Merge pull request #492 from giltirn/develop

Fixes to support CUDA > 13

Grid/lattice/Lattice_slicesum_core.h

@@ -1,7 +1,6 @@
 #pragma once
 
 #if defined(GRID_CUDA)
-
 #include <cub/cub.cuh>
 #define gpucub cub
 #define gpuError_t cudaError_t
@@ -57,8 +56,13 @@ inline void sliceSumReduction_cub_small(const vobj *Data,
   //copy offsets to device
   acceleratorCopyToDeviceAsynch(&offsets[0],d_offsets,sizeof(int)*(rd+1),computeStream);
   
+#if defined(__CUDACC__) && (__CUDACC_VER_MAJOR__ >= 13)
+  #define GRID_CUB_SUM_OP ::cuda::std::plus<>{}
+#else
+  #define GRID_CUB_SUM_OP ::cub::Sum()
+#endif
   
-  gpuError_t gpuErr = gpucub::DeviceSegmentedReduce::Reduce(temp_storage_array, temp_storage_bytes, rb_p,d_out, rd, d_offsets, d_offsets+1, ::gpucub::Sum(), zero_init, computeStream);
+  gpuError_t gpuErr = gpucub::DeviceSegmentedReduce::Reduce(temp_storage_array, temp_storage_bytes, rb_p,d_out, rd, d_offsets, d_offsets+1, GRID_CUB_SUM_OP, zero_init, computeStream);
   if (gpuErr!=gpuSuccess) {
     std::cout << GridLogError << "Lattice_slicesum_gpu.h: Encountered error during gpucub::DeviceSegmentedReduce::Reduce (setup)! Error: " << gpuErr <<std::endl;
     exit(EXIT_FAILURE);
@@ -82,11 +86,13 @@ inline void sliceSumReduction_cub_small(const vobj *Data,
   });
   
   //issue segmented reductions in computeStream
-  gpuErr = gpucub::DeviceSegmentedReduce::Reduce(temp_storage_array, temp_storage_bytes, rb_p, d_out, rd, d_offsets, d_offsets+1,::gpucub::Sum(), zero_init, computeStream);
+  gpuErr = gpucub::DeviceSegmentedReduce::Reduce(temp_storage_array, temp_storage_bytes, rb_p, d_out, rd, d_offsets, d_offsets+1, GRID_CUB_SUM_OP, zero_init, computeStream);
   if (gpuErr!=gpuSuccess) {
     std::cout << GridLogError << "Lattice_slicesum_gpu.h: Encountered error during gpucub::DeviceSegmentedReduce::Reduce! Error: " << gpuErr <<std::endl;
     exit(EXIT_FAILURE);
   }
+
+#undef GRID_CUB_SUM_OP
   
   acceleratorCopyFromDeviceAsynch(d_out,&lvSum[0],rd*sizeof(vobj),computeStream);
   

Grid/threads/Accelerator.h

@@ -96,7 +96,9 @@ void     acceleratorInit(void);
 
 #ifdef GRID_CUDA
 
+NAMESPACE_END(Grid);
 #include <cuda.h>
+NAMESPACE_BEGIN(Grid);
 
 #ifdef __CUDA_ARCH__
 #define GRID_SIMT

scripts/filelist

@@ -11,7 +11,7 @@ CCFILES=`find . -name '*.cc' -not -path '*/instantiation/*/*' -not -path '*/gamm
 
 ZWILS_FERMION_FILES=` find . -name '*.cc' -path '*/instantiation/*' -path '*/instantiation/ZWilsonImpl*' `
 WILS_FERMION_FILES=`  find . -name '*.cc' -path '*/instantiation/*' -path '*/instantiation/WilsonImpl*' `
-STAG_FERMION_FILES=`  find . -name '*.cc' -path '*/instantiation/*' -path '*/instantiation/Staggered*' `
+STAG_FERMION_FILES=`  find . -name '*.cc' -path '*/instantiation/*' -path '*/instantiation/StaggeredImpl*' `
 GP_FERMION_FILES=`    find . -name '*.cc' -path '*/instantiation/*' -path '*/instantiation/Gparity*' `
 ADJ_FERMION_FILES=`   find . -name '*.cc' -path '*/instantiation/*' -path '*/instantiation/WilsonAdj*' `
 TWOIND_FERMION_FILES=`find . -name '*.cc' -path '*/instantiation/*' -path '*/instantiation/WilsonTwoIndex*'`

systems/Frontier/sourceme.sh

@@ -2,27 +2,15 @@
 echo spack
 . /autofs/nccs-svm1_home1/paboyle/Crusher/Grid/spack/share/spack/setup-env.sh
 
-module load cce/15.0.1
 module load amd/7.0.2
-#module load amd/7.1.1
-#module load rocm/7.2.0
-#module load rocm/6.4.2
 module load cray-fftw
 module load craype-accel-amd-gfx90a
+mkdir $HOME/LD_PATH
+ln -s /opt/rocm-6.4.2/lib/libamdhip* $HOME/LD_PATH
 
 #Ugly hacks to get down level software working on current system
 export LD_LIBRARY_PATH=/opt/cray/libfabric/1.20.1/lib64/:$LD_LIBRARY_PATH
 export LD_LIBRARY_PATH=/opt/gcc/mpfr/3.1.4/lib:$LD_LIBRARY_PATH
-export LD_LIBRARY_PATH=`pwd`/:$LD_LIBRARY_PATH
+#export LD_LIBRARY_PATH=`pwd`/:$LD_LIBRARY_PATH
 export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/LD_PATH/
-
-#echo spack load c-lime
-#spack load c-lime
-#module load emacs 
-##module load PrgEnv-gnu
-##module load cray-mpich
-##module load cray-fftw
-##module load craype-accel-amd-gfx90a
-##export LD_LIBRARY_PATH=/opt/gcc/mpfr/3.1.4/lib:$LD_LIBRARY_PATH
-#Hack for lib
-##export LD_LIBRARY_PATH=`pwd`/:$LD_LIBRARY_PATH
+export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/rocm-7.0.2/lib