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_physical_device_properties2',
  38     'VK_KHR_incremental_present',
  39     'VK_KHR_maintenance1',
  40     'VK_KHR_push_descriptor',
  41     'VK_KHR_sampler_mirror_clamp_to_edge',
  42     'VK_KHR_shader_draw_parameters',
  43     'VK_KHR_surface',
  44     'VK_KHR_swapchain',
  45     'VK_KHR_wayland_surface',
  46     'VK_KHR_xcb_surface',
  47     'VK_KHR_xlib_surface',
  48     'VK_KHX_external_memory',
  49     'VK_KHX_external_memory_capabilities',
  50     'VK_KHX_external_memory_fd',
  51     'VK_KHX_external_semaphore',
  52     'VK_KHX_external_semaphore_capabilities',
  53     'VK_KHX_external_semaphore_fd',
  54     'VK_KHX_multiview',
  55 ]
  56
  57 # We generate a static hash table for entry point lookup
  58 # (vkGetProcAddress). We use a linear congruential generator for our hash
  59 # function and a power-of-two size table. The prime numbers are determined
  60 # experimentally.
  61
  62 TEMPLATE_H = Template(textwrap.dedent("""\
  63     /* This file generated from ${filename}, don't edit directly. */
  64
  65     struct anv_dispatch_table {
  66        union {
  67           void *entrypoints[${len(entrypoints)}];
  68           struct {
  69           % for _, name, _, _, _, guard in entrypoints:
  70             % if guard is not None:
  71     #ifdef ${guard}
  72               PFN_vk${name} ${name};
  73     #else
  74               void *${name};
  75     # endif
  76             % else:
  77               PFN_vk${name} ${name};
  78             % endif
  79           % endfor
  80           };
  81        };
  82     };
  83
  84     % for type_, name, args, num, h, guard in entrypoints:
  85       % if guard is not None:
  86     #ifdef ${guard}
  87       % endif
  88       ${type_} anv_${name}(${args});
  89       ${type_} gen7_${name}(${args});
  90       ${type_} gen75_${name}(${args});
  91       ${type_} gen8_${name}(${args});
  92       ${type_} gen9_${name}(${args});
  93       % if guard is not None:
  94     #endif // ${guard}
  95       % endif
  96     % endfor
  97     """), output_encoding='utf-8')
  98
  99 TEMPLATE_C = Template(textwrap.dedent(u"""\
 100     /*
 101      * Copyright © 2015 Intel Corporation
 102      *
 103      * Permission is hereby granted, free of charge, to any person obtaining a
 104      * copy of this software and associated documentation files (the "Software"),
 105      * to deal in the Software without restriction, including without limitation
 106      * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 107      * and/or sell copies of the Software, and to permit persons to whom the
 108      * Software is furnished to do so, subject to the following conditions:
 109      *
 110      * The above copyright notice and this permission notice (including the next
 111      * paragraph) shall be included in all copies or substantial portions of the
 112      * Software.
 113      *
 114      * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 115      * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 116      * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 117      * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 118      * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 119      * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 120      * IN THE SOFTWARE.
 121      */
 122
 123     /* This file generated from ${filename}, don't edit directly. */
 124
 125     #include "anv_private.h"
 126
 127     struct anv_entrypoint {
 128        uint32_t name;
 129        uint32_t hash;
 130     };
 131
 132     /* We use a big string constant to avoid lots of reloctions from the entry
 133      * point table to lots of little strings. The entries in the entry point table
 134      * store the index into this big string.
 135      */
 136
 137     static const char strings[] =
 138     % for _, name, _, _, _, _ in entrypoints:
 139         "vk${name}\\0"
 140     % endfor
 141     ;
 142
 143     static const struct anv_entrypoint entrypoints[] = {
 144     % for _, _, _, num, h, _ in entrypoints:
 145         { ${offsets[num]}, ${'{:0=#8x}'.format(h)} },
 146     % endfor
 147     };
 148
 149     /* Weak aliases for all potential implementations. These will resolve to
 150      * NULL if they're not defined, which lets the resolve_entrypoint() function
 151      * either pick the correct entry point.
 152      */
 153
 154     % for layer in ['anv', 'gen7', 'gen75', 'gen8', 'gen9']:
 155       % for type_, name, args, _, _, guard in entrypoints:
 156         % if guard is not None:
 157     #ifdef ${guard}
 158         % endif
 159         ${type_} ${layer}_${name}(${args}) __attribute__ ((weak));
 160         % if guard is not None:
 161     #endif // ${guard}
 162         % endif
 163       % endfor
 164
 165       const struct anv_dispatch_table ${layer}_layer = {
 166       % for _, name, args, _, _, guard in entrypoints:
 167         % if guard is not None:
 168     #ifdef ${guard}
 169         % endif
 170         .${name} = ${layer}_${name},
 171         % if guard is not None:
 172     #endif // ${guard}
 173         % endif
 174       % endfor
 175       };
 176     % endfor
 177
 178     static void * __attribute__ ((noinline))
 179     anv_resolve_entrypoint(const struct gen_device_info *devinfo, uint32_t index)
 180     {
 181        if (devinfo == NULL) {
 182           return anv_layer.entrypoints[index];
 183        }
 184
 185        switch (devinfo->gen) {
 186        case 9:
 187           if (gen9_layer.entrypoints[index])
 188              return gen9_layer.entrypoints[index];
 189           /* fall through */
 190        case 8:
 191           if (gen8_layer.entrypoints[index])
 192              return gen8_layer.entrypoints[index];
 193           /* fall through */
 194        case 7:
 195           if (devinfo->is_haswell && gen75_layer.entrypoints[index])
 196              return gen75_layer.entrypoints[index];
 197
 198           if (gen7_layer.entrypoints[index])
 199              return gen7_layer.entrypoints[index];
 200           /* fall through */
 201        case 0:
 202           return anv_layer.entrypoints[index];
 203        default:
 204           unreachable("unsupported gen\\n");
 205        }
 206     }
 207
 208     /* Hash table stats:
 209      * size ${hash_size} entries
 210      * collisions entries:
 211     % for i in xrange(10):
 212      *     ${i}${'+' if i == 9 else ''}     ${collisions[i]}
 213     % endfor
 214      */
 215
 216     #define none ${'{:#x}'.format(none)}
 217     static const uint16_t map[] = {
 218     % for i in xrange(0, hash_size, 8):
 219       % for j in xrange(i, i + 8):
 220         ## This is 6 because the 0x is counted in the length
 221         % if mapping[j] & 0xffff == 0xffff:
 222           none,
 223         % else:
 224           ${'{:0=#6x}'.format(mapping[j] & 0xffff)},
 225         % endif
 226       % endfor
 227     % endfor
 228     };
 229
 230     void *
 231     anv_lookup_entrypoint(const struct gen_device_info *devinfo, const char *name)
 232     {
 233        static const uint32_t prime_factor = ${prime_factor};
 234        static const uint32_t prime_step = ${prime_step};
 235        const struct anv_entrypoint *e;
 236        uint32_t hash, h, i;
 237        const char *p;
 238
 239        hash = 0;
 240        for (p = name; *p; p++)
 241           hash = hash * prime_factor + *p;
 242
 243        h = hash;
 244        do {
 245           i = map[h & ${hash_mask}];
 246           if (i == none)
 247              return NULL;
 248           e = &entrypoints[i];
 249           h += prime_step;
 250        } while (e->hash != hash);
 251
 252        if (strcmp(name, strings + e->name) != 0)
 253           return NULL;
 254
 255        return anv_resolve_entrypoint(devinfo, i);
 256     }"""), output_encoding='utf-8')
 257
 258 NONE = 0xffff
 259 HASH_SIZE = 256
 260 U32_MASK = 2**32 - 1
 261 HASH_MASK = HASH_SIZE - 1
 262
 263 PRIME_FACTOR = 5024183
 264 PRIME_STEP = 19
 265
 266
 267 def cal_hash(name):
 268     """Calculate the same hash value that Mesa will calculate in C."""
 269     return functools.reduce(
 270         lambda h, c: (h * PRIME_FACTOR + ord(c)) & U32_MASK, name, 0)
 271
 272
 273 def get_entrypoints(doc, entrypoints_to_defines):
 274     """Extract the entry points from the registry."""
 275     entrypoints = []
 276
 277     enabled_commands = set()
 278     for feature in doc.findall('./feature'):
 279         assert feature.attrib['api'] == 'vulkan'
 280         if float(feature.attrib['number']) > MAX_API_VERSION:
 281             continue
 282
 283         for command in feature.findall('./require/command'):
 284             enabled_commands.add(command.attrib['name'])
 285
 286     for extension in doc.findall('.extensions/extension'):
 287         if extension.attrib['name'] not in SUPPORTED_EXTENSIONS:
 288             continue
 289
 290         assert extension.attrib['supported'] == 'vulkan'
 291         for command in extension.findall('./require/command'):
 292             enabled_commands.add(command.attrib['name'])
 293
 294     index = 0
 295     for command in doc.findall('./commands/command'):
 296         type = command.find('./proto/type').text
 297         fullname = command.find('./proto/name').text
 298
 299         if fullname not in enabled_commands:
 300             continue
 301
 302         shortname = fullname[2:]
 303         params = (''.join(p.itertext()) for p in command.findall('./param'))
 304         params = ', '.join(params)
 305         guard = entrypoints_to_defines.get(fullname)
 306         entrypoints.append((type, shortname, params, index, cal_hash(fullname), guard))
 307         index += 1
 308
 309     return entrypoints
 310
 311
 312 def get_entrypoints_defines(doc):
 313     """Maps entry points to extension defines."""
 314     entrypoints_to_defines = {}
 315
 316     for extension in doc.findall('./extensions/extension[@protect]'):
 317         define = extension.attrib['protect']
 318
 319         for entrypoint in extension.findall('./require/command'):
 320             fullname = entrypoint.attrib['name']
 321             entrypoints_to_defines[fullname] = define
 322
 323     return entrypoints_to_defines
 324
 325
 326 def gen_code(entrypoints):
 327     """Generate the C code."""
 328     i = 0
 329     offsets = []
 330     for _, name, _, _, _, _ in entrypoints:
 331         offsets.append(i)
 332         i += 2 + len(name) + 1
 333
 334     mapping = [NONE] * HASH_SIZE
 335     collisions = [0] * 10
 336     for _, name, _, num, h, _ in entrypoints:
 337         level = 0
 338         while mapping[h & HASH_MASK] != NONE:
 339             h = h + PRIME_STEP
 340             level = level + 1
 341         if level > 9:
 342             collisions[9] += 1
 343         else:
 344             collisions[level] += 1
 345         mapping[h & HASH_MASK] = num
 346
 347     return TEMPLATE_C.render(entrypoints=entrypoints,
 348                              offsets=offsets,
 349                              collisions=collisions,
 350                              mapping=mapping,
 351                              hash_mask=HASH_MASK,
 352                              prime_step=PRIME_STEP,
 353                              prime_factor=PRIME_FACTOR,
 354                              none=NONE,
 355                              hash_size=HASH_SIZE,
 356                              filename=os.path.basename(__file__))
 357
 358
 359 def main():
 360     parser = argparse.ArgumentParser()
 361     parser.add_argument('--outdir', help='Where to write the files.',
 362                         required=True)
 363     parser.add_argument('--xml', help='Vulkan API XML file.', required=True)
 364     args = parser.parse_args()
 365
 366     doc = et.parse(args.xml)
 367     entrypoints = get_entrypoints(doc, get_entrypoints_defines(doc))
 368
 369     # Manually add CreateDmaBufImageINTEL for which we don't have an extension
 370     # defined.
 371     entrypoints.append(('VkResult', 'CreateDmaBufImageINTEL',
 372                         'VkDevice device, ' +
 373                         'const VkDmaBufImageCreateInfo* pCreateInfo, ' +
 374                         'const VkAllocationCallbacks* pAllocator,' +
 375                         'VkDeviceMemory* pMem,' +
 376                         'VkImage* pImage', len(entrypoints),
 377                         cal_hash('vkCreateDmaBufImageINTEL'), None))
 378
 379     # For outputting entrypoints.h we generate a anv_EntryPoint() prototype
 380     # per entry point.
 381     with open(os.path.join(args.outdir, 'anv_entrypoints.h'), 'wb') as f:
 382         f.write(TEMPLATE_H.render(entrypoints=entrypoints,
 383                                   filename=os.path.basename(__file__)))
 384     with open(os.path.join(args.outdir, 'anv_entrypoints.c'), 'wb') as f:
 385         f.write(gen_code(entrypoints))
 386
 387
 388 if __name__ == '__main__':
 389     main()