*******> update.13

Author: David S. Cerutti

Date: January 21, 2019

Programs: pmemd.cuda

Description: This prepares pmemd.cuda to work with the new Turing architecture,
             and CUDA 10.  Also removes the need to specify -volta during
             configuration to get V100 optimizations.  They are now standard.
--------------------------------------------------------------------------------

 AmberTools/src/configure2                          | 182 +++++++++------------
 1 files changed, 107 insertions(+), 167 deletions(-)

diff --git AmberTools/src/configure2 AmberTools/src/configure2
index 8276f969ef..c5df08d480 100755
--- AmberTools/src/configure2
+++ AmberTools/src/configure2
@@ -165,11 +165,6 @@ cat<<EOD
       -afe           Build alternative AFE implementation of Free Energy Methods
                      (for debugging purposes)
       -gem           Build the pmemd.gem program
-      -volta         Build the NVIDIA CUDA version of pmemd (pmemd.cuda or
-                     pmemd.cuda.MPI) with experimental optimizations for the Volta
-                     architecture. Resulting executable will not work on other
-                     GPU architectures. (this option will be deprecated as the
-                     optimizations become standardized in a later release)
       -mic_native    Builds Intel Xeon Phi native version of pmemd.
                      (EXPERIMENTAL) Requires Intel Compiler 2012 or later.  For
                      linking MKL use -mkl and unset MKL_HOME.
@@ -616,7 +611,6 @@ while [ $# -gt 0 ]; do
         -cuda)          cuda='yes';rism='no';mdgx='no';mtkpp='' ;;
         -afe)           afe='yes';;
         -gem)           pmemd_gem='yes';;
-        -volta)         volta='yes';;
         -mic)           mic='yes';;
         -mic_native)    mic='yes'; build_netcdf='build_netcdf' ;;
         -mic_offload)   mic_offload='yes';mpi='yes';intelmpi='yes';mpinab='mpi';;
@@ -1201,48 +1195,32 @@ if [ "$cuda" = 'yes' ]; then
     if [ -z "$NVCC" ]; then nvcc="$CUDA_HOME/bin/nvcc"; else nvcc="$NVCC"; fi
     #Experimental SM7.0 = V100 and Volta Geforce / GTX Ampere?
     #Note results in executable that will not run on other GPUs
-    if [ "$volta" = 'yes' ]; then
-	sm70flags='-gencode arch=compute_60,code=sm_70 -DVOLTAOPT'
-	sm62flags=''
-	sm61flags=''
-	sm60flags=''
-	sm53flags=''
-	sm52flags=''
-	sm50flags=''
-	sm37flags=''
-	sm35flags=''
-	sm30flags=''
-    else
-	sm70flags='-gencode arch=compute_60,code=sm_70'
-	#SM6.2 = ???
-	sm62flags='-gencode arch=compute_62,code=sm_62'
-	#SM6.1 = GP106 = GTX-1070, GP104 = GTX-1080, GP102 = Titan-X[P]
-	sm61flags='-gencode arch=compute_61,code=sm_61'
-	#SM6.0 = GP100 / P100 = DGX-1
-	sm60flags='-gencode arch=compute_60,code=sm_60'
-	#SM5.3 = GM200 [Grid] = M60, M40?
-	sm53flags='-gencode arch=compute_53,code=sm_53'
-	#SM5.2 = GM200 = GTX-Titan-X, M6000 etc.
-	sm52flags='-gencode arch=compute_52,code=sm_52'
-	#SM5.0 = GM204 = GTX980, 970 etc
-	sm50flags='-gencode arch=compute_50,code=sm_50'
-	#SM3.7 = GK210 = K80
-	sm37flags='-gencode arch=compute_37,code=sm_37'
-	#SM3.5 = GK110 + 110B = K20, K20X, K40, GTX780, GTX-Titan, GTX-Titan-Black, GTX-Titan-Z
-	sm35flags='-gencode arch=compute_35,code=sm_35'
-	#SM3.0 = GK104 = K10, GTX680, 690 etc.
-	sm30flags='-gencode arch=compute_30,code=sm_30'
-    fi
-	
+    sm70flags='-gencode arch=compute_60,code=sm_70, -gencode arch=compute_61,code=sm_70'
+    #SM6.2 = ???
+    sm62flags='-gencode arch=compute_62,code=sm_62'
+    #SM6.1 = GP106 = GTX-1070, GP104 = GTX-1080, GP102 = Titan-X[P]
+    sm61flags='-gencode arch=compute_61,code=sm_61'
+    #SM6.0 = GP100 / P100 = DGX-1
+    sm60flags='-gencode arch=compute_60,code=sm_60'
+    #SM5.3 = GM200 [Grid] = M60, M40?
+    sm53flags='-gencode arch=compute_53,code=sm_53'
+    #SM5.2 = GM200 = GTX-Titan-X, M6000 etc.
+    sm52flags='-gencode arch=compute_52,code=sm_52'
+    #SM5.0 = GM204 = GTX980, 970 etc
+    sm50flags='-gencode arch=compute_50,code=sm_50'
+    #SM3.7 = GK210 = K80
+    sm37flags='-gencode arch=compute_37,code=sm_37'
+    #SM3.5 = GK110 + 110B = K20, K20X, K40, GTX780, GTX-Titan, GTX-Titan-Black, GTX-Titan-Z
+    sm35flags='-gencode arch=compute_35,code=sm_35'
+    #SM3.0 = GK104 = K10, GTX680, 690 etc.
+    sm30flags='-gencode arch=compute_30,code=sm_30'
+        
     cudaversion=`$nvcc --version | grep 'release' | cut -d' ' -f5 | cut -d',' -f1`
     if [ "$cudaversion" = "9.0" -o "$cudaversion" = "9.1"  -o \
-         "$cudaversion" = "9.2" ]; then
+         "$cudaversion" = "9.2" -o "$cudaversion" = "10.0" -o \
+         "$cudaversion" = "10.1" -o "$cudaversion" = "10.2" ]; then
       echo "CUDA Version $cudaversion detected"
-      if [ "$volta" = 'yes' ]; then
-	  echo "Configuring for SM7.0 only with experimental optimizations"
-      else
-          echo "Configuring for SM3.0, SM3.5, SM3.7, SM5.0, SM5.2, SM5.3, SM6.0, SM6.1 and SM7.0"
-      fi
+      echo "Configuring for SM3.0, SM3.5, SM3.7, SM5.0, SM5.2, SM5.3, SM6.0, SM6.1 and SM7.0"
       nvccflags="$sm30flags $sm35flags $sm37flags $sm50flags $sm52flags $sm53flags $sm60flags $sm61flags $sm70flags -Wno-deprecated-declarations"
     elif [ "$cudaversion" = "8.0" ]; then
       echo "CUDA Version $cudaversion detected"
@@ -1256,7 +1234,7 @@ if [ "$cuda" = 'yes' ]; then
       nvccflags="$sm30flags  $sm35flags $sm37flags $sm50flags $sm52flags $sm53flags"
     else
       echo "Error: Unsupported CUDA version $cudaversion detected."
-      echo "       AMBER 18 requires CUDA version 7.5, 8.0, or 9.0 - 9.2"
+      echo "       AMBER 18 requires CUDA version 7.5, 8.0, 9.0-9.2, or 10.0-10.2"
       exit 1
     fi
     nvcc="$nvcc $nvccflags"
@@ -1386,11 +1364,11 @@ gnu)
         # OpenMP support requires GNU compiler version 4.2 or higher.
         if ( [ "$cc_version_major" -eq 4 -a "$cc_version_minor" -lt 2 ] ) \
                 || [ "$cc_version_major" -lt 4 ]; then
-		echo "Error: OpenMP support requires GNU compiler version 4.2 or higher."
-		exit 1
-	fi
+                echo "Error: OpenMP support requires GNU compiler version 4.2 or higher."
+                exit 1
+        fi
 
-	# Enable OpenMP in PMEMD code.
+        # Enable OpenMP in PMEMD code.
         if [ "$mpi" = 'yes' ]; then
           foptflags="$foptflags -fopenmp -D_OPENMP_"
         fi
@@ -1400,9 +1378,9 @@ gnu)
         # OpenMP support requires GNU compiler version 4.2 or higher.
         if ( [ "$cc_version_major" -eq 4 -a "$cc_version_minor" -lt 2 ] ) \
                 || [ "$cc_version_major" -lt 4 ]; then
-		echo "Error: OpenMP support requires GNU compiler version 4.2 or higher."
-		exit 1
-	fi
+                echo "Error: OpenMP support requires GNU compiler version 4.2 or higher."
+                exit 1
+        fi
 
         if [ ! -z $dragonegg ]; then
             echo "OpenMP is incompatible with dragonegg"
@@ -1500,12 +1478,6 @@ gnu)
       fi
 
     fi
-
-    # Volta optimizations
-    if [ "$volta" = 'yes' ]; then
-	pmemd_coptflags="$cflags -DVOLTAOPT"
-        pmemd_foptflags="$fflags -DVOLTAOPT"
-    fi
     ;;
 
 #################### icc #######
@@ -1633,11 +1605,11 @@ intel)
              omp_flag="-qopenmp -DOPENMP"
              flibs_arch="$flibs_arch -qopenmp"
              flibsf_arch="$flibsf_arch -qopenmp"
-	else
+        else
              omp_flag="-openmp -DOPENMP"
              flibs_arch="$flibs_arch -openmp"
              flibsf_arch="$flibsf_arch -openmp"
-	fi
+        fi
       #if [ "$fc_version_major" -lt 11 ] ; then
       #fi
     fi
@@ -1769,12 +1741,6 @@ intel)
        fi
     fi
 
-    # Volta optimizations
-    if [ "$volta" = 'yes' ]; then
-	pmemd_coptflags="$cflags -DVOLTAOPT"
-        pmemd_foptflags="$fflags -DVOLTAOPT"
-    fi
-
     #XEON PHI Specifics
     if [ "$mic" = 'yes' ]; then
       pmemd_foptflags="$pmemd_foptflags -mmic"
@@ -1785,32 +1751,32 @@ intel)
 
     #MIC2 Precision model, optimization flags.
     if [ "$mic2" = 'yes' ]; then
-		if [ "$intelmpi" = 'yes' -a "$pmemd_openmp" = 'yes' ]; then 
-			if [ "$fc_version_major" -le 15 -a "$fc_version_major" -ge 13 ]; then
-		     	pmemd_foptflags="$pmemd_foptflags -DMIC2 -openmp-simd"
-		     	pmemd_coptflags="$pmemd_coptflags -DMIC2 -openmp-simd"
-		    elif [ "$fc_version_major" -ge 16 ]; then
-		     	pmemd_foptflags="$pmemd_foptflags -DMIC2 -qopenmp-simd"
-		     	pmemd_coptflags="$pmemd_coptflags -DMIC2 -qopenmp-simd"
-			fi
-			if [ "$spdp" = 'yes' ]; then
-		       pmemd_foptflags="$pmemd_foptflags -Dpmemd_SPDP -Dfaster_MIC2"
-		       pmemd_coptflags="$pmemd_coptflags -Dpmemd_SPDP" 
-		   		if [ -z "$MKL_HOME" ]; then
-		        	echo "Configure error: Cannot use MIC2_SPDP optimizations without Intel MPI, Intel OpenMP, and Intel MKL on."
-		     		echo "Set MKL_HOME to the Intel MKL root directory"
-		     		echo "Then use: ./configure -mic2_spdp -openmp -intelmpi intel"
-		     		exit 1
-		   		fi	
-		 	else
-		       pmemd_foptflags="$pmemd_foptflags -Dpmemd_DPDP"
-		       pmemd_coptflags="$pmemd_coptflags -Dpmemd_DPDP" 
-		    fi
-		else
-			echo "Configure error: Cannot use MIC2 optimizations without Intel MPI & OpenMP on."
-		 	echo "Use: ./configure -mic2 -openmp -intelmpi intel"
-		 	exit 1
-		fi
+                if [ "$intelmpi" = 'yes' -a "$pmemd_openmp" = 'yes' ]; then 
+                        if [ "$fc_version_major" -le 15 -a "$fc_version_major" -ge 13 ]; then
+                             pmemd_foptflags="$pmemd_foptflags -DMIC2 -openmp-simd"
+                             pmemd_coptflags="$pmemd_coptflags -DMIC2 -openmp-simd"
+                    elif [ "$fc_version_major" -ge 16 ]; then
+                             pmemd_foptflags="$pmemd_foptflags -DMIC2 -qopenmp-simd"
+                             pmemd_coptflags="$pmemd_coptflags -DMIC2 -qopenmp-simd"
+                        fi
+                        if [ "$spdp" = 'yes' ]; then
+                       pmemd_foptflags="$pmemd_foptflags -Dpmemd_SPDP -Dfaster_MIC2"
+                       pmemd_coptflags="$pmemd_coptflags -Dpmemd_SPDP" 
+                                   if [ -z "$MKL_HOME" ]; then
+                                echo "Configure error: Cannot use MIC2_SPDP optimizations without Intel MPI, Intel OpenMP, and Intel MKL on."
+                                     echo "Set MKL_HOME to the Intel MKL root directory"
+                                     echo "Then use: ./configure -mic2_spdp -openmp -intelmpi intel"
+                                     exit 1
+                                   fi        
+                         else
+                       pmemd_foptflags="$pmemd_foptflags -Dpmemd_DPDP"
+                       pmemd_coptflags="$pmemd_coptflags -Dpmemd_DPDP" 
+                    fi
+                else
+                        echo "Configure error: Cannot use MIC2 optimizations without Intel MPI & OpenMP on."
+                         echo "Use: ./configure -mic2 -openmp -intelmpi intel"
+                         exit 1
+                fi
     fi
 
     #midpoint optimization flags.
@@ -1888,24 +1854,24 @@ intel)
     #Enabling OpenMP optimizations for pmemd.MPI
     if [ "$pmemd_openmp" = 'yes' ]; then
       if [ "$intelmpi" = 'yes' -o "$mpi" = 'yes' -o "$mic_native" = 'yes' -o "$mic2" = 'yes' ]; then
-	    if [ "$fc_version_major" -le 15 -a "$fc_version_major" -ge 13 ]; then
-          	pmemd_foptflags="$pmemd_foptflags -openmp -D_OPENMP_"
+            if [ "$fc_version_major" -le 15 -a "$fc_version_major" -ge 13 ]; then
+                  pmemd_foptflags="$pmemd_foptflags -openmp -D_OPENMP_"
             elif [ "$fc_version_major" -ge 16 ]; then
-          	pmemd_foptflags="$pmemd_foptflags -qopenmp -D_OPENMP_"
-	    fi
-    	if [ "$mic" = 'yes' ]; then
-	    if [ "$fc_version_major" -le 15 -a "$fc_version_major" -ge 13 ]; then
-      		pmemd_foptflags="$pmemd_foptflags -opt-streaming-cache-evict=0 -fp-model fast=2 -align array64byte -opt-streaming-stores always -opt-ra-region-strategy=default -opt-malloc-options=4 -opt-gather-scatter-unroll=2" 
+                  pmemd_foptflags="$pmemd_foptflags -qopenmp -D_OPENMP_"
+            fi
+            if [ "$mic" = 'yes' ]; then
+            if [ "$fc_version_major" -le 15 -a "$fc_version_major" -ge 13 ]; then
+                      pmemd_foptflags="$pmemd_foptflags -opt-streaming-cache-evict=0 -fp-model fast=2 -align array64byte -opt-streaming-stores always -opt-ra-region-strategy=default -opt-malloc-options=4 -opt-gather-scatter-unroll=2" 
             elif [ "$fc_version_major" -ge 16 ]; then
-      		pmemd_foptflags="$pmemd_foptflags -qopt-streaming-cache-evict=0 -fp-model fast=2 -align array64byte -qopt-streaming-stores always -qopt-ra-region-strategy=default -qopt-malloc-options=4 -qopt-gather-scatter-unroll=2" 
-	    fi
-	else
-	    if [ "$fc_version_major" -le 15 -a "$fc_version_major" -ge 13 ]; then
-      		pmemd_foptflags="$pmemd_foptflags -opt-streaming-cache-evict=0 -fp-model fast=2 -align array64byte" 
+                      pmemd_foptflags="$pmemd_foptflags -qopt-streaming-cache-evict=0 -fp-model fast=2 -align array64byte -qopt-streaming-stores always -qopt-ra-region-strategy=default -qopt-malloc-options=4 -qopt-gather-scatter-unroll=2" 
+            fi
+        else
+            if [ "$fc_version_major" -le 15 -a "$fc_version_major" -ge 13 ]; then
+                      pmemd_foptflags="$pmemd_foptflags -opt-streaming-cache-evict=0 -fp-model fast=2 -align array64byte" 
             elif [ "$fc_version_major" -ge 16 ]; then
-      		pmemd_foptflags="$pmemd_foptflags -qopt-streaming-cache-evict=0 -fp-model fast=2 -align array64byte" 
-	    fi
-	fi
+                      pmemd_foptflags="$pmemd_foptflags -qopt-streaming-cache-evict=0 -fp-model fast=2 -align array64byte" 
+            fi
+        fi
       fi
     fi