Getting "minimal" SSA from LLVM - llvm

The LLVM's opt -S -mem2reg pass produces the so-called "pruned" SSA -- the form that has all the dead phi functions removed. I would like to keep those phi instructions in the IR, obtaining the "minimal" SSA, but I'm failing to find an easy way to do it.
Am I doomed to implement the whole SSA construction algorithm from scratch or there is a way to do it with existing tools?

LLVM doesn't have any support for forming anything other than pruned SSA form, and it's unlikely to grow such a mechanism. We quite literally don't even do the work to synthesize this information when doing phi placement.

Related

Fast boolean expression evaluator

I have an app that includes a 3 operator (& | !) boolean expression evaluator, with variables and constants. Generally the expressions aren't too long (perhaps 50 terms at the most, but usually a lot less). There can be very many expressions - I'm expecting the upper limit to be around a million. Currently I have a hand written parser with a very simple evaluator that simply recursively traverses the parse tree. One constraint is that this has to be callable from C++. I have no sharing between expressions. I'd like to investigate speeding this up.
I see two avenues of research.
Add sharing and store the state indicating whether an expression node has been evaluated or not.
Extract Common Subexpressions.
Also I would expect that a code generation approach will be faster than an interpretive approach working on parse trees or similar structures. It would probably be fairly straightforward to generate some C++ code, but considering the length of the functions, I don't know if a compiler like GCC will be able to optimize the CSEs.
I've seen that there are a few libraries available for expression evaluation, but in my work environment adding 3rd party libraries is not simple plus they all seem very complicated compared to my needs.
Lastly I've been looking at Antlr4 a bit recently, so that might be appropriate for me. In the past I've worked on C code generation, but I have no experience of using something like LLVM for optimisation and code generation.
Any suggestions for which way to go?
As far as I understood, your question is more about faster expression evaluation than it about faster expression parsing. So my answer will focus on the former. Parsing, after all, should not be the bottleneck as your expression language looks simple enough to implement a manually tuned parser for it.
So, to accelerate your evaluations, you can consider JIT execution of your formulas using LLVM. That is, given your formula F you can (relatively) easily generate corresponding LLVM IR and directly evaluate it. This SMT solver does just that. IR code generation is implemented in a single C++ class here.
Note that the boolean expressions you mentioned are a subset of the SMT language supported by that solver. Additionally, you can easily adjust how aggressive the LLVM optimizer needs to be.
However, IR generation and optimization has its overhead. Therefore, in case a given formula is not evaluated often enough to amortize the initial overhead, then I would recommend direct interpretation instead. You can look in this case for opportunities to find structural similarities and common subexpressions.
As much as I'd like to suggest ANTLR4, I fear it won't meet your performance needs. There is a lot going on under the hood with its adaptive LL(*) algoritms and though there are some common tricks to improve its performance, simply tracing an ANTLR4 interpreter at runtime suggests that unless your current expression evaluator is very inefficient, it is likely faster than ANTLR4, which is an industrial-duty engine meant to support grammars far more complicated than yours. I use ANTLR when a LALR(1) DFA shift-reduce engine won't support my grammar, and take the performance hit in return for the extra parsing power of ANTLR4.

Adding clauses directly to the z3 solver

I have an AIG (and-inverter graph) which I keep modifying and whose satisfiability I need to check in an incremental manner using Z3. I can generate a CNF representation of the AIG and would ideally like to feed these clauses directly to the solver and call it repeatedly from my code. Is there some way that I can directly add clauses (or an AIG) to Z3 solver through C/C++ APIs?
Yes, you can simply assert new assertions, which are internally translated into clauses.
Note that for many incremental solving problems, Z3 does not use an off-the-shelf, dedicated SAT-solver, but it's own SMT solver that incorporates some of the features of SAT solvers, but not all, and which natively handles non-Boolean problems. So, it's not necessarily the case that hacking the solver to inject clauses directly will translate into significantly improved performance.
Z3 also has a dedicated Boolean-only SAT solver and if you're solving purely Boolean problems, this solver is likely much faster. You can force it to use this solver by replacing (check-sat) with (check-sat-using sat), or by running the tactic called 'sat'. The implementation of this solver is in sat_solver.h/.cpp, which would be the prime place to start looking around, if you'd like to hack it.
Z3 also uses it's own implementation of AIGs as a pre-processing step in some tactics, see aig_tactic.h/.cpp.

Coding for ARM NEON: How to start?

I'm looking to optimize C++ code (mainly some for loops) using the NEON capability of computing 4 or 8 array elements at a time. Is there some kind of library or set of functions that can be used in C++ environment?
I use Eclipse IDE in Linux Gentoo to write C++ code.
UPDATE
After reading the answers I did some tests with the software. I compiled my project with the following flags:
-O3 -mcpu=cortex-a9 -ftree-vectorize -mfloat-abi=hard -mfpu=neon
Keep in mind that this project includes extensive libraries such as open frameworks, OpenCV, and OpenNI, and everything was compiled with these flags.
To compile for the ARM board we use a Linaro toolchain cross-compiler, and GCC's version is 4.8.3.
Would you expect this to improve the performance of the project? Because we experienced no changes at all, which is rather weird considering all the answers I read here.
Another question: all the for cycles have an apparent number of iterations, but many of them iterate through custom data types (structs or classes). Can GCC optimize these cycles even though they iterate through custom data types?
EDIT:
From your update, you may misunderstand what the NEON processor does. It is an SIMD (Single Instruction, Multiple Data) vector processor. That means that it is very good at performing an instruction (say "multiply by 4") to several pieces of data at the same time. It also loves to do things like "add all these numbers together" or "add each element of these two lists of numbers to create a third list of numbers." So if you problem looks like those things the NEON processor is going to be huge help.
To get that benefit, you must put your data in very specific formats so that the vector processor can load multiple data simultaneously, process it in parallel, and then write it back out simultaneously. You need to organize things such that the math avoids most conditionals (because looking at the results too soon means a roundtrip to the NEON). Vector programming is a different way of thinking about your program. It's all about pipeline management.
Now, for many very common kinds of problems, the compiler automatically can work all of this out. But it's still about working with numbers, and numbers in particular formats. For example, you almost always need to get all of your numbers into a contiguous block in memory. If you're dealing with fields inside of structs and classes, the NEON can't really help you. It's not a general-purpose "do stuff in parallel" engine. It's an SIMD processor for doing parallel math.
For very high-performance systems, data format is everything. You don't take arbitrary data formats (structs, classes, etc.) and try to make them fast. You figure out the data format that will let you do the most parallel work, and you write your code around that. You make your data contiguous. You avoid memory allocation at all costs. But this isn't really something a simple StackOverflow question can address. High-performance programming is a whole skill set and a different way of thinking about things. It isn't something you get by finding the right compiler flag. As you've found, the defaults are pretty good already.
The real question you should be asking is whether you could reorganize your data so that you can use more of OpenCV. OpenCV already has lots of optimized parallel operations that will almost certainly make good use of the NEON. As much as possible, you want to keep your data in the format that OpenCV works in. That's likely where you're going to get your biggest improvements.
My experience is that it is certainly possible to hand-write NEON assembly that will beat clang and gcc (at least from a couple of years ago, though the compiler certainly continues to improve). Having excellent ARM optimization is not the same as NEON optimization. As #Mats notes, the compiler will generally do an excellent job at obvious cases, but does not always handle every case ideally, and it is certainly possible for even a lightly skilled developer to sometimes beat it, sometimes dramatically. (#wallyk is also correct that hand-tuning assembly is best saved for last; but it can still be very powerful.)
That said, given your statement "Assembly, for which I have absolutely no background, and can't possibly afford to learn at this point," then no, you should not even bother. Without first at least understanding the basics (and a few non-basics) of assembly (and specifically vectorized NEON assembly), there is no point in second-guessing the compiler. Step one of beating the compiler is knowing the target.
If you are willing to learn the target, my favorite introduction is Whirlwind Tour of ARM Assembly. That, plus some other references (below), were enough to let me beat the compiler by 2-3x in my particular problems. On the other hand, they were insufficient enough that when I showed my code to an experienced NEON developer, he looked at it for about three seconds and said "you have a halt right there." Really good assembly is hard, but half-decent assembly can still be better than optimized C++. (Again, every year this gets less true as the compiler writers get better, but it can still be true.)
ARM Assembly language
A few things iOS developers ought to know about the ARM architecture (iPhone-focused, but the principles are the same for all uses.)
ARM NEON support in the ARM compiler
Coding for NEON
One side note, my experience with NEON intrinsics is that they are seldom worth the trouble. If you're going to beat the compiler, you're going to need to actually write full assembly. Most of the time, whatever intrinsic you would have used, the compiler already knew about. Where you get your power is more often in restructuring your loops to best manage your pipeline (and intrinsics don't help there). It's possible this has improved over the last couple of years, but I would expect the improving vector optimizer to outpace the value of intrinsics more than the other way around.
Here's a "mee too" with some blog posts from ARM. FIRST, start with the following to get the background information, including 32-bit ARM (ARMV7 and below), Aarch32 (ARMv8 32-bit ARM) and Aarch64 (ARMv8 64-bit ARM):
ARM NEON programming quick reference
Second, checkout the Coding for NEON series. Its a nice introduction with pictures so things like interleaved loads make sense with a glance.
ARM NEON programming quick reference
Coding for NEON - Part 1: Load and Stores
Coding for NEON - Part 2: Dealing With Leftovers
Coding for NEON - Part 3: Matrix Multiplication
Coding for NEON - Part 4: Shifting Left and Right
Coding for NEON - Part 5: Rearranging Vectors
I also went on Amazon looking for some books on ARM assembly with a treatment of NEON. I could only find two, and neither book's treatment of NEON were impressive. They reduced to a single chapter with the obligatory Matrix example.
I believe ARM Intrinsics are a very good idea. The instrinsics allow you to write code for GCC, Clang and Visual C/C++ compilers. We have one code base that works for ARM Linux distros (like Linaro), some iOS devices (using -arch armv7) and Microsoft gadgets (like Windows Phone and Windows Store Apps).
If you have access to a reasonably modern GCC (GCC 4.8 and upwards) I would recommend giving intrinsics a go. The NEON intrinsics are a set of functions that the compiler knows about, which can be used from C or C++ programs to generate NEON/Advanced SIMD instructions. To gain access to them in your program, it is necessary to #include <arm_neon.h>. The verbose documentation of all available intrinsics is available at http://infocenter.arm.com/help/topic/com.arm.doc.ihi0073a/IHI0073A_arm_neon_intrinsics_ref.pdf , but you may find more user-friendly tutorials elsewhere online.
Advice on this site is generally against the NEON intrinsics, and certainly there are GCC versions which have done a poor job of implementing them, but recent versions do reasonably well (and if you spot bad code generation, please do raise it as a bug - https://gcc.gnu.org/bugzilla/ )
They are an easy way to program to the NEON/Advanced SIMD instruction set, and the performance you can achieve is often rather good. They are also "portable", in that when you move to an AArch64 system, a superset of the intrinsics you can use from ARMv7-A are available. They are also portable across implementations of the ARM architecture, which can vary in their performance characteristics, but which the compiler will model for performance tuning.
The principle benefit of the NEON intrinsics over hand-written assembly, is that the compiler can understand them when performing its various optimization passes. By contrast hand-written assembler is an opaque block to GCC, and will not be optimized. On the other hand, expert assembler programmers can often beat the compiler's register allocation policies, particularly when using the instructions which write to or read from to multiple consecutive registers.
In addition to Wally's answer - and probably should be a comment, but I couldn't make it short enough: ARM has a team of compiler developers whose entire role is to improve the parts of GCC and Clang/llvm that does code generation for ARM CPUs, including features that provides "auto-vectorization" - I have not looked deeply into it, but from my experience on x86 code generation, I'd expect for anything that is relatively easy to vectorize, the compiler should do a deecent job. Some code is hard for the compiler to understand when it can vectorize or not, and may need some "encouragement" - such as unrolling loops or marking conditions as "likely" or "unlikely", etc.
Disclaimer: I work for ARM, but have very little to do with the compilers or even CPUs, as I work for the group that does graphics (where I have some involvement with compilers for the GPUs in the OpenCL part of the GPU driver).
Edit:
Performance, and use of various instruction extensions is really depending on EXACTLY what the code is doing. I'd expect that libraries such as OpenCV is already doing a fair amount of clever stuff in their code (such as both handwritten assembler as compiler intrinsics and generally code that is designed to allow the compiler to already do a good job), so it may not really give you much improvement. I'm not a computer vision expert, so I can't really comment on exactly how much such work is done on OpenCV, but I'd certainly expect the "hottest" points of the code to have been fairly well optimised already.
Also, profile your application. Don't just fiddle with optimisation flags, measure it's performance and use a profiling tool (e.g. the Linux "perf" tool) to measure WHERE your code is spending time. Then see what can be done to that particular code. Is it possible to write a more parallel version of it? Can the compiler help, do you need to write assembler? Is there a different algorithm that does the same thing but in a better way, etc, etc...
Although tweaking compiler options CAN help, and often does, it can give tens of percent, where a change in algorithm can often lead to 10 times or 100 times faster code - assuming of course, your algorithm can be improved!
Understanding what part of your application is taking the time, however, is KEY. It's no point in changing things to make the code that takes 5% of the time 10% faster, when a change somewhere else could make a piece of code that is 30 or 60% of the total time 20% faster. Or optimise some math routine, when 80% of the time is spent on reading a file, where making the buffer twice the size would make it twice as fast...
Although a long time has passed since I submitted this question, I realize that it gathers some interest and I decided to tell what I ended up doing regarding this.
My main goal was to optimize a for-loop which was the bottleneck of the project. So, since I don't know anything about Assembly I decided to give NEON intrinsics a go. I ended up having a 40-50% gain in performance (in this loop alone), and a significant overall improvement in performance of the whole project.
The code does some math to transform a bunch of raw distance data into distance to a plane in millimetres. I use some constants (like _constant05, _fXtoZ) that are not defined here, but they are just constant values defined elsewhere.
As you can see, I'm doing the math for 4 elements at a time, talk about real parallelization :)
unsigned short* frameData = frame.ptr<unsigned short>(_depthLimits.y, _depthLimits.x);
unsigned short step = _runWidth - _actWidth; //because a ROI being processed, not the whole image
cv::Mat distToPlaneMat = cv::Mat::zeros(_runHeight, _runWidth, CV_32F);
float* fltPtr = distToPlaneMat.ptr<float>(_depthLimits.y, _depthLimits.x); //A pointer to the start of the data
for(unsigned short y = _depthLimits.y; y < _depthLimits.y + _depthLimits.height; y++)
{
for (unsigned short x = _depthLimits.x; x < _depthLimits.x + _depthLimits.width - 1; x +=4)
{
float32x4_t projX = {(float)x, (float)(x + 1), (float)(x + 2), (float)(x + 3)};
float32x4_t projY = {(float)y, (float)y, (float)y, (float)y};
framePixels = vld1_u16(frameData);
float32x4_t floatFramePixels = {(float)framePixels[0], (float)framePixels[1], (float)framePixels[2], (float)framePixels[3]};
float32x4_t fNormalizedY = vmlsq_f32(_constant05, projY, _yResInv);
float32x4_t auxfNormalizedX = vmulq_f32(projX, _xResInv);
float32x4_t fNormalizedX = vsubq_f32(auxfNormalizedX, _constant05);
float32x4_t realWorldX = vmulq_f32(fNormalizedX, floatFramePixels);
realWorldX = vmulq_f32(realWorldX, _fXtoZ);
float32x4_t realWorldY = vmulq_f32(fNormalizedY, floatFramePixels);
realWorldY = vmulq_f32(realWorldY, _fYtoZ);
float32x4_t realWorldZ = floatFramePixels;
realWorldX = vsubq_f32(realWorldX, _tlVecX);
realWorldY = vsubq_f32(realWorldY, _tlVecY);
realWorldZ = vsubq_f32(realWorldZ, _tlVecZ);
float32x4_t distAuxX, distAuxY, distAuxZ;
distAuxX = vmulq_f32(realWorldX, _xPlane);
distAuxY = vmulq_f32(realWorldY, _yPlane);
distAuxZ = vmulq_f32(realWorldZ, _zPlane);
float32x4_t distToPlane = vaddq_f32(distAuxX, distAuxY);
distToPlane = vaddq_f32(distToPlane, distAuxZ);
*fltPtr = (float) distToPlane[0];
*(fltPtr + 1) = (float) distToPlane[1];
*(fltPtr + 2) = (float) distToPlane[2];
*(fltPtr + 3) = (float) distToPlane[3];
frameData += 4;
fltPtr += 4;
}
frameData += step;
fltPtr += step;
}
If you don't want to mess with assembly code at all, then tweak the compiler flags to maximally optimize for speed. gcc given the proper ARM target should do this provided the number of loop iterations is apparent.
To check gcc code generation, request assembly output by adding the -S flag.
If after several tries (of reading the gcc documentation and tweaking flags) you still can't get it to produce the code you want, then take the assembly output and edit it to your satisfaction.
Beware of premature optimization. The proper development order is to get the code functional, then see if it needs optimization. Only when the code is stable does it makes sense to do so.
Play with some minimal assembly examples on QEMU to understand the instructions
The following setup does not have many examples yet, but it serves as a neat playground:
v7 examples
v8 examples
setup usage
The examples run on QEMU user mode, which dispenses extra hardware, and the GDB is working just fine.
The asserts are done through the C standard library.
You should be a able to easily extend that setup with new instructions as you learn them.
ARM intrinsincs in particular were asked at: Is there a good reference for ARM Neon intrinsics?

Implementing a simulator for a subset of x86

I wish to implement a simulator for a subset of instructions for the x86 architecture. Given a binary, I wish to disassemble it and run a simulation on the instructions. For that, one would need to look at certain bits of an instruction to decide whether it is a control instruction, arithmetic instruction or a logical instruction and based on that, one must derive the parameters of the operation by looking at the remaining bits. One obvious yet painful way to implement this is by using nested if-else/switch-case statements. Can someone suggest a better methodology for implementing this?
Use a lookup table, perhaps in the form of a std::map.
You can look at the source of an x86 emulator to find an implementation of this idea, already fully written and fleshed out.
Here's one you might try: http://www.dosbox.com/wiki/BuildingDOSBox#1._Grab_the_source
Let me know if this doesn't work out; there are lots to choose from.
In general, with an emulator, I would think that a switch on the opcode would be one way to go. Another good approach would be an 256-entry array of function pointers, corresponding to the first byte of the instruction. That gives a little more separation than a giant switch or if block. Of course you can reuse the functions as needed.
Doing a nested if/else type construct should be fine if you cache the output of the translation. If you are doing simulation you will have relatively few dynamic instructions out of all of the static instructions in the program. So the best performance optimization is to cache the output of the translation and then reuse it when the dynamic instruction executes. Eventually your cache will fill up and you will need to clear it for new entries. But it makes more sense to cache the translation somehow, rather than try to come up with a really fast method of doing the translation in the first place.
As an example QEMU is an emulator that supports a variety of targets that is optimized for performance. You can see how they translate x86 instructions here:
https://github.com/qemu/QEMU/blob/master/target-i386/translate.c#L4076
if QEMU did this for every instruction the performance would be very slow. But since the cache the results it does not matter too much that the first time an instruction is translated there is a complex case statement.

Equation parser efficiency

I sunk about a month of full time into a native C++ equation parser. It works, except it is slow (between 30-100 times slower than a hard-coded equation). What can I change to make it faster?
I read everything I could find on efficient code. In broad strokes:
The parser converts a string equation expression into a list of "operation" objects.
An operation object has two function pointers: a "getSource" and a "evaluate".
To evaluate an equation, all I do is a for loop on the operation list, calling each function in turn.
There isn't a single if / switch encountered when evaluating an equation - all conditionals are handled by the parser when it originally assigned the function pointers.
I tried inlining all the functions to which the function pointers point - no improvement.
Would switching from function pointers to functors help?
How about removing the function pointer framework, and instead creating a full set of derived "operation" classes, each with its own virtual "getSource" and "evaluate" functions? (But doesn't this just move the function pointers into the vtable?)
I have a lot of code. Not sure what to distill / post. Ask for some aspect of it, and ye shall receive.
In your post you don't mention that you have profiled the code. This is the first thing I would do if I were in your shoes. It'll give you a good idea of where the time is spent and where to focus your optimization efforts.
It's hard to tell from your description if the slowness includes parsing, or it is just the interpretation time.
The parser, if you write it as recursive-descent (LL1) should be I/O bound. In other words, the reading of characters by the parser, and construction of your parse tree, should take a lot less time than it takes to simply read the file into a buffer.
The interpretation is another matter.
The speed differential between interpreted and compiled code is usually 10-100 times slower, unless the basic operations themselves are lengthy.
That said, you can still optimize it.
You could profile, but in such a simple case, you could also just single-step the program, in the debugger, at the level of individual instructions.
That way, you are "walking in the computer's shoes" and it will be obvious what can be improved.
Whenever I'm doing what you're doing, that is, providing a language to the user, but I want the language to have fast execution, what I do is this:
I translate the source language into a language I have a compiler for, and then compile it on-the-fly into a .dll (or .exe) and run that.
It's very quick, and I don't need to write an interpreter or worry about how fast it is.
The very first thing is: Profile what actually went wrong. Is the bottleneck in parsing or in evaluation? valgrind offers some tools that can help you here.
If it's in parsing, boost::spirit might help you. If in evaluation, remember that virtual functions can be pretty slow to evaluate. I've made pretty good experiences with recursive boost::variant's.
You know, building an expression recursive descent parser is really easy, the LL(1) grammar for expressions is only a couple of rules. Parsing then becomes a linear affair and everything else can work on the expression tree (while parsing basically); you'd collect the data from the lower nodes and pass it up to the higher nodes for aggregation.
This would avoid altogether function/class pointers to determine the call path at runtime, relying instead of proven recursivity (or you can build an iterative LL parser if you wish).
It seems that you're using a quite complicated data structure (as I understand it, a syntax tree with pointers etc.). Thus, walking through pointer dereference is not very efficient memory-wise (lots of random accesses) and could slow you down significantly. As Mike Dunlavey proposed, you could compile the whole expression at runtime using another language or by embedding a compiler (such as LLVM). For what I know, Microsoft .Net provides this feature (dynamic compilation) with Reflection.Emit and Linq.Expression trees.
This is one of those rare times that I'd advise against profiling just yet. My immediate guess is that the basic structure you're using is the real source of the problem. Profiling the code is rarely worth much until you're reasonably certain the basic structure is reasonable, and it's mostly a matter of finding which parts of that basic structure can be improved. It's not so useful when what you really need to do is throw out most of what you have, and basically start over.
I'd advise converting the input to RPN. To execute this, the only data structure you need is a stack. Basically, when you get to an operand, you push it on the stack. When you encounter an operator, it operates on the items at the top of the stack. When you're done evaluating a well-formed expression, you should have exactly one item on the stack, which is the value of the expression.
Just about the only thing that will usually give better performance than this is to do like #Mike Dunlavey advised, and just generate source code and run it through a "real" compiler. That is, however, a fairly "heavy" solution. If you really need maximum speed, it's clearly the best solution -- but if you just want to improve what you're doing now, converting to RPN and interpreting that will usually give a pretty decent speed improvement for a small amount of code.