mirror of
https://github.com/andre-wojtowicz/blas-benchmarks
synced 2024-11-03 19:45:27 +01:00
76 lines
2.1 KiB
R
76 lines
2.1 KiB
R
# source: https://gist.github.com/andrie/24c9672f1ea39af89c66#file-rro-mkl-benchmark-r
|
|
|
|
runs = 10
|
|
results = data.frame(name=character(), time=numeric())
|
|
|
|
cat(c("Runs :", runs, "\n"))
|
|
|
|
# Initialization
|
|
|
|
set.seed (1)
|
|
m <- 10000
|
|
n <- 5000
|
|
A <- matrix (runif (m*n),m,n)
|
|
|
|
cat("Matrix Multiply : ")
|
|
cumulate = 0
|
|
for (i in 1:runs)
|
|
cumulate = cumulate + as.numeric(system.time (B <- crossprod(A))[3])
|
|
|
|
results = rbind(results, data.frame(name="Matrix Multiply",
|
|
time=cumulate/runs))
|
|
cat(c(cumulate/runs, "\n"))
|
|
|
|
cat("Cholesky Factorization : ")
|
|
cumulate = 0
|
|
for (i in 1:runs)
|
|
cumulate = cumulate + as.numeric(system.time (C <- chol(B))[3])
|
|
|
|
results = rbind(results, data.frame(name="Cholesky Factorization",
|
|
time=cumulate/runs))
|
|
cat(c(cumulate/runs, "\n"))
|
|
|
|
cat("Singular Value Deomposition : ")
|
|
m <- 10000
|
|
n <- 2000
|
|
A <- matrix (runif (m*n),m,n)
|
|
|
|
cumulate = 0
|
|
for (i in 1:runs)
|
|
cumulate = cumulate + as.numeric(system.time (S <- svd (A,nu=0,nv=0))[3])
|
|
|
|
results = rbind(results, data.frame(name="Singular Value Deomposition",
|
|
time=cumulate/runs))
|
|
cat(c(cumulate/runs, "\n"))
|
|
|
|
cat("Principal Components Analysis : ")
|
|
m <- 10000
|
|
n <- 2000
|
|
A <- matrix (runif (m*n),m,n)
|
|
|
|
cumulate = 0
|
|
for (i in 1:runs)
|
|
cumulate = cumulate + as.numeric(system.time (P <- prcomp(A))[3])
|
|
|
|
results = rbind(results, data.frame(name="Principal Components Analysis",
|
|
time=cumulate/runs))
|
|
cat(c(cumulate/runs, "\n"))
|
|
|
|
cat("Linear Discriminant Analysis : ")
|
|
library('MASS')
|
|
g <- 5
|
|
k <- round (m/2)
|
|
A <- data.frame (A, fac=sample (LETTERS[1:g],m,replace=TRUE))
|
|
train <- sample(1:m, k)
|
|
|
|
cumulate = 0
|
|
for (i in 1:runs)
|
|
cumulate = cumulate + as.numeric(system.time (L <- lda(fac ~., data=A, prior=rep(1,g)/g, subset=train))[3])
|
|
|
|
results = rbind(results, data.frame(name="Linear Discriminant Analysis",
|
|
time=cumulate/runs))
|
|
cat(c(cumulate/runs, "\n"))
|
|
|
|
attr(results, "runs") = runs
|
|
saveRDS(results, paste0("test-revolution-", ifelse(exists("blasLibName"), blasLibName, ""), ".rds"))
|