Improvements / fixes.
This commit is contained in:
@ -1,3 +1,4 @@
|
||||
using LinearAlgebra
|
||||
|
||||
function test_cpu_interpreter(nrows; parallel = false)
|
||||
exprs = [
|
||||
@ -13,12 +14,15 @@ function test_cpu_interpreter(nrows; parallel = false)
|
||||
|
||||
# warmup
|
||||
interpret_cpu(exprs, X, p)
|
||||
expr_reps = 100 # for each expr
|
||||
reps= 100
|
||||
|
||||
if parallel
|
||||
t_sec = @elapsed Threads.@threads :static for i in 1:100 interpret_cpu(exprs, X, p) end
|
||||
println("~ $(round(30 * 100 * nrows / 1e9 / t_sec, digits=2)) GFLOPS (single-core) ($(round(peakflops(1000, eltype=Float32, ntrials=1, parallel=false) / 1e9, digits=2)) GFLOPS (peak, single-core))")
|
||||
t_sec = @elapsed fetch.([Threads.@spawn interpret_cpu(exprs, X, p; repetitions=expr_reps) for i in 1:reps])
|
||||
println("~ $(round(30 * reps * expr_reps * nrows / 1e9 / t_sec, digits=2)) GFLOPS ($(Threads.nthreads()) threads) ($(round(peakflops(1000, eltype=Float32, ntrials=1) / 1e9, digits=2)) GFLOPS (peak, single-core))")
|
||||
else
|
||||
t_sec = @elapsed for i in 1:100 interpret_cpu(exprs, X, p) end
|
||||
println("~ $(round(30 * 100 * nrows / 1e9 / t_sec, digits=2)) GFLOPS ($(Threads.nthreads()) threads) ($(round(peakflops(1000, eltype=Float32, ntrials=1, parallel=false) / 1e9, digits=2)) GFLOPS (peak, single-core))")
|
||||
t_sec = @elapsed for i in 1:reps interpret_cpu(exprs, X, p; repetitions=expr_reps) end
|
||||
println("~ $(round(30 * reps * expr_reps * nrows / 1e9 / t_sec, digits=2)) GFLOPS (single-core) ($(round(peakflops(1000, eltype=Float32, ntrials=1) / 1e9, digits=2)) GFLOPS (peak, single-core))")
|
||||
end
|
||||
true
|
||||
end
|
||||
|
Reference in New Issue
Block a user