src/intel/vulkan/anv_entrypoints_gen.py

   1 # coding=utf-8
   2 #
   3 # Copyright © 2015, 2017 Intel Corporation
   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 DEALINGS
  22 # IN THE SOFTWARE.
  23 #
  24
  25 import argparse
  26 import functools
  27 import os
  28 import textwrap
  29 import xml.etree.cElementTree as et
  30
  31 from mako.template import Template
  32
  33 MAX_API_VERSION = 1.0
  34
  35 SUPPORTED_EXTENSIONS = [
  36     'VK_KHR_descriptor_update_template',
  37     'VK_KHR_get_memory_requirements2',
  38     'VK_KHR_get_physical_device_properties2',
  39     'VK_KHR_get_surface_capabilities2',
  40     'VK_KHR_incremental_present',
  41     'VK_KHR_maintenance1',
  42     'VK_KHR_push_descriptor',
  43     'VK_KHR_sampler_mirror_clamp_to_edge',
  44     'VK_KHR_shader_draw_parameters',
  45     'VK_KHR_surface',
  46     'VK_KHR_swapchain',
  47     'VK_KHR_wayland_surface',
  48     'VK_KHR_xcb_surface',
  49     'VK_KHR_xlib_surface',
  50     'VK_KHX_multiview',
  51 ]
  52
  53 # We generate a static hash table for entry point lookup
  54 # (vkGetProcAddress). We use a linear congruential generator for our hash
  55 # function and a power-of-two size table. The prime numbers are determined
  56 # experimentally.
  57
  58 TEMPLATE_H = Template(textwrap.dedent("""\
  59     /* This file generated from ${filename}, don't edit directly. */
  60
  61     struct anv_dispatch_table {
  62        union {
  63           void *entrypoints[${len(entrypoints)}];
  64           struct {
  65           % for _, name, _, _, _, guard in entrypoints:
  66             % if guard is not None:
  67     #ifdef ${guard}
  68               PFN_vk${name} ${name};
  69     #else
  70               void *${name};
  71     # endif
  72             % else:
  73               PFN_vk${name} ${name};
  74             % endif
  75           % endfor
  76           };
  77        };
  78     };
  79
  80     % for type_, name, args, num, h, guard in entrypoints:
  81       % if guard is not None:
  82     #ifdef ${guard}
  83       % endif
  84       ${type_} anv_${name}(${args});
  85       ${type_} gen7_${name}(${args});
  86       ${type_} gen75_${name}(${args});
  87       ${type_} gen8_${name}(${args});
  88       ${type_} gen9_${name}(${args});
  89       ${type_} gen10_${name}(${args});
  90       % if guard is not None:
  91     #endif // ${guard}
  92       % endif
  93     % endfor
  94     """), output_encoding='utf-8')
  95
  96 TEMPLATE_C = Template(textwrap.dedent(u"""\
  97     /*
  98      * Copyright © 2015 Intel Corporation
  99      *
 100      * Permission is hereby granted, free of charge, to any person obtaining a
 101      * copy of this software and associated documentation files (the "Software"),
 102      * to deal in the Software without restriction, including without limitation
 103      * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 104      * and/or sell copies of the Software, and to permit persons to whom the
 105      * Software is furnished to do so, subject to the following conditions:
 106      *
 107      * The above copyright notice and this permission notice (including the next
 108      * paragraph) shall be included in all copies or substantial portions of the
 109      * Software.
 110      *
 111      * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 112      * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 113      * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 114      * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 115      * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 116      * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 117      * IN THE SOFTWARE.
 118      */
 119
 120     /* This file generated from ${filename}, don't edit directly. */
 121
 122     #include "anv_private.h"
 123
 124     struct anv_entrypoint {
 125        uint32_t name;
 126        uint32_t hash;
 127     };
 128
 129     /* We use a big string constant to avoid lots of reloctions from the entry
 130      * point table to lots of little strings. The entries in the entry point table
 131      * store the index into this big string.
 132      */
 133
 134     static const char strings[] =
 135     % for _, name, _, _, _, _ in entrypoints:
 136         "vk${name}\\0"
 137     % endfor
 138     ;
 139
 140     static const struct anv_entrypoint entrypoints[] = {
 141     % for _, _, _, num, h, _ in entrypoints:
 142         { ${offsets[num]}, ${'{:0=#8x}'.format(h)} },
 143     % endfor
 144     };
 145
 146     /* Weak aliases for all potential implementations. These will resolve to
 147      * NULL if they're not defined, which lets the resolve_entrypoint() function
 148      * either pick the correct entry point.
 149      */
 150
 151     % for layer in ['anv', 'gen7', 'gen75', 'gen8', 'gen9', 'gen10']:
 152       % for type_, name, args, _, _, guard in entrypoints:
 153         % if guard is not None:
 154     #ifdef ${guard}
 155         % endif
 156         ${type_} ${layer}_${name}(${args}) __attribute__ ((weak));
 157         % if guard is not None:
 158     #endif // ${guard}
 159         % endif
 160       % endfor
 161
 162       const struct anv_dispatch_table ${layer}_layer = {
 163       % for _, name, args, _, _, guard in entrypoints:
 164         % if guard is not None:
 165     #ifdef ${guard}
 166         % endif
 167         .${name} = ${layer}_${name},
 168         % if guard is not None:
 169     #endif // ${guard}
 170         % endif
 171       % endfor
 172       };
 173     % endfor
 174
 175     static void * __attribute__ ((noinline))
 176     anv_resolve_entrypoint(const struct gen_device_info *devinfo, uint32_t index)
 177     {
 178        if (devinfo == NULL) {
 179           return anv_layer.entrypoints[index];
 180        }
 181
 182        switch (devinfo->gen) {
 183        case 10:
 184           if (gen10_layer.entrypoints[index])
 185              return gen10_layer.entrypoints[index];
 186           /* fall through */
 187        case 9:
 188           if (gen9_layer.entrypoints[index])
 189              return gen9_layer.entrypoints[index];
 190           /* fall through */
 191        case 8:
 192           if (gen8_layer.entrypoints[index])
 193              return gen8_layer.entrypoints[index];
 194           /* fall through */
 195        case 7:
 196           if (devinfo->is_haswell && gen75_layer.entrypoints[index])
 197              return gen75_layer.entrypoints[index];
 198
 199           if (gen7_layer.entrypoints[index])
 200              return gen7_layer.entrypoints[index];
 201           /* fall through */
 202        case 0:
 203           return anv_layer.entrypoints[index];
 204        default:
 205           unreachable("unsupported gen\\n");
 206        }
 207     }
 208
 209     /* Hash table stats:
 210      * size ${hash_size} entries
 211      * collisions entries:
 212     % for i in xrange(10):
 213      *     ${i}${'+' if i == 9 else ''}     ${collisions[i]}
 214     % endfor
 215      */
 216
 217     #define none ${'{:#x}'.format(none)}
 218     static const uint16_t map[] = {
 219     % for i in xrange(0, hash_size, 8):
 220       % for j in xrange(i, i + 8):
 221         ## This is 6 because the 0x is counted in the length
 222         % if mapping[j] & 0xffff == 0xffff:
 223           none,
 224         % else:
 225           ${'{:0=#6x}'.format(mapping[j] & 0xffff)},
 226         % endif
 227       % endfor
 228     % endfor
 229     };
 230
 231     void *
 232     anv_lookup_entrypoint(const struct gen_device_info *devinfo, const char *name)
 233     {
 234        static const uint32_t prime_factor = ${prime_factor};
 235        static const uint32_t prime_step = ${prime_step};
 236        const struct anv_entrypoint *e;
 237        uint32_t hash, h, i;
 238        const char *p;
 239
 240        hash = 0;
 241        for (p = name; *p; p++)
 242           hash = hash * prime_factor + *p;
 243
 244        h = hash;
 245        do {
 246           i = map[h & ${hash_mask}];
 247           if (i == none)
 248              return NULL;
 249           e = &entrypoints[i];
 250           h += prime_step;
 251        } while (e->hash != hash);
 252
 253        if (strcmp(name, strings + e->name) != 0)
 254           return NULL;
 255
 256        return anv_resolve_entrypoint(devinfo, i);
 257     }"""), output_encoding='utf-8')
 258
 259 NONE = 0xffff
 260 HASH_SIZE = 256
 261 U32_MASK = 2**32 - 1
 262 HASH_MASK = HASH_SIZE - 1
 263
 264 PRIME_FACTOR = 5024183
 265 PRIME_STEP = 19
 266
 267
 268 def cal_hash(name):
 269     """Calculate the same hash value that Mesa will calculate in C."""
 270     return functools.reduce(
 271         lambda h, c: (h * PRIME_FACTOR + ord(c)) & U32_MASK, name, 0)
 272
 273
 274 def get_entrypoints(doc, entrypoints_to_defines):
 275     """Extract the entry points from the registry."""
 276     entrypoints = []
 277
 278     enabled_commands = set()
 279     for feature in doc.findall('./feature'):
 280         assert feature.attrib['api'] == 'vulkan'
 281         if float(feature.attrib['number']) > MAX_API_VERSION:
 282             continue
 283
 284         for command in feature.findall('./require/command'):
 285             enabled_commands.add(command.attrib['name'])
 286
 287     for extension in doc.findall('.extensions/extension'):
 288         if extension.attrib['name'] not in SUPPORTED_EXTENSIONS:
 289             continue
 290
 291         assert extension.attrib['supported'] == 'vulkan'
 292         for command in extension.findall('./require/command'):
 293             enabled_commands.add(command.attrib['name'])
 294
 295     index = 0
 296     for command in doc.findall('./commands/command'):
 297         type = command.find('./proto/type').text
 298         fullname = command.find('./proto/name').text
 299
 300         if fullname not in enabled_commands:
 301             continue
 302
 303         shortname = fullname[2:]
 304         params = (''.join(p.itertext()) for p in command.findall('./param'))
 305         params = ', '.join(params)
 306         guard = entrypoints_to_defines.get(fullname)
 307         entrypoints.append((type, shortname, params, index, cal_hash(fullname), guard))
 308         index += 1
 309
 310     return entrypoints
 311
 312
 313 def get_entrypoints_defines(doc):
 314     """Maps entry points to extension defines."""
 315     entrypoints_to_defines = {}
 316
 317     for extension in doc.findall('./extensions/extension[@protect]'):
 318         define = extension.attrib['protect']
 319
 320         for entrypoint in extension.findall('./require/command'):
 321             fullname = entrypoint.attrib['name']
 322             entrypoints_to_defines[fullname] = define
 323
 324     return entrypoints_to_defines
 325
 326
 327 def gen_code(entrypoints):
 328     """Generate the C code."""
 329     i = 0
 330     offsets = []
 331     for _, name, _, _, _, _ in entrypoints:
 332         offsets.append(i)
 333         i += 2 + len(name) + 1
 334
 335     mapping = [NONE] * HASH_SIZE
 336     collisions = [0] * 10
 337     for _, name, _, num, h, _ in entrypoints:
 338         level = 0
 339         while mapping[h & HASH_MASK] != NONE:
 340             h = h + PRIME_STEP
 341             level = level + 1
 342         if level > 9:
 343             collisions[9] += 1
 344         else:
 345             collisions[level] += 1
 346         mapping[h & HASH_MASK] = num
 347
 348     return TEMPLATE_C.render(entrypoints=entrypoints,
 349                              offsets=offsets,
 350                              collisions=collisions,
 351                              mapping=mapping,
 352                              hash_mask=HASH_MASK,
 353                              prime_step=PRIME_STEP,
 354                              prime_factor=PRIME_FACTOR,
 355                              none=NONE,
 356                              hash_size=HASH_SIZE,
 357                              filename=os.path.basename(__file__))
 358
 359
 360 def main():
 361     parser = argparse.ArgumentParser()
 362     parser.add_argument('--outdir', help='Where to write the files.',
 363                         required=True)
 364     parser.add_argument('--xml', help='Vulkan API XML file.', required=True)
 365     args = parser.parse_args()
 366
 367     doc = et.parse(args.xml)
 368     entrypoints = get_entrypoints(doc, get_entrypoints_defines(doc))
 369
 370     # Manually add CreateDmaBufImageINTEL for which we don't have an extension
 371     # defined.
 372     entrypoints.append(('VkResult', 'CreateDmaBufImageINTEL',
 373                         'VkDevice device, ' +
 374                         'const VkDmaBufImageCreateInfo* pCreateInfo, ' +
 375                         'const VkAllocationCallbacks* pAllocator,' +
 376                         'VkDeviceMemory* pMem,' +
 377                         'VkImage* pImage', len(entrypoints),
 378                         cal_hash('vkCreateDmaBufImageINTEL'), None))
 379
 380     # For outputting entrypoints.h we generate a anv_EntryPoint() prototype
 381     # per entry point.
 382     with open(os.path.join(args.outdir, 'anv_entrypoints.h'), 'wb') as f:
 383         f.write(TEMPLATE_H.render(entrypoints=entrypoints,
 384                                   filename=os.path.basename(__file__)))
 385     with open(os.path.join(args.outdir, 'anv_entrypoints.c'), 'wb') as f:
 386         f.write(gen_code(entrypoints))
 387
 388
 389 if __name__ == '__main__':
 390     main()