Merge pull request #2326 from ndtrung81/gpu-update-makefiles

update makefiles in lib/gpu for more recent architectures
2020-08-31 12:30:28 -04:00
parent a4d3b21a43 ecb1f266b5
commit 67a5db03c7
6 changed files with 133 additions and 36 deletions
--- a/lib/gpu/Makefile.linux
+++ b/lib/gpu/Makefile.linux
@ -22,13 +22,13 @@ NVCC = nvcc
 #CUDA_ARCH = -arch=sm_21

 # Kepler hardware
-CUDA_ARCH = -arch=sm_30
+#CUDA_ARCH = -arch=sm_30
 #CUDA_ARCH = -arch=sm_32
 #CUDA_ARCH = -arch=sm_35
 #CUDA_ARCH = -arch=sm_37

 # Maxwell hardware
-#CUDA_ARCH = -arch=sm_50
+CUDA_ARCH = -arch=sm_50
 #CUDA_ARCH = -arch=sm_52

 # Pascal hardware
--- a/lib/gpu/Makefile.linux.double
+++ b/lib/gpu/Makefile.linux.double
@ -7,18 +7,40 @@

 EXTRAMAKE = Makefile.lammps.standard

+ifeq ($(CUDA_HOME),)
 CUDA_HOME = /usr/local/cuda
+endif
+
 NVCC = nvcc

-# Kepler CUDA
-#CUDA_ARCH = -arch=sm_35
-# Tesla CUDA
-CUDA_ARCH = -arch=sm_21
-# newer CUDA
+# obsolete hardware. not supported by current drivers anymore.
 #CUDA_ARCH = -arch=sm_13
-# older CUDA
 #CUDA_ARCH = -arch=sm_10 -DCUDA_PRE_THREE

+# Fermi hardware
+#CUDA_ARCH = -arch=sm_20
+#CUDA_ARCH = -arch=sm_21
+
+# Kepler hardware
+#CUDA_ARCH = -arch=sm_30
+#CUDA_ARCH = -arch=sm_32
+#CUDA_ARCH = -arch=sm_35
+#CUDA_ARCH = -arch=sm_37
+
+# Maxwell hardware
+CUDA_ARCH = -arch=sm_50
+#CUDA_ARCH = -arch=sm_52
+
+# Pascal hardware
+#CUDA_ARCH = -arch=sm_60
+#CUDA_ARCH = -arch=sm_61
+
+# Volta hardware
+#CUDA_ARCH = -arch=sm_70
+
+# Turing hardware
+#CUDA_ARCH = -arch=sm_75
+
 # this setting should match LAMMPS Makefile
 # one of LAMMPS_SMALLBIG (default), LAMMPS_BIGBIG and LAMMPS_SMALLSMALL

@ -33,7 +55,7 @@ CUDA_PRECISION = -D_DOUBLE_DOUBLE

 CUDA_INCLUDE = -I$(CUDA_HOME)/include
 CUDA_LIB = -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib64/stubs
-CUDA_OPTS = -DUNIX -O3 --use_fast_math
+CUDA_OPTS = -DUNIX -O3 --use_fast_math $(LMP_INC) -Xcompiler -fPIC

 CUDR_CPP = mpic++ -DMPI_GERYON -DUCL_NO_EXIT -DMPICH_IGNORE_CXX_SEEK
 CUDR_OPTS = -O2 # -xHost -no-prec-div -ansi-alias
--- a/lib/gpu/Makefile.linux.mixed
+++ b/lib/gpu/Makefile.linux.mixed
@ -7,18 +7,41 @@

 EXTRAMAKE = Makefile.lammps.standard

+ifeq ($(CUDA_HOME),)
 CUDA_HOME = /usr/local/cuda
+endif
+
 NVCC = nvcc

-# Kepler CUDA
-#CUDA_ARCH = -arch=sm_35
-# Tesla CUDA
-CUDA_ARCH = -arch=sm_21
-# newer CUDA
+# obsolete hardware. not supported by current drivers anymore.
 #CUDA_ARCH = -arch=sm_13
 # older CUDA
 #CUDA_ARCH = -arch=sm_10 -DCUDA_PRE_THREE

+# Fermi hardware
+#CUDA_ARCH = -arch=sm_20
+#CUDA_ARCH = -arch=sm_21
+
+# Kepler hardware
+#CUDA_ARCH = -arch=sm_30
+#CUDA_ARCH = -arch=sm_32
+#CUDA_ARCH = -arch=sm_35
+#CUDA_ARCH = -arch=sm_37
+
+# Maxwell hardware
+CUDA_ARCH = -arch=sm_50
+#CUDA_ARCH = -arch=sm_52
+
+# Pascal hardware
+#CUDA_ARCH = -arch=sm_60
+#CUDA_ARCH = -arch=sm_61
+
+# Volta hardware
+#CUDA_ARCH = -arch=sm_70
+
+# Turing hardware
+#CUDA_ARCH = -arch=sm_75
+
 # this setting should match LAMMPS Makefile
 # one of LAMMPS_SMALLBIG (default), LAMMPS_BIGBIG and LAMMPS_SMALLSMALL

@ -33,7 +56,7 @@ CUDA_PRECISION = -D_SINGLE_DOUBLE

 CUDA_INCLUDE = -I$(CUDA_HOME)/include
 CUDA_LIB = -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib64/stubs
-CUDA_OPTS = -DUNIX -O3 --use_fast_math
+CUDA_OPTS = -DUNIX -O3 --use_fast_math $(LMP_INC) -Xcompiler -fPIC

 CUDR_CPP = mpic++ -DMPI_GERYON -DUCL_NO_EXIT -DMPICH_IGNORE_CXX_SEEK
 CUDR_OPTS = -O2 # -xHost -no-prec-div -ansi-alias
--- a/lib/gpu/Makefile.linux.single
+++ b/lib/gpu/Makefile.linux.single
@ -7,18 +7,40 @@

 EXTRAMAKE = Makefile.lammps.standard

+ifeq ($(CUDA_HOME),)
 CUDA_HOME = /usr/local/cuda
+endif
+
 NVCC = nvcc

-# Kepler CUDA
-#CUDA_ARCH = -arch=sm_35
-# Tesla CUDA
-CUDA_ARCH = -arch=sm_21
-# newer CUDA
+# obsolete hardware. not supported by current drivers anymore.
 #CUDA_ARCH = -arch=sm_13
-# older CUDA
 #CUDA_ARCH = -arch=sm_10 -DCUDA_PRE_THREE

+# Fermi hardware
+#CUDA_ARCH = -arch=sm_20
+#CUDA_ARCH = -arch=sm_21
+
+# Kepler hardware
+#CUDA_ARCH = -arch=sm_30
+#CUDA_ARCH = -arch=sm_32
+#CUDA_ARCH = -arch=sm_35
+#CUDA_ARCH = -arch=sm_37
+
+# Maxwell hardware
+CUDA_ARCH = -arch=sm_50
+#CUDA_ARCH = -arch=sm_52
+
+# Pascal hardware
+#CUDA_ARCH = -arch=sm_60
+#CUDA_ARCH = -arch=sm_61
+
+# Volta hardware
+#CUDA_ARCH = -arch=sm_70
+
+# Turing hardware
+#CUDA_ARCH = -arch=sm_75
+
 # this setting should match LAMMPS Makefile
 # one of LAMMPS_SMALLBIG (default), LAMMPS_BIGBIG and LAMMPS_SMALLSMALL

@ -33,7 +55,7 @@ CUDA_PRECISION = -D_SINGLE_SINGLE

 CUDA_INCLUDE = -I$(CUDA_HOME)/include
 CUDA_LIB = -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib64/stubs
-CUDA_OPTS = -DUNIX -O3 --use_fast_math
+CUDA_OPTS = -DUNIX -O3 --use_fast_math $(LMP_INC) -Xcompiler -fPIC

 CUDR_CPP = mpic++ -DMPI_GERYON -DUCL_NO_EXIT -DMPICH_IGNORE_CXX_SEEK
 CUDR_OPTS = -O2 # -xHost -no-prec-div -ansi-alias
--- a/lib/gpu/Makefile.linux_multi
+++ b/lib/gpu/Makefile.linux_multi
@ -13,17 +13,27 @@ endif

 NVCC = nvcc

-# Kepler CUDA
-#CUDA_ARCH = -arch=sm_35
-# newer CUDA
+# obsolete hardware. not supported by current drivers anymore.
 #CUDA_ARCH = -arch=sm_13
-# older CUDA
 #CUDA_ARCH = -arch=sm_10 -DCUDA_PRE_THREE
-CUDA_ARCH = -arch=sm_30

-CUDA_CODE = -gencode arch=compute_60,code=[sm_60,compute_60] -gencode arch=compute_61,code=[sm_61,compute_61] \
-	    -gencode arch=compute_30,code=[sm_30,compute_30] -gencode arch=compute_35,code=[sm_35,compute_35] \
-	    -gencode arch=compute_50,code=[sm_50,compute_50] -gencode arch=compute_52,code=[sm_52,compute_52]
+# Fermi hardware
+#CUDA_ARCH = -arch=sm_20
+#CUDA_ARCH = -arch=sm_21
+
+# Kepler hardware
+#CUDA_ARCH = -arch=sm_30
+#CUDA_ARCH = -arch=sm_32
+#CUDA_ARCH = -arch=sm_35
+#CUDA_ARCH = -arch=sm_37
+
+# Maxwell hardware
+CUDA_ARCH = -arch=sm_50
+#CUDA_ARCH = -arch=sm_52
+
+CUDA_CODE = -gencode arch=compute_50,code=[sm_50,compute_50] -gencode arch=compute_52,code=[sm_52,compute_52] \
+	    -gencode arch=compute_60,code=[sm_60,compute_60] -gencode arch=compute_61,code=[sm_61,compute_61] \
+	    -gencode arch=compute_70,code=[sm_70,compute_70] -gencode arch=compute_75,code=[sm_75,compute_75]

 CUDA_ARCH += $(CUDA_CODE)

--- a/lib/gpu/Makefile.serial
+++ b/lib/gpu/Makefile.serial
@ -13,13 +13,33 @@ endif

 NVCC = nvcc

-# Tesla CUDA
-CUDA_ARCH = -arch=sm_21
-# newer CUDA
+# obsolete hardware. not supported by current drivers anymore.
 #CUDA_ARCH = -arch=sm_13
-# older CUDA
 #CUDA_ARCH = -arch=sm_10 -DCUDA_PRE_THREE
-CUDA_ARCH = -arch=sm_35
+
+# Fermi hardware
+#CUDA_ARCH = -arch=sm_20
+#CUDA_ARCH = -arch=sm_21
+
+# Kepler hardware
+#CUDA_ARCH = -arch=sm_30
+#CUDA_ARCH = -arch=sm_32
+#CUDA_ARCH = -arch=sm_35
+#CUDA_ARCH = -arch=sm_37
+
+# Maxwell hardware
+CUDA_ARCH = -arch=sm_50
+#CUDA_ARCH = -arch=sm_52
+
+# Pascal hardware
+#CUDA_ARCH = -arch=sm_60
+#CUDA_ARCH = -arch=sm_61
+
+# Volta hardware
+#CUDA_ARCH = -arch=sm_70
+
+# Turing hardware
+#CUDA_ARCH = -arch=sm_75

 # this setting should match LAMMPS Makefile
 # one of LAMMPS_SMALLBIG (default), LAMMPS_BIGBIG and LAMMPS_SMALLSMALL
@ -35,7 +55,7 @@ CUDA_PRECISION = -D_SINGLE_DOUBLE

 CUDA_INCLUDE = -I$(CUDA_HOME)/include
 CUDA_LIB = -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib64/stubs -L../../src/STUBS -lmpi_stubs
-CUDA_OPTS = -DUNIX -O3 --use_fast_math $(LMP_INC)
+CUDA_OPTS = -DUNIX -O3 --use_fast_math $(LMP_INC) -Xcompiler -fPIC

 CUDR_CPP = g++ -DMPI_GERYON -DUCL_NO_EXIT -fPIC -I../../src/STUBS
 CUDR_OPTS = -O2 $(LMP_INC) # -xHost -no-prec-div -ansi-alias