bool testit( void * p ) {
return (3 & (intptr_t)p) != 0;
}
TESTIT:___________________________
TRUNC R3,2
__________________________________
BZS DONE
__________________________________
LIS R3,1
__________________________________
DONE:
__________________________________
JUMPS R1
__________________________________
|
• returns true if p is not word aligned
_________________________________________________
8 got this, 5 left it blank; the remainder gave answers indistinguishable from nonsense.
1 got this, 4 forgot to limit the return value to true or false, 4 others forgot to return from the function. 15 left it blank; the remainder gave answers indistinguishable from nonsense.
• PC relative addressing
_________________________________________________
40 got this, 2 left it blank; the remainder gave answers indistinguishable from nonsense.
• This requires TESTIT be local, EXT makes it non-local
_________________________________________________
8 got this, 5 were vague but probably correct, 2 were more confused. 5 left it blank; the remainder gave answers indistinguishable from nonsense.
__________________________________
MOVE R5,R6
__________________________________
SR R5,5
__________________________________
|
(Answer in the space to the right.)
4 got this, 6 used SRU instead of SR, 17 used left shifts instead of right shifts, one invented MOVESR (the Hawk only has MOVESL) and a few made various misspellings. 3 left it blank; the remainder gave answers indistinguishable from nonsense.
LEA R4,R2,A
_________________________________________________
6 got this, 15 omitted R2, 2 used LOAD instead of LEA, 2 used STORE. 3 left it blank; the rest gave answers indistinguishable from nonsense. The number of students who missed this elementary problem makes it clear that many in the class never took the programming assignments seriously.
MOVE R3,R5
ADDSL R3,R4,2
_________________________________________________
6 got this, 3 solved it for an array of bytes (with just one instruction, ADD), 3 more made small clerical errors. 3 left it blank; the remainder gave answers indistinguishable from nonsense. (The exam should have contained an extra blank line for the answer.)
• CMPI is a long instruction, TESTR is short
• They also set the C condition code differently
_________________________________________________
3 clearly noted the long-short difference, 7 noted it vaguely. 3 othrs noted the condition code difference. 4 left it blank; the remainder gave answers indistinguishable from nonsense.
STORE R9,R2,R9SAVE in the receiving sequence
_________________________________________________
LOAD R9,R2,R9SAVE in the return sequence
_________________________________________________
11 got this, 8 only mentioned the return sequence, 2 only mentioned the receiving sequence. 4 left it blank; the remainder gave answers indistinguishable from nonsense.
__________________________________
LIL R5,X
__________________________________
LOADS R5,R5
__________________________________
|
10 got this, 11 only mentioned the first line (loading the address of the variable. 12 only gave R5,X without giving the opcode. 3 left it blank; the remainder gave answers indistinguishable from nonsense. Again, this is the kind of elementary problem that any student who took the programming assignments seriously should have been able to answer with no difficulty.
• The result sign will be wrong if there was an overflow.
_________________________________________________
16 got this, 1 was vague but not wrong. 4 left it blank; the remainder gave answers indistinguishable from nonsense.
LOAD R3,R2,A
__________________________________
LIS R4,5
__________________________________
STORE R4,R3,B
__________________________________
|
2 got this, 20 got partial credit; many messed up the final store by imagining that there was a way to push a constant into memory without first loading a register. Another common error was to assume that the address of a was already in a register. 11 left it blank; the remainder gave answers indistinguishable from nonsense.
• a + ~b + 1
_________________________________________________
7 got this, 18 forgot the final + 1. 6 left it blank; the remainder gave answers indistinguishable from nonsense.
• R3 ≥ R4
_________________________________________________
An easy way to work this out is to work three little examples:
1 - 0 = 1 + ~0 + 1 = 0001 + 1111 + 1 which definitely sets C
0 - 0 = 0 + ~0 + 1 = 0000 + 1111 + 1 which definitely sets C
0 - 1 = 0 + ~1 + 1 = 0000 + 1110 + 1 which does not set C
6 got this, 5 said R3 = R4, 3 said R3 > R4. 3 left it blank; the remainder gave answers indistinguishable from nonsense.
MOVESL R5,R4,12
SL R5,12
SRU R3,8
SR R4,8
OR R3,R5
|
• v = v >> 8
_________________________________________________
5 got this, 2 suggested rotate, 2 suggested left shift, 1 suggested an unsigned shift, 2 said it shrinks v to 48 bits, 3 only focused on the effect on the least significant halfword. 5 left it blank; the remainder gave (frequently very long) answers indistinguishable from nonsense.
• indexed
_________________________________________________
• The transfer from main program to MP8 is by trap
_________________________________________________
6 got this, one said by exception. 7 left it blank; the remainder gave answers that suggested that they might not have put much effort into MP8.
OTHER
LCSAVE = .
. = LOCATION
SOME
CODE
. = LCSAVE
STUFF
|
• at and after LOCATION
_________________________________________________
18 got this. 2 left it blank; the remainder gave answers indistinguishable from nonsense. This suggests that many students saw this construction as boilerplate that they could use without understanding.
• none
_________________________________________________
8 got this. 2 left it blank; the remainder gave answers indistinguishable from nonsense; in many cases, they gave explicit and totally unjustified numbers such as 8, 10 or 12.
• macros are faster
_________________________________________________
14 got this. 4 left it blanik, while the remainder said, with great confidence, that exception handlers are faster.
Note that those who put effort into MP8 would have noticed that exception handlers include blocks of code to save and restore the processor state and to decode the opcode the user tried to use, obviously slowing down execution considerably as part of the rpice of virtualization.
LIS CMPI JSR MOVESL COSET
3 got this. 3 thought LIS was long despite the fact that the trailing S stands for short. 37 thought CMPI was short. 23 thought JSR was short. 38 thought MOVESL was long. 25 thought COSET was long.
Difficulties on this problem are evidence that many students forget about
instruction encoding and learn assembly language as if it were a (wretched)
higher level language with no connection to underlying hardware.
| a | b | c | d |
|---|---|---|---|
| 0 | 0 | 1 | 0 |
| ____ | ____ | ____ | ____ |
| 0 | 1 | 0 | 0 |
| ____ | ____ | ____ | ____ |
| 1 | 0 | 1 | 0 |
| ____ | ____ | ____ | ____ |
| 1 | 0 | 1 | 1 |
| ____ | ____ | ____ | ____ |
20 got each of the above. 15 lost credit for giving nonstandard order for the table rows (as indicated by the contents of the a and b columns. Only 3 earned no credit, and nobody left the truth table blank.
• b = 1 implies f = a
_________________________________________________
6 got this, 4 got partial credit for identifying input b. 5 left it blank; the remainder gave answers indistinguishable from nonsense.
• it is a type D latch
_________________________________________________
2 got this, 6 got partial credit for identifying it as some kind of flipflop.
4 left it blank; the remainder gave answers indistinguishable from nonsense.
GETCHAR:
LIW R4,KEYBOARD
GETCLP:
LOAD R3,R4,KBSTAT
BITTST R3,KBREADY
BBR GETCLP
LOAD R3,R4,KBDATA
JUMPS R1
|
• a polling loop
_________________________________________________
3 got this. 18 said it was a while loop, for partial credit.
4 left it blank; most of the remainder gave answers indistinguishable from
nonsense. Some said it was a for loop, but that is nonsense because
there is nothing in the code resembling incrementing or testing a loop control
variable.
• device registers
_________________________________________________
1 got this; for partial credit, 4 said registers, 1 said I/O. Only 1 left it blank; the remainder gave answers indistinguishable from nonsense.
__________________________________
LOAD R5,R3,NEXT
__________________________________
STORE R5,R4,NEXT
__________________________________
STORE R4,R3,NEXT
__________________________________
|
5 got this. There was very little partial credit. 14 left it blank; the remainder gave answers indistinguishable from nonsense. Many gave answers using only LOAD instructions, therefore making no changes at all to the data structure, or even more bizarrely, using arithmetic or logic operations.
•THERE is too far away
_________________________________________________
14 got this, 2 said vaguely that something was too big, 2 implied a range from 0 to 255 instead of -128 to +127. 3 left it blank; the remainder gave answers indistinguishable from nonsense.
• JUMP THERE
_________________________________________________
12 got this. 4 used JSR, 2 added an index register,
1 omitted the label THERE.
6 left it blank; the remainder gave answers indistinguishable from nonsense.
Among the nonsense, a number offered the idea of replacing BR with
BGT or some other conditional branch.
BEQ SKIP
__________________________________
JUMP THERE
__________________________________
SKIP:
|
2 got this, 3 used BNE instead of BEQ,
2 put the label before the JUMP,.
15 left it blank; the remainder gave answers indistinguishable from nonsense.
Again, many replaced BNE with random conditional branches that tested
other condition codes.
• FF0016
_________________________________________________
3 got this, 13 gave 0000 (wrong displacement), 8 gave 00XX (byte order),
7 gave 00 (just the opcode without displacement).
11 left it blank; the remainder gave answers indistinguishable from nonsense.
F0 __________________________________ 30 __________________________________ FC __________________________________ FF __________________________________ |
None got this fully. 11 earned partial credit; common errors included
6 forgot to give a displacement halfword and
3 who used an index register other than PC (R0).
15 left it blank; the remainder gave answers indistinguishable from nonsense;
for example, giving assembly code or a single hex digit per byte.
• –10.0
_________________________________________________
9 got this, 5 earned partial credit where it was possible to identify a connection between their answer and the problem given. 13 left it blank; the remainder gave answers indistinguishable from nonsense.
C120000016 = 110000010010 ...
sign = 1 meaning negative
exponent = 100000010 meaning +3 (recall that 01111111 = 0)
mantissa = (1).010 ... (the hidden bit is in parentheses)
so the absolute value is 1.010 ... * 23 = 1010.02
= 10.010
• Virtuaized instructions could not reference
_________________________________________________
the user's values of R7 to R15
_________________________________________________
6 got this, 4 earned partial credit for vague hints at the answer. 4 left it blank; the remainder gave answers indistinguishable from nonsense. Many wrong answers referenced Hawk calling conventions, which do not apply to trap service routine entry.
The optimization suggested here would actually
be quite appropriate for speeding up any interrupt or trap handler that did not
need to access user registers the way a virtualized instruction trap handler
does.
TBIT R4,0
BBR BIT0
SL R3,1
BIT0:
TBIT R4,1
BBR BIT1
SL R3,2
BIT1:
TBIT R4,2
BBR BIT2
SL R3,4
BIT2:
TBIT R4,3
BBR BIT3
SL R3,8
BIT3:
TBIT R4,4
BBR BIT4
SL R3,16
BIT4:
|
ADDSR R0,R5,1
_________________________________________________
2 got this. 11 identified my mistake, saying BITTST R5,1, but did not notice that BITTST is a derived instruction. 5 forgot to give the operands, just giving the opcode. 9 left it blank; the remainder gave answers indistinguishable from nonsense.
This code was on midterm II, and a debugged version of it was incorporated
into the solutions to MP6 and MP7.
BCR BIT0
_________________________________________________
10 got this, 4 left off the operand, 2 gave BCS.
8 left it blank; the remainder gave answers indistinguishable from nonsense.
• n = 2i
_________________________________________________
14 got this, 2 earned partial credit. 3 left it blank; the remainder gave answers indistinguishable from nonsense. The nonsense was truly odd, including many mathematical expressions that did not relate n to i or that gave relationships that had no connection to the trivial relationship in the given code.
Again, it is worth emphasizing that this code was on midterm II, and a debugged version of it was incorporated into the solutions to MP6 and MP7.
_________________________________________________
SLU R3,1
_________________________________________________
SR R3,1
_________________________________________________
Alternatively (longer code so a bit slower and using more registers):
LIW R4,#7FFFFFFF
AND R3,R4
8 had good solutions, 12 earned partial credit with errors such as using OR instead of and, using SR instead of SRU, switching the order of the shifts (to zero out the low bit instead of the high bit), or worse, trying to use integer NEG on a floating value. 15 left it blank; the remainder gave answers indistinguishable from nonsense.