Daniel/master-thesis
1
0
Fork 0
Code Issues 3 Pull Requests Actions Packages Projects 1 Releases Wiki Activity

Improvements / fixes.

Browse Source
This commit is contained in:
Gabriel Kronberger
2025-02-19 17:18:43 +01:00
parent 942adb8612
commit f4f39ec47c
2 changed files with 17 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
package/src/ExpressionExecutorCuda.jl
View File

@ -29,7 +29,7 @@ end
# Evaluate Expressions on the CPU
function interpret_cpu(exprs::Vector{Expr}, X::Matrix{Float32}, p::Vector{Vector{Float32}})::Matrix{Float32}
function interpret_cpu(exprs::Vector{Expr}, X::Matrix{Float32}, p::Vector{Vector{Float32}}; repetitions=1)::Matrix{Float32}
@assert axes(exprs) == axes(p)
nrows = size(X, 1)
@ -37,7 +37,14 @@ function interpret_cpu(exprs::Vector{Expr}, X::Matrix{Float32}, p::Vector{Vector
res = Matrix{Float32}(undef, nrows, length(exprs))
for i in eachindex(exprs)
CpuInterpreter.interpret!((@view res[:,i]), exprs[i], X, p[i])
# The interpreter holds the postfix code and buffers for evaluation. It is costly to create
interpreter = CpuInterpreter.Interpreter{Float32}(exprs[i], length(p[i]))
# If an expression has to be evaluated multiple times (e.g. for different parameters),
# it is worthwhile to reuse the interpreter to reduce the number of allocations
for rep in 1:repetitions
CpuInterpreter.interpret!((@view res[:,i]), interpreter, X, p[i])
end
end
res