STEVEN R. BAGLEY ARM: LOOPS AND ADDRESSING

Transcription

1 STEVEN R. BAGLEY ARM: LOOPS AND ADDRESSING

2 FROM C TO ASSEMBLER Plan your assembler programs in C first Much easier to get the logic right in C using ifs, whiles etc. Can then convert those high-level structures into machine code Already seen how to convert if statements At least to start with Difference between the logic behind a program and the syntax of the language

3 IF IN ASSEMBLER Do a CMP to test for the condition Branch if the condition is not met to skip over the code inside the if block Code is then executed only if the condition is met Otherwise we skip over it

4 if EXAMPLES C ASSEMBLER

5 if EXAMPLES if(r0 == 10) { C ASSEMBLER CMP R0, #10 BNE skip skip

6 if EXAMPLES if(r0 == 10) { if(r0 < R1) { C ASSEMBLER CMP R0, #10 BNE skip skip CMP R0, R1 BGE skip skip

7 if EXAMPLES if(r0 == 10) { if(r0 < R1) { if(r0 >= 10) { C ASSEMBLER CMP R0, #10 BNE skip skip CMP R0, R1 BGE skip skip CMP R0, #10 BLT skip skip

8 if EXAMPLES if(r0 == 10) { if(r0 < R1) { if(r0 >= 10) { if(r0!= 0) { C ASSEMBLER CMP R0, #10 BNE skip skip CMP R0, R1 BGE skip skip CMP R0, #10 BLT skip skip CMP R0, #10 BEQ skip skip

9 B main menu coke DEFB G51SYS Vending Machine \n\0 DEFB Have a bottle of Coke\n\0 ALIGN main SWI 1 ; Read a character CMP R0, #49 BNE skip ADR R0, coke SWI 3 skip SWI 2 Beginning of the Coke vending machine from G51PRG label name for skip unimportant but it needs to be unique

10 B main menu coke DEFB G51SYS Vending Machine \n\0 DEFB Have a bottle of Coke\n\0 ALIGN main SWI 1 ; Read a character CMP R0, #49 BNE skip ADR R0, coke SWI 3 skip SWI 2 Beginning of the Coke vending machine from G51PRG label name for skip unimportant but it needs to be unique

11 B main menu coke DEFB G51SYS Vending Machine \n\0 DEFB Have a bottle of Coke\n\0 ALIGN main SWI 1 ; Read a character CMP R0, #49 BNE skip ADR R0, coke SWI 3 skip SWI 2 Beginning of the Coke vending machine from G51PRG label name for skip unimportant but it needs to be unique

12 OR ELSE How do we handle the else clause? Already doing it If it s not true we are branching to skip Can put the else section there But the code that is executed when the condition is true will also get executed

13 OR ELSE But we can easily get around that Put a branch at the end of the true block to skip the else section This would be an unconditional branch

14 B main menu coke error DEFB G51SYS Vending Machine \n\0 DEFB Have a bottle of Coke\n\0 DEFB Incorrect option\n\0 ALIGN main SWI 1 ; Read a character CMP R0, #49 BNE skip ADR R0, coke SWI 3 B end skip ADR R0, error SWI 3 end SWI 2

15 B main menu coke error DEFB G51SYS Vending Machine \n\0 DEFB Have a bottle of Coke\n\0 DEFB Incorrect option\n\0 ALIGN main SWI 1 ; Read a character CMP R0, #49 BNE skip ADR R0, coke SWI 3 B end skip ADR R0, error SWI 3 end SWI 2

16 B main menu coke error DEFB G51SYS Vending Machine \n\0 DEFB Have a bottle of Coke\n\0 DEFB Incorrect option\n\0 ALIGN main SWI 1 ; Read a character CMP R0, #49 BNE skip ADR R0, coke SWI 3 B end skip ADR R0, error SWI 3 end SWI 2

17 COMBINED CONDITIONALS Should be possible to see how we can do combined conditionals too Just use multiple CMP/Bxx pairs Can also do lazy evaluation as with C But need to watch our branch conditions

18 AND CONDITIONALS if(r0 == 1 && R1 == 2) { printf("have \n ); CMP R0, #1 BNE skip CMP R1, #2 BNE skip ADR R0, mesg SWI 3 skip SWI 2 Could optimise this a lot though Highlighted code is only executed when r0 == 1 and r1 == 2 Note how both branches use the same condition code If it s ever false skip it

19 AND CONDITIONALS if(r0 == 1 && R1 == 2) { printf("have \n ); CMP R0, #1 BNE skip CMP R1, #2 BNE skip ADR R0, mesg SWI 3 skip SWI 2 Could optimise this a lot though Highlighted code is only executed when r0 == 1 and r1 == 2 Note how both branches use the same condition code If it s ever false skip it

20 AND CONDITIONALS if(r0 == 1 && R1 == 2) { printf("have \n ); CMP R0, #1 BNE skip CMP R1, #2 BNE skip ADR R0, mesg SWI 3 skip SWI 2 Could optimise this a lot though Highlighted code is only executed when r0 == 1 and r1 == 2 Note how both branches use the same condition code If it s ever false skip it

21 AND CONDITIONALS if(r0 == 1 && R1 == 2) { printf("have \n ); CMP R0, #1 BNE skip CMP R1, #2 BNE skip ADR R0, mesg SWI 3 skip SWI 2 Could optimise this a lot though Highlighted code is only executed when r0 == 1 and r1 == 2 Note how both branches use the same condition code If it s ever false skip it

22 OR COMBINATION if(r0 == 1 R1 == 2) { printf("have \n ); CMP R0, #1 BEQ code CMP R1, #2 BNE skip code ADR R0, mesg SWI 3 skip SWI 2 Could optimise this a lot though Highlighted code is only executed when r0 == 1 or r1 == 2 Note how the first branch branches if equal, the second if not equal Again it s lazy evaluation if the first test is true, then we don t need to do the second test so can just jump into the code

23 OR COMBINATION if(r0 == 1 R1 == 2) { printf("have \n ); CMP R0, #1 BEQ code CMP R1, #2 BNE skip code ADR R0, mesg SWI 3 skip SWI 2 Could optimise this a lot though Highlighted code is only executed when r0 == 1 or r1 == 2 Note how the first branch branches if equal, the second if not equal Again it s lazy evaluation if the first test is true, then we don t need to do the second test so can just jump into the code

24 OR COMBINATION if(r0 == 1 R1 == 2) { printf("have \n ); CMP R0, #1 BEQ code CMP R1, #2 BNE skip code ADR R0, mesg SWI 3 skip SWI 2 Could optimise this a lot though Highlighted code is only executed when r0 == 1 or r1 == 2 Note how the first branch branches if equal, the second if not equal Again it s lazy evaluation if the first test is true, then we don t need to do the second test so can just jump into the code

25 OR COMBINATION if(r0 == 1 R1 == 2) { printf("have \n ); CMP R0, #1 BEQ code CMP R1, #2 BNE skip code ADR R0, mesg SWI 3 skip SWI 2 Could optimise this a lot though Highlighted code is only executed when r0 == 1 or r1 == 2 Note how the first branch branches if equal, the second if not equal Again it s lazy evaluation if the first test is true, then we don t need to do the second test so can just jump into the code

26 LOOPS Important building blocks in programs while loops repeat a block of code whilst the condition holds But again, like if, ARM assembler does not provide a while loop Need to manufacture it out of simple components In ARM this is the B instruction In C, there is an equivalent in the goto instruction

27 LOOPS USING GOTO We can translate our while loop to using gotos Bad programming practice normally But a good halfway stage between structured programming and assembler Remember though that a while loop can execute zero or more times Run through an example in C Show that they both run and produce the same result.

28 while LOOP while(i < 8) { j = j + 3; i = i + 1; Note how we ve turned the structure of the while loop upside down We ve put the test at the end There s reasons for this madness Can then convert this very easily into machine code ASM version assumes i is in R1 and j is in R0

29 while LOOP while(i < 8) { j = j + 3; i = i + 1; goto while_cond; while_loop: j = j + 3; i = i + 1; while_cond: if(i < 8) goto while_loop; Note how we ve turned the structure of the while loop upside down We ve put the test at the end There s reasons for this madness Can then convert this very easily into machine code ASM version assumes i is in R1 and j is in R0

30 while LOOP goto while_cond; while_loop: j = j + 3; i = i + 1; while_cond: if(i < 8) goto while_loop; B while_cond while_loop ADD R0, R0, #3 ADD R1, R1, #1 while_cond CMP R1, #8 BLT while_loop Note how we ve turned the structure of the while loop upside down We ve put the test at the end There s reasons for this madness Can then convert this very easily into machine code ASM version assumes i is in R1 and j is in R0

31 WHILE LOOP Putting the conditional at the end has several advantages Don t have to invert the conditional (as we did with if) Will actually execute less instructions Code for a do while() loop is near identical

32 WHILE LOOP B while_cond while_loop ADD R0, R0, #3 ADD R1, R1, #1 while_cond CMP R1, #8 BLT while_loop while_cond CMP R1, #8 BGE end ADD R0, R0, #3 ADD R1, R1, #1 B while_cond end: ASM version assumes i is in R1 and j is in R0 Uses 4 instructions per loop iteration + 3 to start it off (35) second method takes 5 instructions per loop +2 extra (42) Take a look and see what the compiler does too

33 WORKED EXAMPLE Going to work through an example bit of code now To calculate the Highest Common Factor (HCF) Use Euclid s algorithm

34 EUCLID S HCF ALGORITHM Euclid was a famous Greek mathematician (~300BC) Devised a simple algorithm for finding the Highest Common Factor of two integers while(a!= b) { if(a > b) a = a - b; else b = b - a; Let s convert this into assembler

35 LOOPS Firstly, we need to remove the while loop Don t have such luxuries in assembly Only have branches, the equivalent of C s goto Rewrite our C program to use ifs and goto instead Eek GOTO alert Go and do this, show the program works

36 REGISTERS Now we can start assigning the variables to registers Small number here, so we ll use R0 to be equivalent to a, and R1 to be equivalent to b Actual registers used doesn t matter (more or less) If you run out, you can always store the value in memory and the load it back into a register later Okay now start developing an ARM version alongside the C version split screen then test in Komodo

37 FOR LOOPS Again, we can model a for loop in assembler Already seen how to do this in G51PGA Saw that there is an equivalent while loop for every for loop So when we want to use a for loop we can convert it to a while loop Then convert the while loop to assembler as before

38 for(expr1; expr2; expr3) statement; expr1; while(expr2) { statement; expr3;

39 for(expr1; expr2; expr3) statement; expr1; while(expr2) { statement; expr3; These are equivalent

40 for LOOP for(i = 0; i < 10; i++) { printf( %d\n, i); Note how we ve turned the structure of the while loop upside down We ve put the test at the end There s reasons for this madness Can then convert this very easily into machine code ASM version assumes i is in R1 and j is in R0

41 for LOOP for(i = 0; i < 10; i++) { printf( %d\n, i); i = 0; while(i < 10) { printf( %d\n, i); i++; Note how we ve turned the structure of the while loop upside down We ve put the test at the end There s reasons for this madness Can then convert this very easily into machine code ASM version assumes i is in R1 and j is in R0

42 for LOOP i = 0; while(i < 10) { printf( %d\n, i); i++; Note how we ve turned the structure of the while loop upside down We ve put the test at the end There s reasons for this madness Can then convert this very easily into machine code ASM version assumes i is in R1 and j is in R0

43 for LOOP i = 0; while(i < 10) { printf( %d\n, i); i++; i = 0; goto while_cond; while_loop: printf( %d\n, i); i++; while_cond: if(i < 10) goto while_loop; Note how we ve turned the structure of the while loop upside down We ve put the test at the end There s reasons for this madness Can then convert this very easily into machine code ASM version assumes i is in R1 and j is in R0

44 ADDRESSING Already seen how we can load and store values from memory LDR R0, _a STR R1, _b But this can only load from a fixed address, and the same address every time Okay for global variables, wouldn t work for an array This is known as Absolute Addressing Or at least it would be on most systems

45 ADDRESSING ON ARM Like most CPUs, ARM provides support for various addressing modes for accessing memory At their heart, they are all just mechanisms for automatically calculating the address to be accessed But provide very convenient mechanisms to handle the calculation Unlike most other CPUs, ARM doesn t actually have an Absolute Addressing mode

46 HANDLING LARGE ADDRESSES ARM encodes all instructions in one 32-bit word This encodes the instruction and also what it operates on (registers, immediate values, etc) Different to a CISC chip which will use multiple words to encode the data Much faster to hide numeric constants and data addresses in the instruction themselves

47 ADDRESSING ON ARM But the addresses are 32-bits wide How does ARM hide them inside the the 32-bit instruction when there are only 8- or 12-bits spare? Uses a scheme that generates a useful subset of all the 32-bit values Already seen this with immediate values in MOV instructions amongst others

48 IMMEDIATE VALUE Encoded as an 8-bit value (0-255), and a 4-bit rotation value Rotation shifts the bits around the 32-bits of the register If value cannot be expressed using the above, we can always just load it from memory

49 PSEUDO-INSTRUCTION Assembler recognises a pseudo-instruction which will automatically select whether to use a MOV or to load from memory LDR R1, =0x Gets compiled to either a MOV instruction (if it will fit), or loads it from memory using LDR Assembler designates part of the program a literal pool to store the constants Literal pool must be within ±4KB of the PC But assembler handles it (go demo) Pseudo-instruction because it doesn t exist assembler generates a real instruction based on the program s context

50 USING THE PC TO HELP Rotation isn t the only useful trick in generating a large constant (often intended as an address) Time to think about ADR in more detail Again, ADR is not an ARM instruction, but an assembler pseudo-instruction Assumes that the address to be loaded into a register is near the PC Look at ADR, then see how a similar trick is used by LDR/STR

51 ADR INSTRUCTION If address is in range, then destination register can be set using a singleinstruction by adding/subtracting an offset to the value in PC Offset must be expressible in 8-bits with rotation So ADR R0,one might become SUB R0, PC, #&18 or ADD R0, PC, #&80 Depends on the relationship between the label and the instruction s location

52 ADRL Not all address can be generated by this approach The offset must be expressible as a rotation of 8-bits Also have an ADRL pseudo-instruction This will use two (or more) instructions to generate the address Again usually ADD or SUB instructions Given an example use origin to demo it

53 ADRL IN AASM With aasm, ADRL will only generate a single instruction if possible Otherwise it may need two or more instructions Other assemblers may always generate two instructions The combination of PC-relative addressing with rotated constants is very powerful Can be important when writing code to know exactly how many instructions are executed

54 ORIGIN 0x0100 alpha DEFW 20 ORIGIN 0x1000 beta DEFW 30 ORIGIN 0x1010 ENTRY ADRL R0, alpha ADRL R0, beta ADRL R0, gamma E24F0FC6 SUB R0 PC, #& E2400B03 SUB R0, R0, #&C E24F0020 SUB R0, PC, #& C E28F0EED ADD R0, PC, #&ED E2800A02 ADD R0, R0, #&2000 ORIGIN 0x3EF4 gamma DEFW 40 Various examples of ADRL in action in aasm

55 POSITION INDEPENDENCE One other advantage of writing code like this is it is position-independent That is, providing all the access to addresses are relative to the PC then it doesn t matter where it is loaded in memory As soon as you hard-code an address (e.g. in a variable, or literal pool) then this breaks Doesn t mean the code cannot be loaded anywhere but the thing loading it will need to relocate it to the new memory address