Related
My program consists out of two parts, A and B, both written in C++. B is loaded from a separate DLL, and is capable of running both on the CPU or on the GPU, depending on how it is linked. When the main program is launched, it creates one instance of A, which in turn creates one instance of B (which then either works on the locally available CPUs or on the first GPU).
When launching the program using mpirun (or via slurm, which in turn launches mpirun), for each MPI rank one version of A is created, which in turn creates one version of B for itself. When only one GPU is in the system, this GPU will be used, but what happens if there are multiple GPUs in the system? Are versions of B all placed on the same GPU, regardless if there are several GPUs available, or are they distributed evenly?
Is there any way to influence that behavior? Unfortunately my development machine does not have multiple GPUs, thus I can not test it, except on production.
Slurm supports and understands binding MPI ranks to GPUs through for-example the --gpu-bind option: https://slurm.schedmd.com/gres.html. Assuming that the cluster is correctly configured to enforce GPU affinities, this will then allow you assign one GPU per rank even if there are multiple ranks on a single node.
If you want to be able to test this, you could use for example the cudaGetDevice and cudaGetDeviceProperties calls to get the device luid (local unique id) for each rank and then check that there is no duplication of luids within a node.
I have a basic question about the linux system call.
Why are the system calls not handled just like normal function calls and why is handled via software interrupts?
Is it because, there is no linking process performed for user space application with kernel during the build process of user application?
Linking between separately compiled pieces of code is a minor problem. Shared libraries have had a workaround for it for quite some time (relocatable code, export tables, etc). You pay the cost typically just once, when you load the library in the program.
The bigger problem is that you need to switch the CPU from the unprivileged, user mode into the privileged, kernel mode and you need to do it in a controllable way, without letting user code escape and wreck a havoc on the kernel. And that's typically done with special or designated instructions. You may also benefit from automatic interrupt disabling when transitioning into the kernel, which the x86 int instruction can do for you. Most CPUs have something like this instruction and it's a common way of implementing the system call interface, although not the only one.
If you asked about MS-DOS or the original MINIX, both of which ran on the i8086 in the real address mode, where the kernel couldn't protect itself or other programs from anything because all the memory and system resources were accessible to all code, then there would be less reason in using a special instruction like int, there were no two modes, only one, and in that respect int would be largely equivalent to a simple call (far).
Also noteworthy is the fact that CPUs often handle the following 3 types of events in a very similar fashion:
hardware interrupts from I/O devices
exceptions, errors from code execution (e.g. division by 0, page faults, etc)
system calls
That makes using something like the int instruction a natural choice as your entry and exit points in all of the above handlers would be if not fully then largely identical.
Let's take Python as an example. If I am not mistaken, when you program in it, the computer first "translates" the code to C. Then again, from C to assembly. Assembly is written in machine code. (This is just a vague idea that I have about this so correct me if I am wrong) But what's machine code written in, or, more exactly, how does the processor process its instructions, how does it "find out" what to do?
If I am not mistaken, when you program in it, the computer first "translates" the code to C.
No it doesn't. C is nothing special except that it's the most widespread programming language used for system programming.
The Python interpreter translates the Python code into so called P-Code that's executed by a virtual machine. This virtual machine is the actual interpreter which reads P-Code and every blip of P-Code makes the interpreter execute a predefined codepath. This is not very unlike how native binary machine code controls a CPU. A more modern approach is to translate the P-Code into native machine code.
The CPython interpreter itself is written in C and has been compiled into a native binary. Basically a native binary is just a long series of numbers (opcodes) where each number designates a certain operation. Some opcodes tell the machine that a defined count of numbers following it are not opcodes but parameters.
The CPU itself contains a so called instruction decoder, which reads the native binary number by number and for each opcode it reads it gives power to the circuit of the CPU that implement this particular opcode. there are opcodes, that address memory, opcodes that load data from memory into registers and so on.
how does the processor process its instructions, how does it "find out" what to do?
For every opcode, which is just a binary pattern, there is its own circuit on the CPU. If the pattern of the opcode matches the "switch" that enables this opcode, its circuit processes it.
Here's a WikiBook about it:
http://en.wikibooks.org/wiki/Microprocessor_Design
A few years ago some guy built a whole, working computer from simple function logic and memory ICs, i.e. no microcontroller or similar involved. The whole project called "Big Mess o' Wires" was more or less a CPU built from scratch. The only thing nerdier would have been building that thing from single transistors (which actually wasn't that much more difficult). He also provides a simulator which allows you to see how the CPU works internally, decoding each instruction and executing it: Big Mess o' Wires Simulator
EDIT: Ever since I originally wrote that answer, building a fully fledged CPU from modern, discrete components has been done: For your considereation a MOS6502 (the CPU that powered the Apple II, Commodore C64, NES, BBC Micro and many more) built from discetes: https://monster6502.com/
Machine-code does not "communicate with the processor".
Rather, the processor "knows how to evaluate" machine-code. In the [widespread] Von Neumann architecture this machine-code (program) can be thought of as an index-able array of where each cell contains a machine-code instruction (or data, but let's ignore that for now).
The CPU "looks" at the current instruction (often identified by the PC or Program Counter) and decides what to do (this can either be done directly with transistors/"bare-metal", or it be translated to even lower-level code): this is known as the fetch-decode-execute cycle.
When the instructions are executed side-effects occur such as setting a control flag, putting a value in a register, or jumping to a different index (changing the PC) in the program, etc. See this simple overview of a CPU which covers the above a little bit better.
It is the evaluation of each instruction -- as it is encountered -- and the interaction of side-effects that results in the operation of a traditional processor.
(Of course, modern CPUs are very complex and do lots of neat tricky things!)
That's called microcode. It's the code in the CPU that reads machine code instructions and translate that into low level data flow.
When the CPU for example encounters the add instruction, the microcode describes how it should get the two values, feed them to the ALU to do the calculation, and where to put the result.
Electricity. Circuits, memory, and logic gates.
Also, I believe Python is usually interpreted, not compiled through C → assembly → machine code.
I'm curious, why did Sun decide to make the JVM stack-based and Google decide to make the DalvikVM register-based?
I suppose the JVM can't really assume that a certain number of registers are available on the target platform, since it is supposed to be platform independent. Therefor it just postpones the register-allocation etc, to the JIT compiler. (Correct me if I'm wrong.)
So the Android guys thought, "hey, that's inefficient, let's go for a register based vm right away..."? But wait, there are multiple different android devices, what number of registers did the Dalvik target? Are the Dalvik opcodes hardcoded for a certain number of registers?
Do all current Android devices on the market have about the same number of registers? Or, is there a register re-allocation performed during dex-loading? How does all this fit together?
There are a few attributes of a stack-based VM that fit in well with Java's design goals:
A stack-based design makes very few
assumptions about the target
hardware (registers, CPU features),
so it's easy to implement a VM on a
wide variety of hardware.
Since the operands for instructions
are largely implicit, the object
code will tend to be smaller. This
is important if you're going to be
downloading the code over a slow
network link.
Going with a register-based scheme probably means that Dalvik's code generator doesn't have to work as hard to produce performant code. Running on an extremely register-rich or register-poor architecture would probably handicap Dalvik, but that's not the usual target - ARM is a very middle-of-the-road architecture.
I had also forgotten that the initial version of Dalvik didn't include a JIT at all. If you're going to interpret the instructions directly, then a register-based scheme is probably a winner for interpretation performance.
I can't find a reference, but I think Sun decided for the stack-based bytecode approach because it makes it easy to run the JVM on an architecture with few registers (e.g. IA32).
In Dalvik VM Internals from Google I/O 2008, the Dalvik creator Dan Bornstein gives the following arguments for choosing a register-based VM on slide 35 of the presentation slides:
Register Machine
Why?
avoid instruction dispatch
avoid unnecessary memory access
consume instruction stream efficiently (higher semantic density per instruction)
and on slide 36:
Register Machine
The stats
30% fewer instructions
35% fewer code units
35% more bytes in the instructions stream
but we get to consume two at a time
According to Bornstein this is "a general expectation what you could find when you convert a set of class files to dex files".
The relevant part of the presentation video starts at 25:00.
There is also an insightful paper titled "Virtual Machine Showdown: Stack Versus Registers" by Shi et al. (2005), which explores the differences between stack- and register-based virtual machines.
I don't know why Sun decided to make JVM stack based. Erlangs virtual machine, BEAM is register based for performance reasons. And Dalvik also seem to be register based because of performance reasons.
From Pro Android 2:
Dalvik uses registers as primarily units of data storage instead of the stack. Google is hoping to accomplish 30 percent fewer instructions as a result.
And regarding the code size:
The Dalvik VM takes the generated Java class files and combines them into one or more Dalvik Executables (.dex) files. It reuses duplicate information from multiple class files, effectively reducing the space requirement (uncompressed) by half from traditional .jar file. For example, the .dex file of the web browser app in Android is about 200k, whereas the equivalent uncompressed .jar version is about 500k. The .dex file of the alarm clock is about 50k, and roughly twice that size in its .jar version.
And as I remember Computer Architecture: A Quantitative Approach also conclude that a register machine perform better than a stack based machine.
It came to my attention some emulators and virtual machines use dynamic recompilation. How do they do that? In C i know how to call a function in ram using typecasting (although i never tried) but how does one read opcodes and generate code for it? Does the person need to have premade assembly chunks and copy/batch them together? is the assembly written in C? If so how do you find the length of the code? How do you account for system interrupts?
-edit-
system interrupts and how to (re)compile the data is what i am most interested in. Upon more research i heard of one person (no source available) used js, read the machine code, output js source and use eval to 'compile' the js source. Interesting.
It sounds like i MUST have knowledge of the target platform machine code to dynamically recompile
Yes, absolutely. That is why parts of the Java Virtual Machine must be rewritten (namely, the JIT) for every architecture.
When you write a virtual machine, you have a particular host-architecture in mind, and a particular guest-architecture. A portable VM is better called an emulator, since you would be emulating every instruction of the guest-architecture (guest-registers would be represented as host-variables, rather than host-registers).
When the guest- and host-architectures are the same, like VMWare, there are a ton of (pretty neat) optimizations you can do to speed up the virtualization - today we are at the point that this type of virtual machine is BARELY slower than running directly on the processor. Of course, it is extremely architecture-dependent - you would probably be better off rewriting most of VMWare from scratch than trying to port it.
It's quite possible - though obviously not trivial - to disassemble code from a memory pointer, optimize the code in some way, and then write back the optimized code - either to the original location or to a new location with a jump patched into the original location.
Of course, emulators and VMs don't have to RE-write, they can do this at load-time.
This is a wide open question, not sure where you want to go with it. Wikipedia covers the generic topic with a generic answer. The native code being emulated or virtualized is replaced with native code. The more the code is run the more is replaced.
I think you need to do a few things, first decide if you are talking about an emulation or a virtual machine like a vmware or virtualbox. An emulation the processor and hardware is emulated using software, so the next instruction is read by the emulator, the opcode pulled apart by code and you determine what to do with it. I have been doing some 6502 emulation and static binary translation which is dynamic recompilation but pre processed instead of real time. So your emulator may take a LDA #10, load a with immediate, the emulator sees the load A immediate instruction, knows it has to read the next byte which is the immediate the emulator has a variable in the code for the A register and puts the immediate value in that variable. Before completing the instruction the emulator needs to update the flags, in this case the Zero flag is clear the N flag is clear C and V are untouched. But what if the next instruction was a load X immediate? No big deal right? Well, the load x will also modify the z and n flags, so the next time you execute the load a instruction you may figure out that you dont have to compute the flags because they will be destroyed, it is dead code in the emulation. You can continue with this kind of thinking, say you see code that copies the x register to the a register then pushes the a register on the stack then copies the y register to the a register and pushes on the stack, you could replace that chunk with simply pushing the x and y registers on the stack. Or you may see a couple of add with carries chained together to perform a 16 bit add and store the result in adjacent memory locations. Basically looking for operations that the processor being emulated couldnt do but is easy to do in the emulation. Static binary translation which I suggest you look into before dynamic recompilation, performs this analysis and translation in a static manner, as in, before you run the code. Instead of emulating you translate the opcodes to C for example and remove as much dead code as you can (a nice feature is the C compiler can remove more dead code for you).
Once the concept of emulation and translation are understood then you can try to do it dynamically, it is certainly not trivial. I would suggest trying to again doing a static translation of a binary to the machine code of the target processor, which a good exercise. I wouldnt attempt dynamic run time optimizations until I had succeeded in performing them statically against a/the binary.
virtualization is a different story, you are talking about running the same processor on the same processor. So x86 on an x86 for example. the beauty here is that using non-old x86 processors, you can take the program being virtualized and run the actual opcodes on the actual processor, no emulation. You setup traps built into the processor to catch things, so loading values in AX and adding BX, etc these all happen at real time on the processor, when AX wants to read or write memory it depends on your trap mechanism if the addresses are within the virtual machines ram space, no traps, but lets say the program writes to an address which is the virtualized uart, you have the processor trap that then then vmware or whatever decodes that write and emulates it talking to a real serial port. That one instruction though wasnt realtime it took quite a while to execute. What you could do if you chose to is replace that instruction or set of instructions that write a value to the virtualized serial port and maybe have then write to a different address that could be the real serial port or some other location that is not going to cause a fault causing the vm manager to have to emulate the instruction. Or add some code in the virtual memory space that performs a write to the uart without a trap, and have that code instead branch to this uart write routine. The next time you hit that chunk of code it now runs at real time.
Another thing you can do is for example emulate and as you go translate to a virtual intermediate bytcode, like llvm's. From there you can translate from the intermediate machine to the native machine, eventually replacing large sections of program if not the whole thing. You still have to deal with the peripherals and I/O.
Here's an explaination of how they are doing dynamic recompilation for the 'Rubinius' Ruby interpteter:
http://www.engineyard.com/blog/2010/making-ruby-fast-the-rubinius-jit/
This approach is typically used by environments with an intermediate byte code representation (like Java, .net). The byte code contains enough "high level" structures (high level in terms of higher level than machine code) so that the VM can take chunks out of the byte code and replace it by a compiled memory block. The VM typically decide which part is getting compiled by counting how many times the code was already interpreted, since the compilation itself is a complex and time-consuming process. So it is usefull to only compile the parts which get executed many times.
but how does one read opcodes and generate code for it?
The scheme of the opcodes is defined by the specification of the VM, so the VM opens the program file, and interprets it according to the spec.
Does the person need to have premade assembly chunks and copy/batch them together? is the assembly written in C?
This process is an implementation detail of the VM, typically there is a compiler embedded, which is capable to transform the VM opcode stream into machine code.
How do you account for system interrupts?
Very simple: none. The code in the VM can't interact with real hardware. The VM interact with the OS, and transfer OS events to the code by jumping/calling specific parts inside the interpreted code. Every event in the code or from the OS must pass the VM.
Also hardware virtualization products can use some kind of JIT. A typical use cases in the X86 world is the translation of 16bit real mode code to 32 or 64bit protected mode code to not to be forced to emulate a CPU in real mode. Also a software-only VM replaces jump instructions in the executing code by jumps into the VM control software, which at each branch the following code path for jump instructions scans and them replace, before it jumps to the real code destination. But I doubt if the jump replacement qualifies as JIT compilation.
IIS does this by shadow copying: after compilation it copies assemblies to some temporary place and runs them from temp.
Imagine, that user change some files. Then IIS will recompile asseblies in next steps:
Recompile (all requests handled by old code)
Copies new assemblies (all requests handled by old code)
All new requests will be handled by new code, all requests - by old.
I hope this'd be helpful.
A virtual Machine loads "byte code" or "intermediate language" and not machine code therefore, I suppose, that it just recompiles the byte code more efficiently once it has more runtime data.
http://en.wikipedia.org/wiki/Just-in-time_compilation