Keeping your code in L1 cache - optimization

I have been reading Wikipedia's article on K programming language and this is what I saw:
The small size of the interpreter and compact syntax of the language makes it possible for K applications to fit entirely within the level 1 cache of the processor.
I am intrigued. How is it possible to have the whole program in L1 cache? Say, CPU has 256kb L1 cache. Say my program is way less than that and it needs a very little amount of memory (say, just for the call stack and such). Say, it doesn't need any libraries (although if a program is for an OS, it would need to include kernel32.dll or whatever). And doesn't OS automatically allocates some minimal memory for any program (well, for executable code and stack and heap)?
Thank you.

I think what they're saying is not that the entire program fits in L1 cache, but that all the code that runs most of the time fits in the L1 cache.
Yes, the OS allocates lots of other structures, but those are hit rarely enough to not matter.
Of course, this is all speculation -- I know nothing about the 'K' language.

I believe they are speaking to the advantage that the main executing code will fit in the L1 cache; regardless of the memory allocated to the program. Once the K application is loaded, if it never touches that memory then it doesn't matter if it's allocated in terms of performance (i.e. the perf benefit of being totally in L1 cache).

The interpreter runs as a normal program managed by the OS. The interpreted program runs within the memory space of the interpreter, in the data segment. Many K programs may easily fit into the L1 cache completely, even though the entire interpreter may not. The main interpreter loop will probably fit though.

You confuse all the program code with the most frequently executed code.
For the interpreted languages the interpreter core is certainly among the most frequently executed code. Having most frequently executed code in cache speeds up execution the same way as having most frequently accessed data in cache does.
The key part is "most frequently" - it's not necessary to have all the code/data cached to see a significant acceleration.

Related

Could a tight loop destroy cells of a microcontroller's flash?

It is well-known that Flash memory has limited write endurance, less so that reads could also have an upper limit such as mentioned in this Flash endurance test's Conclusion (3rd point).
On a microcontroller the code is typically stored in Flash, and is executed by fetching code words directly from the Flash cells.(at least this is most commonly so on 8 bit micros, some 32 bit micros might have some small buffer).
Depending on the particular code, it might happen that a location is accessed extremely frequently, such as if on the main execution path there is some busy loop, such as a wait for an interrupt (for example from a timer, synchronizing execution to a fixed interval).This could generate 100K or even more (read) accesses per second on average to a single Flash cell (depending on clock and the particular code).
Could such code actually destroy the cells of the Flash underneath it?(Is there any necessity to be concerned about this particular problem when designing code for microcontrollers? Such as part of a system which is meant to operate for years? Of course the Flash could be periodically verified by CRC, but that doesn't prevent the system failing if it happens, only that the failure will more likely happen in a controlled manner)
Only erasing/writing will affect the memory cells, not reading. You don't need to consider the number of reads when designing the program.
Programmed flash memory does age though, meaning that the value of the cells might not be reliable after a certain amount of years. This is known as data retention and depends mainly on temperature. MCU manufacturers typically specify a worse case in years, assuming that the part is kept in maximum specified ambient temperature.
This is something to consider for products that are expected to live long (> 10 years), particularly in environments where high temperatures can be expected. CRC and/or ECC is a good counter-measure against data retention, although if you do find that a cell has been corrupted, it typically just means that the application should shut down to a non-recoverable safe state.
I know of two techniques to approach this issue:
1) One technique is to set aside a const 32-bit integer variable in the system code. Then calculate a CRC32 checksum of the compiled binary image, and inserting the checksum into the reserved variable using an ELF-editor.
A module in the system software will then calculate a CRC32 over the flash area occupied by the application and compare to the "stored" value.
If you are using GCC, the linker can define a symbol to tell you where the segment stops. This method can detect errors but cannot correct them.
2) Another technique is to use a microcontroller that supports Flash ECC. TI sells Cortex-R4 MCUs which support Flash ECC (Hercules series).
I doubt that this is a practical concern. The article you cited vaguely asserts that this can happen but with no supporting evidence or quantification of the effect. There is a vague, unsupported and unquantified reference in the introduction:
[...] flash degrades over time from erasing/writing (or even just reading, although that decay is slower) [...]
Then again in the conclusion:
We did not check flash decay for reads, but reading also causes long term decay. It would be interesting to see if we can read a spot enough times to cause failure.
The author may be referring to read-disturbance in NAND flash, but microcontrollers do not use NAND flash for code storage/execution since it is not random-access. Read disturb is not a permanent effect, erasing and re-writing the affected block restores endurance. NAND controllers maintain read counts for blocks and automatically copy and erase blocks as necessary. They also employ ECC to detect and correct errors, and identify "write-worn" areas.
There is the potential for long-term "bit-rot" but I doubt that it is caused specifically by reading rather just ageing.
Most RTOS systems spend the majority of their processing time in a do-nothing idle loop, and run happily 24/7 365 days a year. Some processors support a wait-for-interrupt instruction that halts the CPU in the idle loop, but by no means all, and it is not uncommon not to use such an instruction. Processors with flash accelerators or caches may also prevent continuous rapid fetch from a single location, but again that is by no means all microcontrollers.

How to check the Heap and Stack RAM consistency on an embedded system

I'm working on a project using a LEON2 Processor (Sparc V8).
The processor uses 8Mbytes of RAM that need to be consistency checked during the Self-Test of my Boot.
My issue is that my Boot obviously uses a small part of the RAM for its Heap/BSS/Stack which I cannot modify without crashing my application.
My RAM test is very simple, write a certain value to all the RAM address then read them back to be sure the RAM chip can be addressed.
This method can be used for most of the RAM available but how could I safely check for the consistency of the remaining RAM?
Generally a RAM test that needs to test every single byte will be done as one of the first things that happens when the processor starts. Often the only other thing that's done before it is the hardware initialization that needs to happen for the RAM test to be able to access RAM.
It'll usually be done in assembly language with interrupts disabled, for one reason because that's about the only way you can ensure that no RAM is used.
If you want to perform a RAM test after that point, you still need to do it pretty early in the system start-up. You could maybe do it in two passes - where any variables/stack/whatever the test needs for its own purposes are in low RAM, and that test tests high RAM. Then have the test run again with it's data in high RAM while it tests low RAM.
Another note: verifying that you read back a certain value written is a simple test that maybe better than nothing, but it can miss certain types of common failures (common particularly with external RAM: missing or cross soldered address lines.
You can find more detailed information about basic RAM tests here:
Jack Ganssle, "Testing RAM in Embedded Systems"
Michael Barr, "Fast Accurate Memory Test Suite"
as I am programming a safety-relevant device, I have to do a full RAM test during operation time.
I split the test in two tests:
Addressing test
you write unique values to the addresses reached by each addressing line and after all values are written, the values are read back and compared to the expected values. This test detects short-circuits (or stuck#low/high) of addressing lines (meaning you want to write 0x55 on address 0xFF40, but due to a short-circuit the value is stored at 0xFF80, you cannot detect this by test 2:
Pattern Test:
You save e.g. the first 4 bytes of RAM in CPU's registers and afterwards you first clear the cells, write 0x55, verify, write 0xAA, verify and restore saved content (you can use other patterns of course) and so on. The reason you have to use the registers is that by using a variable, this variable would be destroyed by that test.
You can even test your stack with this test.
In our project, we test 4 cells at a time and we have to run this test until whole RAM is tested.
I hope that helped a bit.
If you do your testing before the C run time environment is up you can trash the Heap and BSS areas without any problems.
Generally the stack does not get used much during run time setup so you may be able trash it with no ill effect. Just check your system.
If you need to use the stack during testing or need to preserve it just move it to an already tested area adjust the stack pointer. After wards just restore the old stack and continue.
There is no easy ways of doing this once you entered your runtime environment.

Points to be considered while designing or coding for lesser footprint deliverables

Please post the points one should keep in mind while designing or coding for lesser footprint deliverables for embedded systems.
I am not giving compiler or platform details, as I want generic information. But, any specific information on Linux based OS is also welcome.
Depends on how low you want to get. I'm currently coding for fiscal printers, and there's no OS, and the main rule is no dynamic memory allocation. The funny thing is that I still convinced the crew to code fully modern C++ ;).
Actually there are a few rules we decided upon:
no dynamic allocation
hence, no STL
no exception handling (obvious reasons)
There isn't a general answer, only ones specific to language/platform ... but
Small memory footprint ...
Don't use Java, C#/mono, PHP, Perl, Python or anything with garbage collection
Get as close to the metal as feasible, Use C
Do alot of profiling to see where memory is getting allocated, if you are using dynamic allocation
Ensure you prevent heap-fragmentation by allocating sensible chunks and sizes of the heap
Avoid recursive functions especially those that use malloc(). Better allocating a chunk and passing a pointer around.
use free() ;)
Ensure your types are no bigger than required
Turn on compiler optimizations
There will be more.
for real low footprint consider doing Assembly directly.
We all know that Hello World in C or C++ is 20kb+(because of all the default libraries which get linked). In Assembly this overhead is gone. As pointed out in the comments one can reduce the standard libraries quite a bit. However, the fact remains that the code density you can get when coding assembly is much higher than a compiler will generate from a higher language. So for code where every byte matters, use assembly.
also when programming on devices with less capable processors, programming in assembly language might be your only way to do make the program fast enough for it to be realtime enough to (for instance) control machines
When faced with such constraints, it is advisable to pre-allocate memory in order to guarantee that the system will work under load. A design pattern such as "object pooling" can be used to share resources within the system.
The C language enables tight resource (i.e. memory & compute cycles) control. It should be strongly considered.
Avoid recursion as it is easy to abuse and can result in stack overflow conditions.

Why do Cocoa apps use so much memory?

Even the standard blank-window Cocoa app that gets built when you make a new Cocoa project in Xcode uses almost 6 MB of memory. What's the reason for this? Is it possible to make an app use less, or does OS X simply manage memory differently for Cocoa apps?
Not that I'm complaining. I know that performance "hardly matters anymore" (edit: what I mean is, it matters less than readability/maintainability/the programmer's time). I'm just curious.
OS X does a lot of magic with shared memory and copy-on-write pages, so chances are that it doesn't take that much physical RAM for every application.
You can check exactly how memory blocks are mapped by running:
sudo vmmap <PID of the process>
Depends on the all the framework (APIs) you use. Combine that with the VM allocations done by low level ops.
It's only worth trying to reduce the heap alloc (total), as well as the resident size of the code. Making sure your data structs are allocated efficiently and trying to compile with the ever-so-famous "-Os" optimization flag (size optimization). There isn't much you can do about the VM eaten by Cocoa. I wouldn't really worry about it.
This is clearly a 'WTF' moment for developers in general. The question is usually - why does my trivial application use up so much memory.
The answer is down to the underlying framework. You could argue that 6MB is too much, but really, it is nothing.
It's not rare to see computers come with 2GB of memory these days. The stock IMAC is 4GB. The whole point of the computer industry is to use up all the resources a machine has so that it continues to evolve.
Yes you should avoid ineffecincies where possible (Don't load up a 5million point array at start up for instance). But unless your beta demonstrates you fudged up just keep it on the list of todo's.
I'm a bit out on a limb here, but I guess it's because all the libraries that get added have to do quite a bit of setting up and there is no need to garbage collect, so they simply get to waste memory; plus, even if all memory got autoreleased, it would wait until the first idle event, which is after the creation of the window. Delete unneeded libraries/frameworks, or force a garbage collect somewhere after loading the window from the nib and see how much it goes down if you're so concerned.
I am not concerned about it. Some of the memory might be returned later, and the rest is the price you pay for a powerful framework.
A factor which is not directly linked to cocoa but is valid to frameworks in general is that the overhead is not linear. There is usually a fixed and a variable "price", in terms of overhead, to use the framework.
When you create a simple blank window, the fixed overhead is crushing, but when you create an application with tens of windows, dialogs, controls and all, the initial fixed overhead becomes negligible, against the size of the application itself.

Optimizing for ARM: Why different CPUs affects different algorithms differently (and drastically)

I was doing some benchmarks for the performance of code on Windows mobile devices, and noticed that some algorithms were doing significantly better on some hosts, and significantly worse on others. Of course, taking into account the difference in clock speeds.
The statistics for reference (all results are generated from the same binary, compiled by Visual Studio 2005 targeting ARMv4):
Intel XScale PXA270
Algorithm A: 22642 ms
Algorithm B: 29271 ms
ARM1136EJ-S core (embedded in a MSM7201A chip)
Algorithm A: 24874 ms
Algorithm B: 29504 ms
ARM926EJ-S core (embedded in an OMAP 850 chip)
Algorithm A: 70215 ms
Algorithm B: 31652 ms (!)
I checked out floating point as a possible cause, and while algorithm B does use floating point code, it does not use it from the inner loop, and none of the cores seem to have a FPU.
So my question is, what mechanic may be causing this difference, preferrably with suggestions on how to fix/avoid the bottleneck in question.
Thanks in advance.
One possible cause is that the 926 has a shorter pipeline (5 cycles vs. 8 cycles for the 1136, iirc), so branch mispredictions are less costly on the 926.
That said, there are a lot of architectural differences between those processors, too many to say for sure why you see this effect without knowing something about the instructions that you're actually executing.
Clock speed is only one factor. Bus width and latency are big if not bigger factors. Cache is a factor. Speed of the media the program is run from if run from media and not memory.
Is this test using any shared libraries at all at any point in the test or is it all internal code? Fetching shared libraries on media that will vary from platform to platform (even if it is say the same sd card).
Is this the same algorithm compiled separately for each platform or the same binary? You can and will see some compiler induced variation as well. 50% faster and slower can easily come from the same compiler on the same platform by varying compiler settings. If possible you want to execute the same binary, and insure that no shared libraries are used in the loop under test. If not the same binary disassemble the loop under test for each platform and insure that there are no variations other than register selection.
From the data you have presented, its difficult to point the exact problem, but we can share some of the prior experience
Cache setting (check if all the
processors has the same CACHE
setting)
You need to check both D-Cache and I-Cache
For analysis,
Break down your code further, not just as algorithm but at a block level, and try to understand the block that causes the bottle-neck. After you find the block that causes the bottle-neck, try to disassemble the block's source code, and check the assembly. It may help.
Looks like the problem is in cache settings or something memory-related (maybe I-Cache "overflow").
Pipeline stalls, branch miss-predictions usually give less significant differences.
You can try to count some basic operations, executed in each algorithm, for example:
number of "easy" arithmetical/bitwise ops (+-|^&) and shifts by constant
number of shifts by variable
number of multiplications
number of "hard" arithmetics operations (divides, floating point ops)
number of aligned memory reads (32bit)
number of byte memory reads (8bit) (it's slower than 32bit)
number of aligned memory writes (32bit)
number of byte memory writes (8bit)
number of branches
something else, don't remember more :)
And you'll get info, that things get 926 much slower. After this you can check suspicious blocks, making using of them more or less intensive. And you'll get the answer.
Furthermore, it's much better to enable assembly listing generation in VS and use it (but not your high-level source code) as base for research.
p.s.: maybe the problem is in OS/software/firmware? Did you testing on clean system? OS is the same on all devices?