microasm.isa revision 4336
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 print "About to parse tag %s" % opTypes[opNum] 142 opType = OpType(opTypes[opNum]) 143 144 if opType.reg: 145 #Figure out what to do with fixed register operands 146 if opType.reg in ("Ax", "Bx", "Cx", "Dx"): 147 code = opRe.sub("{INTREG_R%s}" % opType.reg.upper(), code) 148 elif opType.reg == "Al": 149 # We need a way to specify register width 150 code = opRe.sub("{INTREG_RAX}", code) 151 else: 152 print "Didn't know how to encode fixed register %s!" % opType.reg 153 elif opType.tag == None or opType.size == None: 154 raise Exception, "Problem parsing operand tag: %s" % opType.tag 155 elif opType.tag in ("C", "D", "G", "P", "S", "T", "V"): 156 # Use the "reg" field of the ModRM byte to select the register 157 code = opRe.sub("{(uint8_t)MODRM_REG}", code) 158 elif opType.tag in ("E", "Q", "W"): 159 # This might refer to memory or to a register. We need to 160 # divide it up farther. 161 regCode = opRe.sub("{(uint8_t)MODRM_RM}", code) 162 regTypes = copy.copy(opTypes) 163 regTypes.pop(-1) 164 # This needs to refer to memory, but we'll fill in the details 165 # later. It needs to take into account unaligned memory 166 # addresses. 167 memCode = opRe.sub("0", code) 168 memTypes = copy.copy(opTypes) 169 memTypes.pop(-1) 170 return doSplitDecode(name, Name, specializeInst, "MODRM_MOD", 171 {"3" : (regCode, regTypes)}, (memCode, memTypes)) 172 elif opType.tag in ("I", "J"): 173 # Immediates are already in the instruction, so don't leave in 174 # those parameters 175 code = opRe.sub("", code) 176 elif opType.tag == "M": 177 # This needs to refer to memory, but we'll fill in the details 178 # later. It needs to take into account unaligned memory 179 # addresses. 180 code = opRe.sub("0", code) 181 elif opType.tag in ("PR", "R", "VR"): 182 # There should probably be a check here to verify that mod 183 # is equal to 11b 184 code = opRe.sub("{(uint8_t)MODRM_RM}", code) 185 else: 186 raise Exception, "Unrecognized tag %s." % opType.tag 187 opTypes.pop(-1) 188 189 # At this point, we've built up "code" to have all the necessary extra 190 # instructions needed to implement whatever types of operands were 191 # specified. Now we'll assemble it it into a microOp sequence. 192 ops = assembleMicro(code) 193 194 # Build a macroop to contain the sequence of microops we've 195 # constructed. The decode block will be used to fill in our 196 # inner decode structure, and the rest will be concatenated and 197 # passed back. 198 return genInst(name, Name, ops) 199}}; 200 201//////////////////////////////////////////////////////////////////// 202// 203// The microcode assembler 204// 205 206let {{ 207 class MicroOpStatement(object): 208 def __init__(self): 209 self.className = '' 210 self.label = '' 211 self.args = [] 212 213 # This converts a list of python bools into 214 # a comma seperated list of C++ bools. 215 def microFlagsText(self, vals): 216 text = "" 217 for val in vals: 218 if val: 219 text += ", true" 220 else: 221 text += ", false" 222 return text 223 224 def getAllocator(self, *microFlags): 225 args = '' 226 for arg in self.args: 227 if arg.has_key("operandConst"): 228 args += ", %s" % arg["operandConst"] 229 elif arg.has_key("operandCode"): 230 args += ", %s" % arg["operandCode"] 231 elif arg.has_key("operandLabel"): 232 raise Exception, "Found a label while creating allocator string." 233 else: 234 raise Exception, "Unrecognized operand type." 235 return 'new %s(machInst%s%s)' % (self.className, self.microFlagsText(microFlags), args) 236}}; 237 238let {{ 239 def buildLabelDict(ops): 240 labels = {} 241 micropc = 0 242 for op in ops: 243 if op.label: 244 labels[op.label] = count 245 micropc += 1 246 return labels 247}}; 248 249let{{ 250 def assembleMicro(code): 251 # This function takes in a block of microcode assembly and returns 252 # a python list of objects which describe it. 253 254 # Keep this around in case we need it later 255 orig_code = code 256 # A list of the statements we've found thus far 257 statements = [] 258 259 # Regular expressions to pull each piece of the statement out at a 260 # time. Each expression expects the thing it's looking for to be at 261 # the beginning of the line, so the previous component is stripped 262 # before continuing. 263 labelRe = re.compile(r'^[ \t]*(?P<label>[a-zA-Z_]\w*)[ \t]:') 264 lineRe = re.compile(r'^(?P<line>[^\n][^\n]*)$') 265 classRe = re.compile(r'^[ \t]*(?P<className>[a-zA-Z_]\w*)') 266 # This recognizes three different flavors of operands: 267 # 1. Raw decimal numbers composed of digits between 0 and 9 268 # 2. Code beginning with "{" and continuing until the first "}" 269 # ^ This one might need revising 270 # 3. A label, which starts with a capital or small letter, or 271 # underscore, which is optionally followed by a sequence of 272 # capital or small letters, underscores, or digts between 0 and 9 273 opRe = re.compile( \ 274 r'^[ \t]*((?P<operandLabel>[a-zA-Z_]\w*)|(?P<operandConst>[0-9][0-9]*)|(\{(?P<operandCode>[^}]*)\}))') 275 lineMatch = lineRe.search(code) 276 while lineMatch != None: 277 statement = MicroOpStatement() 278 # Get a line and seperate it from the rest of the code 279 line = lineMatch.group("line") 280 orig_line = line 281 # print "Parsing line %s" % line 282 code = lineRe.sub('', code, 1) 283 284 # Find the label, if any 285 labelMatch = labelRe.search(line) 286 if labelMatch != None: 287 statement.label = labelMatch.group("label") 288 # print "Found label %s." % statement.label 289 # Clear the label from the statement 290 line = labelRe.sub('', line, 1) 291 292 # Find the class name which is roughly equivalent to the op name 293 classMatch = classRe.search(line) 294 if classMatch == None: 295 raise Exception, "Couldn't find class name in statement: %s" \ 296 % orig_line 297 else: 298 statement.className = classMatch.group("className") 299 # print "Found class name %s." % statement.className 300 301 # Clear the class name from the statement 302 line = classRe.sub('', line, 1) 303 304 #Find as many arguments as you can 305 statement.args = [] 306 opMatch = opRe.search(line) 307 while opMatch is not None: 308 statement.args.append({}) 309 # args is a list of dicts which collect different 310 # representations of operand values. Different forms might be 311 # needed in different places, for instance to replace a label 312 # with an offset. 313 for opType in ("operandLabel", "operandConst", "operandCode"): 314 if opMatch.group(opType): 315 statement.args[-1][opType] = opMatch.group(opType) 316 if len(statement.args[-1]) == 0: 317 print "Problem parsing operand in statement: %s" \ 318 % orig_line 319 line = opRe.sub('', line, 1) 320 # print "Found operand %s." % statement.args[-1] 321 opMatch = opRe.search(line) 322 # print "Found operands", statement.args 323 324 # Add this statement to our collection 325 statements.append(statement) 326 327 # Get the next line 328 lineMatch = lineRe.search(code) 329 330 # Decode the labels into displacements 331 labels = buildLabelDict(statements) 332 micropc = 0 333 for statement in statements: 334 for arg in statement.args: 335 if arg.has_key("operandLabel"): 336 if not labels.has_key(arg["operandLabel"]): 337 raise Exception, "Unrecognized label: %s." % arg["operandLabel"] 338 # This is assuming that intra microcode branches go to 339 # the next micropc + displacement, or 340 # micropc + 1 + displacement. 341 arg["operandConst"] = labels[arg["operandLabel"]] - micropc - 1 342 micropc += 1 343 return statements 344}}; 345