Merge branch 'lwJi:main' into main

EinsteinToolkit · Mar 22, 2024 · c415e61 · c415e61
2 parents 0fc34bf + 0df04eb
commit c415e61
Show file tree

Hide file tree

Showing 5 changed files with 12 additions and 11 deletions.
diff --git a/scripts/actions-cpu-real64.cfg b/scripts/actions-cpu-real64.cfg
@@ -72,9 +72,7 @@ LAPACK_DIR = /usr
 # LORENE_DIR = /usr/local
 MPI_DIR = /usr
 NSIMD_DIR = /usr/local
-NSIMD_INC_DIRS = /usr/local/include
-NSIMD_LIB_DIRS = /usr/local/lib
-NSIMD_SIMD = SSE2
+NSIMD_SIMD = AVX2
 OPENBLAS_DIR = /usr
 OPENPMD_API_DIR = /usr/local
 # OPENSSL_DIR = /usr

diff --git a/scripts/actions-cuda-real64.cfg b/scripts/actions-cuda-real64.cfg
@@ -24,7 +24,7 @@ CFLAGS = -pipe -g -std=gnu11
 #   Call parameter type does not match function signature!
 #     %tmp = load double, double* %x.addr, align 8, !dbg !1483
 #     float  %1 = call i32 @__isnanf(double %tmp), !dbg !1483
-CXXFLAGS = -pipe -g0 -std=c++17 --compiler-options -std=gnu++17 --expt-relaxed-constexpr --extended-lambda --gpu-architecture sm_75 --forward-unknown-to-host-compiler --Werror ext-lambda-captures-this --relocatable-device-code=true --objdir-as-tempdir
+CXXFLAGS = -pipe -g0 -std=c++17 --compiler-options -std=gnu++17 --expt-relaxed-constexpr --extended-lambda --gpu-architecture sm_70 --forward-unknown-to-host-compiler --Werror ext-lambda-captures-this --relocatable-device-code=true --objdir-as-tempdir
 FPPFLAGS = -traditional
 F90FLAGS = -pipe -g -fcray-pointer -ffixed-line-length-none
 LIBS = nvToolsExt gfortran
@@ -88,9 +88,7 @@ LAPACK_DIR = /usr
 # LORENE_DIR = /usr/local
 MPI_DIR = /usr
 NSIMD_DIR = /usr/local
-NSIMD_INC_DIRS = /usr/local/include
-NSIMD_LIB_DIRS = /usr/local/lib
-NSIMD_SIMD = SSE2
+NSIMD_SIMD = AVX2
 OPENBLAS_DIR = /usr
 OPENPMD_API_DIR = /usr/local
 # OPENSSL_DIR = /usr

diff --git a/scripts/actions-cuda-real64.ini b/scripts/actions-cuda-real64.ini
@@ -22,6 +22,9 @@ runscript       = actions-cuda-real64.run
 make            = make -j @MAKEJOBS@
 makejobs        = 2
 disabled-thorns = <<EOT
+    # Not needed with CUDA
+    ExternalLibraries/NSIMD
+
     ExternalLibraries/BLAS
     ExternalLibraries/LAPACK
     ExternalLibraries/LORENE

diff --git a/scripts/actions-rocm-real64.cfg b/scripts/actions-rocm-real64.cfg
@@ -23,7 +23,7 @@ FPPFLAGS = -traditional
 F90FLAGS = -pipe -g -fcray-pointer -ffixed-line-length-none
 
 SYS_INC_DIRS = /opt/rocm/include
-LIBDIRS = /usr/local/lib /opt/rocm/hip/lib
+LIBDIRS = /usr/local/lib /opt/rocm/lib
 LIBS = amdhip64   dl hwloc ltdl open-pal open-rte udev util   gfortran
 
 C_LINE_DIRECTIVES = yes
@@ -84,9 +84,7 @@ LAPACK_DIR = /usr
 # LORENE_DIR = /usr/local
 MPI_DIR = /usr
 NSIMD_DIR = /usr/local
-NSIMD_INC_DIRS = /usr/local/include
-NSIMD_LIB_DIRS = /usr/local/lib
-NSIMD_SIMD = SSE2
+NSIMD_SIMD = AVX2
 OPENBLAS_DIR = /usr
 OPENPMD_API_DIR = /usr/local
 # OPENSSL_DIR = /usr

diff --git a/scripts/actions-rocm-real64.ini b/scripts/actions-rocm-real64.ini
@@ -22,6 +22,10 @@ runscript       = actions-rocm-real64.run
 make            = make -j @MAKEJOBS@
 makejobs        = 2
 disabled-thorns = <<EOT
+    # This change breaks CI... why?
+    # # Not needed with rocm
+    # ExternalLibraries/NSIMD
+
     CactusTest/TestMath         # long double exists but does not quite work
     ExternalLibraries/BLAS
     ExternalLibraries/LAPACK