add ifdefs to enable NVHPC as device compiler (#3801)

## Summary This works around limitations in the NVHPC device compiler by disabling `int128` support and avoiding a static local device variable. This enables experimental use of NVHPC 24.3+ as the unified host and device compiler for CUDA. ## Additional background #3591
AMReX-Codes · Mar 14, 2024 · 201e6a9 · 201e6a9
1 parent 5c849fd
commit 201e6a9
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 1 deletion.
diff --git a/Src/Base/AMReX_Algorithm.H b/Src/Base/AMReX_Algorithm.H
@@ -289,7 +289,11 @@ int clz (T x) noexcept;
 AMREX_GPU_HOST_DEVICE AMREX_FORCE_INLINE
 int clz_generic (std::uint8_t x) noexcept
 {
+#if !defined(__NVCOMPILER)
     static constexpr int clz_lookup[16] = { 4, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0 };
+#else
+    constexpr int clz_lookup[16] = { 4, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0 };
+#endif
     auto upper = x >> 4;
     auto lower = x & 0xF;
     return upper ? clz_lookup[upper] : 4 + clz_lookup[lower];

diff --git a/Src/Base/AMReX_INT.H b/Src/Base/AMReX_INT.H
@@ -31,7 +31,7 @@ namespace amrex {
 }
 #endif
 
-#if (defined(__x86_64) || defined (__aarch64__)) && !defined(_WIN32) && (defined(__GNUC__) || defined(__clang__))
+#if (defined(__x86_64) || defined (__aarch64__)) && !defined(_WIN32) && (defined(__GNUC__) || defined(__clang__)) && !defined(__NVCOMPILER)
 
 #define AMREX_INT128_SUPPORTED 1
-Original file line number
+Diff line change
@@ Expand Up / @@ -31,7 +31,7 @@ namespace amrex { @@
     }
     #endif
-    #if (defined(__x86_64) || defined (__aarch64__)) && !defined(_WIN32) && (defined(__GNUC__) || defined(__clang__))
+    #if (defined(__x86_64) || defined (__aarch64__)) && !defined(_WIN32) && (defined(__GNUC__) || defined(__clang__)) && !defined(__NVCOMPILER)
     #define AMREX_INT128_SUPPORTED 1
@@ Expand Down @@