Reworking how x86's isa description works. I'm adopting the following definitions to make figuring out what's what a little easier:

MicroOp: A single operation actually implemented in hardware.
MacroOp: A collection of microops which are executed as a unit.
Instruction: An architected instruction which can be implemented with a macroop or a microop.

--HG--
extra : convert_revision : 1cfc8409cc686c75220767839f55a30551aa6f13

1 files changed, 147 insertions, 2 deletions

diff --git a/src/arch/x86/isa/microasm.isa b/src/arch/x86/isa/microasm.isa
index 711ebf667..6d428881e 100644
--- a/src/arch/x86/isa/microasm.isa
+++ b/src/arch/x86/isa/microasm.isa
@@ -57,11 +57,154 @@
 
 ////////////////////////////////////////////////////////////////////
 //
-//  Code to "assemble" microcode sequences
+//  Code to "specialize" a microcode sequence to use a particular
+//  variety of operands
 //
 
 let {{
-    class MicroOpStatement:
+    # This builds either a regular or macro op to implement the sequence of
+    # ops we give it.
+    def genInst(name, Name, ops):
+        # If we can implement this instruction with exactly one microop, just
+        # use that directly.
+        newStmnt = ''
+        if len(ops) == 1:
+            decode_block = "return (X86StaticInst *)(%s);" % \
+                            ops[0].getAllocator()
+            return ('', '', decode_block, '')
+        else:
+            # Build a macroop to contain the sequence of microops we've
+            # been given.
+            return genMacroOp(name, Name, ops)
+}};
+
+let {{
+    # This code builds up a decode block which decodes based on switchval.
+    # vals is a dict which matches case values with what should be decoded to.
+    # builder is called on the exploded contents of "vals" values to generate
+    # whatever code should be used.
+    def doSplitDecode(name, Name, builder, switchVal, vals, default = None):
+        header_output = ''
+        decoder_output = ''
+        decode_block = 'switch(%s) {\n' % switchVal
+        exec_output = ''
+        for (val, todo) in vals.items():
+            (new_header_output,
+             new_decoder_output,
+             new_decode_block,
+             new_exec_output) = builder(name, Name, *todo)
+            header_output += new_header_output
+            decoder_output += new_decoder_output
+            decode_block += '\tcase %s: %s\n' % (val, new_decode_block)
+            exec_output += new_exec_output
+        if default:
+            (new_header_output,
+             new_decoder_output,
+             new_decode_block,
+             new_exec_output) = builder(name, Name, *default)
+            header_output += new_header_output
+            decoder_output += new_decoder_output
+            decode_block += '\tdefault: %s\n' % new_decode_block
+            exec_output += new_exec_output
+        decode_block += '}\n'
+        return (header_output, decoder_output, decode_block, exec_output)
+}};
+
+let {{
+    class OpType(object):
+        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]*))")
+        def __init__(self, opTypeString):
+            match = OpType.parser.search(opTypeString)
+            if match == None:
+                raise Exception, "Problem parsing operand type %s" % opTypeString
+            self.reg = match.group("reg")
+            self.tag = match.group("tag")
+            self.size = match.group("size")
+}};
+
+let {{
+
+    # This function specializes the given piece of code to use a particular
+    # set of argument types described by "opTypes". These are "implemented"
+    # in reverse order.
+    def specializeInst(name, Name, code, opTypes):
+        opNum = len(opTypes) - 1
+        while len(opTypes):
+            # print "Building a composite op with tags", opTypes
+            # print "And code", code
+            opNum = len(opTypes) - 1
+            # A regular expression to find the operand placeholders we're
+            # interested in.
+            opRe = re.compile("%%(?P<operandNum>%d)(?=[^0-9]|$)" % opNum)
+
+            # Parse the operand type strign we're working with
+            print "About to parse tag %s" % opTypes[opNum]
+            opType = OpType(opTypes[opNum])
+
+            if opType.reg:
+                #Figure out what to do with fixed register operands
+                if opType.reg in ("Ax", "Bx", "Cx", "Dx"):
+                    code = opRe.sub("{INTREG_R%s}" % opType.reg.upper(), code)
+                elif opType.reg == "Al":
+                    # We need a way to specify register width
+                    code = opRe.sub("{INTREG_RAX}", code)
+                else:
+                    print "Didn't know how to encode fixed register %s!" % opType.reg
+            elif opType.tag == None or opType.size == None:
+                raise Exception, "Problem parsing operand tag: %s" % opType.tag
+            elif opType.tag in ("C", "D", "G", "P", "S", "T", "V"):
+                # Use the "reg" field of the ModRM byte to select the register
+                code = opRe.sub("{(uint8_t)MODRM_REG}", code)
+            elif opType.tag in ("E", "Q", "W"):
+                # This might refer to memory or to a register. We need to
+                # divide it up farther.
+                regCode = opRe.sub("{(uint8_t)MODRM_RM}", code)
+                regTypes = copy.copy(opTypes)
+                regTypes.pop(-1)
+                # This needs to refer to memory, but we'll fill in the details
+                # later. It needs to take into account unaligned memory
+                # addresses.
+                memCode = opRe.sub("0", code)
+                memTypes = copy.copy(opTypes)
+                memTypes.pop(-1)
+                return doSplitDecode(name, Name, specializeInst, "MODRM_MOD",
+                    {"3" : (regCode, regTypes)}, (memCode, memTypes))
+            elif opType.tag in ("I", "J"):
+                # Immediates are already in the instruction, so don't leave in
+                # those parameters
+                code = opRe.sub("", code)
+            elif opType.tag == "M":
+                # This needs to refer to memory, but we'll fill in the details
+                # later. It needs to take into account unaligned memory
+                # addresses.
+                code = opRe.sub("0", code)
+            elif opType.tag in ("PR", "R", "VR"):
+                # There should probably be a check here to verify that mod
+                # is equal to 11b
+                code = opRe.sub("{(uint8_t)MODRM_RM}", code)
+            else:
+                raise Exception, "Unrecognized tag %s." % opType.tag
+            opTypes.pop(-1)
+
+        # At this point, we've built up "code" to have all the necessary extra
+        # instructions needed to implement whatever types of operands were
+        # specified. Now we'll assemble it it into a microOp sequence.
+        ops = assembleMicro(code)
+
+        # Build a macroop to contain the sequence of microops we've
+        # constructed. The decode block will be used to fill in our
+        # inner decode structure, and the rest will be concatenated and
+        # passed back.
+        return genInst(name, Name, ops)
+}};
+
+////////////////////////////////////////////////////////////////////
+//
+//  The microcode assembler
+//
+
+let {{
+    class MicroOpStatement(object):
         def __init__(self):
             self.className = ''
             self.label = ''
@@ -101,7 +244,9 @@ let {{
                 labels[op.label] = count
             micropc += 1
         return labels
+}};
 
+let{{
     def assembleMicro(code):
         # This function takes in a block of microcode assembly and returns
         # a python list of objects which describe it.