|
| 1 | +#!/usr/bin/env python3 |
| 2 | +import re |
| 3 | +import sys |
| 4 | +import argparse |
| 5 | + |
| 6 | +# Increased bucket size to handle overlapping wildcards |
| 7 | +MAX_BUCKET_SIZE = 64 |
| 8 | +TABLE_SIZE = 4096 |
| 9 | + |
| 10 | +# Bits [27:20] and [7:4] |
| 11 | +HASH_BITS_MASK = 0x0FF000F0 |
| 12 | + |
| 13 | + |
| 14 | +class Instruction: |
| 15 | + def __init__(self, name, mnemonic, bitstring, array_index): |
| 16 | + self.name = name |
| 17 | + self.mnemonic = mnemonic |
| 18 | + self.bitstring = bitstring |
| 19 | + self.array_index = array_index |
| 20 | + self.mask = 0 |
| 21 | + self.expected = 0 |
| 22 | + self.parse_bits() |
| 23 | + |
| 24 | + def parse_bits(self): |
| 25 | + if len(self.bitstring) != 32: |
| 26 | + print( |
| 27 | + f"Error: Bitstring length {len(self.bitstring)} invalid for {self.name}" |
| 28 | + ) |
| 29 | + sys.exit(1) |
| 30 | + |
| 31 | + for i, char in enumerate(self.bitstring): |
| 32 | + bit_pos = 31 - i |
| 33 | + if char == "0": |
| 34 | + self.mask |= 1 << bit_pos |
| 35 | + elif char == "1": |
| 36 | + self.mask |= 1 << bit_pos |
| 37 | + self.expected |= 1 << bit_pos |
| 38 | + # Variable bits (c, n, d, m, etc) leave mask as 0 |
| 39 | + |
| 40 | + |
| 41 | +def parse_inc_file(input_path): |
| 42 | + instructions = [] |
| 43 | + regex = re.compile(r'INST\(\s*([A-Za-z0-9_]+),\s*"(.*?)",\s*"(.*?)"\s*\)') |
| 44 | + |
| 45 | + try: |
| 46 | + with open(input_path, "r") as f: |
| 47 | + lines = f.readlines() |
| 48 | + except FileNotFoundError: |
| 49 | + print(f"Error: Could not find input file: {input_path}") |
| 50 | + sys.exit(1) |
| 51 | + |
| 52 | + index_counter = 0 |
| 53 | + for line in lines: |
| 54 | + line = line.strip() |
| 55 | + if not line or line.startswith("//"): |
| 56 | + continue |
| 57 | + |
| 58 | + match = regex.search(line) |
| 59 | + if match: |
| 60 | + inst = Instruction( |
| 61 | + match.group(1), match.group(2), match.group(3), index_counter |
| 62 | + ) |
| 63 | + instructions.append(inst) |
| 64 | + index_counter += 1 |
| 65 | + return instructions |
| 66 | + |
| 67 | + |
| 68 | +def generate_lookup_table(instructions): |
| 69 | + buckets = {i: [] for i in range(TABLE_SIZE)} |
| 70 | + |
| 71 | + # Iterate over every possible hash index to determine which instructions belong in it |
| 72 | + for i in range(TABLE_SIZE): |
| 73 | + # Reconstruct the 32-bit value that would generate this hash index |
| 74 | + # Hash algorithm: (Major << 4) | Minor |
| 75 | + # Major is bits [27:20], Minor is bits [7:4] |
| 76 | + |
| 77 | + major_val = (i >> 4) & 0xFF |
| 78 | + minor_val = i & 0x0F |
| 79 | + |
| 80 | + # Create a "Probe" value with the hash bits set |
| 81 | + probe_val = (major_val << 20) | (minor_val << 4) |
| 82 | + |
| 83 | + for inst in instructions: |
| 84 | + # Check if this instruction matches this hash index. |
| 85 | + # An instruction matches if its FIXED bits (mask) match the Probe bits |
| 86 | + # for the specific positions used by the hash. |
| 87 | + |
| 88 | + relevant_mask = inst.mask & HASH_BITS_MASK |
| 89 | + relevant_expected = inst.expected & HASH_BITS_MASK |
| 90 | + |
| 91 | + if (probe_val & relevant_mask) == relevant_expected: |
| 92 | + buckets[i].append(inst) |
| 93 | + |
| 94 | + if len(buckets[i]) > MAX_BUCKET_SIZE: |
| 95 | + print( |
| 96 | + f"FATAL ERROR: Bucket {i:#05x} overflowed! Size: {len(buckets[i])}" |
| 97 | + ) |
| 98 | + print( |
| 99 | + "This means too many instructions map to the same hash index." |
| 100 | + ) |
| 101 | + sys.exit(1) |
| 102 | + |
| 103 | + return buckets |
| 104 | + |
| 105 | + |
| 106 | +def write_decoder_table_h_file(path): |
| 107 | + print(f"Generating decoder table header file: {path}") |
| 108 | + with open(path, "w") as f: |
| 109 | + f.write("/* GENERATED FILE - DO NOT EDIT */\n") |
| 110 | + f.write("/* This file is generated by scripts/generate_jit_assets.py */\n") |
| 111 | + f.write("#ifndef POUND_JIT_DECODER_ARM32_GENERATED_H\n") |
| 112 | + f.write("#define POUND_JIT_DECODER_ARM32_GENERATED_H\n\n") |
| 113 | + f.write('#include "arm32.h"\n') |
| 114 | + f.write("#include <stddef.h>\n\n") |
| 115 | + f.write(f"#define LOOKUP_TABLE_MAX_BUCKET_SIZE {MAX_BUCKET_SIZE}U\n\n") |
| 116 | + f.write("typedef struct {\n") |
| 117 | + f.write( |
| 118 | + " const pvm_jit_decoder_arm32_instruction_info_t *instructions[LOOKUP_TABLE_MAX_BUCKET_SIZE];\n" |
| 119 | + ) |
| 120 | + f.write(" size_t count;\n") |
| 121 | + f.write("} decode_bucket_t;\n\n") |
| 122 | + f.write( |
| 123 | + f"extern const decode_bucket_t g_decoder_lookup_table[{TABLE_SIZE}];\n\n" |
| 124 | + ) |
| 125 | + f.write("#endif\n") |
| 126 | + |
| 127 | + |
| 128 | +def write_opcodes_header(path, instructions): |
| 129 | + """Generates the arm32_opcodes.h file with a unique enum for each mnemonic.""" |
| 130 | + print(f"Generating opcode header file: {path}") |
| 131 | + seen = set() |
| 132 | + with open(path, "w") as f: |
| 133 | + f.write("/* GENERATED FILE - DO NOT EDIT */\n") |
| 134 | + f.write("/* This file is generated by scripts/generate_jit_assets.py */\n") |
| 135 | + f.write("#ifndef POUND_JIT_DECODER_ARM32_OPCODES_H\n") |
| 136 | + f.write("#define POUND_JIT_DECODER_ARM32_OPCODES_H\n\n") |
| 137 | + f.write("typedef enum {\n") |
| 138 | + for inst in instructions: |
| 139 | + enum_name = f" PVM_A32_OP_{inst.name.upper()},\n" |
| 140 | + if enum_name not in seen: |
| 141 | + f.write(enum_name) |
| 142 | + seen.add(enum_name) |
| 143 | + |
| 144 | + f.write(" PVM_A32_OP_STOP,\n") |
| 145 | + f.write("} pvm_jit_decoder_arm32_opcode_t;\n\n") |
| 146 | + f.write("#endif // POUND_JIT_DECODER_ARM32_OPCODES_H\n") |
| 147 | + |
| 148 | + |
| 149 | +def write_decoder_table_c_file(path, instructions, buckets): |
| 150 | + """Writes the decoder C file, now including the opcode enum.""" |
| 151 | + print(f"Generating decoder table source file: {path}") |
| 152 | + with open(path, "w") as f: |
| 153 | + f.write("/* GENERATED FILE - DO NOT EDIT */\n") |
| 154 | + f.write("/* This file is generated by scripts/generate_jit_assets.py */\n") |
| 155 | + f.write('#include "arm32.h"\n') |
| 156 | + f.write('#include "arm32_table.h"\n\n') |
| 157 | + f.write( |
| 158 | + f"static const pvm_jit_decoder_arm32_instruction_info_t g_instructions[{len(instructions)}] = {{\n" |
| 159 | + ) |
| 160 | + for inst in instructions: |
| 161 | + f.write( |
| 162 | + f' {{ "{inst.mnemonic}", "{inst.bitstring}", PVM_A32_OP_{inst.name.upper()}, {inst.mask:#010x}U, {inst.expected:#010x}U }},\n' |
| 163 | + ) |
| 164 | + f.write("};\n") |
| 165 | + |
| 166 | + f.write(f"const decode_bucket_t g_decoder_lookup_table[{TABLE_SIZE}] = {{\n") |
| 167 | + |
| 168 | + for i in range(TABLE_SIZE): |
| 169 | + if len(buckets[i]) > 0: |
| 170 | + f.write(f" [{i:#05x}] = {{ .instructions = {{ ") |
| 171 | + for inst in buckets[i]: |
| 172 | + f.write(f"&g_instructions[{inst.array_index}], ") |
| 173 | + f.write(f"}}, .count = {len(buckets[i])}U }},\n") |
| 174 | + f.write("};\n") |
| 175 | + |
| 176 | + |
| 177 | +def write_interpreter_handler_table(path, instructions): |
| 178 | + """Generates the dispatch table.""" |
| 179 | + print(f"Generating interpreter handler table: {path}") |
| 180 | + seen = set() |
| 181 | + with open(path, "w") as f: |
| 182 | + f.write("/* GENERATED FILE - DO NOT EDIT */\n") |
| 183 | + f.write("/* This file is generated by scripts/generate_jit_assets.py */\n") |
| 184 | + for inst in instructions: |
| 185 | + enum_name = f"PVM_A32_OP_{inst.name.upper()}" |
| 186 | + if enum_name not in seen: |
| 187 | + f.write(f" [{enum_name}] = &&HANDLER_{enum_name},\n") |
| 188 | + seen.add(enum_name) |
| 189 | + |
| 190 | + |
| 191 | +def write_interpreter_handler_skeletons(path, instructions): |
| 192 | + """Generates a skeleton file for handlers.""" |
| 193 | + |
| 194 | + print(f"Generating new skeleton file: {path}") |
| 195 | + seen = set() |
| 196 | + with open(path, "w") as f: |
| 197 | + f.write("/*\n") |
| 198 | + f.write(" * GENERATED FILE - DO NOT EDIT\n") |
| 199 | + f.write(" * This file is generated by scripts/generate_jit_assets.py \n") |
| 200 | + f.write( |
| 201 | + " * This file contains pre-generated, empty handler blocks for the every instruction.\n" |
| 202 | + ) |
| 203 | + f.write(" */\n\n") |
| 204 | + for inst in instructions: |
| 205 | + enum_name = f"HANDLER(PVM_A32_OP_{inst.name.upper()}): {{\n" |
| 206 | + if enum_name not in seen: |
| 207 | + f.write(enum_name) |
| 208 | + seen.add(enum_name) |
| 209 | + f.write(f" // TODO: Implement handler for {inst.mnemonic}\n") |
| 210 | + f.write(" DISPATCH();\n") |
| 211 | + f.write("}\n\n") |
| 212 | + |
| 213 | + f.write(f"HANDLER(PVM_A32_OP_STOP): {{\n") |
| 214 | + f.write(f" // TODO: Implement handler for PVM_A32_OP_STOP\n") |
| 215 | + f.write(" DISPATCH();\n") |
| 216 | + f.write("}\n\n") |
| 217 | + |
| 218 | + |
| 219 | +def main(): |
| 220 | + parser = argparse.ArgumentParser(description="Generate ARM32 Decoder Tables") |
| 221 | + parser.add_argument("input") |
| 222 | + parser.add_argument("--out-opcodes-h") |
| 223 | + parser.add_argument("--out-decoder-c") |
| 224 | + parser.add_argument("--out-decoder-h") |
| 225 | + parser.add_argument("--out-handler-table-inc") |
| 226 | + parser.add_argument("--out-handler-skeletons-inc") |
| 227 | + |
| 228 | + args = parser.parse_args() |
| 229 | + |
| 230 | + instructions = parse_inc_file(args.input) |
| 231 | + buckets = generate_lookup_table(instructions) |
| 232 | + |
| 233 | + # Generate all necessary files |
| 234 | + write_opcodes_header(args.out_opcodes_h, instructions) |
| 235 | + write_decoder_table_c_file(args.out_decoder_c, instructions, buckets) |
| 236 | + write_decoder_table_h_file(args.out_decoder_h) |
| 237 | + write_interpreter_handler_table(args.out_handler_table_inc, instructions) |
| 238 | + write_interpreter_handler_skeletons(args.out_handler_skeletons_inc, instructions) |
| 239 | + |
| 240 | + |
| 241 | +if __name__ == "__main__": |
| 242 | + main() |
0 commit comments