Superscalar and Out of Order execution, and Register Renaming

Superscalar history

So the 1980's roll on. Machines get deeper pipelines and better forwarding, until the machines usually initiate a new instruction every clock cycle. Eventually, even one cycle per instruction (CPI) is not fast enough. Sadly, one cycle per instruction is the theoretical limit of pipelining--it's the upper bound everybody's trying to reach by avoiding pipeline stalls, hazards, and flushes.

So as the 1990's arrived, designers began adding "wide" superscalar execution, where multiple instructions start every clock cycle. Nowadays, the metric is "instructions per cycle" (IPC), which can be substantially more than one. (Similarly, at some point travel went from gallons per mile to miles per gallon, which is good!)

To get multiple instructions running at once, you need different "execution units", so more arithmetic can execute at the same time. For example, 1994's PowerPC 601 could issue one integer, one floating-point, and one load/store instruction every clock cycle.

You can see superscalar effects in the instruction timings of any modern CPU--many instructions simply cost no time at all. For example, each these assembly programs take the exact same number of nanoseconds on my Pentium 4:

ret

(Try this in NetRun now!) -> Takes 4.4ns

mov eax,8
ret

(Try this in NetRun now!) -> Takes 4.4ns

mov ecx,7
mov eax,8
ret

(Try this in NetRun now!) -> Takes 4.4ns

mov ecx,7
mov eax,ecx
ret

(Try this in NetRun now!) -> Takes 4.4ns

mov ecx,7
add eax,ecx
ret

(Try this in NetRun now!) -> Takes 4.4ns

mov ecx,7
imul eax,ecx
ret

(Try this in NetRun now!) -> Takes 4.4ns

add ecx,7
imul eax,ecx
ret

(Try this in NetRun now!) -> Takes 4.4ns

Yet this one is actually slower by two clock cycles:

mov ecx,2
mov eax,3
mov edx,4
ret

(Try this in NetRun now!)

The limiting factor for this particular instruction sequence seems to just be the number of instructions--the Pentium 4's control unit can't handle three writes per clock cycle.

You can avoid the substantial call/return overhead and see a little more timing detail using a tight loop like this one:

mov ecx,1000
start:
add eax,1
dec ecx
jne start
ret

(Try this in NetRun now!)

This takes about two clocks to go around each time. You can even add one more "add" instructions without changing the loop time (ah, superscalar!). Because the Pentium 4 uses both rising and trailing edges of the clock, a 2.8GHz Pentium 4 (like the main NetRun machine) can take half-integer multiples of the clock period, 0.357ns. So your timings are 0.357ns for one clock, 0.536ns for 1.5 clocks (like the loop above without the "add"), 0.714ns for 2.0 clocks (like the loop above), 0.893ns for 2.5 clocks, etc.

Data Dependencies & Register Renaming

The biggest limit to superscalar execution, overall, is data dependencies: RAW, WAR, and WAW, also called "hazards." (Read it!)

RAR (read after read) is not a dependency, since reads can happen in any order.
WAW (write after write) is where I write a value after you write it. The machine just has to make sure the correct value ends up in the register, which is my value. They call this an "output dependency".
WAR (write after read) is where I write a value after you read it. The machine just has to make sure you get your value before I destroy it. They call this an "antidependency".
RAW (read after write) is the hard one: I read a value after you write it. This means I need your value, so I really have to wait for you. They call these "true dependencies".

You can eliminate the "fake" dependencies WAW and WAR using register renaming (also called Tomasulo's Algorithm): a typical implementation is to hand out a new renamed register for every write operation. Modern CPUs rename the architectural registers (like eax through edi) out to a huge set of *hundreds* of possible physical registers. Also, because most values are passed directly arithmetic-to-arithmetic (with register bypass), today a "register" is basically just an identifier that the arithmetic units use to communicate.

There are several interesting transformations you can use to expose and reduce dependencies, such as Software Pipelining. A naive application of these is unlikely to provide much benefit, since the compiler and CPU can do naive transformations already; but often you can exploit features of the algorithm to improve performance.

Why this matters for software

Here's the obvious way to compute the factorial of 12:

int i, fact=1;
for (i=1;i<=12;i++) {
	fact*=i;
}
return fact;
(Try this in NetRun now!)

Here's a modified version where we separately compute even and odd factorials, then multiply them at the end:

int i, factO=1, factE=1;
for (i=1;i<=12;i+=2) {
	factO*=i;
	factE*=i+1;
}
return factO*factE;
(Try this in NetRun now!)

This modification makes the code "superscalar friendly", so it's possible to execute the loop's multiply instructions simultaniously. Note that this isn't simply a loop unrolling, which gives a net loss of performance, it's a higher-level transformation to expose parallelism in the problem.

Hardware	Obvious	Superscalar	Savings	Discussion
Intel 486, 50MHz, 1991	5000ns	5400ns	-10%	Classic non-pipelined CPU: many clocks/instruction. The superscalar transform just makes the code slower, because the hardware isn't superscalar.
Intel Pentium III, 1133MHz, 2002	59.6ns	50.1ns	+16%	Pipelined CPU: the integer unit is fully pipelined, so we get one instruction per clock cycle. The P3 is also weakly superscalar, but the benefit is small.
Intel Pentium 4, 2.8Ghz, 2005	22.6ns	15.0ns	+33%	Virtually all of the improvement here is due to the P4's much higher clock rate.
Intel Q6600, 2.4GHz, 2008	16.7ns	9.4ns	+43%	Lower clock rate, but fewer pipeline stages leads to better overall performance.
Intel Sandy Bridge i5-2400 3.1Ghz 2011	11.8ns	5.3ns	+55%	Higher clock rate and better tuned superscalar execution. Superscalar transform gives a substantial benefit--with everything else getting faster, the remaining dependencies become more and more important.

The Future

It's really interesting to watch the evolution of modern microarchitectures:

The Intel 8008, from 1972, has a single shared internal bus used for everything: instruction fetch, execution, memory accesses. This means it takes many clock cycles per instruction, to interleave access to the bus.
The Intel 386, from the 1980's, added pipelining. The big difference is you now have many separate busses across the machine, and need to fetch instructions in big blocks (otherwise fetch interferes with execution memory access).
The Intel Pentium from early 1990 added superscalar execution, so now there are multiple arithmetic units and a dependency-checking control unit.
A recent CPU like the Intel Core2 from 2000's has renamed registers and a reorder buffer, as described below. This is about as far as you can take performance without changing the software.
Since about 2005 we've had several separate processor cores per chip, which requires multicore software.
Future CPU/GPU hybrids, like AMD's 2011 "APU A8", might combine several CPU cores with a serious graphics processor, to tackle both sequential and parallel programs.

Recently, CPU designers have been shedding their most complex features in favor of higher parallelism, so the machines of the future may look like the machines of the past!