Home | History | Annotate | Download | only in unit
      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_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       unsigned output_format_size;
    176       unsigned output_normalized = 0;
    177 
    178       if (!output_format_desc
    179             || !output_format_desc->fetch_rgba_float
    180             || !output_format_desc->pack_rgba_float
    181             || output_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
    182             || output_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
    183             || !translate_is_output_format_supported(output_format))
    184          continue;
    185 
    186       for(i = 0; i < output_format_desc->nr_channels; ++i)
    187       {
    188          if(output_format_desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT)
    189             output_normalized |= (1 << output_format_desc->channel[i].normalized);
    190       }
    191 
    192       output_format_size = util_format_get_stride(output_format, 1);
    193 
    194       for (input_format = 1; input_format < PIPE_FORMAT_COUNT; ++input_format)
    195       {
    196          const struct util_format_description* input_format_desc = util_format_description(input_format);
    197          unsigned input_format_size;
    198          struct translate* translate[2];
    199          unsigned fail = 0;
    200          unsigned used_generic = 0;
    201          unsigned input_normalized = 0;
    202          boolean input_is_float = FALSE;
    203 
    204          if (!input_format_desc
    205                || !input_format_desc->fetch_rgba_float
    206                || !input_format_desc->pack_rgba_float
    207                || input_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
    208                || input_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
    209                || !translate_is_output_format_supported(input_format))
    210             continue;
    211 
    212          input_format_size = util_format_get_stride(input_format, 1);
    213 
    214          for(i = 0; i < input_format_desc->nr_channels; ++i)
    215          {
    216             if(input_format_desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT)
    217             {
    218                input_is_float = 1;
    219                input_normalized |= 1 << 1;
    220             }
    221             else
    222                input_normalized |= (1 << input_format_desc->channel[i].normalized);
    223          }
    224 
    225          if(((input_normalized | output_normalized) == 3)
    226                || ((input_normalized & 1) && (output_normalized & 1)
    227                      && input_format_size * output_format_desc->nr_channels > output_format_size * input_format_desc->nr_channels))
    228             continue;
    229 
    230          key.element[0].input_format = input_format;
    231          key.element[0].output_format = output_format;
    232          key.output_stride = output_format_size;
    233          translate[0] = create_fn(&key);
    234          if (!translate[0])
    235             continue;
    236 
    237          key.element[0].input_format = output_format;
    238          key.element[0].output_format = input_format;
    239          key.output_stride = input_format_size;
    240          translate[1] = create_fn(&key);
    241          if(!translate[1])
    242          {
    243             used_generic = 1;
    244             translate[1] = translate_generic_create(&key);
    245             if(!translate[1])
    246                continue;
    247          }
    248 
    249          for(i = 1; i < 5; ++i)
    250             memset(buffer[i], 0xcd - (0x22 * i), 4096);
    251 
    252          if(input_is_float && input_format_desc->channel[0].size == 32)
    253             buffer[0] = (unsigned char*)float_buffer;
    254          else if(input_is_float && input_format_desc->channel[0].size == 64)
    255             buffer[0] = (unsigned char*)double_buffer;
    256          else if(input_is_float && input_format_desc->channel[0].size == 16)
    257             buffer[0] = (unsigned char*)half_buffer;
    258          else if(input_is_float)
    259             abort();
    260          else
    261             buffer[0] = byte_buffer;
    262 
    263          translate[0]->set_buffer(translate[0], 0, buffer[0], input_format_size, count - 1);
    264          translate[0]->run_elts(translate[0], elts, count, 0, 0, buffer[1]);
    265          translate[1]->set_buffer(translate[1], 0, buffer[1], output_format_size, count - 1);
    266          translate[1]->run_elts(translate[1], elts, count, 0, 0, buffer[2]);
    267          translate[0]->set_buffer(translate[0], 0, buffer[2], input_format_size, count - 1);
    268          translate[0]->run_elts(translate[0], elts, count, 0, 0, buffer[3]);
    269          translate[1]->set_buffer(translate[1], 0, buffer[3], output_format_size, count - 1);
    270          translate[1]->run_elts(translate[1], elts, count, 0, 0, buffer[4]);
    271 
    272          for (i = 0; i < count; ++i)
    273          {
    274             float a[4];
    275             float b[4];
    276             input_format_desc->fetch_rgba_float(a, buffer[2] + i * input_format_size, 0, 0);
    277             input_format_desc->fetch_rgba_float(b, buffer[4] + i * input_format_size, 0, 0);
    278 
    279             for (j = 0; j < count; ++j)
    280             {
    281                float d = a[j] - b[j];
    282                if (d > error || d < -error)
    283                {
    284                   fail = 1;
    285                   break;
    286                }
    287             }
    288          }
    289 
    290          printf("%s%s: %s -> %s -> %s -> %s -> %s\n",
    291                fail ? "FAIL" : "PASS",
    292                used_generic ? "[GENERIC]" : "",
    293                input_format_desc->name, output_format_desc->name, input_format_desc->name, output_format_desc->name, input_format_desc->name);
    294 
    295          if (1)
    296          {
    297             for (i = 0; i < ARRAY_SIZE(buffer); ++i)
    298             {
    299                unsigned format_size = (i & 1) ? output_format_size : input_format_size;
    300                printf("%c ", (i == 2 || i == 4) ? '*' : ' ');
    301                for (j = 0; j < count; ++j)
    302                {
    303                   for (k = 0; k < format_size; ++k)
    304                   {
    305                      printf("%02x", buffer[i][j * format_size + k]);
    306                   }
    307                   printf(" ");
    308                }
    309                printf("\n");
    310             }
    311          }
    312 
    313          if (!fail)
    314             ++passed;
    315          ++total;
    316 
    317          if(translate[1])
    318             translate[1]->release(translate[1]);
    319          translate[0]->release(translate[0]);
    320       }
    321    }
    322 
    323    printf("%u/%u tests passed for translate_%s\n", passed, total, argv[1]);
    324    return passed != total;
    325 }
    326