From 26e2c4f617ba1c47caa8e5c1fe812a221f411df7 Mon Sep 17 00:00:00 2001
From: Morris Hafner <mhafner@nvidia.com>
Date: Fri, 30 Aug 2024 12:34:48 +0000
Subject: [PATCH] Add nvc++ support

---
 Eigen/src/Core/arch/NEON/Complex.h | 15 ++++++---------
 Eigen/src/Core/util/Macros.h       |  7 +++++++
 cmake/EigenTesting.cmake           |  3 ++-
 3 files changed, 15 insertions(+), 10 deletions(-)

diff --git a/Eigen/src/Core/arch/NEON/Complex.h b/Eigen/src/Core/arch/NEON/Complex.h
index a0467113c..4190d1bd1 100644
--- a/Eigen/src/Core/arch/NEON/Complex.h
+++ b/Eigen/src/Core/arch/NEON/Complex.h
@@ -502,13 +502,10 @@ EIGEN_STRONG_INLINE Packet2cf pexp<Packet2cf>(const Packet2cf& a) {
 //---------- double ----------
 #if EIGEN_ARCH_ARM64 && !EIGEN_APPLE_DOUBLE_NEON_BUG
 
-// See bug 1325, clang fails to call vld1q_u64.
-#if EIGEN_COMP_CLANG || EIGEN_COMP_CASTXML || EIGEN_COMP_CPE
-static uint64x2_t p2ul_CONJ_XOR = {0x0, 0x8000000000000000};
-#else
-const uint64_t p2ul_conj_XOR_DATA[] = {0x0, 0x8000000000000000};
-static uint64x2_t p2ul_CONJ_XOR = vld1q_u64(p2ul_conj_XOR_DATA);
-#endif
+inline uint64x2_t p2ul_CONJ_XOR() {
+  static const uint64_t p2ul_conj_XOR_DATA[] = {0x0, 0x8000000000000000};
+  return vld1q_u64(p2ul_conj_XOR_DATA);
+}
 
 struct Packet1cd {
   EIGEN_STRONG_INLINE Packet1cd() {}
@@ -592,7 +589,7 @@ EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) {
 
 template <>
 EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) {
-  return Packet1cd(vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(a.v), p2ul_CONJ_XOR)));
+  return Packet1cd(vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(a.v), p2ul_CONJ_XOR())));
 }
 
 #ifdef __ARM_FEATURE_COMPLEX
@@ -622,7 +619,7 @@ EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1c
   // Multiply the imag a with b
   v2 = vmulq_f64(v2, b.v);
   // Conjugate v2
-  v2 = vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(v2), p2ul_CONJ_XOR));
+  v2 = vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(v2), p2ul_CONJ_XOR()));
   // Swap real/imag elements in v2.
   v2 = preverse<Packet2d>(v2);
   // Add and return the result
diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h
index 91c821bd5..226844f64 100644
--- a/Eigen/src/Core/util/Macros.h
+++ b/Eigen/src/Core/util/Macros.h
@@ -196,6 +196,13 @@
 #define EIGEN_COMP_PGI 0
 #endif
 
+/// \internal EIGEN_COMP_NVHPC set to NVHPC version if the compiler is nvc++
+#if defined(__NVCOMPILER)
+#define EIGEN_COMP_NVHPC (__NVCOMPILER_MAJOR__ * 100 + __NVCOMPILER_MINOR__)
+#else
+#define EIGEN_COMP_NVHPC 0
+#endif
+
 /// \internal EIGEN_COMP_ARM set to 1 if the compiler is ARM Compiler
 #if defined(__CC_ARM) || defined(__ARMCC_VERSION)
 #define EIGEN_COMP_ARM 1
diff --git a/cmake/EigenTesting.cmake b/cmake/EigenTesting.cmake
index 9f75c1b8a..a1488fc47 100644
--- a/cmake/EigenTesting.cmake
+++ b/cmake/EigenTesting.cmake
@@ -75,7 +75,8 @@ macro(ei_add_test_internal testname testname_with_suffix)
 
   # let the user pass flags.
   if(${ARGC} GREATER 2)
-    target_compile_options(${targetname} PRIVATE ${ARGV2})
+    separate_arguments(compile_options NATIVE_COMMAND ${ARGV2})
+    target_compile_options(${targetname} PRIVATE ${compile_options})
   endif()
 
   if(EIGEN_TEST_CUSTOM_CXX_FLAGS)