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