (k = 0; k < format_size; ++k)
 308                   {
 309                      printf("%02x", buffer[i][j * format_size + k]);
 310                   }
 311                   printf(" ");
 312                }
 313                printf("\n");
 314             }
 315          }
 316
 317          if (!fail)
 318             ++passed;
 319          ++total;
 320
 321          if(translate[1])
 322             translate[1]->release(translate[1]);
 323          translate[0]->release(translate[0]);
 324       }
 325    }
 326
 327    printf("%u/%u tests passed for translate_%s\n", passed, total, argv[1]);
 328    return passed != total;
 329 }