diff options
Diffstat (limited to 'src/intel')
| -rw-r--r-- | src/intel/vulkan/anv_entrypoints_gen.py | 340 |
1 files changed, 167 insertions, 173 deletions
diff --git a/src/intel/vulkan/anv_entrypoints_gen.py b/src/intel/vulkan/anv_entrypoints_gen.py index 7dd98f960ce..bf6b20951ff 100644 --- a/src/intel/vulkan/anv_entrypoints_gen.py +++ b/src/intel/vulkan/anv_entrypoints_gen.py @@ -88,93 +88,7 @@ TEMPLATE_H = Template(textwrap.dedent("""\ % endfor """)) -NONE = 0xffff -HASH_SIZE = 256 -U32_MASK = 2**32 - 1 -HASH_MASK = HASH_SIZE - 1 - -PRIME_FACTOR = 5024183 -PRIME_STEP = 19 - -opt_header = False -opt_code = False - - -def hash(name): - h = 0 - for c in name: - h = (h * PRIME_FACTOR + ord(c)) & U32_MASK - - return h - - -def print_guard_start(guard): - if guard is not None: - print "#ifdef {0}".format(guard) - - -def print_guard_end(guard): - if guard is not None: - print "#endif // {0}".format(guard) - - -def get_entrypoints(doc, entrypoints_to_defines): - """Extract the entry points from the registry.""" - entrypoints = [] - - enabled_commands = set() - for feature in doc.findall('./feature'): - assert feature.attrib['api'] == 'vulkan' - if float(feature.attrib['number']) > MAX_API_VERSION: - continue - - for command in feature.findall('./require/command'): - enabled_commands.add(command.attrib['name']) - - for extension in doc.findall('.extensions/extension'): - if extension.attrib['name'] not in SUPPORTED_EXTENSIONS: - continue - - assert extension.attrib['supported'] == 'vulkan' - for command in extension.findall('./require/command'): - enabled_commands.add(command.attrib['name']) - - index = 0 - for command in doc.findall('./commands/command'): - type = command.find('./proto/type').text - fullname = command.find('./proto/name').text - - if fullname not in enabled_commands: - continue - - shortname = fullname[2:] - params = (''.join(p.itertext()) for p in command.findall('./param')) - params = ', '.join(params) - if fullname in entrypoints_to_defines: - guard = entrypoints_to_defines[fullname] - else: - guard = None - entrypoints.append((type, shortname, params, index, hash(fullname), guard)) - index += 1 - - return entrypoints - - -def get_entrypoints_defines(doc): - """Maps entry points to extension defines.""" - entrypoints_to_defines = {} - extensions = doc.findall('./extensions/extension') - for extension in extensions: - define = extension.get('protect') - entrypoints = extension.findall('./require/command') - for entrypoint in entrypoints: - fullname = entrypoint.get('name') - entrypoints_to_defines[fullname] = define - return entrypoints_to_defines - - -def gen_code(entrypoints): - print textwrap.dedent("""\ +TEMPLATE_C = Template(textwrap.dedent(u"""\ /* * Copyright © 2015 Intel Corporation * @@ -198,58 +112,61 @@ def gen_code(entrypoints): * IN THE SOFTWARE. */ - /* This file generated from {}, don't edit directly. */ + /* This file generated from ${filename}, don't edit directly. */ #include "anv_private.h" - struct anv_entrypoint {{ + struct anv_entrypoint { uint32_t name; uint32_t hash; - }}; + }; /* We use a big string constant to avoid lots of reloctions from the entry * point table to lots of little strings. The entries in the entry point table * store the index into this big string. */ - static const char strings[] =""".format(os.path.basename(__file__))) - - offsets = [] - i = 0 - for type, name, args, num, h, guard in entrypoints: - print " \"vk%s\\0\"" % name - offsets.append(i) - i += 2 + len(name) + 1 - print " ;" - - # Now generate the table of all entry points - - print "\nstatic const struct anv_entrypoint entrypoints[] = {" - for type, name, args, num, h, guard in entrypoints: - print " { %5d, 0x%08x }," % (offsets[num], h) - print "};\n" + static const char strings[] = + % for _, name, _, _, _, _ in entrypoints: + "vk${name}\\0" + % endfor + ; - print textwrap.dedent(""" + static const struct anv_entrypoint entrypoints[] = { + % for _, _, _, num, h, _ in entrypoints: + { ${offsets[num]}, ${'{:0=#8x}'.format(h)} }, + % endfor + }; /* Weak aliases for all potential implementations. These will resolve to * NULL if they're not defined, which lets the resolve_entrypoint() function * either pick the correct entry point. */ - """) - - for layer in ["anv", "gen7", "gen75", "gen8", "gen9"]: - for type, name, args, num, h, guard in entrypoints: - print_guard_start(guard) - print "%s %s_%s(%s) __attribute__ ((weak));" % (type, layer, name, args) - print_guard_end(guard) - print "\nconst struct anv_dispatch_table %s_layer = {" % layer - for type, name, args, num, h, guard in entrypoints: - print_guard_start(guard) - print " .%s = %s_%s," % (name, layer, name) - print_guard_end(guard) - print "};\n" - - print textwrap.dedent(""" + + % for layer in ['anv', 'gen7', 'gen75', 'gen8', 'gen9']: + % for type_, name, args, _, _, guard in entrypoints: + % if guard is not None: + #ifdef ${guard} + % endif + ${type_} ${layer}_${name}(${args}) __attribute__ ((weak)); + % if guard is not None: + #endif // ${guard} + % endif + % endfor + + const struct anv_dispatch_table ${layer}_layer = { + % for _, name, args, _, _, guard in entrypoints: + % if guard is not None: + #ifdef ${guard} + % endif + .${name} = ${layer}_${name}, + % if guard is not None: + #endif // ${guard} + % endif + % endfor + }; + % endfor + static void * __attribute__ ((noinline)) anv_resolve_entrypoint(const struct gen_device_info *devinfo, uint32_t index) { @@ -279,60 +196,34 @@ def gen_code(entrypoints): unreachable("unsupported gen\\n"); } } - """) - - # Now generate the hash table used for entry point look up. This is a - # uint16_t table of entry point indices. We use 0xffff to indicate an entry - # in the hash table is empty. - - map = [NONE] * HASH_SIZE - collisions = [0] * 10 - for type, name, args, num, h, guard in entrypoints: - level = 0 - while map[h & HASH_MASK] != NONE: - h = h + PRIME_STEP - level = level + 1 - if level > 9: - collisions[9] += 1 - else: - collisions[level] += 1 - map[h & HASH_MASK] = num - - print "/* Hash table stats:" - print " * size %d entries" % HASH_SIZE - print " * collisions entries" - for i in xrange(10): - if i == 9: - plus = "+" - else: - plus = " " - print " * %2d%s %4d" % (i, plus, collisions[i]) - print " */\n" - - print "#define none 0x%04x\n" % NONE - - print "static const uint16_t map[] = {" - for i in xrange(0, HASH_SIZE, 8): - print " ", - for j in xrange(i, i + 8): - if map[j] & 0xffff == 0xffff: - print " none,", - else: - print "0x%04x," % (map[j] & 0xffff), - print - - print "};" + /* Hash table stats: + * size ${hash_size} entries + * collisions entries: + % for i in xrange(10): + * ${i}${'+' if i == 9 else ''} ${collisions[i]} + % endfor + */ - # Finally we generate the hash table lookup function. The hash function and - # linear probing algorithm matches the hash table generated above. + #define none ${'{:#x}'.format(none)} + static const uint16_t map[] = { + % for i in xrange(0, hash_size, 8): + % for j in xrange(i, i + 8): + ## This is 6 because the 0x is counted in the length + % if mapping[j] & 0xffff == 0xffff: + none, + % else: + ${'{:0=#6x}'.format(mapping[j] & 0xffff)}, + % endif + % endfor + % endfor + }; - print textwrap.dedent(""" void * anv_lookup_entrypoint(const struct gen_device_info *devinfo, const char *name) { - static const uint32_t prime_factor = %d; - static const uint32_t prime_step = %d; + static const uint32_t prime_factor = ${prime_factor}; + static const uint32_t prime_step = ${prime_step}; const struct anv_entrypoint *e; uint32_t hash, h, i; const char *p; @@ -343,7 +234,7 @@ def gen_code(entrypoints): h = hash; do { - i = map[h & %d]; + i = map[h & ${hash_mask}]; if (i == none) return NULL; e = &entrypoints[i]; @@ -354,8 +245,111 @@ def gen_code(entrypoints): return NULL; return anv_resolve_entrypoint(devinfo, i); - } - """) % (PRIME_FACTOR, PRIME_STEP, HASH_MASK) + }"""), output_encoding='utf-8') + +NONE = 0xffff +HASH_SIZE = 256 +U32_MASK = 2**32 - 1 +HASH_MASK = HASH_SIZE - 1 + +PRIME_FACTOR = 5024183 +PRIME_STEP = 19 + + +def hash(name): + h = 0 + for c in name: + h = (h * PRIME_FACTOR + ord(c)) & U32_MASK + + return h + + +def get_entrypoints(doc, entrypoints_to_defines): + """Extract the entry points from the registry.""" + entrypoints = [] + + enabled_commands = set() + for feature in doc.findall('./feature'): + assert feature.attrib['api'] == 'vulkan' + if float(feature.attrib['number']) > MAX_API_VERSION: + continue + + for command in feature.findall('./require/command'): + enabled_commands.add(command.attrib['name']) + + for extension in doc.findall('.extensions/extension'): + if extension.attrib['name'] not in SUPPORTED_EXTENSIONS: + continue + + assert extension.attrib['supported'] == 'vulkan' + for command in extension.findall('./require/command'): + enabled_commands.add(command.attrib['name']) + + index = 0 + for command in doc.findall('./commands/command'): + type = command.find('./proto/type').text + fullname = command.find('./proto/name').text + + if fullname not in enabled_commands: + continue + + shortname = fullname[2:] + params = (''.join(p.itertext()) for p in command.findall('./param')) + params = ', '.join(params) + if fullname in entrypoints_to_defines: + guard = entrypoints_to_defines[fullname] + else: + guard = None + entrypoints.append((type, shortname, params, index, hash(fullname), guard)) + index += 1 + + return entrypoints + + +def get_entrypoints_defines(doc): + """Maps entry points to extension defines.""" + entrypoints_to_defines = {} + extensions = doc.findall('./extensions/extension') + for extension in extensions: + define = extension.get('protect') + entrypoints = extension.findall('./require/command') + for entrypoint in entrypoints: + fullname = entrypoint.get('name') + entrypoints_to_defines[fullname] = define + return entrypoints_to_defines + + +def gen_code(entrypoints): + """Generate the C code.""" + i = 0 + offsets = [] + for _, name, _, _, _, _ in entrypoints: + offsets.append(i) + i += 2 + len(name) + 1 + + mapping = [NONE] * HASH_SIZE + collisions = [0] * 10 + for _, name, _, num, h, _ in entrypoints: + level = 0 + while mapping[h & HASH_MASK] != NONE: + h = h + PRIME_STEP + level = level + 1 + if level > 9: + collisions[9] += 1 + else: + collisions[level] += 1 + mapping[h & HASH_MASK] = num + + print TEMPLATE_C.render(entrypoints=entrypoints, + offsets=offsets, + collisions=collisions, + mapping=mapping, + hash_mask=HASH_MASK, + prime_step=PRIME_STEP, + prime_factor=PRIME_FACTOR, + none=NONE, + hash_size=HASH_SIZE, + filename=os.path.basename(__file__)) def main(): |