Lecture 9
1
x86 64 Instructions
April 29, 2015
Alignment
• As mentioned previously, we will use 8 bytes (64 bits) for all non-structure data types. This is a terribly
simplistic approach.
• As such, all data will be implicitly 8-byte aligned.
• End of input argument area of stack frame is 16-byte aligned (by ABI).
2
Instruction Set
• See handout.
3
Subroutines
• Subroutines will begin with a prologue to setup the stack.
.text
.global main
.type
main, @function
main:
pushq
%rbp
movq
%rsp, %rbp
; store callee-saved registers (pushq)
subq
$56, %rsp
– Set following section as executable text (setting stays until another .section pseudo-operation
is specified).
– Make name accessible externally.
– Specify the type for the linker.
– Define the function.
– Save stack pointer (rsp) as frame pointer (rbp).
– Move stack pointer.
• Each subroutine also has an epilogue.
addq $56, %rsp ; only if stored registers
; restore callee-saved registers (popq)
movq %rbp, %rsp
popq %rbp
ret
.size
main, .-main
–
–
–
–
–
Move stack pointer to saved registers.
Pop callee-saved registers.
Restore prior stack pointer.
Restore prior frame pointer.
Set size of subroutine (diff of current location and initial def of main).
1
4
Globals
• When generating assembly, we don’t necessarily know where a global variable will be placed (and
therefore what it’s address will be).
• Actual addressing will be taken care of by the linker (and then the loader).
• This is good. It means that the assembly can just use symbols.
• A global can be defined by a .comm directive (specifying the size and alignment).
.comm glob i,8,8
• This symbol can then be coupled with the rip register to create an address.
movq
glob i(%rip), %rdx
• Value is now in %rdx.
• Can use name as immediate to store the address in a register.
movq
5
$glob i, %rdx
Strings (printf/scanf )
• Similar to the idea above. Create a label that corresponds to the desired string.
.section
.rodata
.LLC0:
.string "%ld\n"
• Then use that label to get the address of the string.
mov
mov
mov
call
$.LLC0, %rdi
$0, %rsi
$0, %rax
printf
; for varargs functions -- # of vector registers used
• Same idea for scanf, but need address for second argument.
2
Sampling
Arithmetic
addq
subq
imulq
imulq
idivq
(r1 | i32 | mem), (r2 | mem)
(r1 | i32 | mem), (r2 | mem)
(r1 | mem), r2
i32 , (r1 | mem), r2
(r | mem)
r2 += r1
r2 -= r1
r2 *= r1
r2 = r1 * i
rax = rdx:rax / r, rdx = rdx:rax % r
Boolean
andq
orq
xorq
(r1 | i32 ), r2
(r1 | i32 ), r2
(r1 | i32 ), r2
r2 &= r1
r2 |= r1
r2 ∧= r1
Comparison and Branching
cmp
jmp
je
jg
jge
jl
jle
jne
(r1 | i32 ), r2
label
label
label
label
label
label
label
tmp = r2 - r1, sets flags
Loads and Stores
movq disp(base, offset, scalar), r
movq (r | i32 ), disp(base, offset, scalar)
load from base + offset * scalar + disp into r
store from r to base + offset * scalar + disp
Invocation
call
ret
label
Conditional Moves
cmoveq (r1
cmovgq (r1
cmovgeq(r1
cmovlq (r1
cmovleq (r1
cmovneq(r1
|
|
|
|
|
|
mem),
mem),
mem),
mem),
mem),
mem),
r2
r2
r2
r2
r2
r2
move
move
move
move
move
move
if
if
if
if
if
if
equal
greater
greater or equal
less
less or equal
not equal
Shifts
shrq
cl, (r | mem)
logical right shift cl times