ca65 Macros for Self Modifying Code <author>Christian Krüger <abstract> The 'smc.inc' macro package for ca65 eases the use, increases the safeness and self-explanation of 'self-modifying-code' (SMC). </abstract> <!-- Table of contents --> <toc> <!-- Begin the document --> <sect>Overview<p> When reading assembler sources, self modifying code is often hard to identify and applying it needs a lot of discipline. Since the cacheless 6502 is a thankful target of such kind of code, the macro package will not only reduce this complexness, but also document the use. The resulting source is more self-explanatory and so easier to maintain. While for general purposes SMC is not a desired form for implementations, it can be quite useful for a small range of scenarios. Normally SMC will be introduced when optimizing code in respect to: <itemize> <item>speed and/or <item>size. </itemize> Please mind that SMC can only be applied for code in RAM, which means that a general purpose library with SMC excludes ROM targets! The ca65 SMC macro package consists of two files: <itemize> <item><tt>smc.inc</tt> <item><tt>opcodes.inc</tt> </itemize> The latter is only needed if you also plan to modify opcodes and not only data within your code. <sect>Usage<p> The use of the macros is quite simple: Original: <tscreen><verb> PHA JSR SUBROUTINE PLA </verb></tscreen> By applying SMC, the speed will now be increased by once cycle: SMC: <tscreen><verb> SMC_StoreValue RestoreAccu JSR SUBROUTINE SMC RestoreAccu, { LDA #SMC_Value } </verb></tscreen> The first line stores the value of the accu into the '<tt>RestoreAccu</tt>' labeled SMC target. Please note: <enum> <item> for all SMC store or transfer operations, a second argument can be given. This determines the register for the operation: '<tt>SMC_StoreValue Label, y</tt>' will store the value of the Y-register. If the second argument is missing, the accu will be used automatically. <item> The label targets a 'special SMC namespace'. It fits only to destinations which are introduced with the macro '<tt>SMC</tt>'. A normal label '<tt>RestoreAccu</tt>' wouldn't match and could even coexist (even if you should abstain from doing so). <item> The macro '<tt>SMC_StoreValue</tt>' takes care, that the store operation will occur on the value-position of a SMC-instruction. As you will see, other macros influence other instruction part positions. There is no consistency check, if the targeted SMC instruction actually contains a value. Storing a 'value' on an immplied SMC instruction would corrupt the following memory cell! </enum> The second line needs no further explanation, this is just a placeholder for some code in the example. The third line is the code line which is about to be modified. It has to start with the '<tt>SMC</tt>' macro and must be labeled, so that the modification can be designated. Then the unmodified code is given in curly braces. Please note the usage of the value placeholder 'SMC_Value'. Using such a placeholder has two advantages: <enum> <item> The code is better documented. It is clearly visible that the given value is about to be changed. <item> When examining an (initial) disassembly (e.g. in a debugger), these placeholders can be better identified: They are fixed and, you may notice that below, quite eye catching defined. </enum> <sect1>Argument placeholders<p> There are four kinds of placeholders: <descrip> <label id="Address placeholder"> <tag><tt>SMC_AbsAdr</tt></tag> Used to indicate an address. The value is '<tt>$FADE</tt>'. Example: <tt>STA SMC_AbsAdr</tt> <label id="Zero-Page-Address placeholder"> <tag><tt>SMC_ZpAdr</tt></tag> Used to indicate a zero-page-address. The value is '<tt>$00</tt>'. Example: <tt>LDA SMC_ZpAdr</tt> <label id="Opcode placeholder"> <tag><tt>SMC_Opcode</tt></tag> Used to indicate an instruction. The value is '<tt>NOP</tt>'. Example: <tt>SMC_Opcode</tt> <label id="Immediate value placeholder"> <tag><tt>SMC_Value</tt></tag> Used to indicate a value. The value is '<tt>$42</tt>'. Example: <tt>LDX #SMC_Value</tt> </descrip> Attention: Often code is modified after the initial use - where using the placeholders does not makes sense. Please mind also, that in very variable expressions (e.g. opcode and argument is about to be changed), placeholders can lead to unidentifyable code for a debugger/disassembler: <tt>SMC Example, { SMC_Opcode SMC_AbsAdr } </tt> Since the opcode is '<tt/NOP/', the value '<tt/$DE/' from '<tt/$FADE/' will interpreted as opcode in a disassembler too. This breaks the correct disassembly, because '<tt/$DE/' is interpreted as '<tt/DEC abx/'. Establishing a valid placeholder instruction may be better: <tt>SMC Example, { sta SMC_AbsAdr } ; Note: Opcode will be modified too!</tt> <sect1>Accessing opcodes<p> Some macros are designed to access the instruction of a code line. To increase readability, please use the opcodes as defined in the '<tt>opcodes.inc</tt>' file. <descrip> <label id="Transfer opcode"> <tag><tt>SMC_TransferOpcode label, opcode (, register)</tt></tag> Loads and store an opcode to given SMC instruction. Example: <tscreen><verb> SMC SumRegister, { LDA #10 } JSR OUTPUT SMC_TransferOpcode SumRegister, OPC_ADC_imm, x </verb></tscreen> The macro above will load the opcode '<tt>ADC #</tt>' into the x - register and stores it at the place of the '<tt>LDA #</tt>'. <label id="Load opcode"> <tag><tt>SMC_LoadOpcode label (, register)</tt></tag> Loads the opcode of a SMC line to the given register. Example: <tscreen><verb> SMC ShiftOrNothing, { LSL } SMC_LoadOpcode ShiftOrNothing, y CPY #OPC_NOP BEQ Exit </verb></tscreen> <label id="Store opcode"> <tag><tt>SMC_StoreOpcode label (, register)</tt></tag> Stores the value of the given register at the opcode place of a SMC line. Example: <tscreen><verb> SetBoldMode: LDA #OPC_INX SMC_StoreOpcode AdaptCharWidth SMC_StoreOpcode AdaptUnderlineWidth RTS ... SMC AdaptCharWidth, { NOP } ... SMC AdaptUnderlineWidth, { NOP } </verb></tscreen> </descrip> <sect1>Accessing arguments<p> These marcos are determined to get, set and change arguments of instructions: <descrip> <label id="Change branch"> <tag><tt>SMC_ChangeBranch label, destination (, register)</tt></tag> Used to modify the destination of a branch instruction. If the address offset exceeds the supported range of 8-bit of the 6502, a error will be thrown. Example: <tscreen><verb> Disable Handler: SMC_ChangeBranch BranchToHandler, Exit RTS ... LDA warning SMC BranchToHandler, { BNE Handler } Exit: RTS </verb></tscreen> <label id="Transfer value"> <tag><tt>SMC_TransferValue label, value (, register)</tt></tag> Changes the value of a SMC line. Example: <tscreen><verb> ClearDefault: SMC_TransferValue LoadDefault, 0 RTS ... SMC LoadDefault, { LDX #25 } </verb></tscreen> <label id="Load value"> <tag><tt>SMC_LoadValue label (, register)</tt></tag> Retrieves the value of a SMC line. Example: <tscreen><verb> ShowDefault: SMC_LoadValue LoadDefault JSR PrintValue RTS ... SMC LoadDefault, { LDX #25 } </verb></tscreen> <label id="Store value"> <tag><tt>SMC_StoreValue label (, register)</tt></tag> Stores the value in the register to given SMC line. Example: <tscreen><verb> InitCounters: LDY #0 SMC_StoreValue GetI, y SMC_StoreValue GetJ, y SMC_StoreValue GetK, y ... SMC GetI, { LDX #SMC_Value } ... SMC GetJ, { LDX #SMC_Value } ... SMC GetK, { LDX #SMC_Value } </verb></tscreen> <label id="Transfer low-byte"> <tag><tt>SMC_TransferLowByte label, value (, register)</tt></tag> Does the same as '<tt>SMC_TransferValue</tt>' but should be used for low-bytes of addresses for better readability. Example: <tscreen><verb> ActivateSecondDataSet: SMC_TransferLowByte LoadData, $40 RTS ... SMC LoadData, { LDA $2000 } </verb></tscreen> <label id="Load low-byte"> <tag><tt>SMC_LoadLowByte label (, register)</tt></tag> Does the same as '<tt>SMC_LoadValue</tt>' but should be used for low-bytes of addresses for better readability. Example: <tscreen><verb> IsSecondDataSetActive: SMC_LoadLowByte LoadData, y CPY #$40 BNE NotActive ... SMC LoadData, { LDA $2000 } </verb></tscreen> <label id="Store low-byte"> <tag><tt>SMC_StoreLowByte label (, register)</tt></tag> Does the same as '<tt>SMC_StoreValue</tt>' but should be used for low-bytes of addresses for better readability. Example: <tscreen><verb> InitStructureBaseAddresses: LDX #0 SMC_StoreLowByte GetPlayerGraphic, x SMC_StoreLowByte GetObjectGraphic, x SMC_StoreLowByte StoreCollisionData, x RTS ... SMC GetPlayerGraphic, { LDX $2000 } ... SMC GetObjectGraphic, { LDA $2100,x } ... SMC StoreCollisionData, { STY $2200 } </verb></tscreen> <label id="Transfer high-byte"> <tag><tt>SMC_TransferHighByte label, value (, register)</tt></tag> Loads and stores the given value via the named register to the high-byte address portion of an SMC-instruction. Example: <tscreen><verb> PlaySFX: SMC GetVolume { LDA $3200,x } STA SoundOut INX BNE PlaySFX ... PlayOtherSound: SMC_TransferHighByte GetVolume, $34 </verb></tscreen> <label id="Load high-byte"> <tag><tt>SMC_LoadHighByte label (, register)</tt></tag> Loads the high-byte part of an SMC-instruction address to the given register. Example: <tscreen><verb> PlaySFX: SMC GetVolume { LDA $3200,x } ... SMC_LoadHighByte GetVolume cmp #$34 beq OtherSoundPlaying ... </verb></tscreen> <label id="Store high-byte"> <tag><tt>SMC_StoreHighByte label (, register)</tt></tag> Stores the high-byte address part of an SMC-instruction from the given register. Example: <tscreen><verb> SetupLevel2: LDX #(>Level2Base) SMC_StoreHighByte GetLevelData, x SMC_StoreHighByte GetScreenData, x SMC_StoreHighByte GetSoundData, x RTS ... SMC GetLevelData, { LDA Level1Base+Data } ... SMC GetScreenData, { LDA Level1Base+Screen, x } ... SMC GetSoundData, { LDA Level1Base+Sound, y } </verb></tscreen> <label id="Transfer single address"> <tag><tt>SMC_TransferAddressSingle label, address (, register)</tt></tag> Transfers the contents of the given address via the given register to the designated SMC instruction. Example: <tscreen><verb> PrintHello: SMC_TransferAddressSingle GetChar, #HelloMsg ... LDX #0 NextChar: SMC GetChar, { LDA SMC_AbsAdr, x } BEQ leave JSR CharOut INX BNE NextChar </verb></tscreen> <label id="Transfer address"> <tag><tt>SMC_TransferAddress label, address</tt></tag> Loads contents of given address to A/X and stores the result to SMC instruction. Allows reuse of register contents by using '<tt>SMC_StoreAddress</tt>' for multiple SMC instruction modifications. Example: <tscreen><verb> SMC_TransferAddress JumpTo, #CloseChannel ... SMC JumpTo, { JMP OpenChannel } </verb></tscreen> <label id="Store address"> <tag><tt>SMC_StoreAddress label</tt></tag> Stores the address value in a/x to a SMC instruction address position. Example: <tscreen><verb> SMC_StoreAddress GetData ... SMC GetData, { LDA SMC_AbsAdr } </verb></tscreen> </descrip> <sect1>Operational macros<p> These marcos are determined to let read/modify/write opcodes work on parts of SMC instructions. <descrip> <label id="Operate on value"> <tag><tt>SMC_OperateOnValue opcode, label</tt></tag> Let given opcode work on the value part of a SMC instruction. Example: <tscreen><verb> SMC_OperateOnValue ASL, LoadMask ; shift mask to left ... SMC LoadMask, { LDA #$20 } </verb></tscreen> <label id="Operate on low-byte"> <tag><tt>SMC_OperateOnLowByte opcode, label</tt></tag> Same as '<tt/SMC_OperateOnValue/' but renamed for better readability when accessing low-bytes of address. Example: <tscreen><verb> SMC_OperateOnLowByte DEC, AccessData ... SMC AccessData, { LDX Data } </verb></tscreen> <label id="Operate on high-byte"> <tag><tt>SMC_OperateOnHighByte opcode, label</tt></tag> Let the given opcode work on the high-byte part on a SMC-instruction. Example: <tscreen><verb> NextPage: SMC_OperateOnHighByte INC, GetPageData ... SMC GetPageData, { LDA SourceData, X } </verb></tscreen> </descrip> <sect1>Scope macros<p> These marcos are determined to export and import SMC labels out of the current file scope. Please handle with care! If you cannot abstain from leaving the file scope, you should at least document the exported SMC lines very well. On import side no checking is available if the SMC line is correct accessed (e.g. invalid access to the value of an implied instruction)! <descrip> <label id="Export SMC line under given name"> <tag><tt>SMC_Export alias, label</tt></tag> SMC label will be exported under given alias. Example: <tscreen><verb> .proc GetValue SMC LoadValue, { LDA #12 } rts .endproc SMC_Export GetValueLoader, GetValue::LoadValue </verb></tscreen> <label id="Import SMC alias"> <tag><tt>SMC_Import alias</tt></tag> SMC line is made accessible under given alias. Example: <tscreen><verb> SMC_Import GetValueLoader ... SMC_TransferValue GetValueLoader, #47 ... </verb></tscreen> </descrip> <sect>A complex example<p> Let's have a look on a quite sophisticated example for the usage of SMC. It not only modifies code, but also the modification of the code is modified - allowing reuse of some instructions. <descrip> <tag/The code is from my 'memset()'implementation:/ <tscreen><verb> 1: ... 2: SMC_StoreAddress StoreAccuFirstSection 3: 4: StoreToFirstSection: 5: SMC StoreAccuFirstSection, { sta SMC_AbsAdr, Y } 6: ... 7: RestoreCodeBranchBaseAdr: 8: SMC FirstIncHighByte, { SMC_OperateOnHighByte inc, StoreAccuFirstSection } ; code will be overwritten to 'beq RestoreCode' (*) 9: ... 10: SMC_TransferOpcode FirstIncHighByte, OPC_BEQ , x ; change code marked above with (*) 11: SMC_TransferValue FirstIncHighByte, #(restoreCode - RestoreCodeBranchBaseAdr-2), x ; set relative address to 'RestoreCode' 12: ... 13: restoreCode: 14: SMC_TransferOpcode FirstIncHighByte, OPC_INC_abs , x ; restore original code... 15: SMC_TransferValue FirstIncHighByte, #(<(StoreToFirstSection+2)), x ; (second byte of inc contained low-byte of address) 16: ... </verb></tscreen> <tag/Some explanation:/ Line 2: The register pair A/X contains an address, which is stored on the address location of a SMC line called 'StoreAccuFirstSection'. According to cc65's calling convention, the low-byte is in accu while the high-byte is in the X-register. Line 5: The (modified) address is accessed. Line 8: We have a line here, which is about to be modified (it begins with SMC), but itself modifies code. Please note: Contrary to the rest of SMC-line modifying macros, the 'OperateOn'-macros just expand their given arguments into a single instruction line. These can be changed of course too. Line 10,11: These lines construct a branch operation for line 8: The X-register will be used to change it from 'inc StoreAccuFirstSection+2' (high-byte operation) to 'beq restoreCode'. Please note: To calculate the relative branch offset, we introduced a second label ('RestoreCodeBranchBaseAdr') for to calculate it. Some could also use the internal name of the SMC label, but you should abstain to do so - it may be changed in the future... Line 14,15: The original code from line 8 is reestablished. </descrip> </article>