GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-09-17 20:03:17 +08:00
Code Issues Packages Projects Releases Wiki Activity

cherry-pick !1682 Add nvc++ support into 3.4

Browse Source
This commit is contained in:
Morris Hafner 2024-11-04 17:55:47 +00:00 committed by Rasmus Munk Larsen
parent 9df21dc8b4
commit 3e7bcf54f7
3 changed files with 14 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
Eigen/src/Core/arch/NEON/Complex.h
View File

@ -373,13 +373,10 @@ template<> EIGEN_STRONG_INLINE Packet2cf psqrt<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
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
{
@ -449,7 +446,7 @@ template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a)
{ return Packet1cd(pnegate<Packet2d>(a.v)); }
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()))); }
template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
{
@ -464,7 +461,7 @@ template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, con
// 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

7
Eigen/src/Core/util/Macros.h
View File

@ -179,6 +179,13 @@
#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

1
cmake/EigenTesting.cmake
View File

@ -81,6 +81,7 @@ macro(ei_add_test_internal testname testname_with_suffix)
# let the user pass flags.
if(${ARGC} GREATER 2)
separate_arguments(compile_options NATIVE_COMMAND ${ARGV2})
ei_add_target_property(${targetname} COMPILE_FLAGS "${ARGV2}")
endif()