Determine whether memory location is in CPU cache - memory

It is possible for an operating system to determine whether a page of memory is in DRAM or in swap; for example, simply try to access it and if a page fault occurs, it wasn't.
However, is the same thing possible with CPU cache?
Is there any efficient way to tell whether a given memory location has been loaded into a cache line, or to know when it does so?

In general, I don't think this is possible. It works for DRAM and the pagefile since that is an OS managed resource, cache is managed by the CPU itself.
The OS could do a tight timing loop of a memory read and try to see if it completes fast enough to be in the cache or if it had to go out to main memory - this would be very error prone.
On multi-core/multi-proc systems, there are cache coherency protocols that are used between processors to determine when to they need to invalidate each other's caches, I suppose you could have a custom device that would snoop this protocol that the OS would query.
What are you trying to do? If you want to force something into memory, current x86 processors support prefetching memory into the cache in a non-blocking way, for instance with Visual C++ you could use _mm_prefetch to fetch a line into the cache.
EDIT:
I haven't done this myself, so use at your own risk. To determine cache misses for profiling, you may be able to use some architecture-specific registers. http://download.intel.com/design/processor/manuals/253669.pdf, Appendix A gives "Performance Tuning Events". This can't be used to determine if an individual address is in the cache or when it is loaded in the cache, but can be used for overall stats. I believe this is what vTune (a phenomenal profiler for this level) uses.

If you try to determine this yourself then the very act of running your program could invalidate the relevant cache lines, hence rendering your measurements useless.
This is one of those cases that mirrors the scientific principle that you cannot measure something without affecting that which you are measuring.

X86
dont know how to tell if address IS in cache
BUT here is how to tell if address WAS in cache
rdtsc
save timestamp
mov eax,address
rdtsc read timestamp counter
calculate timestamp difference
if < threshold then was in cache
threshold has to be determined from documentation or empirically
some machines have cache hit/miss counters which would serve equally well

Related

Is memory outside each core always conceptually flat/uniform/synchronous in a multiprocessor system?

Multi processor systems perform "real" memory operations (those that influence definitive executions, not just speculative execution) out of order and asynchronously as waiting for global synchronization of global state would needlessly stall all executions nearly all the time. On the other hand, immediately outside each individual core, it seems that the memory system, starting with L1 cache, is purely synchronous, consistent, flat from the allowed behavior point of view (allowed semantics); obviously timing depends on the cache size and behavior.
So on a CPU there on one extreme are named "registers" which are private by definition, and on the other extreme there is memory which is shared; it seems a shame that outside the minuscule space of registers, which have peculiar naming or addressing mode, the memory is always global, shared and globally synchronous, and effectively entirely subject to all fences, even if it's memory used as unnamed registers, for the purpose of storing more data than would fit in the few registers, without a possibility of being examined by other threads (except by debugging with ptrace which obviously stalls, halts, serializes and stores the complete observable state of an execution).
Is that always the case on modern computers (modern = those that can reasonably support C++ and Java)?
Why doesn't the dedicated L1 cache provide register-like semantics for those memory units that are only used by a particular core? The cache must track which memory is shared, no matter what. Memory operations on such local data doesn't have to be stalled when strict global ordering of memory operations are needed, as no other core is observing it, and the cache has the power to stall such external accesses if needed. The cache would just have to know which memory units are private (non globally readable) until a stall of out of order operations, which makes then consistent (the cache would probably need a way to ask the core to serialize operations and publish a consistent state in memory).
Do all CPU stall and synchronize all memory accesses on a fence or synchronizing operation?
Can the memory be used as an almost infinite register resource not subject to fencing?
In practice, a single core operating on memory that no other threads are accessing doesn't slow down much in order to maintain global memory semantics, vs. how a uniprocessor system could be designed.
But on a big multi-socket system, especially x86, cache-coherency (snooping the other socket) is part of what makes memory latency worse for cache misses than on a single-socket system, though. (For accesses that miss in private caches).
Yes, all multi-core systems that you can run a single multi-threaded program on have coherent shared memory between all cores, using some variant of the MESI cache-coherency protocol. (Any exceptions to this rule are considered exotic and have to be programmed specially.)
Huge systems with multiple separate coherency domains that require explicit flushing are more like a tightly-coupled cluster for efficient message passing, not an SMP system. (Normal NUMA multi-socket systems are cache-coherent: Is mov + mfence safe on NUMA? goes into detail for x86 specifically.)
While a core has a cache line in MESI Modified or Exclusive state, it can modify it without notifying other cores about changes. M and E states in one cache mean that no other caches in the system have any valid copy of the line. But loads and stores still have to respect the memory model, e.g. an x86 core still has to commit stores to L1d cache in program order.
L1d and L2 are part of a modern CPU core, but you're right that L1d is not actually modified speculatively. It can be read speculatively.
Most of what you're asking about is handled by a store buffer with store forwarding, allowing store/reload to execute without waiting for the store to become globally visible.
what is a store buffer? and Size of store buffers on Intel hardware? What exactly is a store buffer?
A store buffer is essential for decoupling speculative out-of-order execution (writing data+address into the store buffer) from in-order commit to globally-visible L1d cache.
It's very important even for an in-order core, otherwise cache-miss stores would stall execution. And generally you want a store buffer to coalesce consecutive narrow stores into a single wider cache write, especially for weakly-ordered uarches that can do so aggressively; many non-x86 microarchitectures only have fully efficient commit to cache for aligned 4-byte or wider chunks.
On a strongly-ordered memory model, speculative out-of-order loads and checking later to see if any other core invalidated the line before we're "allowed" to read it is also essential for high performance, allowing hit-under-miss for out-of-order exec to continue instead of one cache miss load stalling all other loads.
There are some limitations to this model:
limited store-buffer size means we don't have much private store/reload space
a strongly-ordered memory model stops private stores from committing to L1d out of order, so a store to a shared variable that has to wait for the line from another core could result in the store buffer filling up with private stores.
memory barrier instructions like x86 mfence or lock add, or ARM dsb ish have to drain the store buffer, so stores to (and reloads from) thread-private memory that's not in practice shared still has to wait for stores you care about to become globally visible.
conversely, waiting for shared store you care about to become visible (with a barrier or a release-store) has to also wait for private memory operations even if they're independent.
the memory is always global, shared and globally synchronous, and
effectively entirely subject to all fences, even if it's memory used
as unnamed registers,
I'm not sure what you mean here. If a thread is accessing private data (i.e., not shared with any other thread), then there is almost no need for memory fence instructions1. Fences are used to control the order in which memory accesses from one core are seen by other cores.
Why doesn't the dedicated L1 cache provide register-like semantics for
those memory units that are only used by a particular execution unit?
I think (if I understand you correctly) what you're describing is called a scratchpad memory (SPM), which is a hardware memory structure that is mapped to the architectural physical address space or has its own physical address space. The software can directly access any location in an SPM, similar to main memory. However, unlike main memory, SPM has a higher bandwidth and/or lower latency than main memory, but is typically much smaller in size.
SPM is much simpler than a cache because it doesn't need tags, MSHRs, a replacement policy, or hardware prefetchers. In addition, the coherence of SPM works like main memory, i.e., it comes into play only when there are multiple processors.
SPM has been used in many commercial hardware accelerators such as GPUs, DSPs, and manycore processor. One example I am familiar with is the MCDRAM of the Knights Landing (KNL) manycore processor, which can be configured to work as near memory (i.e., an SPM), a last-level cache for main memory, or as a hybrid. The portion of the MCDRAM that is configured to work as SPM is mapped to the same physical address space as DRAM and the L2 cache (which is private to each tile) becomes the last-level cache for that portion of MCDRAM. If there is a portion of MCDRAM that is configured as a cache for DRAM, then it would be the last-level cache of DRAM only and not the SPM portion. MCDRAM has a much higher bandwdith than DRAM, but the latency is about the same.
In general, SPM can be placed anywhere in the memory hierarchy. For example, it could placed at the same level as the L1 cache. SPM improves performance and reduces energy consumption when there is no or little need to move data between SPM and DRAM.
SPM is very suitable for systems with real-time requirements because it provides guarantees regarding the maximum latency and/or lowest bandwdith, which is necessary to determine with certainty whether real-time constraints can be met.
SPM is not very suitable for general-purpose desktop or server systems where they can be multiple applications running concurrently. Such systems don't have real-time requirements and, currently, the average bandwdith demand doesn't justify the cost of including something like MCDRAM. Moreover, using an SPM at the L1 or L2 level imposes size constraints on the SPM and the caches and makes difficult for the OS and applications to exploit such a memory hierarchy.
Intel Optance DC memory can be mapped to the physical address space, but it is at the same level as main memory, so it's not considered as an SPM.
Footnotes:
(1) Memory fences may still be needed in single-thread (or uniprocessor) scenarios. For example, if you want to measure the execution time of a specific region of code on an out-of-order processor, it may be necessary to wrap the region between two suitable fence instructions. Fences are also required when communicating with an I/O device through write-combining memory-mapped I/O pages to ensure that all earlier stores have reached the device.

How does the cpu decide which data it puts in what memory (ram, cache, registers)?

When the cpu is executing a program, does it move all data through the memory pipeline? Then any piece of data would be moved from ram->cache->registers so all data that's executed goes in the cpu registers at some point. Or does it somehow select the code it puts in those faster memory types, or can you as a programmer select specific code you want to keep in, for example, the cache for optimization?
The answer to this question is an entire course in itself! A very brief summary of what (usually) happens is that:
You, the programmer, specify what goes in RAM. Well, the compiler does it on your behalf, but you're in control of this by how you declare your variables.
Whenever your code accesses a variable the CPU's MMU will check if the value is in the cache and if it is not, then it will fetch the 'line' that contains the variable from RAM into the cache. Some CPU instruction sets may allow you to prevent it from doing so (causing a stall) for specific low-frequecy operations, but it requires very low-level code to do so. When you update a value, the MMU will perform a 'cache flush' operation, committing the cached memory to RAM. Again, you can affect how and when this happens by low-level code. It will also depend on the MMU configuration such as whether the cache is write-through, etc.
If you are going to do any kind of operation on the value that will require it being used by an ALU (arithmetic Logic Unit) or similar, then it will be loaded into an appropriate register from the cache. Which register will depend on the instruction the compiler generated.
Some CPUs support Dynamic Memory Access (DMA), which provides a shortcut for operations that do not really require the CPU to be involved. These include memory-to-memory copies and the transfer of data between memory and memory-mapped peripheral control blocks (such as UARTs and other I/O blocks). These will cause data to be moved, read or written in RAM without actually affecting the CPU core at all.
At a higher level, some operating systems that support multiple processes will save the RAM allocated to the current process to the hard disk when the process is swapped out, and load it back in again from the disk when the process runs again. (This is why you may find 'Page Files' on your C: drive and the options to limit their size.) This allows all of the running processes to utilise most of the available RAM, even though they can't actually share it all simultaneously. Paging is yet another subject worthy of a course on its own. (Thanks to Leeor for mentioning this.)

How to reserve memory for my application and leave a specified amount remaining?

I'm planning an application which will involve loading many pictures at one time and thus requires a large chunk of memory. For example, I might have 50 image objects created at once, taking a total of 1GB of RAM. But when the user goes to load 20 more pictures, I'd like to make sure that amount of memory is already reserved and ready.
Now this part might seem a little backwards from normal. Rather than specifying how much memory my application shall reserve, instead I need to specify how much memory to leave free for other applications, and adjust my application's memory periodically according to this specification. I must say I've never worked with reserving memory at all, and especially won't know how to leave this remaining available memory.
So for example, if the computer has 2048 MB of RAM, and the option is set to leave 50 MB free for other applications, and there is already 10MB of RAM being used by other apps, then it should reserve 2048-50-10 = 1988 MB for my app.
The trouble I foresee is suppose the user opens another application which requires 1GB. My app has to catch this and shrink its self.
Does this even sound like a feasible approach? Basically, I need to make sure there is as much memory reserved as possible at any given time, while leaving a decent amount available for other apps. Would it make a significant impact on performance if I do this, or not much at all? I might be loading and unloading images at rapid paces, and I don't want it to reserve/free this memory on demand, I want it to stay reserved.
+1 for Sertac's mentioning of how SQL Server rides the line of allocating memory it needs, but releasing memory when Windows complains.
Applications can receive Window's complaints by using the CreateMemoryResourceNotification:
hLowMemory := CreateMemoryResourceNotification(LowMemoryResourceNotification);
Applications can use memory resource notification events to scale the
memory usage as appropriate. If available memory is low, the
application can reduce its working set. If available memory is high,
the application can allocate more memory.
Any thread of the calling
process can specify the memory resource notification handle in a call
to the QueryMemoryResourceNotification function or one of the wait functions.
The state of the object is signaled when the specified
memory condition exists. This is a system-wide event, so all
applications receive notification when the object is signaled. Note
that there is a range of memory availability where neither the
LowMemoryResourceNotification or HighMemoryResourceNotification object
is signaled. In this case, applications should attempt to keep the
memory use constant.
But it's also worth mentioning that you might as well allocate memory that you need. Your operating system has a very sophisiticated set of algorithms to swap out the least used memory when memory pressure is high. You can take advantage of this by simply allocating all the memory that you need. When Windows starts to run low, it will find those pages of memory that you are using the least and swap them out to disk. (This is how a well-known reverse proxy works).
The only thing left is to decide if you want to free some images when Windows says it's running low on RAM. But if you're not using the memory, it is going to be swapped out to disk for you.
It's not realistic to account for other apps. Just ignore them. The system will page things in and out as needed. If you really wanted to do this you'd have to dynamically adapt to other processes as they start and finish. That's really not realistic. What's more it's not practical to inquire of other processes how much memory they need. Leave it all to the system.
Set a budget for your app and make sure you don't exceed it. Keep the most recently used images in memory and when you approach your memory budget throw away the least recently used images to make space.
If you are stressing the available resources then make sure you use FastMM and enable LARGE_ADDRESS_AWARE for your app so that you get 4GB address space when running on a 64 bit OS.

Short question about CUDA memory access

hey there,
assuming I have a problem where each thread calculates something (reading some parameters out of the constant memory and using them for calculation) and than stores it to a global memory matrix. this matrix gets never read, just writing access... is there now any sense of using shared memory first to store all the calculated values in and than later write them to the global memory? I think no because the writes to global memory stay the same in complete, so the writes to shared memory just add to the writes which I had before already....
Thanks!
There can be, depending on the access patterns in the kernel code. Using a shared memory buffer to "stage" output can be a useful way of ensure writes are coalesced, when the naive write would not be coalesced. This was pretty crucial for performance in the first couple of generations of CUDA compatible hardware (G80/G90). In newer hardware, the case for this is a lot less strong. Fermi cards have a pretty effective L1 and L2 cache scheme which can (within reason) get close to what used to be only achievable using shared memory without any extra code.
There isn't really a general answer to this question, because it depends a lot of the specifics of what any given code does, and what target hardware it is expected to run well on.

Can a Memcached daemon ever free() unused memory, without terminating the process?

I believe that you can't force a running Memcached instance to de-allocate memory, short of terminating that Memcached instance (and freeing all of the memory it held). Does anyone know of a definitive piece of documentation, or even a mailing list or blog posting from a reliable source, that can confirm or deny this impression?
As I understand it, a Memcached process initially allocates a chunk of memory (the exact initial allocation size is configurable), and then monotonically increases its memory utilization over its lifetime, limited by the daemon's maximum memory allocation size (also configurable). At no point does the Memcached daemon ever free any memory, regardless of whether the daemon has any ongoing need for the memory it holds.
I know that this question might sound a little whiny, with a tone of "I DEMAND that open source project X support my specific need!" That's not it, at all--I'm purely interested in the exact technical answer, here, and I swear I'm not harshing on Memcached. For the curious, this question came out of a discussion about possible methods for gracefully juggling multiple Memcached instances on a single server, given an application where the cost of a cache flush can be quite high.
However, I'd appreciate it if you save your application suggestions/advice for a different question (re-architecting my application, using a different caching implementation, etc.). I do appreciate a good brainstorm, but I think this question will be most valuable if it stays focused on the technical specifics of how Memcached does and does not work. If you don't have the answer to this specific question, there is probably still value in what you have to say, but I'd guess that there's a different, better place to post the more speculative comments/suggestions/advice.
This is probably the hardest problem we have to solve for memcached currently (well, a variation of it, anyway).
Freeing a chunk of memory requires us to know that a) nothing within the chunk is in use and b) nothing will start using it while we're in the process of purging it for reuse/freeing. I've heard some really good ideas for how we might solve our slab rebalancing problems which is basically the same, except we're not trying to free the memory, but to give it to something else (a common problem in a few large installations).
Also, whether free actually reduces the RSS of your process is implementation dependent. In many cases, a malloc/fill/free will leave the memory mapped in (unless your allocator uses mmap instead of sbrk).
I'm pretty sure this isn't possible with memcached. I don't see any technical reason why it couldn't be implemented though. Lock cache operations, expire enough keys to reach the desired size, update the size, unlock. (I'm sure there's nicer ways to avoid blocking the server during that time.)
The standard and default mechanism of memory management in memcached is slab allocator. It means that memory is being allocated for the process and never released to the operating system. Basically, when memory is no longer used to store some data, it is being held by the process in order to be reused later, when needed. However, the operating system releases memory allocated by the process when it is finished. That is why memory is being released when you kill/stop the memcached.
There is a compile-time option in memcached to enable malloc/free mechanism. So that when free() is called, memory might be released to operating system (this depends on C standard library implementation). But doing so might hurt a good fragmentation and performance.
Please read more about the issue here:
Why not use malloc/free
Memcached memory management

Resources