ARM Instruction Set(Part 1): Initialization

Million Instruction Per Second  is a measurement of program execution speed based on the number of millions of instructions. MIPS is computed as the instruction count divided by the product of the execution time and 10^6.

CPI – Clock Cycles Per Instructions

f – Clock frequency

Note: MIPS also refer to a RISC ISA developed by MIPS Technology

The idea that instructions and data of many types can be stored in memory as numbers, leading to stored-program computers.This sole concept enables today’s computers to run/share a wide variety of programs just by executing from the memory.

Endianness is one of the distinguishing factors which defines an architecture.

It’s based on whether LSB or MSB is stored first in the memory.

It’s generally of two types: Little Endian and Big Endian.

Little Endian: Here, numbers are stored in the following formats:

MSB………………………………………LSB

Big Endian: Here, numbers are stored in the following formats:

LSB………………………………………MSB

Most of the primitive architectures were Little Endian based(ARM was one of them), but with the latest architectures endianness is configurable.

But, How simply storing the numbers in a reciprocated format helps?

While adding two numbers, we generally move by adding LSB first then MSB.

On an architecture like Intel 8088, where size of data bus(8 bits) is not equal to size of registers(16 bits), storing in little endian format helps in decreasing operations while adding as LSB is fetched first the MSB. So in a clock cycle we can perform both fetching and arithmetic operation.

Big-endian numbers, on the other hand, can be compared faster then little endian numbers. Since there first bit(MSB) is the sign bit itself we can easily compare on basis of their signs.

Big-endian: IBM,SPARC, Motorola
Little-endian: Intel,DEC
Supporting both:MIPS, PowerPC

Note: 

• PowerPC ( Performance Optimization With Enhanced RISC – Performance Computing) is a RISC instruction set architecture created by Apple–IBM–Motorola alliance, known as AIM.

• Scalable Processor Architecture (SPARC) is a reduced instruction set computing (RISC) instruction set architecture (ISA) originally developed by Sun Microsystems

Further reading: https://people.cs.umass.edu/~verts/cs32/endian.html

Note: Before moving forward I would like to clear a very basic doubt that i had when reading it for first time:

What’s the difference between ‘Registers’ and Memory?
Here, by registers I mean the registers internal to a CPU.  while by, memory i mean RAM.
ARM has 16 internal registers ranging from r0,r1………r12,SP,LR,PC.
Obviously, access to registers is faster then memory thus by using registers for common operations reduces code spill and memory traffic.
MIPS64 has 32 such General purpose registers.

TYPE OF INSTRUCTION SETS:

1. General purpose register architecture

   Let’s take an example of a higher-level statement A=B+C. This statement can be visualized in this architecture as :
   LDR R1,B
   LDR R2,C
   ADD R3,R1,R2
   STR R3,A

   i.e here values of B and C (assuming these variables are in memory) are loaded into registers and then operations are performed on registers itself.
   This is one of the most widely used architecture with an added advantage that the variable has to loaded from memory only once thus reducing LOAD/STORE instructions.
   It is further of 3 types:

   • Register-Register type(2 address):
   • Here, both operands are registers and output/result of the operation will overwrite one of the register. 
   • For ex: ADD R1,R2,R1;
   • Register-Register type(3 address): 
   • Here, both operands are registers and result is stored in another register.
   • For ex: ADD R1,R2,R3;
   • Register-Memory Type:
   • Here, one operand is a register and other is the content of a memory location.

   There is another type of architecture Memory-Memory Type in which both the operands come from memory, but it’s obsolete nowadays. Both X86 and ARM are general-purpose register type machines , but X86 has Memory-Register Architecture along with some instructions from 2 address Register-register type while ARM(and most of RISC machines) has Register-Register Architecture(both 2 and 3 addresses).

   Image courtesy: http://images.slideplayer.com/28/9379293/slides/slide_3.jpg

2. Stack architecture

   Here, The instruction A=B+C will be decoded as:
   PUSH B;
   PUSH C;
   ADD
   POP A;
   i.e, here instead of registers stack is used. Variables are loaded in and from stack and operations are performed there itself.
   It’s very rare to find stack architecture in modern chips rather it’s quite popular in designing interpreters like Java ByteCode

3. Accumulator architecture

   It’s the second most popular architecture after general purpose register(GPR) architecture.
   Here, A=B+C will be converted to:

   LOAD B into Accumulator
   ADD C to Accumulator
   Store Accumulator to A

   Here, Accumulator can be any register stored directly within ALU. Thus, Accessing it is very fast.  It has a drawback that variable need to be loaded from memory again and again in case of multiple operation with same variable.
   One of the operand in this architecture is mostly accumulator.

   Image courtesy: https://uditagarwal.in/wp-content/uploads/2017/08/isa-architecture-4-638-300×225.jpg