microasm.isa revision 4344
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 code builds up a decode block which decodes based on switchval. 66 # vals is a dict which matches case values with what should be decoded to. 67 # builder is called on the exploded contents of "vals" values to generate 68 # whatever code should be used. 69 def doSplitDecode(name, Name, builder, switchVal, vals, default = None): 70 header_output = '' 71 decoder_output = '' 72 decode_block = 'switch(%s) {\n' % switchVal 73 exec_output = '' 74 for (val, todo) in vals.items(): 75 (new_header_output, 76 new_decoder_output, 77 new_decode_block, 78 new_exec_output) = builder(name, Name, *todo) 79 header_output += new_header_output 80 decoder_output += new_decoder_output 81 decode_block += '\tcase %s: %s\n' % (val, new_decode_block) 82 exec_output += new_exec_output 83 if default: 84 (new_header_output, 85 new_decoder_output, 86 new_decode_block, 87 new_exec_output) = builder(name, Name, *default) 88 header_output += new_header_output 89 decoder_output += new_decoder_output 90 decode_block += '\tdefault: %s\n' % new_decode_block 91 exec_output += new_exec_output 92 decode_block += '}\n' 93 return (header_output, decoder_output, decode_block, exec_output) 94}}; 95 96let {{ 97 class OpType(object): 98 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]*))") 99 def __init__(self, opTypeString): 100 match = OpType.parser.search(opTypeString) 101 if match == None: 102 raise Exception, "Problem parsing operand type %s" % opTypeString 103 self.reg = match.group("reg") 104 self.tag = match.group("tag") 105 self.size = match.group("size") 106}}; 107 108let {{ 109 110 # This function specializes the given piece of code to use a particular 111 # set of argument types described by "opTypes". These are "implemented" 112 # in reverse order. 113 def specializeInst(name, Name, code, opTypes): 114 opNum = len(opTypes) - 1 115 while len(opTypes): 116 # print "Building a composite op with tags", opTypes 117 # print "And code", code 118 opNum = len(opTypes) - 1 119 # A regular expression to find the operand placeholders we're 120 # interested in. 121 opRe = re.compile("\\^(?P<operandNum>%d)(?=[^0-9]|$)" % opNum) 122 123 # Parse the operand type strign we're working with 124 opType = OpType(opTypes[opNum]) 125 126 if opType.reg: 127 #Figure out what to do with fixed register operands 128 if opType.reg in ("Ax", "Bx", "Cx", "Dx"): 129 code = opRe.sub("%%{INTREG_R%s}" % opType.reg.upper(), code) 130 elif opType.reg == "Al": 131 # We need a way to specify register width 132 code = opRe.sub("%{INTREG_RAX}", code) 133 else: 134 print "Didn't know how to encode fixed register %s!" % opType.reg 135 elif opType.tag == None or opType.size == None: 136 raise Exception, "Problem parsing operand tag: %s" % opType.tag 137 elif opType.tag in ("C", "D", "G", "P", "S", "T", "V"): 138 # Use the "reg" field of the ModRM byte to select the register 139 code = opRe.sub("%{(uint8_t)MODRM_REG}", code) 140 elif opType.tag in ("E", "Q", "W"): 141 # This might refer to memory or to a register. We need to 142 # divide it up farther. 143 regCode = opRe.sub("%{(uint8_t)MODRM_RM}", code) 144 regTypes = copy.copy(opTypes) 145 regTypes.pop(-1) 146 # This needs to refer to memory, but we'll fill in the details 147 # later. It needs to take into account unaligned memory 148 # addresses. 149 memCode = opRe.sub("%0", code) 150 memTypes = copy.copy(opTypes) 151 memTypes.pop(-1) 152 return doSplitDecode(name, Name, specializeInst, "MODRM_MOD", 153 {"3" : (regCode, regTypes)}, (memCode, memTypes)) 154 elif opType.tag in ("I", "J"): 155 # Immediates are already in the instruction, so don't leave in 156 # those parameters 157 code = opRe.sub("${IMMEDIATE}", code) 158 elif opType.tag == "M": 159 # This needs to refer to memory, but we'll fill in the details 160 # later. It needs to take into account unaligned memory 161 # addresses. 162 code = opRe.sub("%0", code) 163 elif opType.tag in ("PR", "R", "VR"): 164 # There should probably be a check here to verify that mod 165 # is equal to 11b 166 code = opRe.sub("%{(uint8_t)MODRM_RM}", code) 167 else: 168 raise Exception, "Unrecognized tag %s." % opType.tag 169 opTypes.pop(-1) 170 171 # At this point, we've built up "code" to have all the necessary extra 172 # instructions needed to implement whatever types of operands were 173 # specified. Now we'll assemble it it into a StaticInst. 174 return assembleMicro(name, Name, code) 175}}; 176 177//////////////////////////////////////////////////////////////////// 178// 179// The microcode assembler 180// 181 182let {{ 183 # These are used when setting up microops so that they can specialize their 184 # base class template properly. 185 RegOpType = "RegisterOperand" 186 ImmOpType = "ImmediateOperand" 187}}; 188 189let {{ 190 class MicroOpStatement(object): 191 def __init__(self): 192 self.className = '' 193 self.label = '' 194 self.args = [] 195 196 # This converts a list of python bools into 197 # a comma seperated list of C++ bools. 198 def microFlagsText(self, vals): 199 text = "" 200 for val in vals: 201 if val: 202 text += ", true" 203 else: 204 text += ", false" 205 return text 206 207 def getAllocator(self, *microFlags): 208 args = '' 209 signature = "<" 210 emptySig = True 211 for arg in self.args: 212 if not emptySig: 213 signature += ", " 214 emptySig = False 215 if arg.has_key("operandImm"): 216 args += ", %s" % arg["operandImm"] 217 signature += ImmOpType 218 elif arg.has_key("operandReg"): 219 args += ", %s" % arg["operandReg"] 220 signature += RegOpType 221 elif arg.has_key("operandLabel"): 222 raise Exception, "Found a label while creating allocator string." 223 else: 224 raise Exception, "Unrecognized operand type." 225 signature += ">" 226 return 'new %s%s(machInst%s%s)' % (self.className, signature, self.microFlagsText(microFlags), args) 227}}; 228 229let{{ 230 def assembleMicro(name, Name, code): 231 232 # This function takes in a block of microcode assembly and returns 233 # a python list of objects which describe it. 234 235 # Keep this around in case we need it later 236 orig_code = code 237 # A list of the statements we've found thus far 238 statements = [] 239 240 # Regular expressions to pull each piece of the statement out at a 241 # time. Each expression expects the thing it's looking for to be at 242 # the beginning of the line, so the previous component is stripped 243 # before continuing. 244 labelRe = re.compile(r'^[ \t]*(?P<label>\w\w*)[ \t]:') 245 lineRe = re.compile(r'^(?P<line>[^\n][^\n]*)$') 246 classRe = re.compile(r'^[ \t]*(?P<className>[a-zA-Z_]\w*)') 247 # This recognizes three different flavors of operands: 248 # 1. Raw decimal numbers composed of digits between 0 and 9 249 # 2. Code beginning with "{" and continuing until the first "}" 250 # ^ This one might need revising 251 # 3. A label, which starts with a capital or small letter, or 252 # underscore, which is optionally followed by a sequence of 253 # capital or small letters, underscores, or digts between 0 and 9 254 opRe = re.compile( \ 255 r'^[ \t]*((\@(?P<operandLabel0>\w\w*))|' + 256 r'(\@\{(?P<operandLabel1>[^}]*)\})|' + 257 r'(\%(?P<operandReg0>\w\w*))|' + 258 r'(\%\{(?P<operandReg1>[^}]*)\})|' + 259 r'(\$(?P<operandImm0>\w\w*))|' + 260 r'(\$\{(?P<operandImm1>[^}]*)\}))') 261 lineMatch = lineRe.search(code) 262 while lineMatch != None: 263 statement = MicroOpStatement() 264 # Get a line and seperate it from the rest of the code 265 line = lineMatch.group("line") 266 orig_line = line 267 # print "Parsing line %s" % line 268 code = lineRe.sub('', code, 1) 269 270 # Find the label, if any 271 labelMatch = labelRe.search(line) 272 if labelMatch != None: 273 statement.label = labelMatch.group("label") 274 # print "Found label %s." % statement.label 275 # Clear the label from the statement 276 line = labelRe.sub('', line, 1) 277 278 # Find the class name which is roughly equivalent to the op name 279 classMatch = classRe.search(line) 280 if classMatch == None: 281 raise Exception, "Couldn't find class name in statement: %s" \ 282 % orig_line 283 else: 284 statement.className = classMatch.group("className") 285 # print "Found class name %s." % statement.className 286 287 # Clear the class name from the statement 288 line = classRe.sub('', line, 1) 289 290 #Find as many arguments as you can 291 statement.args = [] 292 opMatch = opRe.search(line) 293 while opMatch is not None: 294 statement.args.append({}) 295 # args is a list of dicts which collect different 296 # representations of operand values. Different forms might be 297 # needed in different places, for instance to replace a label 298 # with an offset. 299 for opType in ("operandLabel0", "operandReg0", "operandImm0", 300 "operandLabel1", "operandReg1", "operandImm1"): 301 if opMatch.group(opType): 302 statement.args[-1][opType[:-1]] = opMatch.group(opType) 303 if len(statement.args[-1]) == 0: 304 print "Problem parsing operand in statement: %s" \ 305 % orig_line 306 line = opRe.sub('', line, 1) 307 # print "Found operand %s." % statement.args[-1] 308 opMatch = opRe.search(line) 309 # print "Found operands", statement.args 310 311 # Add this statement to our collection 312 statements.append(statement) 313 314 # Get the next line 315 lineMatch = lineRe.search(code) 316 317 # Decode the labels into displacements 318 319 labels = {} 320 micropc = 0 321 for statement in statements: 322 if statement.label: 323 labels[statement.label] = count 324 micropc += 1 325 micropc = 0 326 for statement in statements: 327 for arg in statement.args: 328 if arg.has_key("operandLabel"): 329 if not labels.has_key(arg["operandLabel"]): 330 raise Exception, "Unrecognized label: %s." % arg["operandLabel"] 331 # This is assuming that intra microcode branches go to 332 # the next micropc + displacement, or 333 # micropc + 1 + displacement. 334 arg["operandImm"] = labels[arg["operandLabel"]] - micropc - 1 335 micropc += 1 336 337 # If we can implement this instruction with exactly one microop, just 338 # use that directly. 339 if len(statements) == 1: 340 decode_block = "return %s;" % \ 341 statements[0].getAllocator() 342 return ('', '', decode_block, '') 343 else: 344 # Build a macroop to contain the sequence of microops we've 345 # been given. 346 return genMacroOp(name, Name, statements) 347}}; 348