famicom party

5. Introducing 6502 Assembly

The "Hello World" program you created in Chapter 3 is probably unlike any other you have ever seen: 44 lines of code, mostly composed of three-letter acronyms and numbers, just to change the color of the screen. What did you copy and paste?

As you can probably guess from the title of this chapter, helloworld.asm contained assembly code. I prefer to use the .asm extension for assembly files, but the choice is pretty arbitrary. Some developers prefer to use .s for their assembly files. Ultimately, the choice is yours - ca65 will assemble any file extension as long as the contents are assembly code. Assembly is one layer of abstraction above machine code. Writing assembly requires detailed knowledge of a processor's instruction set, but it is far more readable than machine code's stream of bytes. Let's look at some example assembly code to see how it works:

.proc main
  LDX PPUSTATUS
  LDX #$3f
  STX PPUADDR
  LDX #$00
  STX PPUADDR

copy_palettes:
  LDA palettes,x
  STA PPUDATA
  INX
  CPX #$20  ; 32 colors total
  BNE copy_palettes

Separation of instructions and data

The three-letter words in all caps on lines 2-6 and 9-13 are called opcodes. Each one represents an instruction from the processor's instruction set, but instead of referring to them by number, they now have names. LDA, for example, means "load accumulator". We'll be learning a few dozen opcodes over the course of this book; there are 56 "official" 6502 assembly opcodes in total.

Having instructions represented with short "words" means that we can now distinguish between instructions and data visually, instead of having to work through the code byte-by-byte. Anything to the right of an opcode is data that goes with that instruction. We call data for an instruction an operand. On line 3, for example, LDX #$3f is an instruction to "load the X register with the hex value $3f". LDX is the opcode, and #$3f is the operand. Here, and throughout this book, I will be presenting opcodes in all-caps, but this is a personal choice. The assembler will accept lowercase opcodes as well, and some developers prefer lowercase. The same applies to operand values - "$3f" can also be written as "$3F".

Constants and labels

While the separation of instructions and data is incredibly useful, assembly gives us several other tools as well. The operands on lines 2, 4, 6, and 10, in all-caps, are constants defined elsewhere. When the assembler (ca65) is run, it will replace the name of the constant (e.g. PPUSTATUS) with its value (e.g. $2002). You can make a constant by using the equal sign: PPUSTATUS = $2002.

Assembly also lets us set labels - named places in our code that can be referenced later. On line 8, a new label, copy_palettes, is defined. What makes a label a label is the colon (:) that it is followed by. On line 13, the label is used as an operand to the BNE opcode. We'll talk about the specifics of how it works later, but line 13 is essentially repeating the code from lines 8-13 again. When the assembler is run, it replaces labels in your assembly code with actual memory addresses in the machine code output.

Comments

You can add comments to your assembly code using the semicolon (;) character. Anything from the semicolon to the end of the line is treated as a comment and will be stripped out when you run your code through an assembler. Comments are useful for reminding yourself what a certain piece of code is meant to do. In our example code above, you can see a comment on line 12.

Assembler directives

Finally, assembly code gives us directives - instructions for the assembler that affect the conversion from assembly to machine code, rather than instructions for the processor where the machine code will be executed. Directives start with a period (.). The example code uses the .proc directive on line 1 to indicate a new lexical scope (more on that later). Another common directive is .byte, which indicates that the following bytes should be copied raw into the machine code output rather than trying to process them as opcodes or operands.

Your First Opcodes: Data Movement

Now that we have seen what 6502 assembly looks like, let's start learning it! I have broken out the opcodes that you will learn over the course of the book into seven major groups. The first group comprises instructions that move data between registers and memory.

Loading data: LDA, LDX, LDY

The "LD" commands load data into a register. As a reminder, the 6502 has three registers to work with: "A" or "accumulator", which can do math, and "X" and "Y", the "index registers". LDA loads data into the accumulator, LDX loads data into the X register, and LDY loads data into the Y register.

There are two main sources for this data: loading from a memory address and loading a given value. Which source is used is based on how you write the operand for the opcode. If you use a 16-bit value (four hex digits), it will load the contents of that memory address. If you use a hash sign (#) followed by an 8-bit value (two hex digits), it will load the exact value you used. Here is an example:

LDA $3f00  ; load contents of memory address $3f00
           ; into the accumulator
LDA #$3f   ; load the value $3f into the accumulator

These different operand formats are called addressing modes. The 6502 can use eleven different addressing modes (though most opcodes are limited to a subset of them). In case you're curious, the eleven addressing modes are Accumulator, Immediate, Implied, Relative, Absolute, Zeropage, Indirect, Absolute Indexed, Zeropage Indexed, Indexed Indirect, and Indirect Indexed.
The two modes we've just seen are absolute mode (providing a memory address) and immediate mode (providing an exact value). We will learn more addressing modes as needed over the course of the book.

When your code is run through the assembler, the different addressing modes actually translate to different entries in the instruction set. LDA $3f00 becomes ad 00 3f - "ad" is the instruction set number of the "LDA absolute mode" instruction, and $3f00 is placed in little-endian order. LDA #$3f, though, becomes a9 3f - "a9" being the instruction set number for "LDA immediate mode". The assembler is smart enough to insert the correct instruction set number based on how you write the operand, so you don't need to worry about which "LDA" you need to use.

Storing data: STA, STX, STY

The "ST" opcodes do the reverse of the "LD" opcodes - they store the contents of a register to a memory address. STA stores the contents of the accumulator to the location specified by its operand, STX stores the contents of the X register, and STY stores the contents of the Y register. The ST instructions cannot use immediate mode, since it doesn't make sense to store the contents of a register into a literal number. After a store operation, the register you stored from keeps its same value, allowing you to store the same register value in a different location right away.

Transferring data: TAX, TAY, TXA, TYA

Finally, the "T" instructions transfer data from one register to another. These opcodes are all read as "transfer from register to register" - TAX, for example, is "transfer from accumulator to X register". The "transfer" in these instructions is more accurately described as a "copy", since after one of these instructions, both registers will have the value of the first register.

A small example

Now that you've learned your first ten opcodes (and two addressing modes), let's look at a small example that makes use of them.

LDA #$a7
TAY
STY $3f00

What does this code do? Let's take it line-by-line:

  1. First, we load the value $a7 into the accumulator. LDA is the opcode that loads data into the accumulator. The # in front of $a7 indicates that we want to use immediate mode, so this is a number to put directly into the accumulator, not a memory address.
  2. Next, we copy the value $a7 from the accumulator to the Y register. Now both A and Y have the same value.
  3. Finally, we store the value of the Y register ($a7) at memory address $3f00.

Back to the Test Project

You now understand enough assembly to work out what most of the test project is doing. Let's look specifically at the main portion, reproduced here:

.proc main
  LDX $2002
  LDX #$3f
  STX $2006
  LDX #$00
  STX $2006
  LDA #$29
  STA $2007
  LDA #%00011110
  STA $2001
forever:
  JMP forever
.endproc

Most of this code is made up of loads and stores. The memory addresses that we load from or store to are $2001, $2002, $2006, and $2007. The immediate-mode values that we load are $3f, $00, $29, and %00011110 (a binary value, unlike the hex values we use everywhere else).

Looking more closely at the immediate-mode loads, notice that the first two are $3f00, the address in PPU memory where palettes begin, followed by $29, the green color that we are using in the background. This code is telling the PPU to store $29 at address $3f00, but how?

Memory-Mapped I/O

On the NES, addresses in the $2000-$6000 range are reserved for use as memory-mapped I/O (or "MMIO") addresses. "I/O" is "input/output" - sending data between different devices. "Memory-mapped" means that these interfaces to other devices are mapped to memory addresses - in other words, certain memory addresses are not memory at all, but rather connections to other devices.

Memory addresses in the low $2000s correspond to connections to the PPU. If you want to learn more about the PPU's MMIO addresses (or any other NES topic, for that matter), check out the NESDev Wiki. While it is not the best resource to learn from, it is an invaluable reference to the system, backed by meticulous research from its contributors. There are four MMIO addresses in use in our code; let's take a look at what each one does (along with the name each address is commonly known by).

$2006: PPUADDR and $2007: PPUDATA

The NES CPU (where your code is being executed) doesn't have direct access to the PPU's memory. Instead, CPU memory address $2006 lets your code select an address in PPU memory, and $2007 lets your code write a byte of data to that address. To set the address you want to write to, store two bytes of data to PPUADDR - first the "high" (left) byte, followed by the "low" (right) byte. Here's how our test project does that:

  LDX #$3f
  STX $2006
  LDX #$00
  STX $2006

This code first stores the byte $3f to $2006, then the byte $00 to $2006 - in other words, it sets the address for any following writes to PPU memory to $3f00, which is the address of the first color of the first palette.

To store data at the selected PPU memory address, store a byte to PPUDATA:

  LDA #$29
  STA $2007

This writes the byte $29 (which represents "green") to the memory address we selected before ($3f00). Each time you store a byte to PPUDATA, the memory address for the next store is incremented by one. If the next lines of code in the program were LDA #$21 and STA $2007, the byte $21 would be written to PPU address $3f01 even though we did not do anything with PPUADDR.

$2002: PPUSTATUS

PPUSTATUS is a read-only MMIO address. When you load from $2002, the resulting byte gives you information about what the PPU is currently doing. Reading from PPUSTATUS has one useful side-effect, as well: it resets the "address latch" for PPUADDR. It takes two writes to PPUADDR to fully specify a memory address, and it is possible that your code might make one write but never get around to doing the second write. Reading from PPUSTATUS makes it so that the next write to PPUADDR will always be considered a "high" byte of an address.

In our test project, we read ("load") from PPUSTATUS before attempting to write an address to PPUADDR: This process - reading from PPUSTATUS, writing two bytes to PPUADDR, and writing bytes to PPUDATA - is something that we will be using all the time. Anything that changes what is displayed to the screen uses this process to tell the PPU what to draw, and virtually everything in a game is going to change what is displayed to the screen. Learning this process well now will be very useful to you in future chapters.

  LDX $2002
  LDX #$3f
  STX $2006
  LDX #$00
  STX $2006
  LDA #$29
  STA $2007

$2001: PPUMASK

There's still one more thing our test project has to do after it tells the PPU to use color $29 as the first color of the first palette - it has to tell the PPU to actually start drawing things to the screen! PPUMASK allows your code to give the PPU instructions about what to draw, as well as set some tweaks to how colors are displayed. The byte that you store to PPUMASK is a set of eight bit flags, where each bit in the byte acts as an on/off switch for a particular property. Here is what each bit does. (Remember, the bits that make up a byte are numbered 0-7, with bit 0 all the way on the right, and bit 7 all the way on the left.)

PPUMASK bit flags. Source: NESDev Wiki
Bit # Effect
0 Greyscale mode enable
(0: normal color, 1: greyscale)
1 Left edge (8px) background enable
(0: hide, 1: show)
2 Left edge (8px) foreground enable
(0: hide, 1: show)
3 Background enable
4 Foreground enable
5 Emphasize red
6 Emphasize green
7 Emphasize blue

Before we look at the test project, a few notes on these options. Bits 1 and 2 enable or disable the display of graphical elements in the left-most eight pixels of the screen. Some games choose to disable these to help avoid flicker during scrolling, which we will learn more about later. Bits 5, 6, and 7 allow your code to "emphasize" certain colors - making one of red, green, or blue brighter and making the other two colors darker. Using one of the emphasis bits essentially applies a color tint to the screen. Using all three at the same time makes the entire screen darker, which many games use as a way to create a transition from one area to another.

Let's look again at our test project's code. What value does our test project write to PPUMASK?

  LDA #%00011110
  STA $2001

Here is what options we are setting, bit-by-bit:

Bit # Value Effect
0 0 Greyscale mode disabled
1 1 Show leftmost 8px of background
2 1 Show leftmost 8px of foreground
3 1 Background enabled
4 1 Foreground enabled
5 0 No red emphasis
6 0 No green emphasis
7 0 No blue emphasis

Since our test project turns on background rendering in its write to PPUMASK, our green background shows up after this line.

Wrapping Up the main Code

Our test project has written a color to PPU memory and turned on display rendering, so it's all done, right?

Not so fast. Remember, the CPU fetches and executes instructions one at a time, continuously. Unless we write code to keep the processor busy, it will continue reading (empty) memory and "executing" it, which could lead to disastrous results. Thankfully, there's an easy solution to this problem:

forever:
  JMP forever

We will learn the details of how this works later, but essentially these two lines of code create a label (forever), and then tell the CPU to fetch that label as the next instruction to execute. The CPU goes into an infinite loop, and your project happily does nothing but display its green background.

We've now covered everything in main (though we still haven't talked about what .proc and .endproc are doing), but there's much more in the test project to discuss. Next, we will learn about interrupt vectors and how our code initializes the NES when it first powers on.

Homework

Now that you know how the test project sets a background color, try modifying it to display a different color. I've included the color chart from Chapter 4 below. Don't forget to re-assemble and re-link your code before opening it in Nintaco.

The NES color palette.