You are on page 1of 30

IT403

Introduction to computer
architecture
Branch instructios
● Conditional branches
– Execute if some condition holds
– Transfer the program control to a different point
based on a condition
● Unconditional branch
– Take the branch without checking any condtion
– Transfer the program flow to a different point
– Also called as jumps
Branch equal
● Syntax: beq r1, r2, <target>
● Sematics:
– Transfer the program control to the given target address of r1
== r2
– Target is a label in the code (which will later be replaced with
an address)
● Examples
beq r1, r2, if
…..
if:
Branch NOT equal
● Syntax: bne r1, r2, <target>
● Sematics:
– Transfer the program control to the given target address of r1
!= r2
– Target is a label in the code (which will later be replaced with
an address)
● Examples
bne r1, r2, if
…..
if:
Other banch instructions
● branch greater than zero
– Syntax: bgtz r1, <target>
– Sematics: take the branch if r1 > 0
● Branch greater than or equal to zero
– Syntax: bgez r1, <target>
– Sematics: take the branch if r1 >= 0
Other banch instructions ...
● branch less than zero
– Syntax: bltz r1, <target>
– Sematics: take the branch if r1 < 0
● Branch less than or equal to zero
– Syntax: blez r1, <target>
– Sematics: take the branch if r1 <= 0
Converting to instructions
● Some of the conditions provided by the higher
level language are much more expressive than
this
● Compiler will convert then to suitable
operations and use these hardware instructions
to implement them.
● Example: if(x + 1 > y)
Thinking time …
● How to convert the following C-code into
assembly
if(x+1 < y && y > 10)
z ++;
else
z –;
Note on the target
● The taget given in the assembly code in encoded using
16bits
– The address field in the j-type instructions
● There is a limit to the target address
– It should be within the 216 byte limit
● If the actual function is outside of this limit another
instruction will be placed there to make the real branch.
● The label will be replaced with the actual address at the link
time
– Assembel → link → final executable
Address encoding
● Size aligned address:
– Almost all computers requires you to align data to its size
– That is, if the data item is 4bytes its address should be divisible
by 4 (or 2 LSBs should be zero)
● Since all instructions are 32bit and they are stored aligned
to the size the last two bits of any instruction should be
zero
– So, we do not include the last 2bits in the encoding (or divide the
address by 4)
– When taking the branch, recompute the address by adding
2more bits (or multiply with 4)
Jump instruction
● In addition to conditional branches MIPS also provide
jump which transfer the control always
● Two type of jump
– Syntax: j <taget address>
– Sematics: jump to the given address. Address has a limited
range (you cannot go to any place in the address space)

– Synaxt: jr r1
– Sematics: jump to the address given in the register. No limit on
the range. However, need to use an additional register for this.
Thinking time …
● Are these enogh?
– Can we implement all branches provided in a
typical programming language

● How about function calls?


What is special about function calls?
● Function calls are esentially transfer of control (with few
other things)
– Go and execute the called function
● Additional feature
1)Need to pass arguments
2)Need to get a result back
3)Need to come back to where we were
● The 1,2 above are handled by the calling convention (which
I will not cover this completely in this course)
● Need some hardware support for the 3 point above
The jump and link instruction
● The jump and link instruction provide the means for function
calls
● It is essentialy a unconditional branch, but the returnn address
will be saved in a given register
● Synax: jal <function>
● Sematics: jump to the function, load PC+4 to r31
● Idea: so I know where to get back
● Once the function finishes execution it can jump register to r31
● jalr jump and link register is also available
A typical function
foo:
save callee saved register (ones you would use)
do the operation
restore the callee saved register (ones you messed up fix
them with original values)
jr r31
● Notes:
– Registers will be saved on the stack
– Not all registers need to be saved
Calling convention
● It is a convention developed by compiler
designers for a given programming language
and architecture, which says
– How to pass the arguments
– How to get the results back
– What registers needs to be preserved
– What registers can be trashed
MIPS calling convention for C
● R0 → always contains 0
● R1 → reserved for the assemble (not used by C code) (called at)
● R2 – R3 → used to return values from a subrotine. If return value
takes one register only R2 is used (V0 – V1)
● R4 – R7 → used to pass arguments to function (A0 – A4)
● R8 – R15, R24, R25 → no need to preserve
● R16 – R23, R28 – R30→need to be preserved (R29 = Stack
Pointer)
● R31 → contains the return address
● R27 – R18 → reserved for the operating system
Excercise
● Conver the following C-code to assembly
int main() int main()
{ {
int a = add(1, 2); int a[10];
return 0; addone(a, 10);
} return 0;
}
int add(int i, int j)
{
return i+j;
void addone(int *a, int limit)
} {
for(int i=0; i<limit; i++)
a[i] += 1;
}
Load instructions
● Loading a word is simple
– What you load from the memory will fit into the register
– Example: on a 32bit maching the registers will be 32bits
and lw/sw will take 32bits from/to memory
● Some data types takes less than 32 bits
– Example: from c: shorts, bytes …
● How to deal with these ones
– Issue: read 16bits from memory and put to a 32bit
register
Signed and unsigned loads
● MIPS provides two types of loads (applicable
when loading sub-word)
● Signed load:
– Basic idea: reapeat the signed bit to fill the MSBs
as requred
– Example: when loading two bytes (16bits) from
memory to a four byte (32bit) register repeat the
MSB from what is loaded to fill the additional 16bits
Effect of signed load
● Find the 8bit 2's complement of -9
● Do a signed load of this number to a 16bit
register
● What is the 16bit value (using 16bit 2's
complement)?
When to use what
● If your data type uses 2's complement then
loads should be signed
– Example: from C, short, byte …
● If your data type is not 2's complement then
loads should be unsigned
– Example: from unsigned short, …
Unsigned loads
● Unsigned loads will fill the additional bits with 0
● Example: if you load 16bits from memory to a
32bit register
– Bottom 16bits (LSB) will be filled with what is
loaded from memory
– Top 16bits (MSB) will be filled with zeros
– So that the unsigned value will be the same
Loading sub-words
● Load half signed → lh
● Load half unsigned → lhu
● Load byte signed → lb
● Load byte unsigned → lbu
● Exercise: Convert the following C-code to
assembly
short a[10];
for(int i=0; i<10; i++) a[i] = a[i] + 10;
Loading immediate values
● MIPS provide instructions to load immediate values
● The rules on signed and unsigned loads will be
applicable here as well
● Value is limited by the 16bits in the I-type
instructions
● Instruction
– li → load the immediate value, signed
– liu → load immediate unsigned
Storing subwords
● Storing subwords is simpler that loading them
● You simply select a suitable number of LSBs
– Exampe: sh (store half) will store the 16LSB bits
– Top 16bits will be lost
Operations with immediate values
● MIPS ISA provide operations with immediate values
– That is values in the instruction itself
● addi → addi r1, r2, 8 (add 8 to r2 and store in r1)
● addiu → addiu r1, r2, 8 (do an unsigned addition of
r2 and 8 and store in r1)
● Other instructions: ori, andi, xori
– These do not have the u variant (they are anyway
bitwise operations)
Shift instructions
● MIPS provides instructions to shift values
● Synatax: sll r1, r1, 3 (shift logically left)
● Semantics: shitf the value in r1 by 3 possitions to the left
and store in r1
● Can be used for multipication by powers of two

● Syntax: sllv r1, r1, r2 (shift logically left value)


● Smantics: shift the value in r1 by given value in r2 and
store in r1
Other instructions
● We will not discuss about them
● They are there to handle exceptions
– Exception → something that should not have happen,
something unexpected by the code
– Example: a packet can through the network, someone
pulled put the USB stick ….
● ISA provides instructions to handling these conditions
● They are, in most cases privilleged → that is to say
not all applications can execute them.
Summary
● There are lots of instructions
– Would vary depending on the architecture
– No one will be able to remember all of them
● However the concepts would be the same
– Examlpe: jal in MIPS vs call in x86
● Use the given MIPS instruction sheet when you
need to refer to any instruction

You might also like