mirror of
https://github.com/fadden/6502bench.git
synced 2024-12-01 07:50:37 +00:00
4d06bb24eb
Removed unnecessary parenthesis from Common-style expressions, which are used by 64tass and ACME.
1077 lines
50 KiB
C#
1077 lines
50 KiB
C#
/*
|
|
* Copyright 2019 faddenSoft
|
|
*
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
*/
|
|
using System;
|
|
using System.Collections.Generic;
|
|
using System.Diagnostics;
|
|
using System.Reflection;
|
|
using System.Text;
|
|
using System.Web.Script.Serialization;
|
|
|
|
using Asm65;
|
|
using CommonUtil;
|
|
|
|
namespace SourceGen {
|
|
/// <summary>
|
|
/// Data pseudo-op formatter. Long operands, notably strings and dense hex blocks, may
|
|
/// be broken across multiple lines.
|
|
///
|
|
/// Assembler output will use Opcode and Operand, emitting multiple lines of ASC, HEX,
|
|
/// etc. The display list may treat it as a single item that is split across
|
|
/// multiple lines.
|
|
/// </summary>
|
|
public class PseudoOp {
|
|
private const int MAX_OPERAND_LEN = 64;
|
|
|
|
/// <summary>
|
|
/// One piece of the pseudo-instruction.
|
|
/// </summary>
|
|
public struct PseudoOut {
|
|
/// <summary>
|
|
/// Opcode. Same for all entries in the list.
|
|
/// </summary>
|
|
public string Opcode { get; set; }
|
|
|
|
/// <summary>
|
|
/// Formatted form of this piece of the operand.
|
|
/// </summary>
|
|
public string Operand { get; set; }
|
|
|
|
/// <summary>
|
|
/// Copy constructor.
|
|
/// </summary>
|
|
public PseudoOut(PseudoOut src) {
|
|
Opcode = src.Opcode;
|
|
Operand = src.Operand;
|
|
}
|
|
}
|
|
|
|
#region PseudoOpNames
|
|
|
|
/// <summary>
|
|
/// Pseudo-op name collection. Instances are immutable.
|
|
/// </summary>
|
|
public class PseudoOpNames {
|
|
public string EquDirective { get; private set; }
|
|
public string VarDirective { get; private set; }
|
|
public string OrgDirective { get; private set; }
|
|
public string RegWidthDirective { get; private set; }
|
|
|
|
public string DefineData1 { get; private set; }
|
|
public string DefineData2 { get; private set; }
|
|
public string DefineData3 { get; private set; }
|
|
public string DefineData4 { get; private set; }
|
|
public string DefineBigData2 { get; private set; }
|
|
public string DefineBigData3 { get; private set; }
|
|
public string DefineBigData4 { get; private set; }
|
|
public string Fill { get; private set; }
|
|
public string Dense { get; private set; }
|
|
public string Junk { get; private set; }
|
|
public string Align { get; private set; }
|
|
public string StrGeneric { get; private set; }
|
|
public string StrReverse { get; private set; }
|
|
public string StrLen8 { get; private set; }
|
|
public string StrLen16 { get; private set; }
|
|
public string StrNullTerm { get; private set; }
|
|
public string StrDci { get; private set; }
|
|
|
|
/// <summary>
|
|
/// Constructs an empty PseudoOp, for deserialization.
|
|
/// </summary>
|
|
public PseudoOpNames() : this(new Dictionary<string, string>()) { }
|
|
|
|
/// <summary>
|
|
/// Constructor. Pass in a dictionary with name/value pairs. Unknown names
|
|
/// will be ignored, missing names will be assigned the empty string.
|
|
/// </summary>
|
|
/// <param name="dict">Dictionary of values.</param>
|
|
public PseudoOpNames(Dictionary<string, string> dict) {
|
|
foreach (PropertyInfo prop in GetType().GetProperties()) {
|
|
dict.TryGetValue(prop.Name, out string value);
|
|
if (value == null) {
|
|
value = string.Empty;
|
|
}
|
|
prop.SetValue(this, value);
|
|
}
|
|
}
|
|
|
|
public static bool operator ==(PseudoOpNames a, PseudoOpNames b) {
|
|
if (ReferenceEquals(a, b)) {
|
|
return true; // same object, or both null
|
|
}
|
|
if (ReferenceEquals(a, null) || ReferenceEquals(b, null)) {
|
|
return false; // one is null
|
|
}
|
|
return a.EquDirective == b.EquDirective &&
|
|
a.VarDirective == b.VarDirective &&
|
|
a.OrgDirective == b.OrgDirective &&
|
|
a.RegWidthDirective == b.RegWidthDirective &&
|
|
a.DefineData1 == b.DefineData1 &&
|
|
a.DefineData2 == b.DefineData2 &&
|
|
a.DefineData3 == b.DefineData3 &&
|
|
a.DefineData4 == b.DefineData4 &&
|
|
a.DefineBigData2 == b.DefineBigData2 &&
|
|
a.DefineBigData3 == b.DefineBigData3 &&
|
|
a.DefineBigData4 == b.DefineBigData4 &&
|
|
a.Fill == b.Fill &&
|
|
a.Dense == b.Dense &&
|
|
a.Junk == b.Junk &&
|
|
a.Align == b.Align &&
|
|
a.StrGeneric == b.StrGeneric &&
|
|
a.StrReverse == b.StrReverse &&
|
|
a.StrLen8 == b.StrLen8 &&
|
|
a.StrLen16 == b.StrLen16 &&
|
|
a.StrNullTerm == b.StrNullTerm &&
|
|
a.StrDci == b.StrDci;
|
|
}
|
|
public static bool operator !=(PseudoOpNames a, PseudoOpNames b) {
|
|
return !(a == b);
|
|
}
|
|
public override bool Equals(object obj) {
|
|
return obj is PseudoOpNames && this == (PseudoOpNames)obj;
|
|
}
|
|
public override int GetHashCode() {
|
|
// should be enough
|
|
return (EquDirective == null ? 0 : EquDirective.GetHashCode()) ^
|
|
(OrgDirective == null ? 0 : OrgDirective.GetHashCode()) ^
|
|
(DefineData1 == null ? 0 : DefineData1.GetHashCode()) ^
|
|
(Fill == null ? 0 : Fill.GetHashCode());
|
|
}
|
|
|
|
public string GetDefineData(int width) {
|
|
switch (width) {
|
|
case 1: return DefineData1;
|
|
case 2: return DefineData2;
|
|
case 3: return DefineData3;
|
|
case 4: return DefineData4;
|
|
default: Debug.Assert(false); return ".?!!";
|
|
}
|
|
}
|
|
public string GetDefineBigData(int width) {
|
|
switch (width) {
|
|
case 1: return DefineData1;
|
|
case 2: return DefineBigData2;
|
|
case 3: return DefineBigData3;
|
|
case 4: return DefineBigData4;
|
|
default: Debug.Assert(false); return ".!!?";
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Merges the non-null, non-empty strings.
|
|
/// </summary>
|
|
public static PseudoOpNames Merge(PseudoOpNames basePon, PseudoOpNames newPon) {
|
|
Dictionary<string, string> baseDict = PropsToDict(basePon);
|
|
Dictionary<string, string> newDict = PropsToDict(newPon);
|
|
|
|
foreach (KeyValuePair<string, string> kvp in newDict) {
|
|
if (string.IsNullOrEmpty(kvp.Value)) {
|
|
continue;
|
|
}
|
|
baseDict[kvp.Key] = kvp.Value;
|
|
}
|
|
|
|
return new PseudoOpNames(baseDict);
|
|
}
|
|
|
|
private static Dictionary<string, string> PropsToDict(PseudoOpNames pon) {
|
|
Dictionary<string, string> dict = new Dictionary<string, string>();
|
|
foreach (PropertyInfo prop in pon.GetType().GetProperties()) {
|
|
string value = (string)prop.GetValue(pon);
|
|
if (!string.IsNullOrEmpty(value)) {
|
|
dict[prop.Name] = value;
|
|
}
|
|
}
|
|
return dict;
|
|
}
|
|
|
|
public string Serialize() {
|
|
// This results in a JSON-encoded string being stored in a JSON-encoded file,
|
|
// which means a lot of double-quote escaping. We could do something here
|
|
// that stored more nicely but it doesn't seem worth the effort.
|
|
JavaScriptSerializer ser = new JavaScriptSerializer();
|
|
Dictionary<string, string> dict = PropsToDict(this);
|
|
return ser.Serialize(dict);
|
|
}
|
|
|
|
public static PseudoOpNames Deserialize(string cereal) {
|
|
JavaScriptSerializer ser = new JavaScriptSerializer();
|
|
try {
|
|
Dictionary<string, string> dict =
|
|
ser.Deserialize<Dictionary<string, string>>(cereal);
|
|
return new PseudoOpNames(dict);
|
|
} catch (Exception ex) {
|
|
Debug.WriteLine("PseudoOpNames deserialization failed: " + ex.Message);
|
|
return new PseudoOpNames();
|
|
}
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Returns a PseudoOpNames instance with some reasonable defaults for on-screen display.
|
|
/// </summary>
|
|
public static PseudoOpNames DefaultPseudoOpNames { get; } =
|
|
new PseudoOpNames(new Dictionary<string, string> {
|
|
{ "EquDirective", ".eq" },
|
|
{ "VarDirective", ".var" },
|
|
{ "OrgDirective", ".org" },
|
|
{ "RegWidthDirective", ".rwid" },
|
|
|
|
{ "DefineData1", ".dd1" },
|
|
{ "DefineData2", ".dd2" },
|
|
{ "DefineData3", ".dd3" },
|
|
{ "DefineData4", ".dd4" },
|
|
{ "DefineBigData2", ".dbd2" },
|
|
{ "DefineBigData3", ".dbd3" },
|
|
{ "DefineBigData4", ".dbd4" },
|
|
{ "Fill", ".fill" },
|
|
{ "Dense", ".bulk" },
|
|
{ "Junk", ".junk" },
|
|
{ "Align", ".align" },
|
|
|
|
{ "StrGeneric", ".str" },
|
|
{ "StrReverse", ".rstr" },
|
|
{ "StrLen8", ".l1str" },
|
|
{ "StrLen16", ".l2str" },
|
|
{ "StrNullTerm", ".zstr" },
|
|
{ "StrDci", ".dstr" }
|
|
});
|
|
|
|
#endregion PseudoOpNames
|
|
|
|
/// <summary>
|
|
/// Computes the number of lines of output required to hold the formatted output.
|
|
/// </summary>
|
|
/// <param name="formatter">Format definition.</param>
|
|
/// <param name="dfd">Data format descriptor.</param>
|
|
/// <returns>Line count.</returns>
|
|
public static int ComputeRequiredLineCount(Formatter formatter, PseudoOpNames opNames,
|
|
FormatDescriptor dfd, byte[] data, int offset) {
|
|
if (dfd.IsString) {
|
|
Debug.Assert(false); // shouldn't be calling here anymore
|
|
List<string> lines = FormatStringOp(formatter, opNames, dfd, data,
|
|
offset, out string popcode);
|
|
return lines.Count;
|
|
}
|
|
|
|
switch (dfd.FormatType) {
|
|
case FormatDescriptor.Type.Default:
|
|
case FormatDescriptor.Type.NumericLE:
|
|
case FormatDescriptor.Type.NumericBE:
|
|
case FormatDescriptor.Type.Fill:
|
|
case FormatDescriptor.Type.Junk:
|
|
return 1;
|
|
case FormatDescriptor.Type.Dense: {
|
|
// no delimiter, two output bytes per input byte
|
|
int maxLen = MAX_OPERAND_LEN;
|
|
int textLen = dfd.Length * formatter.CharsPerDenseByte;
|
|
return (textLen + maxLen - 1) / maxLen;
|
|
}
|
|
default:
|
|
Debug.Assert(false);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Generates a pseudo-op statement for the specified data operation.
|
|
///
|
|
/// For most operations, only one output line will be generated. For larger items,
|
|
/// like dense hex, the value may be split into multiple lines. The sub-index
|
|
/// indicates which line should be formatted.
|
|
/// </summary>
|
|
/// <param name="formatter">Format definition.</param>
|
|
/// <param name="opNames">Table of pseudo-op names.</param>
|
|
/// <param name="symbolTable">Project symbol table.</param>
|
|
/// <param name="labelMap">Symbol label map. May be null.</param>
|
|
/// <param name="dfd">Data format descriptor.</param>
|
|
/// <param name="data">File data array.</param>
|
|
/// <param name="offset">Start offset.</param>
|
|
/// <param name="subIndex">For multi-line items, which line.</param>
|
|
public static PseudoOut FormatDataOp(Formatter formatter, PseudoOpNames opNames,
|
|
SymbolTable symbolTable, Dictionary<string, string> labelMap,
|
|
FormatDescriptor dfd, byte[] data, int offset, int subIndex) {
|
|
if (dfd == null) {
|
|
// should never happen
|
|
//Debug.Assert(false, "Null dfd at offset+" + offset.ToString("x6"));
|
|
PseudoOut failed = new PseudoOut();
|
|
failed.Opcode = failed.Operand = "!FAILED!+" + offset.ToString("x6");
|
|
return failed;
|
|
}
|
|
|
|
int length = dfd.Length;
|
|
Debug.Assert(length > 0);
|
|
|
|
// All outputs for a given offset show the same offset and length, even for
|
|
// multi-line items.
|
|
PseudoOut po = new PseudoOut();
|
|
|
|
if (dfd.IsString) {
|
|
Debug.Assert(false); // shouldn't be calling here anymore
|
|
List<string> lines = FormatStringOp(formatter, opNames, dfd, data,
|
|
offset, out string popcode);
|
|
po.Opcode = popcode;
|
|
po.Operand = lines[subIndex];
|
|
} else {
|
|
switch (dfd.FormatType) {
|
|
case FormatDescriptor.Type.Default:
|
|
if (length != 1) {
|
|
// This shouldn't happen.
|
|
Debug.Assert(false);
|
|
length = 1;
|
|
}
|
|
po.Opcode = opNames.GetDefineData(length);
|
|
int operand = RawData.GetWord(data, offset, length, false);
|
|
po.Operand = formatter.FormatHexValue(operand, length * 2);
|
|
break;
|
|
case FormatDescriptor.Type.NumericLE:
|
|
po.Opcode = opNames.GetDefineData(length);
|
|
operand = RawData.GetWord(data, offset, length, false);
|
|
po.Operand = FormatNumericOperand(formatter, symbolTable, labelMap,
|
|
dfd, operand, length, FormatNumericOpFlags.None);
|
|
break;
|
|
case FormatDescriptor.Type.NumericBE:
|
|
po.Opcode = opNames.GetDefineBigData(length);
|
|
operand = RawData.GetWord(data, offset, length, true);
|
|
po.Operand = FormatNumericOperand(formatter, symbolTable, labelMap,
|
|
dfd, operand, length, FormatNumericOpFlags.None);
|
|
break;
|
|
case FormatDescriptor.Type.Fill:
|
|
po.Opcode = opNames.Fill;
|
|
po.Operand = length + "," + formatter.FormatHexValue(data[offset], 2);
|
|
break;
|
|
case FormatDescriptor.Type.Junk:
|
|
if (dfd.FormatSubType != FormatDescriptor.SubType.None) {
|
|
po.Opcode = opNames.Align;
|
|
int alignPow = FormatDescriptor.AlignmentToPower(dfd.FormatSubType);
|
|
po.Operand = formatter.FormatHexValue(1 << alignPow, 2) +
|
|
" (" + length.ToString() + " bytes)";
|
|
} else {
|
|
po.Opcode = opNames.Junk;
|
|
po.Operand = length.ToString();
|
|
}
|
|
break;
|
|
case FormatDescriptor.Type.Dense: {
|
|
int maxPerLine = MAX_OPERAND_LEN / formatter.CharsPerDenseByte;
|
|
offset += subIndex * maxPerLine;
|
|
length -= subIndex * maxPerLine;
|
|
if (length > maxPerLine) {
|
|
length = maxPerLine;
|
|
}
|
|
po.Opcode = opNames.Dense;
|
|
po.Operand = formatter.FormatDenseHex(data, offset, length);
|
|
//List<PseudoOut> outList = new List<PseudoOut>();
|
|
//GenerateTextLines(text, "", "", po, outList);
|
|
//po = outList[subIndex];
|
|
}
|
|
break;
|
|
default:
|
|
Debug.Assert(false);
|
|
po.Opcode = ".???";
|
|
po.Operand = "$" + data[offset].ToString("x2");
|
|
break;
|
|
}
|
|
}
|
|
|
|
return po;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Adds an additional annotation to an EQU directive, indicating whether the symbol
|
|
/// is a constant or an address, and (if address) how many bytes it spans.
|
|
/// </summary>
|
|
/// <param name="formatter">Formatter object.</param>
|
|
/// <param name="operand">Formatted operand string.</param>
|
|
/// <param name="defSym">Project/platform/variable symbol.</param>
|
|
/// <returns></returns>
|
|
public static string AnnotateEquDirective(Formatter formatter, string operand,
|
|
DefSymbol defSym) {
|
|
string typeStr;
|
|
if (defSym.IsConstant) {
|
|
if (defSym.SymbolSource == Symbol.Source.Variable) {
|
|
typeStr = Res.Strings.EQU_STACK_RELATIVE;
|
|
} else {
|
|
typeStr = Res.Strings.EQU_CONSTANT;
|
|
}
|
|
} else {
|
|
typeStr = Res.Strings.EQU_ADDRESS;
|
|
}
|
|
|
|
if (!defSym.HasWidth && !defSym.IsConstant) {
|
|
// It's an address without an explicit width, do not annotate.
|
|
return operand;
|
|
}
|
|
|
|
StringBuilder sb = new StringBuilder(operand.Length + typeStr.Length + 16);
|
|
sb.Append(operand);
|
|
|
|
int spacesNeeded = 7 - operand.Length;
|
|
do { // always output at least one space
|
|
sb.Append(' ');
|
|
} while (--spacesNeeded > 0);
|
|
|
|
sb.Append("{");
|
|
sb.Append(typeStr);
|
|
if (defSym.HasWidth) {
|
|
sb.Append('/');
|
|
sb.Append(defSym.DataDescriptor.Length);
|
|
}
|
|
sb.Append("}");
|
|
return sb.ToString();
|
|
}
|
|
|
|
/// <summary>
|
|
/// Converts a collection of bytes that represent a string into an array of formatted
|
|
/// string operands.
|
|
/// </summary>
|
|
/// <param name="formatter">Formatter object.</param>
|
|
/// <param name="opNames">Pseudo-opcode name table.</param>
|
|
/// <param name="dfd">Format descriptor.</param>
|
|
/// <param name="data">File data.</param>
|
|
/// <param name="offset">Offset, within data, of start of string.</param>
|
|
/// <param name="popcode">Pseudo-opcode string.</param>
|
|
/// <returns>Array of operand strings.</returns>
|
|
public static List<string> FormatStringOp(Formatter formatter, PseudoOpNames opNames,
|
|
FormatDescriptor dfd, byte[] data, int offset, out string popcode) {
|
|
|
|
int hiddenLeadingBytes = 0;
|
|
int trailingBytes = 0;
|
|
StringOpFormatter.ReverseMode revMode = StringOpFormatter.ReverseMode.Forward;
|
|
Formatter.DelimiterSet delSet = formatter.Config.mStringDelimiters;
|
|
Formatter.DelimiterDef delDef;
|
|
|
|
CharEncoding.Convert charConv;
|
|
switch (dfd.FormatSubType) {
|
|
case FormatDescriptor.SubType.Ascii:
|
|
if (dfd.FormatType == FormatDescriptor.Type.StringDci) {
|
|
charConv = CharEncoding.ConvertLowAndHighAscii;
|
|
} else {
|
|
charConv = CharEncoding.ConvertAscii;
|
|
}
|
|
delDef = delSet.Get(CharEncoding.Encoding.Ascii);
|
|
break;
|
|
case FormatDescriptor.SubType.HighAscii:
|
|
if (dfd.FormatType == FormatDescriptor.Type.StringDci) {
|
|
charConv = CharEncoding.ConvertLowAndHighAscii;
|
|
} else {
|
|
charConv = CharEncoding.ConvertHighAscii;
|
|
}
|
|
delDef = delSet.Get(CharEncoding.Encoding.HighAscii);
|
|
break;
|
|
case FormatDescriptor.SubType.C64Petscii:
|
|
if (dfd.FormatType == FormatDescriptor.Type.StringDci) {
|
|
charConv = CharEncoding.ConvertLowAndHighC64Petscii;
|
|
} else {
|
|
charConv = CharEncoding.ConvertC64Petscii;
|
|
}
|
|
delDef = delSet.Get(CharEncoding.Encoding.C64Petscii);
|
|
break;
|
|
case FormatDescriptor.SubType.C64Screen:
|
|
if (dfd.FormatType == FormatDescriptor.Type.StringDci) {
|
|
charConv = CharEncoding.ConvertLowAndHighC64ScreenCode;
|
|
} else {
|
|
charConv = CharEncoding.ConvertC64ScreenCode;
|
|
}
|
|
delDef = delSet.Get(CharEncoding.Encoding.C64ScreenCode);
|
|
break;
|
|
default:
|
|
Debug.Assert(false);
|
|
charConv = CharEncoding.ConvertAscii;
|
|
delDef = delSet.Get(CharEncoding.Encoding.Ascii);
|
|
break;
|
|
}
|
|
|
|
if (delDef == null) {
|
|
delDef = Formatter.DOUBLE_QUOTE_DELIM;
|
|
}
|
|
|
|
switch (dfd.FormatType) {
|
|
case FormatDescriptor.Type.StringGeneric:
|
|
// Generic character data.
|
|
popcode = opNames.StrGeneric;
|
|
break;
|
|
case FormatDescriptor.Type.StringReverse:
|
|
// Character data, full width specified by formatter. Show characters
|
|
// in reverse order.
|
|
popcode = opNames.StrReverse;
|
|
revMode = StringOpFormatter.ReverseMode.FullReverse;
|
|
break;
|
|
case FormatDescriptor.Type.StringNullTerm:
|
|
// Character data with a terminating null. Don't show the null byte.
|
|
popcode = opNames.StrNullTerm;
|
|
trailingBytes = 1;
|
|
//if (strLen == 0) {
|
|
// showHexZeroes = 1;
|
|
//}
|
|
break;
|
|
case FormatDescriptor.Type.StringL8:
|
|
// Character data with a leading length byte. Don't show the length.
|
|
hiddenLeadingBytes = 1;
|
|
//if (strLen == 0) {
|
|
// showHexZeroes = 1;
|
|
//}
|
|
popcode = opNames.StrLen8;
|
|
break;
|
|
case FormatDescriptor.Type.StringL16:
|
|
// Character data with a leading length word. Don't show the length.
|
|
Debug.Assert(dfd.Length > 1);
|
|
hiddenLeadingBytes = 2;
|
|
//if (strLen == 0) {
|
|
// showHexZeroes = 2;
|
|
//}
|
|
popcode = opNames.StrLen16;
|
|
break;
|
|
case FormatDescriptor.Type.StringDci:
|
|
// High bit on last byte is flipped.
|
|
popcode = opNames.StrDci;
|
|
break;
|
|
default:
|
|
Debug.Assert(false);
|
|
popcode = ".!!!";
|
|
break;
|
|
}
|
|
|
|
StringOpFormatter stropf = new StringOpFormatter(formatter, delDef,
|
|
StringOpFormatter.RawOutputStyle.CommaSep, MAX_OPERAND_LEN, charConv);
|
|
stropf.FeedBytes(data, offset + hiddenLeadingBytes,
|
|
dfd.Length - hiddenLeadingBytes - trailingBytes, 0, revMode);
|
|
|
|
return stropf.Lines;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Special formatting flags for the FormatNumericOperand() method.
|
|
/// </summary>
|
|
[Flags]
|
|
public enum FormatNumericOpFlags {
|
|
None = 0,
|
|
IsPcRel = 1, // opcode is PC relative, e.g. branch or PER
|
|
IsAbsolutePBR = 1 << 1, // operand implicitly uses 'K' on 65816 (JMP/JSR)
|
|
HasHashPrefix = 1 << 2, // operand has a leading '#', reducing ambiguity
|
|
OmitLabelPrefixSuffix = 1 << 3, // don't show annotation char or non-unique prefix
|
|
}
|
|
|
|
/// <summary>
|
|
/// Converts a FormatDescriptor SubType to a CharEncoding.Encoding value.
|
|
/// </summary>
|
|
/// <param name="subType">FormatDescriptor sub-type.</param>
|
|
/// <returns>The corresponding CharEncoding.Encoding value, or Encoding.Unknown
|
|
/// if the sub-type isn't a character encoding.</returns>
|
|
public static CharEncoding.Encoding SubTypeToEnc(FormatDescriptor.SubType subType) {
|
|
switch (subType) {
|
|
case FormatDescriptor.SubType.Ascii:
|
|
return CharEncoding.Encoding.Ascii;
|
|
case FormatDescriptor.SubType.HighAscii:
|
|
return CharEncoding.Encoding.HighAscii;
|
|
case FormatDescriptor.SubType.C64Petscii:
|
|
return CharEncoding.Encoding.C64Petscii;
|
|
case FormatDescriptor.SubType.C64Screen:
|
|
return CharEncoding.Encoding.C64ScreenCode;
|
|
default:
|
|
return CharEncoding.Encoding.Unknown;
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Format a numeric operand value according to the specified sub-format.
|
|
/// </summary>
|
|
/// <param name="formatter">Text formatter.</param>
|
|
/// <param name="symbolTable">Full table of project symbols.</param>
|
|
/// <param name="labelMap">Symbol label remap, for local label conversion. May be
|
|
/// null.</param>
|
|
/// <param name="dfd">Operand format descriptor.</param>
|
|
/// <param name="operandValue">Operand's value. For most things this comes directly
|
|
/// out of the code, for relative branches it's a 24-bit absolute address.</param>
|
|
/// <param name="operandLen">Length of operand, in bytes. For an instruction, this
|
|
/// does not include the opcode byte. For a relative branch, this will be 2.</param>
|
|
/// <param name="flags">Special handling.</param>
|
|
public static string FormatNumericOperand(Formatter formatter, SymbolTable symbolTable,
|
|
Dictionary<string, string> labelMap, FormatDescriptor dfd, int operandValue,
|
|
int operandLen, FormatNumericOpFlags flags) {
|
|
return FormatNumericOperand(formatter, symbolTable, null, labelMap, dfd, -1,
|
|
operandValue, operandLen, flags);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Format a numeric operand value according to the specified sub-format. This
|
|
/// version takes additional arguments to support local variables.
|
|
/// </summary>
|
|
/// <param name="formatter">Text formatter.</param>
|
|
/// <param name="symbolTable">Full table of project symbols.</param>
|
|
/// <param name="lvLookup">Local variable lookup object. May be null if not
|
|
/// formatting an instruction.</param>
|
|
/// <param name="labelMap">Symbol label remap, for local label conversion. May be
|
|
/// null.</param>
|
|
/// <param name="dfd">Operand format descriptor.</param>
|
|
/// <param name="offset">Offset of start of instruction or data pseudo-op, for
|
|
/// variable name lookup. Okay to pass -1 when not formatting an instruction.</param>
|
|
/// <param name="operandValue">Operand's value. For most things this comes directly
|
|
/// out of the code, for relative branches it's a 24-bit absolute address.</param>
|
|
/// <param name="operandLen">Length of operand, in bytes. For an instruction, this
|
|
/// does not include the opcode byte. For a relative branch, this will be 2.</param>
|
|
/// <param name="flags">Special handling.</param>
|
|
public static string FormatNumericOperand(Formatter formatter, SymbolTable symbolTable,
|
|
LocalVariableLookup lvLookup, Dictionary<string, string> labelMap,
|
|
FormatDescriptor dfd, int offset, int operandValue, int operandLen,
|
|
FormatNumericOpFlags flags) {
|
|
Debug.Assert(operandLen > 0);
|
|
int hexMinLen = operandLen * 2;
|
|
|
|
switch (dfd.FormatSubType) {
|
|
case FormatDescriptor.SubType.None:
|
|
case FormatDescriptor.SubType.Hex:
|
|
case FormatDescriptor.SubType.Address:
|
|
return formatter.FormatHexValue(operandValue, hexMinLen);
|
|
case FormatDescriptor.SubType.Decimal:
|
|
return formatter.FormatDecimalValue(operandValue);
|
|
case FormatDescriptor.SubType.Binary:
|
|
return formatter.FormatBinaryValue(operandValue, hexMinLen * 4);
|
|
case FormatDescriptor.SubType.Ascii:
|
|
case FormatDescriptor.SubType.HighAscii:
|
|
case FormatDescriptor.SubType.C64Petscii:
|
|
case FormatDescriptor.SubType.C64Screen:
|
|
CharEncoding.Encoding enc = SubTypeToEnc(dfd.FormatSubType);
|
|
return formatter.FormatCharacterValue(operandValue, enc);
|
|
case FormatDescriptor.SubType.Symbol:
|
|
if (lvLookup != null && dfd.SymbolRef.IsVariable) {
|
|
Debug.Assert(operandLen == 1); // only doing 8-bit stuff
|
|
DefSymbol defSym = lvLookup.GetSymbol(offset, dfd.SymbolRef);
|
|
if (defSym != null) {
|
|
// For local variables we're doing a trivial add and don't
|
|
// wrap, so the "common" format works for everybody.
|
|
StringBuilder sb = new StringBuilder();
|
|
FormatNumericSymbolCommon(formatter, defSym, null,
|
|
dfd, operandValue, operandLen, flags, sb);
|
|
return sb.ToString();
|
|
} else {
|
|
Debug.WriteLine("Local variable format failed");
|
|
Debug.Assert(false);
|
|
return formatter.FormatHexValue(operandValue, hexMinLen);
|
|
}
|
|
} else if (symbolTable.TryGetNonVariableValue(dfd.SymbolRef.Label,
|
|
out Symbol sym)) {
|
|
StringBuilder sb = new StringBuilder();
|
|
|
|
switch (formatter.ExpressionMode) {
|
|
case Formatter.FormatConfig.ExpressionMode.Common:
|
|
FormatNumericSymbolCommon(formatter, sym, labelMap,
|
|
dfd, operandValue, operandLen, flags, sb);
|
|
break;
|
|
case Formatter.FormatConfig.ExpressionMode.Cc65:
|
|
FormatNumericSymbolCc65(formatter, sym, labelMap,
|
|
dfd, operandValue, operandLen, flags, sb);
|
|
break;
|
|
case Formatter.FormatConfig.ExpressionMode.Merlin:
|
|
FormatNumericSymbolMerlin(formatter, sym, labelMap,
|
|
dfd, operandValue, operandLen, flags, sb);
|
|
break;
|
|
default:
|
|
Debug.Assert(false, "Unknown expression mode " +
|
|
formatter.ExpressionMode);
|
|
return "???";
|
|
}
|
|
|
|
return sb.ToString();
|
|
} else {
|
|
return formatter.FormatHexValue(operandValue, hexMinLen);
|
|
}
|
|
default:
|
|
// should not see REMOVE or ASCII_GENERIC here
|
|
Debug.Assert(false);
|
|
return "???";
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Format the symbol and adjustment using common expression syntax.
|
|
/// </summary>
|
|
private static void FormatNumericSymbolCommon(Formatter formatter, Symbol sym,
|
|
Dictionary<string, string> labelMap, FormatDescriptor dfd,
|
|
int operandValue, int operandLen, FormatNumericOpFlags flags, StringBuilder sb) {
|
|
// We could have some simple code that generated correct output, shifting and
|
|
// masking every time, but that's ugly and annoying. For single-byte ops we can
|
|
// just use the byte-select operators, for wider ops we get only as fancy as we
|
|
// need to be.
|
|
|
|
// Start by remapping the label, if necessary. The remapped label may have a
|
|
// local-variable prefix character.
|
|
string symLabel = sym.Label;
|
|
if (labelMap != null && labelMap.TryGetValue(symLabel, out string newLabel)) {
|
|
symLabel = newLabel;
|
|
}
|
|
if (sym.IsVariable) {
|
|
symLabel = formatter.FormatVariableLabel(symLabel);
|
|
}
|
|
|
|
// Now put the prefix/suffix back on if desired. We don't want to mess with it
|
|
// if it's from the assembler.
|
|
if ((flags & FormatNumericOpFlags.OmitLabelPrefixSuffix) == 0) {
|
|
symLabel = Symbol.ConvertLabelForDisplay(symLabel, sym.LabelAnno,
|
|
true, formatter);
|
|
}
|
|
|
|
if (operandLen == 1) {
|
|
// Use the byte-selection operator to get the right piece. In 64tass the
|
|
// selection operator has a very low precedence, similar to Merlin 32.
|
|
int symbolValue;
|
|
string selOp;
|
|
if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.Bank) {
|
|
symbolValue = (sym.Value >> 16) & 0xff;
|
|
if (formatter.Config.mBankSelectBackQuote) {
|
|
selOp = "`";
|
|
} else {
|
|
selOp = "^";
|
|
}
|
|
} else if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.High) {
|
|
symbolValue = (sym.Value >> 8) & 0xff;
|
|
selOp = ">";
|
|
} else {
|
|
symbolValue = sym.Value & 0xff;
|
|
if (symbolValue == sym.Value) {
|
|
selOp = string.Empty;
|
|
} else {
|
|
selOp = "<";
|
|
}
|
|
}
|
|
|
|
operandValue &= 0xff;
|
|
|
|
if (operandValue != symbolValue &&
|
|
dfd.SymbolRef.ValuePart != WeakSymbolRef.Part.Low) {
|
|
// Adjustment is required to an upper-byte part.
|
|
sb.Append('(');
|
|
sb.Append(selOp);
|
|
sb.Append(symLabel);
|
|
sb.Append(')');
|
|
} else {
|
|
// no adjustment required, or no byte-selection
|
|
sb.Append(selOp);
|
|
sb.Append(symLabel);
|
|
}
|
|
int adjustment = operandValue - symbolValue;
|
|
sb.Append(formatter.FormatAdjustment(adjustment));
|
|
} else if (operandLen <= 4) {
|
|
// Operands and values should be 8/16/24 bit unsigned quantities. 32-bit
|
|
// support is really there so you can have a 24-bit pointer in a 32-bit hole.
|
|
// Might need to adjust this if 32-bit signed quantities become interesting.
|
|
uint mask = 0xffffffff >> ((4 - operandLen) * 8);
|
|
int rightShift, symbolValue;
|
|
if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.Bank) {
|
|
symbolValue = (sym.Value >> 16);
|
|
rightShift = 16;
|
|
} else if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.High) {
|
|
symbolValue = (sym.Value >> 8);
|
|
rightShift = 8;
|
|
} else {
|
|
symbolValue = sym.Value;
|
|
rightShift = 0;
|
|
}
|
|
bool hasShift = (rightShift != 0);
|
|
|
|
if ((flags & FormatNumericOpFlags.IsPcRel) != 0) {
|
|
// PC-relative operands are funny, because an 8- or 16-bit value is always
|
|
// expanded to 24 bits. We output a 16-bit value that the assembler will
|
|
// convert back to 8-bit or 16-bit. In any event, the bank byte is never
|
|
// relevant to our computations.
|
|
operandValue &= 0xffff;
|
|
symbolValue &= 0xffff;
|
|
}
|
|
|
|
if ((flags & FormatNumericOpFlags.IsAbsolutePBR) != 0) {
|
|
if ((operandValue & 0x00ff0000) == (symbolValue & 0x00ff0000)) {
|
|
// JMP or JSR to something within the same bank. We don't need to
|
|
// mask the value.
|
|
symbolValue &= 0xffff;
|
|
} else {
|
|
// This is an absolute JMP/JSR to an out-of-bank location, which is
|
|
// bogus and should probably be prevented at a higher level. We handle
|
|
// it by altering the mask so an adjustment that covers the difference
|
|
// in bank values is generated.
|
|
mask = 0x00ffffff;
|
|
}
|
|
}
|
|
|
|
bool needMask = false;
|
|
if (symbolValue > mask) {
|
|
// Post-shift value won't fit in an operand-size box.
|
|
symbolValue = (int) (symbolValue & mask);
|
|
needMask = true;
|
|
}
|
|
|
|
operandValue = (int)(operandValue & mask);
|
|
int adjustment = operandValue - symbolValue;
|
|
|
|
// Possibilities:
|
|
// label
|
|
// label + adj
|
|
// label >> rightShift
|
|
// (label >> rightShift) + adj
|
|
// label & mask
|
|
// (label & mask) + adj
|
|
// (label >> rightShift) & mask
|
|
// ((label >> rightShift) & mask) + adj
|
|
|
|
sb.Append(symLabel);
|
|
|
|
if (rightShift != 0) {
|
|
sb.Append(" >> ");
|
|
sb.Append(rightShift.ToString());
|
|
}
|
|
|
|
if (needMask) {
|
|
if (rightShift != 0) {
|
|
sb.Insert(0, '(');
|
|
sb.Append(')');
|
|
}
|
|
|
|
sb.Append(" & ");
|
|
sb.Append(formatter.FormatHexValue((int)mask, 2));
|
|
}
|
|
|
|
if (adjustment != 0) {
|
|
if (needMask || rightShift != 0) {
|
|
sb.Insert(0, '(');
|
|
sb.Append(')');
|
|
}
|
|
|
|
sb.Append(formatter.FormatAdjustment(adjustment));
|
|
}
|
|
|
|
// Starting with a '(' makes it look like an indirect operand, so we need
|
|
// to prefix the expression with a no-op addition.
|
|
if (sb[0] == '(' && (flags & FormatNumericOpFlags.HasHashPrefix) == 0) {
|
|
sb.Insert(0, "0+");
|
|
}
|
|
} else {
|
|
Debug.Assert(false, "bad numeric len");
|
|
sb.Append("?????");
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Format the symbol and adjustment using cc65 expression syntax.
|
|
/// </summary>
|
|
private static void FormatNumericSymbolCc65(Formatter formatter, Symbol sym,
|
|
Dictionary<string, string> labelMap, FormatDescriptor dfd,
|
|
int operandValue, int operandLen, FormatNumericOpFlags flags, StringBuilder sb) {
|
|
// The key difference between cc65 and other assemblers with general expressions
|
|
// is that the bitwise shift and AND operators have higher precedence than the
|
|
// arithmetic ops like add and subtract. (The bitwise ops are equal to multiply
|
|
// and divide.) This means that, if we want to mask off the low 16 bits and add one
|
|
// to a label, we can write "start & $ffff + 1" rather than "(start & $ffff) + 1".
|
|
//
|
|
// This is particularly convenient for PEA, since "PEA (start & $ffff) + 1" looks like
|
|
// we're trying to use a (non-existent) indirect form of PEA. We can write things
|
|
// in a simpler way.
|
|
|
|
int adjustment, symbolValue;
|
|
|
|
string symLabel = sym.Label;
|
|
if (labelMap != null && labelMap.TryGetValue(symLabel, out string newLabel)) {
|
|
symLabel = newLabel;
|
|
}
|
|
if ((flags & FormatNumericOpFlags.OmitLabelPrefixSuffix) == 0) {
|
|
symLabel = Symbol.ConvertLabelForDisplay(symLabel, sym.LabelAnno,
|
|
true, formatter);
|
|
}
|
|
|
|
if (operandLen == 1) {
|
|
// Use the byte-selection operator to get the right piece.
|
|
string selOp;
|
|
if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.Bank) {
|
|
symbolValue = (sym.Value >> 16) & 0xff;
|
|
selOp = "^";
|
|
} else if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.High) {
|
|
symbolValue = (sym.Value >> 8) & 0xff;
|
|
selOp = ">";
|
|
} else {
|
|
symbolValue = sym.Value & 0xff;
|
|
if (symbolValue == sym.Value) {
|
|
selOp = string.Empty;
|
|
} else {
|
|
selOp = "<";
|
|
}
|
|
}
|
|
sb.Append(selOp);
|
|
sb.Append(symLabel);
|
|
|
|
operandValue &= 0xff;
|
|
} else if (operandLen <= 4) {
|
|
uint mask = 0xffffffff >> ((4 - operandLen) * 8);
|
|
string shOp;
|
|
if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.Bank) {
|
|
symbolValue = (sym.Value >> 16);
|
|
shOp = " >> 16";
|
|
} else if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.High) {
|
|
symbolValue = (sym.Value >> 8);
|
|
shOp = " >> 8";
|
|
} else {
|
|
symbolValue = sym.Value;
|
|
shOp = "";
|
|
}
|
|
|
|
if ((flags & FormatNumericOpFlags.IsPcRel) != 0) {
|
|
// PC-relative operands are funny, because an 8- or 16-bit value is always
|
|
// expanded to 24 bits. We output a 16-bit value that the assembler will
|
|
// convert back to 8-bit or 16-bit. In any event, the bank byte is never
|
|
// relevant to our computations.
|
|
operandValue &= 0xffff;
|
|
symbolValue &= 0xffff;
|
|
}
|
|
|
|
sb.Append(symLabel);
|
|
sb.Append(shOp);
|
|
if (symbolValue > mask) {
|
|
// Post-shift value won't fit in an operand-size box.
|
|
symbolValue = (int)(symbolValue & mask);
|
|
sb.Append(" & ");
|
|
sb.Append(formatter.FormatHexValue((int)mask, 2));
|
|
}
|
|
operandValue = (int)(operandValue & mask);
|
|
|
|
// If we've added stuff, and we're going to add an adjustment later, stick
|
|
// an extra space in between for readability.
|
|
if (sb.Length != symLabel.Length && operandValue != symbolValue) {
|
|
sb.Append(' ');
|
|
}
|
|
} else {
|
|
Debug.Assert(false, "bad numeric len");
|
|
sb.Append("?????");
|
|
symbolValue = 0;
|
|
}
|
|
|
|
adjustment = operandValue - symbolValue;
|
|
|
|
sb.Append(formatter.FormatAdjustment(adjustment));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Format the symbol and adjustment using Merlin expression syntax.
|
|
/// </summary>
|
|
private static void FormatNumericSymbolMerlin(Formatter formatter, Symbol sym,
|
|
Dictionary<string, string> labelMap, FormatDescriptor dfd,
|
|
int operandValue, int operandLen, FormatNumericOpFlags flags, StringBuilder sb) {
|
|
// Merlin expressions are compatible with the original 8-bit Merlin. They're
|
|
// evaluated from left to right, with (almost) no regard for operator precedence.
|
|
//
|
|
// The part-selection operators differ from "simple" in two ways:
|
|
// (1) They always happen last. If FOO=$10f0, "#>FOO+$18" == $11. One of the
|
|
// few cases where left-to-right evaluation is overridden.
|
|
// (2) They select words, not bytes. If FOO=$123456, "#>FOO" is $1234. This is
|
|
// best thought of as a shift operator, rather than byte-selection. For
|
|
// 8-bit code this doesn't matter.
|
|
//
|
|
// This behavior leads to simpler expressions for simple symbol adjustments.
|
|
|
|
string symLabel = sym.Label;
|
|
if (labelMap != null && labelMap.TryGetValue(symLabel, out string newLabel)) {
|
|
symLabel = newLabel;
|
|
}
|
|
if ((flags & FormatNumericOpFlags.OmitLabelPrefixSuffix) == 0) {
|
|
symLabel = Symbol.ConvertLabelForDisplay(symLabel, sym.LabelAnno,
|
|
true, formatter);
|
|
}
|
|
|
|
int adjustment;
|
|
|
|
// If we add or subtract an adjustment, it will be done on the full value, which
|
|
// is then shifted to the appropriate part. So we need to left-shift the operand
|
|
// value to match. We fill in the low bytes with the contents of the symbol, so
|
|
// that the adjustment doesn't include unnecessary values. (For example, let
|
|
// FOO=$10f0, with operand "#>FOO" ($10). We shift the operand to get $1000, then
|
|
// OR in the low byte to get $10f0, so that when we subtract we get adjustment==0.)
|
|
int adjOperand, keepLen;
|
|
if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.Bank) {
|
|
adjOperand = operandValue << 16 | (int)(sym.Value & 0xff00ffff);
|
|
keepLen = 3;
|
|
} else if (dfd.SymbolRef.ValuePart == WeakSymbolRef.Part.High) {
|
|
adjOperand = (operandValue << 8) | (sym.Value & 0xff);
|
|
keepLen = 2;
|
|
} else {
|
|
adjOperand = operandValue;
|
|
keepLen = 1;
|
|
}
|
|
|
|
keepLen = Math.Max(keepLen, operandLen);
|
|
adjustment = adjOperand - sym.Value;
|
|
|
|
if (keepLen == 1) {
|
|
int origAdjust = adjustment;
|
|
adjustment %= 256;
|
|
// Adjust for aesthetics. The assembler implicitly applies a modulo operation,
|
|
// so we can use the value closest to zero.
|
|
if (adjustment > 127) {
|
|
adjustment = -(256 - adjustment) /*% 256*/;
|
|
} else if (adjustment < -128) {
|
|
adjustment = (256 + adjustment) /*% 256*/;
|
|
}
|
|
|
|
// We have a problem with ambiguous direct-page arguments if the adjusted
|
|
// value crosses a bank boundary. For example, "LDA $fff0+24" is computed
|
|
// as $010008, which is too big for a DP arg, so Merlin treats it as absolute
|
|
// (LDA $0008) instead of DP. If Merlin had done the implicit "& $ffff" before
|
|
// testing the value for DP range, this would behave correctly. Unfortunately
|
|
// there is no "force DP" modifier, so we either need to add an explicit mask
|
|
// or just punt and use the original adjustment.
|
|
//
|
|
// Note DP is only relevant for bank zero.
|
|
//
|
|
// TODO(someday): we only need to do this for ambiguous DP. If the instruction
|
|
// is imm or doesn't have an abs equivalent, or it's a fixed-width data item
|
|
// like .DD1, we can still use the nicer-looking adjustment. We don't currently
|
|
// pass the OpDef in here.
|
|
if ((sym.Value & 0xff0000) == 0 && ((sym.Value + adjustment) & 0xff0000) != 0) {
|
|
adjustment = origAdjust;
|
|
}
|
|
} else if (keepLen == 2) {
|
|
adjustment %= 65536;
|
|
if (adjustment > 32767) {
|
|
adjustment = -(65536 - adjustment) /*% 65536*/;
|
|
} else if (adjustment < -32768) {
|
|
adjustment = (65536 + adjustment) /*% 65536*/;
|
|
}
|
|
}
|
|
|
|
// Use the label from sym, not dfd's weak ref; might be different if label
|
|
// comparisons are case-insensitive.
|
|
switch (dfd.SymbolRef.ValuePart) {
|
|
case WeakSymbolRef.Part.Unknown:
|
|
case WeakSymbolRef.Part.Low:
|
|
// For Merlin, "<" is effectively a no-op. We can put it in for
|
|
// aesthetics when grabbing the low byte of a 16-bit value.
|
|
if ((operandLen == 1) && sym.Value > 0xff) {
|
|
sb.Append('<');
|
|
}
|
|
sb.Append(symLabel);
|
|
break;
|
|
case WeakSymbolRef.Part.High:
|
|
sb.Append('>');
|
|
sb.Append(symLabel);
|
|
break;
|
|
case WeakSymbolRef.Part.Bank:
|
|
sb.Append('^');
|
|
sb.Append(symLabel);
|
|
break;
|
|
default:
|
|
Debug.Assert(false, "bad part");
|
|
sb.Append("???");
|
|
break;
|
|
}
|
|
|
|
sb.Append(formatter.FormatAdjustment(adjustment));
|
|
}
|
|
}
|
|
}
|