translate_test.c revision 4d946c4e8adf3f0ac447b6a9a6caf17392b816cd 
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
33int main(int argc, char** argv)
34{
35   struct translate *(*create_fn)(const struct translate_key *key) = 0;
36
37   struct translate_key key;
38   unsigned output_format;
39   unsigned input_format;
40   unsigned buffer_size = 4096;
41   unsigned char* buffer[5];
42   unsigned char* byte_buffer;
43   float* float_buffer;
44   double* double_buffer;
45   unsigned count = 4;
46   unsigned i, j, k;
47   unsigned passed = 0;
48   unsigned total = 0;
49   const float error = 0.03125;
50
51   create_fn = 0;
52
53   util_cpu_detect();
54
55   if(argc <= 1)
56   {}
57   else if (!strcmp(argv[1], "generic"))
58      create_fn = translate_generic_create;
59   else if (!strcmp(argv[1], "x86"))
60      create_fn = translate_sse2_create;
61   else if (!strcmp(argv[1], "nosse"))
62   {
63      util_cpu_caps.has_sse = 0;
64      util_cpu_caps.has_sse2 = 0;
65      util_cpu_caps.has_sse3 = 0;
66      util_cpu_caps.has_sse4_1 = 0;
67      create_fn = translate_sse2_create;
68   }
69   else if (!strcmp(argv[1], "sse"))
70   {
71      if(!util_cpu_caps.has_sse || !rtasm_cpu_has_sse())
72      {
73         printf("Error: CPU doesn't support SSE (test with qemu)\n");
74         return 2;
75      }
76      util_cpu_caps.has_sse2 = 0;
77      util_cpu_caps.has_sse3 = 0;
78      util_cpu_caps.has_sse4_1 = 0;
79      create_fn = translate_sse2_create;
80   }
81   else if (!strcmp(argv[1], "sse2"))
82   {
83      if(!util_cpu_caps.has_sse2 || !rtasm_cpu_has_sse())
84      {
85         printf("Error: CPU doesn't support SSE2 (test with qemu)\n");
86         return 2;
87      }
88      util_cpu_caps.has_sse3 = 0;
89      util_cpu_caps.has_sse4_1 = 0;
90      create_fn = translate_sse2_create;
91   }
92   else if (!strcmp(argv[1], "sse3"))
93   {
94      if(!util_cpu_caps.has_sse3 || !rtasm_cpu_has_sse())
95      {
96         printf("Error: CPU doesn't support SSE3 (test with qemu)\n");
97         return 2;
98      }
99      util_cpu_caps.has_sse4_1 = 0;
100      create_fn = translate_sse2_create;
101   }
102   else if (!strcmp(argv[1], "sse4.1"))
103   {
104      if(!util_cpu_caps.has_sse4_1 || !rtasm_cpu_has_sse())
105      {
106         printf("Error: CPU doesn't support SSE4.1 (test with qemu)\n");
107         return 2;
108      }
109      create_fn = translate_sse2_create;
110   }
111
112   if (!create_fn)
113   {
114      printf("Usage: ./translate_test [generic|x86|nosse|sse|sse2|sse3|sse4.1]\n");
115      return 2;
116   }
117
118   for (i = 1; i < Elements(buffer); ++i)
119      buffer[i] = align_malloc(buffer_size, 4096);
120
121   byte_buffer = align_malloc(buffer_size, 4096);
122   float_buffer = align_malloc(buffer_size, 4096);
123   double_buffer = align_malloc(buffer_size, 4096);
124
125   key.nr_elements = 1;
126   key.element[0].input_buffer = 0;
127   key.element[0].input_offset = 0;
128   key.element[0].output_offset = 0;
129   key.element[0].type = TRANSLATE_ELEMENT_NORMAL;
130   key.element[0].instance_divisor = 0;
131
132   srand48(4359025);
133
134   /* avoid negative values that work badly when converted to unsigned format*/
135   for (i = 0; i < buffer_size / sizeof(unsigned); ++i)
136      ((unsigned*)byte_buffer)[i] = mrand48() & 0x7f7f7f7f;
137
138   for (i = 0; i < buffer_size / sizeof(float); ++i)
139      float_buffer[i] = (float)drand48();
140
141   for (i = 0; i < buffer_size / sizeof(double); ++i)
142      double_buffer[i] = drand48();
143
144   for (output_format = 1; output_format < PIPE_FORMAT_COUNT; ++output_format)
145   {
146      const struct util_format_description* output_format_desc = util_format_description(output_format);
147      unsigned output_format_size;
148      unsigned output_normalized = 0;
149
150      if (!output_format_desc
151            || !output_format_desc->fetch_rgba_float
152            || !output_format_desc->pack_rgba_float
153            || output_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
154            || output_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
155            || !translate_is_output_format_supported(output_format))
156         continue;
157
158      for(i = 0; i < output_format_desc->nr_channels; ++i)
159      {
160         if(output_format_desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT)
161            output_normalized |= (1 << output_format_desc->channel[i].normalized);
162      }
163
164      output_format_size = util_format_get_stride(output_format, 1);
165
166      for (input_format = 1; input_format < PIPE_FORMAT_COUNT; ++input_format)
167      {
168         const struct util_format_description* input_format_desc = util_format_description(input_format);
169         unsigned input_format_size;
170         struct translate* translate[2];
171         unsigned fail = 0;
172         unsigned used_generic = 0;
173         unsigned input_normalized = 0;
174         boolean input_is_float = FALSE;
175
176         if (!input_format_desc
177               || !input_format_desc->fetch_rgba_float
178               || !input_format_desc->pack_rgba_float
179               || input_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
180               || input_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
181               || !translate_is_output_format_supported(input_format))
182            continue;
183
184         input_format_size = util_format_get_stride(input_format, 1);
185
186         for(i = 0; i < input_format_desc->nr_channels; ++i)
187         {
188            if(input_format_desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT)
189            {
190               input_is_float = 1;
191               input_normalized |= 1 << 1;
192            }
193            else
194               input_normalized |= (1 << input_format_desc->channel[i].normalized);
195         }
196
197         if(((input_normalized | output_normalized) == 3)
198               || ((input_normalized & 1) && (output_normalized & 1)
199                     && input_format_size * output_format_desc->nr_channels > output_format_size * input_format_desc->nr_channels))
200            continue;
201
202         key.element[0].input_format = input_format;
203         key.element[0].output_format = output_format;
204         key.output_stride = output_format_size;
205         translate[0] = create_fn(&key);
206         if (!translate[0])
207            continue;
208
209         key.element[0].input_format = output_format;
210         key.element[0].output_format = input_format;
211         key.output_stride = input_format_size;
212         translate[1] = create_fn(&key);
213         if(!translate[1])
214         {
215            used_generic = 1;
216            translate[1] = translate_generic_create(&key);
217            if(!translate[1])
218               continue;
219         }
220
221         for(i = 1; i < 5; ++i)
222            memset(buffer[i], 0xcd - (0x22 * i), 4096);
223
224         if(input_is_float && input_format_desc->channel[0].size == 32)
225            buffer[0] = (unsigned char*)float_buffer;
226         else if(input_is_float && input_format_desc->channel[0].size == 64)
227            buffer[0] = (unsigned char*)double_buffer;
228         else if(input_is_float)
229            abort();
230         else
231            buffer[0] = byte_buffer;
232
233         translate[0]->set_buffer(translate[0], 0, buffer[0], input_format_size, ~0);
234         translate[0]->run(translate[0], 0, count, 0, buffer[1]);
235         translate[1]->set_buffer(translate[1], 0, buffer[1], output_format_size, ~0);
236         translate[1]->run(translate[1], 0, count, 0, buffer[2]);
237         translate[0]->set_buffer(translate[0], 0, buffer[2], input_format_size, ~0);
238         translate[0]->run(translate[0], 0, count, 0, buffer[3]);
239         translate[1]->set_buffer(translate[1], 0, buffer[3], output_format_size, ~0);
240         translate[1]->run(translate[1], 0, count, 0, buffer[4]);
241
242         for (i = 0; i < count; ++i)
243         {
244            float a[4];
245            float b[4];
246            input_format_desc->fetch_rgba_float(a, buffer[2] + i * input_format_size, 0, 0);
247            input_format_desc->fetch_rgba_float(b, buffer[4] + i * input_format_size, 0, 0);
248
249            for (j = 0; j < count; ++j)
250            {
251               float d = a[j] - b[j];
252               if (d > error || d < -error)
253               {
254                  fail = 1;
255                  break;
256               }
257            }
258         }
259
260         printf("%s%s: %s -> %s -> %s -> %s -> %s\n",
261               fail ? "FAIL" : "PASS",
262               used_generic ? "[GENERIC]" : "",
263               input_format_desc->name, output_format_desc->name, input_format_desc->name, output_format_desc->name, input_format_desc->name);
264
265         if (1)
266         {
267            for (i = 0; i < Elements(buffer); ++i)
268            {
269               unsigned format_size = (i & 1) ? output_format_size : input_format_size;
270               printf("%c ", (i == 2 || i == 4) ? '*' : ' ');
271               for (j = 0; j < count; ++j)
272               {
273                  for (k = 0; k < format_size; ++k)
274                  {
275                     printf("%02x", buffer[i][j * format_size + k]);
276                  }
277                  printf(" ");
278               }
279               printf("\n");
280            }
281         }
282
283         if (!fail)
284            ++passed;
285         ++total;
286
287         if(translate[1])
288            translate[1]->release(translate[1]);
289         translate[0]->release(translate[0]);
290      }
291   }
292
293   printf("%u/%u tests passed for translate_%s\n", passed, total, argv[1]);
294   return passed != total;
295}
296