write-back/through
cache address separation

Q24.2 - PA size
Q26.1 - virtual address size
memory segment

kernel vs user mode

kinds of exceptions
exception/signal handlers
setjmp/longjmp (difference between)

volatility of storage
	volatile storage loses its data on (sustained) power loss


pipeline diagrams with Bubble/Stall 
	with registers
	and mispredictions

optimizing code

	for i
		for j
			a[i,j] = b[j,i]
	
	for ii by 16
		for jj by 16
			for i from ii to ii+16
				for j from jj to jj+16
					a[i,j] = b[j,i]

	suppose sizeof(a[0]) = 12
	suppose sizeof(b[0]) = 12
	suppose L1 cache is 64KB
	
	12 * blocksize^2 * 2 < L1 size
	block size <= floor(sqrt(64KB / 24))


Multiple accumulators
	allow instruction-level parallelism
	- Func Unit (issue / capacity) > (latency)
	- have access to operands fast enough to exercise parallelism

sum = 0
for i from 1 to 100000
	sum += i

sum1 = 0
sum2 = 0
sum3 = 0
sum4 = 0
for i from 1 to 100000 by 4
	sum1 += i
	sum2 += i+1
	sum3 += i+2
	sum4 += i+3
sum = sum1+sum2+sum3+sum4



if(x)     if(x != 0)



for(i = 0; ; ) -- ANSI C
for(int i = 0; ; ) -- not ANSI C