I have a loop and inside a have a inner loop. How can I optimise it please in order to optimise execution time like avoiding accessing to memory many times to the same thing and avoid the maximum possible the addition and multiplication.
int n,m,x1,y1,x2,y2,cnst;
int N = 9600;
int M = 1800;
int temp11,temp12,temp13,temp14;
int temp21,temp22,temp23,temp24;
int *arr1 = new int [32000]; // suppose it's already filled
int *arr2 = new int [32000];// suppose it's already filled
int sumFirst = 0;
int maxFirst = 0;
int indexFirst = 0;
int sumSecond = 0;
int maxSecond = 0;
int indexSecond = 0;
int jump = 2400;
for( n = 0; n < N; n++)
{
temp14 = 0;
temp24 = 0;
for( m = 0; m < M; m++)
{
x1 = m + cnst;
y1 = m + n + cnst;
temp11 = arr1[x1];
temp12 = arr2[y1];
temp13 = temp11 * temp12;
temp14+= temp13;
x2 = m + cnst + jump;
y2 = m + n + cnst + jump;
temp21 = arr1[x2];
temp22 = arr2[y2];
temp23 = temp21 * temp22;
temp24+= temp23;
}
sumFirst += temp14;
if (temp14 > maxFirst)
{
maxFirst = temp14;
indexFirst = m;
}
sumSecond += temp24;
if (temp24 > maxSecond)
{
maxSecond = temp24;
indexSecond = n;
}
}
// At the end we use sum , index and max for first and second;
You are multiplying array elements and accumulating the result.
This can be optimized by:
SIMD (doing multiple operations at a single CPU step)
Parallel execution (using multiple physical/logical CPUs at once)
Look for CPU-specific SIMD way of doing this. Like _mm_mul_epi32 from SSE4.1 can possibly be used on x86-64. Before trying to write your own SIMD version with compiler intrinsics, make sure the compiler doesn't do it already for you.
As for parallel execution, look into omp, or using C++17 parallel accumulate.
Related
This question already has answers here:
C: using clock() to measure time in multi-threaded programs
(2 answers)
Closed 2 years ago.
I am implementing pattern matching algorithm, by moving template gradient info over entire target's gradient image , that too at each rotation (-60 to 60). I have already saved the template info for each rotation ,i.e. 121 templates are already preprocessed and saved.
But the issue is, this is consuming lot of time (approx 110ms), so decided to split the matching at set of rotations (-60 to -30 , -30 to 0, 0 to 30 and 30 to 60) into 4 threads, but threading is taking more time that single process (approx 115ms to 120ms).
Snippet of code is...
#define MAXTARGETNUM 64
MatchResultA totalResultsTemp[MAXTARGETNUM];
void CShapeMatch::match(ShapeInfo *ShapeInfoVec, search_region SearchRegion, float MinScore, float Greediness, int width,int height, int16_t *pBufGradX ,int16_t *pBufGradY,float *pBufMag, bool corr)
{
MatchResultA resultsPerDeg[MAXTARGETNUM];
....
....
int startX = SearchRegion.StartX;
int startY = SearchRegion.StartY;
int endX = SearchRegion.EndX;
int endY = SearchRegion.EndY;
float AngleStep = SearchRegion.AngleStep;
float AngleStart = SearchRegion.AngleStart;
float AngleStop = SearchRegion.AngleStop;
int startIndex = (int)(ShapeInfoVec[0].AngleNum/2) + ShapeInfoVec[0].AngleNum%2+(int)AngleStart/AngleStep;
int stopIndex = (int)(ShapeInfoVec[0].AngleNum/2) + ShapeInfoVec[0].AngleNum%2+(int)AngleStop/AngleStep;
for (int k = startIndex; k < stopIndex ; k++){
....
for(int j = startY; j < endY; j++){
for(int i = startX; i < endX; i++){
for(int m = 0; m < ShapeInfoVec[k].NoOfCordinates; m++)
{
curX = i + (ShapeInfoVec[k].Coordinates + m)->x; // template X coordinate
curY = j + (ShapeInfoVec[k].Coordinates + m)->y ; // template Y coordinate
iTx = *(ShapeInfoVec[k].EdgeDerivativeX + m); // template X derivative
iTy = *(ShapeInfoVec[k].EdgeDerivativeY + m); // template Y derivative
iTm = *(ShapeInfoVec[k].EdgeMagnitude + m); // template gradients magnitude
if(curX < 0 ||curY < 0||curX > width-1 ||curY > height-1)
continue;
offSet = curY*width + curX;
iSx = *(pBufGradX + offSet); // get corresponding X derivative from source image
iSy = *(pBufGradY + offSet); // get corresponding Y derivative from source image
iSm = *(pBufMag + offSet);
if (PartialScore > MinScore)
{
float Angle = ShapeInfoVec[k].Angel;
bool hasFlag = false;
for(int n = 0; n < resultsNumPerDegree; n++)
{
if(abs(resultsPerDeg[n].CenterLocX - i) < 5 && abs(resultsPerDeg[n].CenterLocY - j) < 5)
{
hasFlag = true;
if(resultsPerDeg[n].ResultScore < PartialScore)
{
resultsPerDeg[n].Angel = Angle;
resultsPerDeg[n].CenterLocX = i;
resultsPerDeg[n].CenterLocY = j;
resultsPerDeg[n].ResultScore = PartialScore;
break;
}
}
}
if(!hasFlag)
{
resultsPerDeg[resultsNumPerDegree].Angel = Angle;
resultsPerDeg[resultsNumPerDegree].CenterLocX = i;
resultsPerDeg[resultsNumPerDegree].CenterLocY = j;
resultsPerDeg[resultsNumPerDegree].ResultScore = PartialScore;
resultsNumPerDegree ++;
}
minScoreTemp = minScoreTemp < PartialScore ? PartialScore : minScoreTemp;
}
}
}
for(int i = 0; i < resultsNumPerDegree; i++)
{
mtx.lock();
totalResultsTemp[totalResultsNum] = resultsPerDeg[i];
totalResultsNum++;
mtx.unlock();
}
n++;
}
void CallerFunction(){
int16_t *pBufGradX = (int16_t *) malloc(bufferSize * sizeof(int16_t));
int16_t *pBufGradY = (int16_t *) malloc(bufferSize * sizeof(int16_t));
float *pBufMag = (float *) malloc(bufferSize * sizeof(float));
clock_t start = clock();
float temp_stop = SearchRegion->AngleStop;
SearchRegion->AngleStop = -30;
thread t1(&CShapeMatch::match, this, ShapeInfoVec, *SearchRegion, MinScore, Greediness, width, height, pBufGradX ,pBufGradY,pBufMag, corr);
SearchRegion->AngleStart = -30;
SearchRegion->AngleStop=0;
thread t2(&CShapeMatch::match, this, ShapeInfoVec, *SearchRegion, MinScore, Greediness, width, height, pBufGradX ,pBufGradY,pBufMag, corr);
SearchRegion->AngleStart = 0;
SearchRegion->AngleStop=30;
thread t3(&CShapeMatch::match, this, ShapeInfoVec, *SearchRegion, MinScore, Greediness,width, height, pBufGradX ,pBufGradY,pBufMag, corr);
SearchRegion->AngleStart = 30;
SearchRegion->AngleStop=temp_stop;
thread t4(&CShapeMatch::match, this, ShapeInfoVec, *SearchRegion, MinScore, Greediness,width, height, pBufGradX ,pBufGradY,pBufMag, corr);
t1.join();
t2.join();
t3.join();
t4.join();
clock_t end = clock();
cout << 1000*(double)(end-start)/CLOCKS_PER_SEC << endl;
}
As we can see there are plenty of heap access but they just are read-only. Only totalResultTemp and totalResultNum are shared global resource on which write are performed.
My PC configuration is,
i5-7200U CPU # 2.50GHz 4 cores
4 Gig RAM
Ubuntu 18
for(int i = 0; i < resultsNumPerDegree; i++)
{
mtx.lock();
totalResultsTemp[totalResultsNum] = resultsPerDeg[i];
totalResultsNum++;
mtx.unlock();
}
You writing into static array, and mutexes are really time consuming. Instead of creating locks try to use std::atomic_int, or in my opinion even better, just pass to function exact place where to store result, so problem with sync is not your problem anymore
POSIX Threads in c/c++ are not concurrent since the time assigned by the operative system to each parent process must be split into the number of threads it has. Thus, your algorithm is executing only core. To leverage multicore technology, you must use OpenMP. This interface library let you split your algorithm in different physic cores. This is a good OpenMP tutorial
I profiled my code and the most expensive part of the code is the loop included in the post. I want to improve the performance of this loop using AVX. I have tried manually unrolling the loop and, while that does improve performance, the improvements are not satisfactory.
int N = 100000000;
int8_t* data = new int8_t[N];
for(int i = 0; i< N; i++) { data[i] = 1 ;}
std::array<float, 10> f = {1,2,3,4,5,6,7,8,9,10};
std::vector<float> output(N, 0);
int k = 0;
for (int i = k; i < N; i = i + 2) {
for (int j = 0; j < 10; j++, k = j + 1) {
output[i] += f[j] * data[i - k];
output[i + 1] += f[j] * data[i - k + 1];
}
}
Could I have some guidance on how to approach this.
I would assume that data is a large input array of signed bytes, and f is a small array of floats of length 10, and output is the large output array of floats. Your code goes out of bounds for the first 10 iterations by i, so I will start i from 10 instead. Here is a clean version of the original code:
int s = 10;
for (int i = s; i < N; i += 2) {
for (int j = 0; j < 10; j++) {
output[i] += f[j] * data[i-j-1];
output[i+1] += f[j] * data[i-j];
}
}
As it turns out, processing two iterations by i does not change anything, so we simplify it further to:
for (int i = s; i < N; i++)
for (int j = 0; j < 10; j++)
output[i] += f[j] * data[i-j-1];
This version of code (along with declarations of input/output data) should have been present in the question itself, without others having to clean/simplify the mess.
Now it is obvious that this code applies one-dimensional convolution filter, which is a very common thing in signal processing. For instance, it can by computed in Python using numpy.convolve function. The kernel has very small length 10, so Fast Fourier Transform won't provide any benefits compared to bruteforce approach. Given that the problem is well-known, you can read a lot of articles on vectorizing small-kernel convolution. I will follow the article by hgomersall.
First, let's get rid of reverse indexing. Obviously, we can reverse the kernel before running the main algorithm. After that, we have to compute the so-called cross-correlation instead of convolution. In simple words, we move the kernel array along the input array, and compute the dot product between them for every possible offset.
std::reverse(f.data(), f.data() + 10);
for (int i = s; i < N; i++) {
int b = i-10;
float res = 0.0;
for (int j = 0; j < 10; j++)
res += f[j] * data[b+j];
output[i] = res;
}
In order to vectorize it, let's compute 8 consecutive dot products at once. Recall that we can pack eight 32-bit float numbers into one 256-bit AVX register. We will vectorize the outer loop by i, which means that:
The loop by i will be advanced by 8 every iteration.
Every value inside the outer loop turns into a 8-element pack, such that k-th element of the pack holds this value for (i+k)-th iteration of the outer loop from the scalar version.
Here is the resulting code:
//reverse the kernel
__m256 revKernel[10];
for (size_t i = 0; i < 10; i++)
revKernel[i] = _mm256_set1_ps(f[9-i]); //every component will have same value
//note: you have to compute the last 16 values separately!
for (size_t i = s; i + 16 <= N; i += 8) {
int b = i-10;
__m256 res = _mm256_setzero_ps();
for (size_t j = 0; j < 10; j++) {
//load: data[b+j], data[b+j+1], data[b+j+2], ..., data[b+j+15]
__m128i bytes = _mm_loadu_si128((__m128i*)&data[b+j]);
//convert first 8 bytes of loaded 16-byte pack into 8 floats
__m256 floats = _mm256_cvtepi32_ps(_mm256_cvtepi8_epi32(bytes));
//compute res = res + floats * revKernel[j] elementwise
res = _mm256_fmadd_ps(revKernel[j], floats, res);
}
//store 8 values packed in res into: output[i], output[i+1], ..., output[i+7]
_mm256_storeu_ps(&output[i], res);
}
For 100 millions of elements, this code takes about 120 ms on my machine, while the original scalar implementation took 850 ms. Beware: I have Ryzen 1600 CPU, so results on Intel CPUs may be somewhat different.
Now if you really want to unroll something, the inner loop by 10 kernel elements is the perfect place. Here is how it is done:
__m256 revKernel[10];
for (size_t i = 0; i < 10; i++)
revKernel[i] = _mm256_set1_ps(f[9-i]);
for (size_t i = s; i + 16 <= N; i += 8) {
size_t b = i-10;
__m256 res = _mm256_setzero_ps();
#define DOIT(j) {\
__m128i bytes = _mm_loadu_si128((__m128i*)&data[b+j]); \
__m256 floats = _mm256_cvtepi32_ps(_mm256_cvtepi8_epi32(bytes)); \
res = _mm256_fmadd_ps(revKernel[j], floats, res); \
}
DOIT(0);
DOIT(1);
DOIT(2);
DOIT(3);
DOIT(4);
DOIT(5);
DOIT(6);
DOIT(7);
DOIT(8);
DOIT(9);
_mm256_storeu_ps(&output[i], res);
}
It takes 110 ms on my machine (slightly better that the first vectorized version).
The simple copy of all elements (with conversion from bytes to floats) takes 40 ms for me, which means that this code is not memory-bound yet, and there is still some room for improvement left.
Well, here is my code and I am having a problem because my n is not increasing:
#define N 100
#define N_EQUATIONS 18 + 2
//initial values
int v = 1;
int cai = 2;
int caSR = 3;
int nai = 4;
int ki = 5;
int dvdt = 18;
double V_init = -87.5;
double Cai_init=1.0e-4;
double cansr=1.2;
double cajsr=cansr;
double CaSR_init = cansr + cajsr;
double Nai_init = 7;
double Ki_init = 145;
double u[N + 1][N_EQUATIONS + 1];
double Im[N + 1];
int main () {
int n = 0;
for ( n = 0; n <= N; n++) {
printf("n=%.18f\n", n);
u[n][v] = V_init;
//printf("t=%.18f\n", u[n][v]);
u[n][cai] = Cai_init;
//printf("cai=%.18f\n", u[n][cai]);
u[n][caSR] = CaSR_init;
u[n][nai] = Nai_init;
u[n][ki] = Ki_init;
u[n][dvdt] = 0.0;//check it
tapend[n] = 0.0;
tapstart[n] = 0.0;
}
}
Sorry if it is a stupid question and the answer is staring me at the eyes..
P.S. see the new revised code
You are probably just confused because your printf is incorrect:
printf("n=%.18f\n", n);
should be, e.g.
printf("n=%18d\n", n);
Currently you just print garbage in your loop (0 in your case, it seems, but it could be anything), so this may give the incorrect impression that n is not incrementing correctly.
Note that if you enable compiler warnings (and compiler warnings should always be enabled), then the compiler would have pointed out this mistake to you. Always enable compiler warnings and always take notice of any warnings, understand them, and fix them.
I wrote a (probably-inefficient, but anyway..) Rcpp code using inline to simulate a stochastic SEIR model.
The serial version compiles and works perfectly, but since I need to simulate from it a large number of times and since it seems to me like an embarrassingly parallel problem (just need to simulate again for other parameter values and return a matrix with the results) I tried to add #pragma omp parallel for and to compile with -fopenmp -lgomp but ... boom!
I get a segfault even for very small examples!
I tried to add setenv("OMP_STACKSIZE","24M",1); and values well over 24M but still the segfault happens.
I'll explain briefly the code since it's a bit long (I tried to shorten it but the result change and I can't reproduce it..):
I have two nested loops, the inner one execute the model for a given parameter set and the outer one changes the parameters.
The only reason a race condition might happen is if the code were trying to execute set of instructions inside inner the loop in parallel (which cannot be done because of the model structure, on iteration t it depends on iteration t-1) and not to parallelize the outer, but if I'm not mistaken that is what the parallel for constructor does for default if put just outside the outer...
This is basically the form of the code I'm trying to run:
mat result(n_param,T_MAX);
#pragma omp parallel for
for(int i=0,i<n_param_set;i++){
t=0;
rowvec jnk(T_MAX);
while(t < T_MAX){
...
jnk(t) = something(jnk(t-1));
...
t++;
}
result.row(i)=jnk;
}
return wrap(result);
And my question is: How I tell the compiler that I just want to compute in parallel the outer loop (even distributing them statically like n_loops/n_threads for each thread) and not the inner one (which is actually non-parallelizable)?
The real code is a bit more involved and I'll present it here for the sake of reproducibility if you're really willing, but I'm only asking about the behavior of OpenMP. Please notice that the only OpenMP instruction appears at line 122.
library(Rcpp);library(RcppArmadillo);library(inline)
misc='
#include <math.h>
#define _USE_MATH_DEFINES
#include <omp.h>
using namespace arma;
template <typename T> int sgn(T val) {
return (T(0) < val) - (val < T(0));
}
uvec rmultinomial(int n,vec prob)
{
int K = prob.n_elem;
uvec rN = zeros<uvec>(K);
double p_tot = sum(prob);
double pp;
for(int k = 0; k < K-1; k++) {
if(prob(k)>0) {
pp = prob[k] / p_tot;
rN(k) = ((pp < 1.) ? (rbinom(1,(double) n, pp))(0) : n);
n -= rN[k];
} else
rN[k] = 0;
if(n <= 0) /* we have all*/
return rN;
p_tot -= prob[k]; /* i.e. = sum(prob[(k+1):K]) */
}
rN[K-1] = n;
return rN;
}
'
model_and_summary='
mat SEIR_sim_plus_summaries()
{
vec alpha;
alpha << 0.002 << 0.0045;
vec beta;
beta << 0.01 << 0.01;
vec gamma;
gamma << 1.0/14.0 << 1.0/14.0;
vec sigma;
sigma << 1.0/(3.5) << 1.0/(3.5);
vec phi;
phi << 0.8 << 0.8;
int S_0 = 800;
int E_0 = 100;
int I_0 = 100;
int R_0 = 0;
int pop = 1000;
double tau = 0.01;
double t_0 = 0;
vec obs_time;
obs_time << 1 << 2 << 3 << 4 << 5 << 6 << 7 << 8 << 9 << 10 << 11 << 12 << 13 << 14 << 15 << 16 << 17 << 18 << 19 << 20 << 21 << 22 << 23 << 24;
const int n_obs = obs_time.n_elem;
const int n_part = alpha.n_elem;
mat stat(n_part,6);
//#pragma omp parallel for
for(int k=0;k<n_part;k++) {
ivec INC_i(n_obs);
ivec INC_o(n_obs);
// Event variables
double alpha_t;
int nX; //current number of people moving
vec rates(8);
uvec trans(4); // current transitions, e.g. from S to E,I,R,Universe
vec r(4); // rates e.g. from S to E, I, R, Univ.
/*********************** Initialize **********************/
int S_curr = S_0;
int S_prev = S_0;
int E_curr = E_0;
int E_prev = E_0;
int I_curr = I_0;
int I_prev = I_0;
int R_curr = R_0;
int R_prev = R_0;
int IncI_curr = 0;
int IncI_prev = 0;
int IncO_curr = 0;
int IncO_prev = 0;
double t_curr = t_0;
int t_idx =0;
while( t_idx < n_obs ) {
// next time preparation
t_curr += tau;
S_prev = S_curr;
E_prev = E_curr;
I_prev = I_curr;
R_prev = R_curr;
IncI_prev = IncI_curr;
IncO_prev = IncO_curr;
/*********************** description (rates) of the events **********************/
alpha_t = alpha(k)*(1+phi(k)*sin(2*M_PI*(t_curr+0)/52)); //real contact rate, time expressed in weeks
rates(0) = (alpha_t * ((double)I_curr / (double)pop ) * ((double)S_curr)); //e+1, s-1, r,i one s get infected (goes in E, not yey infectous)
rates(1) = (sigma(k) * E_curr); //e-1, i+1, r,s one exposed become infectous (goes in I) INCIDENCE!!
rates(2) = (gamma(k) * I_curr); //i-1, s,e, r+1 one i recover
rates(3) = (beta(k) * I_curr); //i-1, s, r,e one i dies
rates(4) = (beta(k) * R_curr); //i,e, s, r-1 one r dies
rates(5) = (beta(k) * E_curr); //e-1, s, r,i one e dies
rates(6) = (beta(k) * S_curr); //s-1 e, i ,r one s dies
rates(7) = (beta(k) * pop); //s+1 one susc is born
// Let the events occour
/*********************** S compartement **********************/
if((rates(0)+rates(6))>0){
nX = rbinom(1,S_prev,1-exp(-(rates(0)+rates(6))*tau))(0);
r(0) = rates(0)/(rates(0)+rates(6)); r(1) = 0.0; r(2) = 0; r(3) = rates(6)/(rates(0)+rates(6));
trans = rmultinomial(nX, r);
S_curr -= nX;
E_curr += trans(0);
I_curr += trans(1);
R_curr += trans(2);
//trans(3) contains dead individual, who disappear...we could avoid this using sequential conditional binomial
}
/*********************** E compartement **********************/
if((rates(1)+rates(5))>0){
nX = rbinom(1,E_prev,1-exp(-(rates(1)+rates(5))*tau))(0);
r(0) = 0.0; r(1) = rates(1)/(rates(1)+rates(5)); r(2) = 0.0; r(3) = rates(5)/(rates(1)+rates(5));
trans = rmultinomial(nX, r);
S_curr += trans(0);
E_curr -= nX;
I_curr += trans(1);
R_curr += trans(2);
IncI_curr += trans(1);
}
/*********************** I compartement **********************/
if((rates(2)+rates(3))>0){
nX = rbinom(1,I_prev,1-exp(-(rates(2)+rates(3))*tau))(0);
r(0) = 0.0; r(1) = 0.0; r(2) = rates(2)/(rates(2)+rates(3)); r(3) = rates(3)/(rates(2)+rates(3));
trans = rmultinomial(nX, r);
S_curr += trans(0);
E_curr += trans(1);
I_curr -= nX;
R_curr += trans(2);
IncO_curr += trans(2);
}
/*********************** R compartement **********************/
if(rates(4)>0){
nX = rbinom(1,R_prev,1-exp(-rates(4)*tau))(0);
r(0) = 0.0; r(1) = 0.0; r(2) = 0.0; r(3) = rates(4)/rates(4);
trans = rmultinomial(nX, r);
S_curr += trans(0);
E_curr += trans(1);
I_curr += trans(2);
R_curr -= nX;
}
/*********************** Universe **********************/
S_curr += pop - (S_curr+E_curr+I_curr+R_curr); //it should be poisson, but since the pop is fixed...
/*********************** Save & Continue **********************/
// Check if the time is interesting for us
if(t_curr > obs_time[t_idx]){
INC_i(t_idx) = IncI_curr;
INC_o(t_idx) = IncO_curr;
IncI_curr = IncI_prev = 0;
IncO_curr = IncO_prev = 0;
t_idx++;
}
//else just go on...
}
/*********************** Finished - Starting w/ stats **********************/
// INC_i is the useful variable, how can I change its reference withour copying it?
ivec incidence = INC_i; //just so if I want to use INC_o i have to change just this...
//Scan the epidemics to recover the summary stats (naively divide the data each 52 weeks)
double n_years = ceil((double)obs_time(n_obs-1)/52.0);
vec mu_attack(n_years);
vec ratio_attack(n_years-1);
vec peak(n_years);
vec atk(52);
peak(0)=0.0;
vec tmpExplo(52); //explosiveness
vec explo(n_years);
int year=0;
int week;
for(week=0 ; week<n_obs ; week++){
if(week - 52*year > 51){
mu_attack(year) = sum( atk )/(double)pop;
if(year>0)
ratio_attack(year-1) = mu_attack(year)/mu_attack(year-1);
for(int i=0;i<52;i++){
if(atk(i)>(peak(year)/2.0)){
tmpExplo(i) = 1.0;
} else {
tmpExplo(i) = 0.0;
}
}
explo(year) = sum(tmpExplo);
year++;
peak(year)=0.0;
}
atk(week-52*year) = incidence(week);
if( peak(year) < incidence(week) )
peak(year)=incidence(week);
}
if(week - 52*year > 51){
mu_attack(year) = sum( atk )/(double)pop;
} else {
ivec idx(52);
for(int i=0;i<52;i++)
{ idx(i) = i; } //take just the updated ones...
vec tmp = atk.elem(find(idx<(week - 52*year)));
mu_attack(year) = sum( tmp )/((double)pop * (tmp.n_elem/52.0));
ratio_attack(year-1) = mu_attack(year)/mu_attack(year-1);
for(int i=0;i<tmp.n_elem;i++){
if(tmp(i)>(peak(year)/2.0)){
tmpExplo(i) = 1.0;
} else {
tmpExplo(i) = 0.0;
}
}
for(int i=tmp.n_elem;i<52;i++)
tmpExplo(i) = 0.0; //to reset the others
explo(year) = sum(tmpExplo);
}
double correlation2;
double correlation4;
vec autocorr = acf(peak);
/***** ACF *****/
if(n_years<3){
correlation2=0.0;
correlation4=0.0;
} else {
if(n_years<5){
correlation2 = autocorr(1);
correlation4 = 0.0;
} else {
correlation2 = autocorr(1);
correlation4 = autocorr(3);
}
}
rowvec jnk(6);
jnk << sum(mu_attack)/(year+1.0)
<< (sum( log(ratio_attack)%log(ratio_attack) )/(n_years-1)) - (pow(sum( log(ratio_attack) )/(n_years-1),2))
<< correlation2 << correlation4 << max(peak) << sum(explo)/n_years;
stat.row(k) = jnk;
}
return stat;
}
'
main='
std::cout << "max_num_threads " << omp_get_max_threads() << std::endl;
RNGScope scope;
mat summaries = SEIR_sim_plus_summaries();
return wrap(summaries);
'
plug = getPlugin("RcppArmadillo")
## modify the plugin for Rcpp to support OpenMP
plug$env$PKG_CXXFLAGS <- paste('-fopenmp', plug$env$PKG_CXXFLAGS)
plug$env$PKG_LIBS <- paste('-fopenmp -lgomp', plug$env$PKG_LIBS)
SEIR_sim_summary = cxxfunction(sig=signature(),main,settings=plug,inc = paste(misc,model_and_summary),verbose=TRUE)
SEIR_sim_summary()
Thanks for the help!
NB: before you ask, I slightly modified the Rcpp multinomial sampling function just because I liked that way more than the one using pointer...not any other particular reason! :)
The core pseudo-random number generators (PRNGs) in R are not designed to be used in multithreaded environments. That is, their state is stored in a static array (dummy from src/main/PRNG.c) and therefore is shared among all threads. Moreover several other static structures are used to store states for the higher-level interfaces to the core PRNGs.
A possible solution could be that you put each call to rnorm() or other sampling functions inside named critical sections with all having the same name, e.g.:
...
#pragma omp critical(random)
rN(k) = ((pp < 1.) ? (rbinom(1,(double) n, pp))(0) : n);
...
if((rates(0)+rates(6))>0){
#pragma omp critical(random)
nX = rbinom(1,S_prev,1-exp(-(rates(0)+rates(6))*tau))(0);
...
Note that the critical construct operates on the structured block following it and therefore locks the entire statement. If a random number is being drawn inline inside a call to a time-consuming function, e.g.
#pragma omp critical(random)
x = slow_computation(rbinom(...));
this is better transformed to:
#pragma omp critical(random)
rb = rbinom(...);
x = slow_computation(rb);
That way only the rb = rbinom(...); statement will be protected.
I have a huge vector<vector<int>> (18M x 128). Frequently I want to take 2 rows of this vector and compare them by this function:
int getDiff(int indx1, int indx2) {
int result = 0;
int pplus, pminus, tmp;
for (int k = 0; k < 128; k += 2) {
pplus = nodeL[indx2][k] - nodeL[indx1][k];
pminus = nodeL[indx1][k + 1] - nodeL[indx2][k + 1];
tmp = max(pplus, pminus);
if (tmp > result) {
result = tmp;
}
}
return result;
}
As you see, the function, loops through the two row vectors does some subtraction and at the end returns a maximum. This function will be used a million times, so I was wondering if it can be accelerated through SSE instructions. I use Ubuntu 12.04 and gcc.
Of course it is microoptimization but it would helpful if you could provide some help, since I know nothing about SSE. Thanks in advance
Benchmark:
int nofTestCases = 10000000;
vector<int> nodeIds(nofTestCases);
vector<int> goalNodeIds(nofTestCases);
vector<int> results(nofTestCases);
for (int l = 0; l < nofTestCases; l++) {
nodeIds[l] = randomNodeID(18000000);
goalNodeIds[l] = randomNodeID(18000000);
}
double time, result;
time = timestamp();
for (int l = 0; l < nofTestCases; l++) {
results[l] = getDiff2(nodeIds[l], goalNodeIds[l]);
}
result = timestamp() - time;
cout << result / nofTestCases << "s" << endl;
time = timestamp();
for (int l = 0; l < nofTestCases; l++) {
results[l] = getDiff(nodeIds[l], goalNodeIds[l]);
}
result = timestamp() - time;
cout << result / nofTestCases << "s" << endl;
where
int randomNodeID(int n) {
return (int) (rand() / (double) (RAND_MAX + 1.0) * n);
}
/** Returns a timestamp ('now') in seconds (incl. a fractional part). */
inline double timestamp() {
struct timeval tp;
gettimeofday(&tp, NULL);
return double(tp.tv_sec) + tp.tv_usec / 1000000.;
}
FWIW I put together a pure SSE version (SSE4.1) which seems to run around 20% faster than the original scalar code on a Core i7:
#include <smmintrin.h>
int getDiff_SSE(int indx1, int indx2)
{
int result[4] __attribute__ ((aligned(16))) = { 0 };
const int * const p1 = &nodeL[indx1][0];
const int * const p2 = &nodeL[indx2][0];
const __m128i vke = _mm_set_epi32(0, -1, 0, -1);
const __m128i vko = _mm_set_epi32(-1, 0, -1, 0);
__m128i vresult = _mm_set1_epi32(0);
for (int k = 0; k < 128; k += 4)
{
__m128i v1, v2, vmax;
v1 = _mm_loadu_si128((__m128i *)&p1[k]);
v2 = _mm_loadu_si128((__m128i *)&p2[k]);
v1 = _mm_xor_si128(v1, vke);
v2 = _mm_xor_si128(v2, vko);
v1 = _mm_sub_epi32(v1, vke);
v2 = _mm_sub_epi32(v2, vko);
vmax = _mm_add_epi32(v1, v2);
vresult = _mm_max_epi32(vresult, vmax);
}
_mm_store_si128((__m128i *)result, vresult);
return max(max(max(result[0], result[1]), result[2]), result[3]);
}
You probably can get the compiler to use SSE for this. Will it make the code quicker? Probably not. The reason being is that there is a lot of memory access compared to computation. The CPU is much faster than the memory and a trivial implementation of the above will already have the CPU stalling when it's waiting for data to arrive over the system bus. Making the CPU faster will just increase the amount of waiting it does.
The declaration of nodeL can have an effect on the performance so it's important to choose an efficient container for your data.
There is a threshold where optimising does have a benfit, and that's when you're doing more computation between memory reads - i.e. the time between memory reads is much greater. The point at which this occurs depends a lot on your hardware.
It can be helpful, however, to optimise the code if you've got non-memory constrained tasks that can run in prarallel so that the CPU is kept busy whilst waiting for the data.
This will be faster. Double dereference of vector of vectors is expensive. Caching one of the dereferences will help. I know it's not answering the posted question but I think it will be a more helpful answer.
int getDiff(int indx1, int indx2) {
int result = 0;
int pplus, pminus, tmp;
const vector<int>& nodetemp1 = nodeL[indx1];
const vector<int>& nodetemp2 = nodeL[indx2];
for (int k = 0; k < 128; k += 2) {
pplus = nodetemp2[k] - nodetemp1[k];
pminus = nodetemp1[k + 1] - nodetemp2[k + 1];
tmp = max(pplus, pminus);
if (tmp > result) {
result = tmp;
}
}
return result;
}
A couple of things to look at. One is the amount of data you are passing around. That will cause a bigger issue than the trivial calculation.
I've tried to rewrite it using SSE instructions (AVX) using library here
The original code on my system ran in 11.5s
With Neil Kirk's optimisation, it went down to 10.5s
EDIT: Tested the code with a debugger rather than in my head!
int getDiff(std::vector<std::vector<int>>& nodeL,int row1, int row2) {
Vec4i result(0);
const std::vector<int>& nodetemp1 = nodeL[row1];
const std::vector<int>& nodetemp2 = nodeL[row2];
Vec8i mask(-1,0,-1,0,-1,0,-1,0);
for (int k = 0; k < 128; k += 8) {
Vec8i nodeA(nodetemp1[k],nodetemp1[k+1],nodetemp1[k+2],nodetemp1[k+3],nodetemp1[k+4],nodetemp1[k+5],nodetemp1[k+6],nodetemp1[k+7]);
Vec8i nodeB(nodetemp2[k],nodetemp2[k+1],nodetemp2[k+2],nodetemp2[k+3],nodetemp2[k+4],nodetemp2[k+5],nodetemp2[k+6],nodetemp2[k+7]);
Vec8i tmp = select(mask,nodeB-nodeA,nodeA-nodeB);
Vec4i tmp_a(tmp[0],tmp[2],tmp[4],tmp[6]);
Vec4i tmp_b(tmp[1],tmp[3],tmp[5],tmp[7]);
Vec4i max_tmp = max(tmp_a,tmp_b);
result = select(max_tmp > result,max_tmp,result);
}
return horizontal_add(result);
}
The lack of branching speeds it up to 9.5s but still data is the biggest impact.
If you want to speed it up more, try to change the data structure to a single array/vector rather than a 2D one (a.l.a. std::vector) as that will reduce cache pressure.
EDIT
I thought of something - you could add a custom allocator to ensure you allocate the 2*18M vectors in a contiguous block of memory which allows you to keep the data structure and still go through it quickly. But you'd need to profile it to be sure
EDIT 2: Tested the code with a debugger rather than in my head!
Sorry Alex, this should be better. Not sure it will be faster than what the compiler can do. I still maintain that it's memory access that's the issue, so I would still try the single array approach. Give this a go though.