From what I understand:
ICR (Instruction Completion Rate): Is (# of instructions / time)
Instruction Throughput: Is usually an average of the number of instructions completed each clock cycle.
IPC (Instructions Per Clock): Is how many instructions are being completing each clock cycle. (Maybe this is usually an average?)
I'm confused on these definitions, I'm definitely looking for clarification. They might even be wrong, I've been having a tough time finding clear definitions of them.
How does the instruction completion rate affect overall performance of the processor?
How is Instruction Throughput affected compared to IPC?
Instruction throughput is typically used with respect to a specific type of instruction and is meant to provide instruction scheduling information in the context of structural hazards. For example, one might say "this fully pipelined multiplier has a latency of three cycles and an instruction throughput of one". Repeat rate is the inverse of throughput.
IPC describes performance per cycle, while your definition of instruction completion rate describes performance directly (independent of clock frequency).
(Of course, the performance value of "instruction" depends on the instruction set, the compiler, and the application — all of which influence the number of (and types of) instructions executed to complete a task. In addition, the relative performance of different instructions can depend on the hardware implementation; this can, in turn, drive compilation changes and sometimes application programming changes and even ISA changes.)
Related
I have an x86-64 Linux program which I am attempting to optimize via perf. The perf report shows the hottest instructions are scalar conversions from double to long with a memory argument, for example:
cvttsd2si (%rax, %rdi, 8), %rcx
which corresponds to C code like:
extern double *array;
long val = (long)array[idx];
(This is an unusual bottleneck but the code itself is very unusual.)
To inform optimizations I want to know if these instructions are hot because of the load from memory, or because of the arithmetic conversion itself. What's the best way to answer this question? What other data should I collect and how should I proceed to optimize this?
Some things I have looked at already. CPU counter results show 1.5% cache misses per instruction:
30686845493287 cache-references
2140314044614 cache-misses # 6.975 % of all cache refs
52970546 page-faults
1244774326560850 instructions
194784478522820 branches
2573746947618 branch-misses # 1.32% of all branches
52970546 faults
Top-down performance monitors show we are primarily backend-bound:
frontend bound retiring bad speculation backend bound
10.1% 25.9% 4.5% 59.5%
Ad-hoc measurement with top shows all CPUs pegged at 100% suggesting we are not waiting on memory.
A final note of interest: when run on AWS EC2, the code is dramatically slower (44%) on AMD vs Intel with the same core count. (Tested on Ice Lake 8375C vs EPYC 7R13). What could explain this discrepancy?
Thank you for any help.
To inform optimizations I want to know if these instructions are hot because of the load from memory, or because of the arithmetic conversion itself. What's the best way to answer this question?
I think there is two main reason for this instruction to be slow. 1. There is a dependency chain and the latency of this instruction is a problem since the processor is waiting on it to execute other instructions. 2. There is a cache miss (saturating the memory with such instruction is improbable unless many cores are doing memory-based operations).
First of all, tracking what is going on on a specific instruction is hard (especially if the instruction is not executed a lot of time). You need to use precise events to track the root of the problem, that is, events for which the exact instruction addresses that caused the event are available. Only a (small) subset of all events are precise one.
Regarding (1), the latency of the instruction should be about 12 cycles on both architecture (although it might be slightly more on the AMD processor, I do not expect a 44% difference). The target processor are able to execute multiple instruction at the same time in a given cycle. Instructions are executed on different port and are also pipelined. The port usage matters to understand what is going on. This means all the instruction in the hot loop matters. You cannot isolate this specific instruction. Modern processors are insanely complex so a basic analysis can be tricky. On Ice Lake processors, you can measure the average port usage with events like UOPS_DISPATCHED.PORT_XXX where XXX can be 0, 1, 2_3, 4_9, 5, 6, 7_8. Only the first three matters for this instruction. The EXE_ACTIVITY.XXX events may also be useful. You should check if a port is saturated and which one. AFAIK, none of these events are precise so you can only analyse a block of code (typically the hot loop). On Zen 3, the ports are FP23 and FP45. IDK what are the useful events on this architecture (I am not very familiar with it).
On Ice Lake, you can check the FRONTEND_RETIRED.LATENCY_GE_XXX events where XXX is a power of two integer (which should be precise one so you can see if this instruction is the one impacting the events). This help you to see whether the front-end or the back-end is the limiting factor.
Regarding (2), you can check the latency of the memory accesses as well as the number of L1/L2/L3 cache hits/misses. On Ice Lake, you can use events like MEM_LOAD_RETIRED.XXX where XXX can be for example L1_MISS L1_HIT, L2_MISS, L2_HIT, L3_MISS and L3_HIT. Still on Ice Lake, t may be useful to track the latency of the memory operation with MEM_TRANS_RETIRED.LOAD_LATENCY_GT_XXX where XXX is again a power of two integer.
You can also use LLVM-MCA to simulate the scheduling of the loop instruction statically on the target architecture (do not consider branches). This is very useful to understand deeply what the scheduler can do pretty easily.
What could explain this discrepancy?
The latency and reciprocal throughput should be about the same on the two platform or at least close. That being said, for the same core count, the two certainly do not operate at the same frequency. If this is not coming from that, then I doubt this instruction is actually the problem alone (tricky scheduling issues, wrong/inaccurate profiling results, etc.).
CPU counter results show 1.5% cache misses per instruction
The thing is the cache-misses event is certainly not very informative here. Indeed, it references the last-level cache (L3) misses. Thus, it does not give any information about the L1/L2 misses (previous events do).
how should I proceed to optimize this?
If the code is latency bound, the solution is to first break any dependency chain in this loop. Unrolling the loop dans rewriting it so to make it more SIMD-friendly can help a lot to improve performance (the reciprocal throughput of this instruction is about 1 cycle as opposed to 12 for the latency so there is a room for improvements in this case).
If the code is memory bound, they you should care about data locality. Data should fit in the L1 cache if possible. There are many tips to do so but it is hard to guide you without more context. This includes for example sorting data, reordering loop iterations, using smaller data types.
There are many possible source of weird unusual unexpected behaviours that can occurs. If such a thing happens, then it is nearly impossible to understand what is going on without the exact code executed. All details matter in this case.
My question primarily applies to firestorm/icestorm (because that's the hardware I have), but I am curious about what other representative arm cores do too. Arm has strange pre- and post-incremented addressing modes. If I have (for instance) two post-incremented loads from the same register, will the second depend on the first, or is the CPU smart enough to perform them in parallel?
AFAIK the exact behaviour of the M1 execution units is mainly undocumented. Still, there is certainly a dependency chain in this case. In fact, it would be very hard to break it and the design of modern processors make this even harder: the decoders, execution units, schedulers are distinct units and it would be insane to dynamically adapt the scheduling based on the instructions executed in parallel by execution units so to be able to break the chain in this particular case. Not to mention that instructions are pipelined and it generally takes few cycles for them to be committed. Furthermore, the time of the instructions is variable based on the fetched memory location. Finally, even this would be the case, the Firestorm documents does not mention such a feedback loop (see below for the links). Another possible solution for a processor to optimize such a pattern is to fuse the microinstructions so to combine the increment and add more parallelism but this is pretty complex to do for a relatively small improvement and there is no evidence showing Firestorm can do that so far (see here for more information about Firestorm instruction fusion/elimitation).
The M1 big cores (Apple's Firestorm) are designed to be massively parallel. They have 6 ALUs per core so they can execute a lot instructions in parallel on each core (possibly at the expense of a higher latency). However, this design tends to require a lot more transistors than current mainstream x86 Intel/AMD alternative (Alderlake/XX-Cove architecture put aside). Thus, the cores operate at a significantly lower frequency so to keep the energy consumption low. This means dependency chains are significantly more expensive on such an architecture compared to others unless there are enough independent instructions to be execute in parallel on the critical path. For more information about how CPUs works please thread Modern Microprocessors - A 90-Minute Guide!. For more information about the M1 processors and especially the Firestorm architecture, please read this deep analysis.
Note that Icestorm cores are designed to be energy efficient so they are far less parallel and thus having a dependency chain should be less critical on such a core. Still, having less dependency is often a good idea.
As for other ARM processors, recent core architecture are not as parallel as Firestorm. For example, the Cortex-A77 and Neoverse V1 have "only" 4 ALUs (which is already quite good). One need to also care about the latency of each instruction actually used in a given code. This information is available on the ARM website and AFAIK not yet published for Apple processors (one need to benchmark the instructions).
As for the pre VS post increment, I expect them to take the same time (same latency and throughput), especially on big cores like Firestorm (that try to reduce the latency of most frequent instruction at the expense of more transistors). However, the actual scheduling of the instruction for a given code can cause one to be slower than the other if the latency is not hidden by other instructions.
I received an answer to this on IRC: such usage will be fairly fast (makes sense when you consider it corresponds to typical looping patterns; good if the loop-carried dependency doesn't hurt too much), but it is still better to avoid it if possible, as it takes up rename bandwidth.
I think I have a decent understanding of the difference between latency and throughput, in general. However, the implications of latency on instruction throughput are unclear to me for Intel Intrinsics, particularly when using multiple intrinsic calls sequentially (or nearly sequentially).
For example, let's consider:
_mm_cmpestrc
This has a latency of 11, and a throughput of 7 on a Haswell processor. If I ran this instruction in a loop, would I get a continuous per cycle-output after 11 cycles? Since this would require 11 instructions to be running at a time, and since I have a throughput of 7, do I run out of "execution units"?
I am not sure how to use latency and throughput other than to get an impression of how long a single instruction will take relative to a different version of the code.
For a much more complete picture of CPU performance, see Agner Fog's microarchitecture guide and instruction tables. (Also his Optimizing C++ and Optimizing Assembly guides are excellent). See also other links in the x86 tag wiki, especially Intel's optimization manual.
See also
https://uops.info/ for accurate tables collected programmatically from microbenchmarks, so they're free from editing errors like Agner's tables sometimes have.
How many CPU cycles are needed for each assembly instruction?
and What considerations go into predicting latency for operations on modern superscalar processors and how can I calculate them by hand? for more details about using instruction-cost numbers.
What is the efficient way to count set bits at a position or lower? For an example of analyzing short sequences of asm in terms of front-end uops, back-end ports, and latency.
Modern Microprocessors: A 90-Minute Guide! very good intro to the basics of CPU pipelines and HW design constraints like power.
Latency and throughput for a single instruction are not actually enough to get a useful picture for a loop that uses a mix of vector instructions. Those numbers don't tell you which intrinsics (asm instructions) compete with each other for throughput resources (i.e. whether they need the same execution port or not). They're only sufficient for super-simple loops that e.g. load / do one thing / store, or e.g. sum an array with _mm_add_ps or _mm_add_epi32.
You can use multiple accumulators to get more instruction-level parallelism, but you're still only using one intrinsic so you do have enough information to see that e.g. CPUs before Skylake can only sustain a throughput of one _mm_add_ps per clock, while SKL can start two per clock cycle (reciprocal throughput of one per 0.5c). It can run ADDPS on both its fully-pipelined FMA execution units, instead of having a single dedicated FP-add unit, hence the better throughput but worse latency than Haswell (3c lat, one per 1c tput).
Since _mm_add_ps has a latency of 4 cycles on Skylake, that means 8 vector-FP add operations can be in flight at once. So you need 8 independent vector accumulators (which you add to each other at the end) to expose that much parallelism. (e.g. manually unroll your loop with 8 separate __m256 sum0, sum1, ... variables. Compiler-driven unrolling (compile with -funroll-loops -ffast-math) will often use the same register, but loop overhead wasn't the problem).
Those numbers also leave out the third major dimension of Intel CPU performance: fused-domain uop throughput. Most instructions decode to a single uop, but some decode to multiple uops. (Especially the SSE4.2 string instructions like the _mm_cmpestrc you mentioned: PCMPESTRI is 8 uops on Skylake). Even if there's no bottleneck on any specific execution port, you can still bottleneck on the frontend's ability to keep the out-of-order core fed with work to do. Intel Sandybridge-family CPUs can issue up to 4 fused-domain uops per clock, and in practice can often come close to that when other bottlenecks don't occur. (See Is performance reduced when executing loops whose uop count is not a multiple of processor width? for some interesting best-case frontend throughput tests for different loop sizes.) Since load/store instructions use different execution ports than ALU instructions, this can be the bottleneck when data is hot in L1 cache.
And unless you look at the compiler-generated asm, you won't know how many extra MOVDQA instructions the compiler had to use to copy data between registers, to work around the fact that without AVX, most instructions replace their first source register with the result. (i.e. destructive destination). You also won't know about loop overhead from any scalar operations in the loop.
I think I have a decent understanding of the difference between latency and throughput
Your guesses don't seem to make sense, so you're definitely missing something.
CPUs are pipelined, and so are the execution units inside them. A "fully pipelined" execution unit can start a new operation every cycle (throughput = one per clock)
(reciprocal) Throughput is how often an operation can start when no data dependencies force it to wait, e.g. one per 7 cycles for this instruction.
Latency is how long it takes for the results of one operation to be ready, and usually matters only when it's part of a loop-carried dependency chain.
If the next iteration of a loop operates independently from the previous, then out-of-order execution can "see" far enough ahead to find the instruction-level parallelism between two iterations and keep itself busy, bottlenecking only on throughput.
See also Latency bounds and throughput bounds for processors for operations that must occur in sequence for an example of a practice problem from CS:APP with a diagram of two dep chains, one also depending on results from the other.
I have disassembled a small C++ program compiled with MSVC v140 and am trying to estimate the cycles per instruction in order to better understand how code design impacts performance. I've been following Mike Acton's CppCon 2014 talk on "Data-Oriented Design and C++", specifically the portion I've linked to.
In it, he points out these lines:
movss 8(%rbx), %xmm1
movss 12(%rbx), %xmm0
He then claims that these 2 x 32-bit reads are probably on the same cache line therefore cost roughly ~200 cycles.
The Intel 64 and IA-32 Architectures Optimization Reference Manual has been a great resource, specifically "Appendix C - Instruction Latency and Throughput". However on page C-15 in "Table C-16. Streaming SIMD Extension Single-precision Floating-point Instructions" it states that movss is only 1 cycle (unless I'm understanding what latency means here wrong... if so, how do I read this thing?)
I know that a theoretical prediction of execution time will never be correct, but nevertheless this is important to learn. How are these two commands 200 cycles, and how can I learn to reason about execution time beyond this snippet?
I've started to read some things on CPU pipelining... maybe the majority of the cycles are being picked up there?
PS: I'm not interested in actually measuring hardware performance counters here. I'm just looking to learn how to reasonable sight read ASM and cycles.
As you already pointed out, the theoretical throughput and latency of a MOVSS instruction is at 1 cycle. You were looking at the right document (Intel Optimization Manual). Agner Fog (mentioned in the comments) measured the same numbers in his Intruction Tables for Intel CPUs (AMD is has a higher latency).
This leads us to the first problem: What specific microarchitecture are you investigating? This can make a big difference, even for the same vendor. Agner Fog reports that MOVSS has a 2-6cy latency on AMD Bulldozer depending on the source and destination (register vs memory). This is important to keep in mind when looking into performance of computer architectures.
The 200cy are most likely cache misses, as already pointed out be dwelch in the comments. The numbers you get from the Optimization Manual for any memory accessing instructions are all under the assumption that the data resides in the first level cache (L1). Now, if you have never touched the data by previous instructions the cache line (64 bytes with Intel and AMD x86) will need to be loaded from memory into the last level cache, form there into the second level cache, then into L1 and finally into the XMM register (within 1 cycle). Transfers between L3-L2 and L2-L1 have a throughput (not latency!) of two cycles per cache line on current Intel microarchitectures. And the memory bandwidth can be used to estimate the throughput between L3 and memory (e.g, a 2 GHz CPU with an achievable memory bandwidth of 40 GB/s will have a throughput of 3.2 cycles per cache line). Cache lines or memory blocks are typically the smallest unit caches and memory can operate on, they differ between microarchitectures and may even be different within the architecture, depending on the cache level (L1, L2 and so on).
Now this is all throughput and not latency, which will not help you estimate what you have described above. To verify this, you would need to execute the instructions over and over (for at least 1/10s) to get cycle accurate measurements. By changing the instructions you can decide if you want to measure for latency (by including dependencies between instructions) or throughput (by having the instructions input independent of the result of previous instructions). To measure for caches and memory accesses you would need to predict if an access is going to a cache or not, this can be done using layer conditions.
A tool to estimate instruction execution (both latency and throughput) for Intel CPUs is the Intel Architecture Code Analyzer, which supports multiple microarchitectures up to Haswell. The latency predictions are to be taken with the grain of salt, since it much harder to estimate latency than throughput.
I have the source code written and I want to measure efficiency as how many clock cycles it takes to complete a particular task. Where can I learn how many clock cycles different commands take? Does every command take the same amount of time on 8086?
RDTSC is the high-resolution clock fetch instruction.
Bear in mind that cache misses, context switches, instruction reordering and pipelining, and multicore contention can all interfere with the results.
Clock cycles and efficiency are not the same thing.
For efficiency of code you need to consider, in particular, how the memory is utalised, in particular the differing levels of the cache. Also important is the branching prediction of the code etc. You want a profiler that tells you these things, ideally one that gives you profile specific information: examples are CodeAnalyst for AMD chips.
To answer your question, particular base instructions do have a given (average) number of cycles (AMD release the approximate numbers for the basic maths functions in their maths library). These numbers are a poor place to start optimising code, however.