get_phi_kernel_info.py 8.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
#!/bin/python

# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
# 
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
# 
#     http://www.apache.org/licenses/LICENSE-2.0
# 
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import argparse
import json
import yaml
W
Wilber 已提交
20
from typing import List, Dict, Any
21 22 23


def parse_args():
24
    parser = argparse.ArgumentParser("gather phi kernel and infermate info")
25 26 27 28
    parser.add_argument(
        "--paddle_root_path",
        type=str,
        required=True,
W
Wilber 已提交
29
        help="root path of paddle src[WORK_PATH/Paddle].")
30 31 32 33
    parser.add_argument(
        "--kernel_info_file",
        type=str,
        required=True,
34
        help="kernel info file generated by get_phi_kernel_function.sh.")
35 36 37 38
    parser.add_argument(
        "--infermeta_wrap_file",
        type=str,
        required=True,
W
Wilber 已提交
39 40 41 42 43
        help="inferMeta wrap info file.")
    parser.add_argument(
        "--generate_file",
        type=str,
        required=True,
44
        default="../paddle/infrt/kernel/phi/infershaped/infershaped_kernel_launchers.cc",
W
Wilber 已提交
45
        help="generated file.")
46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61
    args = parser.parse_args()
    return args


def get_api_yaml_info(file_path):
    f = open(file_path + "/python/paddle/utils/code_gen/api.yaml", "r")
    cont = f.read()
    return yaml.load(cont, Loader=yaml.FullLoader)


def get_kernel_info(file_path):
    f = open(file_path, "r")
    cont = f.readlines()
    return [l.strip() for l in cont]


62
def merge(infer_meta_data, kernel_data, wrap_data):
63 64
    meta_map = {}
    for api in infer_meta_data:
65
        if "kernel" not in api or "infer_meta" not in api:
66 67
            continue
        meta_map[api["kernel"]["func"]] = api["infer_meta"]["func"]
68 69 70 71
    wrap_map = {}
    for l in wrap_data:
        wrap_map[l.split()[0]] = l.split()[1]

72 73 74
    full_kernel_data = []
    for l in kernel_data:
        key = l.split()[0]
75 76 77 78 79
        if key in meta_map:
            if key in meta_map:
                full_kernel_data.append((l + " " + wrap_map[key]).split())
            else:
                full_kernel_data.append((l + " " + meta_map[key]).split())
80 81 82 83 84 85
        else:
            full_kernel_data.append((l + " unknown").split())

    return full_kernel_data


W
Wilber 已提交
86
def gen_warn_info():
87
    return """// Generated by tools/infrt/gen_phi_kernel_register.py for infrt.
W
Wilber 已提交
88 89 90 91 92 93
// DO NOT edit or include it within paddle.
"""


def gen_include_headers():
    return """
94 95
#include "paddle/infrt/kernel/phi/infershaped/infershaped_kernel_launchers.h"
#include "paddle/infrt/kernel/phi/infershaped/phi_kernel_launcher.h"
96 97 98 99
#include "paddle/phi/backends/all_context.h"
#include "paddle/phi/include/kernels.h"
#include "paddle/phi/include/infermeta.h"
#include "paddle/phi/infermeta/generated.h"
W
Wilber 已提交
100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116
"""


def gen_namespace():
    return ("""
namespace infrt {
namespace kernel {

""", """

}  // namespace kernel
}  // namespace infrt
""")


def gen_context(val):
    if val == "CPU":
117
        return "phi::CPUContext"
W
Wilber 已提交
118
    # elif val == "GPU":
119
    #     return "phi::GPUContext"
W
Wilber 已提交
120
    # elif val == "XPU":
121
    #     return "phi::XPUContext"
W
Wilber 已提交
122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
    else:
        # raise Exception(f"Unknown context type {val}")
        return ""


def gen_layout(val):
    if val == "ALL_LAYOUT":
        return 'any'
    else:
        # TODO(wilber): now only process ALL_LAYOUT
        raise Exception(f"Unknown layout type {val}")


def gen_kernel_func(val, ctx_name, dtype_name):
    if '<' in val and '>' in val:
        st = val.index('<')
        ed = val.index('>')
        func_name = val[:st]
        template_name = val[st + 1:ed]
141 142
        if 'phi::' in template_name:
            return "&phi::" + val
W
Wilber 已提交
143
        else:
144
            return "&phi::" + func_name + "<phi::" + template_name + ">"
W
Wilber 已提交
145
    else:
146
        return "&phi::" + val + "<" + dtype_name + ", " + ctx_name + ">"
W
Wilber 已提交
147 148 149 150 151 152


def gen_dtype(vals: List[str]):
    ir_dtypes, origin_dtypes = [], []
    for val in vals:
        if val == "float":
153
            ir_dtypes.append("float32")
W
Wilber 已提交
154 155
            origin_dtypes.append("float")
        elif val == "double":
156
            ir_dtypes.append("float64")
W
Wilber 已提交
157 158
            origin_dtypes.append("double")
        elif val == "float16":
159
            ir_dtypes.append("float16")
W
Wilber 已提交
160 161 162 163 164
            origin_dtypes.append("paddle::experimental::float16")
        elif val == "bfloat16":
            ir_dtypes.append("bf16")
            origin_dtypes.append("paddle::experimental::bfloat16")
        elif val == "bool":
165
            ir_dtypes.append("bool")
W
Wilber 已提交
166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212
            origin_dtypes.append("bool")
        elif val == "int8_t":
            ir_dtypes.append("int8")
            origin_dtypes.append("int8_t")
        elif val == "uint8_t":
            ir_dtypes.append("uint8")
            origin_dtypes.append("uint8_t")
        elif val == "int16_t":
            ir_dtypes.append("int16")
            origin_dtypes.append("int16_t")
        elif val == "int" or val == "int32_t":
            ir_dtypes.append("int32")
            origin_dtypes.append("int32_t")
        elif val == "int64_t":
            ir_dtypes.append("int64")
            origin_dtypes.append("int64_t")
        elif val == "complex<float>" or val == "complex64":
            ir_dtypes.append("complex64")
            origin_dtypes.append("paddle::experimental::complex64")
        elif val == "complex<double>" or val == "complex128":
            ir_dtypes.append("complex128")
            origin_dtypes.append("paddle::experimental::complex128")
        elif val == "ALL_DTYPE":
            ir_dtypes.append("all")
            origin_dtypes.append("all")
        else:
            if "VA_ARGS" in val:
                continue
            raise Exception(f"Unknown data type {val}")
    return ir_dtypes, origin_dtypes


# TODO(wilber): Now only process CPUContext.
def gen_register_info(resources: List[List[str]]):
    """
    resources: [['add', 'CPU', 'ALL_LAYOUT', 'AddKernel', 'float', 'double', '...'(varaidic types), 'ElementwiseInferMeta'], ...]
    """
    res = "void RegisterInferShapeLaunchers(host_context::KernelRegistry* registry) {"
    for item in resources:
        # The output string is polluted by C++ macros, here the \ is removed
        update_item = [v.strip('\\') for v in item]

        ctx_name = gen_context(update_item[1])
        if (ctx_name == ""):
            continue
        update_item[2] = gen_layout(update_item[2])
        ir_dtypes, origin_dtypes = gen_dtype(update_item[4:-1])
213
        infer_shape_func = "&phi::" + update_item[-1]
W
Wilber 已提交
214 215 216 217 218 219 220 221

        if update_item[-1] == "unknown":
            # TODO(wilber): handle the unknown inferShape func.
            continue

        for ir_dtype, origin_dtype in zip(ir_dtypes, origin_dtypes):
            kernel_func = gen_kernel_func(update_item[3], ctx_name,
                                          origin_dtype)
222 223
            ir_name = 'phi_cpu.' + update_item[0].lower(
            ) + '.' + ir_dtype + '.' + update_item[2].lower()
W
Wilber 已提交
224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242
            res += f"""
  registry->AddKernel("{ir_name}","""

            res += f"""
    std::bind(&KernelLauncherFunc<decltype({kernel_func}),
                                  {kernel_func},
                                  decltype({infer_shape_func}),
                                  {infer_shape_func}>,
              KernelLauncher<decltype({kernel_func}),
                                  {kernel_func},
                                  decltype({infer_shape_func}),
                                  {infer_shape_func}>(),
              std::placeholders::_1));
"""

    res += "\n}"
    return res


243 244
def gen_phi_kernel_register_code(resources: List[List[str]],
                                 src_file_path: str):
W
Wilber 已提交
245 246 247 248 249 250 251 252 253 254
    source_file = open(src_file_path, 'w')
    source_file.write(gen_warn_info())
    source_file.write(gen_include_headers())
    namespace = gen_namespace()
    source_file.write(namespace[0])
    source_file.write(gen_register_info(resources))
    source_file.write(namespace[1])
    source_file.close()


255 256 257 258
if __name__ == "__main__":
    args = parse_args()
    infer_meta_data = get_api_yaml_info(args.paddle_root_path)
    kernel_data = get_kernel_info(args.kernel_info_file)
259 260
    info_meta_wrap_data = get_kernel_info(args.infermeta_wrap_file)
    out = merge(infer_meta_data, kernel_data, info_meta_wrap_data)
261
    gen_phi_kernel_register_code(out, args.generate_file)