Deprecate teh PVC directory and make a PVC-OEM generic PVC target with

no queueing system dependency -- just interactive scripts

Grid/qcd/utils/WilsonLoops.h

@@ -464,7 +464,8 @@ public:
   //U_padded: the gauge link fields padded out using the PaddedCell class
   //Cell: the padded cell class
   //gStencil: the precomputed generalized local stencil for the staple
-  static void StaplePaddedAll(std::vector<GaugeMat> &staple, const std::vector<GaugeMat> &U_padded, const PaddedCell &Cell, const GeneralLocalStencil &gStencil) {
+  static void StaplePaddedAll(std::vector<GaugeMat> &staple, const std::vector<GaugeMat> &U_padded, const PaddedCell &Cell, const GeneralLocalStencil &gStencil)
+  {
     double t0 = usecond();
     assert(U_padded.size() == Nd); assert(staple.size() == Nd);
     assert(U_padded[0].Grid() == (GridBase*)Cell.grids.back());
@@ -489,7 +490,7 @@ public:
 	autoView( gStaple_v , gStaple, AcceleratorWrite);
 	auto gStencil_v = gStencil.View();
 	
-	accelerator_for(ss, ggrid->oSites(), ggrid->Nsimd(), {
+	accelerator_for(ss, ggrid->oSites(), (size_t)ggrid->Nsimd(), {
 	    decltype(coalescedRead(Ug_dirs_v[0][0])) stencil_ss;
 	    stencil_ss = Zero();
 	    int off = outer_off;
@@ -1201,7 +1202,7 @@ public:
 	autoView( gStaple_v , gStaple, AcceleratorWrite);
 	auto gStencil_v = gStencil.View();
 
-	accelerator_for(ss, ggrid->oSites(), ggrid->Nsimd(), {
+	accelerator_for(ss, ggrid->oSites(), (size_t)ggrid->Nsimd(), {
 	    decltype(coalescedRead(Ug_dirs_v[0][0])) stencil_ss;
 	    stencil_ss = Zero();
 	    int s=offset;

Grid/stencil/GeneralLocalStencil.h

@@ -43,7 +43,7 @@ class GeneralLocalStencilView {
   int                               _npoints; // Move to template param?
   GeneralStencilEntry*  _entries_p;
 
-  accelerator_inline GeneralStencilEntry * GetEntry(int point,int osite) { 
+  accelerator_inline GeneralStencilEntry * GetEntry(int point,int osite) const { 
     return & this->_entries_p[point+this->_npoints*osite]; 
   }
 

configure.ac

@@ -41,7 +41,7 @@ AC_PROG_RANLIB
 
 ############### Get compiler informations
 AC_LANG([C++])
-AX_CXX_COMPILE_STDCXX(14,noext,mandatory)
+AX_CXX_COMPILE_STDCXX(17,noext,mandatory)
 AX_COMPILER_VENDOR
 AC_DEFINE_UNQUOTED([CXX_COMP_VENDOR],["$ax_cv_cxx_compiler_vendor"],
       [vendor of C++ compiler that will compile the code])

m4/ax_cxx_compile_stdcxx_14.m4

@@ -0,0 +1,34 @@
+# =============================================================================
+#  https://www.gnu.org/software/autoconf-archive/ax_cxx_compile_stdcxx_14.html
+# =============================================================================
+#
+# SYNOPSIS
+#
+#   AX_CXX_COMPILE_STDCXX_14([ext|noext], [mandatory|optional])
+#
+# DESCRIPTION
+#
+#   Check for baseline language coverage in the compiler for the C++14
+#   standard; if necessary, add switches to CXX and CXXCPP to enable
+#   support.
+#
+#   This macro is a convenience alias for calling the AX_CXX_COMPILE_STDCXX
+#   macro with the version set to C++14.  The two optional arguments are
+#   forwarded literally as the second and third argument respectively.
+#   Please see the documentation for the AX_CXX_COMPILE_STDCXX macro for
+#   more information.  If you want to use this macro, you also need to
+#   download the ax_cxx_compile_stdcxx.m4 file.
+#
+# LICENSE
+#
+#   Copyright (c) 2015 Moritz Klammler <moritz@klammler.eu>
+#
+#   Copying and distribution of this file, with or without modification, are
+#   permitted in any medium without royalty provided the copyright notice
+#   and this notice are preserved. This file is offered as-is, without any
+#   warranty.
+
+#serial 5
+
+AX_REQUIRE_DEFINED([AX_CXX_COMPILE_STDCXX])
+AC_DEFUN([AX_CXX_COMPILE_STDCXX_14], [AX_CXX_COMPILE_STDCXX([14], [$1], [$2])])

systems/PVC-OEM/benchmarks/select_gpu.sh

@@ -1,9 +1,8 @@
 #!/bin/bash
 
 num_tile=2
-
-gpu_id=$(( (MPI_LOCAL_RANKID % num_tile ) ))
-tile_id=$((MPI_LOCAL_RANKID / num_tile))
+gpu_id=$(( (MPI_LOCALRANKID / num_tile ) ))
+tile_id=$((MPI_LOCALRANKID % num_tile))
 
 export ZE_AFFINITY_MASK=$gpu_id.$tile_id
 

systems/PVC/benchmarks/run-1tile.sh

@@ -1,62 +0,0 @@
-#!/bin/sh
-##SBATCH -p PVC-SPR-QZEH 
-##SBATCH -p PVC-ICX-QZNW
-#SBATCH -p QZ1J-ICX-PVC
-##SBATCH -p QZ1J-SPR-PVC-2C
-
-#source /nfs/site/home/paboylex/ATS/GridNew/Grid/systems/PVC-nightly/setup.sh
-
-export NT=8
-
-export I_MPI_OFFLOAD=1
-export I_MPI_OFFLOAD_TOPOLIB=level_zero
-export I_MPI_OFFLOAD_DOMAIN_SIZE=-1
-
-# export IGC_EnableLSCFenceUGMBeforeEOT=0
-# export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file=False"
-export SYCL_DEVICE_FILTER=gpu,level_zero
-#export IGC_ShaderDumpEnable=1 
-#export IGC_DumpToCurrentDir=1
-export I_MPI_OFFLOAD_CELL=tile
-export EnableImplicitScaling=0
-export EnableWalkerPartition=0
-export ZE_AFFINITY_MASK=0.0
-mpiexec -launcher ssh -n 1 -host localhost  ./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 32.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 1 --device-mem 32768
-
-export ZE_AFFINITY_MASK=0
-export I_MPI_OFFLOAD_CELL=device
-export EnableImplicitScaling=1
-export EnableWalkerPartition=1
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-#mpiexec -launcher ssh -n 2 -host localhost  vtune -collect gpu-hotspots -knob gpu-sampling-interval=1 -data-limit=0 -r ./vtune_run4 -- ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-overlap --shm-mpi 1
-
-#mpiexec  -launcher ssh -n 1 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-overlap --shm-mpi 1
-
-#mpiexec  -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 1
-
-#mpiexec  -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-overlap --shm-mpi 1
-
-#mpiexec  -launcher ssh -n 2 -host localhost ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 0
-
-#mpirun -np 2 ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 16.32.32.64 --accelerator-threads $NT --comms-sequential --shm-mpi 0
-#mpirun -np 2 ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT --comms-sequential --shm-mpi 1
-

systems/PVC/benchmarks/run-2tile-mpi.sh

@@ -1,33 +0,0 @@
-#!/bin/bash
-##SBATCH -p PVC-SPR-QZEH 
-##SBATCH -p PVC-ICX-QZNW
-
-#SBATCH -p QZ1J-ICX-PVC
-
-#source /nfs/site/home/paboylex/ATS/GridNew/Grid/systems/PVC-nightly/setup.sh
-
-export NT=16
-
-# export IGC_EnableLSCFenceUGMBeforeEOT=0
-# export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file=False"
-#export IGC_ShaderDumpEnable=1 
-#export IGC_DumpToCurrentDir=1
-export I_MPI_OFFLOAD=1
-export I_MPI_OFFLOAD_TOPOLIB=level_zero
-export I_MPI_OFFLOAD_DOMAIN_SIZE=-1
-export SYCL_DEVICE_FILTER=gpu,level_zero
-export I_MPI_OFFLOAD_CELL=tile
-export EnableImplicitScaling=0
-export EnableWalkerPartition=0
-#export SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=1
-#export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
-export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE=0
-
-for i in 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
-do
-mpiexec -launcher ssh -n 2 -host localhost  ./wrap.sh ./Benchmark_dwf_fp32 --mpi 1.1.1.2 --grid 32.32.32.64 --accelerator-threads $NT  --shm-mpi 0  --device-mem 32768 > 1.1.1.2.log$i
-mpiexec -launcher ssh -n 2 -host localhost  ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT  --shm-mpi 0  --device-mem 32768 > 2.1.1.1.log$i 
-done
-
-mpiexec -launcher ssh -n 2 -host localhost  ./wrap.sh ./Benchmark_dwf_fp32 --mpi 2.1.1.1 --grid 64.32.32.32 --accelerator-threads $NT --comms-sequential --shm-mpi 0
-

systems/PVC/benchmarks/wrap.sh

@@ -1,9 +0,0 @@
-#!/bin/sh
-
-export ZE_AFFINITY_MASK=0.$MPI_LOCALRANKID
-
-echo Ranke $MPI_LOCALRANKID ZE_AFFINITY_MASK is $ZE_AFFINITY_MASK
-
-
-  $@
-

systems/PVC/config-command

@@ -1,16 +0,0 @@
-INSTALL=/nfs/site/home/paboylx/prereqs/
-../../configure \
-	--enable-simd=GPU \
-	--enable-gen-simd-width=64 \
-	--enable-comms=mpi-auto \
-	--disable-accelerator-cshift \
-	--disable-gparity \
-	--disable-fermion-reps \
-	--enable-shm=nvlink \
-	--enable-accelerator=sycl \
-	--enable-unified=no \
-	MPICXX=mpicxx \
-	CXX=dpcpp \
-	LDFLAGS="-fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$INSTALL/lib" \
-	CXXFLAGS="-fsycl-unnamed-lambda -fsycl -no-fma -I$INSTALL/include -Wno-tautological-compare"
-

systems/PVC/setup.sh

@@ -1,18 +0,0 @@
-export https_proxy=http://proxy-chain.intel.com:911
-#export LD_LIBRARY_PATH=/nfs/site/home/azusayax/install/lib:$LD_LIBRARY_PATH
-export LD_LIBRARY_PATH=$HOME/prereqs/lib/:$LD_LIBRARY_PATH
-
-module load intel-release
-module load intel-comp-rt/embargo-ci-neo
-
-#source /opt/intel/oneapi/PVC_setup.sh
-#source /opt/intel/oneapi/ATS_setup.sh
-#module load intel-nightly/20230331
-#module load intel-comp-rt/ci-neo-master/026093
-
-#module load intel/mpich
-module load intel/mpich/pvc45.3
-export PATH=~/ATS/pti-gpu/tools/onetrace/:$PATH
-
-#clsh embargo-ci-neo-022845
-#source /opt/intel/vtune_amplifier/amplxe-vars.sh

systems/Sunspot/benchmarks/bench.pbs

@@ -20,7 +20,7 @@ unset OMP_PLACES
 
 cd $PBS_O_WORKDIR
 
-qsub jobscript.pbs
+#qsub jobscript.pbs
 
 echo Jobid: $PBS_JOBID
 echo Running on host `hostname`
@@ -44,3 +44,4 @@ CMD="mpiexec -np ${NTOTRANKS} -ppn ${NRANKS} -d ${NDEPTH} --cpu-bind=depth -enva
 	./Benchmark_dwf_fp32 --mpi 1.1.2.6 --grid 16.32.64.192 --comms-overlap \
 	--shm-mpi 0 --shm 2048 --device-mem 32000 --accelerator-threads 32"
 
+$CMD

systems/Sunspot/benchmarks/gpu_tile_compact.sh

@@ -45,8 +45,8 @@ echo "rank $PALS_RANKID ; local rank $PALS_LOCAL_RANKID ; ZE_AFFINITY_MASK=$ZE_A
 
 if [ $PALS_LOCAL_RANKID = 0 ]
 then
-    onetrace --chrome-device-timeline "$@"
-#    "$@"
+#    onetrace --chrome-device-timeline "$@"
+    "$@"
 else
 "$@"
 fi

systems/Sunspot/config-command

@@ -11,6 +11,6 @@ TOOLS=$HOME/tools
 	--enable-unified=no \
 	MPICXX=mpicxx \
 	CXX=icpx \
-	LDFLAGS="-fiopenmp -fsycl -fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -lapmidg -L$TOOLS/lib64/" \
+	LDFLAGS="-fiopenmp -fsycl -fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$TOOLS/lib64/" \
 	CXXFLAGS="-fiopenmp -fsycl-unnamed-lambda -fsycl -I$INSTALL/include -Wno-tautological-compare -I$HOME/ -I$TOOLS/include"