The ISA Level

The ISA Level

• The Instruction Set Architecture (ISA) is positioned between the microarchtecture level and the operating system level. Historically, this was the only level. It is the interface between the software and the

hardware. Programs written in high-level languages (C,

C++, FORTRAN 90, etc.) are compiled into a common intermediate form - the ISA level - which the hardware executes directly.

The ISA Level

The ISA Level

How do we decide what the ISA should support?• Ideally, ask both hardware and software engineers

the features they would like to have and reach a compromise.

• In practice, the most important feature is backward compatibility.

• A good ISA should should define a set of instructions that can be implemented effectively in current and future technologies.

• It should also provide a clean target for compiled code.

The ISA Level

For some architectures, the ISA level is specified by a formal defining document (e.g. ARM v7).• This allows different implementations of the same

architecture.• Such defining documents contain normative

sections, which impose requirements, and informative sections, that are intended to help the reader but are not part of the formal specification.

• A Core i7 ISA reference is available at http://developer.intel.com

The ISA Level

There are at least two modes at the ISA level:• Kernel mode is intended to run the operating

system and allows all instructions to be executed.• User mode is intended to run application programs

and does not permit certain sensitive instructions to be executed.

All computers divide memory up into cells that have consecutive addresses.• The most common cell size is 8 bits (called a byte).• The reason for using 8 bits is that ASCII codes are 7

bits (add one bit for parity).

Memory Models

Bytes are grouped into 4-byte (32 bit) or 8-byte (64 bit) words with instructions available for manipulating entire words.

Many architectures require words to be aligned on their natural boundaries.• This allows more efficient memory operations.• Reading words at arbitrary addresses requires extra

logic on the chip, making it bigger and more expensive.

• The Core i7 does not require alignment in order to retain compatibility with the 8088, however unaligned references require two memory references.

The ISA Level

Registers

• All computers have some registers visible at the ISA level. These control execution of the program hold temporary results are used for other purposes

• ISA level registers can be divided into two categories: special-purpose (PC, SP, etc.) general-purpose (local variables, etc.)

Registers

On some machines the general-purpose registers are completely symmetric and interchangeable.

It is common for compilers and operating systems to adopt conventions about how the general-purpose registers are used.

Many special-purpose registers are available only in kernel mode (those that control cache, I/O devices and other hardware).

Registers

• One control register that is a hybrid kernel/user is the flags register or PSW (Program Status Word). This register contains various condition bits:

N - Set when the result was Negative Z - Set when the result was Zero V - Set when the result caused an oVerflow C - Set when the result caused a Carry out of the leftmost

bit A - Set when there was a carry out of bit 3 (Auxiliary

carry) P - Set when the result had even parity

Core i7 ISA Level

The Core i7 has three operating modes, two of which make it look like an 8088. In real mode, features added since the 8088 are turned off. • An application error causes the machine to crash.

Virtual 8086 mode makes it possible to run old 8088 programs in a protected way.• A special isolated environment that acts like an 8088 is

created, except that if the program crashes, the operating system is informed.

• Used when an MS-DOS window is opened. Protected mode has four privilege levels controlled by

bits in the PSW

Core i7 Registers

The first four registers are general-purpose:• EAX is the main arithmetic register.• EBX is good for holding pointers• ECX plays a role in looping• EDX is needed for multiplication and division

(holding half of the 64-bit products and dividends) These registers contain 8- and 16-bit registers

in the low-order bits for manipulation of 8- and 16-bit quantities, respectively.

Core i7 Registers

The next three registers are general-purpose:• ESI and EDI hold pointers into memory, especially

for the hardware string manipulation routines.• EBP is a pointer register (typically to the base of the

current stack frame, like LV in JVM).• ESP is the stack pointer.

CS through GS are segment registers. EIP is the program counter EFLAGS is the PSW.

Core i7 Primary Registers

Overview of the OMAP4430 ARM ISA Level

• The version 7 ARM’s general registers.

Overview of the ATmega168 AVR ISA Level

• On-chip register and memory organization for the ATmega168

Core i7 Data Types

The Core i7 supports two’s complement integers, unsigned integers, binary coded decimal numbers, and IEEE 754 floating-point numbers

It handles 8 and 16 bit integers as well. 64-bit mode with 64-bit registers and

operations Operands do not have to be aligned in memory,

but better performance is obtained if they are. There are also instructions for manipulating 8-

bit ASCII character strings: copying and searching.

Core i7 Numeric Data Types

Data Types on the OMAP4430 ARM CPU

Data Types on the ATmega168 AVR CPU

Instruction Formats

• An instruction consists of an opcode, usually with some additional information such as where operands come from, and where results go.

• The general subject of specifying where the operands are is called addressing.

• Several possible formats for level 2 instructions are shown on the next slide.

Common Instruction Formats

Instruction Formats

• On some machines, all instructions have the same length; on others there may be many different lengths.

• Instructions may be shorter than, the same length as, or longer than the word length. Having a single instruction length is simpler

and makes decoding easier, but is less efficient.

Common Instruction Formats

Expanding Opcodes

We will now examine tradeoffs involving both opcodes and addresses.

Consider an (n + k) bit instruction with a k-bit opcode and a single n-bit address.• This instruction allows 2k different operations and 2n

addressable memory cells.• Alternatively, the same n + k bits could be broken

up into a (k - 1) bit opcode and an (n + 1) bit address, meaning half as many instructions and either twice as much addressable memory or the same amount of memory with twice the resolution.

Expanding Opcodes

The concept of a expanding opcode can best be seen through an example.

Consider a machine in which instructions are 16 bits long and addresses are 4 bits long.• This might be reasonable on a machine that has 16

registers on which all arithmetic operations take place.

• One design would be a 4-bit opcode and three addresses in each instruction, giving 16 three-address instructions.

Expanding Opcodes

Expanding Opcodes

However, if the designers need 15 three-address instructions, 14 two-address instructions, 31 one-address instructions, and 16 instructions with no address at all, they can use opcodes 0 to 14 as three-address instructions but interpret opcode 15 differently.• Opcode 15 means that the opcode is contained in

bits 8 to 15 instead of 12 to 15.

Expanding Opcodes

Core i7 Instruction Formats

OMAP4430 ARM CPU Instruction Formats

• The 32-bit ARM instruction formats.

ATmega168 AVR Instruction Formats