Computer Systems: A Programmer’s Perspective
1. A tour of computer systems
Preprocessor -> Compiler -> Assembler -> Linker
OS:
- protects the hardware from misuse
- provide simple and uniform interface to the hardware
via processes, virtual memory and files
Processes:
- context: current values of the PC, register file, contents from main memory…
- context switch: saves the context of the current process and restores the one of another process.
Kernel:
- always resident in memory
- takes control on syscalls
Files:
- bytestring
- IO devices are modeled as files
Amdahl’s Law: to speed up a system, speed up a very large fraction of it
# T: time # S: x speedup # alpha: fraction # k: x improvement T_new = (1 - alpha)*T_old + (alpha*T_old)/k = T_old((1 - alpha) = alpha/k) S = 1/((1 - alpha) + a/k)Trhead-Level Concurrency:
- each its own L1, L2 caches
- L1 split in: (data) d-cache, (instructions) i-cache
- L3 is unified
- Hyperthread: 2 threads per core (2 copies of some parts of the hardware like PC and register files)
Instruction-Level Parallelism: multiple instructions at one time
Single-Instruction, Multiple-Data (SIMD) Parallelism: e.g. add 8 floats in parallel.
2. Representing and Manipulating Information
Big Endian:
Adress: 0x100 0x101 0x102 0x103 Value: 01 23 45 67 Little Endian:
Adress: 0x100 0x101 0x102 0x103 Value: 67 45 23 01
3. Machine-Level Representation of Programs
Instruction Set Architecture: defines format and behavior of a machine-level program
Memory adressses used by a a machine-level program are virtual adresses
Program Counter: adress of the next instruction to be executed
Integer file: 16 named locations storing 64-bit values
Condition code registers: status information about recent instructions. Usefull for control and data flow.
Vector registers: one or more integer or floating point.
x86:
Instruction length 1..15 (bytes)
suffixes:
type Intel asm suffix size (bytes) u8 byte b 1 u16 word w 2 u32 double word l 4 u64 quad word q 8 *u8 quad word q 8 f32 single precision s 4 f64 double precision q 8 general-purpose registers:
( %al ... %bpl ), %spl, ( %r8b ... %r15b ) (8 bits ) ( %ax ... %bp ), %sp , ( %r8w ... %r15w ) (16 bits) ( %eax ... %ebp ), %esp, ( %r8d ... %r15d ) (32 bits) ( %rax ... %rbp ), %rsp, ( %r8 ... %r15 ) (64 bits)operands:
expression meaning %rax register value 0x104 memory adress $0x104 literal 4(%rax) adress pointed by the register + 4 offset 9(%rax, %rdx) 9 + (adress in register_a) + (value in register_d) (%rax, %rdx, 4) (adress in register_a) + (value in register_d) * 4 mov:
movl $0x4050, %eax # immediate to register, 4 bytes movw %bp, %sp # register to register, 2 bytes movb (%rdi, %rcx), %eax # memory to register, 1 bytes movb $-17, (%eax) # immediate to memory, 1 bytes movq $0x4050, -12(%rbp) # register to memory, 8 bytes movabsq $0x4050, %rbp # 64bit immediate to register (only), 8 bytes- movz
- same as move but remaining bytes will be filled to zero.
- Register or memory as source and register as destination.
movzbw # byte to word movzbl # byte to long-word movzwl # word to long-word movzbq # byte to quartet movzwq # word to quartet # long-word to quartet with movl- movs
- same as move but with sign extension.
- Register or memory as source and register as destination.
movsbw # byte to word movsbl # byte to long-word movswl # word to long-word movsbq # byte to quartet movswq # word to quartet movslq # long-word to quartet- stack
- %rsp (stack pointer)
pushq %rax # push value to the stack popq %rdx # pop value from the stack to a register# Push subq $8, %rsp movq %rbp, (%rsp) # Pop movq (%rsp), %rax addq $8, %rsp