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

benchmarking: minor improvements to interpreter
Some checks are pending
CI / Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }} - ${{ github.event_name }} (x64, ubuntu-latest, 1.10) (push) Waiting to run
Details
CI / Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }} - ${{ github.event_name }} (x64, ubuntu-latest, 1.6) (push) Waiting to run
Details
CI / Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }} - ${{ github.event_name }} (x64, ubuntu-latest, pre) (push) Waiting to run
Details

Browse Source
This commit is contained in:
Daniel
2025-05-20 14:42:52 +02:00
parent 250deb334c
commit 2701f1aa38
2 changed files with 6 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
package/src/Interpreter.jl
View File

@ -24,17 +24,17 @@ function interpret(cudaExprs, numExprs::Integer, exprsInnerLength::Integer,
cudaResults = CuArray{Float32}(undef, variableColumns, numExprs)
# Start kernel for each expression to ensure that no warp is working on different expressions
@inbounds Threads.@threads for i in 1:numExprs # multithreaded to speedup dispatching (seems to have improved performance)
numThreads = min(variableColumns, 121)
numBlocks = cld(variableColumns, numThreads)
numThreads = min(variableColumns, 121)
numBlocks = cld(variableColumns, numThreads)
@inbounds Threads.@threads for i in 1:numExprs # multithreaded to speedup dispatching (seems to have improved performance)
@cuda threads=numThreads blocks=numBlocks fastmath=true interpret_expression(cudaExprs, cudaVars, cudaParams, cudaResults, cudaStepsize, i)
end
return cudaResults
end
const MAX_STACK_SIZE = 25 # The depth of the stack to store the values and intermediate results
const MAX_STACK_SIZE = 10 # The depth of the stack to store the values and intermediate results
function interpret_expression(expressions::CuDeviceArray{ExpressionElement}, variables::CuDeviceArray{Float32}, parameters::CuDeviceArray{Float32}, results::CuDeviceArray{Float32}, stepsize::CuDeviceArray{Int}, exprIndex::Int)
varSetIndex = (blockIdx().x - 1) * blockDim().x + threadIdx().x # ctaid.x * ntid.x + tid.x (1-based)
@inbounds variableCols = length(variables) / stepsize[3] # number of variable sets
@ -92,7 +92,6 @@ function interpret_expression(expressions::CuDeviceArray{ExpressionElement}, var
elseif opcode == SQRT
operationStack[operationStackTop] = sqrt(operationStack[operationStackTop])
elseif opcode == INV
# operationStack[operationStackTop] = 1f0 / operationStack[operationStackTop]
operationStack[operationStackTop] = inv(operationStack[operationStackTop])
end
else

4
package/test/PerformanceTuning.jl
View File

@ -29,10 +29,10 @@ expr_reps = 1
@testset "Interpreter Tuning" begin
# CUDA.@profile interpret_gpu(exprs, X, parameters; repetitions=expr_reps)
CUDA.@profile interpret_gpu(exprs, X, parameters; repetitions=expr_reps)
end
@testset "Transpiler Tuning" begin
CUDA.@profile evaluate_gpu(exprs, X, parameters; repetitions=expr_reps)
# CUDA.@profile evaluate_gpu(exprs, X, parameters; repetitions=expr_reps)
end