search for: sum5

...(log(A)))-((n1-1)*log(v1)+(n2-1)*log(v2)+(n3-1)*log(v3)) #calculate C C=1+(1/(3*(k-1))*(((1/(n1-1))+(1/(n2-1))+(1/(n3-1)))-(1/(N-k)))) #calculate layard estimator xbar1=mean(g1) xbar2=mean(g2) xbar3=mean(g3) sum1=sum((g1-xbar1)^4) sum2=sum((g2-xbar2)^4) sum3=sum((g3-xbar3)^4) sum4=sum((g1-xbar1)^2) sum5=sum((g2-xbar2)^2) sum6=sum((g3-xbar3)^2) y= (N*(sum1+sum2+sum3))/((sum4+sum5+sum6)^2) #calculate bartlett modified statistic bar2=B/(C*(1/2)*(y-1)) bar2 pv[i]<-pchisq(bar2,2,lower=FALSE) } mean(pv<0.01) mean(pv<0.05) -- View this message in context: http://r.789695.n4.nabble.com/simulati...

Hi, when I compile (clang -O3) and optimize (opt -O3) C-code like this: sum1 = val1 + val2; sum2 = val3 + val4; sum3 = val5 + val6; sum4 = val7 + val8; sum5 = sum1 + sum2; sum6 = sum3 + sum4; sum7 = sum5 + sum6; sum += sum7; I get bitcode like this: if.end152: ; preds = %if.then150, %if.else146, %if.end137 %val8.0 = phi i32 [ -2048, %if.then150 ], [ %conv38, %if.else146 ], [ 2047, %if.end137 ]...

I'm observing that base::sum(x, na.rm=FALSE) for typeof(x) == "double" is much more time consuming when there are missing values versus when there are not. I'm observing this on both Window and Linux, but it's quite surprising to me. Currently, my main suspect is settings in on how R was built. The second suspect is my brain. I hope that someone can clarify the below