Merge pull request #29 from bd4/pr/oneapi-2023

Pr/oneapi 2023

.github/workflows/cuda.yml

@@ -0,0 +1,19 @@
+name: CUDA Tests
+
+on:
+  push:
+    branches: [ master, 'pr/*' ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build-cuda:
+    runs-on: ubuntu-latest
+    container: nvidia/cuda:11.0.3-devel-ubuntu20.04
+    env:
+      DEBIAN_FRONTEND: noninteractive
+      BACKEND: cuda
+    steps:
+    - uses: actions/checkout@v3
+    - name: make
+      run: cd $BACKEND; make -j4

.github/workflows/hip.yml

@@ -0,0 +1,19 @@
+name: HIP Tests
+
+on:
+  push:
+    branches: [ master, 'pr/*' ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build-hip:
+    runs-on: ubuntu-latest
+    container: rocm/dev-ubuntu-22.04:5.4.2-complete
+    env:
+      DEBIAN_FRONTEND: noninteractive
+      BACKEND: hip
+    steps:
+    - uses: actions/checkout@v3
+    - name: make
+      run: cd $BACKEND; make -j4

.github/workflows/omp.yml

@@ -0,0 +1,27 @@
+name: OMP Tests
+
+on:
+  push:
+    branches: [ master, 'pr/*' ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  omp-run-cpu:
+    runs-on: ubuntu-latest
+    env:
+      DEBIAN_FRONTEND: noninteractive
+      CC: gcc
+      BACKEND: openmp-threading
+    steps:
+    - uses: actions/checkout@v3
+      with:
+        submodules: true
+    - name: make
+      run: cd $BACKEND; make -j4
+    - name: run nuclide
+      run: ./$BACKEND/XSBench -m event -s small -G nuclide
+    - name: run hash
+      run: ./$BACKEND/XSBench -m event -s small -G hash
+    - name: run unionized
+      run: ./$BACKEND/XSBench -m event -s small -G unionized

.github/workflows/oneapi.yml

@@ -0,0 +1,45 @@
+name: SYCL Tests
+
+on:
+  push:
+    branches: [ master, 'pr/*' ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  sycl-oneapi-run-cpu:
+    runs-on: ubuntu-latest
+    container: intel/oneapi-basekit:2023.1.0-devel-ubuntu22.04
+    env:
+      DEBIAN_FRONTEND: noninteractive
+      ONEAPI_DEVICE_SELECTOR: opencl:cpu
+    steps:
+    - uses: actions/checkout@v3
+      with:
+        submodules: true
+    - name: sycl-ls
+      run: |
+        which sycl-ls
+        sycl-ls
+    - name: make sycl
+      run: cd sycl; CC=oneapi make -j4
+    - name: run sycl nuclide
+      run: ./sycl/XSBench -m event -s small -G nuclide
+    - name: run sycl hash
+      run: ./sycl/XSBench -m event -s small -G hash
+    - name: run sycl unionized
+      run: ./sycl/XSBench -m event -s small -G unionized
+    - name: make openmp-offload
+      run: cd openmp-offload; COMPILER=intel make -j4
+    - name: run openmp-offload nuclide
+      run: ./openmp-offload/XSBench -m event -s small -G nuclide
+    - name: run openmp-offload hash
+      run: ./openmp-offload/XSBench -m event -s small -G hash
+    - name: run openmp-offload unionized
+      run: ./openmp-offload/XSBench -m event -s small -G unionized
+    - name: install opencl packages
+      run: |
+        apt-get -y update && apt-get install -y --no-install-recommends intel-oneapi-runtime-opencl clinfo opencl-headers intel-opencl-icd
+        clinfo
+    - name: make opencl
+      run: cd opencl; COMPILER=gnu make -j4

.gitignore

@@ -0,0 +1,2 @@
+*.o
+*/XSBench

opencl/Makefile

@@ -2,11 +2,11 @@
 # User Options
 #===============================================================================
 
-COMPILER    = gnu
-OPTIMIZE    = yes
-DEBUG       = no
-PROFILE     = no
-MPI         = no
+COMPILER    ?= gnu
+OPTIMIZE    ?= yes
+DEBUG       ?= no
+PROFILE     ?= no
+MPI         ?= no
 
 #===============================================================================
 # Program name & source code list

openmp-offload/Makefile

@@ -2,11 +2,11 @@
 # User Options
 #===============================================================================
 
-COMPILER    = llvm
-OPTIMIZE    = yes
-DEBUG       = no
-PROFILE     = no
-MPI         = no
+COMPILER    ?= llvm
+OPTIMIZE    ?= yes
+DEBUG       ?= no
+PROFILE     ?= no
+MPI         ?= no
 
 #===============================================================================
 # Program name & source code list

sycl/Makefile

@@ -2,11 +2,11 @@
 # User Options
 #===============================================================================
 
-CC          = llvm
-OPTIMIZE    = yes
-DEBUG       = no
-PROFILE     = no
-MPI         = no
+CC          ?= llvm
+OPTIMIZE    ?= yes
+DEBUG       ?= no
+PROFILE     ?= no
+MPI         ?= no
 
 #===============================================================================
 # Program name & source code list
@@ -29,7 +29,7 @@ obj = $(source:.cpp=.o)
 #===============================================================================
 
 # Standard Flags
-CFLAGS := -std=c++14 -Wall
+CFLAGS := -std=c++17 -Wall
 
 # Linker Flags
 LDFLAGS = -lm
@@ -48,6 +48,17 @@ ifeq ($(CC),llvm)
   LDFLAGS += -lOpenCL
 endif
 
+# IntelOneAPI Compiler
+ifeq ($(CC),oneapi)
+  CC = icpx
+  CFLAGS += -fsycl
+endif
+
+ifeq ($(CC),opensycl)
+  CC = syclcc
+  CFLAGS += -fsycl
+endif
+
 # Debug Flags
 ifeq ($(DEBUG),yes)
   CFLAGS += -g

sycl/Simulation.cpp

@@ -12,7 +12,7 @@
 ////////////////////////////////////////////////////////////////////////////////////
 
 // use SYCL namespace to reduce symbol names
-using namespace cl::sycl;
+using namespace sycl;
 unsigned long long run_event_based_simulation(Inputs in, SimulationData SD, int mype, double * kernel_init_time)
 {
 
@@ -48,10 +48,8 @@ unsigned long long run_event_based_simulation(Inputs in, SimulationData SD, int
 	{
 		// create a queue using the default device for the platform (cpu, gpu)
 
-		queue sycl_q{default_selector()};
-		//queue sycl_q{gpu_selector()};
-		//queue sycl_q{cpu_selector()};
-		if(mype == 0 ) printf("Running on: %s\n", sycl_q.get_device().get_info<cl::sycl::info::device::name>().c_str());
+		queue sycl_q{default_selector_v};
+		if(mype == 0 ) printf("Running on: %s\n", sycl_q.get_device().get_info<sycl::info::device::name>().c_str());
 		if(mype == 0 ) printf("Initializing device buffers and JIT compiling kernel...\n");
 
 		////////////////////////////////////////////////////////////////////////////////
@@ -84,7 +82,7 @@ unsigned long long run_event_based_simulation(Inputs in, SimulationData SD, int
 		// For the unionized grid, this is a large array. Enforce that it is able to be allocated on the
 		// OpenCL device (as some OpenCL devices have fairly low limts here for some reason...)
 		size_t index_grid_allocation_sz = ceil((SD.length_index_grid * sizeof(int)));
-		assert( index_grid_allocation_sz <= sycl_q.get_device().get_info<cl::sycl::info::device::max_mem_alloc_size>() );
+		assert( index_grid_allocation_sz <= sycl_q.get_device().get_info<sycl::info::device::max_mem_alloc_size>() );
 		buffer<int, 1> index_grid_d(SD.index_grid, (unsigned long long ) SD.length_index_grid);
 
 		////////////////////////////////////////////////////////////////////////////////

sycl/XSbench_header.h

@@ -10,7 +10,7 @@
 #include<assert.h>
 #include<stdint.h>
 #include <chrono> 
-#include <CL/sycl.hpp>
+#include <sycl/sycl.hpp>
 
 // Papi Header
 #ifdef PAPI