[brassic]

Early RISC processors read data from memory 32 bits (4 bytes) at a time, and these reads had to be aligned on 4-byte boundaries. This was a feature of the memory architecture, not just the processor.

Thus an aligned read of a 32-bit integer took one memory access; an unaligned read took two, which took twice as long. This killed performance.

Rather than quietly performing badly, the processor threw an exception to encourage you to fix your code.

The ARM2 worked a bit differently. It ignored the bottom two bits of the address when it read a value from memory. When the read was complete the value was rotated by the value of the bottom 2 bits multiplied by 8. This had the effect of putting the byte referenced by the full address in the bottom 8 bits of the 32-bit register. A flag in the instruction let you optionally mask off the top 24 bits to simulate a byte read.