Fix FP flags for NVHPC and add docs

CMakeLists.txt

@@ -37,7 +37,10 @@ endif ()
 
 # setup some defaults flags for everything
 set(DEFAULT_DEBUG_FLAGS -O2)
-set(DEFAULT_RELEASE_FLAGS -O3 -march=native -ffast-math)
+set(DEFAULT_RELEASE_FLAGS -O3
+        $<$<OR:$<CXX_COMPILER_ID:NVHPC>,$<CXX_COMPILER_ID:PGI>>:-fast>
+        $<$<AND:$<NOT:$<CXX_COMPILER_ID:NVHPC>>,$<NOT:$<CXX_COMPILER_ID:PGI>>>:-ffast-math>
+)
 
 macro(hint_flag FLAG DESCRIPTION)
     if (NOT DEFINED ${FLAG})
@@ -283,7 +286,7 @@ if ((CMAKE_GENERATOR MATCHES "Unix Makefiles") OR (CMAKE_GENERATOR MATCHES "Ninj
             -DSRC_DIR_PREFIX="${CMAKE_SOURCE_DIR}/src"
             -DOUT_DIR_PREFIX="CMakeFiles/${EXE_NAME}.dir"
             -P ${CMAKE_SOURCE_DIR}/cmake/extract_compile_command.cmake
-            )
+    )
     add_dependencies(${EXE_NAME} extract_compile_commands)
 else ()
     set(COMPILE_COMMANDS COMPILE_FLAGS)

README.md

@@ -40,7 +40,7 @@ $ cd miniBUDE
 
 # configure the build, build type defaults to Release
 # The -DMODEL flag is required
-$ cmake -Bbuild -H. -DMODEL=<model> <model specific flags prefixed with -D...>
+$ cmake -Bbuild -H. -DMODEL=<model> -DCXX_EXTRA_FLAGS=-march=native <more model specific flags prefixed with -D...>
 
 # compile
 $ cmake --build build
@@ -54,6 +54,13 @@ The source for each model's implementations are located in `./src/<model>`.
 
 Currently available models are:
 
+**For optional benchmark performance, please set `-DCXX_EXTRA_FLAGS=-march=native` or
+equivalent. `-march=native` is no longer added by default to prevent bad arch selection on ARM
+platforms.**
+
+**This benchmark should be compiled with relaxed FP rules (e.g. `-O3 -ffast-math` or `-Ofast`); a
+suitable flags will set automatically if compiling with GCC/Clang and NVHPC.**
+
 ```
 omp;ocl;std-indices;std-ranges;hip;cuda;kokkos;sycl;acc;raja;tbb;thrust
 ```