AMReX_FLATTEN_FOR (AMReX-Codes#3855)

A new build option to force flattening of ParallelFor and similar functions for host device. The default is yes unless it's a debug build. The CMake option is AMReX_FLATTEN_FOR, whereas for GNU make, it's USE_FLATTEN_FOR.
Jiasense · Mar 28, 2024 · 8d57ebc · 8d57ebc
1 parent 21a7a66
commit 8d57ebc
Show file tree

Hide file tree

Showing 10 changed files with 61 additions and 0 deletions.
diff --git a/.github/workflows/gcc.yml b/.github/workflows/gcc.yml
@@ -305,6 +305,7 @@ jobs:
         cmake ..                        \
             -DCMAKE_BUILD_TYPE=Release  \
             -DCMAKE_VERBOSE_MAKEFILE=ON \
+            -DAMReX_FLATTEN_FOR=OFF     \
             -DAMReX_ASSERTIONS=ON       \
             -DAMReX_TESTING=ON          \
             -DAMReX_BOUND_CHECK=ON      \

diff --git a/Docs/sphinx_documentation/source/BuildingAMReX.rst b/Docs/sphinx_documentation/source/BuildingAMReX.rst
@@ -532,6 +532,9 @@ The list of available options is reported in the :ref:`table <tab:cmakevar>` bel
    +------------------------------+-------------------------------------------------+-------------------------+-----------------------+
    | AMReX_PROBINIT               |  Enable support for probin file                 | Platform dependent      | YES, NO               |
    +------------------------------+-------------------------------------------------+-------------------------+-----------------------+
+   | AMReX_FLATTEN_FOR            |  Enable flattening of ParallelFor and similar   | YES unless for Debug    | YES, NO               |
+   |                              |  functions for host code                        | build                   |                       |
+   +------------------------------+-------------------------------------------------+-------------------------+-----------------------+
 .. raw:: latex
 
    \end{center}

diff --git a/Src/Base/AMReX_Extension.H b/Src/Base/AMReX_Extension.H
@@ -145,6 +145,12 @@
 #define AMREX_FLATTEN
 #endif
 
+#ifdef AMREX_USE_FLATTEN_FOR
+#define AMREX_ATTRIBUTE_FLATTEN_FOR AMREX_FLATTEN
+#else
+#define AMREX_ATTRIBUTE_FLATTEN_FOR
+#endif
+
 // unroll loop
 #define AMREX_TO_STRING_HELPER(X)   #X
 #define AMREX_TO_STRING(X)          AMREX_TO_STRING_HELPER(X)

diff --git a/Src/Base/AMReX_GpuLaunchFunctsC.H b/Src/Base/AMReX_GpuLaunchFunctsC.H
@@ -68,6 +68,7 @@ void launch (T const& n, L&& f) noexcept
 }
 
 template <typename T, typename L, typename M=std::enable_if_t<std::is_integral_v<T>> >
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void For (T n, L const& f) noexcept
 {
     for (T i = 0; i < n; ++i) {
@@ -96,6 +97,7 @@ void For (Gpu::KernelInfo const&, T n, L&& f) noexcept
 }
 
 template <typename T, typename L, typename M=std::enable_if_t<std::is_integral_v<T>> >
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void ParallelFor (T n, L const& f) noexcept
 {
     AMREX_PRAGMA_SIMD
@@ -125,6 +127,7 @@ void ParallelFor (Gpu::KernelInfo const&, T n, L&& f) noexcept
 }
 
 template <typename L>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void For (Box const& box, L const& f) noexcept
 {
     const auto lo = amrex::lbound(box);
@@ -157,6 +160,7 @@ void For (Gpu::KernelInfo const&, Box const& box, L&& f) noexcept
 }
 
 template <typename L>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void ParallelFor (Box const& box, L const& f) noexcept
 {
     const auto lo = amrex::lbound(box);
@@ -190,6 +194,7 @@ void ParallelFor (Gpu::KernelInfo const&, Box const& box, L&& f) noexcept
 }
 
 template <typename T, typename L, typename M=std::enable_if_t<std::is_integral_v<T>> >
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void For (Box const& box, T ncomp, L const& f) noexcept
 {
     const auto lo = amrex::lbound(box);
@@ -224,6 +229,7 @@ void For (Gpu::KernelInfo const&, Box const& box, T ncomp, L&& f) noexcept
 }
 
 template <typename T, typename L, typename M=std::enable_if_t<std::is_integral_v<T>> >
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void ParallelFor (Box const& box, T ncomp, L const& f) noexcept
 {
     const auto lo = amrex::lbound(box);
@@ -1037,6 +1043,7 @@ void HostDeviceFor (Gpu::KernelInfo const&,
 }
 
 template <typename T, typename L, typename M=std::enable_if_t<std::is_integral_v<T>> >
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void ParallelForRNG (T n, L const& f) noexcept
 {
     for (T i = 0; i < n; ++i) {
@@ -1045,6 +1052,7 @@ void ParallelForRNG (T n, L const& f) noexcept
 }
 
 template <typename L>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void ParallelForRNG (Box const& box, L const& f) noexcept
 {
     const auto lo = amrex::lbound(box);
@@ -1057,6 +1065,7 @@ void ParallelForRNG (Box const& box, L const& f) noexcept
 }
 
 template <typename T, typename L, typename M=std::enable_if_t<std::is_integral_v<T>> >
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void ParallelForRNG (Box const& box, T ncomp, L const& f) noexcept
 {
     const auto lo = amrex::lbound(box);

diff --git a/Src/Base/AMReX_Loop.H b/Src/Base/AMReX_Loop.H
@@ -3,11 +3,13 @@
 #include <AMReX_Config.H>
 
 #include <AMReX_Box.H>
+#include <AMReX_Extension.H>
 
 namespace amrex {
 
 template <class F>
 AMREX_GPU_HOST_DEVICE
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void Loop (Dim3 lo, Dim3 hi, F const& f) noexcept
 {
     for (int k = lo.z; k <= hi.z; ++k) {
@@ -19,6 +21,7 @@ void Loop (Dim3 lo, Dim3 hi, F const& f) noexcept
 
 template <class F>
 AMREX_GPU_HOST_DEVICE
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void Loop (Dim3 lo, Dim3 hi, int ncomp, F const& f) noexcept
 {
     for (int n = 0; n < ncomp; ++n) {
@@ -31,6 +34,7 @@ void Loop (Dim3 lo, Dim3 hi, int ncomp, F const& f) noexcept
 
 template <class F>
 AMREX_GPU_HOST_DEVICE
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopConcurrent (Dim3 lo, Dim3 hi, F const& f) noexcept
 {
     for (int k = lo.z; k <= hi.z; ++k) {
@@ -43,6 +47,7 @@ void LoopConcurrent (Dim3 lo, Dim3 hi, F const& f) noexcept
 
 template <class F>
 AMREX_GPU_HOST_DEVICE
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopConcurrent (Dim3 lo, Dim3 hi, int ncomp, F const& f) noexcept
 {
     for (int n = 0; n < ncomp; ++n) {
@@ -56,6 +61,7 @@ void LoopConcurrent (Dim3 lo, Dim3 hi, int ncomp, F const& f) noexcept
 
 template <class F>
 AMREX_GPU_HOST_DEVICE
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void Loop (Box const& bx, F const& f) noexcept
 {
     const auto lo = amrex::lbound(bx);
@@ -69,6 +75,7 @@ void Loop (Box const& bx, F const& f) noexcept
 
 template <class F>
 AMREX_GPU_HOST_DEVICE
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void Loop (Box const& bx, int ncomp, F const& f) noexcept
 {
     const auto lo = amrex::lbound(bx);
@@ -83,6 +90,7 @@ void Loop (Box const& bx, int ncomp, F const& f) noexcept
 
 template <class F>
 AMREX_GPU_HOST_DEVICE
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopConcurrent (Box const& bx, F const& f) noexcept
 {
     const auto lo = amrex::lbound(bx);
@@ -97,6 +105,7 @@ void LoopConcurrent (Box const& bx, F const& f) noexcept
 
 template <class F>
 AMREX_GPU_HOST_DEVICE
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopConcurrent (Box const& bx, int ncomp, F const& f) noexcept
 {
     const auto lo = amrex::lbound(bx);
@@ -116,6 +125,7 @@ void LoopConcurrent (Box const& bx, int ncomp, F const& f) noexcept
 // of the warning, we have to use the functions below for those situations.
 
 template <class F>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopOnCpu (Dim3 lo, Dim3 hi, F const& f) noexcept
 {
     for (int k = lo.z; k <= hi.z; ++k) {
@@ -126,6 +136,7 @@ void LoopOnCpu (Dim3 lo, Dim3 hi, F const& f) noexcept
 }
 
 template <class F>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopOnCpu (Dim3 lo, Dim3 hi, int ncomp, F const& f) noexcept
 {
     for (int n = 0; n < ncomp; ++n) {
@@ -137,6 +148,7 @@ void LoopOnCpu (Dim3 lo, Dim3 hi, int ncomp, F const& f) noexcept
 }
 
 template <class F>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopConcurrentOnCpu (Dim3 lo, Dim3 hi, F const& f) noexcept
 {
     for (int k = lo.z; k <= hi.z; ++k) {
@@ -148,6 +160,7 @@ void LoopConcurrentOnCpu (Dim3 lo, Dim3 hi, F const& f) noexcept
 }
 
 template <class F>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopConcurrentOnCpu (Dim3 lo, Dim3 hi, int ncomp, F const& f) noexcept
 {
     for (int n = 0; n < ncomp; ++n) {
@@ -160,6 +173,7 @@ void LoopConcurrentOnCpu (Dim3 lo, Dim3 hi, int ncomp, F const& f) noexcept
 }
 
 template <class F>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopOnCpu (Box const& bx, F const& f) noexcept
 {
     const auto lo = amrex::lbound(bx);
@@ -172,6 +186,7 @@ void LoopOnCpu (Box const& bx, F const& f) noexcept
 }
 
 template <class F>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopOnCpu (Box const& bx, int ncomp, F const& f) noexcept
 {
     const auto lo = amrex::lbound(bx);
@@ -185,6 +200,7 @@ void LoopOnCpu (Box const& bx, int ncomp, F const& f) noexcept
 }
 
 template <class F>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopConcurrentOnCpu (Box const& bx, F const& f) noexcept
 {
     const auto lo = amrex::lbound(bx);
@@ -198,6 +214,7 @@ void LoopConcurrentOnCpu (Box const& bx, F const& f) noexcept
 }
 
 template <class F>
+AMREX_ATTRIBUTE_FLATTEN_FOR
 void LoopConcurrentOnCpu (Box const& bx, int ncomp, F const& f) noexcept
 {
     const auto lo = amrex::lbound(bx);

diff --git a/Tools/CMake/AMReXConfig.cmake.in b/Tools/CMake/AMReXConfig.cmake.in
@@ -90,6 +90,7 @@ set(AMReX_HDF5_ZFP_FOUND            @AMReX_HDF5_ZFP@)
 set(AMReX_FPE_FOUND                 @AMReX_FPE@)
 set(AMReX_PIC_FOUND                 @AMReX_PIC@)
 set(AMReX_ASSERTIONS_FOUND          @AMReX_ASSERTIONS@)
+set(AMReX_FLATTEN_FOR_FOUND         @AMReX_FLATTEN_FOR@)
 
 # Profiling options
 set(AMReX_BASEP_FOUND               @AMReX_BASE_PROFILE@)
@@ -141,6 +142,7 @@ set(AMReX_HDF5_ZFP                  @AMReX_HDF5_ZFP@)
 set(AMReX_FPE                       @AMReX_FPE@)
 set(AMReX_PIC                       @AMReX_PIC@)
 set(AMReX_ASSERTIONS                @AMReX_ASSERTIONS@)
+set(AMReX_FLATTEN_FOR               @AMReX_FLATTEN_FOR@)
 
 # Profiling options
 set(AMReX_BASE_PROFILE              @AMReX_BASE_PROFILE@)

diff --git a/Tools/CMake/AMReXOptions.cmake b/Tools/CMake/AMReXOptions.cmake
@@ -377,6 +377,13 @@ endif ()
 
 print_option( AMReX_ASSERTIONS )
 
+if ( "${CMAKE_BUILD_TYPE}" MATCHES "Debug" )
+   option( AMReX_FLATTEN_FOR "Enable flattening of ParallelFor and other similar functions" OFF)
+else ()
+   option( AMReX_FLATTEN_FOR "Enable flattening of ParallelFor and other similar functions" ON)
+endif ()
+print_option( AMReX_FLATTEN_FOR )
+
 option(AMReX_BOUND_CHECK  "Enable bound checking in Array4 class" OFF)
 print_option( AMReX_BOUND_CHECK )
 

diff --git a/Tools/CMake/AMReXSetDefines.cmake b/Tools/CMake/AMReXSetDefines.cmake
@@ -79,6 +79,9 @@ add_amrex_define( AMREX_${CMAKE_SYSTEM_NAME} )
 #  Assertions
 add_amrex_define( AMREX_USE_ASSERTION NO_LEGACY IF AMReX_ASSERTIONS )
 
+# Flatten
+add_amrex_define( AMREX_USE_FLATTEN_FOR NO_LEGACY IF AMReX_FLATTEN_FOR )
+
 # Bound checking
 add_amrex_define( AMREX_BOUND_CHECK NO_LEGACY IF AMReX_BOUND_CHECK )
 

diff --git a/Tools/CMake/AMReX_Config_ND.H.in b/Tools/CMake/AMReX_Config_ND.H.in
@@ -32,6 +32,7 @@
 #cmakedefine BL_SPACEDIM @D@
 #endif
 #cmakedefine AMREX_USE_ASSERTION
+#cmakedefine AMREX_USE_FLATTEN_FOR
 #cmakedefine AMREX_BOUND_CHECK
 #cmakedefine AMREX_EXPORT_DYNAMIC
 #cmakedefine BL_FORT_USE_UNDERSCORE

diff --git a/Tools/GNUMake/Make.defs b/Tools/GNUMake/Make.defs
@@ -311,6 +311,12 @@ else
   USE_FORCE_INLINE := FALSE
 endif
 
+ifdef USE_FLATTEN_FOR
+  USE_FLATTEN_FOR := $(strip $(USE_FLATTEN_FOR))
+else
+  USE_FLATTEN_FOR := TRUE
+endif
+
 ifdef WARN_ALL
   WARN_ALL := $(strip $(WARN_ALL))
 else
@@ -714,6 +720,12 @@ ifeq ($(USE_FORCE_INLINE),TRUE)
     CPPFLAGS += -DAMREX_USE_FORCE_INLINE
 endif
 
+ifeq ($(USE_FLATTEN_FOR),TRUE)
+ifneq ($(DEBUG),TRUE)
+    CPPFLAGS += -DAMREX_USE_FLATTEN_FOR
+endif
+endif
+
 ifeq ($(USE_ACC),TRUE)
 
   USE_GPU := TRUE