Matrix multiplication via std::vector is 10 times slower than numpy - performance

Although it is known that using nested std::vector to represent matrices is a bad idea, let's use it for now since it is flexible and many existing functions can handle std::vector.
I thought, in small cases, the speed difference can be ignored. But it turned out that vector<vector<double>> is 10+ times slower than numpy.dot().
Let A and B be matrices whose size is sizexsize. Assuming square matrices is just for simplicity. (We don't intend to limit discussion to the square matrices case.) We initialize each matrix in a deterministic way, and finally calculate C = A * B.
We define "calculation time" as the time elapsed just to calculate C = A * B. In other words, various overheads are not included.
Python3 code
import numpy as np
import time
import sys
if (len(sys.argv) != 2):
print("Pass `size` as an argument.", file = sys.stderr);
sys.exit(1);
size = int(sys.argv[1]);
A = np.ndarray((size, size));
B = np.ndarray((size, size));
for i in range(size):
for j in range(size):
A[i][j] = i * 3.14 + j
B[i][j] = i * 3.14 - j
start = time.time()
C = np.dot(A, B);
print("{:.3e}".format(time.time() - start), file = sys.stderr);
C++ code
using namespace std;
#include <iostream>
#include <vector>
#include <chrono>
int main(int argc, char **argv) {
if (argc != 2) {
cerr << "Pass `size` as an argument.\n";
return 1;
}
const unsigned size = atoi(argv[1]);
vector<vector<double>> A(size, vector<double>(size));
vector<vector<double>> B(size, vector<double>(size));
for (int i = 0; i < size; ++i) {
for (int j = 0; j < size; ++j) {
A[i][j] = i * 3.14 + j;
B[i][j] = i * 3.14 - j;
}
}
auto start = chrono::system_clock::now();
vector<vector<double>> C(size, vector<double>(size, /* initial_value = */ 0));
for (int i = 0; i < size; ++i) {
for (int j = 0; j < size; ++j) {
for (int k = 0; k < size; ++k) {
C[i][j] += A[i][k] * B[k][j];
}
}
}
cerr << scientific;
cerr.precision(3);
cerr << chrono::duration<double>(chrono::system_clock::now() - start).count() << "\n";
}
C++ code (multithreaded)
We also wrote a multithreaded version of C++ code since numpy.dot() is automatically calculated in parallel.
You can get all the codes from GitHub.
Result
C++ version is 10+ times slower than Python 3 (with numpy) version.
matrix_size: 200x200
--------------- Time in seconds ---------------
C++ (not multithreaded): 8.45e-03
C++ (1 thread): 8.66e-03
C++ (2 threads): 4.68e-03
C++ (3 threads): 3.14e-03
C++ (4 threads): 2.43e-03
Python 3: 4.07e-04
-----------------------------------------------
matrix_size: 400x400
--------------- Time in seconds ---------------
C++ (not multithreaded): 7.011e-02
C++ (1 thread): 6.985e-02
C++ (2 threads): 3.647e-02
C++ (3 threads): 2.462e-02
C++ (4 threads): 1.915e-02
Python 3: 1.466e-03
-----------------------------------------------
Question
Is there any way to make the C++ implementation faster?
Optimizations I Tried
swap calculation order -> at most 3.5 times faster (not than numpy code but than C++ code)
optimization 1 plus partial unroll -> at most 4.5 times faster, but this can be done only when size is known in advance No. As pointed out in this comment, size is not needed to be known. We can just limit the max value of loop variables of unrolled loops and process remaining elements with normal loops. See my implementation for example.
optimization 2, plus minimizing the call of C[i][j] by introducing a simple variable sum -> at most 5.2 times faster. The implementation is here. This result implies std::vector::operator[] is un-ignorably slow.
optimization 3, plus g++ -march=native flag -> at most 6.2 times faster (By the way, we use -O3 of course.)
Optimization 3, plus reducing the call of operator [] by introducing a pointer to an element of A since A's elements are sequentially accessed in the unrolled loop. -> At most 6.2 times faster, and a little little bit faster than Optimization 4. The code is shown below.
g++ -funroll-loops flag to unroll for loops -> no change
g++ #pragma GCC unroll n -> no change
g++ -flto flag to turn on link time optimizations -> no change
Block Algorithm -> no change
transpose B to avoid cache miss -> no change
long linear std::vector instead of nested std::vector<std::vector>, swap calculation order, block algorithm, and partial unroll -> at most 2.2 times faster
Optimization 1, plus PGO(profile-guided optimization) -> 4.7 times faster
Optimization 3, plus PGO -> same as Optimization 3
Optimization 3, plus g++ specific __builtin_prefetch() -> same as Optimization 3
Current Status
(originally) 13.06 times slower -> (currently) 2.10 times slower
Again, you can get all the codes on GitHub. But let us cite some codes, all of which are functions called from the multithreaded version of C++ code.
Original Code (GitHub)
void f(const vector<vector<double>> &A, const vector<vector<double>> &B, vector<vector<double>> &C, unsigned row_start, unsigned row_end) {
const unsigned j_max = B[0].size();
const unsigned k_max = B.size();
for (int i = row_start; i < row_end; ++i) {
for (int j = 0; j < j_max; ++j) {
for (int k = 0; k < k_max; ++k) {
C[i][j] += A[i][k] * B[k][j];
}
}
}
}
Current Best Code (GitHub)
This is the implementation of the Optimization 5 above.
void f(const vector<vector<double>> &A, const vector<vector<double>> &B, vector<vector<double>> &C, unsigned row_start, unsigned row_end) {
static const unsigned num_unroll = 5;
const unsigned j_max = B[0].size();
const unsigned k_max_for_unrolled_loop = B.size() / num_unroll * num_unroll;
const unsigned k_max = B.size();
for (int i = row_start; i < row_end; ++i) {
for (int k = 0; k < k_max_for_unrolled_loop; k += num_unroll) {
for (int j = 0; j < j_max; ++j) {
const double *p = A[i].data() + k;
double sum;
sum = *p++ * B[k][j];
sum += *p++ * B[k+1][j];
sum += *p++ * B[k+2][j];
sum += *p++ * B[k+3][j];
sum += *p++ * B[k+4][j];
C[i][j] += sum;
}
}
for (int k = k_max_for_unrolled_loop; k < k_max; ++k) {
const double a = A[i][k];
for (int j = 0; j < j_max; ++j) {
C[i][j] += a * B[k][j];
}
}
}
}
We've tried many optimizations since we first posted this question. We spent whole two days struggling with this problem, and finally reached the point where we have no more idea how to optimize the current best code. We doubt more complex algorithms like Strassen's will do it better since cases we handle are not large and each operation on std::vector is so expensive that, as we've seen, just reducing the call of [] improved the performance well.
We (want to) believe we can make it better, though.

Matrix multiplication is relativly easy to optimize. However if you want to get to decent cpu utilization it becomes tricky because you need deep knowledge of the hardware you are using. The steps to implement a fast matmul kernel are the following:
Use SIMDInstructions
Use Register Blocking and fetch multiple data at once
Optimize for your chache lines (mainly L2 and L3)
Parallelize your code to use multiple threads
Under this linke is a very good ressource, that explains all the nasty details:
https://gist.github.com/nadavrot/5b35d44e8ba3dd718e595e40184d03f0
If you want more indepth advise leave a comment.

Related

Why can't clang and gcc optimize away this int-to-float conversion?

Consider the following code:
void foo(float* __restrict__ a)
{
int i; float val;
for (i = 0; i < 100; i++) {
val = 2 * i;
a[i] = val;
}
}
void bar(float* __restrict__ a)
{
int i; float val = 0.0;
for (i = 0; i < 100; i++) {
a[i] = val;
val += 2.0;
}
}
They're based on Examples 7.26a and 7.26b in Agner Fog's Optimizing software in C++ and should do the same thing; bar is more "efficient" as written in the sense that we don't do an integer-to-float conversion at every iteration, but rather a float addition which is cheaper (on x86_64).
Here are the clang and gcc results on these two functions (with no vectorization and unrolling).
Question: It seems to me that the optimization of replacing a multiplication by the loop index with an addition of a constant value - when this is beneficial - should be carried out by compilers, even if (or perhaps especially if) there's a type conversion involved. Why is this not happening for these two functions?
Note that if we use int's rather than float's:
void foo(int* __restrict__ a)
{
int i; int val = 0;
for (i = 0; i < 100; i++) {
val = 2 * i;
a[i] = val;
}
}
void bar(int* __restrict__ a)
{
int i; int val = 0;
for (i = 0; i < 100; i++) {
a[i] = val;
val += 2;
}
}
Both clang and gcc perform the expected optimization, albeit not quite in the same way (see this question).
You are looking for enabling induction variable optimization for floating point numbers. This optimization is generally unsafe in floating point land as it changes program semantics. In your example it'll work because both initial value (0.0) and step (2.0) can be precisely represented in IEEE format but this is a rare case in practice.
It could be enabled under -ffast-math but it seems this wasn't considered as important case in GCC as it rejects non-integral induction variables early on (see tree-scalar-evolution.c).
If you believe that this is an important usecase you might consider filing request at GCC Bugzilla.

OpenMP Code Not Scaling due to overheads and cache issues

struct xnode
{
float *mat;
};
void testScaling( )
{
int N = 1000000; ///total num matrices
int dim = 10;
//memory for matrices
std::vector<xnode> nodeArray(N);
for( int k = 0; k < N; ++k )
nodeArray[k].mat = new float [dim*dim];
//memory for Y
std::vector<float*> Y(N,0);
for( int k = 0; k < N; ++k )
Y[k] = new float [dim];
//shared X
float* X = new float [dim];
for(int i = 0; i < dim; ++i ) X[i] = 1.0;
//init mats
for( int k = 0; k < N; ++k )
{
for( int i=0; i<dim*dim; ++i )
nodeArray[k].mat[i] = 0.25+((float)i)/3;
}
int NTIMES = 500;
//gemv args
char trans = 'N';
int lda = dim;
int incx = 1;
float alpha =1 , beta = 0;
//threads
int thr[4];
thr[0] =1 ; thr[1] = 2; thr[2] = 4; thr[3] = 8;
for( int t = 0; t<4; ++t )//test for nthreads
{
int nthreads = thr[t];
double t_1 = omp_get_wtime();
for( int ii = 0; ii < NTIMES; ++ii )//do matvec NTIMES
{
#pragma omp parallel for num_threads(nthreads)
for( int k=0; k<N; ++k )
{
//compute Y[k] = mat[k] * X;
GEMV(&trans, &dim, &dim, &alpha, nodeArray[k].mat, &lda, X, &incx, &beta, Y[k], &incx);
//GEMV(&trans, &dim, &dim, &alpha, nodeArray[0].mat, &lda, X, &incx, &beta, Y[k], &incx);
}
}
double t_2 = omp_get_wtime();
std::cout << "Threads " << nthreads << " time " << (t_2-t_1)/NTIMES << std::endl;
}
//clear memory
for( int k = 0; k < N; ++k )
{
delete [] nodeArray[k].mat;
delete [] Y[k];
}
delete [] X;
}
The above code parallelizes the matrix-vector product of N matrices of size dim, and stores results in N output vectors. The average of 500 products is taken as the time per matrix-vector product. The matrix-vector products in the above example are all of equal size and thus the threads should be perfectly balanced - we should achieve a performance scaling close to ideal 8x. The following are the observations (Machine – Intel Xeon 3.1Ghz.2 processors,8cores each, HyperThreading enabled, Windows, VS2012, Intel MKL, Intel OMP library).
OBSERVATION 1:
dim=10 N=1000000
Threads 1 - time 0.138068s
Threads 2 - time 0.0729147s
Threads 4 - time 0.0360527s
Threads 8 - time 0.0224268s (6.1x on 8threads)
OBSERVATION 2 :
dim=20 N=1000000
Threads 1 time 0.326617
Threads 2 time 0.185706
Threads 4 time 0.0886508
Threads 8 time 0.0733666 (4.5x on 8 threads).
Note – I ran VTune on this case. It showed CPUTime 267.8sec, Overhead time 43 sec, Spin time – 8 sec. The overhead time is all spent in a libiomp function (intel library). 8Threads/1Thread scaling is poor for such cases.
Next - in the gemv for loop, we change nodeArray[k].mat to nodeArray[0].mat (see commented statement), so that only the first matrix is used for all the matrix-vector products.
OBSERVATION 3
dim=20 N=1000000
Threads 1 time 0.152298 (The serial time is halved)
Threads 2 time 0.0769173
Threads 4 time 0.0384086
Threads 8 time 0.019336 (7.87x on 8 threads)
Thus I get almost ideal scaling - why is this behavior? VTune says that a significant portion of CPU time is spent in synchronization and thread overhead. Here it seems there is no relation between the load balancing and thread synchronization. As matrix size is increased the granularity should increase and thread overhead should be proportionately small. But as we increase from size 10 to 20 the scaling is weakening. When we use nodeArray[0].mat (only the first matrix) for doing all the matrix-vector products the cache is updated only once (since the compiler knows this during optimization) and we get near ideal scaling. Thus the synchronization overhead seems to be related to some cache related issue. I have tried a number of other things like setting KMP_AFFINITY and varying load distribution but that did not buy me anything.
My questions are:
1. I dont have a clear idea about how does the cache performance affect openMP thread synchronization. Can someone explain this?
2. Can anything be done about improving the scaling and reducing the overhead?
Thanks

Improving the Efficiency of Compact/Scatter in CUDA

Summary:
Any ideas about how to further improve upon the basic scatter operation in CUDA? Especially if one knows it will only be used to compact a larger array into a smaller one? or why the below methods of vectorizing memory ops and shared memory didn't work? I feel like there may be something fundamental I am missing and any help would be appreciated.
EDIT 03/09/15: So I found this Parallel For All Blog post "Optimized Filtering with Warp-Aggregated Atomics". I had assumed atomics would be intrinsically slower for this purpose, however I was wrong - especially since I don't think I care about maintaining element order in the array during my simulation. I'll have to think about it some more and then implement it to see what happens!
EDIT 01/04/16: I realized I never wrote about my results. Unfortunately in that Parallel for All Blog post they compared the global atomic method for compact to the Thrust prefix-sum compact method, which is actually quite slow. CUB's Device::IF is much faster than Thrust's - as is the prefix-sum version I wrote using CUB's Device::Scan + custom code. The warp-aggregrate global atomic method is still faster by about 5-10%, but nowhere near the 3-4x faster I had been hoping for based on the results in the blog. I'm still using the prefix-sum method as while maintaining element order is not necessary, I prefer the consistency of the prefix-sum results and the advantage from the atomics is not very big. I still try various methods to improve compact, but so far only marginal improvements (2%) at best for dramatically increased code complexity.
Details:
I am writing a simulation in CUDA where I compact out elements I am no longer interested in simulating every 40-60 time steps. From profiling it seems that the scatter op takes up the most amount of time when compacting - more so than the filter kernel or the prefix sum. Right now I use a pretty basic scatter function:
__global__ void scatter_arrays(float * new_freq, const float * const freq, const int * const flag, const int * const scan_Index, const int freq_Index){
int myID = blockIdx.x*blockDim.x + threadIdx.x;
for(int id = myID; id < freq_Index; id+= blockDim.x*gridDim.x){
if(flag[id]){
new_freq[scan_Index[id]] = freq[id];
}
}
}
freq_Index is the number of elements in the old array. The flag array is the result from the filter. Scan_ID is the result from the prefix sum on the flag array.
Attempts I've made to improve it are to read the flagged frequencies into shared memory first and then write from shared memory to global memory - the idea being that the writes to global memory would be more coalesced amongst the warps (e.g. instead of thread 0 writing to position 0 and thread 128 writing to position 1, thread 0 would write to 0 and thread 1 would write to 1). I also tried vectorizing the reads and the writes - instead of reading and writing floats/ints I read/wrote float4/int4 from the global arrays when possible, so four numbers at a time. This I thought might speed up the scatter by having fewer memory ops transferring larger amounts of memory. The "kitchen sink" code with both vectorized memory loads/stores and shared memory is below:
const int compact_threads = 256;
__global__ void scatter_arrays2(float * new_freq, const float * const freq, const int * const flag, const int * const scan_Index, const int freq_Index){
int gID = blockIdx.x*blockDim.x + threadIdx.x; //global ID
int tID = threadIdx.x; //thread ID within block
__shared__ float row[4*compact_threads];
__shared__ int start_index[1];
__shared__ int end_index[1];
float4 myResult;
int st_index;
int4 myFlag;
int4 index;
for(int id = gID; id < freq_Index/4; id+= blockDim.x*gridDim.x){
if(tID == 0){
index = reinterpret_cast<const int4*>(scan_Index)[id];
myFlag = reinterpret_cast<const int4*>(flag)[id];
start_index[0] = index.x;
st_index = index.x;
myResult = reinterpret_cast<const float4*>(freq)[id];
if(myFlag.x){ row[0] = myResult.x; }
if(myFlag.y){ row[index.y-st_index] = myResult.y; }
if(myFlag.z){ row[index.z-st_index] = myResult.z; }
if(myFlag.w){ row[index.w-st_index] = myResult.w; }
}
__syncthreads();
if(tID > 0){
myFlag = reinterpret_cast<const int4*>(flag)[id];
st_index = start_index[0];
index = reinterpret_cast<const int4*>(scan_Index)[id];
myResult = reinterpret_cast<const float4*>(freq)[id];
if(myFlag.x){ row[index.x-st_index] = myResult.x; }
if(myFlag.y){ row[index.y-st_index] = myResult.y; }
if(myFlag.z){ row[index.z-st_index] = myResult.z; }
if(myFlag.w){ row[index.w-st_index] = myResult.w; }
if(tID == blockDim.x -1 || gID == mutations_Index/4 - 1){ end_index[0] = index.w + myFlag.w; }
}
__syncthreads();
int count = end_index[0] - st_index;
int rem = st_index & 0x3; //equivalent to modulo 4
int offset = 0;
if(rem){ offset = 4 - rem; }
if(tID < offset && tID < count){
new_mutations_freq[population*new_array_Length+st_index+tID] = row[tID];
}
int tempID = 4*tID+offset;
if((tempID+3) < count){
reinterpret_cast<float4*>(new_freq)[tID] = make_float4(row[tempID],row[tempID+1],row[tempID+2],row[tempID+3]);
}
tempID = tID + offset + (count-offset)/4*4;
if(tempID < count){ new_freq[st_index+tempID] = row[tempID]; }
}
int id = gID + freq_Index/4 * 4;
if(id < freq_Index){
if(flag[id]){
new_freq[scan_Index[id]] = freq[id];
}
}
}
Obviously it gets a bit more complicated. :) While the above kernel seems stable when there are hundreds of thousands of elements in the array, I've noticed a race condition when the array numbers in the tens of millions. I'm still trying to track the bug down.
But regardless, neither method (shared memory or vectorization) together or alone improved performance. I was especially surprised by the lack of benefit from vectorizing the memory ops. It had helped in other functions I had written, though now I am wondering if maybe it helped because it increased Instruction-Level-Parallelism in the calculation steps of those other functions rather than the fewer memory ops.
I found the algorithm mentioned in this poster (similar algorithm also discussed in this paper) works pretty well, especially for compacting large arrays. It uses less memory to do it and is slightly faster than my previous method (5-10%). I put in a few tweaks to the poster's algorithm: 1) eliminating the final warp shuffle reduction in phase 1, can simply sum the elements as they are calculated, 2) giving the function the ability to work over more than just arrays sized as a multiple of 1024 + adding grid-strided loops, and 3) allowing each thread to load their registers simultaneously in phase 3 instead of one at a time. I also use CUB instead of Thrust for Inclusive sum for faster scans. There may be more tweaks I can make, but for now this is good.
//kernel phase 1
int myID = blockIdx.x*blockDim.x + threadIdx.x;
//padded_length is nearest multiple of 1024 > true_length
for(int id = myID; id < (padded_length >> 5); id+= blockDim.x*gridDim.x){
int lnID = threadIdx.x % warp_size;
int warpID = id >> 5;
unsigned int mask;
unsigned int cnt=0;//;//
for(int j = 0; j < 32; j++){
int index = (warpID<<10)+(j<<5)+lnID;
bool pred;
if(index > true_length) pred = false;
else pred = predicate(input[index]);
mask = __ballot(pred);
if(lnID == 0) {
flag[(warpID<<5)+j] = mask;
cnt += __popc(mask);
}
}
if(lnID == 0) counter[warpID] = cnt; //store sum
}
//kernel phase 2 -> CUB Inclusive sum transforms counter array to scan_Index array
//kernel phase 3
int myID = blockIdx.x*blockDim.x + threadIdx.x;
for(int id = myID; id < (padded_length >> 5); id+= blockDim.x*gridDim.x){
int lnID = threadIdx.x % warp_size;
int warpID = id >> 5;
unsigned int predmask;
unsigned int cnt;
predmask = flag[(warpID<<5)+lnID];
cnt = __popc(predmask);
//parallel prefix sum
#pragma unroll
for(int offset = 1; offset < 32; offset<<=1){
unsigned int n = __shfl_up(cnt, offset);
if(lnID >= offset) cnt += n;
}
unsigned int global_index = 0;
if(warpID > 0) global_index = scan_Index[warpID - 1];
for(int i = 0; i < 32; i++){
unsigned int mask = __shfl(predmask, i); //broadcast from thread i
unsigned int sub_group_index = 0;
if(i > 0) sub_group_index = __shfl(cnt, i-1);
if(mask & (1 << lnID)){
compacted_array[global_index + sub_group_index + __popc(mask & ((1 << lnID) - 1))] = input[(warpID<<10)+(i<<5)+lnID];
}
}
}
}
EDIT: There is a newer article by a subset of the poster authors where they examine a faster variation of compact than what is written above. However, their new version is not order preserving, so not useful for myself and I haven't implemented it to test it out. That said, if your project doesn't rely on object order, their newer compact version can probably speed up your algorithm.

rewriting a simple C++ Code snippet into CUDA Code

I have written the following simple C++ code.
#include <iostream>
#include <omp.h>
int main()
{
int myNumber = 0;
int numOfHits = 0;
cout << "Enter my Number Value" << endl;
cin >> myNumber;
#pragma omp parallel for reduction(+:numOfHits)
for(int i = 0; i <= 100000; ++i)
{
for(int j = 0; j <= 100000; ++j)
{
for(int k = 0; k <= 100000; ++k)
{
if(i + j + k == myNumber)
numOfHits++;
}
}
}
cout << "Number of Hits" << numOfHits << endl;
return 0;
}
As you can see I use OpenMP to parallelize the outermost loop. What I would like to do is to rewrite this small code in CUDA. Any help will be much appreciated.
Well, I can give you a quick tutorial, but I won't necessarily write it all for you.
So first of all, you will want to get MS Visual Studio set up with CUDA, which is easy following this guide: http://www.ademiller.com/blogs/tech/2011/05/visual-studio-2010-and-cuda-easier-with-rc2/
Now you will want to read The NVIDIA CUDA Programming Guide (free pdf), documentation, and CUDA by Example (A book I highly recommend for learning CUDA).
But let's say you haven't done that yet, and definitely will later.
This is an extremely arithmetic heavy and data-light computation - actually it can be computed without this brute force method fairly simply, but that isn't the answer you are looking for. I suggest something like this for the kernel:
__global__ void kernel(int* myNumber, int* numOfHits){
//a shared value will be stored on-chip, which is beneficial since this is written to multiple times
//it is shared by all threads
__shared__ int s_hits = 0;
//this identifies the current thread uniquely
int i = (threadIdx.x + blockIdx.x*blockDim.x);
int j = (threadIdx.y + blockIdx.y*blockDim.y);
int k = 0;
//we increment i and j by an amount equal to the number of threads in one dimension of the block, 16 usually, times the number of blocks in one dimension, which can be quite large (but not 100,000)
for(; i < 100000; i += blockDim.x*gridDim.x){
for(; j < 100000; j += blockDim.y*gridDim.y){
//Thanks to talonmies for this simplification
if(0 <= (*myNumber-i-j) && (*myNumber-i-j) < 100000){
//you should actually use atomics for this
//otherwise, the value may change during the 'read, modify, write' process
s_hits++;
}
}
}
//synchronize threads, so we now s_hits is completely updated
__syncthreads();
//again, atomics
//we make sure only one thread per threadblock actually adds in s_hits
if(threadIdx.x == 0 && threadIdx.y == 0)
*numOfHits += s_hits;
return;
}
To launch the kernel, you will want something like this:
dim3 blocks(some_number, some_number, 1); //some_number should be hand-optimized
dim3 threads(16, 16, 1);
kernel<<<blocks, threads>>>(/*args*/);
I know you probably want a quick way to do this, but getting into CUDA isn't really a 'quick' thing. As in, you will need to do some reading and some setup to get it working; past that, the learning curve isn't too high. I haven't told you anything about memory allocation yet, so you will need to do that (although that is simple). If you followed my code, my goal is that you had to read up a bit on shared memory and CUDA, and so you are already kick-started. Good luck!
Disclaimer: I haven't tested my code, and I am not an expert - it could be idiotic.

Fast Modulo 511 and 127

Is there a way, how to make modulo by 511 (and 127) faster than using "%" operator ?
int c = 758 % 511;
int d = 423 % 127;
Here is a way to do fast modulo by 511 assuming that x is at most 32767. It's about twice as fast as x%511. It does the modulo in five steps: two multiply, two addition, one shift.
inline int fast_mod_511(int x) {
int y = (513*x+64)>>18;
return x - 511*y;
}
Here is the theory at how I arrive at this. I posted the code I tested this at the end
Let's consider
y = x/511 = x/(512-1) = x/1000 * 1/(1-1/512).
Let's define z = 512, then
y = x/z*1/(1-1/z).
Using Taylor expansion
y = x/z(1 + 1/z + 1/z^2 + 1/z^3 + ...).
Now if we know that x has a limited range we can cut the expansion. Let's assume x is always less than 2^15=32768. Then we can write
512*512*y = (1+512)*x = 513*x.
After looking at the digits which are significant we arrive at
y = (513*x+64)>>18 //512^2 = 2^18.
We can divide x/511 (assuming x is less than 32768) in three steps:
multiply,
add,
shift.
Here is the code I just to profile this in MSVC2013 64-bit release mode on an Ivy Bridge core.
#include <stdio.h>
#include <stdlib.h>
#include <omp.h>
inline int fast_mod_511(int x) {
int y = (513*x+64)>>18;
return x - 511*y;
}
int main() {
unsigned int i, x;
volatile unsigned int r;
double dtime;
dtime = omp_get_wtime();
for(i=0; i<100000; i++) {
for(int j=0; j<32768; j++) {
r = j%511;
}
}
dtime =omp_get_wtime() - dtime;
printf("time %f\n", dtime);
dtime = omp_get_wtime();
for(i=0; i<100000; i++) {
for(int j=0; j<32768; j++) {
r = fast_mod_511(j);
}
}
dtime =omp_get_wtime() - dtime;
printf("time %f\n", dtime);
}
You can use a lookup table with the solutions pre-stored. If you create an array of a million integers looking up is about twice as fast as actually doing modulo in my C# app.
// fill an array
var mod511 = new int[1000000];
for (int x = 0; x < 1000000; x++) mod511[x] = x % 511;
and instead of using
c = 758 % 511;
you use
c = mod511[758];
This will cost you (possibly a lot of) memory, and will obviously not work if you want to use it for very large numbers also. But it is faster.
If you have to repeat those two modulus operations on a large number of data and your CPU supports SIMD (for example Intel's SSE/AVX/AVX2) then you can vectorize the operations, i.e., do the operations on many data in parallel. You can do this by using intrinsics or inline assembly. Yes the solution will be platform specific but maybe that is fine...

Resources