Memory Hierarchy
Thinking about how the assembler might translate these symbols into memory addresses, we notice an additional factor we have to consider when designing the computer system's instruction set — addressing. If the computer system has a large amount of memory, then our computer must work with long memory addresses. And the longer a memory address is, the longer it takes for the CPU to perform its operations, since it can only execute its instructions once it's received all of the addresses bits. This time delay is further increased by the fact that the CPU also has to access the long address.
Fortunately, von Neumann provided a solution to this problem: Instead of implementing main memory as just one giant block, we implement it as a memory hierarchy:
Memory hierarchy is system for organizing memory. The idea is that we take all of the computer's registers, and divvy them up into four places. A small amount of registers are given to the CPU. These registers are called the processor registers. A slightly larger amount of registers are placed immediately next to the CPU. These registers are called caches. A large amount of registers are placed in main memory, close to the cache but further from the CPU. The largest amount of registers are found in the disk (i.e., secondary memory); close to main memory but the furthest from the CPU.
When arranged in this sequential manner, we see that the closer we are to the CPU, the shorter the address. Thus, the processor registers are the fastest registers to read and write, followed by the cache and main memory. The disk's registers are the furthest from the CPU, and take the longest to read and write.
Processor Registers
Given how fast processor registers are, they're used for a few key
operations. Some of the registers are data registers. These registers
are used to store the necessary data for completing an operation's
execution. For example, if the CPU must execute ADD x y, the data
contents of x and y are stored inside the data registers to perform the
computation.
Some of the registers are address registers. These registers are used
to store some address outside the CPU, perhaps in main memory. For example,
suppose the instruction DEF x y means "assign the value in y to the
address x." To execute this instruction, the CPU might store the value
stored in y in some data register, then store the address correspoding to
y inside an address register.
Addressing Modes
Using memory hierarchy, we have different ways to read and write registers, called addressing modes. These modes are triggered by the way the assembly language instructions are written.
Immediate addressing is the fastest addressing mode. We're explicitly providing data to the CPU:
ADD 1 2
Register addressing is the next fastest addressing mode. If the data's already stored inside the CPU's data registers, we can simply provide the Assembly language's instruction representing the registers:
ADD R1 R2
Direct addressing is the next fastest. Here, we're explicitly specifying the address:
ADD R1 M[321]
Finally, the slowest form of addressing is indirect addressing. Here, we're using symbols to denote addresses, so the CPU must wait to receive the addresses:
ADD x y