complement many

This commit is contained in:
2025-10-11 08:39:17 +09:00
parent 3d651c4a8a
commit 4c2c0363b3
10 changed files with 992 additions and 6 deletions

678
notes/3.md Normal file
View File

@@ -0,0 +1,678 @@
# Machine Level Programming
## History of Intel Processors
* Eveolutionary design: **Backwards compatible** up until `8086` in 1978
* **C**omplex **I**nstruction **S**et **C**omputer (CISC)
**RISC vs CISC**
* CISC has variable length instructions
* RISC has constant length instructions
1. Intel x86(8086)
1. IA32 to IA64
2. (after x86-64) EM64T(almost same as AMD x86-64)
2. AMD x86-64
## C, Assembly, machine code
* ***Architecture***: The parts of a processor design that one needs to understand or write assembly/machine code
* **ISA(Instruction Set Architecture)**
* e.g., x86, IA32, Itanium, x86-64, ARM
* ***Microarchitecture***: Implementation of the architecture
form of code:
* Machine Code: the byte-level programs that a processor executes
* Assembly Code: A text representation of machine code
### Assembly/Machine Code View
Programmer-Visible State (shown by ISAs)
* PC(Program Counter)
* Address of next instruction
* RIP in (x86-64)
* Register file
* Heavily used program data
* Condition codes
* store status information about most recent arithmetic or logical op
* Used for conditional branching
* Memory(external)
Compiling Into Assembly
```c {cmd=gcc, args=[-Og -x c -c $input_file -o 3_1.o]}
long plus(long x, long y);
void sumstore(long x, long y, long *dest) {
long t = plus(x, y);
*dest = t;
}
```
```sh {cmd hide}
while ! [ -f 3_1.o ]; do sleep .1; done; objdump -d 3_1.o
```
### Integer Registers
* In x86-64
`ax` `bx` `cx` `dx` `si` `di` `sp` `bp` (in 8bytes `r` 4bytes `e`)
`r8` `r9` `r10` `r11` `r12` `r13` `r14` `r15` (in 4bytes: add `d`)
* In IA32
`eax`(32bit): 16bit `ax`(`ah`, `al`); origin: accumulate
`ecx`(32bit): 16bit `cx`(`ch`, `cl`); origin: counter
`edx`(32bit): 16bit `bx`(`bh`, `bl`); origin: data
`ebx`(32bit): 16bit `dx`(`dh`, `dl`); origin: base
`esi`(32bit): 16bit `si`(`sih`, `sil`); origin: source index
`edi`(32bit): 16bit `di`(`dil`, `dil`); origin: destination index
`esp`(32bit): 16bit `sp`(`spl`, `spl`); origin: stack pointer
`ebp`(32bit): 16bit `bp`(`bpl`, `bpl`); origin: base pointer
#### understanding `movq`
In x86asm, There are three operand types: **immediate, register, memory**
* immediate: constant integer data like `$0x400` `$-533`
* register: one of 16 integer regs
* e.g., `%rax`, `%r13`
* but `%rsp` is reserved for special use
* memory: 8 bytes of memory at address given by register
* e.g., `(%rax)`
**movq usage**
`movq $src, $dest`
* limit: Cannot do memory-memory transfer with a single instruction(because memory is external device to cpu)
**memory addressing modes**
`(R)` means `Mem[Reg[R]]`
`D(R)` means `Mem[Reg[R]+D]`, constant displacement `D` specifies offset
`movq 8(%rbp), %rdx`
<table>
<tr>
<td>
```c
int swap (long *xp, long *yp) {
long t0 = *xp;
long t1 = *yp;
*xp = t1;
*yp = t0;
}
```
</td>
<td>
```nasm
swap:
movq (%rdi), %rax
movq (%rsi), %rdx
movq %rdx, (%rdi)
movq %rax, (%rsi)
ret
```
</td>
</tr>
</table>
Complete form of memory addressing modes:
`D(Rb, Ri, S)` means `Mem[Reg[Rb] + S*Reg[Ri] + D]`
* `D`: Constant "displacement"
* `Rb`: Base Register
* `Ri`: Index Register
* `S`: Scale Factor(1, 2, 4, or 8)
for example:
| `%rdx` | `%rcx` |
| -------- | -------- |
| `0xf000` | `0x0100` |
* `0x8(%rdx)` = `0xf008`
* `(%rdx, %rcx)` = `0xf100`
* `(%rdx, %rcx, 4)` = `0xf400`
* `0x80(,%rdx, 2)` = `0x1e080`
#### Arithmetic & Logical Operations
* `leaq $src, $dst`
* computing address without memory reference like `p = &x[i]`
* computing arithmetic expression `x + k * y`
* `addq $src, $dst`
* `subq $src, $dst`
* `imulq $src, $dst`
* `salq $src, $dst`
* `sarq $src, $dst`
* `shrq $src, $dst`
* `xorq $src, $dst`
* `andq $src, $dst`
* `orq $src, $dst`
all the above operator operates like `dest = dest # src`
* `incq $dest`
* `decq $dest`
* `negq $dest`
* `notq $dest`
## Control
**Processor State(x86-64, Partial)**
* Temporary data(`%rax`, ...)
* Location of runtime stack(`%rsp`)
* Location of current code control point(`%rip`, instruction point)
* Status of recent tests(`CF`, `ZF`, `SF`, `OF`)
### Condition Codes
* Single bit registers
* `CF` Carry flag (for unsigned)
* `SF` Sign flag (for signed)
* `ZF` Zero flag
* `OF` Overflow flag (for signed)
**Conditional Codes(Implicit Setting)**
Implicit setting is codes are set by arithmetic operations(`addq`, `subq`, `mulq`)
for example: `addq`: `t = a + b`
* `CF` set if carry out from most significant bit or unsigned overflow
* `ZF` set if `t == 0`
* `SF` set if `t < 0` (as signed)
* `OF` set if two's-complement overflow or signed overflow
`(a > 0 && b > 0 && (a + b) < 0) || (a < 0 && b < 0 && (a + b) >= 0)`
The codes are not implictly set by `leaq`, because it is not designed to be used as arithmetic but used as **address calculation**. so it cannot affect to conditional codes.
**Conditional Codes(Explicit Setting)**
The codes are set explictly by compare instruction.
`cmpq b, a` is computing `a - b` without setting destination.
* `CF` set if carry out from most significant bit or unsigned overflow
* `ZF` set if `a == b` or `a - b == 0`
* `SF` set if `(a - b) < 0` (as signed)
* `OF` set if two's-complement overflow or signed overflow
`(a > 0 && b > 0 && (a - b) < 0) || (a < 0 && b < 0 && (a - b) >= 0)`
And explictly set by test instruction
`testq b, a` is computing `a & b` without setting destination.
Sets condition codes based on value of `a & b` it is useful to have one of the operands be a mask.
* `ZF` set when `a & b == 0`
* `SF` set when `a & b < 0`
**Reading Condition Codes**
`setX`: set single byte based on combination of condition codes
| setX | effect | desc |
| ------- | ---------------- | ------------------------- |
| `sete` | `ZF` | Equal / Zero |
| `setne` | `~ZF` | Not Equal / Not Zero |
| `sets` | `SF` | Negative |
| `setns` | `~SF` | Nonnegative |
| `setg` | `~(SF^OF) & ~ZF` | Greater (signed) |
| `setge` | `~(SF^OF)` | Greater or Equal (signed) |
| `setl` | `SF^OF` | Less (signed) |
| `setle` | `(SF^OF) \| ZF` | Less or Equal (signed) |
| `seta` | `~CF & ~ZF` | Above (unsigned) |
| `setb` | `CF` | Below (unsigned) |
it deos not alter remaining bytes of registers. only use 1 byte register(`%al`, `%bl`)
```nasm
cmpq %rsi(y), %rdi(x) # compare x and y
setg %al # set when >(greater)
movzbl %al, %eax # move zero extend byte to long
ret
```
### Conditional Branches
#### Jumping
`jX` jump to different part of code depending on condition codes.
| jX | condition | desc |
| ----- | ---------------- | ------------------------- |
| `jmp` | 1 | Unconditional |
| `je` | `ZF` | Equal / Zero |
| `jne` | `~ZF` | Not Equal / Not Zero |
| `js` | `SF` | Negative |
| `jns` | `~SF` | Nonnegative |
| `jg` | `~(SF^OF) & ~ZF` | Greater (signed) |
| `jge` | `~(SF^OF)` | Greater or Equal (signed) |
| `jl` | `SF^OF` | Less (signed) |
| `jle` | `(SF^OF) \| ZF` | Less or Equal (signed) |
| `ja` | `~CF & ~ZF` | Above (unsigned) |
| `jb` | `CF` | Below (unsigned) |
Old Style Conditional Branch
```c {cmd=gcc args=[-Og -x c -fno-if-conversion -c $input_file -o 3_3.o]}
long absdiff(long x, long y) {
long result;
if (x > y) result = x - y;
else result = y - x;
return result;
}
```
```sh { cmd hide }
while ! [ -f 3_3.o ]; do sleep .1; done; objdump -d 3_3.o -Msuffix
```
**expressing with `goto`**
```c {cmd=gcc args=[-Og -x c -fno-if-conversion -c $input_file -o 3_4.o]}
long absdiff_j(long x, long y) {
long result;
int ntest = x <= y;
if (ntest) goto Else;
result = x-y;
goto Done;
Else:
result = y-x;
Done:
return result;
}
```
#### Conditional Move
But this branchings are very disruptive to instruction flow through pipelines, **Conditional Moves** are highly used because they do not require control transfer.
```c {cmd=gcc args=[-O3 -x c -c $input_file -o 3_5.o]}
long absdiff(long x, long y) {
long result;
if (x > y) result = x - y;
else result = y - x;
return result;
}
```
```sh {cmd hide}
while ! [ -f 3_5.o ]; do sleep .1; done; objdump -d 3_5.o -Msuffix
```
However, there are several *bad cases* for conditional move.
* expansive computations
```c
val = Test(x) ? Hard1(x) : Hard2(x);
```
because both values are get computed. only simple computations are effective for conditional moves.
* risky computations
```c
val = p ? *p : 0;
```
both values get computed may have undesiarable effects.
* Computations with side effects
```c
val = x > 0 ? x*=7 : x+=3;
```
each expression has side-effect.
### Loop
#### do-while
<table>
<tr>
<td>
```c
long pcount_do(unsigned long x) {
long result = 0;
do {
result += x & 0x1;
x >>= 1;
} while (x);
return result;
}
```
</td>
<td>
```c {cmd=gcc args=[-Og -x c -c $input_file -o 3_6.o]}
long pcount_goto(unsigned long x) {
long result = 0;
loop:
result += x & 0x1;
x >>= 1;
if (x) goto loop;
return result;
}
```
</td>
</tr>
</table>
```sh {cmd hide}
while ! [ -f 3_6.o ]; do sleep .1; done; objdump -d 3_6.o -Msuffix
```
**general do-while translation**
<table>
<tr>
<td>
```c
do {
Body
} while (Test);
```
</td>
<td>
```c
loop:
Body
if (Test) goto loop;
```
</td>
</tr>
</table>
#### while
**general while translation#1**
it is called **jump-to-middle translation**, used with `-O0` (or `-Og`) flag.
<table>
<tr>
<td>
```c
while(Test) {
Body
}
```
</td>
<td>
```c
goto test;
loop:
Body
test:
if (Test)
goto loop;
done:
```
</td>
</tr>
</table>
```c {cmd=gcc args=[-Og -x c -c $input_file -o 3_7.o]}
long pcount_while(unsigned long x) {
long result = 0;
while (x) {
result += x & 0x1;
x >>= 1;
}
return result;
}
```
```sh {cmd hide}
echo "jmp-to-middle translation"
while ! [ -f 3_7.o ]; do sleep .1; done; objdump -d 3_7.o -Msuffix
```
**general while translation#2**
while to do-while conversion, used with `-O1` flag.
<table>
<tr>
<td>
```c
while(Test) {
Body
}
```
</td>
<td>
```c
if (!Test) goto done;
do {
Body
} while (Test);
done:
```
</td>
<td>
```c
if (!Test) goto done;
loop:
Body
if (Test) goto loop;
done:
```
</td>
</tr>
</table>
```c {cmd=gcc args=[-O1 -x c -c $input_file -o 3_8.o]}
long pcount_while(unsigned long x) {
long result = 0;
while (x) {
result += x & 0x1;
x >>= 1;
}
return result;
}
```
```sh {cmd hide}
echo "while to do-while conversion"
while ! [ -f 3_8.o ]; do sleep .1; done; objdump -d 3_8.o -Msuffix
```
#### for loop form
<table>
<tr>
<td>
```c
for (init; test; update) {
Body
}
```
</td>
</tr>
</table>
**for-to-while conversion**
<table>
<tr>
<td>
```c
for (Init; Test; Update) {
Body
}
```
</td>
<td>
```c
Init;
while(Test) {
Body
Update;
}
```
</td>
</tr>
<tr>
<td>
```c {cmd=gcc args=[-O3 -x c -c $input_file -o 3_9.o]}
#include <stddef.h>
#define WSIZE 8 * sizeof(int)
long pcount_for(unsigned long x) {
size_t i;
long result = 0;
for (i = 0; i < WSIZE; i++) {
unsigned bit = (x >> i) & 0x1;
result += bit;
}
return result;
}
```
</td> <td>
```c {cmd=gcc args=[-O3 -x c -c $input_file -o 3_10.o]}
#include <stddef.h>
#define WSIZE 8 * sizeof(int)
long pcount_for(unsigned long x) {
size_t i;
long result = 0;
i = 0;
while(i < WSIZE) {
unsigned bit = (x >> i) & 0x1;
result += bit;
i++;
}
return result;
}
```
</td>
</tr>
<tr>
<td>
```sh {cmd hide}
while ! [ -f 3_9.o ]; do sleep .1; done; objdump -d 3_9.o -Msuffix
```
</td>
<td>
```sh {cmd hide}
while ! [ -f 3_10.o ]; do sleep .1; done; objdump -d 3_10.o -Msuffix
```
</td>
</tr>
</table>
for to do-while conversion, initial test can be optimized away.
### Switch
#### Jump Table Structure
Switch form
<table>
<tr>
<td>
```c {cmd=gcc args=[-Og -fno-asynchronous-unwind-tables -fno-stack-protector -x c -S $input_file -o 3_11.s]}
long switch_eg (long x, long y, long z) {
long w = 1;
switch(x) {
case 1:
w = y*z;
break;
case 2:
w = y/z;
/* Fall Through */
case 3:
w += z;
break;
case 5:
case 6:
w -= z;
break;
case 7:
w *= z;
break;
default:
w = 2;
}
return w;
}
```
</td>
<td>
```sh {cmd hide}
while ! [ -f 3_11.s ]; do sleep .1; done; cat 3_11.s
```
</td>
</tr>
</table>
## Procedures
Mechanisms in Procedures
* **Passing control**
* to beginning of procedure code
* back to return point
* **Passing data**
* procedure arguments
* return value
* **Memory management**
* allocate during procedure execution
* deallocate upon return
this mechanisms are all implemented with machine instructions. **x86-64 implementation** of a procedure used only those mechanisms required.
### Stack Structure
**x86-64 Stack**
Region of memory managed with *stack discipline*. It grows toward lower addresses. `%rsp` contains lowest stack address(address of top element).
`pushq $src`
* fetches operand at src
* decrement `%rsp` by 8
* write operand at address given by `%rsp`
`popq $dest`
* read value at address given by `%rsp`
* increment `%rsp` by 8
* store value at dest(must be register)
### Procedure Control Flow
```c {cmd=gcc args=[-Og -x c -c $input_file -o 3_12.o]}
long mult2(long x, long y) {
long t = x * y;
return t;
}
void multstore(long x, long y, long *dest) {
long t = mult2(x, y);
*dest = t;
}
```
```sh {cmd hide}
while ! [ -f 3_12.o ]; do sleep .1; done; objdump -d 3_12.o -Msuffix
```
Procedure call `call label`
* push return address on stack
* jmp to label
Return address:
* Address of the next instruction right after call
Procedure return: `ret`