correctly build offload for default OMP_FLAGS_OFFLOAD_* settings

.github/workflows/main.yaml

@@ -44,7 +44,7 @@ jobs:
     runs-on: ubuntu-22.04
     defaults:
       run:
-        working-directory: ./src/julia/BabelStream.jl
+        working-directory: ./src/julia/JuliaStream.jl
     steps:
       - uses: actions/checkout@v4
       - name: Setup project

README.md

@@ -44,7 +44,7 @@ BabelStream is currently implemented in the following parallel programming model
 - Futhark
 
 This project also contains implementations in alternative languages with different build systems:
-* Julia - [BabelStream.jl](./src/julia/BabelStream.jl)
+* Julia - [JuliaStream.jl](./src/julia/JuliaStream.jl)
 * Java - [java-stream](./src/java/java-stream)
 * Scala - [scala-stream](./src/scala/scala-stream)
 * Rust - [rust-stream](./src/rust/rust-stream)

src/julia/BabelStream.jl/Project.toml → src/julia/JuliaStream.jl/Project.toml

@@ -1,4 +1,4 @@
-name = "BabelStream"
+name = "JuliaStream"
 uuid = "1bdcc9b7-f5ed-4705-bc7b-be1b748ec681"
 authors = ["Wei-Chen Lin <[email protected]>"]
 version = "4.0"
@@ -16,14 +16,4 @@ ROCKernels = "7eb9e9f0-4bd3-4c4c-8bef-26bd9629d9b9"
 oneAPI = "8f75cd03-7ff8-4ecb-9b8f-daf728133b1b"
 
 [compat]
-AMDGPU = "0.4.8"
-ArgParse = "1.1.4"
-CUDA = "4.0.1"
-CUDAKernels = "0.4.7"
-Distributed = "1"
-ExprTools = "0.1.10"
-KernelAbstractions = "0.8.6"
-Parameters = "0.12.3"
-ROCKernels = "0.3.5"
-oneAPI = "1.0.2"
 julia = "1.9"

src/julia/BabelStream.jl/README.md → src/julia/JuliaStream.jl/README.md

@@ -1,4 +1,4 @@
-BabelStream.jl
+JuliaStream.jl
 ==============
 
 This is an implementation of BabelStream in Julia which contains the following variants:
@@ -35,15 +35,15 @@ A set of reduced dependency projects are available for the following backend and
 With Julia on path, run your selected benchmark with:
 
 ```shell
-> cd BabelStream.jl
+> cd JuliaStream.jl
 > julia --project=<BACKEND> -e 'import Pkg; Pkg.instantiate()' # only required on first run
 > julia --project=<BACKEND> src/<IMPL>Stream.jl
 ```
 
 For example. to run the CUDA implementation:
 
 ```shell
-> cd BabelStream.jl
+> cd JuliaStream.jl
 > julia --project=CUDA -e 'import Pkg; Pkg.instantiate()' 
 > julia --project=CUDA src/CUDAStream.jl
 ```
@@ -61,7 +61,7 @@ There may be instances where some packages are locked to an older version becaus
 
 To run the benchmark using the top-level project, run the benchmark with:
 ```shell
-> cd BabelStream.jl
+> cd JuliaStream.jl
 > julia --project -e 'import Pkg; Pkg.instantiate()'  
 > julia --project src/<IMPL>Stream.jl
-```
+```

src/julia/BabelStream.jl/src/JuliaStream.jl → src/julia/JuliaStream.jl/src/JuliaStream.jl

@@ -1,4 +1,4 @@
-module BabelStream
+module JuliaStream
 end
 
-println("Please run benchmarks directly via `julia --project src/<IMPL>Stream.jl`")
+println("Please run benchmarks directly via `julia --project src/<IMPL>Stream.jl`")

src/omp/model.cmake

@@ -51,7 +51,7 @@ set(OMP_FLAGS_OFFLOAD_INTEL
 set(OMP_FLAGS_OFFLOAD_GNU_NVIDIA
         -foffload=nvptx-none)
 set(OMP_FLAGS_OFFLOAD_GNU_AMD
-        -foffload=amdgcn-amdhsa)
+        -foffload=amdgcn-amdhsa -fcf-protection=none)
 set(OMP_FLAGS_OFFLOAD_CLANG_NVIDIA
         -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda -Xopenmp-target=nvptx64-nvidia-cuda)
 set(OMP_FLAGS_OFFLOAD_CLANG_AMD
@@ -165,6 +165,7 @@ macro(setup)
             list(APPEND OMP_FLAGS ${OMP_FLAGS_OFFLOAD_${OFFLOAD_VENDOR}})
 
         elseif (LEN EQUAL 2)
+            register_definitions(OMP_TARGET_GPU)
             #  offload with <vendor:arch> tuple
             list(GET OFFLOAD_TUPLE 0 OFFLOAD_VENDOR)
             list(GET OFFLOAD_TUPLE 1 OFFLOAD_ARCH)