microasm.isa revision 4338
1// -*- mode:c++ -*- 2 3// Copyright (c) 2007 The Hewlett-Packard Development Company 4// All rights reserved. 5// 6// Redistribution and use of this software in source and binary forms, 7// with or without modification, are permitted provided that the 8// following conditions are met: 9// 10// The software must be used only for Non-Commercial Use which means any 11// use which is NOT directed to receiving any direct monetary 12// compensation for, or commercial advantage from such use. Illustrative 13// examples of non-commercial use are academic research, personal study, 14// teaching, education and corporate research & development. 15// Illustrative examples of commercial use are distributing products for 16// commercial advantage and providing services using the software for 17// commercial advantage. 18// 19// If you wish to use this software or functionality therein that may be 20// covered by patents for commercial use, please contact: 21// Director of Intellectual Property Licensing 22// Office of Strategy and Technology 23// Hewlett-Packard Company 24// 1501 Page Mill Road 25// Palo Alto, California 94304 26// 27// Redistributions of source code must retain the above copyright notice, 28// this list of conditions and the following disclaimer. Redistributions 29// in binary form must reproduce the above copyright notice, this list of 30// conditions and the following disclaimer in the documentation and/or 31// other materials provided with the distribution. Neither the name of 32// the COPYRIGHT HOLDER(s), HEWLETT-PACKARD COMPANY, nor the names of its 33// contributors may be used to endorse or promote products derived from 34// this software without specific prior written permission. No right of 35// sublicense is granted herewith. Derivatives of the software and 36// output created using the software may be prepared, but only for 37// Non-Commercial Uses. Derivatives of the software may be shared with 38// others provided: (i) the others agree to abide by the list of 39// conditions herein which includes the Non-Commercial Use restrictions; 40// and (ii) such Derivatives of the software include the above copyright 41// notice to acknowledge the contribution from this software where 42// applicable, this list of conditions and the disclaimer below. 43// 44// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 45// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 46// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 47// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 48// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 49// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 50// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 51// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 52// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 53// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 54// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 55// 56// Authors: Gabe Black 57 58//////////////////////////////////////////////////////////////////// 59// 60// Code to "specialize" a microcode sequence to use a particular 61// variety of operands 62// 63 64let {{ 65 # This builds either a regular or macro op to implement the sequence of 66 # ops we give it. 67 def genInst(name, Name, ops): 68 # If we can implement this instruction with exactly one microop, just 69 # use that directly. 70 newStmnt = '' 71 if len(ops) == 1: 72 decode_block = "return (X86StaticInst *)(%s);" % \ 73 ops[0].getAllocator() 74 return ('', '', decode_block, '') 75 else: 76 # Build a macroop to contain the sequence of microops we've 77 # been given. 78 return genMacroOp(name, Name, ops) 79}}; 80 81let {{ 82 # This code builds up a decode block which decodes based on switchval. 83 # vals is a dict which matches case values with what should be decoded to. 84 # builder is called on the exploded contents of "vals" values to generate 85 # whatever code should be used. 86 def doSplitDecode(name, Name, builder, switchVal, vals, default = None): 87 header_output = '' 88 decoder_output = '' 89 decode_block = 'switch(%s) {\n' % switchVal 90 exec_output = '' 91 for (val, todo) in vals.items(): 92 (new_header_output, 93 new_decoder_output, 94 new_decode_block, 95 new_exec_output) = builder(name, Name, *todo) 96 header_output += new_header_output 97 decoder_output += new_decoder_output 98 decode_block += '\tcase %s: %s\n' % (val, new_decode_block) 99 exec_output += new_exec_output 100 if default: 101 (new_header_output, 102 new_decoder_output, 103 new_decode_block, 104 new_exec_output) = builder(name, Name, *default) 105 header_output += new_header_output 106 decoder_output += new_decoder_output 107 decode_block += '\tdefault: %s\n' % new_decode_block 108 exec_output += new_exec_output 109 decode_block += '}\n' 110 return (header_output, decoder_output, decode_block, exec_output) 111}}; 112 113let {{ 114 class OpType(object): 115 parser = re.compile(r"(?P<tag>[A-Z][A-Z]*)(?P<size>[a-z][a-z]*)|(r(?P<reg>[A-Za-z0-9][A-Za-z0-9]*))") 116 def __init__(self, opTypeString): 117 match = OpType.parser.search(opTypeString) 118 if match == None: 119 raise Exception, "Problem parsing operand type %s" % opTypeString 120 self.reg = match.group("reg") 121 self.tag = match.group("tag") 122 self.size = match.group("size") 123}}; 124 125let {{ 126 127 # This function specializes the given piece of code to use a particular 128 # set of argument types described by "opTypes". These are "implemented" 129 # in reverse order. 130 def specializeInst(name, Name, code, opTypes): 131 opNum = len(opTypes) - 1 132 while len(opTypes): 133 # print "Building a composite op with tags", opTypes 134 # print "And code", code 135 opNum = len(opTypes) - 1 136 # A regular expression to find the operand placeholders we're 137 # interested in. 138 opRe = re.compile("\\^(?P<operandNum>%d)(?=[^0-9]|$)" % opNum) 139 140 # Parse the operand type strign we're working with 141 opType = OpType(opTypes[opNum]) 142 143 if opType.reg: 144 #Figure out what to do with fixed register operands 145 if opType.reg in ("Ax", "Bx", "Cx", "Dx"): 146 code = opRe.sub("%%{INTREG_R%s}" % opType.reg.upper(), code) 147 elif opType.reg == "Al": 148 # We need a way to specify register width 149 code = opRe.sub("%{INTREG_RAX}", code) 150 else: 151 print "Didn't know how to encode fixed register %s!" % opType.reg 152 elif opType.tag == None or opType.size == None: 153 raise Exception, "Problem parsing operand tag: %s" % opType.tag 154 elif opType.tag in ("C", "D", "G", "P", "S", "T", "V"): 155 # Use the "reg" field of the ModRM byte to select the register 156 code = opRe.sub("%{(uint8_t)MODRM_REG}", code) 157 elif opType.tag in ("E", "Q", "W"): 158 # This might refer to memory or to a register. We need to 159 # divide it up farther. 160 regCode = opRe.sub("%{(uint8_t)MODRM_RM}", code) 161 regTypes = copy.copy(opTypes) 162 regTypes.pop(-1) 163 # This needs to refer to memory, but we'll fill in the details 164 # later. It needs to take into account unaligned memory 165 # addresses. 166 memCode = opRe.sub("%0", code) 167 memTypes = copy.copy(opTypes) 168 memTypes.pop(-1) 169 return doSplitDecode(name, Name, specializeInst, "MODRM_MOD", 170 {"3" : (regCode, regTypes)}, (memCode, memTypes)) 171 elif opType.tag in ("I", "J"): 172 # Immediates are already in the instruction, so don't leave in 173 # those parameters 174 code = opRe.sub("${IMMEDIATE}", code) 175 elif opType.tag == "M": 176 # This needs to refer to memory, but we'll fill in the details 177 # later. It needs to take into account unaligned memory 178 # addresses. 179 code = opRe.sub("%0", code) 180 elif opType.tag in ("PR", "R", "VR"): 181 # There should probably be a check here to verify that mod 182 # is equal to 11b 183 code = opRe.sub("%{(uint8_t)MODRM_RM}", code) 184 else: 185 raise Exception, "Unrecognized tag %s." % opType.tag 186 opTypes.pop(-1) 187 188 # At this point, we've built up "code" to have all the necessary extra 189 # instructions needed to implement whatever types of operands were 190 # specified. Now we'll assemble it it into a microOp sequence. 191 ops = assembleMicro(code) 192 193 # Build a macroop to contain the sequence of microops we've 194 # constructed. The decode block will be used to fill in our 195 # inner decode structure, and the rest will be concatenated and 196 # passed back. 197 return genInst(name, Name, ops) 198}}; 199 200//////////////////////////////////////////////////////////////////// 201// 202// The microcode assembler 203// 204 205let {{ 206 class MicroOpStatement(object): 207 def __init__(self): 208 self.className = '' 209 self.label = '' 210 self.args = [] 211 212 # This converts a list of python bools into 213 # a comma seperated list of C++ bools. 214 def microFlagsText(self, vals): 215 text = "" 216 for val in vals: 217 if val: 218 text += ", true" 219 else: 220 text += ", false" 221 return text 222 223 def getAllocator(self, *microFlags): 224 args = '' 225 signature = "<" 226 emptySig = True 227 for arg in self.args: 228 if not emptySig: 229 signature += ", " 230 emptySig = False 231 if arg.has_key("operandImm"): 232 args += ", %s" % arg["operandImm"] 233 signature += ImmOpType 234 elif arg.has_key("operandReg"): 235 args += ", %s" % arg["operandReg"] 236 signature += RegOpType 237 elif arg.has_key("operandLabel"): 238 raise Exception, "Found a label while creating allocator string." 239 else: 240 raise Exception, "Unrecognized operand type." 241 signature += ">" 242 return 'new %s%s(machInst%s%s)' % (self.className, signature, self.microFlagsText(microFlags), args) 243}}; 244 245let {{ 246 def buildLabelDict(ops): 247 labels = {} 248 micropc = 0 249 for op in ops: 250 if op.label: 251 labels[op.label] = count 252 micropc += 1 253 return labels 254}}; 255 256let{{ 257 def assembleMicro(code): 258 # This function takes in a block of microcode assembly and returns 259 # a python list of objects which describe it. 260 261 # Keep this around in case we need it later 262 orig_code = code 263 # A list of the statements we've found thus far 264 statements = [] 265 266 # Regular expressions to pull each piece of the statement out at a 267 # time. Each expression expects the thing it's looking for to be at 268 # the beginning of the line, so the previous component is stripped 269 # before continuing. 270 labelRe = re.compile(r'^[ \t]*(?P<label>\w\w*)[ \t]:') 271 lineRe = re.compile(r'^(?P<line>[^\n][^\n]*)$') 272 classRe = re.compile(r'^[ \t]*(?P<className>[a-zA-Z_]\w*)') 273 # This recognizes three different flavors of operands: 274 # 1. Raw decimal numbers composed of digits between 0 and 9 275 # 2. Code beginning with "{" and continuing until the first "}" 276 # ^ This one might need revising 277 # 3. A label, which starts with a capital or small letter, or 278 # underscore, which is optionally followed by a sequence of 279 # capital or small letters, underscores, or digts between 0 and 9 280 opRe = re.compile( \ 281 r'^[ \t]*((\@(?P<operandLabel0>\w\w*))|' + 282 r'(\@\{(?P<operandLabel1>[^}]*)\})|' + 283 r'(\%(?P<operandReg0>\w\w*))|' + 284 r'(\%\{(?P<operandReg1>[^}]*)\})|' + 285 r'(\$(?P<operandImm0>\w\w*))|' + 286 r'(\$\{(?P<operandImm1>[^}]*)\}))') 287 lineMatch = lineRe.search(code) 288 while lineMatch != None: 289 statement = MicroOpStatement() 290 # Get a line and seperate it from the rest of the code 291 line = lineMatch.group("line") 292 orig_line = line 293 # print "Parsing line %s" % line 294 code = lineRe.sub('', code, 1) 295 296 # Find the label, if any 297 labelMatch = labelRe.search(line) 298 if labelMatch != None: 299 statement.label = labelMatch.group("label") 300 # print "Found label %s." % statement.label 301 # Clear the label from the statement 302 line = labelRe.sub('', line, 1) 303 304 # Find the class name which is roughly equivalent to the op name 305 classMatch = classRe.search(line) 306 if classMatch == None: 307 raise Exception, "Couldn't find class name in statement: %s" \ 308 % orig_line 309 else: 310 statement.className = classMatch.group("className") 311 # print "Found class name %s." % statement.className 312 313 # Clear the class name from the statement 314 line = classRe.sub('', line, 1) 315 316 #Find as many arguments as you can 317 statement.args = [] 318 opMatch = opRe.search(line) 319 while opMatch is not None: 320 statement.args.append({}) 321 # args is a list of dicts which collect different 322 # representations of operand values. Different forms might be 323 # needed in different places, for instance to replace a label 324 # with an offset. 325 for opType in ("operandLabel0", "operandReg0", "operandImm0", 326 "operandLabel1", "operandReg1", "operandImm1"): 327 if opMatch.group(opType): 328 statement.args[-1][opType[:-1]] = opMatch.group(opType) 329 if len(statement.args[-1]) == 0: 330 print "Problem parsing operand in statement: %s" \ 331 % orig_line 332 line = opRe.sub('', line, 1) 333 # print "Found operand %s." % statement.args[-1] 334 opMatch = opRe.search(line) 335 # print "Found operands", statement.args 336 337 # Add this statement to our collection 338 statements.append(statement) 339 340 # Get the next line 341 lineMatch = lineRe.search(code) 342 343 # Decode the labels into displacements 344 labels = buildLabelDict(statements) 345 micropc = 0 346 for statement in statements: 347 for arg in statement.args: 348 if arg.has_key("operandLabel"): 349 if not labels.has_key(arg["operandLabel"]): 350 raise Exception, "Unrecognized label: %s." % arg["operandLabel"] 351 # This is assuming that intra microcode branches go to 352 # the next micropc + displacement, or 353 # micropc + 1 + displacement. 354 arg["operandImm"] = labels[arg["operandLabel"]] - micropc - 1 355 micropc += 1 356 return statements 357}}; 358